mm/fremap.c

   1 /*
   2  *   linux/mm/fremap.c
   3  *
   4  * Explicit pagetable population and nonlinear (random) mappings support.
   5  *
   6  * started by Ingo Molnar, Copyright (C) 2002, 2003
   7  */
   8 #include <linux/backing-dev.h>
   9 #include <linux/mm.h>
  10 #include <linux/swap.h>
  11 #include <linux/file.h>
  12 #include <linux/mman.h>
  13 #include <linux/pagemap.h>
  14 #include <linux/swapops.h>
  15 #include <linux/rmap.h>
  16 #include <linux/module.h>
  17 #include <linux/syscalls.h>
  18
  19 #include <asm/mmu_context.h>
  20 #include <asm/cacheflush.h>
  21 #include <asm/tlbflush.h>
  22
  23 static void zap_pte(struct mm_struct *mm, struct vm_area_struct *vma,
  24                         unsigned long addr, pte_t *ptep)
  25 {
  26         pte_t pte = *ptep;
  27
  28         if (pte_present(pte)) {
  29                 struct page *page;
  30
  31                 flush_cache_page(vma, addr, pte_pfn(pte));
  32                 pte = ptep_clear_flush(vma, addr, ptep);
  33                 page = vm_normal_page(vma, addr, pte);
  34                 if (page) {
  35                         if (pte_dirty(pte))
  36                                 set_page_dirty(page);
  37                         page_remove_rmap(page, vma);
  38                         page_cache_release(page);
  39                         update_hiwater_rss(mm);
  40                         dec_mm_counter(mm, file_rss);
  41                 }
  42         } else {
  43                 if (!pte_file(pte))
  44                         free_swap_and_cache(pte_to_swp_entry(pte));
  45                 pte_clear_not_present_full(mm, addr, ptep, 0);
  46         }
  47 }
  48
  49 /*
  50  * Install a file pte to a given virtual memory address, release any
  51  * previously existing mapping.
  52  */
  53 static int install_file_pte(struct mm_struct *mm, struct vm_area_struct *vma,
  54                 unsigned long addr, unsigned long pgoff, pgprot_t prot)
  55 {
  56         int err = -ENOMEM;
  57         pte_t *pte;
  58         spinlock_t *ptl;
  59
  60         pte = get_locked_pte(mm, addr, &ptl);
  61         if (!pte)
  62                 goto out;
  63
  64         if (!pte_none(*pte))
  65                 zap_pte(mm, vma, addr, pte);
  66
  67         set_pte_at(mm, addr, pte, pgoff_to_pte(pgoff));
  68         /*
  69          * We don't need to run update_mmu_cache() here because the "file pte"
  70          * being installed by install_file_pte() is not a real pte - it's a
  71          * non-present entry (like a swap entry), noting what file offset should
  72          * be mapped there when there's a fault (in a non-linear vma where
  73          * that's not obvious).
  74          */
  75         pte_unmap_unlock(pte, ptl);
  76         err = 0;
  77 out:
  78         return err;
  79 }
  80
  81 static int populate_range(struct mm_struct *mm, struct vm_area_struct *vma,
  82                         unsigned long addr, unsigned long size, pgoff_t pgoff)
  83 {
  84         int err;
  85
  86         do {
  87                 err = install_file_pte(mm, vma, addr, pgoff, vma->vm_page_prot);
  88                 if (err)
  89                         return err;
  90
  91                 size -= PAGE_SIZE;
  92                 addr += PAGE_SIZE;
  93                 pgoff++;
  94         } while (size);
  95
  96         return 0;
  97
  98 }
  99
 100 /**
 101  * sys_remap_file_pages - remap arbitrary pages of an existing VM_SHARED vma
 102  * @start: start of the remapped virtual memory range
 103  * @size: size of the remapped virtual memory range
 104  * @prot: new protection bits of the range (see NOTE)
 105  * @pgoff: to-be-mapped page of the backing store file
 106  * @flags: 0 or MAP_NONBLOCKED - the later will cause no IO.
 107  *
 108  * sys_remap_file_pages remaps arbitrary pages of an existing VM_SHARED vma
 109  * (shared backing store file).
 110  *
 111  * This syscall works purely via pagetables, so it's the most efficient
 112  * way to map the same (large) file into a given virtual window. Unlike
 113  * mmap()/mremap() it does not create any new vmas. The new mappings are
 114  * also safe across swapout.
 115  *
 116  * NOTE: the @prot parameter right now is ignored (but must be zero),
 117  * and the vma's default protection is used. Arbitrary protections
 118  * might be implemented in the future.
 119  */
 120 asmlinkage long sys_remap_file_pages(unsigned long start, unsigned long size,
 121         unsigned long prot, unsigned long pgoff, unsigned long flags)
 122 {
 123         struct mm_struct *mm = current->mm;
 124         struct address_space *mapping;
 125         unsigned long end = start + size;
 126         struct vm_area_struct *vma;
 127         int err = -EINVAL;
 128         int has_write_lock = 0;
 129
 130         if (prot)
 131                 return err;
 132         /*
 133          * Sanitize the syscall parameters:
 134          */
 135         start = start & PAGE_MASK;
 136         size = size & PAGE_MASK;
 137
 138         /* Does the address range wrap, or is the span zero-sized? */
 139         if (start + size <= start)
 140                 return err;
 141
 142         /* Can we represent this offset inside this architecture's pte's? */
 143 #if PTE_FILE_MAX_BITS < BITS_PER_LONG
 144         if (pgoff + (size >> PAGE_SHIFT) >= (1UL << PTE_FILE_MAX_BITS))
 145                 return err;
 146 #endif
 147
 148         /* We need down_write() to change vma->vm_flags. */
 149         down_read(&mm->mmap_sem);
 150  retry:
 151         vma = find_vma(mm, start);
 152
 153         /*
 154          * Make sure the vma is shared, that it supports prefaulting,
 155          * and that the remapped range is valid and fully within
 156          * the single existing vma.  vm_private_data is used as a
 157          * swapout cursor in a VM_NONLINEAR vma.
 158          */
 159         if (!vma || !(vma->vm_flags & VM_SHARED))
 160                 goto out;
 161
 162         if (vma->vm_private_data && !(vma->vm_flags & VM_NONLINEAR))
 163                 goto out;
 164
 165         if (!(vma->vm_flags & VM_CAN_NONLINEAR))
 166                 goto out;
 167
 168         if (end <= start || start < vma->vm_start || end > vma->vm_end)
 169                 goto out;
 170
 171         /* Must set VM_NONLINEAR before any pages are populated. */
 172         if (!(vma->vm_flags & VM_NONLINEAR)) {
 173                 /* Don't need a nonlinear mapping, exit success */
 174                 if (pgoff == linear_page_index(vma, start)) {
 175                         err = 0;
 176                         goto out;
 177                 }
 178
 179                 if (!has_write_lock) {
 180                         up_read(&mm->mmap_sem);
 181                         down_write(&mm->mmap_sem);
 182                         has_write_lock = 1;
 183                         goto retry;
 184                 }
 185                 mapping = vma->vm_file->f_mapping;
 186                 /*
 187                  * page_mkclean doesn't work on nonlinear vmas, so if
 188                  * dirty pages need to be accounted, emulate with linear
 189                  * vmas.
 190                  */
 191                 if (mapping_cap_account_dirty(mapping)) {
 192                         unsigned long addr;
 193                         struct file *file = vma->vm_file;
 194
 195                         flags &= MAP_NONBLOCK;
 196                         get_file(file);
 197                         addr = mmap_region(file, start, size,
 198                                         flags, vma->vm_flags, pgoff, 1);
 199                         fput(file);
 200                         if (IS_ERR_VALUE(addr)) {
 201                                 err = addr;
 202                         } else {
 203                                 BUG_ON(addr != start);
 204                                 err = 0;
 205                         }
 206                         goto out;
 207                 }
 208                 spin_lock(&mapping->i_mmap_lock);
 209                 flush_dcache_mmap_lock(mapping);
 210                 vma->vm_flags |= VM_NONLINEAR;
 211                 vma_prio_tree_remove(vma, &mapping->i_mmap);
 212                 vma_nonlinear_insert(vma, &mapping->i_mmap_nonlinear);
 213                 flush_dcache_mmap_unlock(mapping);
 214                 spin_unlock(&mapping->i_mmap_lock);
 215         }
 216
 217         err = populate_range(mm, vma, start, size, pgoff);
 218         if (!err && !(flags & MAP_NONBLOCK)) {
 219                 if (unlikely(has_write_lock)) {
 220                         downgrade_write(&mm->mmap_sem);
 221                         has_write_lock = 0;
 222                 }
 223                 make_pages_present(start, start+size);
 224         }
 225
 226         /*
 227          * We can't clear VM_NONLINEAR because we'd have to do
 228          * it after ->populate completes, and that would prevent
 229          * downgrading the lock.  (Locks can't be upgraded).
 230          */
 231
 232 out:
 233         if (likely(!has_write_lock))
 234                 up_read(&mm->mmap_sem);
 235         else
 236                 up_write(&mm->mmap_sem);
 237
 238         return err;
 239 }
 240