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/swapops.h>
18 #include <linux/shmem_fs.h>
19 #include <linux/hugetlb.h>
20 #include <linux/pgtable.h>
21 
22 #include <linux/uaccess.h>
23 
mincore_hugetlb(pte_t * pte,unsigned long hmask,unsigned long addr,unsigned long end,struct mm_walk * walk)24 static int mincore_hugetlb(pte_t *pte, unsigned long hmask, unsigned long addr,
25 			unsigned long end, struct mm_walk *walk)
26 {
27 #ifdef CONFIG_HUGETLB_PAGE
28 	unsigned char present;
29 	unsigned char *vec = walk->private;
30 
31 	/*
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;
39 #else
40 	BUG();
41 #endif
42 	return 0;
43 }
44 
45 /*
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  */
mincore_page(struct address_space * mapping,pgoff_t index)51 static unsigned char mincore_page(struct address_space *mapping, pgoff_t index)
52 {
53 	unsigned char present = 0;
54 	struct page *page;
55 
56 	/*
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
60 	 * tmpfs's .fault). So swapped out tmpfs mappings are tested here.
61 	 */
62 	page = find_get_incore_page(mapping, index);
63 	if (page) {
64 		present = PageUptodate(page);
65 		put_page(page);
66 	}
67 
68 	return present;
69 }
70 
__mincore_unmapped_range(unsigned long addr,unsigned long end,struct vm_area_struct * vma,unsigned char * vec)71 static int __mincore_unmapped_range(unsigned long addr, unsigned long end,
72 				struct vm_area_struct *vma, unsigned char *vec)
73 {
74 	unsigned long nr = (end - addr) >> PAGE_SHIFT;
75 	int i;
76 
77 	if (vma->vm_file) {
78 		pgoff_t pgoff;
79 
80 		pgoff = linear_page_index(vma, addr);
81 		for (i = 0; i < nr; i++, pgoff++)
82 			vec[i] = mincore_page(vma->vm_file->f_mapping, pgoff);
83 	} else {
84 		for (i = 0; i < nr; i++)
85 			vec[i] = 0;
86 	}
87 	return nr;
88 }
89 
mincore_unmapped_range(unsigned long addr,unsigned long end,__always_unused int depth,struct mm_walk * walk)90 static int mincore_unmapped_range(unsigned long addr, unsigned long end,
91 				   __always_unused int depth,
92 				   struct mm_walk *walk)
93 {
94 	walk->private += __mincore_unmapped_range(addr, end,
95 						  walk->vma, walk->private);
96 	return 0;
97 }
98 
mincore_pte_range(pmd_t * pmd,unsigned long addr,unsigned long end,struct mm_walk * walk)99 static int mincore_pte_range(pmd_t *pmd, unsigned long addr, unsigned long end,
100 			struct mm_walk *walk)
101 {
102 	spinlock_t *ptl;
103 	struct vm_area_struct *vma = walk->vma;
104 	pte_t *ptep;
105 	unsigned char *vec = walk->private;
106 	int nr = (end - addr) >> PAGE_SHIFT;
107 
108 	ptl = pmd_trans_huge_lock(pmd, vma);
109 	if (ptl) {
110 		memset(vec, 1, nr);
111 		spin_unlock(ptl);
112 		goto out;
113 	}
114 
115 	if (pmd_trans_unstable(pmd)) {
116 		__mincore_unmapped_range(addr, end, vma, vec);
117 		goto out;
118 	}
119 
120 	ptep = pte_offset_map_lock(walk->mm, pmd, addr, &ptl);
121 	for (; addr != end; ptep++, addr += PAGE_SIZE) {
122 		pte_t pte = *ptep;
123 
124 		if (pte_none(pte))
125 			__mincore_unmapped_range(addr, addr + PAGE_SIZE,
126 						 vma, vec);
127 		else if (pte_present(pte))
128 			*vec = 1;
129 		else { /* pte is a swap entry */
130 			swp_entry_t entry = pte_to_swp_entry(pte);
131 
132 			if (non_swap_entry(entry)) {
133 				/*
134 				 * migration or hwpoison entries are always
135 				 * uptodate
136 				 */
137 				*vec = 1;
138 			} else {
139 #ifdef CONFIG_SWAP
140 				*vec = mincore_page(swap_address_space(entry),
141 						    swp_offset(entry));
142 #else
143 				WARN_ON(1);
144 				*vec = 1;
145 #endif
146 			}
147 		}
148 		vec++;
149 	}
150 	pte_unmap_unlock(ptep - 1, ptl);
151 out:
152 	walk->private += nr;
153 	cond_resched();
154 	return 0;
155 }
156 
can_do_mincore(struct vm_area_struct * vma)157 static inline bool can_do_mincore(struct vm_area_struct *vma)
158 {
159 	if (vma_is_anonymous(vma))
160 		return true;
161 	if (!vma->vm_file)
162 		return false;
163 	/*
164 	 * Reveal pagecache information only for non-anonymous mappings that
165 	 * correspond to the files the calling process could (if tried) open
166 	 * for writing; otherwise we'd be including shared non-exclusive
167 	 * mappings, which opens a side channel.
168 	 */
169 	return inode_owner_or_capable(&init_user_ns,
170 				      file_inode(vma->vm_file)) ||
171 	       file_permission(vma->vm_file, MAY_WRITE) == 0;
172 }
173 
174 static const struct mm_walk_ops mincore_walk_ops = {
175 	.pmd_entry		= mincore_pte_range,
176 	.pte_hole		= mincore_unmapped_range,
177 	.hugetlb_entry		= mincore_hugetlb,
178 };
179 
180 /*
181  * Do a chunk of "sys_mincore()". We've already checked
182  * all the arguments, we hold the mmap semaphore: we should
183  * just return the amount of info we're asked for.
184  */
do_mincore(unsigned long addr,unsigned long pages,unsigned char * vec)185 static long do_mincore(unsigned long addr, unsigned long pages, unsigned char *vec)
186 {
187 	struct vm_area_struct *vma;
188 	unsigned long end;
189 	int err;
190 
191 	vma = find_vma(current->mm, addr);
192 	if (!vma || addr < vma->vm_start)
193 		return -ENOMEM;
194 	end = min(vma->vm_end, addr + (pages << PAGE_SHIFT));
195 	if (!can_do_mincore(vma)) {
196 		unsigned long pages = DIV_ROUND_UP(end - addr, PAGE_SIZE);
197 		memset(vec, 1, pages);
198 		return pages;
199 	}
200 	err = walk_page_range(vma->vm_mm, addr, end, &mincore_walk_ops, vec);
201 	if (err < 0)
202 		return err;
203 	return (end - addr) >> PAGE_SHIFT;
204 }
205 
206 /*
207  * The mincore(2) system call.
208  *
209  * mincore() returns the memory residency status of the pages in the
210  * current process's address space specified by [addr, addr + len).
211  * The status is returned in a vector of bytes.  The least significant
212  * bit of each byte is 1 if the referenced page is in memory, otherwise
213  * it is zero.
214  *
215  * Because the status of a page can change after mincore() checks it
216  * but before it returns to the application, the returned vector may
217  * contain stale information.  Only locked pages are guaranteed to
218  * remain in memory.
219  *
220  * return values:
221  *  zero    - success
222  *  -EFAULT - vec points to an illegal address
223  *  -EINVAL - addr is not a multiple of PAGE_SIZE
224  *  -ENOMEM - Addresses in the range [addr, addr + len] are
225  *		invalid for the address space of this process, or
226  *		specify one or more pages which are not currently
227  *		mapped
228  *  -EAGAIN - A kernel resource was temporarily unavailable.
229  */
SYSCALL_DEFINE3(mincore,unsigned long,start,size_t,len,unsigned char __user *,vec)230 SYSCALL_DEFINE3(mincore, unsigned long, start, size_t, len,
231 		unsigned char __user *, vec)
232 {
233 	long retval;
234 	unsigned long pages;
235 	unsigned char *tmp;
236 
237 	start = untagged_addr(start);
238 
239 	/* Check the start address: needs to be page-aligned.. */
240 	if (start & ~PAGE_MASK)
241 		return -EINVAL;
242 
243 	/* ..and we need to be passed a valid user-space range */
244 	if (!access_ok((void __user *) start, len))
245 		return -ENOMEM;
246 
247 	/* This also avoids any overflows on PAGE_ALIGN */
248 	pages = len >> PAGE_SHIFT;
249 	pages += (offset_in_page(len)) != 0;
250 
251 	if (!access_ok(vec, pages))
252 		return -EFAULT;
253 
254 	tmp = (void *) __get_free_page(GFP_USER);
255 	if (!tmp)
256 		return -EAGAIN;
257 
258 	retval = 0;
259 	while (pages) {
260 		/*
261 		 * Do at most PAGE_SIZE entries per iteration, due to
262 		 * the temporary buffer size.
263 		 */
264 		mmap_read_lock(current->mm);
265 		retval = do_mincore(start, min(pages, PAGE_SIZE), tmp);
266 		mmap_read_unlock(current->mm);
267 
268 		if (retval <= 0)
269 			break;
270 		if (copy_to_user(vec, tmp, retval)) {
271 			retval = -EFAULT;
272 			break;
273 		}
274 		pages -= retval;
275 		vec += retval;
276 		start += retval << PAGE_SHIFT;
277 		retval = 0;
278 	}
279 	free_page((unsigned long) tmp);
280 	return retval;
281 }
282