LXR linux/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
  22static 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 */
  62static 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                page_cache_release(page);
  84        }
  85
  86        return present;
  87}
  88
  89static void mincore_unmapped_range(struct vm_area_struct *vma,
  90                                unsigned long addr, unsigned long end,
  91                                unsigned char *vec)
  92{
  93        unsigned long nr = (end - addr) >> PAGE_SHIFT;
  94        int i;
  95
  96        if (vma->vm_file) {
  97                pgoff_t pgoff;
  98
  99                pgoff = linear_page_index(vma, addr);
 100                for (i = 0; i < nr; i++, pgoff++)
 101                        vec[i] = mincore_page(vma->vm_file->f_mapping, pgoff);
 102        } else {
 103                for (i = 0; i < nr; i++)
 104                        vec[i] = 0;
 105        }
 106}
 107
 108static void mincore_pte_range(struct vm_area_struct *vma, pmd_t *pmd,
 109                        unsigned long addr, unsigned long end,
 110                        unsigned char *vec)
 111{
 112        unsigned long next;
 113        spinlock_t *ptl;
 114        pte_t *ptep;
 115
 116        ptep = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl);
 117        do {
 118                pte_t pte = *ptep;
 119                pgoff_t pgoff;
 120
 121                next = addr + PAGE_SIZE;
 122                if (pte_none(pte))
 123                        mincore_unmapped_range(vma, addr, next, vec);
 124                else if (pte_present(pte))
 125                        *vec = 1;
 126                else if (pte_file(pte)) {
 127                        pgoff = pte_to_pgoff(pte);
 128                        *vec = mincore_page(vma->vm_file->f_mapping, pgoff);
 129                } else { /* pte is a swap entry */
 130                        swp_entry_t entry = pte_to_swp_entry(pte);
 131
 132                        if (is_migration_entry(entry)) {
 133                                /* migration entries are always uptodate */
 134                                *vec = 1;
 135                        } else {
 136#ifdef CONFIG_SWAP
 137                                pgoff = entry.val;
 138                                *vec = mincore_page(&swapper_space, pgoff);
 139#else
 140                                WARN_ON(1);
 141                                *vec = 1;
 142#endif
 143                        }
 144                }
 145                vec++;
 146        } while (ptep++, addr = next, addr != end);
 147        pte_unmap_unlock(ptep - 1, ptl);
 148}
 149
 150static void mincore_pmd_range(struct vm_area_struct *vma, pud_t *pud,
 151                        unsigned long addr, unsigned long end,
 152                        unsigned char *vec)
 153{
 154        unsigned long next;
 155        pmd_t *pmd;
 156
 157        pmd = pmd_offset(pud, addr);
 158        do {
 159                next = pmd_addr_end(addr, end);
 160                if (pmd_trans_huge(*pmd)) {
 161                        if (mincore_huge_pmd(vma, pmd, addr, next, vec)) {
 162                                vec += (next - addr) >> PAGE_SHIFT;
 163                                continue;
 164                        }
 165                        /* fall through */
 166                }
 167                if (pmd_none_or_trans_huge_or_clear_bad(pmd))
 168                        mincore_unmapped_range(vma, addr, next, vec);
 169                else
 170                        mincore_pte_range(vma, pmd, addr, next, vec);
 171                vec += (next - addr) >> PAGE_SHIFT;
 172        } while (pmd++, addr = next, addr != end);
 173}
 174
 175static void mincore_pud_range(struct vm_area_struct *vma, pgd_t *pgd,
 176                        unsigned long addr, unsigned long end,
 177                        unsigned char *vec)
 178{
 179        unsigned long next;
 180        pud_t *pud;
 181
 182        pud = pud_offset(pgd, addr);
 183        do {
 184                next = pud_addr_end(addr, end);
 185                if (pud_none_or_clear_bad(pud))
 186                        mincore_unmapped_range(vma, addr, next, vec);
 187                else
 188                        mincore_pmd_range(vma, pud, addr, next, vec);
 189                vec += (next - addr) >> PAGE_SHIFT;
 190        } while (pud++, addr = next, addr != end);
 191}
 192
 193static void mincore_page_range(struct vm_area_struct *vma,
 194                        unsigned long addr, unsigned long end,
 195                        unsigned char *vec)
 196{
 197        unsigned long next;
 198        pgd_t *pgd;
 199
 200        pgd = pgd_offset(vma->vm_mm, addr);
 201        do {
 202                next = pgd_addr_end(addr, end);
 203                if (pgd_none_or_clear_bad(pgd))
 204                        mincore_unmapped_range(vma, addr, next, vec);
 205                else
 206                        mincore_pud_range(vma, pgd, addr, next, vec);
 207                vec += (next - addr) >> PAGE_SHIFT;
 208        } while (pgd++, addr = next, addr != end);
 209}
 210
 211/*
 212 * Do a chunk of "sys_mincore()". We've already checked
 213 * all the arguments, we hold the mmap semaphore: we should
 214 * just return the amount of info we're asked for.
 215 */
 216static long do_mincore(unsigned long addr, unsigned long pages, unsigned char *vec)
 217{
 218        struct vm_area_struct *vma;
 219        unsigned long end;
 220
 221        vma = find_vma(current->mm, addr);
 222        if (!vma || addr < vma->vm_start)
 223                return -ENOMEM;
 224
 225        end = min(vma->vm_end, addr + (pages << PAGE_SHIFT));
 226
 227        if (is_vm_hugetlb_page(vma)) {
 228                mincore_hugetlb_page_range(vma, addr, end, vec);
 229                return (end - addr) >> PAGE_SHIFT;
 230        }
 231
 232        end = pmd_addr_end(addr, end);
 233
 234        if (is_vm_hugetlb_page(vma))
 235                mincore_hugetlb_page_range(vma, addr, end, vec);
 236        else
 237                mincore_page_range(vma, addr, end, vec);
 238
 239        return (end - addr) >> PAGE_SHIFT;
 240}
 241
 242/*
 243 * The mincore(2) system call.
 244 *
 245 * mincore() returns the memory residency status of the pages in the
 246 * current process's address space specified by [addr, addr + len).
 247 * The status is returned in a vector of bytes.  The least significant
 248 * bit of each byte is 1 if the referenced page is in memory, otherwise
 249 * it is zero.
 250 *
 251 * Because the status of a page can change after mincore() checks it
 252 * but before it returns to the application, the returned vector may
 253 * contain stale information.  Only locked pages are guaranteed to
 254 * remain in memory.
 255 *
 256 * return values:
 257 *  zero    - success
 258 *  -EFAULT - vec points to an illegal address
 259 *  -EINVAL - addr is not a multiple of PAGE_CACHE_SIZE
 260 *  -ENOMEM - Addresses in the range [addr, addr + len] are
 261 *              invalid for the address space of this process, or
 262 *              specify one or more pages which are not currently
 263 *              mapped
 264 *  -EAGAIN - A kernel resource was temporarily unavailable.
 265 */
 266SYSCALL_DEFINE3(mincore, unsigned long, start, size_t, len,
 267                unsigned char __user *, vec)
 268{
 269        long retval;
 270        unsigned long pages;
 271        unsigned char *tmp;
 272
 273        /* Check the start address: needs to be page-aligned.. */
 274        if (start & ~PAGE_CACHE_MASK)
 275                return -EINVAL;
 276
 277        /* ..and we need to be passed a valid user-space range */
 278        if (!access_ok(VERIFY_READ, (void __user *) start, len))
 279                return -ENOMEM;
 280
 281        /* This also avoids any overflows on PAGE_CACHE_ALIGN */
 282        pages = len >> PAGE_SHIFT;
 283        pages += (len & ~PAGE_MASK) != 0;
 284
 285        if (!access_ok(VERIFY_WRITE, vec, pages))
 286                return -EFAULT;
 287
 288        tmp = (void *) __get_free_page(GFP_USER);
 289        if (!tmp)
 290                return -EAGAIN;
 291
 292        retval = 0;
 293        while (pages) {
 294                /*
 295                 * Do at most PAGE_SIZE entries per iteration, due to
 296                 * the temporary buffer size.
 297                 */
 298                down_read(&current->mm->mmap_sem);
 299                retval = do_mincore(start, min(pages, PAGE_SIZE), tmp);
 300                up_read(&current->mm->mmap_sem);
 301
 302                if (retval <= 0)
 303                        break;
 304                if (copy_to_user(vec, tmp, retval)) {
 305                        retval = -EFAULT;
 306                        break;
 307                }
 308                pages -= retval;
 309                vec += retval;
 310                start += retval << PAGE_SHIFT;
 311                retval = 0;
 312        }
 313        free_page((unsigned long) tmp);
 314        return retval;
 315}
 316