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 /*
  23  * Later we can get more picky about what "in core" means precisely.
  24  * For now, simply check to see if the page is in the page cache,
  25  * and is up to date; i.e. that no page-in operation would be required
  26  * at this time if an application were to map and access this page.
  27  */
  28 static unsigned char mincore_page(struct address_space *mapping, pgoff_t pgoff)
  29 {
  30         unsigned char present = 0;
  31         struct page *page;
  32
  33         /*
  34          * When tmpfs swaps out a page from a file, any process mapping that
  35          * file will not get a swp_entry_t in its pte, but rather it is like
  36          * any other file mapping (ie. marked !present and faulted in with
  37          * tmpfs's .fault). So swapped out tmpfs mappings are tested here.
  38          *
  39          * However when tmpfs moves the page from pagecache and into swapcache,
  40          * it is still in core, but the find_get_page below won't find it.
  41          * No big deal, but make a note of it.
  42          */
  43         page = find_get_page(mapping, pgoff);
  44         if (page) {
  45                 present = PageUptodate(page);
  46                 page_cache_release(page);
  47         }
  48
  49         return present;
  50 }
  51
  52 /*
  53  * Do a chunk of "sys_mincore()". We've already checked
  54  * all the arguments, we hold the mmap semaphore: we should
  55  * just return the amount of info we're asked for.
  56  */
  57 static long do_mincore(unsigned long addr, unsigned long pages, unsigned char *vec)
  58 {
  59         pgd_t *pgd;
  60         pud_t *pud;
  61         pmd_t *pmd;
  62         pte_t *ptep;
  63         spinlock_t *ptl;
  64         unsigned long nr;
  65         int i;
  66         pgoff_t pgoff;
  67         struct vm_area_struct *vma;
  68
  69         vma = find_vma(current->mm, addr);
  70         if (!vma || addr < vma->vm_start)
  71                 return -ENOMEM;
  72
  73         nr = min(pages, (vma->vm_end - addr) >> PAGE_SHIFT);
  74
  75 #ifdef CONFIG_HUGETLB_PAGE
  76         if (is_vm_hugetlb_page(vma)) {
  77                 struct hstate *h;
  78
  79                 i = 0;
  80                 h = hstate_vma(vma);
  81                 while (1) {
  82                         unsigned char present;
  83                         /*
  84                          * Huge pages are always in RAM for now, but
  85                          * theoretically it needs to be checked.
  86                          */
  87                         ptep = huge_pte_offset(current->mm,
  88                                                addr & huge_page_mask(h));
  89                         present = ptep && !huge_pte_none(huge_ptep_get(ptep));
  90                         while (1) {
  91                                 vec[i++] = present;
  92                                 addr += PAGE_SIZE;
  93                                 /* reach buffer limit */
  94                                 if (i == nr)
  95                                         return nr;
  96                                 /* check hugepage border */
  97                                 if (!(addr & ~huge_page_mask(h)))
  98                                         break;
  99                         }
 100                 }
 101                 return nr;
 102         }
 103 #endif
 104
 105         /*
 106          * Calculate how many pages there are left in the last level of the
 107          * PTE array for our address.
 108          */
 109         nr = min(nr, PTRS_PER_PTE - ((addr >> PAGE_SHIFT) & (PTRS_PER_PTE-1)));
 110
 111         pgd = pgd_offset(vma->vm_mm, addr);
 112         if (pgd_none_or_clear_bad(pgd))
 113                 goto none_mapped;
 114         pud = pud_offset(pgd, addr);
 115         if (pud_none_or_clear_bad(pud))
 116                 goto none_mapped;
 117         pmd = pmd_offset(pud, addr);
 118         if (pmd_none_or_clear_bad(pmd))
 119                 goto none_mapped;
 120
 121         ptep = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl);
 122         for (i = 0; i < nr; i++, ptep++, addr += PAGE_SIZE) {
 123                 pte_t pte = *ptep;
 124
 125                 if (pte_none(pte)) {
 126                         if (vma->vm_file) {
 127                                 pgoff = linear_page_index(vma, addr);
 128                                 vec[i] = mincore_page(vma->vm_file->f_mapping,
 129                                                 pgoff);
 130                         } else
 131                                 vec[i] = 0;
 132                 } else if (pte_present(pte))
 133                         vec[i] = 1;
 134                 else if (pte_file(pte)) {
 135                         pgoff = pte_to_pgoff(pte);
 136                         vec[i] = mincore_page(vma->vm_file->f_mapping, pgoff);
 137                 } else { /* pte is a swap entry */
 138                         swp_entry_t entry = pte_to_swp_entry(pte);
 139
 140                         if (is_migration_entry(entry)) {
 141                                 /* migration entries are always uptodate */
 142                                 vec[i] = 1;
 143                         } else {
 144 #ifdef CONFIG_SWAP
 145                                 pgoff = entry.val;
 146                                 vec[i] = mincore_page(&swapper_space, pgoff);
 147 #else
 148                                 WARN_ON(1);
 149                                 vec[i] = 1;
 150 #endif
 151                         }
 152                 }
 153         }
 154         pte_unmap_unlock(ptep - 1, ptl);
 155
 156         return nr;
 157
 158 none_mapped:
 159         if (vma->vm_file) {
 160                 pgoff = linear_page_index(vma, addr);
 161                 for (i = 0; i < nr; i++, pgoff++)
 162                         vec[i] = mincore_page(vma->vm_file->f_mapping, pgoff);
 163         } else {
 164                 for (i = 0; i < nr; i++)
 165                         vec[i] = 0;
 166         }
 167
 168         return nr;
 169 }
 170
 171 /*
 172  * The mincore(2) system call.
 173  *
 174  * mincore() returns the memory residency status of the pages in the
 175  * current process's address space specified by [addr, addr + len).
 176  * The status is returned in a vector of bytes.  The least significant
 177  * bit of each byte is 1 if the referenced page is in memory, otherwise
 178  * it is zero.
 179  *
 180  * Because the status of a page can change after mincore() checks it
 181  * but before it returns to the application, the returned vector may
 182  * contain stale information.  Only locked pages are guaranteed to
 183  * remain in memory.
 184  *
 185  * return values:
 186  *  zero    - success
 187  *  -EFAULT - vec points to an illegal address
 188  *  -EINVAL - addr is not a multiple of PAGE_CACHE_SIZE
 189  *  -ENOMEM - Addresses in the range [addr, addr + len] are
 190  *              invalid for the address space of this process, or
 191  *              specify one or more pages which are not currently
 192  *              mapped
 193  *  -EAGAIN - A kernel resource was temporarily unavailable.
 194  */
 195 SYSCALL_DEFINE3(mincore, unsigned long, start, size_t, len,
 196                 unsigned char __user *, vec)
 197 {
 198         long retval;
 199         unsigned long pages;
 200         unsigned char *tmp;
 201
 202         /* Check the start address: needs to be page-aligned.. */
 203         if (start & ~PAGE_CACHE_MASK)
 204                 return -EINVAL;
 205
 206         /* ..and we need to be passed a valid user-space range */
 207         if (!access_ok(VERIFY_READ, (void __user *) start, len))
 208                 return -ENOMEM;
 209
 210         /* This also avoids any overflows on PAGE_CACHE_ALIGN */
 211         pages = len >> PAGE_SHIFT;
 212         pages += (len & ~PAGE_MASK) != 0;
 213
 214         if (!access_ok(VERIFY_WRITE, vec, pages))
 215                 return -EFAULT;
 216
 217         tmp = (void *) __get_free_page(GFP_USER);
 218         if (!tmp)
 219                 return -EAGAIN;
 220
 221         retval = 0;
 222         while (pages) {
 223                 /*
 224                  * Do at most PAGE_SIZE entries per iteration, due to
 225                  * the temporary buffer size.
 226                  */
 227                 down_read(&current->mm->mmap_sem);
 228                 retval = do_mincore(start, min(pages, PAGE_SIZE), tmp);
 229                 up_read(&current->mm->mmap_sem);
 230
 231                 if (retval <= 0)
 232                         break;
 233                 if (copy_to_user(vec, tmp, retval)) {
 234                         retval = -EFAULT;
 235                         break;
 236                 }
 237                 pages -= retval;
 238                 vec += retval;
 239                 start += retval << PAGE_SHIFT;
 240                 retval = 0;
 241         }
 242         free_page((unsigned long) tmp);
 243         return retval;
 244 }