xref: /linux/fs/proc/page.c (revision bf070bb0e6c62ba3075db0a666763ba52c677102)
1 // SPDX-License-Identifier: GPL-2.0
2 #include <linux/bootmem.h>
3 #include <linux/compiler.h>
4 #include <linux/fs.h>
5 #include <linux/init.h>
6 #include <linux/ksm.h>
7 #include <linux/mm.h>
8 #include <linux/mmzone.h>
9 #include <linux/huge_mm.h>
10 #include <linux/proc_fs.h>
11 #include <linux/seq_file.h>
12 #include <linux/hugetlb.h>
13 #include <linux/memcontrol.h>
14 #include <linux/mmu_notifier.h>
15 #include <linux/page_idle.h>
16 #include <linux/kernel-page-flags.h>
17 #include <linux/uaccess.h>
18 #include "internal.h"
19 
20 #define KPMSIZE sizeof(u64)
21 #define KPMMASK (KPMSIZE - 1)
22 #define KPMBITS (KPMSIZE * BITS_PER_BYTE)
23 
24 /* /proc/kpagecount - an array exposing page counts
25  *
26  * Each entry is a u64 representing the corresponding
27  * physical page count.
28  */
29 static ssize_t kpagecount_read(struct file *file, char __user *buf,
30 			     size_t count, loff_t *ppos)
31 {
32 	u64 __user *out = (u64 __user *)buf;
33 	struct page *ppage;
34 	unsigned long src = *ppos;
35 	unsigned long pfn;
36 	ssize_t ret = 0;
37 	u64 pcount;
38 
39 	pfn = src / KPMSIZE;
40 	count = min_t(size_t, count, (max_pfn * KPMSIZE) - src);
41 	if (src & KPMMASK || count & KPMMASK)
42 		return -EINVAL;
43 
44 	while (count > 0) {
45 		if (pfn_valid(pfn))
46 			ppage = pfn_to_page(pfn);
47 		else
48 			ppage = NULL;
49 		if (!ppage || PageSlab(ppage))
50 			pcount = 0;
51 		else
52 			pcount = page_mapcount(ppage);
53 
54 		if (put_user(pcount, out)) {
55 			ret = -EFAULT;
56 			break;
57 		}
58 
59 		pfn++;
60 		out++;
61 		count -= KPMSIZE;
62 
63 		cond_resched();
64 	}
65 
66 	*ppos += (char __user *)out - buf;
67 	if (!ret)
68 		ret = (char __user *)out - buf;
69 	return ret;
70 }
71 
72 static const struct file_operations proc_kpagecount_operations = {
73 	.llseek = mem_lseek,
74 	.read = kpagecount_read,
75 };
76 
77 /* /proc/kpageflags - an array exposing page flags
78  *
79  * Each entry is a u64 representing the corresponding
80  * physical page flags.
81  */
82 
83 static inline u64 kpf_copy_bit(u64 kflags, int ubit, int kbit)
84 {
85 	return ((kflags >> kbit) & 1) << ubit;
86 }
87 
88 u64 stable_page_flags(struct page *page)
89 {
90 	u64 k;
91 	u64 u;
92 
93 	/*
94 	 * pseudo flag: KPF_NOPAGE
95 	 * it differentiates a memory hole from a page with no flags
96 	 */
97 	if (!page)
98 		return 1 << KPF_NOPAGE;
99 
100 	k = page->flags;
101 	u = 0;
102 
103 	/*
104 	 * pseudo flags for the well known (anonymous) memory mapped pages
105 	 *
106 	 * Note that page->_mapcount is overloaded in SLOB/SLUB/SLQB, so the
107 	 * simple test in page_mapped() is not enough.
108 	 */
109 	if (!PageSlab(page) && page_mapped(page))
110 		u |= 1 << KPF_MMAP;
111 	if (PageAnon(page))
112 		u |= 1 << KPF_ANON;
113 	if (PageKsm(page))
114 		u |= 1 << KPF_KSM;
115 
116 	/*
117 	 * compound pages: export both head/tail info
118 	 * they together define a compound page's start/end pos and order
119 	 */
120 	if (PageHead(page))
121 		u |= 1 << KPF_COMPOUND_HEAD;
122 	if (PageTail(page))
123 		u |= 1 << KPF_COMPOUND_TAIL;
124 	if (PageHuge(page))
125 		u |= 1 << KPF_HUGE;
126 	/*
127 	 * PageTransCompound can be true for non-huge compound pages (slab
128 	 * pages or pages allocated by drivers with __GFP_COMP) because it
129 	 * just checks PG_head/PG_tail, so we need to check PageLRU/PageAnon
130 	 * to make sure a given page is a thp, not a non-huge compound page.
131 	 */
132 	else if (PageTransCompound(page)) {
133 		struct page *head = compound_head(page);
134 
135 		if (PageLRU(head) || PageAnon(head))
136 			u |= 1 << KPF_THP;
137 		else if (is_huge_zero_page(head)) {
138 			u |= 1 << KPF_ZERO_PAGE;
139 			u |= 1 << KPF_THP;
140 		}
141 	} else if (is_zero_pfn(page_to_pfn(page)))
142 		u |= 1 << KPF_ZERO_PAGE;
143 
144 
145 	/*
146 	 * Caveats on high order pages: page->_refcount will only be set
147 	 * -1 on the head page; SLUB/SLQB do the same for PG_slab;
148 	 * SLOB won't set PG_slab at all on compound pages.
149 	 */
150 	if (PageBuddy(page))
151 		u |= 1 << KPF_BUDDY;
152 	else if (page_count(page) == 0 && is_free_buddy_page(page))
153 		u |= 1 << KPF_BUDDY;
154 
155 	if (PageBalloon(page))
156 		u |= 1 << KPF_BALLOON;
157 
158 	if (page_is_idle(page))
159 		u |= 1 << KPF_IDLE;
160 
161 	u |= kpf_copy_bit(k, KPF_LOCKED,	PG_locked);
162 
163 	u |= kpf_copy_bit(k, KPF_SLAB,		PG_slab);
164 	if (PageTail(page) && PageSlab(compound_head(page)))
165 		u |= 1 << KPF_SLAB;
166 
167 	u |= kpf_copy_bit(k, KPF_ERROR,		PG_error);
168 	u |= kpf_copy_bit(k, KPF_DIRTY,		PG_dirty);
169 	u |= kpf_copy_bit(k, KPF_UPTODATE,	PG_uptodate);
170 	u |= kpf_copy_bit(k, KPF_WRITEBACK,	PG_writeback);
171 
172 	u |= kpf_copy_bit(k, KPF_LRU,		PG_lru);
173 	u |= kpf_copy_bit(k, KPF_REFERENCED,	PG_referenced);
174 	u |= kpf_copy_bit(k, KPF_ACTIVE,	PG_active);
175 	u |= kpf_copy_bit(k, KPF_RECLAIM,	PG_reclaim);
176 
177 	if (PageSwapCache(page))
178 		u |= 1 << KPF_SWAPCACHE;
179 	u |= kpf_copy_bit(k, KPF_SWAPBACKED,	PG_swapbacked);
180 
181 	u |= kpf_copy_bit(k, KPF_UNEVICTABLE,	PG_unevictable);
182 	u |= kpf_copy_bit(k, KPF_MLOCKED,	PG_mlocked);
183 
184 #ifdef CONFIG_MEMORY_FAILURE
185 	u |= kpf_copy_bit(k, KPF_HWPOISON,	PG_hwpoison);
186 #endif
187 
188 #ifdef CONFIG_ARCH_USES_PG_UNCACHED
189 	u |= kpf_copy_bit(k, KPF_UNCACHED,	PG_uncached);
190 #endif
191 
192 	u |= kpf_copy_bit(k, KPF_RESERVED,	PG_reserved);
193 	u |= kpf_copy_bit(k, KPF_MAPPEDTODISK,	PG_mappedtodisk);
194 	u |= kpf_copy_bit(k, KPF_PRIVATE,	PG_private);
195 	u |= kpf_copy_bit(k, KPF_PRIVATE_2,	PG_private_2);
196 	u |= kpf_copy_bit(k, KPF_OWNER_PRIVATE,	PG_owner_priv_1);
197 	u |= kpf_copy_bit(k, KPF_ARCH,		PG_arch_1);
198 
199 	return u;
200 };
201 
202 static ssize_t kpageflags_read(struct file *file, char __user *buf,
203 			     size_t count, loff_t *ppos)
204 {
205 	u64 __user *out = (u64 __user *)buf;
206 	struct page *ppage;
207 	unsigned long src = *ppos;
208 	unsigned long pfn;
209 	ssize_t ret = 0;
210 
211 	pfn = src / KPMSIZE;
212 	count = min_t(unsigned long, count, (max_pfn * KPMSIZE) - src);
213 	if (src & KPMMASK || count & KPMMASK)
214 		return -EINVAL;
215 
216 	while (count > 0) {
217 		if (pfn_valid(pfn))
218 			ppage = pfn_to_page(pfn);
219 		else
220 			ppage = NULL;
221 
222 		if (put_user(stable_page_flags(ppage), out)) {
223 			ret = -EFAULT;
224 			break;
225 		}
226 
227 		pfn++;
228 		out++;
229 		count -= KPMSIZE;
230 
231 		cond_resched();
232 	}
233 
234 	*ppos += (char __user *)out - buf;
235 	if (!ret)
236 		ret = (char __user *)out - buf;
237 	return ret;
238 }
239 
240 static const struct file_operations proc_kpageflags_operations = {
241 	.llseek = mem_lseek,
242 	.read = kpageflags_read,
243 };
244 
245 #ifdef CONFIG_MEMCG
246 static ssize_t kpagecgroup_read(struct file *file, char __user *buf,
247 				size_t count, loff_t *ppos)
248 {
249 	u64 __user *out = (u64 __user *)buf;
250 	struct page *ppage;
251 	unsigned long src = *ppos;
252 	unsigned long pfn;
253 	ssize_t ret = 0;
254 	u64 ino;
255 
256 	pfn = src / KPMSIZE;
257 	count = min_t(unsigned long, count, (max_pfn * KPMSIZE) - src);
258 	if (src & KPMMASK || count & KPMMASK)
259 		return -EINVAL;
260 
261 	while (count > 0) {
262 		if (pfn_valid(pfn))
263 			ppage = pfn_to_page(pfn);
264 		else
265 			ppage = NULL;
266 
267 		if (ppage)
268 			ino = page_cgroup_ino(ppage);
269 		else
270 			ino = 0;
271 
272 		if (put_user(ino, out)) {
273 			ret = -EFAULT;
274 			break;
275 		}
276 
277 		pfn++;
278 		out++;
279 		count -= KPMSIZE;
280 
281 		cond_resched();
282 	}
283 
284 	*ppos += (char __user *)out - buf;
285 	if (!ret)
286 		ret = (char __user *)out - buf;
287 	return ret;
288 }
289 
290 static const struct file_operations proc_kpagecgroup_operations = {
291 	.llseek = mem_lseek,
292 	.read = kpagecgroup_read,
293 };
294 #endif /* CONFIG_MEMCG */
295 
296 static int __init proc_page_init(void)
297 {
298 	proc_create("kpagecount", S_IRUSR, NULL, &proc_kpagecount_operations);
299 	proc_create("kpageflags", S_IRUSR, NULL, &proc_kpageflags_operations);
300 #ifdef CONFIG_MEMCG
301 	proc_create("kpagecgroup", S_IRUSR, NULL, &proc_kpagecgroup_operations);
302 #endif
303 	return 0;
304 }
305 fs_initcall(proc_page_init);
306