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>
18 #include <linux/vs_base.h>
19 #include <linux/vs_memory.h>
21 #include <asm/mmu_context.h>
22 #include <asm/cacheflush.h>
23 #include <asm/tlbflush.h>
25 static int zap_pte(struct mm_struct *mm, struct vm_area_struct *vma,
26 unsigned long addr, pte_t *ptep)
29 struct page *page = NULL;
31 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);
39 page_cache_release(page);
43 free_swap_and_cache(pte_to_swp_entry(pte));
44 pte_clear(mm, addr, ptep);
50 * Install a file page to a given virtual memory address, release any
51 * previously existing mapping.
53 int install_page(struct mm_struct *mm, struct vm_area_struct *vma,
54 unsigned long addr, struct page *page, pgprot_t prot)
63 pte = get_locked_pte(mm, addr, &ptl);
68 * This page may have been truncated. Tell the
72 inode = vma->vm_file->f_mapping->host;
73 size = (i_size_read(inode) + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT;
74 if (!page->mapping || page->index >= size)
77 if (page_mapcount(page) > INT_MAX/2)
79 if (!vx_rsspages_avail(mm, 1))
82 if (pte_none(*pte) || !zap_pte(mm, vma, addr, pte))
83 inc_mm_counter(mm, file_rss);
85 flush_icache_page(vma, page);
86 set_pte_at(mm, addr, pte, mk_pte(page, prot));
87 page_add_file_rmap(page);
89 update_mmu_cache(vma, addr, pte_val);
92 pte_unmap_unlock(pte, ptl);
96 EXPORT_SYMBOL(install_page);
99 * Install a file pte to a given virtual memory address, release any
100 * previously existing mapping.
102 int install_file_pte(struct mm_struct *mm, struct vm_area_struct *vma,
103 unsigned long addr, unsigned long pgoff, pgprot_t prot)
110 pte = get_locked_pte(mm, addr, &ptl);
114 if (!pte_none(*pte) && zap_pte(mm, vma, addr, pte)) {
115 update_hiwater_rss(mm);
116 dec_mm_counter(mm, file_rss);
119 set_pte_at(mm, addr, pte, pgoff_to_pte(pgoff));
121 update_mmu_cache(vma, addr, pte_val);
122 pte_unmap_unlock(pte, ptl);
129 * sys_remap_file_pages - remap arbitrary pages of a shared backing store
130 * file within an existing vma.
131 * @start: start of the remapped virtual memory range
132 * @size: size of the remapped virtual memory range
133 * @prot: new protection bits of the range
134 * @pgoff: to be mapped page of the backing store file
135 * @flags: 0 or MAP_NONBLOCKED - the later will cause no IO.
137 * this syscall works purely via pagetables, so it's the most efficient
138 * way to map the same (large) file into a given virtual window. Unlike
139 * mmap()/mremap() it does not create any new vmas. The new mappings are
140 * also safe across swapout.
142 * NOTE: the 'prot' parameter right now is ignored, and the vma's default
143 * protection is used. Arbitrary protections might be implemented in the
146 asmlinkage long sys_remap_file_pages(unsigned long start, unsigned long size,
147 unsigned long __prot, unsigned long pgoff, unsigned long flags)
149 struct mm_struct *mm = current->mm;
150 struct address_space *mapping;
151 unsigned long end = start + size;
152 struct vm_area_struct *vma;
154 int has_write_lock = 0;
159 * Sanitize the syscall parameters:
161 start = start & PAGE_MASK;
162 size = size & PAGE_MASK;
164 /* Does the address range wrap, or is the span zero-sized? */
165 if (start + size <= start)
168 /* Can we represent this offset inside this architecture's pte's? */
169 #if PTE_FILE_MAX_BITS < BITS_PER_LONG
170 if (pgoff + (size >> PAGE_SHIFT) >= (1UL << PTE_FILE_MAX_BITS))
174 /* We need down_write() to change vma->vm_flags. */
175 down_read(&mm->mmap_sem);
177 vma = find_vma(mm, start);
180 * Make sure the vma is shared, that it supports prefaulting,
181 * and that the remapped range is valid and fully within
182 * the single existing vma. vm_private_data is used as a
183 * swapout cursor in a VM_NONLINEAR vma.
185 if (vma && (vma->vm_flags & VM_SHARED) &&
186 (!vma->vm_private_data || (vma->vm_flags & VM_NONLINEAR)) &&
187 vma->vm_ops && vma->vm_ops->populate &&
188 end > start && start >= vma->vm_start &&
189 end <= vma->vm_end) {
191 /* Must set VM_NONLINEAR before any pages are populated. */
192 if (pgoff != linear_page_index(vma, start) &&
193 !(vma->vm_flags & VM_NONLINEAR)) {
194 if (!has_write_lock) {
195 up_read(&mm->mmap_sem);
196 down_write(&mm->mmap_sem);
200 mapping = vma->vm_file->f_mapping;
201 spin_lock(&mapping->i_mmap_lock);
202 flush_dcache_mmap_lock(mapping);
203 vma->vm_flags |= VM_NONLINEAR;
204 vma_prio_tree_remove(vma, &mapping->i_mmap);
205 vma_nonlinear_insert(vma, &mapping->i_mmap_nonlinear);
206 flush_dcache_mmap_unlock(mapping);
207 spin_unlock(&mapping->i_mmap_lock);
210 err = vma->vm_ops->populate(vma, start, size,
212 pgoff, flags & MAP_NONBLOCK);
215 * We can't clear VM_NONLINEAR because we'd have to do
216 * it after ->populate completes, and that would prevent
217 * downgrading the lock. (Locks can't be upgraded).
220 if (likely(!has_write_lock))
221 up_read(&mm->mmap_sem);
223 up_write(&mm->mmap_sem);