page.c

  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
 24enum kpage_operation {
 25	KPAGE_FLAGS,
 26	KPAGE_COUNT,
 27	KPAGE_CGROUP,
 28};
 29
 30static inline unsigned long get_max_dump_pfn(void)
 31{
 32#ifdef CONFIG_SPARSEMEM
 33	/*
 34	 * The memmap of early sections is completely populated and marked
 35	 * online even if max_pfn does not fall on a section boundary -
 36	 * pfn_to_online_page() will succeed on all pages. Allow inspecting
 37	 * these memmaps.
 38	 */
 39	return round_up(max_pfn, PAGES_PER_SECTION);
 40#else
 41	return max_pfn;
 42#endif
 43}
 44
 45static u64 get_kpage_count(const struct page *page)
 46{
 47	struct page_snapshot ps;
 48	u64 ret;
 49
 50	snapshot_page(&ps, page);
 51
 52	if (IS_ENABLED(CONFIG_PAGE_MAPCOUNT))
 53		ret = folio_precise_page_mapcount(&ps.folio_snapshot,
 54						  &ps.page_snapshot);
 55	else
 56		ret = folio_average_page_mapcount(&ps.folio_snapshot);
 57
 58	return ret;
 59}
 60
 61static ssize_t kpage_read(struct file *file, char __user *buf,
 62		size_t count, loff_t *ppos,
 63		enum kpage_operation op)
 64{
 65	const unsigned long max_dump_pfn = get_max_dump_pfn();
 66	u64 __user *out = (u64 __user *)buf;
 67	struct page *page;
 68	unsigned long src = *ppos;
 69	unsigned long pfn;
 70	ssize_t ret = 0;
 71	u64 info;
 72
 73	pfn = src / KPMSIZE;
 74	if (src & KPMMASK || count & KPMMASK)
 75		return -EINVAL;
 76	if (src >= max_dump_pfn * KPMSIZE)
 77		return 0;
 78	count = min_t(unsigned long, count, (max_dump_pfn * KPMSIZE) - src);
 79
 80	while (count > 0) {
 81		/*
 82		 * TODO: ZONE_DEVICE support requires to identify
 83		 * memmaps that were actually initialized.
 84		 */
 85		page = pfn_to_online_page(pfn);
 86
 87		if (page) {
 88			switch (op) {
 89			case KPAGE_FLAGS:
 90				info = stable_page_flags(page);
 91				break;
 92			case KPAGE_COUNT:
 93				info = get_kpage_count(page);
 94				break;
 95			case KPAGE_CGROUP:
 96				info = page_cgroup_ino(page);
 97				break;
 98			default:
 99				info = 0;
100				break;
101			}
102		} else
103			info = 0;
104
105		if (put_user(info, out)) {
106			ret = -EFAULT;
107			break;
108		}
109
110		pfn++;
111		out++;
112		count -= KPMSIZE;
113
114		cond_resched();
115	}
116
117	*ppos += (char __user *)out - buf;
118	if (!ret)
119		ret = (char __user *)out - buf;
120	return ret;
121}
122
123/* /proc/kpagecount - an array exposing page mapcounts
124 *
125 * Each entry is a u64 representing the corresponding
126 * physical page mapcount.
127 */
128static ssize_t kpagecount_read(struct file *file, char __user *buf,
129		size_t count, loff_t *ppos)
130{
131	return kpage_read(file, buf, count, ppos, KPAGE_COUNT);
132}
133
134static const struct proc_ops kpagecount_proc_ops = {
135	.proc_flags	= PROC_ENTRY_PERMANENT,
136	.proc_lseek	= mem_lseek,
137	.proc_read	= kpagecount_read,
138};
139
140
141static inline u64 kpf_copy_bit(u64 kflags, int ubit, int kbit)
142{
143	return ((kflags >> kbit) & 1) << ubit;
144}
145
146u64 stable_page_flags(const struct page *page)
147{
148	const struct folio *folio;
149	struct page_snapshot ps;
150	unsigned long k;
151	unsigned long mapping;
152	bool is_anon;
153	u64 u = 0;
154
155	/*
156	 * pseudo flag: KPF_NOPAGE
157	 * it differentiates a memory hole from a page with no flags
158	 */
159	if (!page)
160		return 1 << KPF_NOPAGE;
161
162	snapshot_page(&ps, page);
163	folio = &ps.folio_snapshot;
164
165	k = folio->flags.f;
166	mapping = (unsigned long)folio->mapping;
167	is_anon = mapping & FOLIO_MAPPING_ANON;
168
169	/*
170	 * pseudo flags for the well known (anonymous) memory mapped pages
171	 */
172	if (folio_mapped(folio))
173		u |= 1 << KPF_MMAP;
174	if (is_anon) {
175		u |= 1 << KPF_ANON;
176		if (mapping & FOLIO_MAPPING_KSM)
177			u |= 1 << KPF_KSM;
178	}
179
180	/*
181	 * compound pages: export both head/tail info
182	 * they together define a compound page's start/end pos and order
183	 */
184	if (ps.idx == 0)
185		u |= kpf_copy_bit(k, KPF_COMPOUND_HEAD, PG_head);
186	else
187		u |= 1 << KPF_COMPOUND_TAIL;
188	if (folio_test_hugetlb(folio))
189		u |= 1 << KPF_HUGE;
190	else if (folio_test_large(folio) &&
191	         folio_test_large_rmappable(folio)) {
192		/* Note: we indicate any THPs here, not just PMD-sized ones */
193		u |= 1 << KPF_THP;
194	} else if (is_huge_zero_pfn(ps.pfn)) {
195		u |= 1 << KPF_ZERO_PAGE;
196		u |= 1 << KPF_THP;
197	} else if (is_zero_pfn(ps.pfn)) {
198		u |= 1 << KPF_ZERO_PAGE;
199	}
200
201	if (ps.flags & PAGE_SNAPSHOT_PG_BUDDY)
202		u |= 1 << KPF_BUDDY;
203
204	if (folio_test_offline(folio))
205		u |= 1 << KPF_OFFLINE;
206	if (folio_test_pgtable(folio))
207		u |= 1 << KPF_PGTABLE;
208	if (folio_test_slab(folio))
209		u |= 1 << KPF_SLAB;
210
211#if defined(CONFIG_PAGE_IDLE_FLAG) && defined(CONFIG_64BIT)
212	u |= kpf_copy_bit(k, KPF_IDLE,          PG_idle);
213#else
214	if (ps.flags & PAGE_SNAPSHOT_PG_IDLE)
215		u |= 1 << KPF_IDLE;
216#endif
217
218	u |= kpf_copy_bit(k, KPF_LOCKED,	PG_locked);
219	u |= kpf_copy_bit(k, KPF_DIRTY,		PG_dirty);
220	u |= kpf_copy_bit(k, KPF_UPTODATE,	PG_uptodate);
221	u |= kpf_copy_bit(k, KPF_WRITEBACK,	PG_writeback);
222
223	u |= kpf_copy_bit(k, KPF_LRU,		PG_lru);
224	u |= kpf_copy_bit(k, KPF_REFERENCED,	PG_referenced);
225	u |= kpf_copy_bit(k, KPF_ACTIVE,	PG_active);
226	u |= kpf_copy_bit(k, KPF_RECLAIM,	PG_reclaim);
227
228#define SWAPCACHE ((1 << PG_swapbacked) | (1 << PG_swapcache))
229	if ((k & SWAPCACHE) == SWAPCACHE)
230		u |= 1 << KPF_SWAPCACHE;
231	u |= kpf_copy_bit(k, KPF_SWAPBACKED,	PG_swapbacked);
232
233	u |= kpf_copy_bit(k, KPF_UNEVICTABLE,	PG_unevictable);
234	u |= kpf_copy_bit(k, KPF_MLOCKED,	PG_mlocked);
235
236#ifdef CONFIG_MEMORY_FAILURE
237	if (u & (1 << KPF_HUGE))
238		u |= kpf_copy_bit(k, KPF_HWPOISON,	PG_hwpoison);
239	else
240		u |= kpf_copy_bit(ps.page_snapshot.flags.f, KPF_HWPOISON, PG_hwpoison);
241#endif
242
243	u |= kpf_copy_bit(k, KPF_RESERVED,	PG_reserved);
244	u |= kpf_copy_bit(k, KPF_OWNER_2,	PG_owner_2);
245	u |= kpf_copy_bit(k, KPF_PRIVATE,	PG_private);
246	u |= kpf_copy_bit(k, KPF_PRIVATE_2,	PG_private_2);
247	u |= kpf_copy_bit(k, KPF_OWNER_PRIVATE,	PG_owner_priv_1);
248	u |= kpf_copy_bit(k, KPF_ARCH,		PG_arch_1);
249#ifdef CONFIG_ARCH_USES_PG_ARCH_2
250	u |= kpf_copy_bit(k, KPF_ARCH_2,	PG_arch_2);
251#endif
252#ifdef CONFIG_ARCH_USES_PG_ARCH_3
253	u |= kpf_copy_bit(k, KPF_ARCH_3,	PG_arch_3);
254#endif
255
256	return u;
257}
258EXPORT_SYMBOL_GPL(stable_page_flags);
259
260/* /proc/kpageflags - an array exposing page flags
261 *
262 * Each entry is a u64 representing the corresponding
263 * physical page flags.
264 */
265static 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
271static 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
278static 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}
283static 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
290static 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}
299fs_initcall(proc_page_init);