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 void mincore_hugetlb_page_range(struct vm_area_struct *vma,
  23                                 unsigned long addr, unsigned long end,
  24                                 unsigned char *vec)
  25 {
  26 #ifdef CONFIG_HUGETLB_PAGE
  27         struct hstate *h;
  28
  29         h = hstate_vma(vma);
  30         while (1) {
  31                 unsigned char present;
  32                 pte_t *ptep;
  33                 /*
  34                  * Huge pages are always in RAM for now, but
  35                  * theoretically it needs to be checked.
  36                  */
  37                 ptep = huge_pte_offset(current->mm,
  38                                        addr & huge_page_mask(h));
  39                 present = ptep && !huge_pte_none(huge_ptep_get(ptep));
  40                 while (1) {
  41                         *vec = present;
  42                         vec++;
  43                         addr += PAGE_SIZE;
  44                         if (addr == end)
  45                                 return;
  46                         /* check hugepage border */
  47                         if (!(addr & ~huge_page_mask(h)))
  48                                 break;
  49                 }
  50         }
  51 #else
  52         BUG();
  53 #endif
  54 }
  55
  56 /*
  57  * Later we can get more picky about what "in core" means precisely.
  58  * For now, simply check to see if the page is in the page cache,
  59  * and is up to date; i.e. that no page-in operation would be required
  60  * at this time if an application were to map and access this page.
  61  */
  62 static unsigned char mincore_page(struct address_space *mapping, pgoff_t pgoff)
  63 {
  64         unsigned char present = 0;
  65         struct page *page;
  66
  67         /*
  68          * When tmpfs swaps out a page from a file, any process mapping that
  69          * file will not get a swp_entry_t in its pte, but rather it is like
  70          * any other file mapping (ie. marked !present and faulted in with
  71          * tmpfs's .fault). So swapped out tmpfs mappings are tested here.
  72          */
  73         page = find_get_page(mapping, pgoff);
  74 #ifdef CONFIG_SWAP
  75         /* shmem/tmpfs may return swap: account for swapcache page too. */
  76         if (radix_tree_exceptional_entry(page)) {
  77                 swp_entry_t swap = radix_to_swp_entry(page);
  78                 page = find_get_page(&swapper_space, swap.val);
  79         }
  80 #endif
  81         if (page) {
  82                 present = PageUptodate(page);
  83                 if (present)
  84                         present |= (PageReadaheadUnused(page) << 7);
  85                 page_cache_release(page);
  86         }
  87
  88         return present;
  89 }
  90
  91 static void mincore_unmapped_range(struct vm_area_struct *vma,
  92                                 unsigned long addr, unsigned long end,
  93                                 unsigned char *vec)
  94 {
  95         unsigned long nr = (end - addr) >> PAGE_SHIFT;
  96         int i;
  97
  98         if (vma->vm_file) {
  99                 pgoff_t pgoff;
 100
 101                 pgoff = linear_page_index(vma, addr);
 102                 for (i = 0; i < nr; i++, pgoff++)
 103                         vec[i] = mincore_page(vma->vm_file->f_mapping, pgoff);
 104         } else {
 105                 for (i = 0; i < nr; i++)
 106                         vec[i] = 0;
 107         }
 108 }
 109
 110 static void mincore_pte_range(struct vm_area_struct *vma, pmd_t *pmd,
 111                         unsigned long addr, unsigned long end,
 112                         unsigned char *vec)
 113 {
 114         unsigned long next;
 115         spinlock_t *ptl;
 116         pte_t *ptep;
 117
 118         ptep = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl);
 119         do {
 120                 pte_t pte = *ptep;
 121                 pgoff_t pgoff;
 122
 123                 next = addr + PAGE_SIZE;
 124                 if (pte_none(pte))
 125                         mincore_unmapped_range(vma, addr, next, vec);
 126                 else if (pte_present(pte))
 127                         *vec = 1;
 128                 else if (pte_file(pte)) {
 129                         pgoff = pte_to_pgoff(pte);
 130                         *vec = mincore_page(vma->vm_file->f_mapping, pgoff);
 131                 } else { /* pte is a swap entry */
 132                         swp_entry_t entry = pte_to_swp_entry(pte);
 133
 134                         if (is_migration_entry(entry)) {
 135                                 /* migration entries are always uptodate */
 136                                 *vec = 1;
 137                         } else {
 138 #ifdef CONFIG_SWAP
 139                                 pgoff = entry.val;
 140                                 *vec = mincore_page(&swapper_space, pgoff);
 141 #else
 142                                 WARN_ON(1);
 143                                 *vec = 1;
 144 #endif
 145                         }
 146                 }
 147                 vec++;
 148         } while (ptep++, addr = next, addr != end);
 149         pte_unmap_unlock(ptep - 1, ptl);
 150 }
 151
 152 static void mincore_pmd_range(struct vm_area_struct *vma, pud_t *pud,
 153                         unsigned long addr, unsigned long end,
 154                         unsigned char *vec)
 155 {
 156         unsigned long next;
 157         pmd_t *pmd;
 158
 159         pmd = pmd_offset(pud, addr);
 160         do {
 161                 next = pmd_addr_end(addr, end);
 162                 if (pmd_trans_huge(*pmd)) {
 163                         if (mincore_huge_pmd(vma, pmd, addr, next, vec)) {
 164                                 vec += (next - addr) >> PAGE_SHIFT;
 165                                 continue;
 166                         }
 167                         /* fall through */
 168                 }
 169                 if (pmd_none_or_trans_huge_or_clear_bad(pmd))
 170                         mincore_unmapped_range(vma, addr, next, vec);
 171                 else
 172                         mincore_pte_range(vma, pmd, addr, next, vec);
 173                 vec += (next - addr) >> PAGE_SHIFT;
 174         } while (pmd++, addr = next, addr != end);
 175 }
 176
 177 static void mincore_pud_range(struct vm_area_struct *vma, pgd_t *pgd,
 178                         unsigned long addr, unsigned long end,
 179                         unsigned char *vec)
 180 {
 181         unsigned long next;
 182         pud_t *pud;
 183
 184         pud = pud_offset(pgd, addr);
 185         do {
 186                 next = pud_addr_end(addr, end);
 187                 if (pud_none_or_clear_bad(pud))
 188                         mincore_unmapped_range(vma, addr, next, vec);
 189                 else
 190                         mincore_pmd_range(vma, pud, addr, next, vec);
 191                 vec += (next - addr) >> PAGE_SHIFT;
 192         } while (pud++, addr = next, addr != end);
 193 }
 194
 195 static void mincore_page_range(struct vm_area_struct *vma,
 196                         unsigned long addr, unsigned long end,
 197                         unsigned char *vec)
 198 {
 199         unsigned long next;
 200         pgd_t *pgd;
 201
 202         pgd = pgd_offset(vma->vm_mm, addr);
 203         do {
 204                 next = pgd_addr_end(addr, end);
 205                 if (pgd_none_or_clear_bad(pgd))
 206                         mincore_unmapped_range(vma, addr, next, vec);
 207                 else
 208                         mincore_pud_range(vma, pgd, addr, next, vec);
 209                 vec += (next - addr) >> PAGE_SHIFT;
 210         } while (pgd++, addr = next, addr != end);
 211 }
 212
 213 /*
 214  * Do a chunk of "sys_mincore()". We've already checked
 215  * all the arguments, we hold the mmap semaphore: we should
 216  * just return the amount of info we're asked for.
 217  */
 218 static long do_mincore(unsigned long addr, unsigned long pages, unsigned char *vec)
 219 {
 220         struct vm_area_struct *vma;
 221         unsigned long end;
 222
 223         vma = find_vma(current->mm, addr);
 224         if (!vma || addr < vma->vm_start)
 225                 return -ENOMEM;
 226
 227         end = min(vma->vm_end, addr + (pages << PAGE_SHIFT));
 228
 229         if (is_vm_hugetlb_page(vma)) {
 230                 mincore_hugetlb_page_range(vma, addr, end, vec);
 231                 return (end - addr) >> PAGE_SHIFT;
 232         }
 233
 234         end = pmd_addr_end(addr, end);
 235
 236         if (is_vm_hugetlb_page(vma))
 237                 mincore_hugetlb_page_range(vma, addr, end, vec);
 238         else
 239                 mincore_page_range(vma, addr, end, vec);
 240
 241         return (end - addr) >> PAGE_SHIFT;
 242 }
 243
 244 /*
 245  * The mincore(2) system call.
 246  *
 247  * mincore() returns the memory residency status of the pages in the
 248  * current process's address space specified by [addr, addr + len).
 249  * The status is returned in a vector of bytes.  The least significant
 250  * bit of each byte is 1 if the referenced page is in memory, otherwise
 251  * it is zero.
 252  *
 253  * Because the status of a page can change after mincore() checks it
 254  * but before it returns to the application, the returned vector may
 255  * contain stale information.  Only locked pages are guaranteed to
 256  * remain in memory.
 257  *
 258  * return values:
 259  *  zero    - success
 260  *  -EFAULT - vec points to an illegal address
 261  *  -EINVAL - addr is not a multiple of PAGE_CACHE_SIZE
 262  *  -ENOMEM - Addresses in the range [addr, addr + len] are
 263  *              invalid for the address space of this process, or
 264  *              specify one or more pages which are not currently
 265  *              mapped
 266  *  -EAGAIN - A kernel resource was temporarily unavailable.
 267  */
 268 SYSCALL_DEFINE3(mincore, unsigned long, start, size_t, len,
 269                 unsigned char __user *, vec)
 270 {
 271         long retval;
 272         unsigned long pages;
 273         unsigned char *tmp;
 274
 275         /* Check the start address: needs to be page-aligned.. */
 276         if (start & ~PAGE_CACHE_MASK)
 277                 return -EINVAL;
 278
 279         /* ..and we need to be passed a valid user-space range */
 280         if (!access_ok(VERIFY_READ, (void __user *) start, len))
 281                 return -ENOMEM;
 282
 283         /* This also avoids any overflows on PAGE_CACHE_ALIGN */
 284         pages = len >> PAGE_SHIFT;
 285         pages += (len & ~PAGE_MASK) != 0;
 286
 287         if (!access_ok(VERIFY_WRITE, vec, pages))
 288                 return -EFAULT;
 289
 290         tmp = (void *) __get_free_page(GFP_USER);
 291         if (!tmp)
 292                 return -EAGAIN;
 293
 294         retval = 0;
 295         while (pages) {
 296                 /*
 297                  * Do at most PAGE_SIZE entries per iteration, due to
 298                  * the temporary buffer size.
 299                  */
 300                 down_read(&current->mm->mmap_sem);
 301                 retval = do_mincore(start, min(pages, PAGE_SIZE), tmp);
 302                 up_read(&current->mm->mmap_sem);
 303
 304                 if (retval <= 0)
 305                         break;
 306                 if (copy_to_user(vec, tmp, retval)) {
 307                         retval = -EFAULT;
 308                         break;
 309                 }
 310                 pages -= retval;
 311                 vec += retval;
 312                 start += retval << PAGE_SHIFT;
 313                 retval = 0;
 314         }
 315         free_page((unsigned long) tmp);
 316         return retval;
 317 }