Loading...
1/*
2 * Debug helper to dump the current kernel pagetables of the system
3 * so that we can see what the various memory ranges are set to.
4 *
5 * (C) Copyright 2008 Intel Corporation
6 *
7 * Author: Arjan van de Ven <arjan@linux.intel.com>
8 *
9 * This program is free software; you can redistribute it and/or
10 * modify it under the terms of the GNU General Public License
11 * as published by the Free Software Foundation; version 2
12 * of the License.
13 */
14
15#include <linux/debugfs.h>
16#include <linux/mm.h>
17#include <linux/module.h>
18#include <linux/seq_file.h>
19
20#include <asm/pgtable.h>
21
22/*
23 * The dumper groups pagetable entries of the same type into one, and for
24 * that it needs to keep some state when walking, and flush this state
25 * when a "break" in the continuity is found.
26 */
27struct pg_state {
28 int level;
29 pgprot_t current_prot;
30 unsigned long start_address;
31 unsigned long current_address;
32 const struct addr_marker *marker;
33 unsigned long lines;
34 bool to_dmesg;
35 bool check_wx;
36 unsigned long wx_pages;
37};
38
39struct addr_marker {
40 unsigned long start_address;
41 const char *name;
42 unsigned long max_lines;
43};
44
45/* indices for address_markers; keep sync'd w/ address_markers below */
46enum address_markers_idx {
47 USER_SPACE_NR = 0,
48#ifdef CONFIG_X86_64
49 KERNEL_SPACE_NR,
50 LOW_KERNEL_NR,
51 VMALLOC_START_NR,
52 VMEMMAP_START_NR,
53# ifdef CONFIG_X86_ESPFIX64
54 ESPFIX_START_NR,
55# endif
56 HIGH_KERNEL_NR,
57 MODULES_VADDR_NR,
58 MODULES_END_NR,
59#else
60 KERNEL_SPACE_NR,
61 VMALLOC_START_NR,
62 VMALLOC_END_NR,
63# ifdef CONFIG_HIGHMEM
64 PKMAP_BASE_NR,
65# endif
66 FIXADDR_START_NR,
67#endif
68};
69
70/* Address space markers hints */
71static struct addr_marker address_markers[] = {
72 { 0, "User Space" },
73#ifdef CONFIG_X86_64
74 { 0x8000000000000000UL, "Kernel Space" },
75 { PAGE_OFFSET, "Low Kernel Mapping" },
76 { VMALLOC_START, "vmalloc() Area" },
77 { VMEMMAP_START, "Vmemmap" },
78# ifdef CONFIG_X86_ESPFIX64
79 { ESPFIX_BASE_ADDR, "ESPfix Area", 16 },
80# endif
81# ifdef CONFIG_EFI
82 { EFI_VA_END, "EFI Runtime Services" },
83# endif
84 { __START_KERNEL_map, "High Kernel Mapping" },
85 { MODULES_VADDR, "Modules" },
86 { MODULES_END, "End Modules" },
87#else
88 { PAGE_OFFSET, "Kernel Mapping" },
89 { 0/* VMALLOC_START */, "vmalloc() Area" },
90 { 0/*VMALLOC_END*/, "vmalloc() End" },
91# ifdef CONFIG_HIGHMEM
92 { 0/*PKMAP_BASE*/, "Persistent kmap() Area" },
93# endif
94 { 0/*FIXADDR_START*/, "Fixmap Area" },
95#endif
96 { -1, NULL } /* End of list */
97};
98
99/* Multipliers for offsets within the PTEs */
100#define PTE_LEVEL_MULT (PAGE_SIZE)
101#define PMD_LEVEL_MULT (PTRS_PER_PTE * PTE_LEVEL_MULT)
102#define PUD_LEVEL_MULT (PTRS_PER_PMD * PMD_LEVEL_MULT)
103#define PGD_LEVEL_MULT (PTRS_PER_PUD * PUD_LEVEL_MULT)
104
105#define pt_dump_seq_printf(m, to_dmesg, fmt, args...) \
106({ \
107 if (to_dmesg) \
108 printk(KERN_INFO fmt, ##args); \
109 else \
110 if (m) \
111 seq_printf(m, fmt, ##args); \
112})
113
114#define pt_dump_cont_printf(m, to_dmesg, fmt, args...) \
115({ \
116 if (to_dmesg) \
117 printk(KERN_CONT fmt, ##args); \
118 else \
119 if (m) \
120 seq_printf(m, fmt, ##args); \
121})
122
123/*
124 * Print a readable form of a pgprot_t to the seq_file
125 */
126static void printk_prot(struct seq_file *m, pgprot_t prot, int level, bool dmsg)
127{
128 pgprotval_t pr = pgprot_val(prot);
129 static const char * const level_name[] =
130 { "cr3", "pgd", "pud", "pmd", "pte" };
131
132 if (!pgprot_val(prot)) {
133 /* Not present */
134 pt_dump_cont_printf(m, dmsg, " ");
135 } else {
136 if (pr & _PAGE_USER)
137 pt_dump_cont_printf(m, dmsg, "USR ");
138 else
139 pt_dump_cont_printf(m, dmsg, " ");
140 if (pr & _PAGE_RW)
141 pt_dump_cont_printf(m, dmsg, "RW ");
142 else
143 pt_dump_cont_printf(m, dmsg, "ro ");
144 if (pr & _PAGE_PWT)
145 pt_dump_cont_printf(m, dmsg, "PWT ");
146 else
147 pt_dump_cont_printf(m, dmsg, " ");
148 if (pr & _PAGE_PCD)
149 pt_dump_cont_printf(m, dmsg, "PCD ");
150 else
151 pt_dump_cont_printf(m, dmsg, " ");
152
153 /* Bit 7 has a different meaning on level 3 vs 4 */
154 if (level <= 3 && pr & _PAGE_PSE)
155 pt_dump_cont_printf(m, dmsg, "PSE ");
156 else
157 pt_dump_cont_printf(m, dmsg, " ");
158 if ((level == 4 && pr & _PAGE_PAT) ||
159 ((level == 3 || level == 2) && pr & _PAGE_PAT_LARGE))
160 pt_dump_cont_printf(m, dmsg, "PAT ");
161 else
162 pt_dump_cont_printf(m, dmsg, " ");
163 if (pr & _PAGE_GLOBAL)
164 pt_dump_cont_printf(m, dmsg, "GLB ");
165 else
166 pt_dump_cont_printf(m, dmsg, " ");
167 if (pr & _PAGE_NX)
168 pt_dump_cont_printf(m, dmsg, "NX ");
169 else
170 pt_dump_cont_printf(m, dmsg, "x ");
171 }
172 pt_dump_cont_printf(m, dmsg, "%s\n", level_name[level]);
173}
174
175/*
176 * On 64 bits, sign-extend the 48 bit address to 64 bit
177 */
178static unsigned long normalize_addr(unsigned long u)
179{
180#ifdef CONFIG_X86_64
181 return (signed long)(u << 16) >> 16;
182#else
183 return u;
184#endif
185}
186
187/*
188 * This function gets called on a break in a continuous series
189 * of PTE entries; the next one is different so we need to
190 * print what we collected so far.
191 */
192static void note_page(struct seq_file *m, struct pg_state *st,
193 pgprot_t new_prot, int level)
194{
195 pgprotval_t prot, cur;
196 static const char units[] = "BKMGTPE";
197
198 /*
199 * If we have a "break" in the series, we need to flush the state that
200 * we have now. "break" is either changing perms, levels or
201 * address space marker.
202 */
203 prot = pgprot_val(new_prot);
204 cur = pgprot_val(st->current_prot);
205
206 if (!st->level) {
207 /* First entry */
208 st->current_prot = new_prot;
209 st->level = level;
210 st->marker = address_markers;
211 st->lines = 0;
212 pt_dump_seq_printf(m, st->to_dmesg, "---[ %s ]---\n",
213 st->marker->name);
214 } else if (prot != cur || level != st->level ||
215 st->current_address >= st->marker[1].start_address) {
216 const char *unit = units;
217 unsigned long delta;
218 int width = sizeof(unsigned long) * 2;
219 pgprotval_t pr = pgprot_val(st->current_prot);
220
221 if (st->check_wx && (pr & _PAGE_RW) && !(pr & _PAGE_NX)) {
222 WARN_ONCE(1,
223 "x86/mm: Found insecure W+X mapping at address %p/%pS\n",
224 (void *)st->start_address,
225 (void *)st->start_address);
226 st->wx_pages += (st->current_address -
227 st->start_address) / PAGE_SIZE;
228 }
229
230 /*
231 * Now print the actual finished series
232 */
233 if (!st->marker->max_lines ||
234 st->lines < st->marker->max_lines) {
235 pt_dump_seq_printf(m, st->to_dmesg,
236 "0x%0*lx-0x%0*lx ",
237 width, st->start_address,
238 width, st->current_address);
239
240 delta = st->current_address - st->start_address;
241 while (!(delta & 1023) && unit[1]) {
242 delta >>= 10;
243 unit++;
244 }
245 pt_dump_cont_printf(m, st->to_dmesg, "%9lu%c ",
246 delta, *unit);
247 printk_prot(m, st->current_prot, st->level,
248 st->to_dmesg);
249 }
250 st->lines++;
251
252 /*
253 * We print markers for special areas of address space,
254 * such as the start of vmalloc space etc.
255 * This helps in the interpretation.
256 */
257 if (st->current_address >= st->marker[1].start_address) {
258 if (st->marker->max_lines &&
259 st->lines > st->marker->max_lines) {
260 unsigned long nskip =
261 st->lines - st->marker->max_lines;
262 pt_dump_seq_printf(m, st->to_dmesg,
263 "... %lu entr%s skipped ... \n",
264 nskip,
265 nskip == 1 ? "y" : "ies");
266 }
267 st->marker++;
268 st->lines = 0;
269 pt_dump_seq_printf(m, st->to_dmesg, "---[ %s ]---\n",
270 st->marker->name);
271 }
272
273 st->start_address = st->current_address;
274 st->current_prot = new_prot;
275 st->level = level;
276 }
277}
278
279static void walk_pte_level(struct seq_file *m, struct pg_state *st, pmd_t addr,
280 unsigned long P)
281{
282 int i;
283 pte_t *start;
284 pgprotval_t prot;
285
286 start = (pte_t *) pmd_page_vaddr(addr);
287 for (i = 0; i < PTRS_PER_PTE; i++) {
288 prot = pte_flags(*start);
289 st->current_address = normalize_addr(P + i * PTE_LEVEL_MULT);
290 note_page(m, st, __pgprot(prot), 4);
291 start++;
292 }
293}
294
295#if PTRS_PER_PMD > 1
296
297static void walk_pmd_level(struct seq_file *m, struct pg_state *st, pud_t addr,
298 unsigned long P)
299{
300 int i;
301 pmd_t *start;
302 pgprotval_t prot;
303
304 start = (pmd_t *) pud_page_vaddr(addr);
305 for (i = 0; i < PTRS_PER_PMD; i++) {
306 st->current_address = normalize_addr(P + i * PMD_LEVEL_MULT);
307 if (!pmd_none(*start)) {
308 if (pmd_large(*start) || !pmd_present(*start)) {
309 prot = pmd_flags(*start);
310 note_page(m, st, __pgprot(prot), 3);
311 } else {
312 walk_pte_level(m, st, *start,
313 P + i * PMD_LEVEL_MULT);
314 }
315 } else
316 note_page(m, st, __pgprot(0), 3);
317 start++;
318 }
319}
320
321#else
322#define walk_pmd_level(m,s,a,p) walk_pte_level(m,s,__pmd(pud_val(a)),p)
323#define pud_large(a) pmd_large(__pmd(pud_val(a)))
324#define pud_none(a) pmd_none(__pmd(pud_val(a)))
325#endif
326
327#if PTRS_PER_PUD > 1
328
329static void walk_pud_level(struct seq_file *m, struct pg_state *st, pgd_t addr,
330 unsigned long P)
331{
332 int i;
333 pud_t *start;
334 pgprotval_t prot;
335
336 start = (pud_t *) pgd_page_vaddr(addr);
337
338 for (i = 0; i < PTRS_PER_PUD; i++) {
339 st->current_address = normalize_addr(P + i * PUD_LEVEL_MULT);
340 if (!pud_none(*start)) {
341 if (pud_large(*start) || !pud_present(*start)) {
342 prot = pud_flags(*start);
343 note_page(m, st, __pgprot(prot), 2);
344 } else {
345 walk_pmd_level(m, st, *start,
346 P + i * PUD_LEVEL_MULT);
347 }
348 } else
349 note_page(m, st, __pgprot(0), 2);
350
351 start++;
352 }
353}
354
355#else
356#define walk_pud_level(m,s,a,p) walk_pmd_level(m,s,__pud(pgd_val(a)),p)
357#define pgd_large(a) pud_large(__pud(pgd_val(a)))
358#define pgd_none(a) pud_none(__pud(pgd_val(a)))
359#endif
360
361static inline bool is_hypervisor_range(int idx)
362{
363#ifdef CONFIG_X86_64
364 /*
365 * ffff800000000000 - ffff87ffffffffff is reserved for
366 * the hypervisor.
367 */
368 return (idx >= pgd_index(__PAGE_OFFSET) - 16) &&
369 (idx < pgd_index(__PAGE_OFFSET));
370#else
371 return false;
372#endif
373}
374
375static void ptdump_walk_pgd_level_core(struct seq_file *m, pgd_t *pgd,
376 bool checkwx)
377{
378#ifdef CONFIG_X86_64
379 pgd_t *start = (pgd_t *) &init_level4_pgt;
380#else
381 pgd_t *start = swapper_pg_dir;
382#endif
383 pgprotval_t prot;
384 int i;
385 struct pg_state st = {};
386
387 if (pgd) {
388 start = pgd;
389 st.to_dmesg = true;
390 }
391
392 st.check_wx = checkwx;
393 if (checkwx)
394 st.wx_pages = 0;
395
396 for (i = 0; i < PTRS_PER_PGD; i++) {
397 st.current_address = normalize_addr(i * PGD_LEVEL_MULT);
398 if (!pgd_none(*start) && !is_hypervisor_range(i)) {
399 if (pgd_large(*start) || !pgd_present(*start)) {
400 prot = pgd_flags(*start);
401 note_page(m, &st, __pgprot(prot), 1);
402 } else {
403 walk_pud_level(m, &st, *start,
404 i * PGD_LEVEL_MULT);
405 }
406 } else
407 note_page(m, &st, __pgprot(0), 1);
408
409 start++;
410 }
411
412 /* Flush out the last page */
413 st.current_address = normalize_addr(PTRS_PER_PGD*PGD_LEVEL_MULT);
414 note_page(m, &st, __pgprot(0), 0);
415 if (!checkwx)
416 return;
417 if (st.wx_pages)
418 pr_info("x86/mm: Checked W+X mappings: FAILED, %lu W+X pages found.\n",
419 st.wx_pages);
420 else
421 pr_info("x86/mm: Checked W+X mappings: passed, no W+X pages found.\n");
422}
423
424void ptdump_walk_pgd_level(struct seq_file *m, pgd_t *pgd)
425{
426 ptdump_walk_pgd_level_core(m, pgd, false);
427}
428EXPORT_SYMBOL_GPL(ptdump_walk_pgd_level);
429
430void ptdump_walk_pgd_level_checkwx(void)
431{
432 ptdump_walk_pgd_level_core(NULL, NULL, true);
433}
434
435static int __init pt_dump_init(void)
436{
437#ifdef CONFIG_X86_32
438 /* Not a compile-time constant on x86-32 */
439 address_markers[VMALLOC_START_NR].start_address = VMALLOC_START;
440 address_markers[VMALLOC_END_NR].start_address = VMALLOC_END;
441# ifdef CONFIG_HIGHMEM
442 address_markers[PKMAP_BASE_NR].start_address = PKMAP_BASE;
443# endif
444 address_markers[FIXADDR_START_NR].start_address = FIXADDR_START;
445#endif
446
447 return 0;
448}
449
450__initcall(pt_dump_init);
451MODULE_LICENSE("GPL");
452MODULE_AUTHOR("Arjan van de Ven <arjan@linux.intel.com>");
453MODULE_DESCRIPTION("Kernel debugging helper that dumps pagetables");
1// SPDX-License-Identifier: GPL-2.0-only
2/*
3 * Debug helper to dump the current kernel pagetables of the system
4 * so that we can see what the various memory ranges are set to.
5 *
6 * (C) Copyright 2008 Intel Corporation
7 *
8 * Author: Arjan van de Ven <arjan@linux.intel.com>
9 */
10
11#include <linux/debugfs.h>
12#include <linux/kasan.h>
13#include <linux/mm.h>
14#include <linux/init.h>
15#include <linux/sched.h>
16#include <linux/seq_file.h>
17#include <linux/highmem.h>
18#include <linux/pci.h>
19
20#include <asm/e820/types.h>
21#include <asm/pgtable.h>
22
23/*
24 * The dumper groups pagetable entries of the same type into one, and for
25 * that it needs to keep some state when walking, and flush this state
26 * when a "break" in the continuity is found.
27 */
28struct pg_state {
29 int level;
30 pgprot_t current_prot;
31 pgprotval_t effective_prot;
32 unsigned long start_address;
33 unsigned long current_address;
34 const struct addr_marker *marker;
35 unsigned long lines;
36 bool to_dmesg;
37 bool check_wx;
38 unsigned long wx_pages;
39};
40
41struct addr_marker {
42 unsigned long start_address;
43 const char *name;
44 unsigned long max_lines;
45};
46
47/* Address space markers hints */
48
49#ifdef CONFIG_X86_64
50
51enum address_markers_idx {
52 USER_SPACE_NR = 0,
53 KERNEL_SPACE_NR,
54#ifdef CONFIG_MODIFY_LDT_SYSCALL
55 LDT_NR,
56#endif
57 LOW_KERNEL_NR,
58 VMALLOC_START_NR,
59 VMEMMAP_START_NR,
60#ifdef CONFIG_KASAN
61 KASAN_SHADOW_START_NR,
62 KASAN_SHADOW_END_NR,
63#endif
64 CPU_ENTRY_AREA_NR,
65#ifdef CONFIG_X86_ESPFIX64
66 ESPFIX_START_NR,
67#endif
68#ifdef CONFIG_EFI
69 EFI_END_NR,
70#endif
71 HIGH_KERNEL_NR,
72 MODULES_VADDR_NR,
73 MODULES_END_NR,
74 FIXADDR_START_NR,
75 END_OF_SPACE_NR,
76};
77
78static struct addr_marker address_markers[] = {
79 [USER_SPACE_NR] = { 0, "User Space" },
80 [KERNEL_SPACE_NR] = { (1UL << 63), "Kernel Space" },
81 [LOW_KERNEL_NR] = { 0UL, "Low Kernel Mapping" },
82 [VMALLOC_START_NR] = { 0UL, "vmalloc() Area" },
83 [VMEMMAP_START_NR] = { 0UL, "Vmemmap" },
84#ifdef CONFIG_KASAN
85 /*
86 * These fields get initialized with the (dynamic)
87 * KASAN_SHADOW_{START,END} values in pt_dump_init().
88 */
89 [KASAN_SHADOW_START_NR] = { 0UL, "KASAN shadow" },
90 [KASAN_SHADOW_END_NR] = { 0UL, "KASAN shadow end" },
91#endif
92#ifdef CONFIG_MODIFY_LDT_SYSCALL
93 [LDT_NR] = { 0UL, "LDT remap" },
94#endif
95 [CPU_ENTRY_AREA_NR] = { CPU_ENTRY_AREA_BASE,"CPU entry Area" },
96#ifdef CONFIG_X86_ESPFIX64
97 [ESPFIX_START_NR] = { ESPFIX_BASE_ADDR, "ESPfix Area", 16 },
98#endif
99#ifdef CONFIG_EFI
100 [EFI_END_NR] = { EFI_VA_END, "EFI Runtime Services" },
101#endif
102 [HIGH_KERNEL_NR] = { __START_KERNEL_map, "High Kernel Mapping" },
103 [MODULES_VADDR_NR] = { MODULES_VADDR, "Modules" },
104 [MODULES_END_NR] = { MODULES_END, "End Modules" },
105 [FIXADDR_START_NR] = { FIXADDR_START, "Fixmap Area" },
106 [END_OF_SPACE_NR] = { -1, NULL }
107};
108
109#define INIT_PGD ((pgd_t *) &init_top_pgt)
110
111#else /* CONFIG_X86_64 */
112
113enum address_markers_idx {
114 USER_SPACE_NR = 0,
115 KERNEL_SPACE_NR,
116 VMALLOC_START_NR,
117 VMALLOC_END_NR,
118#ifdef CONFIG_HIGHMEM
119 PKMAP_BASE_NR,
120#endif
121#ifdef CONFIG_MODIFY_LDT_SYSCALL
122 LDT_NR,
123#endif
124 CPU_ENTRY_AREA_NR,
125 FIXADDR_START_NR,
126 END_OF_SPACE_NR,
127};
128
129static struct addr_marker address_markers[] = {
130 [USER_SPACE_NR] = { 0, "User Space" },
131 [KERNEL_SPACE_NR] = { PAGE_OFFSET, "Kernel Mapping" },
132 [VMALLOC_START_NR] = { 0UL, "vmalloc() Area" },
133 [VMALLOC_END_NR] = { 0UL, "vmalloc() End" },
134#ifdef CONFIG_HIGHMEM
135 [PKMAP_BASE_NR] = { 0UL, "Persistent kmap() Area" },
136#endif
137#ifdef CONFIG_MODIFY_LDT_SYSCALL
138 [LDT_NR] = { 0UL, "LDT remap" },
139#endif
140 [CPU_ENTRY_AREA_NR] = { 0UL, "CPU entry area" },
141 [FIXADDR_START_NR] = { 0UL, "Fixmap area" },
142 [END_OF_SPACE_NR] = { -1, NULL }
143};
144
145#define INIT_PGD (swapper_pg_dir)
146
147#endif /* !CONFIG_X86_64 */
148
149/* Multipliers for offsets within the PTEs */
150#define PTE_LEVEL_MULT (PAGE_SIZE)
151#define PMD_LEVEL_MULT (PTRS_PER_PTE * PTE_LEVEL_MULT)
152#define PUD_LEVEL_MULT (PTRS_PER_PMD * PMD_LEVEL_MULT)
153#define P4D_LEVEL_MULT (PTRS_PER_PUD * PUD_LEVEL_MULT)
154#define PGD_LEVEL_MULT (PTRS_PER_P4D * P4D_LEVEL_MULT)
155
156#define pt_dump_seq_printf(m, to_dmesg, fmt, args...) \
157({ \
158 if (to_dmesg) \
159 printk(KERN_INFO fmt, ##args); \
160 else \
161 if (m) \
162 seq_printf(m, fmt, ##args); \
163})
164
165#define pt_dump_cont_printf(m, to_dmesg, fmt, args...) \
166({ \
167 if (to_dmesg) \
168 printk(KERN_CONT fmt, ##args); \
169 else \
170 if (m) \
171 seq_printf(m, fmt, ##args); \
172})
173
174/*
175 * Print a readable form of a pgprot_t to the seq_file
176 */
177static void printk_prot(struct seq_file *m, pgprot_t prot, int level, bool dmsg)
178{
179 pgprotval_t pr = pgprot_val(prot);
180 static const char * const level_name[] =
181 { "cr3", "pgd", "p4d", "pud", "pmd", "pte" };
182
183 if (!(pr & _PAGE_PRESENT)) {
184 /* Not present */
185 pt_dump_cont_printf(m, dmsg, " ");
186 } else {
187 if (pr & _PAGE_USER)
188 pt_dump_cont_printf(m, dmsg, "USR ");
189 else
190 pt_dump_cont_printf(m, dmsg, " ");
191 if (pr & _PAGE_RW)
192 pt_dump_cont_printf(m, dmsg, "RW ");
193 else
194 pt_dump_cont_printf(m, dmsg, "ro ");
195 if (pr & _PAGE_PWT)
196 pt_dump_cont_printf(m, dmsg, "PWT ");
197 else
198 pt_dump_cont_printf(m, dmsg, " ");
199 if (pr & _PAGE_PCD)
200 pt_dump_cont_printf(m, dmsg, "PCD ");
201 else
202 pt_dump_cont_printf(m, dmsg, " ");
203
204 /* Bit 7 has a different meaning on level 3 vs 4 */
205 if (level <= 4 && pr & _PAGE_PSE)
206 pt_dump_cont_printf(m, dmsg, "PSE ");
207 else
208 pt_dump_cont_printf(m, dmsg, " ");
209 if ((level == 5 && pr & _PAGE_PAT) ||
210 ((level == 4 || level == 3) && pr & _PAGE_PAT_LARGE))
211 pt_dump_cont_printf(m, dmsg, "PAT ");
212 else
213 pt_dump_cont_printf(m, dmsg, " ");
214 if (pr & _PAGE_GLOBAL)
215 pt_dump_cont_printf(m, dmsg, "GLB ");
216 else
217 pt_dump_cont_printf(m, dmsg, " ");
218 if (pr & _PAGE_NX)
219 pt_dump_cont_printf(m, dmsg, "NX ");
220 else
221 pt_dump_cont_printf(m, dmsg, "x ");
222 }
223 pt_dump_cont_printf(m, dmsg, "%s\n", level_name[level]);
224}
225
226/*
227 * On 64 bits, sign-extend the 48 bit address to 64 bit
228 */
229static unsigned long normalize_addr(unsigned long u)
230{
231 int shift;
232 if (!IS_ENABLED(CONFIG_X86_64))
233 return u;
234
235 shift = 64 - (__VIRTUAL_MASK_SHIFT + 1);
236 return (signed long)(u << shift) >> shift;
237}
238
239static void note_wx(struct pg_state *st)
240{
241 unsigned long npages;
242
243 npages = (st->current_address - st->start_address) / PAGE_SIZE;
244
245#ifdef CONFIG_PCI_BIOS
246 /*
247 * If PCI BIOS is enabled, the PCI BIOS area is forced to WX.
248 * Inform about it, but avoid the warning.
249 */
250 if (pcibios_enabled && st->start_address >= PAGE_OFFSET + BIOS_BEGIN &&
251 st->current_address <= PAGE_OFFSET + BIOS_END) {
252 pr_warn_once("x86/mm: PCI BIOS W+X mapping %lu pages\n", npages);
253 return;
254 }
255#endif
256 /* Account the WX pages */
257 st->wx_pages += npages;
258 WARN_ONCE(__supported_pte_mask & _PAGE_NX,
259 "x86/mm: Found insecure W+X mapping at address %pS\n",
260 (void *)st->start_address);
261}
262
263/*
264 * This function gets called on a break in a continuous series
265 * of PTE entries; the next one is different so we need to
266 * print what we collected so far.
267 */
268static void note_page(struct seq_file *m, struct pg_state *st,
269 pgprot_t new_prot, pgprotval_t new_eff, int level)
270{
271 pgprotval_t prot, cur, eff;
272 static const char units[] = "BKMGTPE";
273
274 /*
275 * If we have a "break" in the series, we need to flush the state that
276 * we have now. "break" is either changing perms, levels or
277 * address space marker.
278 */
279 prot = pgprot_val(new_prot);
280 cur = pgprot_val(st->current_prot);
281 eff = st->effective_prot;
282
283 if (!st->level) {
284 /* First entry */
285 st->current_prot = new_prot;
286 st->effective_prot = new_eff;
287 st->level = level;
288 st->marker = address_markers;
289 st->lines = 0;
290 pt_dump_seq_printf(m, st->to_dmesg, "---[ %s ]---\n",
291 st->marker->name);
292 } else if (prot != cur || new_eff != eff || level != st->level ||
293 st->current_address >= st->marker[1].start_address) {
294 const char *unit = units;
295 unsigned long delta;
296 int width = sizeof(unsigned long) * 2;
297
298 if (st->check_wx && (eff & _PAGE_RW) && !(eff & _PAGE_NX))
299 note_wx(st);
300
301 /*
302 * Now print the actual finished series
303 */
304 if (!st->marker->max_lines ||
305 st->lines < st->marker->max_lines) {
306 pt_dump_seq_printf(m, st->to_dmesg,
307 "0x%0*lx-0x%0*lx ",
308 width, st->start_address,
309 width, st->current_address);
310
311 delta = st->current_address - st->start_address;
312 while (!(delta & 1023) && unit[1]) {
313 delta >>= 10;
314 unit++;
315 }
316 pt_dump_cont_printf(m, st->to_dmesg, "%9lu%c ",
317 delta, *unit);
318 printk_prot(m, st->current_prot, st->level,
319 st->to_dmesg);
320 }
321 st->lines++;
322
323 /*
324 * We print markers for special areas of address space,
325 * such as the start of vmalloc space etc.
326 * This helps in the interpretation.
327 */
328 if (st->current_address >= st->marker[1].start_address) {
329 if (st->marker->max_lines &&
330 st->lines > st->marker->max_lines) {
331 unsigned long nskip =
332 st->lines - st->marker->max_lines;
333 pt_dump_seq_printf(m, st->to_dmesg,
334 "... %lu entr%s skipped ... \n",
335 nskip,
336 nskip == 1 ? "y" : "ies");
337 }
338 st->marker++;
339 st->lines = 0;
340 pt_dump_seq_printf(m, st->to_dmesg, "---[ %s ]---\n",
341 st->marker->name);
342 }
343
344 st->start_address = st->current_address;
345 st->current_prot = new_prot;
346 st->effective_prot = new_eff;
347 st->level = level;
348 }
349}
350
351static inline pgprotval_t effective_prot(pgprotval_t prot1, pgprotval_t prot2)
352{
353 return (prot1 & prot2 & (_PAGE_USER | _PAGE_RW)) |
354 ((prot1 | prot2) & _PAGE_NX);
355}
356
357static void walk_pte_level(struct seq_file *m, struct pg_state *st, pmd_t addr,
358 pgprotval_t eff_in, unsigned long P)
359{
360 int i;
361 pte_t *pte;
362 pgprotval_t prot, eff;
363
364 for (i = 0; i < PTRS_PER_PTE; i++) {
365 st->current_address = normalize_addr(P + i * PTE_LEVEL_MULT);
366 pte = pte_offset_map(&addr, st->current_address);
367 prot = pte_flags(*pte);
368 eff = effective_prot(eff_in, prot);
369 note_page(m, st, __pgprot(prot), eff, 5);
370 pte_unmap(pte);
371 }
372}
373#ifdef CONFIG_KASAN
374
375/*
376 * This is an optimization for KASAN=y case. Since all kasan page tables
377 * eventually point to the kasan_early_shadow_page we could call note_page()
378 * right away without walking through lower level page tables. This saves
379 * us dozens of seconds (minutes for 5-level config) while checking for
380 * W+X mapping or reading kernel_page_tables debugfs file.
381 */
382static inline bool kasan_page_table(struct seq_file *m, struct pg_state *st,
383 void *pt)
384{
385 if (__pa(pt) == __pa(kasan_early_shadow_pmd) ||
386 (pgtable_l5_enabled() &&
387 __pa(pt) == __pa(kasan_early_shadow_p4d)) ||
388 __pa(pt) == __pa(kasan_early_shadow_pud)) {
389 pgprotval_t prot = pte_flags(kasan_early_shadow_pte[0]);
390 note_page(m, st, __pgprot(prot), 0, 5);
391 return true;
392 }
393 return false;
394}
395#else
396static inline bool kasan_page_table(struct seq_file *m, struct pg_state *st,
397 void *pt)
398{
399 return false;
400}
401#endif
402
403#if PTRS_PER_PMD > 1
404
405static void walk_pmd_level(struct seq_file *m, struct pg_state *st, pud_t addr,
406 pgprotval_t eff_in, unsigned long P)
407{
408 int i;
409 pmd_t *start, *pmd_start;
410 pgprotval_t prot, eff;
411
412 pmd_start = start = (pmd_t *)pud_page_vaddr(addr);
413 for (i = 0; i < PTRS_PER_PMD; i++) {
414 st->current_address = normalize_addr(P + i * PMD_LEVEL_MULT);
415 if (!pmd_none(*start)) {
416 prot = pmd_flags(*start);
417 eff = effective_prot(eff_in, prot);
418 if (pmd_large(*start) || !pmd_present(*start)) {
419 note_page(m, st, __pgprot(prot), eff, 4);
420 } else if (!kasan_page_table(m, st, pmd_start)) {
421 walk_pte_level(m, st, *start, eff,
422 P + i * PMD_LEVEL_MULT);
423 }
424 } else
425 note_page(m, st, __pgprot(0), 0, 4);
426 start++;
427 }
428}
429
430#else
431#define walk_pmd_level(m,s,a,e,p) walk_pte_level(m,s,__pmd(pud_val(a)),e,p)
432#define pud_large(a) pmd_large(__pmd(pud_val(a)))
433#define pud_none(a) pmd_none(__pmd(pud_val(a)))
434#endif
435
436#if PTRS_PER_PUD > 1
437
438static void walk_pud_level(struct seq_file *m, struct pg_state *st, p4d_t addr,
439 pgprotval_t eff_in, unsigned long P)
440{
441 int i;
442 pud_t *start, *pud_start;
443 pgprotval_t prot, eff;
444
445 pud_start = start = (pud_t *)p4d_page_vaddr(addr);
446
447 for (i = 0; i < PTRS_PER_PUD; i++) {
448 st->current_address = normalize_addr(P + i * PUD_LEVEL_MULT);
449 if (!pud_none(*start)) {
450 prot = pud_flags(*start);
451 eff = effective_prot(eff_in, prot);
452 if (pud_large(*start) || !pud_present(*start)) {
453 note_page(m, st, __pgprot(prot), eff, 3);
454 } else if (!kasan_page_table(m, st, pud_start)) {
455 walk_pmd_level(m, st, *start, eff,
456 P + i * PUD_LEVEL_MULT);
457 }
458 } else
459 note_page(m, st, __pgprot(0), 0, 3);
460
461 start++;
462 }
463}
464
465#else
466#define walk_pud_level(m,s,a,e,p) walk_pmd_level(m,s,__pud(p4d_val(a)),e,p)
467#define p4d_large(a) pud_large(__pud(p4d_val(a)))
468#define p4d_none(a) pud_none(__pud(p4d_val(a)))
469#endif
470
471static void walk_p4d_level(struct seq_file *m, struct pg_state *st, pgd_t addr,
472 pgprotval_t eff_in, unsigned long P)
473{
474 int i;
475 p4d_t *start, *p4d_start;
476 pgprotval_t prot, eff;
477
478 if (PTRS_PER_P4D == 1)
479 return walk_pud_level(m, st, __p4d(pgd_val(addr)), eff_in, P);
480
481 p4d_start = start = (p4d_t *)pgd_page_vaddr(addr);
482
483 for (i = 0; i < PTRS_PER_P4D; i++) {
484 st->current_address = normalize_addr(P + i * P4D_LEVEL_MULT);
485 if (!p4d_none(*start)) {
486 prot = p4d_flags(*start);
487 eff = effective_prot(eff_in, prot);
488 if (p4d_large(*start) || !p4d_present(*start)) {
489 note_page(m, st, __pgprot(prot), eff, 2);
490 } else if (!kasan_page_table(m, st, p4d_start)) {
491 walk_pud_level(m, st, *start, eff,
492 P + i * P4D_LEVEL_MULT);
493 }
494 } else
495 note_page(m, st, __pgprot(0), 0, 2);
496
497 start++;
498 }
499}
500
501#define pgd_large(a) (pgtable_l5_enabled() ? pgd_large(a) : p4d_large(__p4d(pgd_val(a))))
502#define pgd_none(a) (pgtable_l5_enabled() ? pgd_none(a) : p4d_none(__p4d(pgd_val(a))))
503
504static inline bool is_hypervisor_range(int idx)
505{
506#ifdef CONFIG_X86_64
507 /*
508 * A hole in the beginning of kernel address space reserved
509 * for a hypervisor.
510 */
511 return (idx >= pgd_index(GUARD_HOLE_BASE_ADDR)) &&
512 (idx < pgd_index(GUARD_HOLE_END_ADDR));
513#else
514 return false;
515#endif
516}
517
518static void ptdump_walk_pgd_level_core(struct seq_file *m, pgd_t *pgd,
519 bool checkwx, bool dmesg)
520{
521 pgd_t *start = INIT_PGD;
522 pgprotval_t prot, eff;
523 int i;
524 struct pg_state st = {};
525
526 if (pgd) {
527 start = pgd;
528 st.to_dmesg = dmesg;
529 }
530
531 st.check_wx = checkwx;
532 if (checkwx)
533 st.wx_pages = 0;
534
535 for (i = 0; i < PTRS_PER_PGD; i++) {
536 st.current_address = normalize_addr(i * PGD_LEVEL_MULT);
537 if (!pgd_none(*start) && !is_hypervisor_range(i)) {
538 prot = pgd_flags(*start);
539#ifdef CONFIG_X86_PAE
540 eff = _PAGE_USER | _PAGE_RW;
541#else
542 eff = prot;
543#endif
544 if (pgd_large(*start) || !pgd_present(*start)) {
545 note_page(m, &st, __pgprot(prot), eff, 1);
546 } else {
547 walk_p4d_level(m, &st, *start, eff,
548 i * PGD_LEVEL_MULT);
549 }
550 } else
551 note_page(m, &st, __pgprot(0), 0, 1);
552
553 cond_resched();
554 start++;
555 }
556
557 /* Flush out the last page */
558 st.current_address = normalize_addr(PTRS_PER_PGD*PGD_LEVEL_MULT);
559 note_page(m, &st, __pgprot(0), 0, 0);
560 if (!checkwx)
561 return;
562 if (st.wx_pages)
563 pr_info("x86/mm: Checked W+X mappings: FAILED, %lu W+X pages found.\n",
564 st.wx_pages);
565 else
566 pr_info("x86/mm: Checked W+X mappings: passed, no W+X pages found.\n");
567}
568
569void ptdump_walk_pgd_level(struct seq_file *m, pgd_t *pgd)
570{
571 ptdump_walk_pgd_level_core(m, pgd, false, true);
572}
573
574void ptdump_walk_pgd_level_debugfs(struct seq_file *m, pgd_t *pgd, bool user)
575{
576#ifdef CONFIG_PAGE_TABLE_ISOLATION
577 if (user && boot_cpu_has(X86_FEATURE_PTI))
578 pgd = kernel_to_user_pgdp(pgd);
579#endif
580 ptdump_walk_pgd_level_core(m, pgd, false, false);
581}
582EXPORT_SYMBOL_GPL(ptdump_walk_pgd_level_debugfs);
583
584void ptdump_walk_user_pgd_level_checkwx(void)
585{
586#ifdef CONFIG_PAGE_TABLE_ISOLATION
587 pgd_t *pgd = INIT_PGD;
588
589 if (!(__supported_pte_mask & _PAGE_NX) ||
590 !boot_cpu_has(X86_FEATURE_PTI))
591 return;
592
593 pr_info("x86/mm: Checking user space page tables\n");
594 pgd = kernel_to_user_pgdp(pgd);
595 ptdump_walk_pgd_level_core(NULL, pgd, true, false);
596#endif
597}
598
599void ptdump_walk_pgd_level_checkwx(void)
600{
601 ptdump_walk_pgd_level_core(NULL, NULL, true, false);
602}
603
604static int __init pt_dump_init(void)
605{
606 /*
607 * Various markers are not compile-time constants, so assign them
608 * here.
609 */
610#ifdef CONFIG_X86_64
611 address_markers[LOW_KERNEL_NR].start_address = PAGE_OFFSET;
612 address_markers[VMALLOC_START_NR].start_address = VMALLOC_START;
613 address_markers[VMEMMAP_START_NR].start_address = VMEMMAP_START;
614#ifdef CONFIG_MODIFY_LDT_SYSCALL
615 address_markers[LDT_NR].start_address = LDT_BASE_ADDR;
616#endif
617#ifdef CONFIG_KASAN
618 address_markers[KASAN_SHADOW_START_NR].start_address = KASAN_SHADOW_START;
619 address_markers[KASAN_SHADOW_END_NR].start_address = KASAN_SHADOW_END;
620#endif
621#endif
622#ifdef CONFIG_X86_32
623 address_markers[VMALLOC_START_NR].start_address = VMALLOC_START;
624 address_markers[VMALLOC_END_NR].start_address = VMALLOC_END;
625# ifdef CONFIG_HIGHMEM
626 address_markers[PKMAP_BASE_NR].start_address = PKMAP_BASE;
627# endif
628 address_markers[FIXADDR_START_NR].start_address = FIXADDR_START;
629 address_markers[CPU_ENTRY_AREA_NR].start_address = CPU_ENTRY_AREA_BASE;
630# ifdef CONFIG_MODIFY_LDT_SYSCALL
631 address_markers[LDT_NR].start_address = LDT_BASE_ADDR;
632# endif
633#endif
634 return 0;
635}
636__initcall(pt_dump_init);