dump_pagetables.c

  1// SPDX-License-Identifier: GPL-2.0
  2
  3#include <linux/cpufeature.h>
  4#include <linux/set_memory.h>
  5#include <linux/ptdump.h>
  6#include <linux/seq_file.h>
  7#include <linux/debugfs.h>
  8#include <linux/sort.h>
  9#include <linux/mm.h>
 10#include <linux/kfence.h>
 11#include <linux/kasan.h>
 12#include <asm/kasan.h>
 13#include <asm/abs_lowcore.h>
 14#include <asm/nospec-branch.h>
 15#include <asm/sections.h>
 16#include <asm/maccess.h>
 17
 18static unsigned long max_addr;
 19
 20struct addr_marker {
 21	int is_start;
 22	unsigned long start_address;
 23	unsigned long size;
 24	const char *name;
 25};
 26
 27static struct addr_marker *markers;
 28static unsigned int markers_cnt;
 29
 30struct pg_state {
 31	struct ptdump_state ptdump;
 32	struct seq_file *seq;
 33	int level;
 34	unsigned int current_prot;
 35	bool check_wx;
 36	unsigned long wx_pages;
 37	unsigned long start_address;
 38	const struct addr_marker *marker;
 39};
 40
 41#define pt_dump_seq_printf(m, fmt, args...)	\
 42({						\
 43	struct seq_file *__m = (m);		\
 44						\
 45	if (__m)				\
 46		seq_printf(__m, fmt, ##args);	\
 47})
 48
 49#define pt_dump_seq_puts(m, fmt)		\
 50({						\
 51	struct seq_file *__m = (m);		\
 52						\
 53	if (__m)				\
 54		seq_puts(__m, fmt);		\
 55})
 56
 57static void print_prot(struct seq_file *m, unsigned int pr, int level)
 58{
 59	static const char * const level_name[] =
 60		{ "ASCE", "PGD", "PUD", "PMD", "PTE" };
 61
 62	pt_dump_seq_printf(m, "%s ", level_name[level]);
 63	if (pr & _PAGE_INVALID) {
 64		pt_dump_seq_printf(m, "I\n");
 65		return;
 66	}
 67	pt_dump_seq_puts(m, (pr & _PAGE_PROTECT) ? "RO " : "RW ");
 68	pt_dump_seq_puts(m, (pr & _PAGE_NOEXEC) ? "NX\n" : "X\n");
 69}
 70
 71static void note_prot_wx(struct pg_state *st, unsigned long addr)
 72{
 73	if (!st->check_wx)
 74		return;
 75	if (st->current_prot & _PAGE_INVALID)
 76		return;
 77	if (st->current_prot & _PAGE_PROTECT)
 78		return;
 79	if (st->current_prot & _PAGE_NOEXEC)
 80		return;
 81	/*
 82	 * The first lowcore page is W+X if spectre mitigations are using
 83	 * trampolines or the BEAR enhancements facility is not installed,
 84	 * in which case we have two lpswe instructions in lowcore that need
 85	 * to be executable.
 86	 */
 87	if (addr == PAGE_SIZE && (nospec_uses_trampoline() || !cpu_has_bear()))
 88		return;
 89	WARN_ONCE(IS_ENABLED(CONFIG_DEBUG_WX),
 90		  "s390/mm: Found insecure W+X mapping at address %pS\n",
 91		  (void *)st->start_address);
 92	st->wx_pages += (addr - st->start_address) / PAGE_SIZE;
 93}
 94
 95static void note_page_update_state(struct pg_state *st, unsigned long addr, unsigned int prot, int level)
 96{
 97	struct seq_file *m = st->seq;
 98
 99	while (addr >= st->marker[1].start_address) {
100		st->marker++;
101		pt_dump_seq_printf(m, "---[ %s %s ]---\n", st->marker->name,
102				   st->marker->is_start ? "Start" : "End");
103	}
104	st->start_address = addr;
105	st->current_prot = prot;
106	st->level = level;
107}
108
109static void note_page(struct ptdump_state *pt_st, unsigned long addr, int level, u64 val)
110{
111	int width = sizeof(unsigned long) * 2;
112	static const char units[] = "KMGTPE";
113	const char *unit = units;
114	unsigned long delta;
115	struct pg_state *st;
116	struct seq_file *m;
117	unsigned int prot;
118
119	st = container_of(pt_st, struct pg_state, ptdump);
120	m = st->seq;
121	prot = val & (_PAGE_PROTECT | _PAGE_NOEXEC);
122	if (level == 4 && (val & _PAGE_INVALID))
123		prot = _PAGE_INVALID;
124	/* For pmd_none() & friends val gets passed as zero. */
125	if (level != 4 && !val)
126		prot = _PAGE_INVALID;
127	/* Final flush from generic code. */
128	if (level == -1)
129		addr = max_addr;
130	if (st->level == -1) {
131		pt_dump_seq_puts(m, "---[ Kernel Virtual Address Space ]---\n");
132		note_page_update_state(st, addr, prot, level);
133	} else if (prot != st->current_prot || level != st->level ||
134		   addr >= st->marker[1].start_address) {
135		note_prot_wx(st, addr);
136		pt_dump_seq_printf(m, "0x%0*lx-0x%0*lx ",
137				   width, st->start_address,
138				   width, addr);
139		delta = (addr - st->start_address) >> 10;
140		while (!(delta & 0x3ff) && unit[1]) {
141			delta >>= 10;
142			unit++;
143		}
144		pt_dump_seq_printf(m, "%9lu%c ", delta, *unit);
145		print_prot(m, st->current_prot, st->level);
146		note_page_update_state(st, addr, prot, level);
147	}
148}
149
150static void note_page_pte(struct ptdump_state *pt_st, unsigned long addr, pte_t pte)
151{
152	note_page(pt_st, addr, 4, pte_val(pte));
153}
154
155static void note_page_pmd(struct ptdump_state *pt_st, unsigned long addr, pmd_t pmd)
156{
157	note_page(pt_st, addr, 3, pmd_val(pmd));
158}
159
160static void note_page_pud(struct ptdump_state *pt_st, unsigned long addr, pud_t pud)
161{
162	note_page(pt_st, addr, 2, pud_val(pud));
163}
164
165static void note_page_p4d(struct ptdump_state *pt_st, unsigned long addr, p4d_t p4d)
166{
167	note_page(pt_st, addr, 1, p4d_val(p4d));
168}
169
170static void note_page_pgd(struct ptdump_state *pt_st, unsigned long addr, pgd_t pgd)
171{
172	note_page(pt_st, addr, 0, pgd_val(pgd));
173}
174
175static void note_page_flush(struct ptdump_state *pt_st)
176{
177	pte_t pte_zero = {0};
178
179	note_page(pt_st, 0, -1, pte_val(pte_zero));
180}
181
182bool ptdump_check_wx(void)
183{
184	struct pg_state st = {
185		.ptdump = {
186			.note_page_pte = note_page_pte,
187			.note_page_pmd = note_page_pmd,
188			.note_page_pud = note_page_pud,
189			.note_page_p4d = note_page_p4d,
190			.note_page_pgd = note_page_pgd,
191			.note_page_flush = note_page_flush,
192			.range = (struct ptdump_range[]) {
193				{.start = 0, .end = max_addr},
194				{.start = 0, .end = 0},
195			}
196		},
197		.seq = NULL,
198		.level = -1,
199		.current_prot = 0,
200		.check_wx = true,
201		.wx_pages = 0,
202		.start_address = 0,
203		.marker = (struct addr_marker[]) {
204			{ .start_address =  0, .name = NULL},
205			{ .start_address = -1, .name = NULL},
206		},
207	};
208
209	if (!cpu_has_nx())
210		return true;
211	ptdump_walk_pgd(&st.ptdump, &init_mm, NULL);
212	if (st.wx_pages) {
213		pr_warn("Checked W+X mappings: FAILED, %lu W+X pages found\n", st.wx_pages);
214
215		return false;
216	} else {
217		pr_info("Checked W+X mappings: passed, no %sW+X pages found\n",
218			(nospec_uses_trampoline() || !cpu_has_bear()) ?
219			"unexpected " : "");
220
221		return true;
222	}
223}
224
225#ifdef CONFIG_PTDUMP_DEBUGFS
226static int ptdump_show(struct seq_file *m, void *v)
227{
228	struct pg_state st = {
229		.ptdump = {
230			.note_page_pte = note_page_pte,
231			.note_page_pmd = note_page_pmd,
232			.note_page_pud = note_page_pud,
233			.note_page_p4d = note_page_p4d,
234			.note_page_pgd = note_page_pgd,
235			.note_page_flush = note_page_flush,
236			.range = (struct ptdump_range[]) {
237				{.start = 0, .end = max_addr},
238				{.start = 0, .end = 0},
239			}
240		},
241		.seq = m,
242		.level = -1,
243		.current_prot = 0,
244		.check_wx = false,
245		.wx_pages = 0,
246		.start_address = 0,
247		.marker = markers,
248	};
249
250	mutex_lock(&cpa_mutex);
251	ptdump_walk_pgd(&st.ptdump, &init_mm, NULL);
252	mutex_unlock(&cpa_mutex);
253	return 0;
254}
255DEFINE_SHOW_ATTRIBUTE(ptdump);
256#endif /* CONFIG_PTDUMP_DEBUGFS */
257
258static int ptdump_cmp(const void *a, const void *b)
259{
260	const struct addr_marker *ama = a;
261	const struct addr_marker *amb = b;
262
263	if (ama->start_address > amb->start_address)
264		return 1;
265	if (ama->start_address < amb->start_address)
266		return -1;
267	/*
268	 * If the start addresses of two markers are identical sort markers in an
269	 * order that considers areas contained within other areas correctly.
270	 */
271	if (ama->is_start && amb->is_start) {
272		if (ama->size > amb->size)
273			return -1;
274		if (ama->size < amb->size)
275			return 1;
276		return 0;
277	}
278	if (!ama->is_start && !amb->is_start) {
279		if (ama->size > amb->size)
280			return 1;
281		if (ama->size < amb->size)
282			return -1;
283		return 0;
284	}
285	if (ama->is_start)
286		return 1;
287	if (amb->is_start)
288		return -1;
289	return 0;
290}
291
292static int add_marker(unsigned long start, unsigned long end, const char *name)
293{
294	struct addr_marker *new;
295	size_t newsize;
296
297	newsize = (markers_cnt + 2) * sizeof(*markers);
298	new = kvrealloc(markers, newsize, GFP_KERNEL);
299	if (!new)
300		return -ENOMEM;
301	markers = new;
302	markers[markers_cnt].is_start = 1;
303	markers[markers_cnt].start_address = start;
304	markers[markers_cnt].size = end - start;
305	markers[markers_cnt].name = name;
306	markers_cnt++;
307	markers[markers_cnt].is_start = 0;
308	markers[markers_cnt].start_address = end;
309	markers[markers_cnt].size = end - start;
310	markers[markers_cnt].name = name;
311	markers_cnt++;
312	return 0;
313}
314
315static int pt_dump_init(void)
316{
317#ifdef CONFIG_KFENCE
318	unsigned long kfence_start = (unsigned long)__kfence_pool;
319#endif
320	unsigned long lowcore = (unsigned long)get_lowcore();
321	int rc;
322
323	/*
324	 * Figure out the maximum virtual address being accessible with the
325	 * kernel ASCE. We need this to keep the page table walker functions
326	 * from accessing non-existent entries.
327	 */
328	max_addr = (get_lowcore()->kernel_asce.val & _REGION_ENTRY_TYPE_MASK) >> 2;
329	max_addr = 1UL << (max_addr * 11 + 31);
330	/* start + end markers - must be added first */
331	rc = add_marker(0, -1UL, NULL);
332	rc |= add_marker((unsigned long)_stext, (unsigned long)_end, "Kernel Image");
333	rc |= add_marker(lowcore, lowcore + sizeof(struct lowcore), "Lowcore");
334	rc |= add_marker(__identity_base, __identity_base + ident_map_size, "Identity Mapping");
335	rc |= add_marker((unsigned long)__samode31, (unsigned long)__eamode31, "Amode31 Area");
336	rc |= add_marker(MODULES_VADDR, MODULES_END, "Modules Area");
337	rc |= add_marker(__abs_lowcore, __abs_lowcore + ABS_LOWCORE_MAP_SIZE, "Lowcore Area");
338	rc |= add_marker(__memcpy_real_area, __memcpy_real_area + MEMCPY_REAL_SIZE, "Real Memory Copy Area");
339	rc |= add_marker((unsigned long)vmemmap, (unsigned long)vmemmap + vmemmap_size, "vmemmap Area");
340	rc |= add_marker(VMALLOC_START, VMALLOC_END, "vmalloc Area");
341#ifdef CONFIG_KFENCE
342	rc |= add_marker(kfence_start, kfence_start + KFENCE_POOL_SIZE, "KFence Pool");
343#endif
344#ifdef CONFIG_KMSAN
345	rc |= add_marker(KMSAN_VMALLOC_SHADOW_START, KMSAN_VMALLOC_SHADOW_END, "Kmsan vmalloc Shadow");
346	rc |= add_marker(KMSAN_VMALLOC_ORIGIN_START, KMSAN_VMALLOC_ORIGIN_END, "Kmsan vmalloc Origins");
347	rc |= add_marker(KMSAN_MODULES_SHADOW_START, KMSAN_MODULES_SHADOW_END, "Kmsan Modules Shadow");
348	rc |= add_marker(KMSAN_MODULES_ORIGIN_START, KMSAN_MODULES_ORIGIN_END, "Kmsan Modules Origins");
349#endif
350#ifdef CONFIG_KASAN
351	rc |= add_marker(KASAN_SHADOW_START, KASAN_SHADOW_END, "Kasan Shadow");
352#endif
353	if (rc)
354		goto error;
355	sort(&markers[1], markers_cnt - 1, sizeof(*markers), ptdump_cmp, NULL);
356#ifdef CONFIG_PTDUMP_DEBUGFS
357	debugfs_create_file("kernel_page_tables", 0400, NULL, NULL, &ptdump_fops);
358#endif /* CONFIG_PTDUMP_DEBUGFS */
359	return 0;
360error:
361	kvfree(markers);
362	return -ENOMEM;
363}
364device_initcall(pt_dump_init);