projects/linux·blob master · refs

343 lines · 8.5 KiB

1// SPDX-License-Identifier: GPL-2.0
2/*
3 * linux/mm/mincore.c
4 *
5 * Copyright (C) 1994-2006 Linus Torvalds
6 */
7
8/*
9 * The mincore() system call.
10 */
11#include <linux/pagemap.h>
12#include <linux/gfp.h>
13#include <linux/pagewalk.h>
14#include <linux/mman.h>
15#include <linux/slab.h>
16#include <linux/syscalls.h>
17#include <linux/swap.h>
18#include <linux/leafops.h>
19#include <linux/shmem_fs.h>
20#include <linux/hugetlb.h>
21#include <linux/pgtable.h>
22
23#include <linux/uaccess.h>
24#include "swap.h"
25#include "internal.h"
26
27static int mincore_hugetlb(pte_t *pte, unsigned long hmask, unsigned long addr,
28 unsigned long end, struct mm_walk *walk)
29{
30#ifdef CONFIG_HUGETLB_PAGE
31 const unsigned long nr = (end - addr) >> PAGE_SHIFT;
32 unsigned char resident;
33 spinlock_t *ptl;
34 pte_t ptep;
35
36 ptl = huge_pte_lock(hstate_vma(walk->vma), walk->mm, pte);
37 ptep = huge_ptep_get(walk->mm, addr, pte);
38 resident = !huge_pte_none(ptep) && !pte_is_marker(ptep);
39 memset(walk->private, resident, nr);
40 walk->private += nr;
41 spin_unlock(ptl);
42#else
43 BUG();
44#endif
45 return 0;
46}
47
48static unsigned char mincore_swap(swp_entry_t entry, bool shmem)
49{
50 struct swap_info_struct *si;
51 struct folio *folio = NULL;
52 unsigned char present = 0;
53
54 /*
55 * Shmem mapping may contain swapin error entries, which are
56 * absent. Page table may contain migration or hwpoison
57 * entries which are always uptodate.
58 */
59 if (!softleaf_is_swap(entry))
60 return !shmem;
61
62 if (!IS_ENABLED(CONFIG_SWAP)) {
63 WARN_ON(1);
64 return 0;
65 }
66
67 /*
68 * Shmem mapping lookup is lockless, so we need to grab the swap
69 * device. mincore page table walk locks the PTL, and the swap
70 * device is stable, avoid touching the si for better performance.
71 */
72 if (shmem) {
73 si = get_swap_device(entry);
74 if (!si)
75 return 0;
76 }
77 folio = swap_cache_get_folio(entry);
78 if (shmem)
79 put_swap_device(si);
80 if (folio) {
81 present = folio_test_uptodate(folio);
82 folio_put(folio);
83 }
84
85 return present;
86}
87
88/*
89 * Later we can get more picky about what "in core" means precisely.
90 * For now, simply check to see if the page is in the page cache,
91 * and is up to date; i.e. that no page-in operation would be required
92 * at this time if an application were to map and access this page.
93 */
94static unsigned char mincore_page(struct address_space *mapping, pgoff_t index)
95{
96 unsigned char present;
97 struct folio *folio;
98
99 /*
100 * When tmpfs swaps out a page from a file, any process mapping that
101 * file will not get a swp_entry_t in its pte, but rather it is like
102 * any other file mapping (ie. marked !present and faulted in with
103 * tmpfs's .fault). So swapped out tmpfs mappings are tested here.
104 */
105 folio = filemap_get_entry(mapping, index);
106 if (!folio)
107 return 0;
108
109 if (xa_is_value(folio)) {
110 if (!shmem_mapping(mapping))
111 return 0;
112 return mincore_swap(radix_to_swp_entry(folio), true);
113 }
114 present = folio_test_uptodate(folio);
115 folio_put(folio);
116
117 return present;
118}
119
120static int __mincore_unmapped_range(unsigned long addr, unsigned long end,
121 struct vm_area_struct *vma, unsigned char *vec)
122{
123 unsigned long nr = (end - addr) >> PAGE_SHIFT;
124 int i;
125
126 if (vma->vm_file) {
127 pgoff_t pgoff;
128
129 pgoff = linear_page_index(vma, addr);
130 for (i = 0; i < nr; i++, pgoff++)
131 vec[i] = mincore_page(vma->vm_file->f_mapping, pgoff);
132 } else {
133 for (i = 0; i < nr; i++)
134 vec[i] = 0;
135 }
136 return nr;
137}
138
139static int mincore_unmapped_range(unsigned long addr, unsigned long end,
140 __always_unused int depth,
141 struct mm_walk *walk)
142{
143 walk->private += __mincore_unmapped_range(addr, end,
144 walk->vma, walk->private);
145 return 0;
146}
147
148static int mincore_pud_entry(pud_t *pudp, unsigned long addr, unsigned long end,
149 struct mm_walk *walk)
150{
151 if (pud_is_huge(pudp_get(pudp))) {
152 const unsigned long nr = (end - addr) >> PAGE_SHIFT;
153
154 memset(walk->private, 1, nr);
155 walk->private += nr;
156 walk->action = ACTION_CONTINUE;
157 }
158
159 return 0;
160}
161
162static int mincore_pte_range(pmd_t *pmd, unsigned long addr, unsigned long end,
163 struct mm_walk *walk)
164{
165 spinlock_t *ptl;
166 struct vm_area_struct *vma = walk->vma;
167 pte_t *ptep;
168 unsigned char *vec = walk->private;
169 int nr = (end - addr) >> PAGE_SHIFT;
170 int step, i;
171
172 ptl = pmd_trans_huge_lock(pmd, vma);
173 if (ptl) {
174 memset(vec, 1, nr);
175 spin_unlock(ptl);
176 goto out;
177 }
178
179 ptep = pte_offset_map_lock(walk->mm, pmd, addr, &ptl);
180 if (!ptep) {
181 walk->action = ACTION_AGAIN;
182 return 0;
183 }
184 for (; addr != end; ptep += step, addr += step * PAGE_SIZE) {
185 pte_t pte = ptep_get(ptep);
186
187 step = 1;
188 /* We need to do cache lookup too for markers */
189 if (pte_none(pte) || pte_is_marker(pte))
190 __mincore_unmapped_range(addr, addr + PAGE_SIZE,
191 vma, vec);
192 else if (pte_present(pte)) {
193 unsigned int batch = pte_batch_hint(ptep, pte);
194
195 if (batch > 1) {
196 unsigned int max_nr = (end - addr) >> PAGE_SHIFT;
197
198 step = min_t(unsigned int, batch, max_nr);
199 }
200
201 for (i = 0; i < step; i++)
202 vec[i] = 1;
203 } else { /* pte is a swap entry */
204 const softleaf_t entry = softleaf_from_pte(pte);
205
206 *vec = mincore_swap(entry, false);
207 }
208 vec += step;
209 }
210 pte_unmap_unlock(ptep - 1, ptl);
211out:
212 walk->private += nr;
213 cond_resched();
214 return 0;
215}
216
217static inline bool can_do_mincore(struct vm_area_struct *vma)
218{
219 if (vma_is_anonymous(vma))
220 return true;
221 if (!vma->vm_file)
222 return false;
223 /*
224 * Reveal pagecache information only for non-anonymous mappings that
225 * correspond to the files the calling process could (if tried) open
226 * for writing; otherwise we'd be including shared non-exclusive
227 * mappings, which opens a side channel.
228 */
229 return file_owner_or_capable(vma->vm_file) ||
230 file_permission(vma->vm_file, MAY_WRITE) == 0;
231}
232
233static const struct mm_walk_ops mincore_walk_ops = {
234 .pud_entry = mincore_pud_entry,
235 .pmd_entry = mincore_pte_range,
236 .pte_hole = mincore_unmapped_range,
237 .hugetlb_entry = mincore_hugetlb,
238 .walk_lock = PGWALK_RDLOCK,
239};
240
241/*
242 * Do a chunk of "sys_mincore()". We've already checked
243 * all the arguments, we hold the mmap semaphore: we should
244 * just return the amount of info we're asked for.
245 */
246static long do_mincore(unsigned long addr, unsigned long pages, unsigned char *vec)
247{
248 struct vm_area_struct *vma;
249 unsigned long end;
250 int err;
251
252 vma = vma_lookup(current->mm, addr);
253 if (!vma)
254 return -ENOMEM;
255 end = min(vma->vm_end, addr + (pages << PAGE_SHIFT));
256 if (!can_do_mincore(vma)) {
257 unsigned long pages = DIV_ROUND_UP(end - addr, PAGE_SIZE);
258 memset(vec, 1, pages);
259 return pages;
260 }
261
262 err = walk_page_range_vma(vma, addr, end, &mincore_walk_ops, vec);
263 if (err < 0)
264 return err;
265 return (end - addr) >> PAGE_SHIFT;
266}
267
268/*
269 * The mincore(2) system call.
270 *
271 * mincore() returns the memory residency status of the pages in the
272 * current process's address space specified by [addr, addr + len).
273 * The status is returned in a vector of bytes. The least significant
274 * bit of each byte is 1 if the referenced page is in memory, otherwise
275 * it is zero.
276 *
277 * Because the status of a page can change after mincore() checks it
278 * but before it returns to the application, the returned vector may
279 * contain stale information. Only locked pages are guaranteed to
280 * remain in memory.
281 *
282 * return values:
283 * zero - success
284 * -EFAULT - vec points to an illegal address
285 * -EINVAL - addr is not a multiple of PAGE_SIZE
286 * -ENOMEM - Addresses in the range [addr, addr + len] are
287 * invalid for the address space of this process, or
288 * specify one or more pages which are not currently
289 * mapped
290 * -EAGAIN - A kernel resource was temporarily unavailable.
291 */
292SYSCALL_DEFINE3(mincore, unsigned long, start, size_t, len,
293 unsigned char __user *, vec)
294{
295 long retval;
296 unsigned long pages;
297 unsigned char *tmp;
298
299 start = untagged_addr(start);
300
301 /* Check the start address: needs to be page-aligned.. */
302 if (unlikely(start & ~PAGE_MASK))
303 return -EINVAL;
304
305 /* ..and we need to be passed a valid user-space range */
306 if (!access_ok((void __user *) start, len))
307 return -ENOMEM;
308
309 /* This also avoids any overflows on PAGE_ALIGN */
310 pages = len >> PAGE_SHIFT;
311 pages += (offset_in_page(len)) != 0;
312
313 if (!access_ok(vec, pages))
314 return -EFAULT;
315
316 tmp = kmalloc(PAGE_SIZE, GFP_KERNEL);
317 if (!tmp)
318 return -EAGAIN;
319
320 retval = 0;
321 while (pages) {
322 /*
323 * Do at most PAGE_SIZE entries per iteration, due to
324 * the temporary buffer size.
325 */
326 mmap_read_lock(current->mm);
327 retval = do_mincore(start, min(pages, PAGE_SIZE), tmp);
328 mmap_read_unlock(current->mm);
329
330 if (retval <= 0)
331 break;
332 if (copy_to_user(vec, tmp, retval)) {
333 retval = -EFAULT;
334 break;
335 }
336 pages -= retval;
337 vec += retval;
338 start += retval << PAGE_SHIFT;
339 retval = 0;
340 }
341 kfree(tmp);
342 return retval;
343}