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/syscalls.h>
  17 #include <linux/mmu_notifier.h>
  18
  19 #include <asm/mmu_context.h>
  20 #include <asm/cacheflush.h>
  21 #include <asm/tlbflush.h>
  22
  23 #include "internal.h"
  24
  25 static void zap_pte(struct mm_struct *mm, struct vm_area_struct *vma,
  26                         unsigned long addr, pte_t *ptep)
  27 {
  28         pte_t pte = *ptep;
  29
  30         if (pte_present(pte)) {
  31                 struct page *page;
  32
  33                 flush_cache_page(vma, addr, pte_pfn(pte));
  34                 pte = ptep_clear_flush(vma, addr, ptep);
  35                 page = vm_normal_page(vma, addr, pte);
  36                 if (page) {
  37                         if (pte_dirty(pte))
  38                                 set_page_dirty(page);
  39                         page_remove_rmap(page);
  40                         page_cache_release(page);
  41                         update_hiwater_rss(mm);
  42                         dec_mm_counter(mm, MM_FILEPAGES);
  43                 }
  44         } else {
  45                 if (!pte_file(pte))
  46                         free_swap_and_cache(pte_to_swp_entry(pte));
  47                 pte_clear_not_present_full(mm, addr, ptep, 0);
  48         }
  49 }
  50
  51 /*
  52  * Install a file pte to a given virtual memory address, release any
  53  * previously existing mapping.
  54  */
  55 static int install_file_pte(struct mm_struct *mm, struct vm_area_struct *vma,
  56                 unsigned long addr, unsigned long pgoff, pgprot_t prot)
  57 {
  58         int err = -ENOMEM;
  59         pte_t *pte;
  60         spinlock_t *ptl;
  61
  62         pte = get_locked_pte(mm, addr, &ptl);
  63         if (!pte)
  64                 goto out;
  65
  66         if (!pte_none(*pte))
  67                 zap_pte(mm, vma, addr, pte);
  68
  69         set_pte_at(mm, addr, pte, pgoff_to_pte(pgoff));
  70         /*
  71          * We don't need to run update_mmu_cache() here because the "file pte"
  72          * being installed by install_file_pte() is not a real pte - it's a
  73          * non-present entry (like a swap entry), noting what file offset should
  74          * be mapped there when there's a fault (in a non-linear vma where
  75          * that's not obvious).
  76          */
  77         pte_unmap_unlock(pte, ptl);
  78         err = 0;
  79 out:
  80         return err;
  81 }
  82
  83 static int populate_range(struct mm_struct *mm, struct vm_area_struct *vma,
  84                         unsigned long addr, unsigned long size, pgoff_t pgoff)
  85 {
  86         int err;
  87
  88         do {
  89                 err = install_file_pte(mm, vma, addr, pgoff, vma->vm_page_prot);
  90                 if (err)
  91                         return err;
  92
  93                 size -= PAGE_SIZE;
  94                 addr += PAGE_SIZE;
  95                 pgoff++;
  96         } while (size);
  97
  98         return 0;
  99
 100 }
 101
 102 /**
 103  * sys_remap_file_pages - remap arbitrary pages of an existing VM_SHARED vma
 104  * @start: start of the remapped virtual memory range
 105  * @size: size of the remapped virtual memory range
 106  * @prot: new protection bits of the range (see NOTE)
 107  * @pgoff: to-be-mapped page of the backing store file
 108  * @flags: 0 or MAP_NONBLOCKED - the later will cause no IO.
 109  *
 110  * sys_remap_file_pages remaps arbitrary pages of an existing VM_SHARED vma
 111  * (shared backing store file).
 112  *
 113  * This syscall works purely via pagetables, so it's the most efficient
 114  * way to map the same (large) file into a given virtual window. Unlike
 115  * mmap()/mremap() it does not create any new vmas. The new mappings are
 116  * also safe across swapout.
 117  *
 118  * NOTE: the @prot parameter right now is ignored (but must be zero),
 119  * and the vma's default protection is used. Arbitrary protections
 120  * might be implemented in the future.
 121  */
 122 SYSCALL_DEFINE5(remap_file_pages, unsigned long, start, unsigned long, size,
 123                 unsigned long, prot, unsigned long, pgoff, unsigned long, flags)
 124 {
 125         struct mm_struct *mm = current->mm;
 126         struct address_space *mapping;
 127         struct vm_area_struct *vma;
 128         int err = -EINVAL;
 129         int has_write_lock = 0;
 130
 131         if (prot)
 132                 return err;
 133         /*
 134          * Sanitize the syscall parameters:
 135          */
 136         start = start & PAGE_MASK;
 137         size = size & PAGE_MASK;
 138
 139         /* Does the address range wrap, or is the span zero-sized? */
 140         if (start + size <= start)
 141                 return err;
 142
 143         /* Does pgoff wrap? */
 144         if (pgoff + (size >> PAGE_SHIFT) < pgoff)
 145                 return err;
 146
 147         /* Can we represent this offset inside this architecture's pte's? */
 148 #if PTE_FILE_MAX_BITS < BITS_PER_LONG
 149         if (pgoff + (size >> PAGE_SHIFT) >= (1UL << PTE_FILE_MAX_BITS))
 150                 return err;
 151 #endif
 152
 153         /* We need down_write() to change vma->vm_flags. */
 154         down_read(&mm->mmap_sem);
 155  retry:
 156         vma = find_vma(mm, start);
 157
 158         /*
 159          * Make sure the vma is shared, that it supports prefaulting,
 160          * and that the remapped range is valid and fully within
 161          * the single existing vma.  vm_private_data is used as a
 162          * swapout cursor in a VM_NONLINEAR vma.
 163          */
 164         if (!vma || !(vma->vm_flags & VM_SHARED))
 165                 goto out;
 166
 167         if (vma->vm_private_data && !(vma->vm_flags & VM_NONLINEAR))
 168                 goto out;
 169
 170         if (!(vma->vm_flags & VM_CAN_NONLINEAR))
 171                 goto out;
 172
 173         if (start < vma->vm_start || start + size > vma->vm_end)
 174                 goto out;
 175
 176         /* Must set VM_NONLINEAR before any pages are populated. */
 177         if (!(vma->vm_flags & VM_NONLINEAR)) {
 178                 /* Don't need a nonlinear mapping, exit success */
 179                 if (pgoff == linear_page_index(vma, start)) {
 180                         err = 0;
 181                         goto out;
 182                 }
 183
 184                 if (!has_write_lock) {
 185                         up_read(&mm->mmap_sem);
 186                         down_write(&mm->mmap_sem);
 187                         has_write_lock = 1;
 188                         goto retry;
 189                 }
 190                 mapping = vma->vm_file->f_mapping;
 191                 /*
 192                  * page_mkclean doesn't work on nonlinear vmas, so if
 193                  * dirty pages need to be accounted, emulate with linear
 194                  * vmas.
 195                  */
 196                 if (mapping_cap_account_dirty(mapping)) {
 197                         unsigned long addr;
 198                         struct file *file = vma->vm_file;
 199
 200                         flags &= MAP_NONBLOCK;
 201                         get_file(file);
 202                         addr = mmap_region(file, start, size,
 203                                         flags, vma->vm_flags, pgoff);
 204                         fput(file);
 205                         if (IS_ERR_VALUE(addr)) {
 206                                 err = addr;
 207                         } else {
 208                                 BUG_ON(addr != start);
 209                                 err = 0;
 210                         }
 211                         goto out;
 212                 }
 213                 mutex_lock(&mapping->i_mmap_mutex);
 214                 flush_dcache_mmap_lock(mapping);
 215                 vma->vm_flags |= VM_NONLINEAR;
 216                 vma_prio_tree_remove(vma, &mapping->i_mmap);
 217                 vma_nonlinear_insert(vma, &mapping->i_mmap_nonlinear);
 218                 flush_dcache_mmap_unlock(mapping);
 219                 mutex_unlock(&mapping->i_mmap_mutex);
 220         }
 221
 222         if (vma->vm_flags & VM_LOCKED) {
 223                 /*
 224                  * drop PG_Mlocked flag for over-mapped range
 225                  */
 226                 vm_flags_t saved_flags = vma->vm_flags;
 227                 munlock_vma_pages_range(vma, start, start + size);
 228                 vma->vm_flags = saved_flags;
 229         }
 230
 231         mmu_notifier_invalidate_range_start(mm, start, start + size);
 232         err = populate_range(mm, vma, start, size, pgoff);
 233         mmu_notifier_invalidate_range_end(mm, start, start + size);
 234         if (!err && !(flags & MAP_NONBLOCK)) {
 235                 if (vma->vm_flags & VM_LOCKED) {
 236                         /*
 237                          * might be mapping previously unmapped range of file
 238                          */
 239                         mlock_vma_pages_range(vma, start, start + size);
 240                 } else {
 241                         if (unlikely(has_write_lock)) {
 242                                 downgrade_write(&mm->mmap_sem);
 243                                 has_write_lock = 0;
 244                         }
 245                         make_pages_present(start, start+size);
 246                 }
 247         }
 248
 249         /*
 250          * We can't clear VM_NONLINEAR because we'd have to do
 251          * it after ->populate completes, and that would prevent
 252          * downgrading the lock.  (Locks can't be upgraded).
 253          */
 254
 255 out:
 256         if (likely(!has_write_lock))
 257                 up_read(&mm->mmap_sem);
 258         else
 259                 up_write(&mm->mmap_sem);
 260
 261         return err;
 262 }