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 #ifdef CONFIG_SWAP
  83                 if (radix_tree_exceptional_entry(page)) {
  84                         swp_entry_t swap = radix_to_swp_entry(page);
  85                         page = find_get_page(&swapper_space, swap.val);
  86                 }
  87 #endif
  88                 present = PageUptodate(page);
  89                 page_cache_release(page);
  90         }
  91
  92         return present;
  93 }
  94
  95 static void mincore_unmapped_range(struct vm_area_struct *vma,
  96                                 unsigned long addr, unsigned long end,
  97                                 unsigned char *vec)
  98 {
  99         unsigned long nr = (end - addr) >> PAGE_SHIFT;
 100         int i;
 101
 102         if (vma->vm_file) {
 103                 pgoff_t pgoff;
 104
 105                 pgoff = linear_page_index(vma, addr);
 106                 for (i = 0; i < nr; i++, pgoff++)
 107                         vec[i] = mincore_page(vma->vm_file->f_mapping, pgoff);
 108         } else {
 109                 for (i = 0; i < nr; i++)
 110                         vec[i] = 0;
 111         }
 112 }
 113
 114 static void mincore_pte_range(struct vm_area_struct *vma, pmd_t *pmd,
 115                         unsigned long addr, unsigned long end,
 116                         unsigned char *vec)
 117 {
 118         unsigned long next;
 119         spinlock_t *ptl;
 120         pte_t *ptep;
 121
 122         ptep = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl);
 123         do {
 124                 pte_t pte = *ptep;
 125                 pgoff_t pgoff;
 126
 127                 next = addr + PAGE_SIZE;
 128                 if (pte_none(pte))
 129                         mincore_unmapped_range(vma, addr, next, vec);
 130                 else if (pte_present(pte))
 131                         *vec = 1;
 132                 else if (pte_file(pte)) {
 133                         pgoff = pte_to_pgoff(pte);
 134                         *vec = mincore_page(vma->vm_file->f_mapping, pgoff);
 135                 } else { /* pte is a swap entry */
 136                         swp_entry_t entry = pte_to_swp_entry(pte);
 137
 138                         if (is_migration_entry(entry)) {
 139                                 /* migration entries are always uptodate */
 140                                 *vec = 1;
 141                         } else {
 142 #ifdef CONFIG_SWAP
 143                                 pgoff = entry.val;
 144                                 *vec = mincore_page(&swapper_space, pgoff);
 145 #else
 146                                 WARN_ON(1);
 147                                 *vec = 1;
 148 #endif
 149                         }
 150                 }
 151                 vec++;
 152         } while (ptep++, addr = next, addr != end);
 153         pte_unmap_unlock(ptep - 1, ptl);
 154 }
 155
 156 static void mincore_pmd_range(struct vm_area_struct *vma, pud_t *pud,
 157                         unsigned long addr, unsigned long end,
 158                         unsigned char *vec)
 159 {
 160         unsigned long next;
 161         pmd_t *pmd;
 162
 163         pmd = pmd_offset(pud, addr);
 164         do {
 165                 next = pmd_addr_end(addr, end);
 166                 if (pmd_trans_huge(*pmd)) {
 167                         if (mincore_huge_pmd(vma, pmd, addr, next, vec)) {
 168                                 vec += (next - addr) >> PAGE_SHIFT;
 169                                 continue;
 170                         }
 171                         /* fall through */
 172                 }
 173                 if (pmd_none_or_clear_bad(pmd))
 174                         mincore_unmapped_range(vma, addr, next, vec);
 175                 else
 176                         mincore_pte_range(vma, pmd, addr, next, vec);
 177                 vec += (next - addr) >> PAGE_SHIFT;
 178         } while (pmd++, addr = next, addr != end);
 179 }
 180
 181 static void mincore_pud_range(struct vm_area_struct *vma, pgd_t *pgd,
 182                         unsigned long addr, unsigned long end,
 183                         unsigned char *vec)
 184 {
 185         unsigned long next;
 186         pud_t *pud;
 187
 188         pud = pud_offset(pgd, addr);
 189         do {
 190                 next = pud_addr_end(addr, end);
 191                 if (pud_none_or_clear_bad(pud))
 192                         mincore_unmapped_range(vma, addr, next, vec);
 193                 else
 194                         mincore_pmd_range(vma, pud, addr, next, vec);
 195                 vec += (next - addr) >> PAGE_SHIFT;
 196         } while (pud++, addr = next, addr != end);
 197 }
 198
 199 static void mincore_page_range(struct vm_area_struct *vma,
 200                         unsigned long addr, unsigned long end,
 201                         unsigned char *vec)
 202 {
 203         unsigned long next;
 204         pgd_t *pgd;
 205
 206         pgd = pgd_offset(vma->vm_mm, addr);
 207         do {
 208                 next = pgd_addr_end(addr, end);
 209                 if (pgd_none_or_clear_bad(pgd))
 210                         mincore_unmapped_range(vma, addr, next, vec);
 211                 else
 212                         mincore_pud_range(vma, pgd, addr, next, vec);
 213                 vec += (next - addr) >> PAGE_SHIFT;
 214         } while (pgd++, addr = next, addr != end);
 215 }
 216
 217 /*
 218  * Do a chunk of "sys_mincore()". We've already checked
 219  * all the arguments, we hold the mmap semaphore: we should
 220  * just return the amount of info we're asked for.
 221  */
 222 static long do_mincore(unsigned long addr, unsigned long pages, unsigned char *vec)
 223 {
 224         struct vm_area_struct *vma;
 225         unsigned long end;
 226
 227         vma = find_vma(current->mm, addr);
 228         if (!vma || addr < vma->vm_start)
 229                 return -ENOMEM;
 230
 231         end = min(vma->vm_end, addr + (pages << PAGE_SHIFT));
 232
 233         if (is_vm_hugetlb_page(vma)) {
 234                 mincore_hugetlb_page_range(vma, addr, end, vec);
 235                 return (end - addr) >> PAGE_SHIFT;
 236         }
 237
 238         end = pmd_addr_end(addr, end);
 239
 240         if (is_vm_hugetlb_page(vma))
 241                 mincore_hugetlb_page_range(vma, addr, end, vec);
 242         else
 243                 mincore_page_range(vma, addr, end, vec);
 244
 245         return (end - addr) >> PAGE_SHIFT;
 246 }
 247
 248 /*
 249  * The mincore(2) system call.
 250  *
 251  * mincore() returns the memory residency status of the pages in the
 252  * current process's address space specified by [addr, addr + len).
 253  * The status is returned in a vector of bytes.  The least significant
 254  * bit of each byte is 1 if the referenced page is in memory, otherwise
 255  * it is zero.
 256  *
 257  * Because the status of a page can change after mincore() checks it
 258  * but before it returns to the application, the returned vector may
 259  * contain stale information.  Only locked pages are guaranteed to
 260  * remain in memory.
 261  *
 262  * return values:
 263  *  zero    - success
 264  *  -EFAULT - vec points to an illegal address
 265  *  -EINVAL - addr is not a multiple of PAGE_CACHE_SIZE
 266  *  -ENOMEM - Addresses in the range [addr, addr + len] are
 267  *              invalid for the address space of this process, or
 268  *              specify one or more pages which are not currently
 269  *              mapped
 270  *  -EAGAIN - A kernel resource was temporarily unavailable.
 271  */
 272 SYSCALL_DEFINE3(mincore, unsigned long, start, size_t, len,
 273                 unsigned char __user *, vec)
 274 {
 275         long retval;
 276         unsigned long pages;
 277         unsigned char *tmp;
 278
 279         /* Check the start address: needs to be page-aligned.. */
 280         if (start & ~PAGE_CACHE_MASK)
 281                 return -EINVAL;
 282
 283         /* ..and we need to be passed a valid user-space range */
 284         if (!access_ok(VERIFY_READ, (void __user *) start, len))
 285                 return -ENOMEM;
 286
 287         /* This also avoids any overflows on PAGE_CACHE_ALIGN */
 288         pages = len >> PAGE_SHIFT;
 289         pages += (len & ~PAGE_MASK) != 0;
 290
 291         if (!access_ok(VERIFY_WRITE, vec, pages))
 292                 return -EFAULT;
 293
 294         tmp = (void *) __get_free_page(GFP_USER);
 295         if (!tmp)
 296                 return -EAGAIN;
 297
 298         retval = 0;
 299         while (pages) {
 300                 /*
 301                  * Do at most PAGE_SIZE entries per iteration, due to
 302                  * the temporary buffer size.
 303                  */
 304                 down_read(&current->mm->mmap_sem);
 305                 retval = do_mincore(start, min(pages, PAGE_SIZE), tmp);
 306                 up_read(&current->mm->mmap_sem);
 307
 308                 if (retval <= 0)
 309                         break;
 310                 if (copy_to_user(vec, tmp, retval)) {
 311                         retval = -EFAULT;
 312                         break;
 313                 }
 314                 pages -= retval;
 315                 vec += retval;
 316                 start += retval << PAGE_SHIFT;
 317                 retval = 0;
 318         }
 319         free_page((unsigned long) tmp);
 320         return retval;
 321 }