4 * Explicit pagetable population and nonlinear (random) mappings support.
6 * started by Ingo Molnar, Copyright (C) 2002, 2003
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>
19 #include <asm/mmu_context.h>
20 #include <asm/cacheflush.h>
21 #include <asm/tlbflush.h>
23 static int zap_pte(struct mm_struct *mm, struct vm_area_struct *vma,
24 unsigned long addr, pte_t *ptep)
27 struct page *page = NULL;
29 if (pte_present(pte)) {
30 unsigned long pfn = pte_pfn(pte);
31 flush_cache_page(vma, addr, pfn);
32 pte = ptep_clear_flush(vma, addr, ptep);
33 if (unlikely(!pfn_valid(pfn))) {
34 print_bad_pte(vma, pte, addr);
37 page = pfn_to_page(pfn);
40 page_remove_rmap(page);
41 page_cache_release(page);
44 free_swap_and_cache(pte_to_swp_entry(pte));
45 pte_clear(mm, addr, ptep);
52 * Install a file page to a given virtual memory address, release any
53 * previously existing mapping.
55 int install_page(struct mm_struct *mm, struct vm_area_struct *vma,
56 unsigned long addr, struct page *page, pgprot_t prot)
67 BUG_ON(vma->vm_flags & VM_RESERVED);
69 pgd = pgd_offset(mm, addr);
70 spin_lock(&mm->page_table_lock);
72 pud = pud_alloc(mm, pgd, addr);
76 pmd = pmd_alloc(mm, pud, addr);
80 pte = pte_alloc_map(mm, pmd, addr);
85 * This page may have been truncated. Tell the
89 inode = vma->vm_file->f_mapping->host;
90 size = (i_size_read(inode) + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT;
91 if (!page->mapping || page->index >= size)
94 if (page_mapcount(page) > INT_MAX/2)
97 if (pte_none(*pte) || !zap_pte(mm, vma, addr, pte))
98 inc_mm_counter(mm, file_rss);
100 flush_icache_page(vma, page);
101 set_pte_at(mm, addr, pte, mk_pte(page, prot));
102 page_add_file_rmap(page);
105 update_mmu_cache(vma, addr, pte_val);
109 spin_unlock(&mm->page_table_lock);
112 EXPORT_SYMBOL(install_page);
116 * Install a file pte to a given virtual memory address, release any
117 * previously existing mapping.
119 int install_file_pte(struct mm_struct *mm, struct vm_area_struct *vma,
120 unsigned long addr, unsigned long pgoff, pgprot_t prot)
129 BUG_ON(vma->vm_flags & VM_RESERVED);
131 pgd = pgd_offset(mm, addr);
132 spin_lock(&mm->page_table_lock);
134 pud = pud_alloc(mm, pgd, addr);
138 pmd = pmd_alloc(mm, pud, addr);
142 pte = pte_alloc_map(mm, pmd, addr);
146 if (!pte_none(*pte) && zap_pte(mm, vma, addr, pte))
147 dec_mm_counter(mm, file_rss);
149 set_pte_at(mm, addr, pte, pgoff_to_pte(pgoff));
152 update_mmu_cache(vma, addr, pte_val);
153 spin_unlock(&mm->page_table_lock);
157 spin_unlock(&mm->page_table_lock);
163 * sys_remap_file_pages - remap arbitrary pages of a shared backing store
164 * file within an existing vma.
165 * @start: start of the remapped virtual memory range
166 * @size: size of the remapped virtual memory range
167 * @prot: new protection bits of the range
168 * @pgoff: to be mapped page of the backing store file
169 * @flags: 0 or MAP_NONBLOCKED - the later will cause no IO.
171 * this syscall works purely via pagetables, so it's the most efficient
172 * way to map the same (large) file into a given virtual window. Unlike
173 * mmap()/mremap() it does not create any new vmas. The new mappings are
174 * also safe across swapout.
176 * NOTE: the 'prot' parameter right now is ignored, and the vma's default
177 * protection is used. Arbitrary protections might be implemented in the
180 asmlinkage long sys_remap_file_pages(unsigned long start, unsigned long size,
181 unsigned long __prot, unsigned long pgoff, unsigned long flags)
183 struct mm_struct *mm = current->mm;
184 struct address_space *mapping;
185 unsigned long end = start + size;
186 struct vm_area_struct *vma;
188 int has_write_lock = 0;
193 * Sanitize the syscall parameters:
195 start = start & PAGE_MASK;
196 size = size & PAGE_MASK;
198 /* Does the address range wrap, or is the span zero-sized? */
199 if (start + size <= start)
202 /* Can we represent this offset inside this architecture's pte's? */
203 #if PTE_FILE_MAX_BITS < BITS_PER_LONG
204 if (pgoff + (size >> PAGE_SHIFT) >= (1UL << PTE_FILE_MAX_BITS))
208 /* We need down_write() to change vma->vm_flags. */
209 down_read(&mm->mmap_sem);
211 vma = find_vma(mm, start);
214 * Make sure the vma is shared, that it supports prefaulting,
215 * and that the remapped range is valid and fully within
216 * the single existing vma. vm_private_data is used as a
217 * swapout cursor in a VM_NONLINEAR vma (unless VM_RESERVED
218 * or VM_LOCKED, but VM_LOCKED could be revoked later on).
220 if (vma && (vma->vm_flags & VM_SHARED) &&
221 (!vma->vm_private_data ||
222 (vma->vm_flags & (VM_NONLINEAR|VM_RESERVED))) &&
223 vma->vm_ops && vma->vm_ops->populate &&
224 end > start && start >= vma->vm_start &&
225 end <= vma->vm_end) {
227 /* Must set VM_NONLINEAR before any pages are populated. */
228 if (pgoff != linear_page_index(vma, start) &&
229 !(vma->vm_flags & VM_NONLINEAR)) {
230 if (!has_write_lock) {
231 up_read(&mm->mmap_sem);
232 down_write(&mm->mmap_sem);
236 mapping = vma->vm_file->f_mapping;
237 spin_lock(&mapping->i_mmap_lock);
238 flush_dcache_mmap_lock(mapping);
239 vma->vm_flags |= VM_NONLINEAR;
240 vma_prio_tree_remove(vma, &mapping->i_mmap);
241 vma_nonlinear_insert(vma, &mapping->i_mmap_nonlinear);
242 flush_dcache_mmap_unlock(mapping);
243 spin_unlock(&mapping->i_mmap_lock);
246 err = vma->vm_ops->populate(vma, start, size,
248 pgoff, flags & MAP_NONBLOCK);
251 * We can't clear VM_NONLINEAR because we'd have to do
252 * it after ->populate completes, and that would prevent
253 * downgrading the lock. (Locks can't be upgraded).
256 if (likely(!has_write_lock))
257 up_read(&mm->mmap_sem);
259 up_write(&mm->mmap_sem);