4 * Explicit pagetable population and nonlinear (random) mappings support.
6 * started by Ingo Molnar, Copyright (C) 2002, 2003
8 #include <linux/backing-dev.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 #include <linux/mmu_notifier.h>
20 #include <asm/mmu_context.h>
21 #include <asm/cacheflush.h>
22 #include <asm/tlbflush.h>
24 static void zap_pte(struct mm_struct *mm, struct vm_area_struct *vma,
25 unsigned long addr, pte_t *ptep)
29 if (pte_present(pte)) {
32 flush_cache_page(vma, addr, pte_pfn(pte));
33 pte = ptep_clear_flush(vma, addr, ptep);
34 page = vm_normal_page(vma, addr, pte);
38 page_remove_rmap(page, vma);
39 page_cache_release(page);
40 update_hiwater_rss(mm);
41 dec_mm_counter(mm, file_rss);
45 free_swap_and_cache(pte_to_swp_entry(pte));
46 pte_clear_not_present_full(mm, addr, ptep, 0);
51 * Install a file pte to a given virtual memory address, release any
52 * previously existing mapping.
54 static int install_file_pte(struct mm_struct *mm, struct vm_area_struct *vma,
55 unsigned long addr, unsigned long pgoff, pgprot_t prot)
61 pte = get_locked_pte(mm, addr, &ptl);
66 zap_pte(mm, vma, addr, pte);
68 set_pte_at(mm, addr, pte, pgoff_to_pte(pgoff));
70 * We don't need to run update_mmu_cache() here because the "file pte"
71 * being installed by install_file_pte() is not a real pte - it's a
72 * non-present entry (like a swap entry), noting what file offset should
73 * be mapped there when there's a fault (in a non-linear vma where
74 * that's not obvious).
76 pte_unmap_unlock(pte, ptl);
82 static int populate_range(struct mm_struct *mm, struct vm_area_struct *vma,
83 unsigned long addr, unsigned long size, pgoff_t pgoff)
88 err = install_file_pte(mm, vma, addr, pgoff, vma->vm_page_prot);
102 * sys_remap_file_pages - remap arbitrary pages of an existing VM_SHARED vma
103 * @start: start of the remapped virtual memory range
104 * @size: size of the remapped virtual memory range
105 * @prot: new protection bits of the range (see NOTE)
106 * @pgoff: to-be-mapped page of the backing store file
107 * @flags: 0 or MAP_NONBLOCKED - the later will cause no IO.
109 * sys_remap_file_pages remaps arbitrary pages of an existing VM_SHARED vma
110 * (shared backing store file).
112 * This syscall works purely via pagetables, so it's the most efficient
113 * way to map the same (large) file into a given virtual window. Unlike
114 * mmap()/mremap() it does not create any new vmas. The new mappings are
115 * also safe across swapout.
117 * NOTE: the @prot parameter right now is ignored (but must be zero),
118 * and the vma's default protection is used. Arbitrary protections
119 * might be implemented in the future.
121 asmlinkage long sys_remap_file_pages(unsigned long start, unsigned long size,
122 unsigned long prot, unsigned long pgoff, unsigned long flags)
124 struct mm_struct *mm = current->mm;
125 struct address_space *mapping;
126 unsigned long end = start + size;
127 struct vm_area_struct *vma;
129 int has_write_lock = 0;
134 * Sanitize the syscall parameters:
136 start = start & PAGE_MASK;
137 size = size & PAGE_MASK;
139 /* Does the address range wrap, or is the span zero-sized? */
140 if (start + size <= start)
143 /* Can we represent this offset inside this architecture's pte's? */
144 #if PTE_FILE_MAX_BITS < BITS_PER_LONG
145 if (pgoff + (size >> PAGE_SHIFT) >= (1UL << PTE_FILE_MAX_BITS))
149 /* We need down_write() to change vma->vm_flags. */
150 down_read(&mm->mmap_sem);
152 vma = find_vma(mm, start);
155 * Make sure the vma is shared, that it supports prefaulting,
156 * and that the remapped range is valid and fully within
157 * the single existing vma. vm_private_data is used as a
158 * swapout cursor in a VM_NONLINEAR vma.
160 if (!vma || !(vma->vm_flags & VM_SHARED))
163 if (vma->vm_private_data && !(vma->vm_flags & VM_NONLINEAR))
166 if (!(vma->vm_flags & VM_CAN_NONLINEAR))
169 if (end <= start || start < vma->vm_start || end > vma->vm_end)
172 /* Must set VM_NONLINEAR before any pages are populated. */
173 if (!(vma->vm_flags & VM_NONLINEAR)) {
174 /* Don't need a nonlinear mapping, exit success */
175 if (pgoff == linear_page_index(vma, start)) {
180 if (!has_write_lock) {
181 up_read(&mm->mmap_sem);
182 down_write(&mm->mmap_sem);
186 mapping = vma->vm_file->f_mapping;
188 * page_mkclean doesn't work on nonlinear vmas, so if
189 * dirty pages need to be accounted, emulate with linear
192 if (mapping_cap_account_dirty(mapping)) {
194 struct file *file = vma->vm_file;
196 flags &= MAP_NONBLOCK;
198 addr = mmap_region(file, start, size,
199 flags, vma->vm_flags, pgoff, 1);
201 if (IS_ERR_VALUE(addr)) {
204 BUG_ON(addr != start);
209 spin_lock(&mapping->i_mmap_lock);
210 flush_dcache_mmap_lock(mapping);
211 vma->vm_flags |= VM_NONLINEAR;
212 vma_prio_tree_remove(vma, &mapping->i_mmap);
213 vma_nonlinear_insert(vma, &mapping->i_mmap_nonlinear);
214 flush_dcache_mmap_unlock(mapping);
215 spin_unlock(&mapping->i_mmap_lock);
218 mmu_notifier_invalidate_range_start(mm, start, start + size);
219 err = populate_range(mm, vma, start, size, pgoff);
220 mmu_notifier_invalidate_range_end(mm, start, start + size);
221 if (!err && !(flags & MAP_NONBLOCK)) {
222 if (unlikely(has_write_lock)) {
223 downgrade_write(&mm->mmap_sem);
226 make_pages_present(start, start+size);
230 * We can't clear VM_NONLINEAR because we'd have to do
231 * it after ->populate completes, and that would prevent
232 * downgrading the lock. (Locks can't be upgraded).
236 if (likely(!has_write_lock))
237 up_read(&mm->mmap_sem);
239 up_write(&mm->mmap_sem);