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 
get_max_dump_pfn(void)25 static inline unsigned long get_max_dump_pfn(void)
26 {
27 #ifdef CONFIG_SPARSEMEM
28 	/*
29 	 * The memmap of early sections is completely populated and marked
30 	 * online even if max_pfn does not fall on a section boundary -
31 	 * pfn_to_online_page() will succeed on all pages. Allow inspecting
32 	 * these memmaps.
33 	 */
34 	return round_up(max_pfn, PAGES_PER_SECTION);
35 #else
36 	return max_pfn;
37 #endif
38 }
39 
40 /* /proc/kpagecount - an array exposing page mapcounts
41  *
42  * Each entry is a u64 representing the corresponding
43  * physical page mapcount.
44  */
kpagecount_read(struct file * file,char __user * buf,size_t count,loff_t * ppos)45 static ssize_t kpagecount_read(struct file *file, char __user *buf,
46 			     size_t count, loff_t *ppos)
47 {
48 	const unsigned long max_dump_pfn = get_max_dump_pfn();
49 	u64 __user *out = (u64 __user *)buf;
50 	unsigned long src = *ppos;
51 	unsigned long pfn;
52 	ssize_t ret = 0;
53 
54 	pfn = src / KPMSIZE;
55 	if (src & KPMMASK || count & KPMMASK)
56 		return -EINVAL;
57 	if (src >= max_dump_pfn * KPMSIZE)
58 		return 0;
59 	count = min_t(unsigned long, count, (max_dump_pfn * KPMSIZE) - src);
60 
61 	while (count > 0) {
62 		struct page *page;
63 		u64 mapcount = 0;
64 
65 		/*
66 		 * TODO: ZONE_DEVICE support requires to identify
67 		 * memmaps that were actually initialized.
68 		 */
69 		page = pfn_to_online_page(pfn);
70 		if (page) {
71 			struct folio *folio = page_folio(page);
72 
73 			if (IS_ENABLED(CONFIG_PAGE_MAPCOUNT))
74 				mapcount = folio_precise_page_mapcount(folio, page);
75 			else
76 				mapcount = folio_average_page_mapcount(folio);
77 		}
78 
79 		if (put_user(mapcount, out)) {
80 			ret = -EFAULT;
81 			break;
82 		}
83 
84 		pfn++;
85 		out++;
86 		count -= KPMSIZE;
87 
88 		cond_resched();
89 	}
90 
91 	*ppos += (char __user *)out - buf;
92 	if (!ret)
93 		ret = (char __user *)out - buf;
94 	return ret;
95 }
96 
97 static const struct proc_ops kpagecount_proc_ops = {
98 	.proc_flags	= PROC_ENTRY_PERMANENT,
99 	.proc_lseek	= mem_lseek,
100 	.proc_read	= kpagecount_read,
101 };
102 
103 /* /proc/kpageflags - an array exposing page flags
104  *
105  * Each entry is a u64 representing the corresponding
106  * physical page flags.
107  */
108 
kpf_copy_bit(u64 kflags,int ubit,int kbit)109 static inline u64 kpf_copy_bit(u64 kflags, int ubit, int kbit)
110 {
111 	return ((kflags >> kbit) & 1) << ubit;
112 }
113 
stable_page_flags(const struct page * page)114 u64 stable_page_flags(const struct page *page)
115 {
116 	const struct folio *folio;
117 	unsigned long k;
118 	unsigned long mapping;
119 	bool is_anon;
120 	u64 u = 0;
121 
122 	/*
123 	 * pseudo flag: KPF_NOPAGE
124 	 * it differentiates a memory hole from a page with no flags
125 	 */
126 	if (!page)
127 		return 1 << KPF_NOPAGE;
128 	folio = page_folio(page);
129 
130 	k = folio->flags;
131 	mapping = (unsigned long)folio->mapping;
132 	is_anon = mapping & PAGE_MAPPING_ANON;
133 
134 	/*
135 	 * pseudo flags for the well known (anonymous) memory mapped pages
136 	 */
137 	if (page_mapped(page))
138 		u |= 1 << KPF_MMAP;
139 	if (is_anon) {
140 		u |= 1 << KPF_ANON;
141 		if (mapping & PAGE_MAPPING_KSM)
142 			u |= 1 << KPF_KSM;
143 	}
144 
145 	/*
146 	 * compound pages: export both head/tail info
147 	 * they together define a compound page's start/end pos and order
148 	 */
149 	if (page == &folio->page)
150 		u |= kpf_copy_bit(k, KPF_COMPOUND_HEAD, PG_head);
151 	else
152 		u |= 1 << KPF_COMPOUND_TAIL;
153 	if (folio_test_hugetlb(folio))
154 		u |= 1 << KPF_HUGE;
155 	else if (folio_test_large(folio) &&
156 	         folio_test_large_rmappable(folio)) {
157 		/* Note: we indicate any THPs here, not just PMD-sized ones */
158 		u |= 1 << KPF_THP;
159 	} else if (is_huge_zero_folio(folio)) {
160 		u |= 1 << KPF_ZERO_PAGE;
161 		u |= 1 << KPF_THP;
162 	} else if (is_zero_folio(folio)) {
163 		u |= 1 << KPF_ZERO_PAGE;
164 	}
165 
166 	/*
167 	 * Caveats on high order pages: PG_buddy and PG_slab will only be set
168 	 * on the head page.
169 	 */
170 	if (PageBuddy(page))
171 		u |= 1 << KPF_BUDDY;
172 	else if (page_count(page) == 0 && is_free_buddy_page(page))
173 		u |= 1 << KPF_BUDDY;
174 
175 	if (PageOffline(page))
176 		u |= 1 << KPF_OFFLINE;
177 	if (PageTable(page))
178 		u |= 1 << KPF_PGTABLE;
179 	if (folio_test_slab(folio))
180 		u |= 1 << KPF_SLAB;
181 
182 #if defined(CONFIG_PAGE_IDLE_FLAG) && defined(CONFIG_64BIT)
183 	u |= kpf_copy_bit(k, KPF_IDLE,          PG_idle);
184 #else
185 	if (folio_test_idle(folio))
186 		u |= 1 << KPF_IDLE;
187 #endif
188 
189 	u |= kpf_copy_bit(k, KPF_LOCKED,	PG_locked);
190 	u |= kpf_copy_bit(k, KPF_DIRTY,		PG_dirty);
191 	u |= kpf_copy_bit(k, KPF_UPTODATE,	PG_uptodate);
192 	u |= kpf_copy_bit(k, KPF_WRITEBACK,	PG_writeback);
193 
194 	u |= kpf_copy_bit(k, KPF_LRU,		PG_lru);
195 	u |= kpf_copy_bit(k, KPF_REFERENCED,	PG_referenced);
196 	u |= kpf_copy_bit(k, KPF_ACTIVE,	PG_active);
197 	u |= kpf_copy_bit(k, KPF_RECLAIM,	PG_reclaim);
198 
199 #define SWAPCACHE ((1 << PG_swapbacked) | (1 << PG_swapcache))
200 	if ((k & SWAPCACHE) == SWAPCACHE)
201 		u |= 1 << KPF_SWAPCACHE;
202 	u |= kpf_copy_bit(k, KPF_SWAPBACKED,	PG_swapbacked);
203 
204 	u |= kpf_copy_bit(k, KPF_UNEVICTABLE,	PG_unevictable);
205 	u |= kpf_copy_bit(k, KPF_MLOCKED,	PG_mlocked);
206 
207 #ifdef CONFIG_MEMORY_FAILURE
208 	if (u & (1 << KPF_HUGE))
209 		u |= kpf_copy_bit(k, KPF_HWPOISON,	PG_hwpoison);
210 	else
211 		u |= kpf_copy_bit(page->flags, KPF_HWPOISON,	PG_hwpoison);
212 #endif
213 
214 	u |= kpf_copy_bit(k, KPF_RESERVED,	PG_reserved);
215 	u |= kpf_copy_bit(k, KPF_OWNER_2,	PG_owner_2);
216 	u |= kpf_copy_bit(k, KPF_PRIVATE,	PG_private);
217 	u |= kpf_copy_bit(k, KPF_PRIVATE_2,	PG_private_2);
218 	u |= kpf_copy_bit(k, KPF_OWNER_PRIVATE,	PG_owner_priv_1);
219 	u |= kpf_copy_bit(k, KPF_ARCH,		PG_arch_1);
220 #ifdef CONFIG_ARCH_USES_PG_ARCH_2
221 	u |= kpf_copy_bit(k, KPF_ARCH_2,	PG_arch_2);
222 #endif
223 #ifdef CONFIG_ARCH_USES_PG_ARCH_3
224 	u |= kpf_copy_bit(k, KPF_ARCH_3,	PG_arch_3);
225 #endif
226 
227 	return u;
228 };
229 
kpageflags_read(struct file * file,char __user * buf,size_t count,loff_t * ppos)230 static ssize_t kpageflags_read(struct file *file, char __user *buf,
231 			     size_t count, loff_t *ppos)
232 {
233 	const unsigned long max_dump_pfn = get_max_dump_pfn();
234 	u64 __user *out = (u64 __user *)buf;
235 	unsigned long src = *ppos;
236 	unsigned long pfn;
237 	ssize_t ret = 0;
238 
239 	pfn = src / KPMSIZE;
240 	if (src & KPMMASK || count & KPMMASK)
241 		return -EINVAL;
242 	if (src >= max_dump_pfn * KPMSIZE)
243 		return 0;
244 	count = min_t(unsigned long, count, (max_dump_pfn * KPMSIZE) - src);
245 
246 	while (count > 0) {
247 		/*
248 		 * TODO: ZONE_DEVICE support requires to identify
249 		 * memmaps that were actually initialized.
250 		 */
251 		struct page *page = pfn_to_online_page(pfn);
252 
253 		if (put_user(stable_page_flags(page), out)) {
254 			ret = -EFAULT;
255 			break;
256 		}
257 
258 		pfn++;
259 		out++;
260 		count -= KPMSIZE;
261 
262 		cond_resched();
263 	}
264 
265 	*ppos += (char __user *)out - buf;
266 	if (!ret)
267 		ret = (char __user *)out - buf;
268 	return ret;
269 }
270 
271 static const struct proc_ops kpageflags_proc_ops = {
272 	.proc_flags	= PROC_ENTRY_PERMANENT,
273 	.proc_lseek	= mem_lseek,
274 	.proc_read	= kpageflags_read,
275 };
276 
277 #ifdef CONFIG_MEMCG
kpagecgroup_read(struct file * file,char __user * buf,size_t count,loff_t * ppos)278 static ssize_t kpagecgroup_read(struct file *file, char __user *buf,
279 				size_t count, loff_t *ppos)
280 {
281 	const unsigned long max_dump_pfn = get_max_dump_pfn();
282 	u64 __user *out = (u64 __user *)buf;
283 	struct page *ppage;
284 	unsigned long src = *ppos;
285 	unsigned long pfn;
286 	ssize_t ret = 0;
287 	u64 ino;
288 
289 	pfn = src / KPMSIZE;
290 	if (src & KPMMASK || count & KPMMASK)
291 		return -EINVAL;
292 	if (src >= max_dump_pfn * KPMSIZE)
293 		return 0;
294 	count = min_t(unsigned long, count, (max_dump_pfn * KPMSIZE) - src);
295 
296 	while (count > 0) {
297 		/*
298 		 * TODO: ZONE_DEVICE support requires to identify
299 		 * memmaps that were actually initialized.
300 		 */
301 		ppage = pfn_to_online_page(pfn);
302 
303 		if (ppage)
304 			ino = page_cgroup_ino(ppage);
305 		else
306 			ino = 0;
307 
308 		if (put_user(ino, out)) {
309 			ret = -EFAULT;
310 			break;
311 		}
312 
313 		pfn++;
314 		out++;
315 		count -= KPMSIZE;
316 
317 		cond_resched();
318 	}
319 
320 	*ppos += (char __user *)out - buf;
321 	if (!ret)
322 		ret = (char __user *)out - buf;
323 	return ret;
324 }
325 
326 static const struct proc_ops kpagecgroup_proc_ops = {
327 	.proc_flags	= PROC_ENTRY_PERMANENT,
328 	.proc_lseek	= mem_lseek,
329 	.proc_read	= kpagecgroup_read,
330 };
331 #endif /* CONFIG_MEMCG */
332 
proc_page_init(void)333 static int __init proc_page_init(void)
334 {
335 	proc_create("kpagecount", S_IRUSR, NULL, &kpagecount_proc_ops);
336 	proc_create("kpageflags", S_IRUSR, NULL, &kpageflags_proc_ops);
337 #ifdef CONFIG_MEMCG
338 	proc_create("kpagecgroup", S_IRUSR, NULL, &kpagecgroup_proc_ops);
339 #endif
340 	return 0;
341 }
342 fs_initcall(proc_page_init);
343