xref: /openbmc/linux/arch/s390/kernel/kprobes.c (revision 240e6d25)
1 // SPDX-License-Identifier: GPL-2.0+
2 /*
3  *  Kernel Probes (KProbes)
4  *
5  * Copyright IBM Corp. 2002, 2006
6  *
7  * s390 port, used ppc64 as template. Mike Grundy <grundym@us.ibm.com>
8  */
9 
10 #define pr_fmt(fmt) "kprobes: " fmt
11 
12 #include <linux/moduleloader.h>
13 #include <linux/kprobes.h>
14 #include <linux/ptrace.h>
15 #include <linux/preempt.h>
16 #include <linux/stop_machine.h>
17 #include <linux/kdebug.h>
18 #include <linux/uaccess.h>
19 #include <linux/extable.h>
20 #include <linux/module.h>
21 #include <linux/slab.h>
22 #include <linux/hardirq.h>
23 #include <linux/ftrace.h>
24 #include <asm/set_memory.h>
25 #include <asm/sections.h>
26 #include <asm/dis.h>
27 #include "entry.h"
28 
29 DEFINE_PER_CPU(struct kprobe *, current_kprobe);
30 DEFINE_PER_CPU(struct kprobe_ctlblk, kprobe_ctlblk);
31 
32 struct kretprobe_blackpoint kretprobe_blacklist[] = { };
33 
34 DEFINE_INSN_CACHE_OPS(s390_insn);
35 
36 static int insn_page_in_use;
37 
38 void *alloc_insn_page(void)
39 {
40 	void *page;
41 
42 	page = module_alloc(PAGE_SIZE);
43 	if (!page)
44 		return NULL;
45 	__set_memory((unsigned long) page, 1, SET_MEMORY_RO | SET_MEMORY_X);
46 	return page;
47 }
48 
49 static void *alloc_s390_insn_page(void)
50 {
51 	if (xchg(&insn_page_in_use, 1) == 1)
52 		return NULL;
53 	return &kprobes_insn_page;
54 }
55 
56 static void free_s390_insn_page(void *page)
57 {
58 	xchg(&insn_page_in_use, 0);
59 }
60 
61 struct kprobe_insn_cache kprobe_s390_insn_slots = {
62 	.mutex = __MUTEX_INITIALIZER(kprobe_s390_insn_slots.mutex),
63 	.alloc = alloc_s390_insn_page,
64 	.free = free_s390_insn_page,
65 	.pages = LIST_HEAD_INIT(kprobe_s390_insn_slots.pages),
66 	.insn_size = MAX_INSN_SIZE,
67 };
68 
69 static void copy_instruction(struct kprobe *p)
70 {
71 	kprobe_opcode_t insn[MAX_INSN_SIZE];
72 	s64 disp, new_disp;
73 	u64 addr, new_addr;
74 	unsigned int len;
75 
76 	len = insn_length(*p->addr >> 8);
77 	memcpy(&insn, p->addr, len);
78 	p->opcode = insn[0];
79 	if (probe_is_insn_relative_long(&insn[0])) {
80 		/*
81 		 * For pc-relative instructions in RIL-b or RIL-c format patch
82 		 * the RI2 displacement field. We have already made sure that
83 		 * the insn slot for the patched instruction is within the same
84 		 * 2GB area as the original instruction (either kernel image or
85 		 * module area). Therefore the new displacement will always fit.
86 		 */
87 		disp = *(s32 *)&insn[1];
88 		addr = (u64)(unsigned long)p->addr;
89 		new_addr = (u64)(unsigned long)p->ainsn.insn;
90 		new_disp = ((addr + (disp * 2)) - new_addr) / 2;
91 		*(s32 *)&insn[1] = new_disp;
92 	}
93 	s390_kernel_write(p->ainsn.insn, &insn, len);
94 }
95 NOKPROBE_SYMBOL(copy_instruction);
96 
97 static int s390_get_insn_slot(struct kprobe *p)
98 {
99 	/*
100 	 * Get an insn slot that is within the same 2GB area like the original
101 	 * instruction. That way instructions with a 32bit signed displacement
102 	 * field can be patched and executed within the insn slot.
103 	 */
104 	p->ainsn.insn = NULL;
105 	if (is_kernel((unsigned long)p->addr))
106 		p->ainsn.insn = get_s390_insn_slot();
107 	else if (is_module_addr(p->addr))
108 		p->ainsn.insn = get_insn_slot();
109 	return p->ainsn.insn ? 0 : -ENOMEM;
110 }
111 NOKPROBE_SYMBOL(s390_get_insn_slot);
112 
113 static void s390_free_insn_slot(struct kprobe *p)
114 {
115 	if (!p->ainsn.insn)
116 		return;
117 	if (is_kernel((unsigned long)p->addr))
118 		free_s390_insn_slot(p->ainsn.insn, 0);
119 	else
120 		free_insn_slot(p->ainsn.insn, 0);
121 	p->ainsn.insn = NULL;
122 }
123 NOKPROBE_SYMBOL(s390_free_insn_slot);
124 
125 /* Check if paddr is at an instruction boundary */
126 static bool can_probe(unsigned long paddr)
127 {
128 	unsigned long addr, offset = 0;
129 	kprobe_opcode_t insn;
130 	struct kprobe *kp;
131 
132 	if (paddr & 0x01)
133 		return false;
134 
135 	if (!kallsyms_lookup_size_offset(paddr, NULL, &offset))
136 		return false;
137 
138 	/* Decode instructions */
139 	addr = paddr - offset;
140 	while (addr < paddr) {
141 		if (copy_from_kernel_nofault(&insn, (void *)addr, sizeof(insn)))
142 			return false;
143 
144 		if (insn >> 8 == 0) {
145 			if (insn != BREAKPOINT_INSTRUCTION) {
146 				/*
147 				 * Note that QEMU inserts opcode 0x0000 to implement
148 				 * software breakpoints for guests. Since the size of
149 				 * the original instruction is unknown, stop following
150 				 * instructions and prevent setting a kprobe.
151 				 */
152 				return false;
153 			}
154 			/*
155 			 * Check if the instruction has been modified by another
156 			 * kprobe, in which case the original instruction is
157 			 * decoded.
158 			 */
159 			kp = get_kprobe((void *)addr);
160 			if (!kp) {
161 				/* not a kprobe */
162 				return false;
163 			}
164 			insn = kp->opcode;
165 		}
166 		addr += insn_length(insn >> 8);
167 	}
168 	return addr == paddr;
169 }
170 
171 int arch_prepare_kprobe(struct kprobe *p)
172 {
173 	if (!can_probe((unsigned long)p->addr))
174 		return -EINVAL;
175 	/* Make sure the probe isn't going on a difficult instruction */
176 	if (probe_is_prohibited_opcode(p->addr))
177 		return -EINVAL;
178 	if (s390_get_insn_slot(p))
179 		return -ENOMEM;
180 	copy_instruction(p);
181 	return 0;
182 }
183 NOKPROBE_SYMBOL(arch_prepare_kprobe);
184 
185 struct swap_insn_args {
186 	struct kprobe *p;
187 	unsigned int arm_kprobe : 1;
188 };
189 
190 static int swap_instruction(void *data)
191 {
192 	struct swap_insn_args *args = data;
193 	struct kprobe *p = args->p;
194 	u16 opc;
195 
196 	opc = args->arm_kprobe ? BREAKPOINT_INSTRUCTION : p->opcode;
197 	s390_kernel_write(p->addr, &opc, sizeof(opc));
198 	return 0;
199 }
200 NOKPROBE_SYMBOL(swap_instruction);
201 
202 void arch_arm_kprobe(struct kprobe *p)
203 {
204 	struct swap_insn_args args = {.p = p, .arm_kprobe = 1};
205 
206 	stop_machine_cpuslocked(swap_instruction, &args, NULL);
207 }
208 NOKPROBE_SYMBOL(arch_arm_kprobe);
209 
210 void arch_disarm_kprobe(struct kprobe *p)
211 {
212 	struct swap_insn_args args = {.p = p, .arm_kprobe = 0};
213 
214 	stop_machine_cpuslocked(swap_instruction, &args, NULL);
215 }
216 NOKPROBE_SYMBOL(arch_disarm_kprobe);
217 
218 void arch_remove_kprobe(struct kprobe *p)
219 {
220 	s390_free_insn_slot(p);
221 }
222 NOKPROBE_SYMBOL(arch_remove_kprobe);
223 
224 static void enable_singlestep(struct kprobe_ctlblk *kcb,
225 			      struct pt_regs *regs,
226 			      unsigned long ip)
227 {
228 	struct per_regs per_kprobe;
229 
230 	/* Set up the PER control registers %cr9-%cr11 */
231 	per_kprobe.control = PER_EVENT_IFETCH;
232 	per_kprobe.start = ip;
233 	per_kprobe.end = ip;
234 
235 	/* Save control regs and psw mask */
236 	__ctl_store(kcb->kprobe_saved_ctl, 9, 11);
237 	kcb->kprobe_saved_imask = regs->psw.mask &
238 		(PSW_MASK_PER | PSW_MASK_IO | PSW_MASK_EXT);
239 
240 	/* Set PER control regs, turns on single step for the given address */
241 	__ctl_load(per_kprobe, 9, 11);
242 	regs->psw.mask |= PSW_MASK_PER;
243 	regs->psw.mask &= ~(PSW_MASK_IO | PSW_MASK_EXT);
244 	regs->psw.addr = ip;
245 }
246 NOKPROBE_SYMBOL(enable_singlestep);
247 
248 static void disable_singlestep(struct kprobe_ctlblk *kcb,
249 			       struct pt_regs *regs,
250 			       unsigned long ip)
251 {
252 	/* Restore control regs and psw mask, set new psw address */
253 	__ctl_load(kcb->kprobe_saved_ctl, 9, 11);
254 	regs->psw.mask &= ~PSW_MASK_PER;
255 	regs->psw.mask |= kcb->kprobe_saved_imask;
256 	regs->psw.addr = ip;
257 }
258 NOKPROBE_SYMBOL(disable_singlestep);
259 
260 /*
261  * Activate a kprobe by storing its pointer to current_kprobe. The
262  * previous kprobe is stored in kcb->prev_kprobe. A stack of up to
263  * two kprobes can be active, see KPROBE_REENTER.
264  */
265 static void push_kprobe(struct kprobe_ctlblk *kcb, struct kprobe *p)
266 {
267 	kcb->prev_kprobe.kp = __this_cpu_read(current_kprobe);
268 	kcb->prev_kprobe.status = kcb->kprobe_status;
269 	__this_cpu_write(current_kprobe, p);
270 }
271 NOKPROBE_SYMBOL(push_kprobe);
272 
273 /*
274  * Deactivate a kprobe by backing up to the previous state. If the
275  * current state is KPROBE_REENTER prev_kprobe.kp will be non-NULL,
276  * for any other state prev_kprobe.kp will be NULL.
277  */
278 static void pop_kprobe(struct kprobe_ctlblk *kcb)
279 {
280 	__this_cpu_write(current_kprobe, kcb->prev_kprobe.kp);
281 	kcb->kprobe_status = kcb->prev_kprobe.status;
282 }
283 NOKPROBE_SYMBOL(pop_kprobe);
284 
285 void arch_prepare_kretprobe(struct kretprobe_instance *ri, struct pt_regs *regs)
286 {
287 	ri->ret_addr = (kprobe_opcode_t *) regs->gprs[14];
288 	ri->fp = NULL;
289 
290 	/* Replace the return addr with trampoline addr */
291 	regs->gprs[14] = (unsigned long) &__kretprobe_trampoline;
292 }
293 NOKPROBE_SYMBOL(arch_prepare_kretprobe);
294 
295 static void kprobe_reenter_check(struct kprobe_ctlblk *kcb, struct kprobe *p)
296 {
297 	switch (kcb->kprobe_status) {
298 	case KPROBE_HIT_SSDONE:
299 	case KPROBE_HIT_ACTIVE:
300 		kprobes_inc_nmissed_count(p);
301 		break;
302 	case KPROBE_HIT_SS:
303 	case KPROBE_REENTER:
304 	default:
305 		/*
306 		 * A kprobe on the code path to single step an instruction
307 		 * is a BUG. The code path resides in the .kprobes.text
308 		 * section and is executed with interrupts disabled.
309 		 */
310 		pr_err("Failed to recover from reentered kprobes.\n");
311 		dump_kprobe(p);
312 		BUG();
313 	}
314 }
315 NOKPROBE_SYMBOL(kprobe_reenter_check);
316 
317 static int kprobe_handler(struct pt_regs *regs)
318 {
319 	struct kprobe_ctlblk *kcb;
320 	struct kprobe *p;
321 
322 	/*
323 	 * We want to disable preemption for the entire duration of kprobe
324 	 * processing. That includes the calls to the pre/post handlers
325 	 * and single stepping the kprobe instruction.
326 	 */
327 	preempt_disable();
328 	kcb = get_kprobe_ctlblk();
329 	p = get_kprobe((void *)(regs->psw.addr - 2));
330 
331 	if (p) {
332 		if (kprobe_running()) {
333 			/*
334 			 * We have hit a kprobe while another is still
335 			 * active. This can happen in the pre and post
336 			 * handler. Single step the instruction of the
337 			 * new probe but do not call any handler function
338 			 * of this secondary kprobe.
339 			 * push_kprobe and pop_kprobe saves and restores
340 			 * the currently active kprobe.
341 			 */
342 			kprobe_reenter_check(kcb, p);
343 			push_kprobe(kcb, p);
344 			kcb->kprobe_status = KPROBE_REENTER;
345 		} else {
346 			/*
347 			 * If we have no pre-handler or it returned 0, we
348 			 * continue with single stepping. If we have a
349 			 * pre-handler and it returned non-zero, it prepped
350 			 * for changing execution path, so get out doing
351 			 * nothing more here.
352 			 */
353 			push_kprobe(kcb, p);
354 			kcb->kprobe_status = KPROBE_HIT_ACTIVE;
355 			if (p->pre_handler && p->pre_handler(p, regs)) {
356 				pop_kprobe(kcb);
357 				preempt_enable_no_resched();
358 				return 1;
359 			}
360 			kcb->kprobe_status = KPROBE_HIT_SS;
361 		}
362 		enable_singlestep(kcb, regs, (unsigned long) p->ainsn.insn);
363 		return 1;
364 	} /* else:
365 	   * No kprobe at this address and no active kprobe. The trap has
366 	   * not been caused by a kprobe breakpoint. The race of breakpoint
367 	   * vs. kprobe remove does not exist because on s390 as we use
368 	   * stop_machine to arm/disarm the breakpoints.
369 	   */
370 	preempt_enable_no_resched();
371 	return 0;
372 }
373 NOKPROBE_SYMBOL(kprobe_handler);
374 
375 /*
376  * Function return probe trampoline:
377  *	- init_kprobes() establishes a probepoint here
378  *	- When the probed function returns, this probe
379  *		causes the handlers to fire
380  */
381 static void __used kretprobe_trampoline_holder(void)
382 {
383 	asm volatile(".global __kretprobe_trampoline\n"
384 		     "__kretprobe_trampoline: bcr 0,0\n");
385 }
386 
387 /*
388  * Called when the probe at kretprobe trampoline is hit
389  */
390 static int trampoline_probe_handler(struct kprobe *p, struct pt_regs *regs)
391 {
392 	regs->psw.addr = __kretprobe_trampoline_handler(regs, NULL);
393 	/*
394 	 * By returning a non-zero value, we are telling
395 	 * kprobe_handler() that we don't want the post_handler
396 	 * to run (and have re-enabled preemption)
397 	 */
398 	return 1;
399 }
400 NOKPROBE_SYMBOL(trampoline_probe_handler);
401 
402 /*
403  * Called after single-stepping.  p->addr is the address of the
404  * instruction whose first byte has been replaced by the "breakpoint"
405  * instruction.  To avoid the SMP problems that can occur when we
406  * temporarily put back the original opcode to single-step, we
407  * single-stepped a copy of the instruction.  The address of this
408  * copy is p->ainsn.insn.
409  */
410 static void resume_execution(struct kprobe *p, struct pt_regs *regs)
411 {
412 	struct kprobe_ctlblk *kcb = get_kprobe_ctlblk();
413 	unsigned long ip = regs->psw.addr;
414 	int fixup = probe_get_fixup_type(p->ainsn.insn);
415 
416 	if (fixup & FIXUP_PSW_NORMAL)
417 		ip += (unsigned long) p->addr - (unsigned long) p->ainsn.insn;
418 
419 	if (fixup & FIXUP_BRANCH_NOT_TAKEN) {
420 		int ilen = insn_length(p->ainsn.insn[0] >> 8);
421 		if (ip - (unsigned long) p->ainsn.insn == ilen)
422 			ip = (unsigned long) p->addr + ilen;
423 	}
424 
425 	if (fixup & FIXUP_RETURN_REGISTER) {
426 		int reg = (p->ainsn.insn[0] & 0xf0) >> 4;
427 		regs->gprs[reg] += (unsigned long) p->addr -
428 				   (unsigned long) p->ainsn.insn;
429 	}
430 
431 	disable_singlestep(kcb, regs, ip);
432 }
433 NOKPROBE_SYMBOL(resume_execution);
434 
435 static int post_kprobe_handler(struct pt_regs *regs)
436 {
437 	struct kprobe_ctlblk *kcb = get_kprobe_ctlblk();
438 	struct kprobe *p = kprobe_running();
439 
440 	if (!p)
441 		return 0;
442 
443 	if (kcb->kprobe_status != KPROBE_REENTER && p->post_handler) {
444 		kcb->kprobe_status = KPROBE_HIT_SSDONE;
445 		p->post_handler(p, regs, 0);
446 	}
447 
448 	resume_execution(p, regs);
449 	pop_kprobe(kcb);
450 	preempt_enable_no_resched();
451 
452 	/*
453 	 * if somebody else is singlestepping across a probe point, psw mask
454 	 * will have PER set, in which case, continue the remaining processing
455 	 * of do_single_step, as if this is not a probe hit.
456 	 */
457 	if (regs->psw.mask & PSW_MASK_PER)
458 		return 0;
459 
460 	return 1;
461 }
462 NOKPROBE_SYMBOL(post_kprobe_handler);
463 
464 static int kprobe_trap_handler(struct pt_regs *regs, int trapnr)
465 {
466 	struct kprobe_ctlblk *kcb = get_kprobe_ctlblk();
467 	struct kprobe *p = kprobe_running();
468 	const struct exception_table_entry *entry;
469 
470 	switch(kcb->kprobe_status) {
471 	case KPROBE_HIT_SS:
472 	case KPROBE_REENTER:
473 		/*
474 		 * We are here because the instruction being single
475 		 * stepped caused a page fault. We reset the current
476 		 * kprobe and the nip points back to the probe address
477 		 * and allow the page fault handler to continue as a
478 		 * normal page fault.
479 		 */
480 		disable_singlestep(kcb, regs, (unsigned long) p->addr);
481 		pop_kprobe(kcb);
482 		preempt_enable_no_resched();
483 		break;
484 	case KPROBE_HIT_ACTIVE:
485 	case KPROBE_HIT_SSDONE:
486 		/*
487 		 * In case the user-specified fault handler returned
488 		 * zero, try to fix up.
489 		 */
490 		entry = s390_search_extables(regs->psw.addr);
491 		if (entry && ex_handle(entry, regs))
492 			return 1;
493 
494 		/*
495 		 * fixup_exception() could not handle it,
496 		 * Let do_page_fault() fix it.
497 		 */
498 		break;
499 	default:
500 		break;
501 	}
502 	return 0;
503 }
504 NOKPROBE_SYMBOL(kprobe_trap_handler);
505 
506 int kprobe_fault_handler(struct pt_regs *regs, int trapnr)
507 {
508 	int ret;
509 
510 	if (regs->psw.mask & (PSW_MASK_IO | PSW_MASK_EXT))
511 		local_irq_disable();
512 	ret = kprobe_trap_handler(regs, trapnr);
513 	if (regs->psw.mask & (PSW_MASK_IO | PSW_MASK_EXT))
514 		local_irq_restore(regs->psw.mask & ~PSW_MASK_PER);
515 	return ret;
516 }
517 NOKPROBE_SYMBOL(kprobe_fault_handler);
518 
519 /*
520  * Wrapper routine to for handling exceptions.
521  */
522 int kprobe_exceptions_notify(struct notifier_block *self,
523 			     unsigned long val, void *data)
524 {
525 	struct die_args *args = (struct die_args *) data;
526 	struct pt_regs *regs = args->regs;
527 	int ret = NOTIFY_DONE;
528 
529 	if (regs->psw.mask & (PSW_MASK_IO | PSW_MASK_EXT))
530 		local_irq_disable();
531 
532 	switch (val) {
533 	case DIE_BPT:
534 		if (kprobe_handler(regs))
535 			ret = NOTIFY_STOP;
536 		break;
537 	case DIE_SSTEP:
538 		if (post_kprobe_handler(regs))
539 			ret = NOTIFY_STOP;
540 		break;
541 	case DIE_TRAP:
542 		if (!preemptible() && kprobe_running() &&
543 		    kprobe_trap_handler(regs, args->trapnr))
544 			ret = NOTIFY_STOP;
545 		break;
546 	default:
547 		break;
548 	}
549 
550 	if (regs->psw.mask & (PSW_MASK_IO | PSW_MASK_EXT))
551 		local_irq_restore(regs->psw.mask & ~PSW_MASK_PER);
552 
553 	return ret;
554 }
555 NOKPROBE_SYMBOL(kprobe_exceptions_notify);
556 
557 static struct kprobe trampoline = {
558 	.addr = (kprobe_opcode_t *) &__kretprobe_trampoline,
559 	.pre_handler = trampoline_probe_handler
560 };
561 
562 int __init arch_init_kprobes(void)
563 {
564 	return register_kprobe(&trampoline);
565 }
566 
567 int arch_trampoline_kprobe(struct kprobe *p)
568 {
569 	return p->addr == (kprobe_opcode_t *) &__kretprobe_trampoline;
570 }
571 NOKPROBE_SYMBOL(arch_trampoline_kprobe);
572