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          * However when tmpfs moves the page from pagecache and into swapcache,
  74          * it is still in core, but the find_get_page below won't find it.
  75          * No big deal, but make a note of it.
  76          */
  77         page = find_get_page(mapping, pgoff);
  78         if (page) {
  79                 present = PageUptodate(page);
  80                 page_cache_release(page);
  81         }
  82
  83         return present;
  84 }
  85
  86 static void mincore_unmapped_range(struct vm_area_struct *vma,
  87                                 unsigned long addr, unsigned long end,
  88                                 unsigned char *vec)
  89 {
  90         unsigned long nr = (end - addr) >> PAGE_SHIFT;
  91         int i;
  92
  93         if (vma->vm_file) {
  94                 pgoff_t pgoff;
  95
  96                 pgoff = linear_page_index(vma, addr);
  97                 for (i = 0; i < nr; i++, pgoff++)
  98                         vec[i] = mincore_page(vma->vm_file->f_mapping, pgoff);
  99         } else {
 100                 for (i = 0; i < nr; i++)
 101                         vec[i] = 0;
 102         }
 103 }
 104
 105 static void mincore_pte_range(struct vm_area_struct *vma, pmd_t *pmd,
 106                         unsigned long addr, unsigned long end,
 107                         unsigned char *vec)
 108 {
 109         unsigned long next;
 110         spinlock_t *ptl;
 111         pte_t *ptep;
 112
 113         ptep = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl);
 114         do {
 115                 pte_t pte = *ptep;
 116                 pgoff_t pgoff;
 117
 118                 next = addr + PAGE_SIZE;
 119                 if (pte_none(pte))
 120                         mincore_unmapped_range(vma, addr, next, vec);
 121                 else if (pte_present(pte))
 122                         *vec = 1;
 123                 else if (pte_file(pte)) {
 124                         pgoff = pte_to_pgoff(pte);
 125                         *vec = mincore_page(vma->vm_file->f_mapping, pgoff);
 126                 } else { /* pte is a swap entry */
 127                         swp_entry_t entry = pte_to_swp_entry(pte);
 128
 129                         if (is_migration_entry(entry)) {
 130                                 /* migration entries are always uptodate */
 131                                 *vec = 1;
 132                         } else {
 133 #ifdef CONFIG_SWAP
 134                                 pgoff = entry.val;
 135                                 *vec = mincore_page(&swapper_space, pgoff);
 136 #else
 137                                 WARN_ON(1);
 138                                 *vec = 1;
 139 #endif
 140                         }
 141                 }
 142                 vec++;
 143         } while (ptep++, addr = next, addr != end);
 144         pte_unmap_unlock(ptep - 1, ptl);
 145 }
 146
 147 static void mincore_pmd_range(struct vm_area_struct *vma, pud_t *pud,
 148                         unsigned long addr, unsigned long end,
 149                         unsigned char *vec)
 150 {
 151         unsigned long next;
 152         pmd_t *pmd;
 153
 154         pmd = pmd_offset(pud, addr);
 155         do {
 156                 next = pmd_addr_end(addr, end);
 157                 if (pmd_none_or_clear_bad(pmd))
 158                         mincore_unmapped_range(vma, addr, next, vec);
 159                 else
 160                         mincore_pte_range(vma, pmd, addr, next, vec);
 161                 vec += (next - addr) >> PAGE_SHIFT;
 162         } while (pmd++, addr = next, addr != end);
 163 }
 164
 165 static void mincore_pud_range(struct vm_area_struct *vma, pgd_t *pgd,
 166                         unsigned long addr, unsigned long end,
 167                         unsigned char *vec)
 168 {
 169         unsigned long next;
 170         pud_t *pud;
 171
 172         pud = pud_offset(pgd, addr);
 173         do {
 174                 next = pud_addr_end(addr, end);
 175                 if (pud_none_or_clear_bad(pud))
 176                         mincore_unmapped_range(vma, addr, next, vec);
 177                 else
 178                         mincore_pmd_range(vma, pud, addr, next, vec);
 179                 vec += (next - addr) >> PAGE_SHIFT;
 180         } while (pud++, addr = next, addr != end);
 181 }
 182
 183 static void mincore_page_range(struct vm_area_struct *vma,
 184                         unsigned long addr, unsigned long end,
 185                         unsigned char *vec)
 186 {
 187         unsigned long next;
 188         pgd_t *pgd;
 189
 190         pgd = pgd_offset(vma->vm_mm, addr);
 191         do {
 192                 next = pgd_addr_end(addr, end);
 193                 if (pgd_none_or_clear_bad(pgd))
 194                         mincore_unmapped_range(vma, addr, next, vec);
 195                 else
 196                         mincore_pud_range(vma, pgd, addr, next, vec);
 197                 vec += (next - addr) >> PAGE_SHIFT;
 198         } while (pgd++, addr = next, addr != end);
 199 }
 200
 201 /*
 202  * Do a chunk of "sys_mincore()". We've already checked
 203  * all the arguments, we hold the mmap semaphore: we should
 204  * just return the amount of info we're asked for.
 205  */
 206 static long do_mincore(unsigned long addr, unsigned long pages, unsigned char *vec)
 207 {
 208         struct vm_area_struct *vma;
 209         unsigned long end;
 210
 211         vma = find_vma(current->mm, addr);
 212         if (!vma || addr < vma->vm_start)
 213                 return -ENOMEM;
 214
 215         end = min(vma->vm_end, addr + (pages << PAGE_SHIFT));
 216
 217         if (is_vm_hugetlb_page(vma)) {
 218                 mincore_hugetlb_page_range(vma, addr, end, vec);
 219                 return (end - addr) >> PAGE_SHIFT;
 220         }
 221
 222         end = pmd_addr_end(addr, end);
 223
 224         if (is_vm_hugetlb_page(vma))
 225                 mincore_hugetlb_page_range(vma, addr, end, vec);
 226         else
 227                 mincore_page_range(vma, addr, end, vec);
 228
 229         return (end - addr) >> PAGE_SHIFT;
 230 }
 231
 232 /*
 233  * The mincore(2) system call.
 234  *
 235  * mincore() returns the memory residency status of the pages in the
 236  * current process's address space specified by [addr, addr + len).
 237  * The status is returned in a vector of bytes.  The least significant
 238  * bit of each byte is 1 if the referenced page is in memory, otherwise
 239  * it is zero.
 240  *
 241  * Because the status of a page can change after mincore() checks it
 242  * but before it returns to the application, the returned vector may
 243  * contain stale information.  Only locked pages are guaranteed to
 244  * remain in memory.
 245  *
 246  * return values:
 247  *  zero    - success
 248  *  -EFAULT - vec points to an illegal address
 249  *  -EINVAL - addr is not a multiple of PAGE_CACHE_SIZE
 250  *  -ENOMEM - Addresses in the range [addr, addr + len] are
 251  *              invalid for the address space of this process, or
 252  *              specify one or more pages which are not currently
 253  *              mapped
 254  *  -EAGAIN - A kernel resource was temporarily unavailable.
 255  */
 256 SYSCALL_DEFINE3(mincore, unsigned long, start, size_t, len,
 257                 unsigned char __user *, vec)
 258 {
 259         long retval;
 260         unsigned long pages;
 261         unsigned char *tmp;
 262
 263         /* Check the start address: needs to be page-aligned.. */
 264         if (start & ~PAGE_CACHE_MASK)
 265                 return -EINVAL;
 266
 267         /* ..and we need to be passed a valid user-space range */
 268         if (!access_ok(VERIFY_READ, (void __user *) start, len))
 269                 return -ENOMEM;
 270
 271         /* This also avoids any overflows on PAGE_CACHE_ALIGN */
 272         pages = len >> PAGE_SHIFT;
 273         pages += (len & ~PAGE_MASK) != 0;
 274
 275         if (!access_ok(VERIFY_WRITE, vec, pages))
 276                 return -EFAULT;
 277
 278         tmp = (void *) __get_free_page(GFP_USER);
 279         if (!tmp)
 280                 return -EAGAIN;
 281
 282         retval = 0;
 283         while (pages) {
 284                 /*
 285                  * Do at most PAGE_SIZE entries per iteration, due to
 286                  * the temporary buffer size.
 287                  */
 288                 down_read(&current->mm->mmap_sem);
 289                 retval = do_mincore(start, min(pages, PAGE_SIZE), tmp);
 290                 up_read(&current->mm->mmap_sem);
 291
 292                 if (retval <= 0)
 293                         break;
 294                 if (copy_to_user(vec, tmp, retval)) {
 295                         retval = -EFAULT;
 296                         break;
 297                 }
 298                 pages -= retval;
 299                 vec += retval;
 300                 start += retval << PAGE_SHIFT;
 301                 retval = 0;
 302         }
 303         free_page((unsigned long) tmp);
 304         return retval;
 305 }