xref: /openbmc/linux/mm/mincore.c (revision b34e08d5)
1 /*
2  *	linux/mm/mincore.c
3  *
4  * Copyright (C) 1994-2006  Linus Torvalds
5  */
6 
7 /*
8  * The mincore() system call.
9  */
10 #include <linux/pagemap.h>
11 #include <linux/gfp.h>
12 #include <linux/mm.h>
13 #include <linux/mman.h>
14 #include <linux/syscalls.h>
15 #include <linux/swap.h>
16 #include <linux/swapops.h>
17 #include <linux/hugetlb.h>
18 
19 #include <asm/uaccess.h>
20 #include <asm/pgtable.h>
21 
22 static void mincore_hugetlb_page_range(struct vm_area_struct *vma,
23 				unsigned long addr, unsigned long end,
24 				unsigned char *vec)
25 {
26 #ifdef CONFIG_HUGETLB_PAGE
27 	struct hstate *h;
28 
29 	h = hstate_vma(vma);
30 	while (1) {
31 		unsigned char present;
32 		pte_t *ptep;
33 		/*
34 		 * Huge pages are always in RAM for now, but
35 		 * theoretically it needs to be checked.
36 		 */
37 		ptep = huge_pte_offset(current->mm,
38 				       addr & huge_page_mask(h));
39 		present = ptep && !huge_pte_none(huge_ptep_get(ptep));
40 		while (1) {
41 			*vec = present;
42 			vec++;
43 			addr += PAGE_SIZE;
44 			if (addr == end)
45 				return;
46 			/* check hugepage border */
47 			if (!(addr & ~huge_page_mask(h)))
48 				break;
49 		}
50 	}
51 #else
52 	BUG();
53 #endif
54 }
55 
56 /*
57  * Later we can get more picky about what "in core" means precisely.
58  * For now, simply check to see if the page is in the page cache,
59  * and is up to date; i.e. that no page-in operation would be required
60  * at this time if an application were to map and access this page.
61  */
62 static unsigned char mincore_page(struct address_space *mapping, pgoff_t pgoff)
63 {
64 	unsigned char present = 0;
65 	struct page *page;
66 
67 	/*
68 	 * When tmpfs swaps out a page from a file, any process mapping that
69 	 * file will not get a swp_entry_t in its pte, but rather it is like
70 	 * any other file mapping (ie. marked !present and faulted in with
71 	 * tmpfs's .fault). So swapped out tmpfs mappings are tested here.
72 	 */
73 #ifdef CONFIG_SWAP
74 	if (shmem_mapping(mapping)) {
75 		page = find_get_entry(mapping, pgoff);
76 		/*
77 		 * shmem/tmpfs may return swap: account for swapcache
78 		 * page too.
79 		 */
80 		if (radix_tree_exceptional_entry(page)) {
81 			swp_entry_t swp = radix_to_swp_entry(page);
82 			page = find_get_page(swap_address_space(swp), swp.val);
83 		}
84 	} else
85 		page = find_get_page(mapping, pgoff);
86 #else
87 	page = find_get_page(mapping, pgoff);
88 #endif
89 	if (page) {
90 		present = PageUptodate(page);
91 		page_cache_release(page);
92 	}
93 
94 	return present;
95 }
96 
97 static void mincore_unmapped_range(struct vm_area_struct *vma,
98 				unsigned long addr, unsigned long end,
99 				unsigned char *vec)
100 {
101 	unsigned long nr = (end - addr) >> PAGE_SHIFT;
102 	int i;
103 
104 	if (vma->vm_file) {
105 		pgoff_t pgoff;
106 
107 		pgoff = linear_page_index(vma, addr);
108 		for (i = 0; i < nr; i++, pgoff++)
109 			vec[i] = mincore_page(vma->vm_file->f_mapping, pgoff);
110 	} else {
111 		for (i = 0; i < nr; i++)
112 			vec[i] = 0;
113 	}
114 }
115 
116 static void mincore_pte_range(struct vm_area_struct *vma, pmd_t *pmd,
117 			unsigned long addr, unsigned long end,
118 			unsigned char *vec)
119 {
120 	unsigned long next;
121 	spinlock_t *ptl;
122 	pte_t *ptep;
123 
124 	ptep = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl);
125 	do {
126 		pte_t pte = *ptep;
127 		pgoff_t pgoff;
128 
129 		next = addr + PAGE_SIZE;
130 		if (pte_none(pte))
131 			mincore_unmapped_range(vma, addr, next, vec);
132 		else if (pte_present(pte))
133 			*vec = 1;
134 		else if (pte_file(pte)) {
135 			pgoff = pte_to_pgoff(pte);
136 			*vec = mincore_page(vma->vm_file->f_mapping, pgoff);
137 		} else { /* pte is a swap entry */
138 			swp_entry_t entry = pte_to_swp_entry(pte);
139 
140 			if (is_migration_entry(entry)) {
141 				/* migration entries are always uptodate */
142 				*vec = 1;
143 			} else {
144 #ifdef CONFIG_SWAP
145 				pgoff = entry.val;
146 				*vec = mincore_page(swap_address_space(entry),
147 					pgoff);
148 #else
149 				WARN_ON(1);
150 				*vec = 1;
151 #endif
152 			}
153 		}
154 		vec++;
155 	} while (ptep++, addr = next, addr != end);
156 	pte_unmap_unlock(ptep - 1, ptl);
157 }
158 
159 static void mincore_pmd_range(struct vm_area_struct *vma, pud_t *pud,
160 			unsigned long addr, unsigned long end,
161 			unsigned char *vec)
162 {
163 	unsigned long next;
164 	pmd_t *pmd;
165 
166 	pmd = pmd_offset(pud, addr);
167 	do {
168 		next = pmd_addr_end(addr, end);
169 		if (pmd_trans_huge(*pmd)) {
170 			if (mincore_huge_pmd(vma, pmd, addr, next, vec)) {
171 				vec += (next - addr) >> PAGE_SHIFT;
172 				continue;
173 			}
174 			/* fall through */
175 		}
176 		if (pmd_none_or_trans_huge_or_clear_bad(pmd))
177 			mincore_unmapped_range(vma, addr, next, vec);
178 		else
179 			mincore_pte_range(vma, pmd, addr, next, vec);
180 		vec += (next - addr) >> PAGE_SHIFT;
181 	} while (pmd++, addr = next, addr != end);
182 }
183 
184 static void mincore_pud_range(struct vm_area_struct *vma, pgd_t *pgd,
185 			unsigned long addr, unsigned long end,
186 			unsigned char *vec)
187 {
188 	unsigned long next;
189 	pud_t *pud;
190 
191 	pud = pud_offset(pgd, addr);
192 	do {
193 		next = pud_addr_end(addr, end);
194 		if (pud_none_or_clear_bad(pud))
195 			mincore_unmapped_range(vma, addr, next, vec);
196 		else
197 			mincore_pmd_range(vma, pud, addr, next, vec);
198 		vec += (next - addr) >> PAGE_SHIFT;
199 	} while (pud++, addr = next, addr != end);
200 }
201 
202 static void mincore_page_range(struct vm_area_struct *vma,
203 			unsigned long addr, unsigned long end,
204 			unsigned char *vec)
205 {
206 	unsigned long next;
207 	pgd_t *pgd;
208 
209 	pgd = pgd_offset(vma->vm_mm, addr);
210 	do {
211 		next = pgd_addr_end(addr, end);
212 		if (pgd_none_or_clear_bad(pgd))
213 			mincore_unmapped_range(vma, addr, next, vec);
214 		else
215 			mincore_pud_range(vma, pgd, addr, next, vec);
216 		vec += (next - addr) >> PAGE_SHIFT;
217 	} while (pgd++, addr = next, addr != end);
218 }
219 
220 /*
221  * Do a chunk of "sys_mincore()". We've already checked
222  * all the arguments, we hold the mmap semaphore: we should
223  * just return the amount of info we're asked for.
224  */
225 static long do_mincore(unsigned long addr, unsigned long pages, unsigned char *vec)
226 {
227 	struct vm_area_struct *vma;
228 	unsigned long end;
229 
230 	vma = find_vma(current->mm, addr);
231 	if (!vma || addr < vma->vm_start)
232 		return -ENOMEM;
233 
234 	end = min(vma->vm_end, addr + (pages << PAGE_SHIFT));
235 
236 	if (is_vm_hugetlb_page(vma))
237 		mincore_hugetlb_page_range(vma, addr, end, vec);
238 	else
239 		mincore_page_range(vma, addr, end, vec);
240 
241 	return (end - addr) >> PAGE_SHIFT;
242 }
243 
244 /*
245  * The mincore(2) system call.
246  *
247  * mincore() returns the memory residency status of the pages in the
248  * current process's address space specified by [addr, addr + len).
249  * The status is returned in a vector of bytes.  The least significant
250  * bit of each byte is 1 if the referenced page is in memory, otherwise
251  * it is zero.
252  *
253  * Because the status of a page can change after mincore() checks it
254  * but before it returns to the application, the returned vector may
255  * contain stale information.  Only locked pages are guaranteed to
256  * remain in memory.
257  *
258  * return values:
259  *  zero    - success
260  *  -EFAULT - vec points to an illegal address
261  *  -EINVAL - addr is not a multiple of PAGE_CACHE_SIZE
262  *  -ENOMEM - Addresses in the range [addr, addr + len] are
263  *		invalid for the address space of this process, or
264  *		specify one or more pages which are not currently
265  *		mapped
266  *  -EAGAIN - A kernel resource was temporarily unavailable.
267  */
268 SYSCALL_DEFINE3(mincore, unsigned long, start, size_t, len,
269 		unsigned char __user *, vec)
270 {
271 	long retval;
272 	unsigned long pages;
273 	unsigned char *tmp;
274 
275 	/* Check the start address: needs to be page-aligned.. */
276  	if (start & ~PAGE_CACHE_MASK)
277 		return -EINVAL;
278 
279 	/* ..and we need to be passed a valid user-space range */
280 	if (!access_ok(VERIFY_READ, (void __user *) start, len))
281 		return -ENOMEM;
282 
283 	/* This also avoids any overflows on PAGE_CACHE_ALIGN */
284 	pages = len >> PAGE_SHIFT;
285 	pages += (len & ~PAGE_MASK) != 0;
286 
287 	if (!access_ok(VERIFY_WRITE, vec, pages))
288 		return -EFAULT;
289 
290 	tmp = (void *) __get_free_page(GFP_USER);
291 	if (!tmp)
292 		return -EAGAIN;
293 
294 	retval = 0;
295 	while (pages) {
296 		/*
297 		 * Do at most PAGE_SIZE entries per iteration, due to
298 		 * the temporary buffer size.
299 		 */
300 		down_read(&current->mm->mmap_sem);
301 		retval = do_mincore(start, min(pages, PAGE_SIZE), tmp);
302 		up_read(&current->mm->mmap_sem);
303 
304 		if (retval <= 0)
305 			break;
306 		if (copy_to_user(vec, tmp, retval)) {
307 			retval = -EFAULT;
308 			break;
309 		}
310 		pages -= retval;
311 		vec += retval;
312 		start += retval << PAGE_SHIFT;
313 		retval = 0;
314 	}
315 	free_page((unsigned long) tmp);
316 	return retval;
317 }
318