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/shmem_fs.h>
  18 #include <linux/hugetlb.h>
  19
  20 #include <linux/uaccess.h>
  21 #include <asm/pgtable.h>
  22
  23 static int mincore_hugetlb(pte_t *pte, unsigned long hmask, unsigned long addr,
  24                         unsigned long end, struct mm_walk *walk)
  25 {
  26 #ifdef CONFIG_HUGETLB_PAGE
  27         unsigned char present;
  28         unsigned char *vec = walk->private;
  29
  30         /*
  31          * Hugepages under user process are always in RAM and never
  32          * swapped out, but theoretically it needs to be checked.
  33          */
  34         present = pte && !huge_pte_none(huge_ptep_get(pte));
  35         for (; addr != end; vec++, addr += PAGE_SIZE)
  36                 *vec = present;
  37         walk->private = vec;
  38 #else
  39         BUG();
  40 #endif
  41         return 0;
  42 }
  43
  44 /*
  45  * Later we can get more picky about what "in core" means precisely.
  46  * For now, simply check to see if the page is in the page cache,
  47  * and is up to date; i.e. that no page-in operation would be required
  48  * at this time if an application were to map and access this page.
  49  */
  50 static unsigned char mincore_page(struct address_space *mapping, pgoff_t pgoff)
  51 {
  52         unsigned char present = 0;
  53         struct page *page;
  54
  55         /*
  56          * When tmpfs swaps out a page from a file, any process mapping that
  57          * file will not get a swp_entry_t in its pte, but rather it is like
  58          * any other file mapping (ie. marked !present and faulted in with
  59          * tmpfs's .fault). So swapped out tmpfs mappings are tested here.
  60          */
  61 #ifdef CONFIG_SWAP
  62         if (shmem_mapping(mapping)) {
  63                 page = find_get_entry(mapping, pgoff);
  64                 /*
  65                  * shmem/tmpfs may return swap: account for swapcache
  66                  * page too.
  67                  */
  68                 if (radix_tree_exceptional_entry(page)) {
  69                         swp_entry_t swp = radix_to_swp_entry(page);
  70                         page = find_get_page(swap_address_space(swp),
  71                                              swp_offset(swp));
  72                 }
  73         } else
  74                 page = find_get_page(mapping, pgoff);
  75 #else
  76         page = find_get_page(mapping, pgoff);
  77 #endif
  78         if (page) {
  79                 present = PageUptodate(page);
  80                 put_page(page);
  81         }
  82
  83         return present;
  84 }
  85
  86 static int __mincore_unmapped_range(unsigned long addr, unsigned long end,
  87                                 struct vm_area_struct *vma, unsigned char *vec)
  88 {
  89         unsigned long nr = (end - addr) >> PAGE_SHIFT;
  90         int i;
  91
  92         if (vma->vm_file) {
  93                 pgoff_t pgoff;
  94
  95                 pgoff = linear_page_index(vma, addr);
  96                 for (i = 0; i < nr; i++, pgoff++)
  97                         vec[i] = mincore_page(vma->vm_file->f_mapping, pgoff);
  98         } else {
  99                 for (i = 0; i < nr; i++)
 100                         vec[i] = 0;
 101         }
 102         return nr;
 103 }
 104
 105 static int mincore_unmapped_range(unsigned long addr, unsigned long end,
 106                                    struct mm_walk *walk)
 107 {
 108         walk->private += __mincore_unmapped_range(addr, end,
 109                                                   walk->vma, walk->private);
 110         return 0;
 111 }
 112
 113 static int mincore_pte_range(pmd_t *pmd, unsigned long addr, unsigned long end,
 114                         struct mm_walk *walk)
 115 {
 116         spinlock_t *ptl;
 117         struct vm_area_struct *vma = walk->vma;
 118         pte_t *ptep;
 119         unsigned char *vec = walk->private;
 120         int nr = (end - addr) >> PAGE_SHIFT;
 121
 122         ptl = pmd_trans_huge_lock(pmd, vma);
 123         if (ptl) {
 124                 memset(vec, 1, nr);
 125                 spin_unlock(ptl);
 126                 goto out;
 127         }
 128
 129         if (pmd_trans_unstable(pmd)) {
 130                 __mincore_unmapped_range(addr, end, vma, vec);
 131                 goto out;
 132         }
 133
 134         ptep = pte_offset_map_lock(walk->mm, pmd, addr, &ptl);
 135         for (; addr != end; ptep++, addr += PAGE_SIZE) {
 136                 pte_t pte = *ptep;
 137
 138                 if (pte_none(pte))
 139                         __mincore_unmapped_range(addr, addr + PAGE_SIZE,
 140                                                  vma, vec);
 141                 else if (pte_present(pte))
 142                         *vec = 1;
 143                 else { /* pte is a swap entry */
 144                         swp_entry_t entry = pte_to_swp_entry(pte);
 145
 146                         if (non_swap_entry(entry)) {
 147                                 /*
 148                                  * migration or hwpoison entries are always
 149                                  * uptodate
 150                                  */
 151                                 *vec = 1;
 152                         } else {
 153 #ifdef CONFIG_SWAP
 154                                 *vec = mincore_page(swap_address_space(entry),
 155                                                     swp_offset(entry));
 156 #else
 157                                 WARN_ON(1);
 158                                 *vec = 1;
 159 #endif
 160                         }
 161                 }
 162                 vec++;
 163         }
 164         pte_unmap_unlock(ptep - 1, ptl);
 165 out:
 166         walk->private += nr;
 167         cond_resched();
 168         return 0;
 169 }
 170
 171 /*
 172  * Do a chunk of "sys_mincore()". We've already checked
 173  * all the arguments, we hold the mmap semaphore: we should
 174  * just return the amount of info we're asked for.
 175  */
 176 static long do_mincore(unsigned long addr, unsigned long pages, unsigned char *vec)
 177 {
 178         struct vm_area_struct *vma;
 179         unsigned long end;
 180         int err;
 181         struct mm_walk mincore_walk = {
 182                 .pmd_entry = mincore_pte_range,
 183                 .pte_hole = mincore_unmapped_range,
 184                 .hugetlb_entry = mincore_hugetlb,
 185                 .private = vec,
 186         };
 187
 188         vma = find_vma(current->mm, addr);
 189         if (!vma || addr < vma->vm_start)
 190                 return -ENOMEM;
 191         mincore_walk.mm = vma->vm_mm;
 192         end = min(vma->vm_end, addr + (pages << PAGE_SHIFT));
 193         err = walk_page_range(addr, end, &mincore_walk);
 194         if (err < 0)
 195                 return err;
 196         return (end - addr) >> PAGE_SHIFT;
 197 }
 198
 199 /*
 200  * The mincore(2) system call.
 201  *
 202  * mincore() returns the memory residency status of the pages in the
 203  * current process's address space specified by [addr, addr + len).
 204  * The status is returned in a vector of bytes.  The least significant
 205  * bit of each byte is 1 if the referenced page is in memory, otherwise
 206  * it is zero.
 207  *
 208  * Because the status of a page can change after mincore() checks it
 209  * but before it returns to the application, the returned vector may
 210  * contain stale information.  Only locked pages are guaranteed to
 211  * remain in memory.
 212  *
 213  * return values:
 214  *  zero    - success
 215  *  -EFAULT - vec points to an illegal address
 216  *  -EINVAL - addr is not a multiple of PAGE_SIZE
 217  *  -ENOMEM - Addresses in the range [addr, addr + len] are
 218  *              invalid for the address space of this process, or
 219  *              specify one or more pages which are not currently
 220  *              mapped
 221  *  -EAGAIN - A kernel resource was temporarily unavailable.
 222  */
 223 SYSCALL_DEFINE3(mincore, unsigned long, start, size_t, len,
 224                 unsigned char __user *, vec)
 225 {
 226         long retval;
 227         unsigned long pages;
 228         unsigned char *tmp;
 229
 230         /* Check the start address: needs to be page-aligned.. */
 231         if (start & ~PAGE_MASK)
 232                 return -EINVAL;
 233
 234         /* ..and we need to be passed a valid user-space range */
 235         if (!access_ok(VERIFY_READ, (void __user *) start, len))
 236                 return -ENOMEM;
 237
 238         /* This also avoids any overflows on PAGE_ALIGN */
 239         pages = len >> PAGE_SHIFT;
 240         pages += (offset_in_page(len)) != 0;
 241
 242         if (!access_ok(VERIFY_WRITE, vec, pages))
 243                 return -EFAULT;
 244
 245         tmp = (void *) __get_free_page(GFP_USER);
 246         if (!tmp)
 247                 return -EAGAIN;
 248
 249         retval = 0;
 250         while (pages) {
 251                 /*
 252                  * Do at most PAGE_SIZE entries per iteration, due to
 253                  * the temporary buffer size.
 254                  */
 255                 down_read(&current->mm->mmap_sem);
 256                 retval = do_mincore(start, min(pages, PAGE_SIZE), tmp);
 257                 up_read(&current->mm->mmap_sem);
 258
 259                 if (retval <= 0)
 260                         break;
 261                 if (copy_to_user(vec, tmp, retval)) {
 262                         retval = -EFAULT;
 263                         break;
 264                 }
 265                 pages -= retval;
 266                 vec += retval;
 267                 start += retval << PAGE_SHIFT;
 268                 retval = 0;
 269         }
 270         free_page((unsigned long) tmp);
 271         return retval;
 272 }