xref: /linux/arch/s390/kernel/kprobes.c (revision 67f49869106f78882a8a09b736d4884be85aba18)
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 "kprobes.h"
28 #include "entry.h"
29 
30 DEFINE_PER_CPU(struct kprobe *, current_kprobe);
31 DEFINE_PER_CPU(struct kprobe_ctlblk, kprobe_ctlblk);
32 
33 struct kretprobe_blackpoint kretprobe_blacklist[] = { };
34 
35 static int insn_page_in_use;
36 
37 void *alloc_insn_page(void)
38 {
39 	void *page;
40 
41 	page = module_alloc(PAGE_SIZE);
42 	if (!page)
43 		return NULL;
44 	__set_memory((unsigned long) page, 1, SET_MEMORY_RO | SET_MEMORY_X);
45 	return page;
46 }
47 
48 static void *alloc_s390_insn_page(void)
49 {
50 	if (xchg(&insn_page_in_use, 1) == 1)
51 		return NULL;
52 	return &kprobes_insn_page;
53 }
54 
55 static void free_s390_insn_page(void *page)
56 {
57 	xchg(&insn_page_in_use, 0);
58 }
59 
60 struct kprobe_insn_cache kprobe_s390_insn_slots = {
61 	.mutex = __MUTEX_INITIALIZER(kprobe_s390_insn_slots.mutex),
62 	.alloc = alloc_s390_insn_page,
63 	.free = free_s390_insn_page,
64 	.pages = LIST_HEAD_INIT(kprobe_s390_insn_slots.pages),
65 	.insn_size = MAX_INSN_SIZE,
66 };
67 
68 static void copy_instruction(struct kprobe *p)
69 {
70 	kprobe_opcode_t insn[MAX_INSN_SIZE];
71 	s64 disp, new_disp;
72 	u64 addr, new_addr;
73 	unsigned int len;
74 
75 	len = insn_length(*p->addr >> 8);
76 	memcpy(&insn, p->addr, len);
77 	p->opcode = insn[0];
78 	if (probe_is_insn_relative_long(&insn[0])) {
79 		/*
80 		 * For pc-relative instructions in RIL-b or RIL-c format patch
81 		 * the RI2 displacement field. We have already made sure that
82 		 * the insn slot for the patched instruction is within the same
83 		 * 2GB area as the original instruction (either kernel image or
84 		 * module area). Therefore the new displacement will always fit.
85 		 */
86 		disp = *(s32 *)&insn[1];
87 		addr = (u64)(unsigned long)p->addr;
88 		new_addr = (u64)(unsigned long)p->ainsn.insn;
89 		new_disp = ((addr + (disp * 2)) - new_addr) / 2;
90 		*(s32 *)&insn[1] = new_disp;
91 	}
92 	s390_kernel_write(p->ainsn.insn, &insn, len);
93 }
94 NOKPROBE_SYMBOL(copy_instruction);
95 
96 static int s390_get_insn_slot(struct kprobe *p)
97 {
98 	/*
99 	 * Get an insn slot that is within the same 2GB area like the original
100 	 * instruction. That way instructions with a 32bit signed displacement
101 	 * field can be patched and executed within the insn slot.
102 	 */
103 	p->ainsn.insn = NULL;
104 	if (is_kernel((unsigned long)p->addr))
105 		p->ainsn.insn = get_s390_insn_slot();
106 	else if (is_module_addr(p->addr))
107 		p->ainsn.insn = get_insn_slot();
108 	return p->ainsn.insn ? 0 : -ENOMEM;
109 }
110 NOKPROBE_SYMBOL(s390_get_insn_slot);
111 
112 static void s390_free_insn_slot(struct kprobe *p)
113 {
114 	if (!p->ainsn.insn)
115 		return;
116 	if (is_kernel((unsigned long)p->addr))
117 		free_s390_insn_slot(p->ainsn.insn, 0);
118 	else
119 		free_insn_slot(p->ainsn.insn, 0);
120 	p->ainsn.insn = NULL;
121 }
122 NOKPROBE_SYMBOL(s390_free_insn_slot);
123 
124 /* Check if paddr is at an instruction boundary */
125 static bool can_probe(unsigned long paddr)
126 {
127 	unsigned long addr, offset = 0;
128 	kprobe_opcode_t insn;
129 	struct kprobe *kp;
130 
131 	if (paddr & 0x01)
132 		return false;
133 
134 	if (!kallsyms_lookup_size_offset(paddr, NULL, &offset))
135 		return false;
136 
137 	/* Decode instructions */
138 	addr = paddr - offset;
139 	while (addr < paddr) {
140 		if (copy_from_kernel_nofault(&insn, (void *)addr, sizeof(insn)))
141 			return false;
142 
143 		if (insn >> 8 == 0) {
144 			if (insn != BREAKPOINT_INSTRUCTION) {
145 				/*
146 				 * Note that QEMU inserts opcode 0x0000 to implement
147 				 * software breakpoints for guests. Since the size of
148 				 * the original instruction is unknown, stop following
149 				 * instructions and prevent setting a kprobe.
150 				 */
151 				return false;
152 			}
153 			/*
154 			 * Check if the instruction has been modified by another
155 			 * kprobe, in which case the original instruction is
156 			 * decoded.
157 			 */
158 			kp = get_kprobe((void *)addr);
159 			if (!kp) {
160 				/* not a kprobe */
161 				return false;
162 			}
163 			insn = kp->opcode;
164 		}
165 		addr += insn_length(insn >> 8);
166 	}
167 	return addr == paddr;
168 }
169 
170 int arch_prepare_kprobe(struct kprobe *p)
171 {
172 	if (!can_probe((unsigned long)p->addr))
173 		return -EINVAL;
174 	/* Make sure the probe isn't going on a difficult instruction */
175 	if (probe_is_prohibited_opcode(p->addr))
176 		return -EINVAL;
177 	if (s390_get_insn_slot(p))
178 		return -ENOMEM;
179 	copy_instruction(p);
180 	return 0;
181 }
182 NOKPROBE_SYMBOL(arch_prepare_kprobe);
183 
184 struct swap_insn_args {
185 	struct kprobe *p;
186 	unsigned int arm_kprobe : 1;
187 };
188 
189 static int swap_instruction(void *data)
190 {
191 	struct swap_insn_args *args = data;
192 	struct kprobe *p = args->p;
193 	u16 opc;
194 
195 	opc = args->arm_kprobe ? BREAKPOINT_INSTRUCTION : p->opcode;
196 	s390_kernel_write(p->addr, &opc, sizeof(opc));
197 	return 0;
198 }
199 NOKPROBE_SYMBOL(swap_instruction);
200 
201 void arch_arm_kprobe(struct kprobe *p)
202 {
203 	struct swap_insn_args args = {.p = p, .arm_kprobe = 1};
204 
205 	stop_machine_cpuslocked(swap_instruction, &args, NULL);
206 }
207 NOKPROBE_SYMBOL(arch_arm_kprobe);
208 
209 void arch_disarm_kprobe(struct kprobe *p)
210 {
211 	struct swap_insn_args args = {.p = p, .arm_kprobe = 0};
212 
213 	stop_machine_cpuslocked(swap_instruction, &args, NULL);
214 }
215 NOKPROBE_SYMBOL(arch_disarm_kprobe);
216 
217 void arch_remove_kprobe(struct kprobe *p)
218 {
219 	s390_free_insn_slot(p);
220 }
221 NOKPROBE_SYMBOL(arch_remove_kprobe);
222 
223 static void enable_singlestep(struct kprobe_ctlblk *kcb,
224 			      struct pt_regs *regs,
225 			      unsigned long ip)
226 {
227 	struct per_regs per_kprobe;
228 
229 	/* Set up the PER control registers %cr9-%cr11 */
230 	per_kprobe.control = PER_EVENT_IFETCH;
231 	per_kprobe.start = ip;
232 	per_kprobe.end = ip;
233 
234 	/* Save control regs and psw mask */
235 	__ctl_store(kcb->kprobe_saved_ctl, 9, 11);
236 	kcb->kprobe_saved_imask = regs->psw.mask &
237 		(PSW_MASK_PER | PSW_MASK_IO | PSW_MASK_EXT);
238 
239 	/* Set PER control regs, turns on single step for the given address */
240 	__ctl_load(per_kprobe, 9, 11);
241 	regs->psw.mask |= PSW_MASK_PER;
242 	regs->psw.mask &= ~(PSW_MASK_IO | PSW_MASK_EXT);
243 	regs->psw.addr = ip;
244 }
245 NOKPROBE_SYMBOL(enable_singlestep);
246 
247 static void disable_singlestep(struct kprobe_ctlblk *kcb,
248 			       struct pt_regs *regs,
249 			       unsigned long ip)
250 {
251 	/* Restore control regs and psw mask, set new psw address */
252 	__ctl_load(kcb->kprobe_saved_ctl, 9, 11);
253 	regs->psw.mask &= ~PSW_MASK_PER;
254 	regs->psw.mask |= kcb->kprobe_saved_imask;
255 	regs->psw.addr = ip;
256 }
257 NOKPROBE_SYMBOL(disable_singlestep);
258 
259 /*
260  * Activate a kprobe by storing its pointer to current_kprobe. The
261  * previous kprobe is stored in kcb->prev_kprobe. A stack of up to
262  * two kprobes can be active, see KPROBE_REENTER.
263  */
264 static void push_kprobe(struct kprobe_ctlblk *kcb, struct kprobe *p)
265 {
266 	kcb->prev_kprobe.kp = __this_cpu_read(current_kprobe);
267 	kcb->prev_kprobe.status = kcb->kprobe_status;
268 	__this_cpu_write(current_kprobe, p);
269 }
270 NOKPROBE_SYMBOL(push_kprobe);
271 
272 /*
273  * Deactivate a kprobe by backing up to the previous state. If the
274  * current state is KPROBE_REENTER prev_kprobe.kp will be non-NULL,
275  * for any other state prev_kprobe.kp will be NULL.
276  */
277 static void pop_kprobe(struct kprobe_ctlblk *kcb)
278 {
279 	__this_cpu_write(current_kprobe, kcb->prev_kprobe.kp);
280 	kcb->kprobe_status = kcb->prev_kprobe.status;
281 }
282 NOKPROBE_SYMBOL(pop_kprobe);
283 
284 void arch_prepare_kretprobe(struct kretprobe_instance *ri, struct pt_regs *regs)
285 {
286 	ri->ret_addr = (kprobe_opcode_t *)regs->gprs[14];
287 	ri->fp = (void *)regs->gprs[15];
288 
289 	/* Replace the return addr with trampoline addr */
290 	regs->gprs[14] = (unsigned long)&__kretprobe_trampoline;
291 }
292 NOKPROBE_SYMBOL(arch_prepare_kretprobe);
293 
294 static void kprobe_reenter_check(struct kprobe_ctlblk *kcb, struct kprobe *p)
295 {
296 	switch (kcb->kprobe_status) {
297 	case KPROBE_HIT_SSDONE:
298 	case KPROBE_HIT_ACTIVE:
299 		kprobes_inc_nmissed_count(p);
300 		break;
301 	case KPROBE_HIT_SS:
302 	case KPROBE_REENTER:
303 	default:
304 		/*
305 		 * A kprobe on the code path to single step an instruction
306 		 * is a BUG. The code path resides in the .kprobes.text
307 		 * section and is executed with interrupts disabled.
308 		 */
309 		pr_err("Failed to recover from reentered kprobes.\n");
310 		dump_kprobe(p);
311 		BUG();
312 	}
313 }
314 NOKPROBE_SYMBOL(kprobe_reenter_check);
315 
316 static int kprobe_handler(struct pt_regs *regs)
317 {
318 	struct kprobe_ctlblk *kcb;
319 	struct kprobe *p;
320 
321 	/*
322 	 * We want to disable preemption for the entire duration of kprobe
323 	 * processing. That includes the calls to the pre/post handlers
324 	 * and single stepping the kprobe instruction.
325 	 */
326 	preempt_disable();
327 	kcb = get_kprobe_ctlblk();
328 	p = get_kprobe((void *)(regs->psw.addr - 2));
329 
330 	if (p) {
331 		if (kprobe_running()) {
332 			/*
333 			 * We have hit a kprobe while another is still
334 			 * active. This can happen in the pre and post
335 			 * handler. Single step the instruction of the
336 			 * new probe but do not call any handler function
337 			 * of this secondary kprobe.
338 			 * push_kprobe and pop_kprobe saves and restores
339 			 * the currently active kprobe.
340 			 */
341 			kprobe_reenter_check(kcb, p);
342 			push_kprobe(kcb, p);
343 			kcb->kprobe_status = KPROBE_REENTER;
344 		} else {
345 			/*
346 			 * If we have no pre-handler or it returned 0, we
347 			 * continue with single stepping. If we have a
348 			 * pre-handler and it returned non-zero, it prepped
349 			 * for changing execution path, so get out doing
350 			 * nothing more here.
351 			 */
352 			push_kprobe(kcb, p);
353 			kcb->kprobe_status = KPROBE_HIT_ACTIVE;
354 			if (p->pre_handler && p->pre_handler(p, regs)) {
355 				pop_kprobe(kcb);
356 				preempt_enable_no_resched();
357 				return 1;
358 			}
359 			kcb->kprobe_status = KPROBE_HIT_SS;
360 		}
361 		enable_singlestep(kcb, regs, (unsigned long) p->ainsn.insn);
362 		return 1;
363 	} /* else:
364 	   * No kprobe at this address and no active kprobe. The trap has
365 	   * not been caused by a kprobe breakpoint. The race of breakpoint
366 	   * vs. kprobe remove does not exist because on s390 as we use
367 	   * stop_machine to arm/disarm the breakpoints.
368 	   */
369 	preempt_enable_no_resched();
370 	return 0;
371 }
372 NOKPROBE_SYMBOL(kprobe_handler);
373 
374 void arch_kretprobe_fixup_return(struct pt_regs *regs,
375 				 kprobe_opcode_t *correct_ret_addr)
376 {
377 	/* Replace fake return address with real one. */
378 	regs->gprs[14] = (unsigned long)correct_ret_addr;
379 }
380 NOKPROBE_SYMBOL(arch_kretprobe_fixup_return);
381 
382 /*
383  * Called from __kretprobe_trampoline
384  */
385 void trampoline_probe_handler(struct pt_regs *regs)
386 {
387 	kretprobe_trampoline_handler(regs, (void *)regs->gprs[15]);
388 }
389 NOKPROBE_SYMBOL(trampoline_probe_handler);
390 
391 /* assembler function that handles the kretprobes must not be probed itself */
392 NOKPROBE_SYMBOL(__kretprobe_trampoline);
393 
394 /*
395  * Called after single-stepping.  p->addr is the address of the
396  * instruction whose first byte has been replaced by the "breakpoint"
397  * instruction.  To avoid the SMP problems that can occur when we
398  * temporarily put back the original opcode to single-step, we
399  * single-stepped a copy of the instruction.  The address of this
400  * copy is p->ainsn.insn.
401  */
402 static void resume_execution(struct kprobe *p, struct pt_regs *regs)
403 {
404 	struct kprobe_ctlblk *kcb = get_kprobe_ctlblk();
405 	unsigned long ip = regs->psw.addr;
406 	int fixup = probe_get_fixup_type(p->ainsn.insn);
407 
408 	if (fixup & FIXUP_PSW_NORMAL)
409 		ip += (unsigned long) p->addr - (unsigned long) p->ainsn.insn;
410 
411 	if (fixup & FIXUP_BRANCH_NOT_TAKEN) {
412 		int ilen = insn_length(p->ainsn.insn[0] >> 8);
413 		if (ip - (unsigned long) p->ainsn.insn == ilen)
414 			ip = (unsigned long) p->addr + ilen;
415 	}
416 
417 	if (fixup & FIXUP_RETURN_REGISTER) {
418 		int reg = (p->ainsn.insn[0] & 0xf0) >> 4;
419 		regs->gprs[reg] += (unsigned long) p->addr -
420 				   (unsigned long) p->ainsn.insn;
421 	}
422 
423 	disable_singlestep(kcb, regs, ip);
424 }
425 NOKPROBE_SYMBOL(resume_execution);
426 
427 static int post_kprobe_handler(struct pt_regs *regs)
428 {
429 	struct kprobe_ctlblk *kcb = get_kprobe_ctlblk();
430 	struct kprobe *p = kprobe_running();
431 
432 	if (!p)
433 		return 0;
434 
435 	if (kcb->kprobe_status != KPROBE_REENTER && p->post_handler) {
436 		kcb->kprobe_status = KPROBE_HIT_SSDONE;
437 		p->post_handler(p, regs, 0);
438 	}
439 
440 	resume_execution(p, regs);
441 	pop_kprobe(kcb);
442 	preempt_enable_no_resched();
443 
444 	/*
445 	 * if somebody else is singlestepping across a probe point, psw mask
446 	 * will have PER set, in which case, continue the remaining processing
447 	 * of do_single_step, as if this is not a probe hit.
448 	 */
449 	if (regs->psw.mask & PSW_MASK_PER)
450 		return 0;
451 
452 	return 1;
453 }
454 NOKPROBE_SYMBOL(post_kprobe_handler);
455 
456 static int kprobe_trap_handler(struct pt_regs *regs, int trapnr)
457 {
458 	struct kprobe_ctlblk *kcb = get_kprobe_ctlblk();
459 	struct kprobe *p = kprobe_running();
460 
461 	switch(kcb->kprobe_status) {
462 	case KPROBE_HIT_SS:
463 	case KPROBE_REENTER:
464 		/*
465 		 * We are here because the instruction being single
466 		 * stepped caused a page fault. We reset the current
467 		 * kprobe and the nip points back to the probe address
468 		 * and allow the page fault handler to continue as a
469 		 * normal page fault.
470 		 */
471 		disable_singlestep(kcb, regs, (unsigned long) p->addr);
472 		pop_kprobe(kcb);
473 		preempt_enable_no_resched();
474 		break;
475 	case KPROBE_HIT_ACTIVE:
476 	case KPROBE_HIT_SSDONE:
477 		/*
478 		 * In case the user-specified fault handler returned
479 		 * zero, try to fix up.
480 		 */
481 		if (fixup_exception(regs))
482 			return 1;
483 		/*
484 		 * fixup_exception() could not handle it,
485 		 * Let do_page_fault() fix it.
486 		 */
487 		break;
488 	default:
489 		break;
490 	}
491 	return 0;
492 }
493 NOKPROBE_SYMBOL(kprobe_trap_handler);
494 
495 int kprobe_fault_handler(struct pt_regs *regs, int trapnr)
496 {
497 	int ret;
498 
499 	if (regs->psw.mask & (PSW_MASK_IO | PSW_MASK_EXT))
500 		local_irq_disable();
501 	ret = kprobe_trap_handler(regs, trapnr);
502 	if (regs->psw.mask & (PSW_MASK_IO | PSW_MASK_EXT))
503 		local_irq_restore(regs->psw.mask & ~PSW_MASK_PER);
504 	return ret;
505 }
506 NOKPROBE_SYMBOL(kprobe_fault_handler);
507 
508 /*
509  * Wrapper routine to for handling exceptions.
510  */
511 int kprobe_exceptions_notify(struct notifier_block *self,
512 			     unsigned long val, void *data)
513 {
514 	struct die_args *args = (struct die_args *) data;
515 	struct pt_regs *regs = args->regs;
516 	int ret = NOTIFY_DONE;
517 
518 	if (regs->psw.mask & (PSW_MASK_IO | PSW_MASK_EXT))
519 		local_irq_disable();
520 
521 	switch (val) {
522 	case DIE_BPT:
523 		if (kprobe_handler(regs))
524 			ret = NOTIFY_STOP;
525 		break;
526 	case DIE_SSTEP:
527 		if (post_kprobe_handler(regs))
528 			ret = NOTIFY_STOP;
529 		break;
530 	case DIE_TRAP:
531 		if (!preemptible() && kprobe_running() &&
532 		    kprobe_trap_handler(regs, args->trapnr))
533 			ret = NOTIFY_STOP;
534 		break;
535 	default:
536 		break;
537 	}
538 
539 	if (regs->psw.mask & (PSW_MASK_IO | PSW_MASK_EXT))
540 		local_irq_restore(regs->psw.mask & ~PSW_MASK_PER);
541 
542 	return ret;
543 }
544 NOKPROBE_SYMBOL(kprobe_exceptions_notify);
545 
546 int __init arch_init_kprobes(void)
547 {
548 	return 0;
549 }
550 
551 int arch_trampoline_kprobe(struct kprobe *p)
552 {
553 	return 0;
554 }
555 NOKPROBE_SYMBOL(arch_trampoline_kprobe);
556