1 // SPDX-License-Identifier: GPL-2.0
2 /*
3  *  S390 version
4  *    Copyright IBM Corp. 1999
5  *    Author(s): Hartmut Penner (hp@de.ibm.com)
6  *		 Ulrich Weigand (uweigand@de.ibm.com)
7  *
8  *  Derived from "arch/i386/mm/fault.c"
9  *    Copyright (C) 1995  Linus Torvalds
10  */
11 
12 #include <linux/kernel_stat.h>
13 #include <linux/mmu_context.h>
14 #include <linux/cpufeature.h>
15 #include <linux/perf_event.h>
16 #include <linux/signal.h>
17 #include <linux/sched.h>
18 #include <linux/sched/debug.h>
19 #include <linux/kernel.h>
20 #include <linux/errno.h>
21 #include <linux/string.h>
22 #include <linux/types.h>
23 #include <linux/ptrace.h>
24 #include <linux/mman.h>
25 #include <linux/mm.h>
26 #include <linux/compat.h>
27 #include <linux/smp.h>
28 #include <linux/kdebug.h>
29 #include <linux/init.h>
30 #include <linux/console.h>
31 #include <linux/extable.h>
32 #include <linux/hardirq.h>
33 #include <linux/kprobes.h>
34 #include <linux/uaccess.h>
35 #include <linux/hugetlb.h>
36 #include <linux/kfence.h>
37 #include <linux/pagewalk.h>
38 #include <asm/asm-extable.h>
39 #include <asm/asm-offsets.h>
40 #include <asm/ptrace.h>
41 #include <asm/fault.h>
42 #include <asm/diag.h>
43 #include <asm/irq.h>
44 #include <asm/facility.h>
45 #include <asm/uv.h>
46 #include "../kernel/entry.h"
47 
48 /*
49  * Find out which address space caused the exception.
50  */
51 static bool is_kernel_fault(struct pt_regs *regs)
52 {
53 	union teid teid = { .val = regs->int_parm_long };
54 
55 	if (user_mode(regs))
56 		return false;
57 	if (teid.as == PSW_BITS_AS_SECONDARY)
58 		return false;
59 	return true;
60 }
61 
62 static unsigned long get_fault_address(struct pt_regs *regs)
63 {
64 	union teid teid = { .val = regs->int_parm_long };
65 
66 	return teid.addr * PAGE_SIZE;
67 }
68 
69 static __always_inline bool fault_is_write(struct pt_regs *regs)
70 {
71 	union teid teid = { .val = regs->int_parm_long };
72 
73 	if (test_facility(75))
74 		return teid.fsi == TEID_FSI_STORE;
75 	return false;
76 }
77 
78 static void dump_pagetable(unsigned long asce, unsigned long address)
79 {
80 	unsigned long entry, *table = __va(asce & _ASCE_ORIGIN);
81 
82 	pr_alert("AS:%016lx ", asce);
83 	switch (asce & _ASCE_TYPE_MASK) {
84 	case _ASCE_TYPE_REGION1:
85 		table += (address & _REGION1_INDEX) >> _REGION1_SHIFT;
86 		if (get_kernel_nofault(entry, table))
87 			goto bad;
88 		pr_cont("R1:%016lx ", entry);
89 		if (entry & _REGION_ENTRY_INVALID)
90 			goto out;
91 		table = __va(entry & _REGION_ENTRY_ORIGIN);
92 		fallthrough;
93 	case _ASCE_TYPE_REGION2:
94 		table += (address & _REGION2_INDEX) >> _REGION2_SHIFT;
95 		if (get_kernel_nofault(entry, table))
96 			goto bad;
97 		pr_cont("R2:%016lx ", entry);
98 		if (entry & _REGION_ENTRY_INVALID)
99 			goto out;
100 		table = __va(entry & _REGION_ENTRY_ORIGIN);
101 		fallthrough;
102 	case _ASCE_TYPE_REGION3:
103 		table += (address & _REGION3_INDEX) >> _REGION3_SHIFT;
104 		if (get_kernel_nofault(entry, table))
105 			goto bad;
106 		pr_cont("R3:%016lx ", entry);
107 		if (entry & (_REGION_ENTRY_INVALID | _REGION3_ENTRY_LARGE))
108 			goto out;
109 		table = __va(entry & _REGION_ENTRY_ORIGIN);
110 		fallthrough;
111 	case _ASCE_TYPE_SEGMENT:
112 		table += (address & _SEGMENT_INDEX) >> _SEGMENT_SHIFT;
113 		if (get_kernel_nofault(entry, table))
114 			goto bad;
115 		pr_cont("S:%016lx ", entry);
116 		if (entry & (_SEGMENT_ENTRY_INVALID | _SEGMENT_ENTRY_LARGE))
117 			goto out;
118 		table = __va(entry & _SEGMENT_ENTRY_ORIGIN);
119 	}
120 	table += (address & _PAGE_INDEX) >> PAGE_SHIFT;
121 	if (get_kernel_nofault(entry, table))
122 		goto bad;
123 	pr_cont("P:%016lx ", entry);
124 out:
125 	pr_cont("\n");
126 	return;
127 bad:
128 	pr_cont("BAD\n");
129 }
130 
131 static void dump_fault_info(struct pt_regs *regs)
132 {
133 	union teid teid = { .val = regs->int_parm_long };
134 	unsigned long asce;
135 
136 	pr_alert("Failing address: %016lx TEID: %016lx\n",
137 		 get_fault_address(regs), teid.val);
138 	pr_alert("Fault in ");
139 	switch (teid.as) {
140 	case PSW_BITS_AS_HOME:
141 		pr_cont("home space ");
142 		break;
143 	case PSW_BITS_AS_SECONDARY:
144 		pr_cont("secondary space ");
145 		break;
146 	case PSW_BITS_AS_ACCREG:
147 		pr_cont("access register ");
148 		break;
149 	case PSW_BITS_AS_PRIMARY:
150 		pr_cont("primary space ");
151 		break;
152 	}
153 	pr_cont("mode while using ");
154 	if (is_kernel_fault(regs)) {
155 		asce = get_lowcore()->kernel_asce.val;
156 		pr_cont("kernel ");
157 	} else {
158 		asce = get_lowcore()->user_asce.val;
159 		pr_cont("user ");
160 	}
161 	pr_cont("ASCE.\n");
162 	dump_pagetable(asce, get_fault_address(regs));
163 }
164 
165 int show_unhandled_signals = 1;
166 
167 static const struct ctl_table s390_fault_sysctl_table[] = {
168 	{
169 		.procname	= "userprocess_debug",
170 		.data		= &show_unhandled_signals,
171 		.maxlen		= sizeof(int),
172 		.mode		= 0644,
173 		.proc_handler	= proc_dointvec,
174 	},
175 };
176 
177 static int __init init_s390_fault_sysctls(void)
178 {
179 	register_sysctl_init("kernel", s390_fault_sysctl_table);
180 	return 0;
181 }
182 arch_initcall(init_s390_fault_sysctls);
183 
184 void report_user_fault(struct pt_regs *regs, long signr, int is_mm_fault)
185 {
186 	static DEFINE_RATELIMIT_STATE(rs, DEFAULT_RATELIMIT_INTERVAL, DEFAULT_RATELIMIT_BURST);
187 
188 	if ((task_pid_nr(current) > 1) && !show_unhandled_signals)
189 		return;
190 	if (!unhandled_signal(current, signr))
191 		return;
192 	if (!__ratelimit(&rs))
193 		return;
194 	pr_alert("User process fault: interruption code %04x ilc:%d ",
195 		 regs->int_code & 0xffff, regs->int_code >> 17);
196 	print_vma_addr(KERN_CONT "in ", regs->psw.addr);
197 	pr_cont("\n");
198 	if (is_mm_fault)
199 		dump_fault_info(regs);
200 	show_regs(regs);
201 }
202 
203 static void do_sigsegv(struct pt_regs *regs, int si_code)
204 {
205 	report_user_fault(regs, SIGSEGV, 1);
206 	force_sig_fault(SIGSEGV, si_code, (void __user *)get_fault_address(regs));
207 }
208 
209 static void handle_fault_error_nolock(struct pt_regs *regs, int si_code)
210 {
211 	unsigned long address;
212 	bool is_write;
213 
214 	if (user_mode(regs)) {
215 		if (WARN_ON_ONCE(!si_code))
216 			si_code = SEGV_MAPERR;
217 		return do_sigsegv(regs, si_code);
218 	}
219 	if (fixup_exception(regs))
220 		return;
221 	if (is_kernel_fault(regs)) {
222 		address = get_fault_address(regs);
223 		is_write = fault_is_write(regs);
224 		if (kfence_handle_page_fault(address, is_write, regs))
225 			return;
226 		pr_alert("Unable to handle kernel pointer dereference in virtual kernel address space\n");
227 	} else {
228 		pr_alert("Unable to handle kernel paging request in virtual user address space\n");
229 	}
230 	dump_fault_info(regs);
231 	die(regs, "Oops");
232 }
233 
234 static void handle_fault_error(struct pt_regs *regs, int si_code)
235 {
236 	struct mm_struct *mm = current->mm;
237 
238 	mmap_read_unlock(mm);
239 	handle_fault_error_nolock(regs, si_code);
240 }
241 
242 static void do_sigbus(struct pt_regs *regs)
243 {
244 	force_sig_fault(SIGBUS, BUS_ADRERR, (void __user *)get_fault_address(regs));
245 }
246 
247 /*
248  * This routine handles page faults.  It determines the address,
249  * and the problem, and then passes it off to one of the appropriate
250  * routines.
251  *
252  * interruption code (int_code):
253  *   04       Protection	   ->  Write-Protection  (suppression)
254  *   10       Segment translation  ->  Not present	 (nullification)
255  *   11       Page translation	   ->  Not present	 (nullification)
256  *   3b       Region third trans.  ->  Not present	 (nullification)
257  */
258 static void do_exception(struct pt_regs *regs, int access)
259 {
260 	struct vm_area_struct *vma;
261 	unsigned long address;
262 	struct mm_struct *mm;
263 	unsigned int flags;
264 	vm_fault_t fault;
265 	bool is_write;
266 
267 	/*
268 	 * The instruction that caused the program check has
269 	 * been nullified. Don't signal single step via SIGTRAP.
270 	 */
271 	clear_thread_flag(TIF_PER_TRAP);
272 	if (kprobe_page_fault(regs, 14))
273 		return;
274 	mm = current->mm;
275 	address = get_fault_address(regs);
276 	is_write = fault_is_write(regs);
277 	if (is_kernel_fault(regs) || faulthandler_disabled() || !mm)
278 		return handle_fault_error_nolock(regs, 0);
279 	perf_sw_event(PERF_COUNT_SW_PAGE_FAULTS, 1, regs, address);
280 	flags = FAULT_FLAG_DEFAULT;
281 	if (user_mode(regs))
282 		flags |= FAULT_FLAG_USER;
283 	if (is_write)
284 		access = VM_WRITE;
285 	if (access == VM_WRITE)
286 		flags |= FAULT_FLAG_WRITE;
287 	if (!(flags & FAULT_FLAG_USER))
288 		goto lock_mmap;
289 	vma = lock_vma_under_rcu(mm, address);
290 	if (!vma)
291 		goto lock_mmap;
292 	if (!(vma->vm_flags & access)) {
293 		vma_end_read(vma);
294 		count_vm_vma_lock_event(VMA_LOCK_SUCCESS);
295 		return handle_fault_error_nolock(regs, SEGV_ACCERR);
296 	}
297 	fault = handle_mm_fault(vma, address, flags | FAULT_FLAG_VMA_LOCK, regs);
298 	if (!(fault & (VM_FAULT_RETRY | VM_FAULT_COMPLETED)))
299 		vma_end_read(vma);
300 	if (!(fault & VM_FAULT_RETRY)) {
301 		count_vm_vma_lock_event(VMA_LOCK_SUCCESS);
302 		goto done;
303 	}
304 	count_vm_vma_lock_event(VMA_LOCK_RETRY);
305 	if (fault & VM_FAULT_MAJOR)
306 		flags |= FAULT_FLAG_TRIED;
307 	/* Quick path to respond to signals */
308 	if (fault_signal_pending(fault, regs)) {
309 		if (!user_mode(regs))
310 			handle_fault_error_nolock(regs, 0);
311 		return;
312 	}
313 lock_mmap:
314 retry:
315 	vma = lock_mm_and_find_vma(mm, address, regs);
316 	if (!vma)
317 		return handle_fault_error_nolock(regs, SEGV_MAPERR);
318 	if (unlikely(!(vma->vm_flags & access)))
319 		return handle_fault_error(regs, SEGV_ACCERR);
320 	fault = handle_mm_fault(vma, address, flags, regs);
321 	if (fault_signal_pending(fault, regs)) {
322 		if (!user_mode(regs))
323 			handle_fault_error_nolock(regs, 0);
324 		return;
325 	}
326 	/* The fault is fully completed (including releasing mmap lock) */
327 	if (fault & VM_FAULT_COMPLETED)
328 		return;
329 	if (fault & VM_FAULT_RETRY) {
330 		flags |= FAULT_FLAG_TRIED;
331 		goto retry;
332 	}
333 	mmap_read_unlock(mm);
334 done:
335 	if (!(fault & VM_FAULT_ERROR))
336 		return;
337 	if (fault & VM_FAULT_OOM) {
338 		if (!user_mode(regs))
339 			handle_fault_error_nolock(regs, 0);
340 		else
341 			pagefault_out_of_memory();
342 	} else if (fault & VM_FAULT_SIGSEGV) {
343 		if (!user_mode(regs))
344 			handle_fault_error_nolock(regs, 0);
345 		else
346 			do_sigsegv(regs, SEGV_MAPERR);
347 	} else if (fault & (VM_FAULT_SIGBUS | VM_FAULT_HWPOISON |
348 			    VM_FAULT_HWPOISON_LARGE)) {
349 		if (!user_mode(regs))
350 			handle_fault_error_nolock(regs, 0);
351 		else
352 			do_sigbus(regs);
353 	} else {
354 		pr_emerg("Unexpected fault flags: %08x\n", fault);
355 		BUG();
356 	}
357 }
358 
359 void do_protection_exception(struct pt_regs *regs)
360 {
361 	union teid teid = { .val = regs->int_parm_long };
362 
363 	/*
364 	 * Protection exceptions are suppressing, decrement psw address.
365 	 * The exception to this rule are aborted transactions, for these
366 	 * the PSW already points to the correct location.
367 	 */
368 	if (!(regs->int_code & 0x200))
369 		regs->psw.addr = __rewind_psw(regs->psw, regs->int_code >> 16);
370 	/*
371 	 * Check for low-address protection.  This needs to be treated
372 	 * as a special case because the translation exception code
373 	 * field is not guaranteed to contain valid data in this case.
374 	 */
375 	if (unlikely(!teid.b61)) {
376 		if (user_mode(regs)) {
377 			/* Low-address protection in user mode: cannot happen */
378 			dump_fault_info(regs);
379 			die(regs, "Low-address protection");
380 		}
381 		/*
382 		 * Low-address protection in kernel mode means
383 		 * NULL pointer write access in kernel mode.
384 		 */
385 		return handle_fault_error_nolock(regs, 0);
386 	}
387 	if (unlikely(cpu_has_nx() && teid.b56)) {
388 		regs->int_parm_long = (teid.addr * PAGE_SIZE) | (regs->psw.addr & PAGE_MASK);
389 		return handle_fault_error_nolock(regs, SEGV_ACCERR);
390 	}
391 	do_exception(regs, VM_WRITE);
392 }
393 NOKPROBE_SYMBOL(do_protection_exception);
394 
395 void do_dat_exception(struct pt_regs *regs)
396 {
397 	do_exception(regs, VM_ACCESS_FLAGS);
398 }
399 NOKPROBE_SYMBOL(do_dat_exception);
400 
401 #if IS_ENABLED(CONFIG_PGSTE)
402 
403 void do_secure_storage_access(struct pt_regs *regs)
404 {
405 	union teid teid = { .val = regs->int_parm_long };
406 	unsigned long addr = get_fault_address(regs);
407 	struct vm_area_struct *vma;
408 	struct folio_walk fw;
409 	struct mm_struct *mm;
410 	struct folio *folio;
411 	int rc;
412 
413 	/*
414 	 * Bit 61 indicates if the address is valid, if it is not the
415 	 * kernel should be stopped or SIGSEGV should be sent to the
416 	 * process. Bit 61 is not reliable without the misc UV feature,
417 	 * therefore this needs to be checked too.
418 	 */
419 	if (uv_has_feature(BIT_UV_FEAT_MISC) && !teid.b61) {
420 		/*
421 		 * When this happens, userspace did something that it
422 		 * was not supposed to do, e.g. branching into secure
423 		 * memory. Trigger a segmentation fault.
424 		 */
425 		if (user_mode(regs)) {
426 			send_sig(SIGSEGV, current, 0);
427 			return;
428 		}
429 		/*
430 		 * The kernel should never run into this case and
431 		 * there is no way out of this situation.
432 		 */
433 		panic("Unexpected PGM 0x3d with TEID bit 61=0");
434 	}
435 	if (is_kernel_fault(regs)) {
436 		folio = phys_to_folio(addr);
437 		if (unlikely(!folio_try_get(folio)))
438 			return;
439 		rc = arch_make_folio_accessible(folio);
440 		folio_put(folio);
441 		if (rc)
442 			BUG();
443 	} else {
444 		if (faulthandler_disabled())
445 			return handle_fault_error_nolock(regs, 0);
446 		mm = current->mm;
447 		mmap_read_lock(mm);
448 		vma = find_vma(mm, addr);
449 		if (!vma)
450 			return handle_fault_error(regs, SEGV_MAPERR);
451 		folio = folio_walk_start(&fw, vma, addr, 0);
452 		if (!folio) {
453 			mmap_read_unlock(mm);
454 			return;
455 		}
456 		/* arch_make_folio_accessible() needs a raised refcount. */
457 		folio_get(folio);
458 		rc = arch_make_folio_accessible(folio);
459 		folio_put(folio);
460 		folio_walk_end(&fw, vma);
461 		if (rc)
462 			send_sig(SIGSEGV, current, 0);
463 		mmap_read_unlock(mm);
464 	}
465 }
466 NOKPROBE_SYMBOL(do_secure_storage_access);
467 
468 #endif /* CONFIG_PGSTE */
469