blob: 4723ac8d2fc200d5990f830b6f4c8c0fd2a9bbcf [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 */
Konstantin Khlebnikov0b173bc2012-10-08 16:28:46 -07008#include <linux/export.h>
Alexey Dobriyan4af3c9c2007-10-16 23:29:23 -07009#include <linux/backing-dev.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070010#include <linux/mm.h>
11#include <linux/swap.h>
12#include <linux/file.h>
13#include <linux/mman.h>
14#include <linux/pagemap.h>
15#include <linux/swapops.h>
16#include <linux/rmap.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070017#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);
KAMEZAWA Hiroyukid559db02010-03-05 13:41:39 -080043 dec_mm_counter(mm, MM_FILEPAGES);
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
Konstantin Khlebnikov0b173bc2012-10-08 16:28:46 -070084int generic_file_remap_pages(struct vm_area_struct *vma, unsigned long addr,
85 unsigned long size, pgoff_t pgoff)
Nick Piggin54cb8822007-07-19 01:46:59 -070086{
Konstantin Khlebnikov0b173bc2012-10-08 16:28:46 -070087 struct mm_struct *mm = vma->vm_mm;
Nick Piggin54cb8822007-07-19 01:46:59 -070088 int err;
89
90 do {
91 err = install_file_pte(mm, vma, addr, pgoff, vma->vm_page_prot);
92 if (err)
93 return err;
94
95 size -= PAGE_SIZE;
96 addr += PAGE_SIZE;
97 pgoff++;
98 } while (size);
99
Konstantin Khlebnikov0b173bc2012-10-08 16:28:46 -0700100 return 0;
Nick Piggin54cb8822007-07-19 01:46:59 -0700101}
Konstantin Khlebnikov0b173bc2012-10-08 16:28:46 -0700102EXPORT_SYMBOL(generic_file_remap_pages);
Nick Piggin54cb8822007-07-19 01:46:59 -0700103
Randy Dunlap8d634942007-10-16 23:31:29 -0700104/**
105 * sys_remap_file_pages - remap arbitrary pages of an existing VM_SHARED vma
Linus Torvalds1da177e2005-04-16 15:20:36 -0700106 * @start: start of the remapped virtual memory range
107 * @size: size of the remapped virtual memory range
Randy Dunlap8d634942007-10-16 23:31:29 -0700108 * @prot: new protection bits of the range (see NOTE)
109 * @pgoff: to-be-mapped page of the backing store file
Linus Torvalds1da177e2005-04-16 15:20:36 -0700110 * @flags: 0 or MAP_NONBLOCKED - the later will cause no IO.
111 *
Randy Dunlap8d634942007-10-16 23:31:29 -0700112 * sys_remap_file_pages remaps arbitrary pages of an existing VM_SHARED vma
113 * (shared backing store file).
114 *
115 * This syscall works purely via pagetables, so it's the most efficient
Linus Torvalds1da177e2005-04-16 15:20:36 -0700116 * way to map the same (large) file into a given virtual window. Unlike
117 * mmap()/mremap() it does not create any new vmas. The new mappings are
118 * also safe across swapout.
119 *
Randy Dunlap76824862008-03-19 17:00:40 -0700120 * NOTE: the @prot parameter right now is ignored (but must be zero),
Randy Dunlap8d634942007-10-16 23:31:29 -0700121 * and the vma's default protection is used. Arbitrary protections
122 * might be implemented in the future.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700123 */
Heiko Carstens6a6160a2009-01-14 14:14:15 +0100124SYSCALL_DEFINE5(remap_file_pages, unsigned long, start, unsigned long, size,
125 unsigned long, prot, unsigned long, pgoff, unsigned long, flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700126{
127 struct mm_struct *mm = current->mm;
128 struct address_space *mapping;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700129 struct vm_area_struct *vma;
130 int err = -EINVAL;
131 int has_write_lock = 0;
Michel Lespinassea2362d22013-03-14 16:50:02 -0700132 vm_flags_t vm_flags = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700133
Randy Dunlap8d634942007-10-16 23:31:29 -0700134 if (prot)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700135 return err;
136 /*
137 * Sanitize the syscall parameters:
138 */
139 start = start & PAGE_MASK;
140 size = size & PAGE_MASK;
141
142 /* Does the address range wrap, or is the span zero-sized? */
143 if (start + size <= start)
144 return err;
145
Larry Woodman5ec10552010-09-24 12:04:48 -0400146 /* Does pgoff wrap? */
147 if (pgoff + (size >> PAGE_SHIFT) < pgoff)
148 return err;
149
Linus Torvalds1da177e2005-04-16 15:20:36 -0700150 /* Can we represent this offset inside this architecture's pte's? */
151#if PTE_FILE_MAX_BITS < BITS_PER_LONG
152 if (pgoff + (size >> PAGE_SHIFT) >= (1UL << PTE_FILE_MAX_BITS))
153 return err;
154#endif
155
156 /* We need down_write() to change vma->vm_flags. */
157 down_read(&mm->mmap_sem);
158 retry:
159 vma = find_vma(mm, start);
160
161 /*
162 * Make sure the vma is shared, that it supports prefaulting,
163 * and that the remapped range is valid and fully within
Michel Lespinasse940e7da2013-02-22 16:32:36 -0800164 * the single existing vma.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700165 */
Michel Lespinassea2362d22013-03-14 16:50:02 -0700166 if (!vma || !(vma->vm_flags & VM_SHARED))
Nick Piggin54cb8822007-07-19 01:46:59 -0700167 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700168
Linus Torvaldsdeb521c2012-10-19 13:37:57 -0700169 if (!vma->vm_ops || !vma->vm_ops->remap_pages)
Nick Piggin54cb8822007-07-19 01:46:59 -0700170 goto out;
171
Linus Torvaldse92b05d2010-09-24 14:13:57 -0700172 if (start < vma->vm_start || start + size > vma->vm_end)
Nick Piggin54cb8822007-07-19 01:46:59 -0700173 goto out;
174
175 /* Must set VM_NONLINEAR before any pages are populated. */
176 if (!(vma->vm_flags & VM_NONLINEAR)) {
Michel Lespinasse940e7da2013-02-22 16:32:36 -0800177 /*
178 * vm_private_data is used as a swapout cursor
179 * in a VM_NONLINEAR vma.
180 */
181 if (vma->vm_private_data)
182 goto out;
183
Nick Piggin54cb8822007-07-19 01:46:59 -0700184 /* Don't need a nonlinear mapping, exit success */
185 if (pgoff == linear_page_index(vma, start)) {
186 err = 0;
187 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700188 }
189
Nick Piggin54cb8822007-07-19 01:46:59 -0700190 if (!has_write_lock) {
Michel Lespinasse940e7da2013-02-22 16:32:36 -0800191get_write_lock:
Nick Piggin54cb8822007-07-19 01:46:59 -0700192 up_read(&mm->mmap_sem);
193 down_write(&mm->mmap_sem);
194 has_write_lock = 1;
195 goto retry;
196 }
197 mapping = vma->vm_file->f_mapping;
Miklos Szeredi3ee6daf2007-07-19 01:47:24 -0700198 /*
199 * page_mkclean doesn't work on nonlinear vmas, so if
200 * dirty pages need to be accounted, emulate with linear
201 * vmas.
202 */
203 if (mapping_cap_account_dirty(mapping)) {
204 unsigned long addr;
Al Virocb0942b2012-08-27 14:48:26 -0400205 struct file *file = get_file(vma->vm_file);
Miklos Szeredi3ee6daf2007-07-19 01:47:24 -0700206
Michel Lespinasse18693052013-02-22 16:32:46 -0800207 vm_flags = vma->vm_flags;
208 if (!(flags & MAP_NONBLOCK))
209 vm_flags |= VM_POPULATE;
210 addr = mmap_region(file, start, size, vm_flags, pgoff);
Oleg Nesterov8a459e42008-02-04 22:27:18 -0800211 fput(file);
Miklos Szeredi3ee6daf2007-07-19 01:47:24 -0700212 if (IS_ERR_VALUE(addr)) {
213 err = addr;
214 } else {
215 BUG_ON(addr != start);
216 err = 0;
217 }
218 goto out;
219 }
Peter Zijlstra3d48ae42011-05-24 17:12:06 -0700220 mutex_lock(&mapping->i_mmap_mutex);
Nick Piggin54cb8822007-07-19 01:46:59 -0700221 flush_dcache_mmap_lock(mapping);
222 vma->vm_flags |= VM_NONLINEAR;
Michel Lespinasse6b2dbba2012-10-08 16:31:25 -0700223 vma_interval_tree_remove(vma, &mapping->i_mmap);
Nick Piggin54cb8822007-07-19 01:46:59 -0700224 vma_nonlinear_insert(vma, &mapping->i_mmap_nonlinear);
225 flush_dcache_mmap_unlock(mapping);
Peter Zijlstra3d48ae42011-05-24 17:12:06 -0700226 mutex_unlock(&mapping->i_mmap_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700227 }
Nick Piggin54cb8822007-07-19 01:46:59 -0700228
Michel Lespinasse18693052013-02-22 16:32:46 -0800229 if (!(flags & MAP_NONBLOCK) && !(vma->vm_flags & VM_POPULATE)) {
230 if (!has_write_lock)
231 goto get_write_lock;
232 vma->vm_flags |= VM_POPULATE;
233 }
234
Rik van Rielba470de2008-10-18 20:26:50 -0700235 if (vma->vm_flags & VM_LOCKED) {
236 /*
237 * drop PG_Mlocked flag for over-mapped range
238 */
Michel Lespinasse940e7da2013-02-22 16:32:36 -0800239 if (!has_write_lock)
240 goto get_write_lock;
Michel Lespinassea1ea9542013-02-22 16:32:38 -0800241 vm_flags = vma->vm_flags;
Rik van Rielba470de2008-10-18 20:26:50 -0700242 munlock_vma_pages_range(vma, start, start + size);
Michel Lespinassea1ea9542013-02-22 16:32:38 -0800243 vma->vm_flags = vm_flags;
Rik van Rielba470de2008-10-18 20:26:50 -0700244 }
245
Andrea Arcangelicddb8a52008-07-28 15:46:29 -0700246 mmu_notifier_invalidate_range_start(mm, start, start + size);
Konstantin Khlebnikov0b173bc2012-10-08 16:28:46 -0700247 err = vma->vm_ops->remap_pages(vma, start, size, pgoff);
Andrea Arcangelicddb8a52008-07-28 15:46:29 -0700248 mmu_notifier_invalidate_range_end(mm, start, start + size);
Nick Piggin54cb8822007-07-19 01:46:59 -0700249
250 /*
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).
254 */
255
256out:
Andrew Morton6d7825b2013-03-13 14:59:43 -0700257 if (vma)
258 vm_flags = vma->vm_flags;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700259 if (likely(!has_write_lock))
260 up_read(&mm->mmap_sem);
261 else
262 up_write(&mm->mmap_sem);
Michel Lespinassea1ea9542013-02-22 16:32:38 -0800263 if (!err && ((vm_flags & VM_LOCKED) || !(flags & MAP_NONBLOCK)))
264 mm_populate(start, size);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700265
266 return err;
267}