forked from ~ljy/RK356X_SDK_RELEASE
blame | history | patch | commit | commitdiff
hc
2024-02-19 1c055e55a242a33e574e48be530e06770a210dcd 
 kernel/arch/s390/mm/dump_pagetables.c
....@@ -1,10 +1,13 @@
11 // SPDX-License-Identifier: GPL-2.0
2
+#include <linux/set_memory.h>
3
+#include <linux/ptdump.h>
24 #include <linux/seq_file.h>
35 #include <linux/debugfs.h>
4
-#include <linux/sched.h>
56 #include <linux/mm.h>
7
+#include <linux/kasan.h>
8
+#include <asm/ptdump.h>
9
+#include <asm/kasan.h>
610 #include <asm/sections.h>
7
-#include <asm/pgtable.h>
811 
912 static unsigned long max_addr;
1013 
....@@ -14,222 +17,234 @@
1417 };
1518 
1619 enum address_markers_idx {
17
-	IDENTITY_NR = 0,
20
+	IDENTITY_BEFORE_NR = 0,
21
+	IDENTITY_BEFORE_END_NR,
1822 	KERNEL_START_NR,
1923 	KERNEL_END_NR,
24
+	IDENTITY_AFTER_NR,
25
+	IDENTITY_AFTER_END_NR,
26
+#ifdef CONFIG_KASAN
27
+	KASAN_SHADOW_START_NR,
28
+	KASAN_SHADOW_END_NR,
29
+#endif
2030 	VMEMMAP_NR,
31
+	VMEMMAP_END_NR,
2132 	VMALLOC_NR,
33
+	VMALLOC_END_NR,
2234 	MODULES_NR,
35
+	MODULES_END_NR,
2336 };
2437 
2538 static struct addr_marker address_markers[] = {
26
-	[IDENTITY_NR]	  = {0, "Identity Mapping"},
27
-	[KERNEL_START_NR] = {(unsigned long)_stext, "Kernel Image Start"},
28
-	[KERNEL_END_NR]	  = {(unsigned long)_end, "Kernel Image End"},
29
-	[VMEMMAP_NR]	  = {0, "vmemmap Area"},
30
-	[VMALLOC_NR]	  = {0, "vmalloc Area"},
31
-	[MODULES_NR]	  = {0, "Modules Area"},
39
+	[IDENTITY_BEFORE_NR]	= {0, "Identity Mapping Start"},
40
+	[IDENTITY_BEFORE_END_NR] = {(unsigned long)_stext, "Identity Mapping End"},
41
+	[KERNEL_START_NR]	= {(unsigned long)_stext, "Kernel Image Start"},
42
+	[KERNEL_END_NR]		= {(unsigned long)_end, "Kernel Image End"},
43
+	[IDENTITY_AFTER_NR]	= {(unsigned long)_end, "Identity Mapping Start"},
44
+	[IDENTITY_AFTER_END_NR]	= {0, "Identity Mapping End"},
45
+#ifdef CONFIG_KASAN
46
+	[KASAN_SHADOW_START_NR]	= {KASAN_SHADOW_START, "Kasan Shadow Start"},
47
+	[KASAN_SHADOW_END_NR]	= {KASAN_SHADOW_END, "Kasan Shadow End"},
48
+#endif
49
+	[VMEMMAP_NR]		= {0, "vmemmap Area Start"},
50
+	[VMEMMAP_END_NR]	= {0, "vmemmap Area End"},
51
+	[VMALLOC_NR]		= {0, "vmalloc Area Start"},
52
+	[VMALLOC_END_NR]	= {0, "vmalloc Area End"},
53
+	[MODULES_NR]		= {0, "Modules Area Start"},
54
+	[MODULES_END_NR]	= {0, "Modules Area End"},
3255 	{ -1, NULL }
3356 };
3457 
3558 struct pg_state {
59
+	struct ptdump_state ptdump;
60
+	struct seq_file *seq;
3661 	int level;
3762 	unsigned int current_prot;
63
+	bool check_wx;
64
+	unsigned long wx_pages;
3865 	unsigned long start_address;
39
-	unsigned long current_address;
4066 	const struct addr_marker *marker;
4167 };
68
+
69
+#define pt_dump_seq_printf(m, fmt, args...)	\
70
+({						\
71
+	struct seq_file *__m = (m);		\
72
+						\
73
+	if (__m)				\
74
+		seq_printf(__m, fmt, ##args);	\
75
+})
76
+
77
+#define pt_dump_seq_puts(m, fmt)		\
78
+({						\
79
+	struct seq_file *__m = (m);		\
80
+						\
81
+	if (__m)				\
82
+		seq_printf(__m, fmt);		\
83
+})
4284 
4385 static void print_prot(struct seq_file *m, unsigned int pr, int level)
4486 {
4587 	static const char * const level_name[] =
4688 		{ "ASCE", "PGD", "PUD", "PMD", "PTE" };
4789 
48
-	seq_printf(m, "%s ", level_name[level]);
90
+	pt_dump_seq_printf(m, "%s ", level_name[level]);
4991 	if (pr & _PAGE_INVALID) {
50
-		seq_printf(m, "I\n");
92
+		pt_dump_seq_printf(m, "I\n");
5193 		return;
5294 	}
53
-	seq_puts(m, (pr & _PAGE_PROTECT) ? "RO " : "RW ");
54
-	seq_puts(m, (pr & _PAGE_NOEXEC) ? "NX\n" : "X\n");
95
+	pt_dump_seq_puts(m, (pr & _PAGE_PROTECT) ? "RO " : "RW ");
96
+	pt_dump_seq_puts(m, (pr & _PAGE_NOEXEC) ? "NX\n" : "X\n");
5597 }
5698 
57
-static void note_page(struct seq_file *m, struct pg_state *st,
58
-		     unsigned int new_prot, int level)
99
+static void note_prot_wx(struct pg_state *st, unsigned long addr)
59100 {
60
-	static const char units[] = "KMGTPE";
101
+#ifdef CONFIG_DEBUG_WX
102
+	if (!st->check_wx)
103
+		return;
104
+	if (st->current_prot & _PAGE_INVALID)
105
+		return;
106
+	if (st->current_prot & _PAGE_PROTECT)
107
+		return;
108
+	if (st->current_prot & _PAGE_NOEXEC)
109
+		return;
110
+	/* The first lowcore page is currently still W+X. */
111
+	if (addr == PAGE_SIZE)
112
+		return;
113
+	WARN_ONCE(1, "s390/mm: Found insecure W+X mapping at address %pS\n",
114
+		  (void *)st->start_address);
115
+	st->wx_pages += (addr - st->start_address) / PAGE_SIZE;
116
+#endif /* CONFIG_DEBUG_WX */
117
+}
118
+
119
+static void note_page(struct ptdump_state *pt_st, unsigned long addr, int level, u64 val)
120
+{
61121 	int width = sizeof(unsigned long) * 2;
122
+	static const char units[] = "KMGTPE";
62123 	const char *unit = units;
63
-	unsigned int prot, cur;
64124 	unsigned long delta;
125
+	struct pg_state *st;
126
+	struct seq_file *m;
127
+	unsigned int prot;
65128 
66
-	/*
67
-	 * If we have a "break" in the series, we need to flush the state
68
-	 * that we have now. "break" is either changing perms, levels or
69
-	 * address space marker.
70
-	 */
71
-	prot = new_prot;
72
-	cur = st->current_prot;
73
-
74
-	if (!st->level) {
75
-		/* First entry */
76
-		st->current_prot = new_prot;
129
+	st = container_of(pt_st, struct pg_state, ptdump);
130
+	m = st->seq;
131
+	prot = val & (_PAGE_PROTECT | _PAGE_NOEXEC);
132
+	if (level == 4 && (val & _PAGE_INVALID))
133
+		prot = _PAGE_INVALID;
134
+	/* For pmd_none() & friends val gets passed as zero. */
135
+	if (level != 4 && !val)
136
+		prot = _PAGE_INVALID;
137
+	/* Final flush from generic code. */
138
+	if (level == -1)
139
+		addr = max_addr;
140
+	if (st->level == -1) {
141
+		pt_dump_seq_printf(m, "---[ %s ]---\n", st->marker->name);
142
+		st->start_address = addr;
143
+		st->current_prot = prot;
77144 		st->level = level;
78
-		st->marker = address_markers;
79
-		seq_printf(m, "---[ %s ]---\n", st->marker->name);
80
-	} else if (prot != cur || level != st->level ||
81
-		   st->current_address >= st->marker[1].start_address) {
82
-		/* Print the actual finished series */
83
-		seq_printf(m, "0x%0*lx-0x%0*lx",
84
-			   width, st->start_address,
85
-			   width, st->current_address);
86
-		delta = (st->current_address - st->start_address) >> 10;
145
+	} else if (prot != st->current_prot || level != st->level ||
146
+		   addr >= st->marker[1].start_address) {
147
+		note_prot_wx(st, addr);
148
+		pt_dump_seq_printf(m, "0x%0*lx-0x%0*lx ",
149
+				   width, st->start_address,
150
+				   width, addr);
151
+		delta = (addr - st->start_address) >> 10;
87152 		while (!(delta & 0x3ff) && unit[1]) {
88153 			delta >>= 10;
89154 			unit++;
90155 		}
91
-		seq_printf(m, "%9lu%c ", delta, *unit);
156
+		pt_dump_seq_printf(m, "%9lu%c ", delta, *unit);
92157 		print_prot(m, st->current_prot, st->level);
93
-		if (st->current_address >= st->marker[1].start_address) {
158
+		while (addr >= st->marker[1].start_address) {
94159 			st->marker++;
95
-			seq_printf(m, "---[ %s ]---\n", st->marker->name);
160
+			pt_dump_seq_printf(m, "---[ %s ]---\n", st->marker->name);
96161 		}
97
-		st->start_address = st->current_address;
98
-		st->current_prot = new_prot;
162
+		st->start_address = addr;
163
+		st->current_prot = prot;
99164 		st->level = level;
100165 	}
101166 }
102167 
103
-/*
104
- * The actual page table walker functions. In order to keep the
105
- * implementation of print_prot() short, we only check and pass
106
- * _PAGE_INVALID and _PAGE_PROTECT flags to note_page() if a region,
107
- * segment or page table entry is invalid or read-only.
108
- * After all it's just a hint that the current level being walked
109
- * contains an invalid or read-only entry.
110
- */
111
-static void walk_pte_level(struct seq_file *m, struct pg_state *st,
112
-			   pmd_t *pmd, unsigned long addr)
168
+#ifdef CONFIG_DEBUG_WX
169
+void ptdump_check_wx(void)
113170 {
114
-	unsigned int prot;
115
-	pte_t *pte;
116
-	int i;
171
+	struct pg_state st = {
172
+		.ptdump = {
173
+			.note_page = note_page,
174
+			.range = (struct ptdump_range[]) {
175
+				{.start = 0, .end = max_addr},
176
+				{.start = 0, .end = 0},
177
+			}
178
+		},
179
+		.seq = NULL,
180
+		.level = -1,
181
+		.current_prot = 0,
182
+		.check_wx = true,
183
+		.wx_pages = 0,
184
+		.start_address = 0,
185
+		.marker = (struct addr_marker[]) {
186
+			{ .start_address =  0, .name = NULL},
187
+			{ .start_address = -1, .name = NULL},
188
+		},
189
+	};
117190 
118
-	for (i = 0; i < PTRS_PER_PTE && addr < max_addr; i++) {
119
-		st->current_address = addr;
120
-		pte = pte_offset_kernel(pmd, addr);
121
-		prot = pte_val(*pte) &
122
-			(_PAGE_PROTECT | _PAGE_INVALID | _PAGE_NOEXEC);
123
-		note_page(m, st, prot, 4);
124
-		addr += PAGE_SIZE;
125
-	}
191
+	if (!MACHINE_HAS_NX)
192
+		return;
193
+	ptdump_walk_pgd(&st.ptdump, &init_mm, NULL);
194
+	if (st.wx_pages)
195
+		pr_warn("Checked W+X mappings: FAILED, %lu W+X pages found\n", st.wx_pages);
196
+	else
197
+		pr_info("Checked W+X mappings: passed, no unexpected W+X pages found\n");
126198 }
199
+#endif /* CONFIG_DEBUG_WX */
127200 
128
-static void walk_pmd_level(struct seq_file *m, struct pg_state *st,
129
-			   pud_t *pud, unsigned long addr)
130
-{
131
-	unsigned int prot;
132
-	pmd_t *pmd;
133
-	int i;
134
-
135
-	for (i = 0; i < PTRS_PER_PMD && addr < max_addr; i++) {
136
-		st->current_address = addr;
137
-		pmd = pmd_offset(pud, addr);
138
-		if (!pmd_none(*pmd)) {
139
-			if (pmd_large(*pmd)) {
140
-				prot = pmd_val(*pmd) &
141
-					(_SEGMENT_ENTRY_PROTECT |
142
-					 _SEGMENT_ENTRY_NOEXEC);
143
-				note_page(m, st, prot, 3);
144
-			} else
145
-				walk_pte_level(m, st, pmd, addr);
146
-		} else
147
-			note_page(m, st, _PAGE_INVALID, 3);
148
-		addr += PMD_SIZE;
149
-	}
150
-}
151
-
152
-static void walk_pud_level(struct seq_file *m, struct pg_state *st,
153
-			   p4d_t *p4d, unsigned long addr)
154
-{
155
-	unsigned int prot;
156
-	pud_t *pud;
157
-	int i;
158
-
159
-	for (i = 0; i < PTRS_PER_PUD && addr < max_addr; i++) {
160
-		st->current_address = addr;
161
-		pud = pud_offset(p4d, addr);
162
-		if (!pud_none(*pud))
163
-			if (pud_large(*pud)) {
164
-				prot = pud_val(*pud) &
165
-					(_REGION_ENTRY_PROTECT |
166
-					 _REGION_ENTRY_NOEXEC);
167
-				note_page(m, st, prot, 2);
168
-			} else
169
-				walk_pmd_level(m, st, pud, addr);
170
-		else
171
-			note_page(m, st, _PAGE_INVALID, 2);
172
-		addr += PUD_SIZE;
173
-	}
174
-}
175
-
176
-static void walk_p4d_level(struct seq_file *m, struct pg_state *st,
177
-			   pgd_t *pgd, unsigned long addr)
178
-{
179
-	p4d_t *p4d;
180
-	int i;
181
-
182
-	for (i = 0; i < PTRS_PER_P4D && addr < max_addr; i++) {
183
-		st->current_address = addr;
184
-		p4d = p4d_offset(pgd, addr);
185
-		if (!p4d_none(*p4d))
186
-			walk_pud_level(m, st, p4d, addr);
187
-		else
188
-			note_page(m, st, _PAGE_INVALID, 2);
189
-		addr += P4D_SIZE;
190
-	}
191
-}
192
-
193
-static void walk_pgd_level(struct seq_file *m)
194
-{
195
-	unsigned long addr = 0;
196
-	struct pg_state st;
197
-	pgd_t *pgd;
198
-	int i;
199
-
200
-	memset(&st, 0, sizeof(st));
201
-	for (i = 0; i < PTRS_PER_PGD && addr < max_addr; i++) {
202
-		st.current_address = addr;
203
-		pgd = pgd_offset_k(addr);
204
-		if (!pgd_none(*pgd))
205
-			walk_p4d_level(m, &st, pgd, addr);
206
-		else
207
-			note_page(m, &st, _PAGE_INVALID, 1);
208
-		addr += PGDIR_SIZE;
209
-		cond_resched();
210
-	}
211
-	/* Flush out the last page */
212
-	st.current_address = max_addr;
213
-	note_page(m, &st, 0, 0);
214
-}
215
-
201
+#ifdef CONFIG_PTDUMP_DEBUGFS
216202 static int ptdump_show(struct seq_file *m, void *v)
217203 {
218
-	walk_pgd_level(m);
204
+	struct pg_state st = {
205
+		.ptdump = {
206
+			.note_page = note_page,
207
+			.range = (struct ptdump_range[]) {
208
+				{.start = 0, .end = max_addr},
209
+				{.start = 0, .end = 0},
210
+			}
211
+		},
212
+		.seq = m,
213
+		.level = -1,
214
+		.current_prot = 0,
215
+		.check_wx = false,
216
+		.wx_pages = 0,
217
+		.start_address = 0,
218
+		.marker = address_markers,
219
+	};
220
+
221
+	get_online_mems();
222
+	mutex_lock(&cpa_mutex);
223
+	ptdump_walk_pgd(&st.ptdump, &init_mm, NULL);
224
+	mutex_unlock(&cpa_mutex);
225
+	put_online_mems();
219226 	return 0;
220227 }
228
+DEFINE_SHOW_ATTRIBUTE(ptdump);
229
+#endif /* CONFIG_PTDUMP_DEBUGFS */
221230 
222
-static int ptdump_open(struct inode *inode, struct file *filp)
231
+/*
232
+ * Heapsort from lib/sort.c is not a stable sorting algorithm, do a simple
233
+ * insertion sort to preserve the original order of markers with the same
234
+ * start address.
235
+ */
236
+static void sort_address_markers(void)
223237 {
224
-	return single_open(filp, ptdump_show, NULL);
225
-}
238
+	struct addr_marker tmp;
239
+	int i, j;
226240 
227
-static const struct file_operations ptdump_fops = {
228
-	.open		= ptdump_open,
229
-	.read		= seq_read,
230
-	.llseek		= seq_lseek,
231
-	.release	= single_release,
232
-};
241
+	for (i = 1; i < ARRAY_SIZE(address_markers) - 1; i++) {
242
+		tmp = address_markers[i];
243
+		for (j = i - 1; j >= 0 && address_markers[j].start_address > tmp.start_address; j--)
244
+			address_markers[j + 1] = address_markers[j];
245
+		address_markers[j + 1] = tmp;
246
+	}
247
+}
233248 
234249 static int pt_dump_init(void)
235250 {
....@@ -240,10 +255,17 @@
240255 	 */
241256 	max_addr = (S390_lowcore.kernel_asce & _REGION_ENTRY_TYPE_MASK) >> 2;
242257 	max_addr = 1UL << (max_addr * 11 + 31);
258
+	address_markers[IDENTITY_AFTER_END_NR].start_address = memory_end;
243259 	address_markers[MODULES_NR].start_address = MODULES_VADDR;
260
+	address_markers[MODULES_END_NR].start_address = MODULES_END;
244261 	address_markers[VMEMMAP_NR].start_address = (unsigned long) vmemmap;
262
+	address_markers[VMEMMAP_END_NR].start_address = (unsigned long)vmemmap + vmemmap_size;
245263 	address_markers[VMALLOC_NR].start_address = VMALLOC_START;
264
+	address_markers[VMALLOC_END_NR].start_address = VMALLOC_END;
265
+	sort_address_markers();
266
+#ifdef CONFIG_PTDUMP_DEBUGFS
246267 	debugfs_create_file("kernel_page_tables", 0400, NULL, NULL, &ptdump_fops);
268
+#endif /* CONFIG_PTDUMP_DEBUGFS */
247269 	return 0;
248270 }
249271 device_initcall(pt_dump_init);