xref: /openbmc/linux/arch/arm64/kernel/mte.c (revision 5b7b41cb)
1 // SPDX-License-Identifier: GPL-2.0-only
2 /*
3  * Copyright (C) 2020 ARM Ltd.
4  */
5 
6 #include <linux/bitops.h>
7 #include <linux/kernel.h>
8 #include <linux/mm.h>
9 #include <linux/prctl.h>
10 #include <linux/sched.h>
11 #include <linux/sched/mm.h>
12 #include <linux/string.h>
13 #include <linux/swap.h>
14 #include <linux/swapops.h>
15 #include <linux/thread_info.h>
16 #include <linux/uio.h>
17 
18 #include <asm/cpufeature.h>
19 #include <asm/mte.h>
20 #include <asm/ptrace.h>
21 #include <asm/sysreg.h>
22 
23 static void mte_sync_page_tags(struct page *page, pte_t *ptep, bool check_swap)
24 {
25 	pte_t old_pte = READ_ONCE(*ptep);
26 
27 	if (check_swap && is_swap_pte(old_pte)) {
28 		swp_entry_t entry = pte_to_swp_entry(old_pte);
29 
30 		if (!non_swap_entry(entry) && mte_restore_tags(entry, page))
31 			return;
32 	}
33 
34 	mte_clear_page_tags(page_address(page));
35 }
36 
37 void mte_sync_tags(pte_t *ptep, pte_t pte)
38 {
39 	struct page *page = pte_page(pte);
40 	long i, nr_pages = compound_nr(page);
41 	bool check_swap = nr_pages == 1;
42 
43 	/* if PG_mte_tagged is set, tags have already been initialised */
44 	for (i = 0; i < nr_pages; i++, page++) {
45 		if (!test_and_set_bit(PG_mte_tagged, &page->flags))
46 			mte_sync_page_tags(page, ptep, check_swap);
47 	}
48 }
49 
50 int memcmp_pages(struct page *page1, struct page *page2)
51 {
52 	char *addr1, *addr2;
53 	int ret;
54 
55 	addr1 = page_address(page1);
56 	addr2 = page_address(page2);
57 	ret = memcmp(addr1, addr2, PAGE_SIZE);
58 
59 	if (!system_supports_mte() || ret)
60 		return ret;
61 
62 	/*
63 	 * If the page content is identical but at least one of the pages is
64 	 * tagged, return non-zero to avoid KSM merging. If only one of the
65 	 * pages is tagged, set_pte_at() may zero or change the tags of the
66 	 * other page via mte_sync_tags().
67 	 */
68 	if (test_bit(PG_mte_tagged, &page1->flags) ||
69 	    test_bit(PG_mte_tagged, &page2->flags))
70 		return addr1 != addr2;
71 
72 	return ret;
73 }
74 
75 static void update_sctlr_el1_tcf0(u64 tcf0)
76 {
77 	/* ISB required for the kernel uaccess routines */
78 	sysreg_clear_set(sctlr_el1, SCTLR_EL1_TCF0_MASK, tcf0);
79 	isb();
80 }
81 
82 static void set_sctlr_el1_tcf0(u64 tcf0)
83 {
84 	/*
85 	 * mte_thread_switch() checks current->thread.sctlr_tcf0 as an
86 	 * optimisation. Disable preemption so that it does not see
87 	 * the variable update before the SCTLR_EL1.TCF0 one.
88 	 */
89 	preempt_disable();
90 	current->thread.sctlr_tcf0 = tcf0;
91 	update_sctlr_el1_tcf0(tcf0);
92 	preempt_enable();
93 }
94 
95 static void update_gcr_el1_excl(u64 incl)
96 {
97 	u64 excl = ~incl & SYS_GCR_EL1_EXCL_MASK;
98 
99 	/*
100 	 * Note that 'incl' is an include mask (controlled by the user via
101 	 * prctl()) while GCR_EL1 accepts an exclude mask.
102 	 * No need for ISB since this only affects EL0 currently, implicit
103 	 * with ERET.
104 	 */
105 	sysreg_clear_set_s(SYS_GCR_EL1, SYS_GCR_EL1_EXCL_MASK, excl);
106 }
107 
108 static void set_gcr_el1_excl(u64 incl)
109 {
110 	current->thread.gcr_user_incl = incl;
111 	update_gcr_el1_excl(incl);
112 }
113 
114 void flush_mte_state(void)
115 {
116 	if (!system_supports_mte())
117 		return;
118 
119 	/* clear any pending asynchronous tag fault */
120 	dsb(ish);
121 	write_sysreg_s(0, SYS_TFSRE0_EL1);
122 	clear_thread_flag(TIF_MTE_ASYNC_FAULT);
123 	/* disable tag checking */
124 	set_sctlr_el1_tcf0(SCTLR_EL1_TCF0_NONE);
125 	/* reset tag generation mask */
126 	set_gcr_el1_excl(0);
127 }
128 
129 void mte_thread_switch(struct task_struct *next)
130 {
131 	if (!system_supports_mte())
132 		return;
133 
134 	/* avoid expensive SCTLR_EL1 accesses if no change */
135 	if (current->thread.sctlr_tcf0 != next->thread.sctlr_tcf0)
136 		update_sctlr_el1_tcf0(next->thread.sctlr_tcf0);
137 	update_gcr_el1_excl(next->thread.gcr_user_incl);
138 }
139 
140 void mte_suspend_exit(void)
141 {
142 	if (!system_supports_mte())
143 		return;
144 
145 	update_gcr_el1_excl(current->thread.gcr_user_incl);
146 }
147 
148 long set_mte_ctrl(struct task_struct *task, unsigned long arg)
149 {
150 	u64 tcf0;
151 	u64 gcr_incl = (arg & PR_MTE_TAG_MASK) >> PR_MTE_TAG_SHIFT;
152 
153 	if (!system_supports_mte())
154 		return 0;
155 
156 	switch (arg & PR_MTE_TCF_MASK) {
157 	case PR_MTE_TCF_NONE:
158 		tcf0 = SCTLR_EL1_TCF0_NONE;
159 		break;
160 	case PR_MTE_TCF_SYNC:
161 		tcf0 = SCTLR_EL1_TCF0_SYNC;
162 		break;
163 	case PR_MTE_TCF_ASYNC:
164 		tcf0 = SCTLR_EL1_TCF0_ASYNC;
165 		break;
166 	default:
167 		return -EINVAL;
168 	}
169 
170 	if (task != current) {
171 		task->thread.sctlr_tcf0 = tcf0;
172 		task->thread.gcr_user_incl = gcr_incl;
173 	} else {
174 		set_sctlr_el1_tcf0(tcf0);
175 		set_gcr_el1_excl(gcr_incl);
176 	}
177 
178 	return 0;
179 }
180 
181 long get_mte_ctrl(struct task_struct *task)
182 {
183 	unsigned long ret;
184 
185 	if (!system_supports_mte())
186 		return 0;
187 
188 	ret = task->thread.gcr_user_incl << PR_MTE_TAG_SHIFT;
189 
190 	switch (task->thread.sctlr_tcf0) {
191 	case SCTLR_EL1_TCF0_NONE:
192 		return PR_MTE_TCF_NONE;
193 	case SCTLR_EL1_TCF0_SYNC:
194 		ret |= PR_MTE_TCF_SYNC;
195 		break;
196 	case SCTLR_EL1_TCF0_ASYNC:
197 		ret |= PR_MTE_TCF_ASYNC;
198 		break;
199 	}
200 
201 	return ret;
202 }
203 
204 /*
205  * Access MTE tags in another process' address space as given in mm. Update
206  * the number of tags copied. Return 0 if any tags copied, error otherwise.
207  * Inspired by __access_remote_vm().
208  */
209 static int __access_remote_tags(struct mm_struct *mm, unsigned long addr,
210 				struct iovec *kiov, unsigned int gup_flags)
211 {
212 	struct vm_area_struct *vma;
213 	void __user *buf = kiov->iov_base;
214 	size_t len = kiov->iov_len;
215 	int ret;
216 	int write = gup_flags & FOLL_WRITE;
217 
218 	if (!access_ok(buf, len))
219 		return -EFAULT;
220 
221 	if (mmap_read_lock_killable(mm))
222 		return -EIO;
223 
224 	while (len) {
225 		unsigned long tags, offset;
226 		void *maddr;
227 		struct page *page = NULL;
228 
229 		ret = get_user_pages_remote(mm, addr, 1, gup_flags, &page,
230 					    &vma, NULL);
231 		if (ret <= 0)
232 			break;
233 
234 		/*
235 		 * Only copy tags if the page has been mapped as PROT_MTE
236 		 * (PG_mte_tagged set). Otherwise the tags are not valid and
237 		 * not accessible to user. Moreover, an mprotect(PROT_MTE)
238 		 * would cause the existing tags to be cleared if the page
239 		 * was never mapped with PROT_MTE.
240 		 */
241 		if (!test_bit(PG_mte_tagged, &page->flags)) {
242 			ret = -EOPNOTSUPP;
243 			put_page(page);
244 			break;
245 		}
246 
247 		/* limit access to the end of the page */
248 		offset = offset_in_page(addr);
249 		tags = min(len, (PAGE_SIZE - offset) / MTE_GRANULE_SIZE);
250 
251 		maddr = page_address(page);
252 		if (write) {
253 			tags = mte_copy_tags_from_user(maddr + offset, buf, tags);
254 			set_page_dirty_lock(page);
255 		} else {
256 			tags = mte_copy_tags_to_user(buf, maddr + offset, tags);
257 		}
258 		put_page(page);
259 
260 		/* error accessing the tracer's buffer */
261 		if (!tags)
262 			break;
263 
264 		len -= tags;
265 		buf += tags;
266 		addr += tags * MTE_GRANULE_SIZE;
267 	}
268 	mmap_read_unlock(mm);
269 
270 	/* return an error if no tags copied */
271 	kiov->iov_len = buf - kiov->iov_base;
272 	if (!kiov->iov_len) {
273 		/* check for error accessing the tracee's address space */
274 		if (ret <= 0)
275 			return -EIO;
276 		else
277 			return -EFAULT;
278 	}
279 
280 	return 0;
281 }
282 
283 /*
284  * Copy MTE tags in another process' address space at 'addr' to/from tracer's
285  * iovec buffer. Return 0 on success. Inspired by ptrace_access_vm().
286  */
287 static int access_remote_tags(struct task_struct *tsk, unsigned long addr,
288 			      struct iovec *kiov, unsigned int gup_flags)
289 {
290 	struct mm_struct *mm;
291 	int ret;
292 
293 	mm = get_task_mm(tsk);
294 	if (!mm)
295 		return -EPERM;
296 
297 	if (!tsk->ptrace || (current != tsk->parent) ||
298 	    ((get_dumpable(mm) != SUID_DUMP_USER) &&
299 	     !ptracer_capable(tsk, mm->user_ns))) {
300 		mmput(mm);
301 		return -EPERM;
302 	}
303 
304 	ret = __access_remote_tags(mm, addr, kiov, gup_flags);
305 	mmput(mm);
306 
307 	return ret;
308 }
309 
310 int mte_ptrace_copy_tags(struct task_struct *child, long request,
311 			 unsigned long addr, unsigned long data)
312 {
313 	int ret;
314 	struct iovec kiov;
315 	struct iovec __user *uiov = (void __user *)data;
316 	unsigned int gup_flags = FOLL_FORCE;
317 
318 	if (!system_supports_mte())
319 		return -EIO;
320 
321 	if (get_user(kiov.iov_base, &uiov->iov_base) ||
322 	    get_user(kiov.iov_len, &uiov->iov_len))
323 		return -EFAULT;
324 
325 	if (request == PTRACE_POKEMTETAGS)
326 		gup_flags |= FOLL_WRITE;
327 
328 	/* align addr to the MTE tag granule */
329 	addr &= MTE_GRANULE_MASK;
330 
331 	ret = access_remote_tags(child, addr, &kiov, gup_flags);
332 	if (!ret)
333 		ret = put_user(kiov.iov_len, &uiov->iov_len);
334 
335 	return ret;
336 }
337