1 // SPDX-License-Identifier: GPL-2.0-only
3 * Copyright (C) 2020 ARM Ltd.
6 #include <linux/bitops.h>
7 #include <linux/kernel.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/types.h>
17 #include <linux/uio.h>
19 #include <asm/barrier.h>
20 #include <asm/cpufeature.h>
22 #include <asm/mte-kasan.h>
23 #include <asm/ptrace.h>
24 #include <asm/sysreg.h>
26 static void mte_sync_page_tags(struct page *page, pte_t *ptep, bool check_swap)
28 pte_t old_pte = READ_ONCE(*ptep);
30 if (check_swap && is_swap_pte(old_pte)) {
31 swp_entry_t entry = pte_to_swp_entry(old_pte);
33 if (!non_swap_entry(entry) && mte_restore_tags(entry, page))
37 mte_clear_page_tags(page_address(page));
40 void mte_sync_tags(pte_t *ptep, pte_t pte)
42 struct page *page = pte_page(pte);
43 long i, nr_pages = compound_nr(page);
44 bool check_swap = nr_pages == 1;
46 /* if PG_mte_tagged is set, tags have already been initialised */
47 for (i = 0; i < nr_pages; i++, page++) {
48 if (!test_and_set_bit(PG_mte_tagged, &page->flags))
49 mte_sync_page_tags(page, ptep, check_swap);
53 int memcmp_pages(struct page *page1, struct page *page2)
58 addr1 = page_address(page1);
59 addr2 = page_address(page2);
60 ret = memcmp(addr1, addr2, PAGE_SIZE);
62 if (!system_supports_mte() || ret)
66 * If the page content is identical but at least one of the pages is
67 * tagged, return non-zero to avoid KSM merging. If only one of the
68 * pages is tagged, set_pte_at() may zero or change the tags of the
69 * other page via mte_sync_tags().
71 if (test_bit(PG_mte_tagged, &page1->flags) ||
72 test_bit(PG_mte_tagged, &page2->flags))
73 return addr1 != addr2;
78 u8 mte_get_mem_tag(void *addr)
80 if (!system_supports_mte())
83 asm(__MTE_PREAMBLE "ldg %0, [%0]"
86 return mte_get_ptr_tag(addr);
89 u8 mte_get_random_tag(void)
93 if (!system_supports_mte())
96 asm(__MTE_PREAMBLE "irg %0, %0"
99 return mte_get_ptr_tag(addr);
102 void *mte_set_mem_tag_range(void *addr, size_t size, u8 tag)
106 if ((!system_supports_mte()) || (size == 0))
109 /* Make sure that size is MTE granule aligned. */
110 WARN_ON(size & (MTE_GRANULE_SIZE - 1));
112 /* Make sure that the address is MTE granule aligned. */
113 WARN_ON((u64)addr & (MTE_GRANULE_SIZE - 1));
116 ptr = (void *)__tag_set(ptr, tag);
118 mte_assign_mem_tag_range(ptr, size);
123 static void update_sctlr_el1_tcf0(u64 tcf0)
125 /* ISB required for the kernel uaccess routines */
126 sysreg_clear_set(sctlr_el1, SCTLR_EL1_TCF0_MASK, tcf0);
130 static void set_sctlr_el1_tcf0(u64 tcf0)
133 * mte_thread_switch() checks current->thread.sctlr_tcf0 as an
134 * optimisation. Disable preemption so that it does not see
135 * the variable update before the SCTLR_EL1.TCF0 one.
138 current->thread.sctlr_tcf0 = tcf0;
139 update_sctlr_el1_tcf0(tcf0);
143 static void update_gcr_el1_excl(u64 incl)
145 u64 excl = ~incl & SYS_GCR_EL1_EXCL_MASK;
148 * Note that 'incl' is an include mask (controlled by the user via
149 * prctl()) while GCR_EL1 accepts an exclude mask.
150 * No need for ISB since this only affects EL0 currently, implicit
153 sysreg_clear_set_s(SYS_GCR_EL1, SYS_GCR_EL1_EXCL_MASK, excl);
156 static void set_gcr_el1_excl(u64 incl)
158 current->thread.gcr_user_incl = incl;
159 update_gcr_el1_excl(incl);
162 void flush_mte_state(void)
164 if (!system_supports_mte())
167 /* clear any pending asynchronous tag fault */
169 write_sysreg_s(0, SYS_TFSRE0_EL1);
170 clear_thread_flag(TIF_MTE_ASYNC_FAULT);
171 /* disable tag checking */
172 set_sctlr_el1_tcf0(SCTLR_EL1_TCF0_NONE);
173 /* reset tag generation mask */
177 void mte_thread_switch(struct task_struct *next)
179 if (!system_supports_mte())
182 /* avoid expensive SCTLR_EL1 accesses if no change */
183 if (current->thread.sctlr_tcf0 != next->thread.sctlr_tcf0)
184 update_sctlr_el1_tcf0(next->thread.sctlr_tcf0);
185 update_gcr_el1_excl(next->thread.gcr_user_incl);
188 void mte_suspend_exit(void)
190 if (!system_supports_mte())
193 update_gcr_el1_excl(current->thread.gcr_user_incl);
196 long set_mte_ctrl(struct task_struct *task, unsigned long arg)
199 u64 gcr_incl = (arg & PR_MTE_TAG_MASK) >> PR_MTE_TAG_SHIFT;
201 if (!system_supports_mte())
204 switch (arg & PR_MTE_TCF_MASK) {
205 case PR_MTE_TCF_NONE:
206 tcf0 = SCTLR_EL1_TCF0_NONE;
208 case PR_MTE_TCF_SYNC:
209 tcf0 = SCTLR_EL1_TCF0_SYNC;
211 case PR_MTE_TCF_ASYNC:
212 tcf0 = SCTLR_EL1_TCF0_ASYNC;
218 if (task != current) {
219 task->thread.sctlr_tcf0 = tcf0;
220 task->thread.gcr_user_incl = gcr_incl;
222 set_sctlr_el1_tcf0(tcf0);
223 set_gcr_el1_excl(gcr_incl);
229 long get_mte_ctrl(struct task_struct *task)
233 if (!system_supports_mte())
236 ret = task->thread.gcr_user_incl << PR_MTE_TAG_SHIFT;
238 switch (task->thread.sctlr_tcf0) {
239 case SCTLR_EL1_TCF0_NONE:
240 ret |= PR_MTE_TCF_NONE;
242 case SCTLR_EL1_TCF0_SYNC:
243 ret |= PR_MTE_TCF_SYNC;
245 case SCTLR_EL1_TCF0_ASYNC:
246 ret |= PR_MTE_TCF_ASYNC;
254 * Access MTE tags in another process' address space as given in mm. Update
255 * the number of tags copied. Return 0 if any tags copied, error otherwise.
256 * Inspired by __access_remote_vm().
258 static int __access_remote_tags(struct mm_struct *mm, unsigned long addr,
259 struct iovec *kiov, unsigned int gup_flags)
261 struct vm_area_struct *vma;
262 void __user *buf = kiov->iov_base;
263 size_t len = kiov->iov_len;
265 int write = gup_flags & FOLL_WRITE;
267 if (!access_ok(buf, len))
270 if (mmap_read_lock_killable(mm))
274 unsigned long tags, offset;
276 struct page *page = NULL;
278 ret = get_user_pages_remote(mm, addr, 1, gup_flags, &page,
284 * Only copy tags if the page has been mapped as PROT_MTE
285 * (PG_mte_tagged set). Otherwise the tags are not valid and
286 * not accessible to user. Moreover, an mprotect(PROT_MTE)
287 * would cause the existing tags to be cleared if the page
288 * was never mapped with PROT_MTE.
290 if (!test_bit(PG_mte_tagged, &page->flags)) {
296 /* limit access to the end of the page */
297 offset = offset_in_page(addr);
298 tags = min(len, (PAGE_SIZE - offset) / MTE_GRANULE_SIZE);
300 maddr = page_address(page);
302 tags = mte_copy_tags_from_user(maddr + offset, buf, tags);
303 set_page_dirty_lock(page);
305 tags = mte_copy_tags_to_user(buf, maddr + offset, tags);
309 /* error accessing the tracer's buffer */
315 addr += tags * MTE_GRANULE_SIZE;
317 mmap_read_unlock(mm);
319 /* return an error if no tags copied */
320 kiov->iov_len = buf - kiov->iov_base;
321 if (!kiov->iov_len) {
322 /* check for error accessing the tracee's address space */
333 * Copy MTE tags in another process' address space at 'addr' to/from tracer's
334 * iovec buffer. Return 0 on success. Inspired by ptrace_access_vm().
336 static int access_remote_tags(struct task_struct *tsk, unsigned long addr,
337 struct iovec *kiov, unsigned int gup_flags)
339 struct mm_struct *mm;
342 mm = get_task_mm(tsk);
346 if (!tsk->ptrace || (current != tsk->parent) ||
347 ((get_dumpable(mm) != SUID_DUMP_USER) &&
348 !ptracer_capable(tsk, mm->user_ns))) {
353 ret = __access_remote_tags(mm, addr, kiov, gup_flags);
359 int mte_ptrace_copy_tags(struct task_struct *child, long request,
360 unsigned long addr, unsigned long data)
364 struct iovec __user *uiov = (void __user *)data;
365 unsigned int gup_flags = FOLL_FORCE;
367 if (!system_supports_mte())
370 if (get_user(kiov.iov_base, &uiov->iov_base) ||
371 get_user(kiov.iov_len, &uiov->iov_len))
374 if (request == PTRACE_POKEMTETAGS)
375 gup_flags |= FOLL_WRITE;
377 /* align addr to the MTE tag granule */
378 addr &= MTE_GRANULE_MASK;
380 ret = access_remote_tags(child, addr, &kiov, gup_flags);
382 ret = put_user(kiov.iov_len, &uiov->iov_len);