mm/mincore.c

   1 /*
   2  *      linux/mm/mincore.c
   3  *
   4  * Copyright (C) 1994-2006  Linus Torvalds
   5  */
   6
   7 /*
   8  * The mincore() system call.
   9  */
  10 #include <linux/pagemap.h>
  11 #include <linux/gfp.h>
  12 #include <linux/mm.h>
  13 #include <linux/mman.h>
  14 #include <linux/syscalls.h>
  15 #include <linux/swap.h>
  16 #include <linux/swapops.h>
  17 #include <linux/hugetlb.h>
  18
  19 #include <asm/uaccess.h>
  20 #include <asm/pgtable.h>
  21
  22 static int mincore_hugetlb(pte_t *pte, unsigned long hmask, unsigned long addr,
  23                         unsigned long end, struct mm_walk *walk)
  24 {
  25 #ifdef CONFIG_HUGETLB_PAGE
  26         unsigned char present;
  27         unsigned char *vec = walk->private;
  28
  29         /*
  30          * Hugepages under user process are always in RAM and never
  31          * swapped out, but theoretically it needs to be checked.
  32          */
  33         present = pte && !huge_pte_none(huge_ptep_get(pte));
  34         for (; addr != end; vec++, addr += PAGE_SIZE)
  35                 *vec = present;
  36         walk->private = vec;
  37 #else
  38         BUG();
  39 #endif
  40         return 0;
  41 }
  42
  43 /*
  44  * Later we can get more picky about what "in core" means precisely.
  45  * For now, simply check to see if the page is in the page cache,
  46  * and is up to date; i.e. that no page-in operation would be required
  47  * at this time if an application were to map and access this page.
  48  */
  49 static unsigned char mincore_page(struct address_space *mapping, pgoff_t pgoff)
  50 {
  51         unsigned char present = 0;
  52         struct page *page;
  53
  54         /*
  55          * When tmpfs swaps out a page from a file, any process mapping that
  56          * file will not get a swp_entry_t in its pte, but rather it is like
  57          * any other file mapping (ie. marked !present and faulted in with
  58          * tmpfs's .fault). So swapped out tmpfs mappings are tested here.
  59          */
  60 #ifdef CONFIG_SWAP
  61         if (shmem_mapping(mapping)) {
  62                 page = find_get_entry(mapping, pgoff);
  63                 /*
  64                  * shmem/tmpfs may return swap: account for swapcache
  65                  * page too.
  66                  */
  67                 if (radix_tree_exceptional_entry(page)) {
  68                         swp_entry_t swp = radix_to_swp_entry(page);
  69                         page = find_get_page(swap_address_space(swp), swp.val);
  70                 }
  71         } else
  72                 page = find_get_page(mapping, pgoff);
  73 #else
  74         page = find_get_page(mapping, pgoff);
  75 #endif
  76         if (page) {
  77                 present = PageUptodate(page);
  78                 put_page(page);
  79         }
  80
  81         return present;
  82 }
  83
  84 static int __mincore_unmapped_range(unsigned long addr, unsigned long end,
  85                                 struct vm_area_struct *vma, unsigned char *vec)
  86 {
  87         unsigned long nr = (end - addr) >> PAGE_SHIFT;
  88         int i;
  89
  90         if (vma->vm_file) {
  91                 pgoff_t pgoff;
  92
  93                 pgoff = linear_page_index(vma, addr);
  94                 for (i = 0; i < nr; i++, pgoff++)
  95                         vec[i] = mincore_page(vma->vm_file->f_mapping, pgoff);
  96         } else {
  97                 for (i = 0; i < nr; i++)
  98                         vec[i] = 0;
  99         }
 100         return nr;
 101 }
 102
 103 static int mincore_unmapped_range(unsigned long addr, unsigned long end,
 104                                    struct mm_walk *walk)
 105 {
 106         walk->private += __mincore_unmapped_range(addr, end,
 107                                                   walk->vma, walk->private);
 108         return 0;
 109 }
 110
 111 static int mincore_pte_range(pmd_t *pmd, unsigned long addr, unsigned long end,
 112                         struct mm_walk *walk)
 113 {
 114         spinlock_t *ptl;
 115         struct vm_area_struct *vma = walk->vma;
 116         pte_t *ptep;
 117         unsigned char *vec = walk->private;
 118         int nr = (end - addr) >> PAGE_SHIFT;
 119
 120         ptl = pmd_trans_huge_lock(pmd, vma);
 121         if (ptl) {
 122                 memset(vec, 1, nr);
 123                 spin_unlock(ptl);
 124                 goto out;
 125         }
 126
 127         if (pmd_trans_unstable(pmd)) {
 128                 __mincore_unmapped_range(addr, end, vma, vec);
 129                 goto out;
 130         }
 131
 132         ptep = pte_offset_map_lock(walk->mm, pmd, addr, &ptl);
 133         for (; addr != end; ptep++, addr += PAGE_SIZE) {
 134                 pte_t pte = *ptep;
 135
 136                 if (pte_none(pte))
 137                         __mincore_unmapped_range(addr, addr + PAGE_SIZE,
 138                                                  vma, vec);
 139                 else if (pte_present(pte))
 140                         *vec = 1;
 141                 else { /* pte is a swap entry */
 142                         swp_entry_t entry = pte_to_swp_entry(pte);
 143
 144                         if (non_swap_entry(entry)) {
 145                                 /*
 146                                  * migration or hwpoison entries are always
 147                                  * uptodate
 148                                  */
 149                                 *vec = 1;
 150                         } else {
 151 #ifdef CONFIG_SWAP
 152                                 *vec = mincore_page(swap_address_space(entry),
 153                                         entry.val);
 154 #else
 155                                 WARN_ON(1);
 156                                 *vec = 1;
 157 #endif
 158                         }
 159                 }
 160                 vec++;
 161         }
 162         pte_unmap_unlock(ptep - 1, ptl);
 163 out:
 164         walk->private += nr;
 165         cond_resched();
 166         return 0;
 167 }
 168
 169 /*
 170  * Do a chunk of "sys_mincore()". We've already checked
 171  * all the arguments, we hold the mmap semaphore: we should
 172  * just return the amount of info we're asked for.
 173  */
 174 static long do_mincore(unsigned long addr, unsigned long pages, unsigned char *vec)
 175 {
 176         struct vm_area_struct *vma;
 177         unsigned long end;
 178         int err;
 179         struct mm_walk mincore_walk = {
 180                 .pmd_entry = mincore_pte_range,
 181                 .pte_hole = mincore_unmapped_range,
 182                 .hugetlb_entry = mincore_hugetlb,
 183                 .private = vec,
 184         };
 185
 186         vma = find_vma(current->mm, addr);
 187         if (!vma || addr < vma->vm_start)
 188                 return -ENOMEM;
 189         mincore_walk.mm = vma->vm_mm;
 190         end = min(vma->vm_end, addr + (pages << PAGE_SHIFT));
 191         err = walk_page_range(addr, end, &mincore_walk);
 192         if (err < 0)
 193                 return err;
 194         return (end - addr) >> PAGE_SHIFT;
 195 }
 196
 197 /*
 198  * The mincore(2) system call.
 199  *
 200  * mincore() returns the memory residency status of the pages in the
 201  * current process's address space specified by [addr, addr + len).
 202  * The status is returned in a vector of bytes.  The least significant
 203  * bit of each byte is 1 if the referenced page is in memory, otherwise
 204  * it is zero.
 205  *
 206  * Because the status of a page can change after mincore() checks it
 207  * but before it returns to the application, the returned vector may
 208  * contain stale information.  Only locked pages are guaranteed to
 209  * remain in memory.
 210  *
 211  * return values:
 212  *  zero    - success
 213  *  -EFAULT - vec points to an illegal address
 214  *  -EINVAL - addr is not a multiple of PAGE_SIZE
 215  *  -ENOMEM - Addresses in the range [addr, addr + len] are
 216  *              invalid for the address space of this process, or
 217  *              specify one or more pages which are not currently
 218  *              mapped
 219  *  -EAGAIN - A kernel resource was temporarily unavailable.
 220  */
 221 SYSCALL_DEFINE3(mincore, unsigned long, start, size_t, len,
 222                 unsigned char __user *, vec)
 223 {
 224         long retval;
 225         unsigned long pages;
 226         unsigned char *tmp;
 227
 228         /* Check the start address: needs to be page-aligned.. */
 229         if (start & ~PAGE_MASK)
 230                 return -EINVAL;
 231
 232         /* ..and we need to be passed a valid user-space range */
 233         if (!access_ok(VERIFY_READ, (void __user *) start, len))
 234                 return -ENOMEM;
 235
 236         /* This also avoids any overflows on PAGE_ALIGN */
 237         pages = len >> PAGE_SHIFT;
 238         pages += (offset_in_page(len)) != 0;
 239
 240         if (!access_ok(VERIFY_WRITE, vec, pages))
 241                 return -EFAULT;
 242
 243         tmp = (void *) __get_free_page(GFP_USER);
 244         if (!tmp)
 245                 return -EAGAIN;
 246
 247         retval = 0;
 248         while (pages) {
 249                 /*
 250                  * Do at most PAGE_SIZE entries per iteration, due to
 251                  * the temporary buffer size.
 252                  */
 253                 down_read(&current->mm->mmap_sem);
 254                 retval = do_mincore(start, min(pages, PAGE_SIZE), tmp);
 255                 up_read(&current->mm->mmap_sem);
 256
 257                 if (retval <= 0)
 258                         break;
 259                 if (copy_to_user(vec, tmp, retval)) {
 260                         retval = -EFAULT;
 261                         break;
 262                 }
 263                 pages -= retval;
 264                 vec += retval;
 265                 start += retval << PAGE_SHIFT;
 266                 retval = 0;
 267         }
 268         free_page((unsigned long) tmp);
 269         return retval;
 270 }