mincore.c

  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/syscalls.h>
 16#include <linux/swap.h>
 17#include <linux/leafops.h>
 18#include <linux/shmem_fs.h>
 19#include <linux/hugetlb.h>
 20#include <linux/pgtable.h>
 21
 22#include <linux/uaccess.h>
 23#include "swap.h"
 24#include "internal.h"
 25
 26static int mincore_hugetlb(pte_t *pte, unsigned long hmask, unsigned long addr,
 27			unsigned long end, struct mm_walk *walk)
 28{
 29#ifdef CONFIG_HUGETLB_PAGE
 30	unsigned char present;
 31	unsigned char *vec = walk->private;
 32	spinlock_t *ptl;
 33
 34	ptl = huge_pte_lock(hstate_vma(walk->vma), walk->mm, pte);
 35
 36	/*
 37	 * Hugepages under user process are always in RAM and never
 38	 * swapped out, but theoretically it needs to be checked.
 39	 */
 40	if (!pte) {
 41		present = 0;
 42	} else {
 43		const pte_t ptep = huge_ptep_get(walk->mm, addr, pte);
 44
 45		if (huge_pte_none(ptep) || pte_is_marker(ptep))
 46			present = 0;
 47		else
 48			present = 1;
 49	}
 50
 51	for (; addr != end; vec++, addr += PAGE_SIZE)
 52		*vec = present;
 53	walk->private = vec;
 54	spin_unlock(ptl);
 55#else
 56	BUG();
 57#endif
 58	return 0;
 59}
 60
 61static unsigned char mincore_swap(swp_entry_t entry, bool shmem)
 62{
 63	struct swap_info_struct *si;
 64	struct folio *folio = NULL;
 65	unsigned char present = 0;
 66
 67	if (!IS_ENABLED(CONFIG_SWAP)) {
 68		WARN_ON(1);
 69		return 0;
 70	}
 71
 72	/*
 73	 * Shmem mapping may contain swapin error entries, which are
 74	 * absent. Page table may contain migration or hwpoison
 75	 * entries which are always uptodate.
 76	 */
 77	if (!softleaf_is_swap(entry))
 78		return !shmem;
 79
 80	/*
 81	 * Shmem mapping lookup is lockless, so we need to grab the swap
 82	 * device. mincore page table walk locks the PTL, and the swap
 83	 * device is stable, avoid touching the si for better performance.
 84	 */
 85	if (shmem) {
 86		si = get_swap_device(entry);
 87		if (!si)
 88			return 0;
 89	}
 90	folio = swap_cache_get_folio(entry);
 91	if (shmem)
 92		put_swap_device(si);
 93	/* The swap cache space contains either folio, shadow or NULL */
 94	if (folio && !xa_is_value(folio)) {
 95		present = folio_test_uptodate(folio);
 96		folio_put(folio);
 97	}
 98
 99	return present;
100}
101
102/*
103 * Later we can get more picky about what "in core" means precisely.
104 * For now, simply check to see if the page is in the page cache,
105 * and is up to date; i.e. that no page-in operation would be required
106 * at this time if an application were to map and access this page.
107 */
108static unsigned char mincore_page(struct address_space *mapping, pgoff_t index)
109{
110	unsigned char present = 0;
111	struct folio *folio;
112
113	/*
114	 * When tmpfs swaps out a page from a file, any process mapping that
115	 * file will not get a swp_entry_t in its pte, but rather it is like
116	 * any other file mapping (ie. marked !present and faulted in with
117	 * tmpfs's .fault). So swapped out tmpfs mappings are tested here.
118	 */
119	folio = filemap_get_entry(mapping, index);
120	if (folio) {
121		if (xa_is_value(folio)) {
122			if (shmem_mapping(mapping))
123				return mincore_swap(radix_to_swp_entry(folio),
124						    true);
125			else
126				return 0;
127		}
128		present = folio_test_uptodate(folio);
129		folio_put(folio);
130	}
131
132	return present;
133}
134
135static int __mincore_unmapped_range(unsigned long addr, unsigned long end,
136				struct vm_area_struct *vma, unsigned char *vec)
137{
138	unsigned long nr = (end - addr) >> PAGE_SHIFT;
139	int i;
140
141	if (vma->vm_file) {
142		pgoff_t pgoff;
143
144		pgoff = linear_page_index(vma, addr);
145		for (i = 0; i < nr; i++, pgoff++)
146			vec[i] = mincore_page(vma->vm_file->f_mapping, pgoff);
147	} else {
148		for (i = 0; i < nr; i++)
149			vec[i] = 0;
150	}
151	return nr;
152}
153
154static int mincore_unmapped_range(unsigned long addr, unsigned long end,
155				   __always_unused int depth,
156				   struct mm_walk *walk)
157{
158	walk->private += __mincore_unmapped_range(addr, end,
159						  walk->vma, walk->private);
160	return 0;
161}
162
163static int mincore_pte_range(pmd_t *pmd, unsigned long addr, unsigned long end,
164			struct mm_walk *walk)
165{
166	spinlock_t *ptl;
167	struct vm_area_struct *vma = walk->vma;
168	pte_t *ptep;
169	unsigned char *vec = walk->private;
170	int nr = (end - addr) >> PAGE_SHIFT;
171	int step, i;
172
173	ptl = pmd_trans_huge_lock(pmd, vma);
174	if (ptl) {
175		memset(vec, 1, nr);
176		spin_unlock(ptl);
177		goto out;
178	}
179
180	ptep = pte_offset_map_lock(walk->mm, pmd, addr, &ptl);
181	if (!ptep) {
182		walk->action = ACTION_AGAIN;
183		return 0;
184	}
185	for (; addr != end; ptep += step, addr += step * PAGE_SIZE) {
186		pte_t pte = ptep_get(ptep);
187
188		step = 1;
189		/* We need to do cache lookup too for markers */
190		if (pte_none(pte) || pte_is_marker(pte))
191			__mincore_unmapped_range(addr, addr + PAGE_SIZE,
192						 vma, vec);
193		else if (pte_present(pte)) {
194			unsigned int batch = pte_batch_hint(ptep, pte);
195
196			if (batch > 1) {
197				unsigned int max_nr = (end - addr) >> PAGE_SHIFT;
198
199				step = min_t(unsigned int, batch, max_nr);
200			}
201
202			for (i = 0; i < step; i++)
203				vec[i] = 1;
204		} else { /* pte is a swap entry */
205			const softleaf_t entry = softleaf_from_pte(pte);
206
207			*vec = mincore_swap(entry, false);
208		}
209		vec += step;
210	}
211	pte_unmap_unlock(ptep - 1, ptl);
212out:
213	walk->private += nr;
214	cond_resched();
215	return 0;
216}
217
218static inline bool can_do_mincore(struct vm_area_struct *vma)
219{
220	if (vma_is_anonymous(vma))
221		return true;
222	if (!vma->vm_file)
223		return false;
224	/*
225	 * Reveal pagecache information only for non-anonymous mappings that
226	 * correspond to the files the calling process could (if tried) open
227	 * for writing; otherwise we'd be including shared non-exclusive
228	 * mappings, which opens a side channel.
229	 */
230	return inode_owner_or_capable(&nop_mnt_idmap,
231				      file_inode(vma->vm_file)) ||
232	       file_permission(vma->vm_file, MAY_WRITE) == 0;
233}
234
235static const struct mm_walk_ops mincore_walk_ops = {
236	.pmd_entry		= mincore_pte_range,
237	.pte_hole		= mincore_unmapped_range,
238	.hugetlb_entry		= mincore_hugetlb,
239	.walk_lock		= PGWALK_RDLOCK,
240};
241
242/*
243 * Do a chunk of "sys_mincore()". We've already checked
244 * all the arguments, we hold the mmap semaphore: we should
245 * just return the amount of info we're asked for.
246 */
247static long do_mincore(unsigned long addr, unsigned long pages, unsigned char *vec)
248{
249	struct vm_area_struct *vma;
250	unsigned long end;
251	int err;
252
253	vma = vma_lookup(current->mm, addr);
254	if (!vma)
255		return -ENOMEM;
256	end = min(vma->vm_end, addr + (pages << PAGE_SHIFT));
257	if (!can_do_mincore(vma)) {
258		unsigned long pages = DIV_ROUND_UP(end - addr, PAGE_SIZE);
259		memset(vec, 1, pages);
260		return pages;
261	}
262	err = walk_page_range(vma->vm_mm, 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 = (void *) __get_free_page(GFP_USER);
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	free_page((unsigned long) tmp);
342	return retval;
343}