blob: fc37afe226e65881613e7cd2b5384c521983aef6 [file] [log] [blame]
Greg Kroah-Hartmanb2441312017-11-01 15:07:57 +01001// SPDX-License-Identifier: GPL-2.0
Linus Torvalds1da177e2005-04-16 15:20:36 -07002/*
3 * linux/mm/mincore.c
4 *
Linus Torvalds2f77d102006-12-16 09:44:32 -08005 * Copyright (C) 1994-2006 Linus Torvalds
Linus Torvalds1da177e2005-04-16 15:20:36 -07006 */
7
8/*
9 * The mincore() system call.
10 */
Linus Torvalds1da177e2005-04-16 15:20:36 -070011#include <linux/pagemap.h>
Tejun Heo5a0e3ad2010-03-24 17:04:11 +090012#include <linux/gfp.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070013#include <linux/mm.h>
14#include <linux/mman.h>
15#include <linux/syscalls.h>
Nick Piggin42da9cb2007-02-12 00:51:39 -080016#include <linux/swap.h>
17#include <linux/swapops.h>
Hugh Dickins3a4f8a02017-02-24 14:59:36 -080018#include <linux/shmem_fs.h>
Naoya Horiguchi4f16fc12009-12-14 17:59:58 -080019#include <linux/hugetlb.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070020
Linus Torvalds7c0f6ba2016-12-24 11:46:01 -080021#include <linux/uaccess.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070022#include <asm/pgtable.h>
23
Naoya Horiguchi1e25a272015-02-11 15:28:11 -080024static int mincore_hugetlb(pte_t *pte, unsigned long hmask, unsigned long addr,
25 unsigned long end, struct mm_walk *walk)
Johannes Weinerf4884012010-05-24 14:32:10 -070026{
27#ifdef CONFIG_HUGETLB_PAGE
Naoya Horiguchi1e25a272015-02-11 15:28:11 -080028 unsigned char present;
29 unsigned char *vec = walk->private;
Johannes Weinerf4884012010-05-24 14:32:10 -070030
Naoya Horiguchi1e25a272015-02-11 15:28:11 -080031 /*
32 * Hugepages under user process are always in RAM and never
33 * swapped out, but theoretically it needs to be checked.
34 */
35 present = pte && !huge_pte_none(huge_ptep_get(pte));
36 for (; addr != end; vec++, addr += PAGE_SIZE)
37 *vec = present;
38 walk->private = vec;
Johannes Weinerf4884012010-05-24 14:32:10 -070039#else
40 BUG();
41#endif
Naoya Horiguchi1e25a272015-02-11 15:28:11 -080042 return 0;
Johannes Weinerf4884012010-05-24 14:32:10 -070043}
44
Linus Torvalds1da177e2005-04-16 15:20:36 -070045/*
46 * Later we can get more picky about what "in core" means precisely.
47 * For now, simply check to see if the page is in the page cache,
48 * and is up to date; i.e. that no page-in operation would be required
49 * at this time if an application were to map and access this page.
50 */
Nick Piggin42da9cb2007-02-12 00:51:39 -080051static unsigned char mincore_page(struct address_space *mapping, pgoff_t pgoff)
Linus Torvalds1da177e2005-04-16 15:20:36 -070052{
53 unsigned char present = 0;
Nick Piggin42da9cb2007-02-12 00:51:39 -080054 struct page *page;
Linus Torvalds1da177e2005-04-16 15:20:36 -070055
Nick Piggin42da9cb2007-02-12 00:51:39 -080056 /*
57 * When tmpfs swaps out a page from a file, any process mapping that
58 * file will not get a swp_entry_t in its pte, but rather it is like
59 * any other file mapping (ie. marked !present and faulted in with
Nick Piggin3c18ddd2008-04-28 02:12:10 -070060 * tmpfs's .fault). So swapped out tmpfs mappings are tested here.
Nick Piggin42da9cb2007-02-12 00:51:39 -080061 */
Hugh Dickins31475dd2011-08-03 16:21:27 -070062#ifdef CONFIG_SWAP
Johannes Weiner0cd61442014-04-03 14:47:46 -070063 if (shmem_mapping(mapping)) {
64 page = find_get_entry(mapping, pgoff);
65 /*
66 * shmem/tmpfs may return swap: account for swapcache
67 * page too.
68 */
69 if (radix_tree_exceptional_entry(page)) {
70 swp_entry_t swp = radix_to_swp_entry(page);
Huang Yingf6ab1f72016-10-07 17:00:21 -070071 page = find_get_page(swap_address_space(swp),
72 swp_offset(swp));
Johannes Weiner0cd61442014-04-03 14:47:46 -070073 }
74 } else
75 page = find_get_page(mapping, pgoff);
76#else
77 page = find_get_page(mapping, pgoff);
Hugh Dickins31475dd2011-08-03 16:21:27 -070078#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -070079 if (page) {
80 present = PageUptodate(page);
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +030081 put_page(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -070082 }
83
84 return present;
85}
86
Naoya Horiguchi1e25a272015-02-11 15:28:11 -080087static int __mincore_unmapped_range(unsigned long addr, unsigned long end,
88 struct vm_area_struct *vma, unsigned char *vec)
Linus Torvalds1da177e2005-04-16 15:20:36 -070089{
Johannes Weiner25ef0e52010-05-24 14:32:11 -070090 unsigned long nr = (end - addr) >> PAGE_SHIFT;
Nick Piggin42da9cb2007-02-12 00:51:39 -080091 int i;
Linus Torvalds1da177e2005-04-16 15:20:36 -070092
Johannes Weinerf4884012010-05-24 14:32:10 -070093 if (vma->vm_file) {
94 pgoff_t pgoff;
Linus Torvalds2f77d102006-12-16 09:44:32 -080095
Johannes Weinerf4884012010-05-24 14:32:10 -070096 pgoff = linear_page_index(vma, addr);
97 for (i = 0; i < nr; i++, pgoff++)
98 vec[i] = mincore_page(vma->vm_file->f_mapping, pgoff);
99 } else {
100 for (i = 0; i < nr; i++)
101 vec[i] = 0;
Naoya Horiguchi4f16fc12009-12-14 17:59:58 -0800102 }
Naoya Horiguchi1e25a272015-02-11 15:28:11 -0800103 return nr;
Johannes Weinerf4884012010-05-24 14:32:10 -0700104}
Naoya Horiguchi4f16fc12009-12-14 17:59:58 -0800105
Naoya Horiguchi1e25a272015-02-11 15:28:11 -0800106static int mincore_unmapped_range(unsigned long addr, unsigned long end,
107 struct mm_walk *walk)
Johannes Weinerf4884012010-05-24 14:32:10 -0700108{
Naoya Horiguchi1e25a272015-02-11 15:28:11 -0800109 walk->private += __mincore_unmapped_range(addr, end,
110 walk->vma, walk->private);
111 return 0;
112}
Linus Torvalds2f77d102006-12-16 09:44:32 -0800113
Naoya Horiguchi1e25a272015-02-11 15:28:11 -0800114static int mincore_pte_range(pmd_t *pmd, unsigned long addr, unsigned long end,
115 struct mm_walk *walk)
116{
117 spinlock_t *ptl;
118 struct vm_area_struct *vma = walk->vma;
119 pte_t *ptep;
120 unsigned char *vec = walk->private;
121 int nr = (end - addr) >> PAGE_SHIFT;
122
Kirill A. Shutemovb6ec57f2016-01-21 16:40:25 -0800123 ptl = pmd_trans_huge_lock(pmd, vma);
124 if (ptl) {
Naoya Horiguchi1e25a272015-02-11 15:28:11 -0800125 memset(vec, 1, nr);
126 spin_unlock(ptl);
127 goto out;
128 }
129
130 if (pmd_trans_unstable(pmd)) {
131 __mincore_unmapped_range(addr, end, vma, vec);
132 goto out;
133 }
134
135 ptep = pte_offset_map_lock(walk->mm, pmd, addr, &ptl);
136 for (; addr != end; ptep++, addr += PAGE_SIZE) {
Nick Piggin42da9cb2007-02-12 00:51:39 -0800137 pte_t pte = *ptep;
138
Johannes Weinerf4884012010-05-24 14:32:10 -0700139 if (pte_none(pte))
Naoya Horiguchi1e25a272015-02-11 15:28:11 -0800140 __mincore_unmapped_range(addr, addr + PAGE_SIZE,
141 vma, vec);
Johannes Weinerf4884012010-05-24 14:32:10 -0700142 else if (pte_present(pte))
Johannes Weiner25ef0e52010-05-24 14:32:11 -0700143 *vec = 1;
Kirill A. Shutemov0661a332015-02-10 14:10:04 -0800144 else { /* pte is a swap entry */
Nick Piggin42da9cb2007-02-12 00:51:39 -0800145 swp_entry_t entry = pte_to_swp_entry(pte);
Johannes Weiner6a60f1b2010-05-24 14:32:09 -0700146
Weijie Yangc313dc5d2014-12-12 16:55:07 -0800147 if (non_swap_entry(entry)) {
148 /*
149 * migration or hwpoison entries are always
150 * uptodate
151 */
Johannes Weiner25ef0e52010-05-24 14:32:11 -0700152 *vec = 1;
Nick Piggin42da9cb2007-02-12 00:51:39 -0800153 } else {
Nick Piggin30fcffe2007-02-14 12:35:02 +0100154#ifdef CONFIG_SWAP
Shaohua Li33806f02013-02-22 16:34:37 -0800155 *vec = mincore_page(swap_address_space(entry),
Huang Yingf6ab1f72016-10-07 17:00:21 -0700156 swp_offset(entry));
Nick Piggin30fcffe2007-02-14 12:35:02 +0100157#else
158 WARN_ON(1);
Johannes Weiner25ef0e52010-05-24 14:32:11 -0700159 *vec = 1;
Nick Piggin30fcffe2007-02-14 12:35:02 +0100160#endif
Nick Piggin42da9cb2007-02-12 00:51:39 -0800161 }
162 }
Johannes Weiner25ef0e52010-05-24 14:32:11 -0700163 vec++;
Naoya Horiguchi1e25a272015-02-11 15:28:11 -0800164 }
Johannes Weiner6a60f1b2010-05-24 14:32:09 -0700165 pte_unmap_unlock(ptep - 1, ptl);
Naoya Horiguchi1e25a272015-02-11 15:28:11 -0800166out:
167 walk->private += nr;
168 cond_resched();
169 return 0;
Johannes Weinere48293f2010-05-24 14:32:11 -0700170}
171
Johannes Weinerf4884012010-05-24 14:32:10 -0700172/*
173 * Do a chunk of "sys_mincore()". We've already checked
174 * all the arguments, we hold the mmap semaphore: we should
175 * just return the amount of info we're asked for.
176 */
177static long do_mincore(unsigned long addr, unsigned long pages, unsigned char *vec)
178{
Johannes Weinerf4884012010-05-24 14:32:10 -0700179 struct vm_area_struct *vma;
Johannes Weiner25ef0e52010-05-24 14:32:11 -0700180 unsigned long end;
Naoya Horiguchi1e25a272015-02-11 15:28:11 -0800181 int err;
182 struct mm_walk mincore_walk = {
183 .pmd_entry = mincore_pte_range,
184 .pte_hole = mincore_unmapped_range,
185 .hugetlb_entry = mincore_hugetlb,
186 .private = vec,
187 };
Johannes Weinerf4884012010-05-24 14:32:10 -0700188
189 vma = find_vma(current->mm, addr);
190 if (!vma || addr < vma->vm_start)
191 return -ENOMEM;
Naoya Horiguchi1e25a272015-02-11 15:28:11 -0800192 mincore_walk.mm = vma->vm_mm;
Johannes Weiner25ef0e52010-05-24 14:32:11 -0700193 end = min(vma->vm_end, addr + (pages << PAGE_SHIFT));
Naoya Horiguchi1e25a272015-02-11 15:28:11 -0800194 err = walk_page_range(addr, end, &mincore_walk);
195 if (err < 0)
196 return err;
Johannes Weiner25ef0e52010-05-24 14:32:11 -0700197 return (end - addr) >> PAGE_SHIFT;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700198}
199
200/*
201 * The mincore(2) system call.
202 *
203 * mincore() returns the memory residency status of the pages in the
204 * current process's address space specified by [addr, addr + len).
205 * The status is returned in a vector of bytes. The least significant
206 * bit of each byte is 1 if the referenced page is in memory, otherwise
207 * it is zero.
208 *
209 * Because the status of a page can change after mincore() checks it
210 * but before it returns to the application, the returned vector may
211 * contain stale information. Only locked pages are guaranteed to
212 * remain in memory.
213 *
214 * return values:
215 * zero - success
216 * -EFAULT - vec points to an illegal address
Kirill A. Shutemovea1754a2016-04-01 15:29:48 +0300217 * -EINVAL - addr is not a multiple of PAGE_SIZE
Linus Torvalds1da177e2005-04-16 15:20:36 -0700218 * -ENOMEM - Addresses in the range [addr, addr + len] are
219 * invalid for the address space of this process, or
220 * specify one or more pages which are not currently
221 * mapped
222 * -EAGAIN - A kernel resource was temporarily unavailable.
223 */
Heiko Carstens3480b252009-01-14 14:14:16 +0100224SYSCALL_DEFINE3(mincore, unsigned long, start, size_t, len,
225 unsigned char __user *, vec)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700226{
Linus Torvalds2f77d102006-12-16 09:44:32 -0800227 long retval;
228 unsigned long pages;
229 unsigned char *tmp;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700230
Linus Torvalds2f77d102006-12-16 09:44:32 -0800231 /* Check the start address: needs to be page-aligned.. */
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +0300232 if (start & ~PAGE_MASK)
Linus Torvalds2f77d102006-12-16 09:44:32 -0800233 return -EINVAL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700234
Linus Torvalds2f77d102006-12-16 09:44:32 -0800235 /* ..and we need to be passed a valid user-space range */
236 if (!access_ok(VERIFY_READ, (void __user *) start, len))
237 return -ENOMEM;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700238
Kirill A. Shutemovea1754a2016-04-01 15:29:48 +0300239 /* This also avoids any overflows on PAGE_ALIGN */
Linus Torvalds2f77d102006-12-16 09:44:32 -0800240 pages = len >> PAGE_SHIFT;
Alexander Kuleshove7bbdd02015-11-05 18:46:38 -0800241 pages += (offset_in_page(len)) != 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700242
Linus Torvalds2f77d102006-12-16 09:44:32 -0800243 if (!access_ok(VERIFY_WRITE, vec, pages))
244 return -EFAULT;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700245
Linus Torvalds2f77d102006-12-16 09:44:32 -0800246 tmp = (void *) __get_free_page(GFP_USER);
247 if (!tmp)
Linus Torvalds4fb23e42006-12-16 16:01:50 -0800248 return -EAGAIN;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700249
Linus Torvalds2f77d102006-12-16 09:44:32 -0800250 retval = 0;
251 while (pages) {
252 /*
253 * Do at most PAGE_SIZE entries per iteration, due to
254 * the temporary buffer size.
255 */
256 down_read(&current->mm->mmap_sem);
Johannes Weiner6a60f1b2010-05-24 14:32:09 -0700257 retval = do_mincore(start, min(pages, PAGE_SIZE), tmp);
Linus Torvalds2f77d102006-12-16 09:44:32 -0800258 up_read(&current->mm->mmap_sem);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700259
Linus Torvalds2f77d102006-12-16 09:44:32 -0800260 if (retval <= 0)
261 break;
262 if (copy_to_user(vec, tmp, retval)) {
263 retval = -EFAULT;
264 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700265 }
Linus Torvalds2f77d102006-12-16 09:44:32 -0800266 pages -= retval;
267 vec += retval;
268 start += retval << PAGE_SHIFT;
269 retval = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700270 }
Linus Torvalds2f77d102006-12-16 09:44:32 -0800271 free_page((unsigned long) tmp);
272 return retval;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700273}