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
   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-locking.h>
  16 #include <asm/mmu_context.h>
  17 #include <asm/cacheflush.h>
  18 #include <asm/tlbflush.h>
  19
  20 static inline int zap_pte(struct mm_struct *mm, struct vm_area_struct *vma,
  21                         unsigned long addr, pte_t *ptep)
  22 {
  23         pte_t pte = *ptep;
  24
  25         if (pte_none(pte))
  26                 return 0;
  27         if (pte_present(pte)) {
  28                 unsigned long pfn = pte_pfn(pte);
  29
  30                 flush_cache_page(vma, addr);
  31                 pte = ptep_get_and_clear(ptep);
  32                 if (pfn_valid(pfn)) {
  33                         struct page *page = pfn_to_page(pfn);
  34                         if (!PageReserved(page)) {
  35                                 if (pte_dirty(pte))
  36                                         set_page_dirty(page);
  37                                 page_remove_rmap(page, ptep);
  38                                 page_cache_release(page);
  39                                 mm->rss--;
  40                         }
  41                 }
  42                 return 1;
  43         } else {
  44                 if (!pte_file(pte))
  45                         free_swap_and_cache(pte_to_swp_entry(pte));
  46                 pte_clear(ptep);
  47                 return 0;
  48         }
  49 }
  50
  51 /*
  52  * Install a page to a given virtual memory address, release any
  53  * previously existing mapping.
  54  */
  55 int install_page(struct mm_struct *mm, struct vm_area_struct *vma,
  56                 unsigned long addr, struct page *page, pgprot_t prot)
  57 {
  58         int err = -ENOMEM, flush;
  59         pte_t *pte;
  60         pgd_t *pgd;
  61         pmd_t *pmd;
  62         struct pte_chain *pte_chain;
  63
  64         pte_chain = pte_chain_alloc(GFP_KERNEL);
  65         if (!pte_chain)
  66                 goto err;
  67         pgd = pgd_offset(mm, addr);
  68         spin_lock(&mm->page_table_lock);
  69
  70         pmd = pmd_alloc(mm, pgd, addr);
  71         if (!pmd)
  72                 goto err_unlock;
  73
  74         pte = pte_alloc_map(mm, pmd, addr);
  75         if (!pte)
  76                 goto err_unlock;
  77
  78         flush = zap_pte(mm, vma, addr, pte);
  79
  80         mm->rss++;
  81         flush_icache_page(vma, page);
  82         set_pte(pte, mk_pte(page, prot));
  83         pte_chain = page_add_rmap(page, pte, pte_chain);
  84         pte_unmap(pte);
  85         if (flush)
  86                 flush_tlb_page(vma, addr);
  87         update_mmu_cache(vma, addr, *pte);
  88         spin_unlock(&mm->page_table_lock);
  89         pte_chain_free(pte_chain);
  90         return 0;
  91
  92 err_unlock:
  93         spin_unlock(&mm->page_table_lock);
  94         pte_chain_free(pte_chain);
  95 err:
  96         return err;
  97 }
  98
  99 /***
 100  * sys_remap_file_pages - remap arbitrary pages of a shared backing store
 101  *                        file within an existing 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
 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  * this syscall works purely via pagetables, so it's the most efficient
 109  * way to map the same (large) file into a given virtual window. Unlike
 110  * mmap()/mremap() it does not create any new vmas. The new mappings are
 111  * also safe across swapout.
 112  *
 113  * NOTE: the 'prot' parameter right now is ignored, and the vma's default
 114  * protection is used. Arbitrary protections might be implemented in the
 115  * future.
 116  */
 117 long sys_remap_file_pages(unsigned long start, unsigned long size,
 118         unsigned long __prot, unsigned long pgoff, unsigned long flags)
 119 {
 120         struct mm_struct *mm = current->mm;
 121         unsigned long end = start + size;
 122         struct vm_area_struct *vma;
 123         int err = -EINVAL;
 124
 125         if (__prot)
 126                 return err;
 127         /*
 128          * Sanitize the syscall parameters:
 129          */
 130         start = start & PAGE_MASK;
 131         size = size & PAGE_MASK;
 132
 133         /* Does the address range wrap, or is the span zero-sized? */
 134         if (start + size <= start)
 135                 return err;
 136
 137         /* Can we represent this offset inside this architecture's pte's? */
 138 #if PTE_FILE_MAX_BITS < BITS_PER_LONG
 139         if (pgoff + (size >> PAGE_SHIFT) >= (1UL << PTE_FILE_MAX_BITS))
 140                 return err;
 141 #endif
 142
 143         down_read(&mm->mmap_sem);
 144
 145         vma = find_vma(mm, start);
 146         /*
 147          * Make sure the vma is shared, that it supports prefaulting,
 148          * and that the remapped range is valid and fully within
 149          * the single existing vma:
 150          */
 151         if (vma && (vma->vm_flags & VM_SHARED) &&
 152                 vma->vm_ops && vma->vm_ops->populate &&
 153                         end > start && start >= vma->vm_start &&
 154                                 end <= vma->vm_end)
 155                 err = vma->vm_ops->populate(vma, start, size, vma->vm_page_prot,
 156                                 pgoff, flags & MAP_NONBLOCK);
 157
 158         up_read(&mm->mmap_sem);
 159
 160         return err;
 161 }
 162