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
   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 int 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         struct page *page = NULL;
  28
  29         if (pte_present(pte)) {
  30                 flush_cache_page(vma, addr, pte_pfn(pte));
  31                 pte = ptep_clear_flush(vma, addr, ptep);
  32                 page = vm_normal_page(vma, addr, pte);
  33                 if (page) {
  34                         if (pte_dirty(pte))
  35                                 set_page_dirty(page);
  36                         page_remove_rmap(page);
  37                         page_cache_release(page);
  38                 }
  39         } else {
  40                 if (!pte_file(pte))
  41                         free_swap_and_cache(pte_to_swp_entry(pte));
  42                 pte_clear(mm, addr, ptep);
  43         }
  44         return !!page;
  45 }
  46
  47 /*
  48  * Install a file page to a given virtual memory address, release any
  49  * previously existing mapping.
  50  */
  51 int install_page(struct mm_struct *mm, struct vm_area_struct *vma,
  52                 unsigned long addr, struct page *page, pgprot_t prot)
  53 {
  54         struct inode *inode;
  55         pgoff_t size;
  56         int err = -ENOMEM;
  57         pte_t *pte;
  58         pte_t pte_val;
  59         spinlock_t *ptl;
  60
  61         pte = get_locked_pte(mm, addr, &ptl);
  62         if (!pte)
  63                 goto out;
  64
  65         /*
  66          * This page may have been truncated. Tell the
  67          * caller about it.
  68          */
  69         err = -EINVAL;
  70         inode = vma->vm_file->f_mapping->host;
  71         size = (i_size_read(inode) + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT;
  72         if (!page->mapping || page->index >= size)
  73                 goto unlock;
  74         err = -ENOMEM;
  75         if (page_mapcount(page) > INT_MAX/2)
  76                 goto unlock;
  77
  78         if (pte_none(*pte) || !zap_pte(mm, vma, addr, pte))
  79                 inc_mm_counter(mm, file_rss);
  80
  81         flush_icache_page(vma, page);
  82         set_pte_at(mm, addr, pte, mk_pte(page, prot));
  83         page_add_file_rmap(page);
  84         pte_val = *pte;
  85         update_mmu_cache(vma, addr, pte_val);
  86         err = 0;
  87 unlock:
  88         pte_unmap_unlock(pte, ptl);
  89 out:
  90         return err;
  91 }
  92 EXPORT_SYMBOL(install_page);
  93
  94 /*
  95  * Install a file pte to a given virtual memory address, release any
  96  * previously existing mapping.
  97  */
  98 int install_file_pte(struct mm_struct *mm, struct vm_area_struct *vma,
  99                 unsigned long addr, unsigned long pgoff, pgprot_t prot)
 100 {
 101         int err = -ENOMEM;
 102         pte_t *pte;
 103         pte_t pte_val;
 104         spinlock_t *ptl;
 105
 106         pte = get_locked_pte(mm, addr, &ptl);
 107         if (!pte)
 108                 goto out;
 109
 110         if (!pte_none(*pte) && zap_pte(mm, vma, addr, pte)) {
 111                 update_hiwater_rss(mm);
 112                 dec_mm_counter(mm, file_rss);
 113         }
 114
 115         set_pte_at(mm, addr, pte, pgoff_to_pte(pgoff));
 116         pte_val = *pte;
 117         update_mmu_cache(vma, addr, pte_val);
 118         pte_unmap_unlock(pte, ptl);
 119         err = 0;
 120 out:
 121         return err;
 122 }
 123
 124 /***
 125  * sys_remap_file_pages - remap arbitrary pages of a shared backing store
 126  *                        file within an existing vma.
 127  * @start: start of the remapped virtual memory range
 128  * @size: size of the remapped virtual memory range
 129  * @prot: new protection bits of the range
 130  * @pgoff: to be mapped page of the backing store file
 131  * @flags: 0 or MAP_NONBLOCKED - the later will cause no IO.
 132  *
 133  * this syscall works purely via pagetables, so it's the most efficient
 134  * way to map the same (large) file into a given virtual window. Unlike
 135  * mmap()/mremap() it does not create any new vmas. The new mappings are
 136  * also safe across swapout.
 137  *
 138  * NOTE: the 'prot' parameter right now is ignored, and the vma's default
 139  * protection is used. Arbitrary protections might be implemented in the
 140  * future.
 141  */
 142 asmlinkage long sys_remap_file_pages(unsigned long start, unsigned long size,
 143         unsigned long __prot, unsigned long pgoff, unsigned long flags)
 144 {
 145         struct mm_struct *mm = current->mm;
 146         struct address_space *mapping;
 147         unsigned long end = start + size;
 148         struct vm_area_struct *vma;
 149         int err = -EINVAL;
 150         int has_write_lock = 0;
 151
 152         if (__prot)
 153                 return err;
 154         /*
 155          * Sanitize the syscall parameters:
 156          */
 157         start = start & PAGE_MASK;
 158         size = size & PAGE_MASK;
 159
 160         /* Does the address range wrap, or is the span zero-sized? */
 161         if (start + size <= start)
 162                 return err;
 163
 164         /* Can we represent this offset inside this architecture's pte's? */
 165 #if PTE_FILE_MAX_BITS < BITS_PER_LONG
 166         if (pgoff + (size >> PAGE_SHIFT) >= (1UL << PTE_FILE_MAX_BITS))
 167                 return err;
 168 #endif
 169
 170         /* We need down_write() to change vma->vm_flags. */
 171         down_read(&mm->mmap_sem);
 172  retry:
 173         vma = find_vma(mm, start);
 174
 175         /*
 176          * Make sure the vma is shared, that it supports prefaulting,
 177          * and that the remapped range is valid and fully within
 178          * the single existing vma.  vm_private_data is used as a
 179          * swapout cursor in a VM_NONLINEAR vma.
 180          */
 181         if (vma && (vma->vm_flags & VM_SHARED) &&
 182                 (!vma->vm_private_data || (vma->vm_flags & VM_NONLINEAR)) &&
 183                 vma->vm_ops && vma->vm_ops->populate &&
 184                         end > start && start >= vma->vm_start &&
 185                                 end <= vma->vm_end) {
 186
 187                 /* Must set VM_NONLINEAR before any pages are populated. */
 188                 if (pgoff != linear_page_index(vma, start) &&
 189                     !(vma->vm_flags & VM_NONLINEAR)) {
 190                         if (!has_write_lock) {
 191                                 up_read(&mm->mmap_sem);
 192                                 down_write(&mm->mmap_sem);
 193                                 has_write_lock = 1;
 194                                 goto retry;
 195                         }
 196                         mapping = vma->vm_file->f_mapping;
 197                         spin_lock(&mapping->i_mmap_lock);
 198                         flush_dcache_mmap_lock(mapping);
 199                         vma->vm_flags |= VM_NONLINEAR;
 200                         vma_prio_tree_remove(vma, &mapping->i_mmap);
 201                         vma_nonlinear_insert(vma, &mapping->i_mmap_nonlinear);
 202                         flush_dcache_mmap_unlock(mapping);
 203                         spin_unlock(&mapping->i_mmap_lock);
 204                 }
 205
 206                 err = vma->vm_ops->populate(vma, start, size,
 207                                             vma->vm_page_prot,
 208                                             pgoff, flags & MAP_NONBLOCK);
 209
 210                 /*
 211                  * We can't clear VM_NONLINEAR because we'd have to do
 212                  * it after ->populate completes, and that would prevent
 213                  * downgrading the lock.  (Locks can't be upgraded).
 214                  */
 215         }
 216         if (likely(!has_write_lock))
 217                 up_read(&mm->mmap_sem);
 218         else
 219                 up_write(&mm->mmap_sem);
 220
 221         return err;
 222 }
 223