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 counts
41  *
42  * Each entry is a u64 representing the corresponding
43  * physical page count.
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 	struct page *ppage;
51 	unsigned long src = *ppos;
52 	unsigned long pfn;
53 	ssize_t ret = 0;
54 	u64 pcount;
55 
56 	pfn = src / KPMSIZE;
57 	if (src & KPMMASK || count & KPMMASK)
58 		return -EINVAL;
59 	if (src >= max_dump_pfn * KPMSIZE)
60 		return 0;
61 	count = min_t(unsigned long, count, (max_dump_pfn * KPMSIZE) - src);
62 
63 	while (count > 0) {
64 		/*
65 		 * TODO: ZONE_DEVICE support requires to identify
66 		 * memmaps that were actually initialized.
67 		 */
68 		ppage = pfn_to_online_page(pfn);
69 
70 		if (!ppage || PageSlab(ppage) || page_has_type(ppage))
71 			pcount = 0;
72 		else
73 			pcount = page_mapcount(ppage);
74 
75 		if (put_user(pcount, out)) {
76 			ret = -EFAULT;
77 			break;
78 		}
79 
80 		pfn++;
81 		out++;
82 		count -= KPMSIZE;
83 
84 		cond_resched();
85 	}
86 
87 	*ppos += (char __user *)out - buf;
88 	if (!ret)
89 		ret = (char __user *)out - buf;
90 	return ret;
91 }
92 
93 static const struct proc_ops kpagecount_proc_ops = {
94 	.proc_lseek	= mem_lseek,
95 	.proc_read	= kpagecount_read,
96 };
97 
98 /* /proc/kpageflags - an array exposing page flags
99  *
100  * Each entry is a u64 representing the corresponding
101  * physical page flags.
102  */
103 
kpf_copy_bit(u64 kflags,int ubit,int kbit)104 static inline u64 kpf_copy_bit(u64 kflags, int ubit, int kbit)
105 {
106 	return ((kflags >> kbit) & 1) << ubit;
107 }
108 
stable_page_flags(struct page * page)109 u64 stable_page_flags(struct page *page)
110 {
111 	u64 k;
112 	u64 u;
113 
114 	/*
115 	 * pseudo flag: KPF_NOPAGE
116 	 * it differentiates a memory hole from a page with no flags
117 	 */
118 	if (!page)
119 		return 1 << KPF_NOPAGE;
120 
121 	k = page->flags;
122 	u = 0;
123 
124 	/*
125 	 * pseudo flags for the well known (anonymous) memory mapped pages
126 	 *
127 	 * Note that page->_mapcount is overloaded in SLOB/SLUB/SLQB, so the
128 	 * simple test in page_mapped() is not enough.
129 	 */
130 	if (!PageSlab(page) && page_mapped(page))
131 		u |= 1 << KPF_MMAP;
132 	if (PageAnon(page))
133 		u |= 1 << KPF_ANON;
134 	if (PageKsm(page))
135 		u |= 1 << KPF_KSM;
136 
137 	/*
138 	 * compound pages: export both head/tail info
139 	 * they together define a compound page's start/end pos and order
140 	 */
141 	if (PageHead(page))
142 		u |= 1 << KPF_COMPOUND_HEAD;
143 	if (PageTail(page))
144 		u |= 1 << KPF_COMPOUND_TAIL;
145 	if (PageHuge(page))
146 		u |= 1 << KPF_HUGE;
147 	/*
148 	 * PageTransCompound can be true for non-huge compound pages (slab
149 	 * pages or pages allocated by drivers with __GFP_COMP) because it
150 	 * just checks PG_head/PG_tail, so we need to check PageLRU/PageAnon
151 	 * to make sure a given page is a thp, not a non-huge compound page.
152 	 */
153 	else if (PageTransCompound(page)) {
154 		struct page *head = compound_head(page);
155 
156 		if (PageLRU(head) || PageAnon(head))
157 			u |= 1 << KPF_THP;
158 		else if (is_huge_zero_page(head)) {
159 			u |= 1 << KPF_ZERO_PAGE;
160 			u |= 1 << KPF_THP;
161 		}
162 	} else if (is_zero_pfn(page_to_pfn(page)))
163 		u |= 1 << KPF_ZERO_PAGE;
164 
165 
166 	/*
167 	 * Caveats on high order pages: page->_refcount will only be set
168 	 * -1 on the head page; SLUB/SLQB do the same for PG_slab;
169 	 * SLOB won't set PG_slab at all on compound pages.
170 	 */
171 	if (PageBuddy(page))
172 		u |= 1 << KPF_BUDDY;
173 	else if (page_count(page) == 0 && is_free_buddy_page(page))
174 		u |= 1 << KPF_BUDDY;
175 
176 	if (PageOffline(page))
177 		u |= 1 << KPF_OFFLINE;
178 	if (PageTable(page))
179 		u |= 1 << KPF_PGTABLE;
180 
181 	if (page_is_idle(page))
182 		u |= 1 << KPF_IDLE;
183 
184 	u |= kpf_copy_bit(k, KPF_LOCKED,	PG_locked);
185 
186 	u |= kpf_copy_bit(k, KPF_SLAB,		PG_slab);
187 	if (PageTail(page) && PageSlab(compound_head(page)))
188 		u |= 1 << KPF_SLAB;
189 
190 	u |= kpf_copy_bit(k, KPF_ERROR,		PG_error);
191 	u |= kpf_copy_bit(k, KPF_DIRTY,		PG_dirty);
192 	u |= kpf_copy_bit(k, KPF_UPTODATE,	PG_uptodate);
193 	u |= kpf_copy_bit(k, KPF_WRITEBACK,	PG_writeback);
194 
195 	u |= kpf_copy_bit(k, KPF_LRU,		PG_lru);
196 	u |= kpf_copy_bit(k, KPF_REFERENCED,	PG_referenced);
197 	u |= kpf_copy_bit(k, KPF_ACTIVE,	PG_active);
198 	u |= kpf_copy_bit(k, KPF_RECLAIM,	PG_reclaim);
199 
200 	if (PageSwapCache(page))
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 	u |= kpf_copy_bit(k, KPF_HWPOISON,	PG_hwpoison);
209 #endif
210 
211 #ifdef CONFIG_ARCH_USES_PG_UNCACHED
212 	u |= kpf_copy_bit(k, KPF_UNCACHED,	PG_uncached);
213 #endif
214 
215 	u |= kpf_copy_bit(k, KPF_RESERVED,	PG_reserved);
216 	u |= kpf_copy_bit(k, KPF_MAPPEDTODISK,	PG_mappedtodisk);
217 	u |= kpf_copy_bit(k, KPF_PRIVATE,	PG_private);
218 	u |= kpf_copy_bit(k, KPF_PRIVATE_2,	PG_private_2);
219 	u |= kpf_copy_bit(k, KPF_OWNER_PRIVATE,	PG_owner_priv_1);
220 	u |= kpf_copy_bit(k, KPF_ARCH,		PG_arch_1);
221 #ifdef CONFIG_64BIT
222 	u |= kpf_copy_bit(k, KPF_ARCH_2,	PG_arch_2);
223 #endif
224 
225 	return u;
226 };
227 
kpageflags_read(struct file * file,char __user * buf,size_t count,loff_t * ppos)228 static ssize_t kpageflags_read(struct file *file, char __user *buf,
229 			     size_t count, loff_t *ppos)
230 {
231 	const unsigned long max_dump_pfn = get_max_dump_pfn();
232 	u64 __user *out = (u64 __user *)buf;
233 	struct page *ppage;
234 	unsigned long src = *ppos;
235 	unsigned long pfn;
236 	ssize_t ret = 0;
237 
238 	pfn = src / KPMSIZE;
239 	if (src & KPMMASK || count & KPMMASK)
240 		return -EINVAL;
241 	if (src >= max_dump_pfn * KPMSIZE)
242 		return 0;
243 	count = min_t(unsigned long, count, (max_dump_pfn * KPMSIZE) - src);
244 
245 	while (count > 0) {
246 		/*
247 		 * TODO: ZONE_DEVICE support requires to identify
248 		 * memmaps that were actually initialized.
249 		 */
250 		ppage = pfn_to_online_page(pfn);
251 
252 		if (put_user(stable_page_flags(ppage), out)) {
253 			ret = -EFAULT;
254 			break;
255 		}
256 
257 		pfn++;
258 		out++;
259 		count -= KPMSIZE;
260 
261 		cond_resched();
262 	}
263 
264 	*ppos += (char __user *)out - buf;
265 	if (!ret)
266 		ret = (char __user *)out - buf;
267 	return ret;
268 }
269 
270 static const struct proc_ops kpageflags_proc_ops = {
271 	.proc_lseek	= mem_lseek,
272 	.proc_read	= kpageflags_read,
273 };
274 
275 #ifdef CONFIG_MEMCG
kpagecgroup_read(struct file * file,char __user * buf,size_t count,loff_t * ppos)276 static ssize_t kpagecgroup_read(struct file *file, char __user *buf,
277 				size_t count, loff_t *ppos)
278 {
279 	const unsigned long max_dump_pfn = get_max_dump_pfn();
280 	u64 __user *out = (u64 __user *)buf;
281 	struct page *ppage;
282 	unsigned long src = *ppos;
283 	unsigned long pfn;
284 	ssize_t ret = 0;
285 	u64 ino;
286 
287 	pfn = src / KPMSIZE;
288 	if (src & KPMMASK || count & KPMMASK)
289 		return -EINVAL;
290 	if (src >= max_dump_pfn * KPMSIZE)
291 		return 0;
292 	count = min_t(unsigned long, count, (max_dump_pfn * KPMSIZE) - src);
293 
294 	while (count > 0) {
295 		/*
296 		 * TODO: ZONE_DEVICE support requires to identify
297 		 * memmaps that were actually initialized.
298 		 */
299 		ppage = pfn_to_online_page(pfn);
300 
301 		if (ppage)
302 			ino = page_cgroup_ino(ppage);
303 		else
304 			ino = 0;
305 
306 		if (put_user(ino, out)) {
307 			ret = -EFAULT;
308 			break;
309 		}
310 
311 		pfn++;
312 		out++;
313 		count -= KPMSIZE;
314 
315 		cond_resched();
316 	}
317 
318 	*ppos += (char __user *)out - buf;
319 	if (!ret)
320 		ret = (char __user *)out - buf;
321 	return ret;
322 }
323 
324 static const struct proc_ops kpagecgroup_proc_ops = {
325 	.proc_lseek	= mem_lseek,
326 	.proc_read	= kpagecgroup_read,
327 };
328 #endif /* CONFIG_MEMCG */
329 
proc_page_init(void)330 static int __init proc_page_init(void)
331 {
332 	proc_create("kpagecount", S_IRUSR, NULL, &kpagecount_proc_ops);
333 	proc_create("kpageflags", S_IRUSR, NULL, &kpageflags_proc_ops);
334 #ifdef CONFIG_MEMCG
335 	proc_create("kpagecgroup", S_IRUSR, NULL, &kpagecgroup_proc_ops);
336 #endif
337 	return 0;
338 }
339 fs_initcall(proc_page_init);
340