blob: 72b8fa36143328f728a44de1b5cd53b629653a0a [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
Hugh Dickins88784392014-03-18 17:38:38 -070026static int mm_counter(struct page *page)
27{
28 return PageAnon(page) ? MM_ANONPAGES : MM_FILEPAGES;
29}
30
Nick Piggind0217ac2007-07-19 01:47:03 -070031static void zap_pte(struct mm_struct *mm, struct vm_area_struct *vma,
Linus Torvalds1da177e2005-04-16 15:20:36 -070032 unsigned long addr, pte_t *ptep)
33{
34 pte_t pte = *ptep;
Hugh Dickins88784392014-03-18 17:38:38 -070035 struct page *page;
36 swp_entry_t entry;
Linus Torvalds1da177e2005-04-16 15:20:36 -070037
Linus Torvalds1da177e2005-04-16 15:20:36 -070038 if (pte_present(pte)) {
Linus Torvalds6aab3412005-11-28 14:34:23 -080039 flush_cache_page(vma, addr, pte_pfn(pte));
Linus Torvalds1da177e2005-04-16 15:20:36 -070040 pte = ptep_clear_flush(vma, addr, ptep);
Linus Torvalds6aab3412005-11-28 14:34:23 -080041 page = vm_normal_page(vma, addr, pte);
42 if (page) {
43 if (pte_dirty(pte))
44 set_page_dirty(page);
Hugh Dickins88784392014-03-18 17:38:38 -070045 update_hiwater_rss(mm);
46 dec_mm_counter(mm, mm_counter(page));
Hugh Dickinsedc315f2009-01-06 14:40:11 -080047 page_remove_rmap(page);
Linus Torvalds6aab3412005-11-28 14:34:23 -080048 page_cache_release(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -070049 }
Hugh Dickins88784392014-03-18 17:38:38 -070050 } else { /* zap_pte() is not called when pte_none() */
51 if (!pte_file(pte)) {
52 update_hiwater_rss(mm);
53 entry = pte_to_swp_entry(pte);
54 if (non_swap_entry(entry)) {
55 if (is_migration_entry(entry)) {
56 page = migration_entry_to_page(entry);
57 dec_mm_counter(mm, mm_counter(page));
58 }
59 } else {
60 free_swap_and_cache(entry);
61 dec_mm_counter(mm, MM_SWAPENTS);
62 }
63 }
Zachary Amsden9888a1c2006-09-30 23:29:31 -070064 pte_clear_not_present_full(mm, addr, ptep, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -070065 }
66}
67
68/*
Linus Torvalds1da177e2005-04-16 15:20:36 -070069 * Install a file pte to a given virtual memory address, release any
70 * previously existing mapping.
71 */
Nick Piggind0217ac2007-07-19 01:47:03 -070072static int install_file_pte(struct mm_struct *mm, struct vm_area_struct *vma,
Linus Torvalds1da177e2005-04-16 15:20:36 -070073 unsigned long addr, unsigned long pgoff, pgprot_t prot)
74{
75 int err = -ENOMEM;
Cyrill Gorcunov41bb3472013-08-13 16:00:51 -070076 pte_t *pte, ptfile;
Hugh Dickinsc74df322005-10-29 18:16:23 -070077 spinlock_t *ptl;
Linus Torvalds1da177e2005-04-16 15:20:36 -070078
Linus Torvaldsc9cfcdd2005-11-29 14:03:14 -080079 pte = get_locked_pte(mm, addr, &ptl);
Linus Torvalds1da177e2005-04-16 15:20:36 -070080 if (!pte)
Hugh Dickinsc74df322005-10-29 18:16:23 -070081 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -070082
Cyrill Gorcunov41bb3472013-08-13 16:00:51 -070083 ptfile = pgoff_to_pte(pgoff);
Linus Torvalds1da177e2005-04-16 15:20:36 -070084
Cyrill Gorcunov0bf07332014-06-04 16:06:41 -070085 if (!pte_none(*pte))
Cyrill Gorcunov41bb3472013-08-13 16:00:51 -070086 zap_pte(mm, vma, addr, pte);
Cyrill Gorcunov41bb3472013-08-13 16:00:51 -070087
Cyrill Gorcunov0bf07332014-06-04 16:06:41 -070088 set_pte_at(mm, addr, pte, pte_file_mksoft_dirty(ptfile));
Hugh Dickins668e0d82006-06-23 02:03:45 -070089 /*
90 * We don't need to run update_mmu_cache() here because the "file pte"
91 * being installed by install_file_pte() is not a real pte - it's a
92 * non-present entry (like a swap entry), noting what file offset should
93 * be mapped there when there's a fault (in a non-linear vma where
94 * that's not obvious).
95 */
Hugh Dickinsc74df322005-10-29 18:16:23 -070096 pte_unmap_unlock(pte, ptl);
97 err = 0;
98out:
Linus Torvalds1da177e2005-04-16 15:20:36 -070099 return err;
100}
101
Konstantin Khlebnikov0b173bc2012-10-08 16:28:46 -0700102int generic_file_remap_pages(struct vm_area_struct *vma, unsigned long addr,
103 unsigned long size, pgoff_t pgoff)
Nick Piggin54cb8822007-07-19 01:46:59 -0700104{
Konstantin Khlebnikov0b173bc2012-10-08 16:28:46 -0700105 struct mm_struct *mm = vma->vm_mm;
Nick Piggin54cb8822007-07-19 01:46:59 -0700106 int err;
107
108 do {
109 err = install_file_pte(mm, vma, addr, pgoff, vma->vm_page_prot);
110 if (err)
111 return err;
112
113 size -= PAGE_SIZE;
114 addr += PAGE_SIZE;
115 pgoff++;
116 } while (size);
117
Konstantin Khlebnikov0b173bc2012-10-08 16:28:46 -0700118 return 0;
Nick Piggin54cb8822007-07-19 01:46:59 -0700119}
Konstantin Khlebnikov0b173bc2012-10-08 16:28:46 -0700120EXPORT_SYMBOL(generic_file_remap_pages);
Nick Piggin54cb8822007-07-19 01:46:59 -0700121
Randy Dunlap8d634942007-10-16 23:31:29 -0700122/**
123 * sys_remap_file_pages - remap arbitrary pages of an existing VM_SHARED vma
Linus Torvalds1da177e2005-04-16 15:20:36 -0700124 * @start: start of the remapped virtual memory range
125 * @size: size of the remapped virtual memory range
Randy Dunlap8d634942007-10-16 23:31:29 -0700126 * @prot: new protection bits of the range (see NOTE)
127 * @pgoff: to-be-mapped page of the backing store file
Linus Torvalds1da177e2005-04-16 15:20:36 -0700128 * @flags: 0 or MAP_NONBLOCKED - the later will cause no IO.
129 *
Randy Dunlap8d634942007-10-16 23:31:29 -0700130 * sys_remap_file_pages remaps arbitrary pages of an existing VM_SHARED vma
131 * (shared backing store file).
132 *
133 * This syscall works purely via pagetables, so it's the most efficient
Linus Torvalds1da177e2005-04-16 15:20:36 -0700134 * way to map the same (large) file into a given virtual window. Unlike
135 * mmap()/mremap() it does not create any new vmas. The new mappings are
136 * also safe across swapout.
137 *
Randy Dunlap76824862008-03-19 17:00:40 -0700138 * NOTE: the @prot parameter right now is ignored (but must be zero),
Randy Dunlap8d634942007-10-16 23:31:29 -0700139 * and the vma's default protection is used. Arbitrary protections
140 * might be implemented in the future.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700141 */
Heiko Carstens6a6160a2009-01-14 14:14:15 +0100142SYSCALL_DEFINE5(remap_file_pages, unsigned long, start, unsigned long, size,
143 unsigned long, prot, unsigned long, pgoff, unsigned long, flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700144{
145 struct mm_struct *mm = current->mm;
146 struct address_space *mapping;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700147 struct vm_area_struct *vma;
148 int err = -EINVAL;
149 int has_write_lock = 0;
Michel Lespinassea2362d22013-03-14 16:50:02 -0700150 vm_flags_t vm_flags = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700151
Kirill A. Shutemov33041a02014-06-06 14:38:23 -0700152 pr_warn_once("%s (%d) uses deprecated remap_file_pages() syscall. "
153 "See Documentation/vm/remap_file_pages.txt.\n",
154 current->comm, current->pid);
155
Randy Dunlap8d634942007-10-16 23:31:29 -0700156 if (prot)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700157 return err;
158 /*
159 * Sanitize the syscall parameters:
160 */
161 start = start & PAGE_MASK;
162 size = size & PAGE_MASK;
163
164 /* Does the address range wrap, or is the span zero-sized? */
165 if (start + size <= start)
166 return err;
167
Larry Woodman5ec10552010-09-24 12:04:48 -0400168 /* Does pgoff wrap? */
169 if (pgoff + (size >> PAGE_SHIFT) < pgoff)
170 return err;
171
Linus Torvalds1da177e2005-04-16 15:20:36 -0700172 /* Can we represent this offset inside this architecture's pte's? */
173#if PTE_FILE_MAX_BITS < BITS_PER_LONG
174 if (pgoff + (size >> PAGE_SHIFT) >= (1UL << PTE_FILE_MAX_BITS))
175 return err;
176#endif
177
178 /* We need down_write() to change vma->vm_flags. */
179 down_read(&mm->mmap_sem);
180 retry:
181 vma = find_vma(mm, start);
182
183 /*
184 * Make sure the vma is shared, that it supports prefaulting,
185 * and that the remapped range is valid and fully within
Michel Lespinasse940e7da2013-02-22 16:32:36 -0800186 * the single existing vma.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700187 */
Michel Lespinassea2362d22013-03-14 16:50:02 -0700188 if (!vma || !(vma->vm_flags & VM_SHARED))
Nick Piggin54cb8822007-07-19 01:46:59 -0700189 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700190
Linus Torvaldsdeb521c2012-10-19 13:37:57 -0700191 if (!vma->vm_ops || !vma->vm_ops->remap_pages)
Nick Piggin54cb8822007-07-19 01:46:59 -0700192 goto out;
193
Linus Torvaldse92b05d2010-09-24 14:13:57 -0700194 if (start < vma->vm_start || start + size > vma->vm_end)
Nick Piggin54cb8822007-07-19 01:46:59 -0700195 goto out;
196
197 /* Must set VM_NONLINEAR before any pages are populated. */
198 if (!(vma->vm_flags & VM_NONLINEAR)) {
Michel Lespinasse940e7da2013-02-22 16:32:36 -0800199 /*
200 * vm_private_data is used as a swapout cursor
201 * in a VM_NONLINEAR vma.
202 */
203 if (vma->vm_private_data)
204 goto out;
205
Nick Piggin54cb8822007-07-19 01:46:59 -0700206 /* Don't need a nonlinear mapping, exit success */
207 if (pgoff == linear_page_index(vma, start)) {
208 err = 0;
209 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700210 }
211
Nick Piggin54cb8822007-07-19 01:46:59 -0700212 if (!has_write_lock) {
Michel Lespinasse940e7da2013-02-22 16:32:36 -0800213get_write_lock:
Nick Piggin54cb8822007-07-19 01:46:59 -0700214 up_read(&mm->mmap_sem);
215 down_write(&mm->mmap_sem);
216 has_write_lock = 1;
217 goto retry;
218 }
219 mapping = vma->vm_file->f_mapping;
Miklos Szeredi3ee6daf2007-07-19 01:47:24 -0700220 /*
221 * page_mkclean doesn't work on nonlinear vmas, so if
222 * dirty pages need to be accounted, emulate with linear
223 * vmas.
224 */
225 if (mapping_cap_account_dirty(mapping)) {
226 unsigned long addr;
Al Virocb0942b2012-08-27 14:48:26 -0400227 struct file *file = get_file(vma->vm_file);
Rik van Riel4eb91982014-01-02 12:58:46 -0800228 /* mmap_region may free vma; grab the info now */
229 vm_flags = vma->vm_flags;
Miklos Szeredi3ee6daf2007-07-19 01:47:24 -0700230
Rik van Riel4eb91982014-01-02 12:58:46 -0800231 addr = mmap_region(file, start, size, vm_flags, pgoff);
Oleg Nesterov8a459e42008-02-04 22:27:18 -0800232 fput(file);
Miklos Szeredi3ee6daf2007-07-19 01:47:24 -0700233 if (IS_ERR_VALUE(addr)) {
234 err = addr;
235 } else {
236 BUG_ON(addr != start);
237 err = 0;
238 }
Rik van Riel4eb91982014-01-02 12:58:46 -0800239 goto out_freed;
Miklos Szeredi3ee6daf2007-07-19 01:47:24 -0700240 }
Peter Zijlstra3d48ae42011-05-24 17:12:06 -0700241 mutex_lock(&mapping->i_mmap_mutex);
Nick Piggin54cb8822007-07-19 01:46:59 -0700242 flush_dcache_mmap_lock(mapping);
243 vma->vm_flags |= VM_NONLINEAR;
Michel Lespinasse6b2dbba2012-10-08 16:31:25 -0700244 vma_interval_tree_remove(vma, &mapping->i_mmap);
Nick Piggin54cb8822007-07-19 01:46:59 -0700245 vma_nonlinear_insert(vma, &mapping->i_mmap_nonlinear);
246 flush_dcache_mmap_unlock(mapping);
Peter Zijlstra3d48ae42011-05-24 17:12:06 -0700247 mutex_unlock(&mapping->i_mmap_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700248 }
Nick Piggin54cb8822007-07-19 01:46:59 -0700249
Rik van Rielba470de2008-10-18 20:26:50 -0700250 if (vma->vm_flags & VM_LOCKED) {
251 /*
252 * drop PG_Mlocked flag for over-mapped range
253 */
Michel Lespinasse940e7da2013-02-22 16:32:36 -0800254 if (!has_write_lock)
255 goto get_write_lock;
Michel Lespinassea1ea9542013-02-22 16:32:38 -0800256 vm_flags = vma->vm_flags;
Rik van Rielba470de2008-10-18 20:26:50 -0700257 munlock_vma_pages_range(vma, start, start + size);
Michel Lespinassea1ea9542013-02-22 16:32:38 -0800258 vma->vm_flags = vm_flags;
Rik van Rielba470de2008-10-18 20:26:50 -0700259 }
260
Andrea Arcangelicddb8a52008-07-28 15:46:29 -0700261 mmu_notifier_invalidate_range_start(mm, start, start + size);
Konstantin Khlebnikov0b173bc2012-10-08 16:28:46 -0700262 err = vma->vm_ops->remap_pages(vma, start, size, pgoff);
Andrea Arcangelicddb8a52008-07-28 15:46:29 -0700263 mmu_notifier_invalidate_range_end(mm, start, start + size);
Nick Piggin54cb8822007-07-19 01:46:59 -0700264
265 /*
266 * We can't clear VM_NONLINEAR because we'd have to do
267 * it after ->populate completes, and that would prevent
268 * downgrading the lock. (Locks can't be upgraded).
269 */
270
271out:
Andrew Morton6d7825b2013-03-13 14:59:43 -0700272 if (vma)
273 vm_flags = vma->vm_flags;
Rik van Riel4eb91982014-01-02 12:58:46 -0800274out_freed:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700275 if (likely(!has_write_lock))
276 up_read(&mm->mmap_sem);
277 else
278 up_write(&mm->mmap_sem);
Michel Lespinassea1ea9542013-02-22 16:32:38 -0800279 if (!err && ((vm_flags & VM_LOCKED) || !(flags & MAP_NONBLOCK)))
280 mm_populate(start, size);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700281
282 return err;
283}