blob: 69a37c2bdf815ce5d7d14f90cca4c1185b65dff7 [file] [log] [blame]
Linus Torvalds1da177e2005-04-16 15:20:36 -07001/*
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 */
Alexey Dobriyan4af3c9c2007-10-16 23:29:23 -07008#include <linux/backing-dev.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -07009#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/module.h>
17#include <linux/syscalls.h>
18
19#include <asm/mmu_context.h>
20#include <asm/cacheflush.h>
21#include <asm/tlbflush.h>
22
Nick Piggind0217ac2007-07-19 01:47:03 -070023static void zap_pte(struct mm_struct *mm, struct vm_area_struct *vma,
Linus Torvalds1da177e2005-04-16 15:20:36 -070024 unsigned long addr, pte_t *ptep)
25{
26 pte_t pte = *ptep;
27
Linus Torvalds1da177e2005-04-16 15:20:36 -070028 if (pte_present(pte)) {
Nick Piggind0217ac2007-07-19 01:47:03 -070029 struct page *page;
30
Linus Torvalds6aab3412005-11-28 14:34:23 -080031 flush_cache_page(vma, addr, pte_pfn(pte));
Linus Torvalds1da177e2005-04-16 15:20:36 -070032 pte = ptep_clear_flush(vma, addr, ptep);
Linus Torvalds6aab3412005-11-28 14:34:23 -080033 page = vm_normal_page(vma, addr, pte);
34 if (page) {
35 if (pte_dirty(pte))
36 set_page_dirty(page);
Nick Piggin7de6b802006-12-22 01:09:33 -080037 page_remove_rmap(page, vma);
Linus Torvalds6aab3412005-11-28 14:34:23 -080038 page_cache_release(page);
Nick Piggind0217ac2007-07-19 01:47:03 -070039 update_hiwater_rss(mm);
40 dec_mm_counter(mm, file_rss);
Linus Torvalds1da177e2005-04-16 15:20:36 -070041 }
42 } else {
43 if (!pte_file(pte))
44 free_swap_and_cache(pte_to_swp_entry(pte));
Zachary Amsden9888a1c2006-09-30 23:29:31 -070045 pte_clear_not_present_full(mm, addr, ptep, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -070046 }
47}
48
49/*
Linus Torvalds1da177e2005-04-16 15:20:36 -070050 * Install a file pte to a given virtual memory address, release any
51 * previously existing mapping.
52 */
Nick Piggind0217ac2007-07-19 01:47:03 -070053static int install_file_pte(struct mm_struct *mm, struct vm_area_struct *vma,
Linus Torvalds1da177e2005-04-16 15:20:36 -070054 unsigned long addr, unsigned long pgoff, pgprot_t prot)
55{
56 int err = -ENOMEM;
57 pte_t *pte;
Hugh Dickinsc74df322005-10-29 18:16:23 -070058 spinlock_t *ptl;
Linus Torvalds1da177e2005-04-16 15:20:36 -070059
Linus Torvaldsc9cfcdd2005-11-29 14:03:14 -080060 pte = get_locked_pte(mm, addr, &ptl);
Linus Torvalds1da177e2005-04-16 15:20:36 -070061 if (!pte)
Hugh Dickinsc74df322005-10-29 18:16:23 -070062 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -070063
Nick Piggind0217ac2007-07-19 01:47:03 -070064 if (!pte_none(*pte))
65 zap_pte(mm, vma, addr, pte);
Linus Torvalds1da177e2005-04-16 15:20:36 -070066
67 set_pte_at(mm, addr, pte, pgoff_to_pte(pgoff));
Hugh Dickins668e0d82006-06-23 02:03:45 -070068 /*
69 * We don't need to run update_mmu_cache() here because the "file pte"
70 * being installed by install_file_pte() is not a real pte - it's a
71 * non-present entry (like a swap entry), noting what file offset should
72 * be mapped there when there's a fault (in a non-linear vma where
73 * that's not obvious).
74 */
Hugh Dickinsc74df322005-10-29 18:16:23 -070075 pte_unmap_unlock(pte, ptl);
76 err = 0;
77out:
Linus Torvalds1da177e2005-04-16 15:20:36 -070078 return err;
79}
80
Nick Piggin54cb8822007-07-19 01:46:59 -070081static int populate_range(struct mm_struct *mm, struct vm_area_struct *vma,
82 unsigned long addr, unsigned long size, pgoff_t pgoff)
83{
84 int err;
85
86 do {
87 err = install_file_pte(mm, vma, addr, pgoff, vma->vm_page_prot);
88 if (err)
89 return err;
90
91 size -= PAGE_SIZE;
92 addr += PAGE_SIZE;
93 pgoff++;
94 } while (size);
95
96 return 0;
97
98}
99
Randy Dunlap8d634942007-10-16 23:31:29 -0700100/**
101 * sys_remap_file_pages - remap arbitrary pages of an existing VM_SHARED vma
Linus Torvalds1da177e2005-04-16 15:20:36 -0700102 * @start: start of the remapped virtual memory range
103 * @size: size of the remapped virtual memory range
Randy Dunlap8d634942007-10-16 23:31:29 -0700104 * @prot: new protection bits of the range (see NOTE)
105 * @pgoff: to-be-mapped page of the backing store file
Linus Torvalds1da177e2005-04-16 15:20:36 -0700106 * @flags: 0 or MAP_NONBLOCKED - the later will cause no IO.
107 *
Randy Dunlap8d634942007-10-16 23:31:29 -0700108 * sys_remap_file_pages remaps arbitrary pages of an existing VM_SHARED vma
109 * (shared backing store file).
110 *
111 * This syscall works purely via pagetables, so it's the most efficient
Linus Torvalds1da177e2005-04-16 15:20:36 -0700112 * way to map the same (large) file into a given virtual window. Unlike
113 * mmap()/mremap() it does not create any new vmas. The new mappings are
114 * also safe across swapout.
115 *
Randy Dunlap8d634942007-10-16 23:31:29 -0700116 * NOTE: the 'prot' parameter right now is ignored (but must be zero),
117 * and the vma's default protection is used. Arbitrary protections
118 * might be implemented in the future.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700119 */
120asmlinkage long sys_remap_file_pages(unsigned long start, unsigned long size,
Randy Dunlap8d634942007-10-16 23:31:29 -0700121 unsigned long prot, unsigned long pgoff, unsigned long flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700122{
123 struct mm_struct *mm = current->mm;
124 struct address_space *mapping;
125 unsigned long end = start + size;
126 struct vm_area_struct *vma;
127 int err = -EINVAL;
128 int has_write_lock = 0;
129
Randy Dunlap8d634942007-10-16 23:31:29 -0700130 if (prot)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700131 return err;
132 /*
133 * Sanitize the syscall parameters:
134 */
135 start = start & PAGE_MASK;
136 size = size & PAGE_MASK;
137
138 /* Does the address range wrap, or is the span zero-sized? */
139 if (start + size <= start)
140 return err;
141
142 /* Can we represent this offset inside this architecture's pte's? */
143#if PTE_FILE_MAX_BITS < BITS_PER_LONG
144 if (pgoff + (size >> PAGE_SHIFT) >= (1UL << PTE_FILE_MAX_BITS))
145 return err;
146#endif
147
148 /* We need down_write() to change vma->vm_flags. */
149 down_read(&mm->mmap_sem);
150 retry:
151 vma = find_vma(mm, start);
152
153 /*
154 * Make sure the vma is shared, that it supports prefaulting,
155 * and that the remapped range is valid and fully within
156 * the single existing vma. vm_private_data is used as a
Hugh Dickins101d2be2005-11-21 21:32:16 -0800157 * swapout cursor in a VM_NONLINEAR vma.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700158 */
Nick Piggin54cb8822007-07-19 01:46:59 -0700159 if (!vma || !(vma->vm_flags & VM_SHARED))
160 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700161
Nick Piggin54cb8822007-07-19 01:46:59 -0700162 if (vma->vm_private_data && !(vma->vm_flags & VM_NONLINEAR))
163 goto out;
164
Yan Zhengdd204d62007-10-08 10:05:48 -0700165 if (!(vma->vm_flags & VM_CAN_NONLINEAR))
Nick Piggin54cb8822007-07-19 01:46:59 -0700166 goto out;
167
168 if (end <= start || start < vma->vm_start || end > vma->vm_end)
169 goto out;
170
171 /* Must set VM_NONLINEAR before any pages are populated. */
172 if (!(vma->vm_flags & VM_NONLINEAR)) {
173 /* Don't need a nonlinear mapping, exit success */
174 if (pgoff == linear_page_index(vma, start)) {
175 err = 0;
176 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700177 }
178
Nick Piggin54cb8822007-07-19 01:46:59 -0700179 if (!has_write_lock) {
180 up_read(&mm->mmap_sem);
181 down_write(&mm->mmap_sem);
182 has_write_lock = 1;
183 goto retry;
184 }
185 mapping = vma->vm_file->f_mapping;
Miklos Szeredi3ee6daf2007-07-19 01:47:24 -0700186 /*
187 * page_mkclean doesn't work on nonlinear vmas, so if
188 * dirty pages need to be accounted, emulate with linear
189 * vmas.
190 */
191 if (mapping_cap_account_dirty(mapping)) {
192 unsigned long addr;
Oleg Nesterov8a459e42008-02-04 22:27:18 -0800193 struct file *file = vma->vm_file;
Miklos Szeredi3ee6daf2007-07-19 01:47:24 -0700194
195 flags &= MAP_NONBLOCK;
Oleg Nesterov8a459e42008-02-04 22:27:18 -0800196 get_file(file);
197 addr = mmap_region(file, start, size,
Miklos Szeredi3ee6daf2007-07-19 01:47:24 -0700198 flags, vma->vm_flags, pgoff, 1);
Oleg Nesterov8a459e42008-02-04 22:27:18 -0800199 fput(file);
Miklos Szeredi3ee6daf2007-07-19 01:47:24 -0700200 if (IS_ERR_VALUE(addr)) {
201 err = addr;
202 } else {
203 BUG_ON(addr != start);
204 err = 0;
205 }
206 goto out;
207 }
Nick Piggin54cb8822007-07-19 01:46:59 -0700208 spin_lock(&mapping->i_mmap_lock);
209 flush_dcache_mmap_lock(mapping);
210 vma->vm_flags |= VM_NONLINEAR;
211 vma_prio_tree_remove(vma, &mapping->i_mmap);
212 vma_nonlinear_insert(vma, &mapping->i_mmap_nonlinear);
213 flush_dcache_mmap_unlock(mapping);
214 spin_unlock(&mapping->i_mmap_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700215 }
Nick Piggin54cb8822007-07-19 01:46:59 -0700216
Nick Piggind0217ac2007-07-19 01:47:03 -0700217 err = populate_range(mm, vma, start, size, pgoff);
218 if (!err && !(flags & MAP_NONBLOCK)) {
219 if (unlikely(has_write_lock)) {
220 downgrade_write(&mm->mmap_sem);
221 has_write_lock = 0;
Nick Piggin54cb8822007-07-19 01:46:59 -0700222 }
Nick Piggind0217ac2007-07-19 01:47:03 -0700223 make_pages_present(start, start+size);
224 }
Nick Piggin54cb8822007-07-19 01:46:59 -0700225
226 /*
227 * We can't clear VM_NONLINEAR because we'd have to do
228 * it after ->populate completes, and that would prevent
229 * downgrading the lock. (Locks can't be upgraded).
230 */
231
232out:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700233 if (likely(!has_write_lock))
234 up_read(&mm->mmap_sem);
235 else
236 up_write(&mm->mmap_sem);
237
238 return err;
239}
240