blob: b6ec85abbb39cc22dac7e9c309953c60a22e921c [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>
Andrea Arcangelicddb8a52008-07-28 15:46:29 -070018#include <linux/mmu_notifier.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070019
20#include <asm/mmu_context.h>
21#include <asm/cacheflush.h>
22#include <asm/tlbflush.h>
23
Rik van Rielba470de2008-10-18 20:26:50 -070024#include "internal.h"
25
Nick Piggind0217ac2007-07-19 01:47:03 -070026static void zap_pte(struct mm_struct *mm, struct vm_area_struct *vma,
Linus Torvalds1da177e2005-04-16 15:20:36 -070027 unsigned long addr, pte_t *ptep)
28{
29 pte_t pte = *ptep;
30
Linus Torvalds1da177e2005-04-16 15:20:36 -070031 if (pte_present(pte)) {
Nick Piggind0217ac2007-07-19 01:47:03 -070032 struct page *page;
33
Linus Torvalds6aab3412005-11-28 14:34:23 -080034 flush_cache_page(vma, addr, pte_pfn(pte));
Linus Torvalds1da177e2005-04-16 15:20:36 -070035 pte = ptep_clear_flush(vma, addr, ptep);
Linus Torvalds6aab3412005-11-28 14:34:23 -080036 page = vm_normal_page(vma, addr, pte);
37 if (page) {
38 if (pte_dirty(pte))
39 set_page_dirty(page);
Hugh Dickinsedc315f2009-01-06 14:40:11 -080040 page_remove_rmap(page);
Linus Torvalds6aab3412005-11-28 14:34:23 -080041 page_cache_release(page);
Nick Piggind0217ac2007-07-19 01:47:03 -070042 update_hiwater_rss(mm);
43 dec_mm_counter(mm, file_rss);
Linus Torvalds1da177e2005-04-16 15:20:36 -070044 }
45 } else {
46 if (!pte_file(pte))
47 free_swap_and_cache(pte_to_swp_entry(pte));
Zachary Amsden9888a1c2006-09-30 23:29:31 -070048 pte_clear_not_present_full(mm, addr, ptep, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -070049 }
50}
51
52/*
Linus Torvalds1da177e2005-04-16 15:20:36 -070053 * Install a file pte to a given virtual memory address, release any
54 * previously existing mapping.
55 */
Nick Piggind0217ac2007-07-19 01:47:03 -070056static int install_file_pte(struct mm_struct *mm, struct vm_area_struct *vma,
Linus Torvalds1da177e2005-04-16 15:20:36 -070057 unsigned long addr, unsigned long pgoff, pgprot_t prot)
58{
59 int err = -ENOMEM;
60 pte_t *pte;
Hugh Dickinsc74df322005-10-29 18:16:23 -070061 spinlock_t *ptl;
Linus Torvalds1da177e2005-04-16 15:20:36 -070062
Linus Torvaldsc9cfcdd2005-11-29 14:03:14 -080063 pte = get_locked_pte(mm, addr, &ptl);
Linus Torvalds1da177e2005-04-16 15:20:36 -070064 if (!pte)
Hugh Dickinsc74df322005-10-29 18:16:23 -070065 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -070066
Nick Piggind0217ac2007-07-19 01:47:03 -070067 if (!pte_none(*pte))
68 zap_pte(mm, vma, addr, pte);
Linus Torvalds1da177e2005-04-16 15:20:36 -070069
70 set_pte_at(mm, addr, pte, pgoff_to_pte(pgoff));
Hugh Dickins668e0d82006-06-23 02:03:45 -070071 /*
72 * We don't need to run update_mmu_cache() here because the "file pte"
73 * being installed by install_file_pte() is not a real pte - it's a
74 * non-present entry (like a swap entry), noting what file offset should
75 * be mapped there when there's a fault (in a non-linear vma where
76 * that's not obvious).
77 */
Hugh Dickinsc74df322005-10-29 18:16:23 -070078 pte_unmap_unlock(pte, ptl);
79 err = 0;
80out:
Linus Torvalds1da177e2005-04-16 15:20:36 -070081 return err;
82}
83
Nick Piggin54cb8822007-07-19 01:46:59 -070084static int populate_range(struct mm_struct *mm, struct vm_area_struct *vma,
85 unsigned long addr, unsigned long size, pgoff_t pgoff)
86{
87 int err;
88
89 do {
90 err = install_file_pte(mm, vma, addr, pgoff, vma->vm_page_prot);
91 if (err)
92 return err;
93
94 size -= PAGE_SIZE;
95 addr += PAGE_SIZE;
96 pgoff++;
97 } while (size);
98
99 return 0;
100
101}
102
Randy Dunlap8d634942007-10-16 23:31:29 -0700103/**
104 * sys_remap_file_pages - remap arbitrary pages of an existing VM_SHARED vma
Linus Torvalds1da177e2005-04-16 15:20:36 -0700105 * @start: start of the remapped virtual memory range
106 * @size: size of the remapped virtual memory range
Randy Dunlap8d634942007-10-16 23:31:29 -0700107 * @prot: new protection bits of the range (see NOTE)
108 * @pgoff: to-be-mapped page of the backing store file
Linus Torvalds1da177e2005-04-16 15:20:36 -0700109 * @flags: 0 or MAP_NONBLOCKED - the later will cause no IO.
110 *
Randy Dunlap8d634942007-10-16 23:31:29 -0700111 * sys_remap_file_pages remaps arbitrary pages of an existing VM_SHARED vma
112 * (shared backing store file).
113 *
114 * This syscall works purely via pagetables, so it's the most efficient
Linus Torvalds1da177e2005-04-16 15:20:36 -0700115 * way to map the same (large) file into a given virtual window. Unlike
116 * mmap()/mremap() it does not create any new vmas. The new mappings are
117 * also safe across swapout.
118 *
Randy Dunlap76824862008-03-19 17:00:40 -0700119 * NOTE: the @prot parameter right now is ignored (but must be zero),
Randy Dunlap8d634942007-10-16 23:31:29 -0700120 * and the vma's default protection is used. Arbitrary protections
121 * might be implemented in the future.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700122 */
Heiko Carstens6a6160a2009-01-14 14:14:15 +0100123SYSCALL_DEFINE5(remap_file_pages, unsigned long, start, unsigned long, size,
124 unsigned long, prot, unsigned long, pgoff, unsigned long, flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700125{
126 struct mm_struct *mm = current->mm;
127 struct address_space *mapping;
128 unsigned long end = start + size;
129 struct vm_area_struct *vma;
130 int err = -EINVAL;
131 int has_write_lock = 0;
132
Randy Dunlap8d634942007-10-16 23:31:29 -0700133 if (prot)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700134 return err;
135 /*
136 * Sanitize the syscall parameters:
137 */
138 start = start & PAGE_MASK;
139 size = size & PAGE_MASK;
140
141 /* Does the address range wrap, or is the span zero-sized? */
142 if (start + size <= start)
143 return err;
144
145 /* Can we represent this offset inside this architecture's pte's? */
146#if PTE_FILE_MAX_BITS < BITS_PER_LONG
147 if (pgoff + (size >> PAGE_SHIFT) >= (1UL << PTE_FILE_MAX_BITS))
148 return err;
149#endif
150
151 /* We need down_write() to change vma->vm_flags. */
152 down_read(&mm->mmap_sem);
153 retry:
154 vma = find_vma(mm, start);
155
156 /*
157 * Make sure the vma is shared, that it supports prefaulting,
158 * and that the remapped range is valid and fully within
159 * the single existing vma. vm_private_data is used as a
Hugh Dickins101d2be2005-11-21 21:32:16 -0800160 * swapout cursor in a VM_NONLINEAR vma.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700161 */
Nick Piggin54cb8822007-07-19 01:46:59 -0700162 if (!vma || !(vma->vm_flags & VM_SHARED))
163 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700164
Nick Piggin54cb8822007-07-19 01:46:59 -0700165 if (vma->vm_private_data && !(vma->vm_flags & VM_NONLINEAR))
166 goto out;
167
Yan Zhengdd204d62007-10-08 10:05:48 -0700168 if (!(vma->vm_flags & VM_CAN_NONLINEAR))
Nick Piggin54cb8822007-07-19 01:46:59 -0700169 goto out;
170
171 if (end <= start || start < vma->vm_start || end > vma->vm_end)
172 goto out;
173
174 /* Must set VM_NONLINEAR before any pages are populated. */
175 if (!(vma->vm_flags & VM_NONLINEAR)) {
176 /* Don't need a nonlinear mapping, exit success */
177 if (pgoff == linear_page_index(vma, start)) {
178 err = 0;
179 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700180 }
181
Nick Piggin54cb8822007-07-19 01:46:59 -0700182 if (!has_write_lock) {
183 up_read(&mm->mmap_sem);
184 down_write(&mm->mmap_sem);
185 has_write_lock = 1;
186 goto retry;
187 }
188 mapping = vma->vm_file->f_mapping;
Miklos Szeredi3ee6daf2007-07-19 01:47:24 -0700189 /*
190 * page_mkclean doesn't work on nonlinear vmas, so if
191 * dirty pages need to be accounted, emulate with linear
192 * vmas.
193 */
194 if (mapping_cap_account_dirty(mapping)) {
195 unsigned long addr;
Oleg Nesterov8a459e42008-02-04 22:27:18 -0800196 struct file *file = vma->vm_file;
Miklos Szeredi3ee6daf2007-07-19 01:47:24 -0700197
198 flags &= MAP_NONBLOCK;
Oleg Nesterov8a459e42008-02-04 22:27:18 -0800199 get_file(file);
200 addr = mmap_region(file, start, size,
Mel Gorman5a6fe122009-02-10 14:02:27 +0000201 flags, vma->vm_flags, pgoff);
Oleg Nesterov8a459e42008-02-04 22:27:18 -0800202 fput(file);
Miklos Szeredi3ee6daf2007-07-19 01:47:24 -0700203 if (IS_ERR_VALUE(addr)) {
204 err = addr;
205 } else {
206 BUG_ON(addr != start);
207 err = 0;
208 }
209 goto out;
210 }
Nick Piggin54cb8822007-07-19 01:46:59 -0700211 spin_lock(&mapping->i_mmap_lock);
212 flush_dcache_mmap_lock(mapping);
213 vma->vm_flags |= VM_NONLINEAR;
214 vma_prio_tree_remove(vma, &mapping->i_mmap);
215 vma_nonlinear_insert(vma, &mapping->i_mmap_nonlinear);
216 flush_dcache_mmap_unlock(mapping);
217 spin_unlock(&mapping->i_mmap_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700218 }
Nick Piggin54cb8822007-07-19 01:46:59 -0700219
Rik van Rielba470de2008-10-18 20:26:50 -0700220 if (vma->vm_flags & VM_LOCKED) {
221 /*
222 * drop PG_Mlocked flag for over-mapped range
223 */
224 unsigned int saved_flags = vma->vm_flags;
225 munlock_vma_pages_range(vma, start, start + size);
226 vma->vm_flags = saved_flags;
227 }
228
Andrea Arcangelicddb8a52008-07-28 15:46:29 -0700229 mmu_notifier_invalidate_range_start(mm, start, start + size);
Nick Piggind0217ac2007-07-19 01:47:03 -0700230 err = populate_range(mm, vma, start, size, pgoff);
Andrea Arcangelicddb8a52008-07-28 15:46:29 -0700231 mmu_notifier_invalidate_range_end(mm, start, start + size);
Nick Piggind0217ac2007-07-19 01:47:03 -0700232 if (!err && !(flags & MAP_NONBLOCK)) {
Rik van Rielba470de2008-10-18 20:26:50 -0700233 if (vma->vm_flags & VM_LOCKED) {
234 /*
235 * might be mapping previously unmapped range of file
236 */
237 mlock_vma_pages_range(vma, start, start + size);
238 } else {
239 if (unlikely(has_write_lock)) {
240 downgrade_write(&mm->mmap_sem);
241 has_write_lock = 0;
242 }
243 make_pages_present(start, start+size);
Nick Piggin54cb8822007-07-19 01:46:59 -0700244 }
Nick Piggind0217ac2007-07-19 01:47:03 -0700245 }
Nick Piggin54cb8822007-07-19 01:46:59 -0700246
247 /*
248 * We can't clear VM_NONLINEAR because we'd have to do
249 * it after ->populate completes, and that would prevent
250 * downgrading the lock. (Locks can't be upgraded).
251 */
252
253out:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700254 if (likely(!has_write_lock))
255 up_read(&mm->mmap_sem);
256 else
257 up_write(&mm->mmap_sem);
258
259 return err;
260}