xref: /openbmc/linux/mm/mincore.c (revision 69fad28c)
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/mm.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 
21 #include <linux/uaccess.h>
22 #include <asm/pgtable.h>
23 
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 static int mincore_unmapped_range(unsigned long addr, unsigned long end,
46 				   struct mm_walk *walk)
47 {
48 	unsigned char *vec = walk->private;
49 	unsigned long nr = (end - addr) >> PAGE_SHIFT;
50 
51 	memset(vec, 0, nr);
52 	walk->private += nr;
53 	return 0;
54 }
55 
56 static int mincore_pte_range(pmd_t *pmd, unsigned long addr, unsigned long end,
57 			struct mm_walk *walk)
58 {
59 	spinlock_t *ptl;
60 	struct vm_area_struct *vma = walk->vma;
61 	pte_t *ptep;
62 	unsigned char *vec = walk->private;
63 	int nr = (end - addr) >> PAGE_SHIFT;
64 
65 	ptl = pmd_trans_huge_lock(pmd, vma);
66 	if (ptl) {
67 		memset(vec, 1, nr);
68 		spin_unlock(ptl);
69 		goto out;
70 	}
71 
72 	/* We'll consider a THP page under construction to be there */
73 	if (pmd_trans_unstable(pmd)) {
74 		memset(vec, 1, nr);
75 		goto out;
76 	}
77 
78 	ptep = pte_offset_map_lock(walk->mm, pmd, addr, &ptl);
79 	for (; addr != end; ptep++, addr += PAGE_SIZE) {
80 		pte_t pte = *ptep;
81 
82 		if (pte_none(pte))
83 			*vec = 0;
84 		else if (pte_present(pte))
85 			*vec = 1;
86 		else { /* pte is a swap entry */
87 			swp_entry_t entry = pte_to_swp_entry(pte);
88 
89 			/*
90 			 * migration or hwpoison entries are always
91 			 * uptodate
92 			 */
93 			*vec = !!non_swap_entry(entry);
94 		}
95 		vec++;
96 	}
97 	pte_unmap_unlock(ptep - 1, ptl);
98 out:
99 	walk->private += nr;
100 	cond_resched();
101 	return 0;
102 }
103 
104 /*
105  * Do a chunk of "sys_mincore()". We've already checked
106  * all the arguments, we hold the mmap semaphore: we should
107  * just return the amount of info we're asked for.
108  */
109 static long do_mincore(unsigned long addr, unsigned long pages, unsigned char *vec)
110 {
111 	struct vm_area_struct *vma;
112 	unsigned long end;
113 	int err;
114 	struct mm_walk mincore_walk = {
115 		.pmd_entry = mincore_pte_range,
116 		.pte_hole = mincore_unmapped_range,
117 		.hugetlb_entry = mincore_hugetlb,
118 		.private = vec,
119 	};
120 
121 	vma = find_vma(current->mm, addr);
122 	if (!vma || addr < vma->vm_start)
123 		return -ENOMEM;
124 	mincore_walk.mm = vma->vm_mm;
125 	end = min(vma->vm_end, addr + (pages << PAGE_SHIFT));
126 	err = walk_page_range(addr, end, &mincore_walk);
127 	if (err < 0)
128 		return err;
129 	return (end - addr) >> PAGE_SHIFT;
130 }
131 
132 /*
133  * The mincore(2) system call.
134  *
135  * mincore() returns the memory residency status of the pages in the
136  * current process's address space specified by [addr, addr + len).
137  * The status is returned in a vector of bytes.  The least significant
138  * bit of each byte is 1 if the referenced page is in memory, otherwise
139  * it is zero.
140  *
141  * Because the status of a page can change after mincore() checks it
142  * but before it returns to the application, the returned vector may
143  * contain stale information.  Only locked pages are guaranteed to
144  * remain in memory.
145  *
146  * return values:
147  *  zero    - success
148  *  -EFAULT - vec points to an illegal address
149  *  -EINVAL - addr is not a multiple of PAGE_SIZE
150  *  -ENOMEM - Addresses in the range [addr, addr + len] are
151  *		invalid for the address space of this process, or
152  *		specify one or more pages which are not currently
153  *		mapped
154  *  -EAGAIN - A kernel resource was temporarily unavailable.
155  */
156 SYSCALL_DEFINE3(mincore, unsigned long, start, size_t, len,
157 		unsigned char __user *, vec)
158 {
159 	long retval;
160 	unsigned long pages;
161 	unsigned char *tmp;
162 
163 	/* Check the start address: needs to be page-aligned.. */
164 	if (start & ~PAGE_MASK)
165 		return -EINVAL;
166 
167 	/* ..and we need to be passed a valid user-space range */
168 	if (!access_ok((void __user *) start, len))
169 		return -ENOMEM;
170 
171 	/* This also avoids any overflows on PAGE_ALIGN */
172 	pages = len >> PAGE_SHIFT;
173 	pages += (offset_in_page(len)) != 0;
174 
175 	if (!access_ok(vec, pages))
176 		return -EFAULT;
177 
178 	tmp = (void *) __get_free_page(GFP_USER);
179 	if (!tmp)
180 		return -EAGAIN;
181 
182 	retval = 0;
183 	while (pages) {
184 		/*
185 		 * Do at most PAGE_SIZE entries per iteration, due to
186 		 * the temporary buffer size.
187 		 */
188 		down_read(&current->mm->mmap_sem);
189 		retval = do_mincore(start, min(pages, PAGE_SIZE), tmp);
190 		up_read(&current->mm->mmap_sem);
191 
192 		if (retval <= 0)
193 			break;
194 		if (copy_to_user(vec, tmp, retval)) {
195 			retval = -EFAULT;
196 			break;
197 		}
198 		pages -= retval;
199 		vec += retval;
200 		start += retval << PAGE_SHIFT;
201 		retval = 0;
202 	}
203 	free_page((unsigned long) tmp);
204 	return retval;
205 }
206