1 // SPDX-License-Identifier: GPL-2.0
2 #include <linux/memblock.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/memremap.h>
14 #include <linux/memcontrol.h>
15 #include <linux/mmu_notifier.h>
16 #include <linux/page_idle.h>
17 #include <linux/kernel-page-flags.h>
18 #include <linux/uaccess.h>
19 #include "internal.h"
20 
21 #define KPMSIZE sizeof(u64)
22 #define KPMMASK (KPMSIZE - 1)
23 #define KPMBITS (KPMSIZE * BITS_PER_BYTE)
24 
25 enum kpage_operation {
26 	KPAGE_FLAGS,
27 	KPAGE_COUNT,
28 	KPAGE_CGROUP,
29 };
30 
31 static inline unsigned long get_max_dump_pfn(void)
32 {
33 #ifdef CONFIG_SPARSEMEM
34 	/*
35 	 * The memmap of early sections is completely populated and marked
36 	 * online even if max_pfn does not fall on a section boundary -
37 	 * pfn_to_online_page() will succeed on all pages. Allow inspecting
38 	 * these memmaps.
39 	 */
40 	return round_up(max_pfn, PAGES_PER_SECTION);
41 #else
42 	return max_pfn;
43 #endif
44 }
45 
46 static ssize_t kpage_read(struct file *file, char __user *buf,
47 		size_t count, loff_t *ppos,
48 		enum kpage_operation op)
49 {
50 	const unsigned long max_dump_pfn = get_max_dump_pfn();
51 	u64 __user *out = (u64 __user *)buf;
52 	struct page *page;
53 	unsigned long src = *ppos;
54 	unsigned long pfn;
55 	ssize_t ret = 0;
56 	u64 info;
57 
58 	pfn = src / KPMSIZE;
59 	if (src & KPMMASK || count & KPMMASK)
60 		return -EINVAL;
61 	if (src >= max_dump_pfn * KPMSIZE)
62 		return 0;
63 	count = min_t(unsigned long, count, (max_dump_pfn * KPMSIZE) - src);
64 
65 	while (count > 0) {
66 		/*
67 		 * TODO: ZONE_DEVICE support requires to identify
68 		 * memmaps that were actually initialized.
69 		 */
70 		page = pfn_to_online_page(pfn);
71 
72 		if (page) {
73 			switch (op) {
74 			case KPAGE_FLAGS:
75 				info = stable_page_flags(page);
76 				break;
77 			case KPAGE_COUNT:
78 				if (IS_ENABLED(CONFIG_PAGE_MAPCOUNT))
79 					info = folio_precise_page_mapcount(page_folio(page), page);
80 				else
81 					info = folio_average_page_mapcount(page_folio(page));
82 				break;
83 			case KPAGE_CGROUP:
84 				info = page_cgroup_ino(page);
85 				break;
86 			default:
87 				info = 0;
88 				break;
89 			}
90 		} else
91 			info = 0;
92 
93 		if (put_user(info, out)) {
94 			ret = -EFAULT;
95 			break;
96 		}
97 
98 		pfn++;
99 		out++;
100 		count -= KPMSIZE;
101 
102 		cond_resched();
103 	}
104 
105 	*ppos += (char __user *)out - buf;
106 	if (!ret)
107 		ret = (char __user *)out - buf;
108 	return ret;
109 }
110 
111 /* /proc/kpagecount - an array exposing page mapcounts
112  *
113  * Each entry is a u64 representing the corresponding
114  * physical page mapcount.
115  */
116 static ssize_t kpagecount_read(struct file *file, char __user *buf,
117 		size_t count, loff_t *ppos)
118 {
119 	return kpage_read(file, buf, count, ppos, KPAGE_COUNT);
120 }
121 
122 static const struct proc_ops kpagecount_proc_ops = {
123 	.proc_flags	= PROC_ENTRY_PERMANENT,
124 	.proc_lseek	= mem_lseek,
125 	.proc_read	= kpagecount_read,
126 };
127 
128 
129 static inline u64 kpf_copy_bit(u64 kflags, int ubit, int kbit)
130 {
131 	return ((kflags >> kbit) & 1) << ubit;
132 }
133 
134 u64 stable_page_flags(const struct page *page)
135 {
136 	const struct folio *folio;
137 	unsigned long k;
138 	unsigned long mapping;
139 	bool is_anon;
140 	u64 u = 0;
141 
142 	/*
143 	 * pseudo flag: KPF_NOPAGE
144 	 * it differentiates a memory hole from a page with no flags
145 	 */
146 	if (!page)
147 		return 1 << KPF_NOPAGE;
148 	folio = page_folio(page);
149 
150 	k = folio->flags;
151 	mapping = (unsigned long)folio->mapping;
152 	is_anon = mapping & PAGE_MAPPING_ANON;
153 
154 	/*
155 	 * pseudo flags for the well known (anonymous) memory mapped pages
156 	 */
157 	if (page_mapped(page))
158 		u |= 1 << KPF_MMAP;
159 	if (is_anon) {
160 		u |= 1 << KPF_ANON;
161 		if (mapping & PAGE_MAPPING_KSM)
162 			u |= 1 << KPF_KSM;
163 	}
164 
165 	/*
166 	 * compound pages: export both head/tail info
167 	 * they together define a compound page's start/end pos and order
168 	 */
169 	if (page == &folio->page)
170 		u |= kpf_copy_bit(k, KPF_COMPOUND_HEAD, PG_head);
171 	else
172 		u |= 1 << KPF_COMPOUND_TAIL;
173 	if (folio_test_hugetlb(folio))
174 		u |= 1 << KPF_HUGE;
175 	else if (folio_test_large(folio) &&
176 	         folio_test_large_rmappable(folio)) {
177 		/* Note: we indicate any THPs here, not just PMD-sized ones */
178 		u |= 1 << KPF_THP;
179 	} else if (is_huge_zero_folio(folio)) {
180 		u |= 1 << KPF_ZERO_PAGE;
181 		u |= 1 << KPF_THP;
182 	} else if (is_zero_folio(folio)) {
183 		u |= 1 << KPF_ZERO_PAGE;
184 	}
185 
186 	/*
187 	 * Caveats on high order pages: PG_buddy and PG_slab will only be set
188 	 * on the head page.
189 	 */
190 	if (PageBuddy(page))
191 		u |= 1 << KPF_BUDDY;
192 	else if (page_count(page) == 0 && is_free_buddy_page(page))
193 		u |= 1 << KPF_BUDDY;
194 
195 	if (PageOffline(page))
196 		u |= 1 << KPF_OFFLINE;
197 	if (PageTable(page))
198 		u |= 1 << KPF_PGTABLE;
199 	if (folio_test_slab(folio))
200 		u |= 1 << KPF_SLAB;
201 
202 #if defined(CONFIG_PAGE_IDLE_FLAG) && defined(CONFIG_64BIT)
203 	u |= kpf_copy_bit(k, KPF_IDLE,          PG_idle);
204 #else
205 	if (folio_test_idle(folio))
206 		u |= 1 << KPF_IDLE;
207 #endif
208 
209 	u |= kpf_copy_bit(k, KPF_LOCKED,	PG_locked);
210 	u |= kpf_copy_bit(k, KPF_DIRTY,		PG_dirty);
211 	u |= kpf_copy_bit(k, KPF_UPTODATE,	PG_uptodate);
212 	u |= kpf_copy_bit(k, KPF_WRITEBACK,	PG_writeback);
213 
214 	u |= kpf_copy_bit(k, KPF_LRU,		PG_lru);
215 	u |= kpf_copy_bit(k, KPF_REFERENCED,	PG_referenced);
216 	u |= kpf_copy_bit(k, KPF_ACTIVE,	PG_active);
217 	u |= kpf_copy_bit(k, KPF_RECLAIM,	PG_reclaim);
218 
219 #define SWAPCACHE ((1 << PG_swapbacked) | (1 << PG_swapcache))
220 	if ((k & SWAPCACHE) == SWAPCACHE)
221 		u |= 1 << KPF_SWAPCACHE;
222 	u |= kpf_copy_bit(k, KPF_SWAPBACKED,	PG_swapbacked);
223 
224 	u |= kpf_copy_bit(k, KPF_UNEVICTABLE,	PG_unevictable);
225 	u |= kpf_copy_bit(k, KPF_MLOCKED,	PG_mlocked);
226 
227 #ifdef CONFIG_MEMORY_FAILURE
228 	if (u & (1 << KPF_HUGE))
229 		u |= kpf_copy_bit(k, KPF_HWPOISON,	PG_hwpoison);
230 	else
231 		u |= kpf_copy_bit(page->flags, KPF_HWPOISON,	PG_hwpoison);
232 #endif
233 
234 	u |= kpf_copy_bit(k, KPF_RESERVED,	PG_reserved);
235 	u |= kpf_copy_bit(k, KPF_OWNER_2,	PG_owner_2);
236 	u |= kpf_copy_bit(k, KPF_PRIVATE,	PG_private);
237 	u |= kpf_copy_bit(k, KPF_PRIVATE_2,	PG_private_2);
238 	u |= kpf_copy_bit(k, KPF_OWNER_PRIVATE,	PG_owner_priv_1);
239 	u |= kpf_copy_bit(k, KPF_ARCH,		PG_arch_1);
240 #ifdef CONFIG_ARCH_USES_PG_ARCH_2
241 	u |= kpf_copy_bit(k, KPF_ARCH_2,	PG_arch_2);
242 #endif
243 #ifdef CONFIG_ARCH_USES_PG_ARCH_3
244 	u |= kpf_copy_bit(k, KPF_ARCH_3,	PG_arch_3);
245 #endif
246 
247 	return u;
248 }
249 
250 /* /proc/kpageflags - an array exposing page flags
251  *
252  * Each entry is a u64 representing the corresponding
253  * physical page flags.
254  */
255 static ssize_t kpageflags_read(struct file *file, char __user *buf,
256 		size_t count, loff_t *ppos)
257 {
258 	return kpage_read(file, buf, count, ppos, KPAGE_FLAGS);
259 }
260 
261 static const struct proc_ops kpageflags_proc_ops = {
262 	.proc_flags	= PROC_ENTRY_PERMANENT,
263 	.proc_lseek	= mem_lseek,
264 	.proc_read	= kpageflags_read,
265 };
266 
267 #ifdef CONFIG_MEMCG
268 static ssize_t kpagecgroup_read(struct file *file, char __user *buf,
269 		size_t count, loff_t *ppos)
270 {
271 	return kpage_read(file, buf, count, ppos, KPAGE_CGROUP);
272 }
273 static const struct proc_ops kpagecgroup_proc_ops = {
274 	.proc_flags	= PROC_ENTRY_PERMANENT,
275 	.proc_lseek	= mem_lseek,
276 	.proc_read	= kpagecgroup_read,
277 };
278 #endif /* CONFIG_MEMCG */
279 
280 static int __init proc_page_init(void)
281 {
282 	proc_create("kpagecount", S_IRUSR, NULL, &kpagecount_proc_ops);
283 	proc_create("kpageflags", S_IRUSR, NULL, &kpageflags_proc_ops);
284 #ifdef CONFIG_MEMCG
285 	proc_create("kpagecgroup", S_IRUSR, NULL, &kpagecgroup_proc_ops);
286 #endif
287 	return 0;
288 }
289 fs_initcall(proc_page_init);
290