]>
Commit | Line | Data |
---|---|---|
637ec831 VF |
1 | // SPDX-License-Identifier: GPL-2.0-only |
2 | /* | |
3 | * Copyright (C) 2020 ARM Ltd. | |
4 | */ | |
5 | ||
34bfeea4 | 6 | #include <linux/bitops.h> |
18ddbaa0 | 7 | #include <linux/kernel.h> |
34bfeea4 | 8 | #include <linux/mm.h> |
1c101da8 CM |
9 | #include <linux/prctl.h> |
10 | #include <linux/sched.h> | |
18ddbaa0 | 11 | #include <linux/sched/mm.h> |
4d1a8a2d | 12 | #include <linux/string.h> |
36943aba SP |
13 | #include <linux/swap.h> |
14 | #include <linux/swapops.h> | |
637ec831 | 15 | #include <linux/thread_info.h> |
85f49cae | 16 | #include <linux/types.h> |
18ddbaa0 | 17 | #include <linux/uio.h> |
637ec831 | 18 | |
85f49cae | 19 | #include <asm/barrier.h> |
637ec831 VF |
20 | #include <asm/cpufeature.h> |
21 | #include <asm/mte.h> | |
85f49cae | 22 | #include <asm/mte-kasan.h> |
18ddbaa0 | 23 | #include <asm/ptrace.h> |
637ec831 VF |
24 | #include <asm/sysreg.h> |
25 | ||
36943aba SP |
26 | static void mte_sync_page_tags(struct page *page, pte_t *ptep, bool check_swap) |
27 | { | |
28 | pte_t old_pte = READ_ONCE(*ptep); | |
29 | ||
30 | if (check_swap && is_swap_pte(old_pte)) { | |
31 | swp_entry_t entry = pte_to_swp_entry(old_pte); | |
32 | ||
33 | if (!non_swap_entry(entry) && mte_restore_tags(entry, page)) | |
34 | return; | |
35 | } | |
36 | ||
e5b8d921 VF |
37 | page_kasan_tag_reset(page); |
38 | /* | |
39 | * We need smp_wmb() in between setting the flags and clearing the | |
40 | * tags because if another thread reads page->flags and builds a | |
41 | * tagged address out of it, there is an actual dependency to the | |
42 | * memory access, but on the current thread we do not guarantee that | |
43 | * the new page->flags are visible before the tags were updated. | |
44 | */ | |
45 | smp_wmb(); | |
36943aba SP |
46 | mte_clear_page_tags(page_address(page)); |
47 | } | |
48 | ||
34bfeea4 CM |
49 | void mte_sync_tags(pte_t *ptep, pte_t pte) |
50 | { | |
51 | struct page *page = pte_page(pte); | |
52 | long i, nr_pages = compound_nr(page); | |
36943aba | 53 | bool check_swap = nr_pages == 1; |
34bfeea4 CM |
54 | |
55 | /* if PG_mte_tagged is set, tags have already been initialised */ | |
56 | for (i = 0; i < nr_pages; i++, page++) { | |
57 | if (!test_and_set_bit(PG_mte_tagged, &page->flags)) | |
36943aba | 58 | mte_sync_page_tags(page, ptep, check_swap); |
34bfeea4 CM |
59 | } |
60 | } | |
61 | ||
4d1a8a2d CM |
62 | int memcmp_pages(struct page *page1, struct page *page2) |
63 | { | |
64 | char *addr1, *addr2; | |
65 | int ret; | |
66 | ||
67 | addr1 = page_address(page1); | |
68 | addr2 = page_address(page2); | |
69 | ret = memcmp(addr1, addr2, PAGE_SIZE); | |
70 | ||
71 | if (!system_supports_mte() || ret) | |
72 | return ret; | |
73 | ||
74 | /* | |
75 | * If the page content is identical but at least one of the pages is | |
76 | * tagged, return non-zero to avoid KSM merging. If only one of the | |
77 | * pages is tagged, set_pte_at() may zero or change the tags of the | |
78 | * other page via mte_sync_tags(). | |
79 | */ | |
80 | if (test_bit(PG_mte_tagged, &page1->flags) || | |
81 | test_bit(PG_mte_tagged, &page2->flags)) | |
82 | return addr1 != addr2; | |
83 | ||
84 | return ret; | |
85 | } | |
86 | ||
85f49cae VF |
87 | u8 mte_get_mem_tag(void *addr) |
88 | { | |
89 | if (!system_supports_mte()) | |
90 | return 0xFF; | |
91 | ||
92 | asm(__MTE_PREAMBLE "ldg %0, [%0]" | |
93 | : "+r" (addr)); | |
94 | ||
95 | return mte_get_ptr_tag(addr); | |
96 | } | |
97 | ||
98 | u8 mte_get_random_tag(void) | |
99 | { | |
100 | void *addr; | |
101 | ||
102 | if (!system_supports_mte()) | |
103 | return 0xFF; | |
104 | ||
105 | asm(__MTE_PREAMBLE "irg %0, %0" | |
106 | : "+r" (addr)); | |
107 | ||
108 | return mte_get_ptr_tag(addr); | |
109 | } | |
110 | ||
111 | void *mte_set_mem_tag_range(void *addr, size_t size, u8 tag) | |
112 | { | |
113 | void *ptr = addr; | |
114 | ||
115 | if ((!system_supports_mte()) || (size == 0)) | |
116 | return addr; | |
117 | ||
118 | /* Make sure that size is MTE granule aligned. */ | |
119 | WARN_ON(size & (MTE_GRANULE_SIZE - 1)); | |
120 | ||
121 | /* Make sure that the address is MTE granule aligned. */ | |
122 | WARN_ON((u64)addr & (MTE_GRANULE_SIZE - 1)); | |
123 | ||
124 | tag = 0xF0 | tag; | |
125 | ptr = (void *)__tag_set(ptr, tag); | |
126 | ||
127 | mte_assign_mem_tag_range(ptr, size); | |
128 | ||
129 | return ptr; | |
130 | } | |
131 | ||
bfc62c59 VF |
132 | void mte_enable_kernel(void) |
133 | { | |
134 | /* Enable MTE Sync Mode for EL1. */ | |
135 | sysreg_clear_set(sctlr_el1, SCTLR_ELx_TCF_MASK, SCTLR_ELx_TCF_SYNC); | |
136 | isb(); | |
137 | } | |
138 | ||
1c101da8 CM |
139 | static void update_sctlr_el1_tcf0(u64 tcf0) |
140 | { | |
141 | /* ISB required for the kernel uaccess routines */ | |
142 | sysreg_clear_set(sctlr_el1, SCTLR_EL1_TCF0_MASK, tcf0); | |
143 | isb(); | |
144 | } | |
145 | ||
146 | static void set_sctlr_el1_tcf0(u64 tcf0) | |
147 | { | |
148 | /* | |
149 | * mte_thread_switch() checks current->thread.sctlr_tcf0 as an | |
150 | * optimisation. Disable preemption so that it does not see | |
151 | * the variable update before the SCTLR_EL1.TCF0 one. | |
152 | */ | |
153 | preempt_disable(); | |
154 | current->thread.sctlr_tcf0 = tcf0; | |
155 | update_sctlr_el1_tcf0(tcf0); | |
156 | preempt_enable(); | |
157 | } | |
158 | ||
af5ce952 CM |
159 | static void update_gcr_el1_excl(u64 incl) |
160 | { | |
161 | u64 excl = ~incl & SYS_GCR_EL1_EXCL_MASK; | |
162 | ||
163 | /* | |
164 | * Note that 'incl' is an include mask (controlled by the user via | |
165 | * prctl()) while GCR_EL1 accepts an exclude mask. | |
166 | * No need for ISB since this only affects EL0 currently, implicit | |
167 | * with ERET. | |
168 | */ | |
169 | sysreg_clear_set_s(SYS_GCR_EL1, SYS_GCR_EL1_EXCL_MASK, excl); | |
170 | } | |
171 | ||
172 | static void set_gcr_el1_excl(u64 incl) | |
173 | { | |
174 | current->thread.gcr_user_incl = incl; | |
175 | update_gcr_el1_excl(incl); | |
176 | } | |
177 | ||
637ec831 VF |
178 | void flush_mte_state(void) |
179 | { | |
180 | if (!system_supports_mte()) | |
181 | return; | |
182 | ||
183 | /* clear any pending asynchronous tag fault */ | |
184 | dsb(ish); | |
185 | write_sysreg_s(0, SYS_TFSRE0_EL1); | |
186 | clear_thread_flag(TIF_MTE_ASYNC_FAULT); | |
1c101da8 CM |
187 | /* disable tag checking */ |
188 | set_sctlr_el1_tcf0(SCTLR_EL1_TCF0_NONE); | |
af5ce952 CM |
189 | /* reset tag generation mask */ |
190 | set_gcr_el1_excl(0); | |
1c101da8 CM |
191 | } |
192 | ||
193 | void mte_thread_switch(struct task_struct *next) | |
194 | { | |
195 | if (!system_supports_mte()) | |
196 | return; | |
197 | ||
198 | /* avoid expensive SCTLR_EL1 accesses if no change */ | |
199 | if (current->thread.sctlr_tcf0 != next->thread.sctlr_tcf0) | |
200 | update_sctlr_el1_tcf0(next->thread.sctlr_tcf0); | |
af5ce952 | 201 | update_gcr_el1_excl(next->thread.gcr_user_incl); |
1c101da8 CM |
202 | } |
203 | ||
39d08e83 CM |
204 | void mte_suspend_exit(void) |
205 | { | |
206 | if (!system_supports_mte()) | |
207 | return; | |
208 | ||
209 | update_gcr_el1_excl(current->thread.gcr_user_incl); | |
210 | } | |
211 | ||
93f067f6 | 212 | long set_mte_ctrl(struct task_struct *task, unsigned long arg) |
1c101da8 CM |
213 | { |
214 | u64 tcf0; | |
93f067f6 | 215 | u64 gcr_incl = (arg & PR_MTE_TAG_MASK) >> PR_MTE_TAG_SHIFT; |
1c101da8 CM |
216 | |
217 | if (!system_supports_mte()) | |
218 | return 0; | |
219 | ||
220 | switch (arg & PR_MTE_TCF_MASK) { | |
221 | case PR_MTE_TCF_NONE: | |
222 | tcf0 = SCTLR_EL1_TCF0_NONE; | |
223 | break; | |
224 | case PR_MTE_TCF_SYNC: | |
225 | tcf0 = SCTLR_EL1_TCF0_SYNC; | |
226 | break; | |
227 | case PR_MTE_TCF_ASYNC: | |
228 | tcf0 = SCTLR_EL1_TCF0_ASYNC; | |
229 | break; | |
230 | default: | |
231 | return -EINVAL; | |
232 | } | |
233 | ||
93f067f6 CM |
234 | if (task != current) { |
235 | task->thread.sctlr_tcf0 = tcf0; | |
236 | task->thread.gcr_user_incl = gcr_incl; | |
237 | } else { | |
238 | set_sctlr_el1_tcf0(tcf0); | |
239 | set_gcr_el1_excl(gcr_incl); | |
240 | } | |
1c101da8 CM |
241 | |
242 | return 0; | |
243 | } | |
244 | ||
93f067f6 | 245 | long get_mte_ctrl(struct task_struct *task) |
1c101da8 | 246 | { |
af5ce952 CM |
247 | unsigned long ret; |
248 | ||
1c101da8 CM |
249 | if (!system_supports_mte()) |
250 | return 0; | |
251 | ||
93f067f6 | 252 | ret = task->thread.gcr_user_incl << PR_MTE_TAG_SHIFT; |
af5ce952 | 253 | |
93f067f6 | 254 | switch (task->thread.sctlr_tcf0) { |
1c101da8 | 255 | case SCTLR_EL1_TCF0_NONE: |
929c1f33 PC |
256 | ret |= PR_MTE_TCF_NONE; |
257 | break; | |
1c101da8 | 258 | case SCTLR_EL1_TCF0_SYNC: |
af5ce952 CM |
259 | ret |= PR_MTE_TCF_SYNC; |
260 | break; | |
1c101da8 | 261 | case SCTLR_EL1_TCF0_ASYNC: |
af5ce952 CM |
262 | ret |= PR_MTE_TCF_ASYNC; |
263 | break; | |
1c101da8 CM |
264 | } |
265 | ||
af5ce952 | 266 | return ret; |
637ec831 | 267 | } |
18ddbaa0 CM |
268 | |
269 | /* | |
270 | * Access MTE tags in another process' address space as given in mm. Update | |
271 | * the number of tags copied. Return 0 if any tags copied, error otherwise. | |
272 | * Inspired by __access_remote_vm(). | |
273 | */ | |
274 | static int __access_remote_tags(struct mm_struct *mm, unsigned long addr, | |
275 | struct iovec *kiov, unsigned int gup_flags) | |
276 | { | |
277 | struct vm_area_struct *vma; | |
278 | void __user *buf = kiov->iov_base; | |
279 | size_t len = kiov->iov_len; | |
280 | int ret; | |
281 | int write = gup_flags & FOLL_WRITE; | |
282 | ||
283 | if (!access_ok(buf, len)) | |
284 | return -EFAULT; | |
285 | ||
286 | if (mmap_read_lock_killable(mm)) | |
287 | return -EIO; | |
288 | ||
289 | while (len) { | |
290 | unsigned long tags, offset; | |
291 | void *maddr; | |
292 | struct page *page = NULL; | |
293 | ||
294 | ret = get_user_pages_remote(mm, addr, 1, gup_flags, &page, | |
295 | &vma, NULL); | |
296 | if (ret <= 0) | |
297 | break; | |
298 | ||
299 | /* | |
300 | * Only copy tags if the page has been mapped as PROT_MTE | |
301 | * (PG_mte_tagged set). Otherwise the tags are not valid and | |
302 | * not accessible to user. Moreover, an mprotect(PROT_MTE) | |
303 | * would cause the existing tags to be cleared if the page | |
304 | * was never mapped with PROT_MTE. | |
305 | */ | |
306 | if (!test_bit(PG_mte_tagged, &page->flags)) { | |
307 | ret = -EOPNOTSUPP; | |
308 | put_page(page); | |
309 | break; | |
310 | } | |
311 | ||
312 | /* limit access to the end of the page */ | |
313 | offset = offset_in_page(addr); | |
314 | tags = min(len, (PAGE_SIZE - offset) / MTE_GRANULE_SIZE); | |
315 | ||
316 | maddr = page_address(page); | |
317 | if (write) { | |
318 | tags = mte_copy_tags_from_user(maddr + offset, buf, tags); | |
319 | set_page_dirty_lock(page); | |
320 | } else { | |
321 | tags = mte_copy_tags_to_user(buf, maddr + offset, tags); | |
322 | } | |
323 | put_page(page); | |
324 | ||
325 | /* error accessing the tracer's buffer */ | |
326 | if (!tags) | |
327 | break; | |
328 | ||
329 | len -= tags; | |
330 | buf += tags; | |
331 | addr += tags * MTE_GRANULE_SIZE; | |
332 | } | |
333 | mmap_read_unlock(mm); | |
334 | ||
335 | /* return an error if no tags copied */ | |
336 | kiov->iov_len = buf - kiov->iov_base; | |
337 | if (!kiov->iov_len) { | |
338 | /* check for error accessing the tracee's address space */ | |
339 | if (ret <= 0) | |
340 | return -EIO; | |
341 | else | |
342 | return -EFAULT; | |
343 | } | |
344 | ||
345 | return 0; | |
346 | } | |
347 | ||
348 | /* | |
349 | * Copy MTE tags in another process' address space at 'addr' to/from tracer's | |
350 | * iovec buffer. Return 0 on success. Inspired by ptrace_access_vm(). | |
351 | */ | |
352 | static int access_remote_tags(struct task_struct *tsk, unsigned long addr, | |
353 | struct iovec *kiov, unsigned int gup_flags) | |
354 | { | |
355 | struct mm_struct *mm; | |
356 | int ret; | |
357 | ||
358 | mm = get_task_mm(tsk); | |
359 | if (!mm) | |
360 | return -EPERM; | |
361 | ||
362 | if (!tsk->ptrace || (current != tsk->parent) || | |
363 | ((get_dumpable(mm) != SUID_DUMP_USER) && | |
364 | !ptracer_capable(tsk, mm->user_ns))) { | |
365 | mmput(mm); | |
366 | return -EPERM; | |
367 | } | |
368 | ||
369 | ret = __access_remote_tags(mm, addr, kiov, gup_flags); | |
370 | mmput(mm); | |
371 | ||
372 | return ret; | |
373 | } | |
374 | ||
375 | int mte_ptrace_copy_tags(struct task_struct *child, long request, | |
376 | unsigned long addr, unsigned long data) | |
377 | { | |
378 | int ret; | |
379 | struct iovec kiov; | |
380 | struct iovec __user *uiov = (void __user *)data; | |
381 | unsigned int gup_flags = FOLL_FORCE; | |
382 | ||
383 | if (!system_supports_mte()) | |
384 | return -EIO; | |
385 | ||
386 | if (get_user(kiov.iov_base, &uiov->iov_base) || | |
387 | get_user(kiov.iov_len, &uiov->iov_len)) | |
388 | return -EFAULT; | |
389 | ||
390 | if (request == PTRACE_POKEMTETAGS) | |
391 | gup_flags |= FOLL_WRITE; | |
392 | ||
393 | /* align addr to the MTE tag granule */ | |
394 | addr &= MTE_GRANULE_MASK; | |
395 | ||
396 | ret = access_remote_tags(child, addr, &kiov, gup_flags); | |
397 | if (!ret) | |
398 | ret = put_user(kiov.iov_len, &uiov->iov_len); | |
399 | ||
400 | return ret; | |
401 | } |