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
get_max_dump_pfn(void)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
get_kpage_count(const struct page * page)46 static u64 get_kpage_count(const struct page *page)
47 {
48 struct page_snapshot ps;
49 u64 ret;
50
51 snapshot_page(&ps, page);
52
53 if (IS_ENABLED(CONFIG_PAGE_MAPCOUNT))
54 ret = folio_precise_page_mapcount(&ps.folio_snapshot,
55 &ps.page_snapshot);
56 else
57 ret = folio_average_page_mapcount(&ps.folio_snapshot);
58
59 return ret;
60 }
61
kpage_read(struct file * file,char __user * buf,size_t count,loff_t * ppos,enum kpage_operation op)62 static ssize_t kpage_read(struct file *file, char __user *buf,
63 size_t count, loff_t *ppos,
64 enum kpage_operation op)
65 {
66 const unsigned long max_dump_pfn = get_max_dump_pfn();
67 u64 __user *out = (u64 __user *)buf;
68 struct page *page;
69 unsigned long src = *ppos;
70 unsigned long pfn;
71 ssize_t ret = 0;
72 u64 info;
73
74 pfn = src / KPMSIZE;
75 if (src & KPMMASK || count & KPMMASK)
76 return -EINVAL;
77 if (src >= max_dump_pfn * KPMSIZE)
78 return 0;
79 count = min_t(unsigned long, count, (max_dump_pfn * KPMSIZE) - src);
80
81 while (count > 0) {
82 /*
83 * TODO: ZONE_DEVICE support requires to identify
84 * memmaps that were actually initialized.
85 */
86 page = pfn_to_online_page(pfn);
87
88 if (page) {
89 switch (op) {
90 case KPAGE_FLAGS:
91 info = stable_page_flags(page);
92 break;
93 case KPAGE_COUNT:
94 info = get_kpage_count(page);
95 break;
96 case KPAGE_CGROUP:
97 info = page_cgroup_ino(page);
98 break;
99 default:
100 info = 0;
101 break;
102 }
103 } else
104 info = 0;
105
106 if (put_user(info, out)) {
107 ret = -EFAULT;
108 break;
109 }
110
111 pfn++;
112 out++;
113 count -= KPMSIZE;
114
115 cond_resched();
116 }
117
118 *ppos += (char __user *)out - buf;
119 if (!ret)
120 ret = (char __user *)out - buf;
121 return ret;
122 }
123
124 /* /proc/kpagecount - an array exposing page mapcounts
125 *
126 * Each entry is a u64 representing the corresponding
127 * physical page mapcount.
128 */
kpagecount_read(struct file * file,char __user * buf,size_t count,loff_t * ppos)129 static ssize_t kpagecount_read(struct file *file, char __user *buf,
130 size_t count, loff_t *ppos)
131 {
132 return kpage_read(file, buf, count, ppos, KPAGE_COUNT);
133 }
134
135 static const struct proc_ops kpagecount_proc_ops = {
136 .proc_flags = PROC_ENTRY_PERMANENT,
137 .proc_lseek = mem_lseek,
138 .proc_read = kpagecount_read,
139 };
140
141
kpf_copy_bit(u64 kflags,int ubit,int kbit)142 static inline u64 kpf_copy_bit(u64 kflags, int ubit, int kbit)
143 {
144 return ((kflags >> kbit) & 1) << ubit;
145 }
146
stable_page_flags(const struct page * page)147 u64 stable_page_flags(const struct page *page)
148 {
149 const struct folio *folio;
150 struct page_snapshot ps;
151 unsigned long k;
152 unsigned long mapping;
153 bool is_anon;
154 u64 u = 0;
155
156 /*
157 * pseudo flag: KPF_NOPAGE
158 * it differentiates a memory hole from a page with no flags
159 */
160 if (!page)
161 return 1 << KPF_NOPAGE;
162
163 snapshot_page(&ps, page);
164 folio = &ps.folio_snapshot;
165
166 k = folio->flags;
167 mapping = (unsigned long)folio->mapping;
168 is_anon = mapping & FOLIO_MAPPING_ANON;
169
170 /*
171 * pseudo flags for the well known (anonymous) memory mapped pages
172 */
173 if (folio_mapped(folio))
174 u |= 1 << KPF_MMAP;
175 if (is_anon) {
176 u |= 1 << KPF_ANON;
177 if (mapping & FOLIO_MAPPING_KSM)
178 u |= 1 << KPF_KSM;
179 }
180
181 /*
182 * compound pages: export both head/tail info
183 * they together define a compound page's start/end pos and order
184 */
185 if (ps.idx == 0)
186 u |= kpf_copy_bit(k, KPF_COMPOUND_HEAD, PG_head);
187 else
188 u |= 1 << KPF_COMPOUND_TAIL;
189 if (folio_test_hugetlb(folio))
190 u |= 1 << KPF_HUGE;
191 else if (folio_test_large(folio) &&
192 folio_test_large_rmappable(folio)) {
193 /* Note: we indicate any THPs here, not just PMD-sized ones */
194 u |= 1 << KPF_THP;
195 } else if (is_huge_zero_pfn(ps.pfn)) {
196 u |= 1 << KPF_ZERO_PAGE;
197 u |= 1 << KPF_THP;
198 } else if (is_zero_pfn(ps.pfn)) {
199 u |= 1 << KPF_ZERO_PAGE;
200 }
201
202 if (ps.flags & PAGE_SNAPSHOT_PG_BUDDY)
203 u |= 1 << KPF_BUDDY;
204
205 if (folio_test_offline(folio))
206 u |= 1 << KPF_OFFLINE;
207 if (folio_test_pgtable(folio))
208 u |= 1 << KPF_PGTABLE;
209 if (folio_test_slab(folio))
210 u |= 1 << KPF_SLAB;
211
212 #if defined(CONFIG_PAGE_IDLE_FLAG) && defined(CONFIG_64BIT)
213 u |= kpf_copy_bit(k, KPF_IDLE, PG_idle);
214 #else
215 if (ps.flags & PAGE_SNAPSHOT_PG_IDLE)
216 u |= 1 << KPF_IDLE;
217 #endif
218
219 u |= kpf_copy_bit(k, KPF_LOCKED, PG_locked);
220 u |= kpf_copy_bit(k, KPF_DIRTY, PG_dirty);
221 u |= kpf_copy_bit(k, KPF_UPTODATE, PG_uptodate);
222 u |= kpf_copy_bit(k, KPF_WRITEBACK, PG_writeback);
223
224 u |= kpf_copy_bit(k, KPF_LRU, PG_lru);
225 u |= kpf_copy_bit(k, KPF_REFERENCED, PG_referenced);
226 u |= kpf_copy_bit(k, KPF_ACTIVE, PG_active);
227 u |= kpf_copy_bit(k, KPF_RECLAIM, PG_reclaim);
228
229 #define SWAPCACHE ((1 << PG_swapbacked) | (1 << PG_swapcache))
230 if ((k & SWAPCACHE) == SWAPCACHE)
231 u |= 1 << KPF_SWAPCACHE;
232 u |= kpf_copy_bit(k, KPF_SWAPBACKED, PG_swapbacked);
233
234 u |= kpf_copy_bit(k, KPF_UNEVICTABLE, PG_unevictable);
235 u |= kpf_copy_bit(k, KPF_MLOCKED, PG_mlocked);
236
237 #ifdef CONFIG_MEMORY_FAILURE
238 if (u & (1 << KPF_HUGE))
239 u |= kpf_copy_bit(k, KPF_HWPOISON, PG_hwpoison);
240 else
241 u |= kpf_copy_bit(ps.page_snapshot.flags, KPF_HWPOISON, PG_hwpoison);
242 #endif
243
244 u |= kpf_copy_bit(k, KPF_RESERVED, PG_reserved);
245 u |= kpf_copy_bit(k, KPF_OWNER_2, PG_owner_2);
246 u |= kpf_copy_bit(k, KPF_PRIVATE, PG_private);
247 u |= kpf_copy_bit(k, KPF_PRIVATE_2, PG_private_2);
248 u |= kpf_copy_bit(k, KPF_OWNER_PRIVATE, PG_owner_priv_1);
249 u |= kpf_copy_bit(k, KPF_ARCH, PG_arch_1);
250 #ifdef CONFIG_ARCH_USES_PG_ARCH_2
251 u |= kpf_copy_bit(k, KPF_ARCH_2, PG_arch_2);
252 #endif
253 #ifdef CONFIG_ARCH_USES_PG_ARCH_3
254 u |= kpf_copy_bit(k, KPF_ARCH_3, PG_arch_3);
255 #endif
256
257 return u;
258 }
259
260 /* /proc/kpageflags - an array exposing page flags
261 *
262 * Each entry is a u64 representing the corresponding
263 * physical page flags.
264 */
kpageflags_read(struct file * file,char __user * buf,size_t count,loff_t * ppos)265 static ssize_t kpageflags_read(struct file *file, char __user *buf,
266 size_t count, loff_t *ppos)
267 {
268 return kpage_read(file, buf, count, ppos, KPAGE_FLAGS);
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 return kpage_read(file, buf, count, ppos, KPAGE_CGROUP);
282 }
283 static const struct proc_ops kpagecgroup_proc_ops = {
284 .proc_flags = PROC_ENTRY_PERMANENT,
285 .proc_lseek = mem_lseek,
286 .proc_read = kpagecgroup_read,
287 };
288 #endif /* CONFIG_MEMCG */
289
proc_page_init(void)290 static int __init proc_page_init(void)
291 {
292 proc_create("kpagecount", S_IRUSR, NULL, &kpagecount_proc_ops);
293 proc_create("kpageflags", S_IRUSR, NULL, &kpageflags_proc_ops);
294 #ifdef CONFIG_MEMCG
295 proc_create("kpagecgroup", S_IRUSR, NULL, &kpagecgroup_proc_ops);
296 #endif
297 return 0;
298 }
299 fs_initcall(proc_page_init);
300