Loading...
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#include <linux/kprobes.h>
11#include <linux/ptrace.h>
12#include <linux/preempt.h>
13#include <linux/stop_machine.h>
14#include <linux/kdebug.h>
15#include <linux/uaccess.h>
16#include <linux/extable.h>
17#include <linux/module.h>
18#include <linux/slab.h>
19#include <linux/hardirq.h>
20#include <linux/ftrace.h>
21#include <asm/set_memory.h>
22#include <asm/sections.h>
23#include <asm/dis.h>
24
25DEFINE_PER_CPU(struct kprobe *, current_kprobe);
26DEFINE_PER_CPU(struct kprobe_ctlblk, kprobe_ctlblk);
27
28struct kretprobe_blackpoint kretprobe_blacklist[] = { };
29
30DEFINE_INSN_CACHE_OPS(s390_insn);
31
32static int insn_page_in_use;
33static char insn_page[PAGE_SIZE] __aligned(PAGE_SIZE);
34
35static void *alloc_s390_insn_page(void)
36{
37 if (xchg(&insn_page_in_use, 1) == 1)
38 return NULL;
39 set_memory_x((unsigned long) &insn_page, 1);
40 return &insn_page;
41}
42
43static void free_s390_insn_page(void *page)
44{
45 set_memory_nx((unsigned long) page, 1);
46 xchg(&insn_page_in_use, 0);
47}
48
49struct kprobe_insn_cache kprobe_s390_insn_slots = {
50 .mutex = __MUTEX_INITIALIZER(kprobe_s390_insn_slots.mutex),
51 .alloc = alloc_s390_insn_page,
52 .free = free_s390_insn_page,
53 .pages = LIST_HEAD_INIT(kprobe_s390_insn_slots.pages),
54 .insn_size = MAX_INSN_SIZE,
55};
56
57static void copy_instruction(struct kprobe *p)
58{
59 unsigned long ip = (unsigned long) p->addr;
60 s64 disp, new_disp;
61 u64 addr, new_addr;
62
63 if (ftrace_location(ip) == ip) {
64 /*
65 * If kprobes patches the instruction that is morphed by
66 * ftrace make sure that kprobes always sees the branch
67 * "jg .+24" that skips the mcount block or the "brcl 0,0"
68 * in case of hotpatch.
69 */
70 ftrace_generate_nop_insn((struct ftrace_insn *)p->ainsn.insn);
71 p->ainsn.is_ftrace_insn = 1;
72 } else
73 memcpy(p->ainsn.insn, p->addr, insn_length(*p->addr >> 8));
74 p->opcode = p->ainsn.insn[0];
75 if (!probe_is_insn_relative_long(p->ainsn.insn))
76 return;
77 /*
78 * For pc-relative instructions in RIL-b or RIL-c format patch the
79 * RI2 displacement field. We have already made sure that the insn
80 * slot for the patched instruction is within the same 2GB area
81 * as the original instruction (either kernel image or module area).
82 * Therefore the new displacement will always fit.
83 */
84 disp = *(s32 *)&p->ainsn.insn[1];
85 addr = (u64)(unsigned long)p->addr;
86 new_addr = (u64)(unsigned long)p->ainsn.insn;
87 new_disp = ((addr + (disp * 2)) - new_addr) / 2;
88 *(s32 *)&p->ainsn.insn[1] = new_disp;
89}
90NOKPROBE_SYMBOL(copy_instruction);
91
92static inline int is_kernel_addr(void *addr)
93{
94 return addr < (void *)_end;
95}
96
97static 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_addr(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}
111NOKPROBE_SYMBOL(s390_get_insn_slot);
112
113static void s390_free_insn_slot(struct kprobe *p)
114{
115 if (!p->ainsn.insn)
116 return;
117 if (is_kernel_addr(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}
123NOKPROBE_SYMBOL(s390_free_insn_slot);
124
125int arch_prepare_kprobe(struct kprobe *p)
126{
127 if ((unsigned long) p->addr & 0x01)
128 return -EINVAL;
129 /* Make sure the probe isn't going on a difficult instruction */
130 if (probe_is_prohibited_opcode(p->addr))
131 return -EINVAL;
132 if (s390_get_insn_slot(p))
133 return -ENOMEM;
134 copy_instruction(p);
135 return 0;
136}
137NOKPROBE_SYMBOL(arch_prepare_kprobe);
138
139int arch_check_ftrace_location(struct kprobe *p)
140{
141 return 0;
142}
143
144struct swap_insn_args {
145 struct kprobe *p;
146 unsigned int arm_kprobe : 1;
147};
148
149static int swap_instruction(void *data)
150{
151 struct swap_insn_args *args = data;
152 struct ftrace_insn new_insn, *insn;
153 struct kprobe *p = args->p;
154 size_t len;
155
156 new_insn.opc = args->arm_kprobe ? BREAKPOINT_INSTRUCTION : p->opcode;
157 len = sizeof(new_insn.opc);
158 if (!p->ainsn.is_ftrace_insn)
159 goto skip_ftrace;
160 len = sizeof(new_insn);
161 insn = (struct ftrace_insn *) p->addr;
162 if (args->arm_kprobe) {
163 if (is_ftrace_nop(insn))
164 new_insn.disp = KPROBE_ON_FTRACE_NOP;
165 else
166 new_insn.disp = KPROBE_ON_FTRACE_CALL;
167 } else {
168 ftrace_generate_call_insn(&new_insn, (unsigned long)p->addr);
169 if (insn->disp == KPROBE_ON_FTRACE_NOP)
170 ftrace_generate_nop_insn(&new_insn);
171 }
172skip_ftrace:
173 s390_kernel_write(p->addr, &new_insn, len);
174 return 0;
175}
176NOKPROBE_SYMBOL(swap_instruction);
177
178void arch_arm_kprobe(struct kprobe *p)
179{
180 struct swap_insn_args args = {.p = p, .arm_kprobe = 1};
181
182 stop_machine_cpuslocked(swap_instruction, &args, NULL);
183}
184NOKPROBE_SYMBOL(arch_arm_kprobe);
185
186void arch_disarm_kprobe(struct kprobe *p)
187{
188 struct swap_insn_args args = {.p = p, .arm_kprobe = 0};
189
190 stop_machine_cpuslocked(swap_instruction, &args, NULL);
191}
192NOKPROBE_SYMBOL(arch_disarm_kprobe);
193
194void arch_remove_kprobe(struct kprobe *p)
195{
196 s390_free_insn_slot(p);
197}
198NOKPROBE_SYMBOL(arch_remove_kprobe);
199
200static void enable_singlestep(struct kprobe_ctlblk *kcb,
201 struct pt_regs *regs,
202 unsigned long ip)
203{
204 struct per_regs per_kprobe;
205
206 /* Set up the PER control registers %cr9-%cr11 */
207 per_kprobe.control = PER_EVENT_IFETCH;
208 per_kprobe.start = ip;
209 per_kprobe.end = ip;
210
211 /* Save control regs and psw mask */
212 __ctl_store(kcb->kprobe_saved_ctl, 9, 11);
213 kcb->kprobe_saved_imask = regs->psw.mask &
214 (PSW_MASK_PER | PSW_MASK_IO | PSW_MASK_EXT);
215
216 /* Set PER control regs, turns on single step for the given address */
217 __ctl_load(per_kprobe, 9, 11);
218 regs->psw.mask |= PSW_MASK_PER;
219 regs->psw.mask &= ~(PSW_MASK_IO | PSW_MASK_EXT);
220 regs->psw.addr = ip;
221}
222NOKPROBE_SYMBOL(enable_singlestep);
223
224static void disable_singlestep(struct kprobe_ctlblk *kcb,
225 struct pt_regs *regs,
226 unsigned long ip)
227{
228 /* Restore control regs and psw mask, set new psw address */
229 __ctl_load(kcb->kprobe_saved_ctl, 9, 11);
230 regs->psw.mask &= ~PSW_MASK_PER;
231 regs->psw.mask |= kcb->kprobe_saved_imask;
232 regs->psw.addr = ip;
233}
234NOKPROBE_SYMBOL(disable_singlestep);
235
236/*
237 * Activate a kprobe by storing its pointer to current_kprobe. The
238 * previous kprobe is stored in kcb->prev_kprobe. A stack of up to
239 * two kprobes can be active, see KPROBE_REENTER.
240 */
241static void push_kprobe(struct kprobe_ctlblk *kcb, struct kprobe *p)
242{
243 kcb->prev_kprobe.kp = __this_cpu_read(current_kprobe);
244 kcb->prev_kprobe.status = kcb->kprobe_status;
245 __this_cpu_write(current_kprobe, p);
246}
247NOKPROBE_SYMBOL(push_kprobe);
248
249/*
250 * Deactivate a kprobe by backing up to the previous state. If the
251 * current state is KPROBE_REENTER prev_kprobe.kp will be non-NULL,
252 * for any other state prev_kprobe.kp will be NULL.
253 */
254static void pop_kprobe(struct kprobe_ctlblk *kcb)
255{
256 __this_cpu_write(current_kprobe, kcb->prev_kprobe.kp);
257 kcb->kprobe_status = kcb->prev_kprobe.status;
258}
259NOKPROBE_SYMBOL(pop_kprobe);
260
261void arch_prepare_kretprobe(struct kretprobe_instance *ri, struct pt_regs *regs)
262{
263 ri->ret_addr = (kprobe_opcode_t *) regs->gprs[14];
264
265 /* Replace the return addr with trampoline addr */
266 regs->gprs[14] = (unsigned long) &kretprobe_trampoline;
267}
268NOKPROBE_SYMBOL(arch_prepare_kretprobe);
269
270static void kprobe_reenter_check(struct kprobe_ctlblk *kcb, struct kprobe *p)
271{
272 switch (kcb->kprobe_status) {
273 case KPROBE_HIT_SSDONE:
274 case KPROBE_HIT_ACTIVE:
275 kprobes_inc_nmissed_count(p);
276 break;
277 case KPROBE_HIT_SS:
278 case KPROBE_REENTER:
279 default:
280 /*
281 * A kprobe on the code path to single step an instruction
282 * is a BUG. The code path resides in the .kprobes.text
283 * section and is executed with interrupts disabled.
284 */
285 pr_err("Invalid kprobe detected.\n");
286 dump_kprobe(p);
287 BUG();
288 }
289}
290NOKPROBE_SYMBOL(kprobe_reenter_check);
291
292static int kprobe_handler(struct pt_regs *regs)
293{
294 struct kprobe_ctlblk *kcb;
295 struct kprobe *p;
296
297 /*
298 * We want to disable preemption for the entire duration of kprobe
299 * processing. That includes the calls to the pre/post handlers
300 * and single stepping the kprobe instruction.
301 */
302 preempt_disable();
303 kcb = get_kprobe_ctlblk();
304 p = get_kprobe((void *)(regs->psw.addr - 2));
305
306 if (p) {
307 if (kprobe_running()) {
308 /*
309 * We have hit a kprobe while another is still
310 * active. This can happen in the pre and post
311 * handler. Single step the instruction of the
312 * new probe but do not call any handler function
313 * of this secondary kprobe.
314 * push_kprobe and pop_kprobe saves and restores
315 * the currently active kprobe.
316 */
317 kprobe_reenter_check(kcb, p);
318 push_kprobe(kcb, p);
319 kcb->kprobe_status = KPROBE_REENTER;
320 } else {
321 /*
322 * If we have no pre-handler or it returned 0, we
323 * continue with single stepping. If we have a
324 * pre-handler and it returned non-zero, it prepped
325 * for changing execution path, so get out doing
326 * nothing more here.
327 */
328 push_kprobe(kcb, p);
329 kcb->kprobe_status = KPROBE_HIT_ACTIVE;
330 if (p->pre_handler && p->pre_handler(p, regs)) {
331 pop_kprobe(kcb);
332 preempt_enable_no_resched();
333 return 1;
334 }
335 kcb->kprobe_status = KPROBE_HIT_SS;
336 }
337 enable_singlestep(kcb, regs, (unsigned long) p->ainsn.insn);
338 return 1;
339 } /* else:
340 * No kprobe at this address and no active kprobe. The trap has
341 * not been caused by a kprobe breakpoint. The race of breakpoint
342 * vs. kprobe remove does not exist because on s390 as we use
343 * stop_machine to arm/disarm the breakpoints.
344 */
345 preempt_enable_no_resched();
346 return 0;
347}
348NOKPROBE_SYMBOL(kprobe_handler);
349
350/*
351 * Function return probe trampoline:
352 * - init_kprobes() establishes a probepoint here
353 * - When the probed function returns, this probe
354 * causes the handlers to fire
355 */
356static void __used kretprobe_trampoline_holder(void)
357{
358 asm volatile(".global kretprobe_trampoline\n"
359 "kretprobe_trampoline: bcr 0,0\n");
360}
361
362/*
363 * Called when the probe at kretprobe trampoline is hit
364 */
365static int trampoline_probe_handler(struct kprobe *p, struct pt_regs *regs)
366{
367 struct kretprobe_instance *ri;
368 struct hlist_head *head, empty_rp;
369 struct hlist_node *tmp;
370 unsigned long flags, orig_ret_address;
371 unsigned long trampoline_address;
372 kprobe_opcode_t *correct_ret_addr;
373
374 INIT_HLIST_HEAD(&empty_rp);
375 kretprobe_hash_lock(current, &head, &flags);
376
377 /*
378 * It is possible to have multiple instances associated with a given
379 * task either because an multiple functions in the call path
380 * have a return probe installed on them, and/or more than one return
381 * return probe was registered for a target function.
382 *
383 * We can handle this because:
384 * - instances are always inserted at the head of the list
385 * - when multiple return probes are registered for the same
386 * function, the first instance's ret_addr will point to the
387 * real return address, and all the rest will point to
388 * kretprobe_trampoline
389 */
390 ri = NULL;
391 orig_ret_address = 0;
392 correct_ret_addr = NULL;
393 trampoline_address = (unsigned long) &kretprobe_trampoline;
394 hlist_for_each_entry_safe(ri, tmp, head, hlist) {
395 if (ri->task != current)
396 /* another task is sharing our hash bucket */
397 continue;
398
399 orig_ret_address = (unsigned long) ri->ret_addr;
400
401 if (orig_ret_address != trampoline_address)
402 /*
403 * This is the real return address. Any other
404 * instances associated with this task are for
405 * other calls deeper on the call stack
406 */
407 break;
408 }
409
410 kretprobe_assert(ri, orig_ret_address, trampoline_address);
411
412 correct_ret_addr = ri->ret_addr;
413 hlist_for_each_entry_safe(ri, tmp, head, hlist) {
414 if (ri->task != current)
415 /* another task is sharing our hash bucket */
416 continue;
417
418 orig_ret_address = (unsigned long) ri->ret_addr;
419
420 if (ri->rp && ri->rp->handler) {
421 ri->ret_addr = correct_ret_addr;
422 ri->rp->handler(ri, regs);
423 }
424
425 recycle_rp_inst(ri, &empty_rp);
426
427 if (orig_ret_address != trampoline_address)
428 /*
429 * This is the real return address. Any other
430 * instances associated with this task are for
431 * other calls deeper on the call stack
432 */
433 break;
434 }
435
436 regs->psw.addr = orig_ret_address;
437
438 kretprobe_hash_unlock(current, &flags);
439
440 hlist_for_each_entry_safe(ri, tmp, &empty_rp, hlist) {
441 hlist_del(&ri->hlist);
442 kfree(ri);
443 }
444 /*
445 * By returning a non-zero value, we are telling
446 * kprobe_handler() that we don't want the post_handler
447 * to run (and have re-enabled preemption)
448 */
449 return 1;
450}
451NOKPROBE_SYMBOL(trampoline_probe_handler);
452
453/*
454 * Called after single-stepping. p->addr is the address of the
455 * instruction whose first byte has been replaced by the "breakpoint"
456 * instruction. To avoid the SMP problems that can occur when we
457 * temporarily put back the original opcode to single-step, we
458 * single-stepped a copy of the instruction. The address of this
459 * copy is p->ainsn.insn.
460 */
461static void resume_execution(struct kprobe *p, struct pt_regs *regs)
462{
463 struct kprobe_ctlblk *kcb = get_kprobe_ctlblk();
464 unsigned long ip = regs->psw.addr;
465 int fixup = probe_get_fixup_type(p->ainsn.insn);
466
467 /* Check if the kprobes location is an enabled ftrace caller */
468 if (p->ainsn.is_ftrace_insn) {
469 struct ftrace_insn *insn = (struct ftrace_insn *) p->addr;
470 struct ftrace_insn call_insn;
471
472 ftrace_generate_call_insn(&call_insn, (unsigned long) p->addr);
473 /*
474 * A kprobe on an enabled ftrace call site actually single
475 * stepped an unconditional branch (ftrace nop equivalent).
476 * Now we need to fixup things and pretend that a brasl r0,...
477 * was executed instead.
478 */
479 if (insn->disp == KPROBE_ON_FTRACE_CALL) {
480 ip += call_insn.disp * 2 - MCOUNT_INSN_SIZE;
481 regs->gprs[0] = (unsigned long)p->addr + sizeof(*insn);
482 }
483 }
484
485 if (fixup & FIXUP_PSW_NORMAL)
486 ip += (unsigned long) p->addr - (unsigned long) p->ainsn.insn;
487
488 if (fixup & FIXUP_BRANCH_NOT_TAKEN) {
489 int ilen = insn_length(p->ainsn.insn[0] >> 8);
490 if (ip - (unsigned long) p->ainsn.insn == ilen)
491 ip = (unsigned long) p->addr + ilen;
492 }
493
494 if (fixup & FIXUP_RETURN_REGISTER) {
495 int reg = (p->ainsn.insn[0] & 0xf0) >> 4;
496 regs->gprs[reg] += (unsigned long) p->addr -
497 (unsigned long) p->ainsn.insn;
498 }
499
500 disable_singlestep(kcb, regs, ip);
501}
502NOKPROBE_SYMBOL(resume_execution);
503
504static int post_kprobe_handler(struct pt_regs *regs)
505{
506 struct kprobe_ctlblk *kcb = get_kprobe_ctlblk();
507 struct kprobe *p = kprobe_running();
508
509 if (!p)
510 return 0;
511
512 if (kcb->kprobe_status != KPROBE_REENTER && p->post_handler) {
513 kcb->kprobe_status = KPROBE_HIT_SSDONE;
514 p->post_handler(p, regs, 0);
515 }
516
517 resume_execution(p, regs);
518 pop_kprobe(kcb);
519 preempt_enable_no_resched();
520
521 /*
522 * if somebody else is singlestepping across a probe point, psw mask
523 * will have PER set, in which case, continue the remaining processing
524 * of do_single_step, as if this is not a probe hit.
525 */
526 if (regs->psw.mask & PSW_MASK_PER)
527 return 0;
528
529 return 1;
530}
531NOKPROBE_SYMBOL(post_kprobe_handler);
532
533static int kprobe_trap_handler(struct pt_regs *regs, int trapnr)
534{
535 struct kprobe_ctlblk *kcb = get_kprobe_ctlblk();
536 struct kprobe *p = kprobe_running();
537 const struct exception_table_entry *entry;
538
539 switch(kcb->kprobe_status) {
540 case KPROBE_HIT_SS:
541 case KPROBE_REENTER:
542 /*
543 * We are here because the instruction being single
544 * stepped caused a page fault. We reset the current
545 * kprobe and the nip points back to the probe address
546 * and allow the page fault handler to continue as a
547 * normal page fault.
548 */
549 disable_singlestep(kcb, regs, (unsigned long) p->addr);
550 pop_kprobe(kcb);
551 preempt_enable_no_resched();
552 break;
553 case KPROBE_HIT_ACTIVE:
554 case KPROBE_HIT_SSDONE:
555 /*
556 * We increment the nmissed count for accounting,
557 * we can also use npre/npostfault count for accounting
558 * these specific fault cases.
559 */
560 kprobes_inc_nmissed_count(p);
561
562 /*
563 * We come here because instructions in the pre/post
564 * handler caused the page_fault, this could happen
565 * if handler tries to access user space by
566 * copy_from_user(), get_user() etc. Let the
567 * user-specified handler try to fix it first.
568 */
569 if (p->fault_handler && p->fault_handler(p, regs, trapnr))
570 return 1;
571
572 /*
573 * In case the user-specified fault handler returned
574 * zero, try to fix up.
575 */
576 entry = s390_search_extables(regs->psw.addr);
577 if (entry) {
578 regs->psw.addr = extable_fixup(entry);
579 return 1;
580 }
581
582 /*
583 * fixup_exception() could not handle it,
584 * Let do_page_fault() fix it.
585 */
586 break;
587 default:
588 break;
589 }
590 return 0;
591}
592NOKPROBE_SYMBOL(kprobe_trap_handler);
593
594int kprobe_fault_handler(struct pt_regs *regs, int trapnr)
595{
596 int ret;
597
598 if (regs->psw.mask & (PSW_MASK_IO | PSW_MASK_EXT))
599 local_irq_disable();
600 ret = kprobe_trap_handler(regs, trapnr);
601 if (regs->psw.mask & (PSW_MASK_IO | PSW_MASK_EXT))
602 local_irq_restore(regs->psw.mask & ~PSW_MASK_PER);
603 return ret;
604}
605NOKPROBE_SYMBOL(kprobe_fault_handler);
606
607/*
608 * Wrapper routine to for handling exceptions.
609 */
610int kprobe_exceptions_notify(struct notifier_block *self,
611 unsigned long val, void *data)
612{
613 struct die_args *args = (struct die_args *) data;
614 struct pt_regs *regs = args->regs;
615 int ret = NOTIFY_DONE;
616
617 if (regs->psw.mask & (PSW_MASK_IO | PSW_MASK_EXT))
618 local_irq_disable();
619
620 switch (val) {
621 case DIE_BPT:
622 if (kprobe_handler(regs))
623 ret = NOTIFY_STOP;
624 break;
625 case DIE_SSTEP:
626 if (post_kprobe_handler(regs))
627 ret = NOTIFY_STOP;
628 break;
629 case DIE_TRAP:
630 if (!preemptible() && kprobe_running() &&
631 kprobe_trap_handler(regs, args->trapnr))
632 ret = NOTIFY_STOP;
633 break;
634 default:
635 break;
636 }
637
638 if (regs->psw.mask & (PSW_MASK_IO | PSW_MASK_EXT))
639 local_irq_restore(regs->psw.mask & ~PSW_MASK_PER);
640
641 return ret;
642}
643NOKPROBE_SYMBOL(kprobe_exceptions_notify);
644
645static struct kprobe trampoline = {
646 .addr = (kprobe_opcode_t *) &kretprobe_trampoline,
647 .pre_handler = trampoline_probe_handler
648};
649
650int __init arch_init_kprobes(void)
651{
652 return register_kprobe(&trampoline);
653}
654
655int arch_trampoline_kprobe(struct kprobe *p)
656{
657 return p->addr == (kprobe_opcode_t *) &kretprobe_trampoline;
658}
659NOKPROBE_SYMBOL(arch_trampoline_kprobe);
1/*
2 * Kernel Probes (KProbes)
3 *
4 * This program is free software; you can redistribute it and/or modify
5 * it under the terms of the GNU General Public License as published by
6 * the Free Software Foundation; either version 2 of the License, or
7 * (at your option) any later version.
8 *
9 * This program is distributed in the hope that it will be useful,
10 * but WITHOUT ANY WARRANTY; without even the implied warranty of
11 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
12 * GNU General Public License for more details.
13 *
14 * You should have received a copy of the GNU General Public License
15 * along with this program; if not, write to the Free Software
16 * Foundation, Inc., 59 Temple Place - Suite 330, Boston, MA 02111-1307, USA.
17 *
18 * Copyright (C) IBM Corporation, 2002, 2006
19 *
20 * s390 port, used ppc64 as template. Mike Grundy <grundym@us.ibm.com>
21 */
22
23#include <linux/kprobes.h>
24#include <linux/ptrace.h>
25#include <linux/preempt.h>
26#include <linux/stop_machine.h>
27#include <linux/kdebug.h>
28#include <linux/uaccess.h>
29#include <asm/cacheflush.h>
30#include <asm/sections.h>
31#include <linux/module.h>
32#include <linux/slab.h>
33#include <linux/hardirq.h>
34
35DEFINE_PER_CPU(struct kprobe *, current_kprobe);
36DEFINE_PER_CPU(struct kprobe_ctlblk, kprobe_ctlblk);
37
38struct kretprobe_blackpoint kretprobe_blacklist[] = { };
39
40static int __kprobes is_prohibited_opcode(kprobe_opcode_t *insn)
41{
42 switch (insn[0] >> 8) {
43 case 0x0c: /* bassm */
44 case 0x0b: /* bsm */
45 case 0x83: /* diag */
46 case 0x44: /* ex */
47 case 0xac: /* stnsm */
48 case 0xad: /* stosm */
49 return -EINVAL;
50 }
51 switch (insn[0]) {
52 case 0x0101: /* pr */
53 case 0xb25a: /* bsa */
54 case 0xb240: /* bakr */
55 case 0xb258: /* bsg */
56 case 0xb218: /* pc */
57 case 0xb228: /* pt */
58 case 0xb98d: /* epsw */
59 return -EINVAL;
60 }
61 return 0;
62}
63
64static int __kprobes get_fixup_type(kprobe_opcode_t *insn)
65{
66 /* default fixup method */
67 int fixup = FIXUP_PSW_NORMAL;
68
69 switch (insn[0] >> 8) {
70 case 0x05: /* balr */
71 case 0x0d: /* basr */
72 fixup = FIXUP_RETURN_REGISTER;
73 /* if r2 = 0, no branch will be taken */
74 if ((insn[0] & 0x0f) == 0)
75 fixup |= FIXUP_BRANCH_NOT_TAKEN;
76 break;
77 case 0x06: /* bctr */
78 case 0x07: /* bcr */
79 fixup = FIXUP_BRANCH_NOT_TAKEN;
80 break;
81 case 0x45: /* bal */
82 case 0x4d: /* bas */
83 fixup = FIXUP_RETURN_REGISTER;
84 break;
85 case 0x47: /* bc */
86 case 0x46: /* bct */
87 case 0x86: /* bxh */
88 case 0x87: /* bxle */
89 fixup = FIXUP_BRANCH_NOT_TAKEN;
90 break;
91 case 0x82: /* lpsw */
92 fixup = FIXUP_NOT_REQUIRED;
93 break;
94 case 0xb2: /* lpswe */
95 if ((insn[0] & 0xff) == 0xb2)
96 fixup = FIXUP_NOT_REQUIRED;
97 break;
98 case 0xa7: /* bras */
99 if ((insn[0] & 0x0f) == 0x05)
100 fixup |= FIXUP_RETURN_REGISTER;
101 break;
102 case 0xc0:
103 if ((insn[0] & 0x0f) == 0x00 || /* larl */
104 (insn[0] & 0x0f) == 0x05) /* brasl */
105 fixup |= FIXUP_RETURN_REGISTER;
106 break;
107 case 0xeb:
108 if ((insn[2] & 0xff) == 0x44 || /* bxhg */
109 (insn[2] & 0xff) == 0x45) /* bxleg */
110 fixup = FIXUP_BRANCH_NOT_TAKEN;
111 break;
112 case 0xe3: /* bctg */
113 if ((insn[2] & 0xff) == 0x46)
114 fixup = FIXUP_BRANCH_NOT_TAKEN;
115 break;
116 }
117 return fixup;
118}
119
120int __kprobes arch_prepare_kprobe(struct kprobe *p)
121{
122 if ((unsigned long) p->addr & 0x01)
123 return -EINVAL;
124
125 /* Make sure the probe isn't going on a difficult instruction */
126 if (is_prohibited_opcode(p->addr))
127 return -EINVAL;
128
129 p->opcode = *p->addr;
130 memcpy(p->ainsn.insn, p->addr, ((p->opcode >> 14) + 3) & -2);
131
132 return 0;
133}
134
135struct ins_replace_args {
136 kprobe_opcode_t *ptr;
137 kprobe_opcode_t opcode;
138};
139
140static int __kprobes swap_instruction(void *aref)
141{
142 struct kprobe_ctlblk *kcb = get_kprobe_ctlblk();
143 unsigned long status = kcb->kprobe_status;
144 struct ins_replace_args *args = aref;
145
146 kcb->kprobe_status = KPROBE_SWAP_INST;
147 probe_kernel_write(args->ptr, &args->opcode, sizeof(args->opcode));
148 kcb->kprobe_status = status;
149 return 0;
150}
151
152void __kprobes arch_arm_kprobe(struct kprobe *p)
153{
154 struct ins_replace_args args;
155
156 args.ptr = p->addr;
157 args.opcode = BREAKPOINT_INSTRUCTION;
158 stop_machine(swap_instruction, &args, NULL);
159}
160
161void __kprobes arch_disarm_kprobe(struct kprobe *p)
162{
163 struct ins_replace_args args;
164
165 args.ptr = p->addr;
166 args.opcode = p->opcode;
167 stop_machine(swap_instruction, &args, NULL);
168}
169
170void __kprobes arch_remove_kprobe(struct kprobe *p)
171{
172}
173
174static void __kprobes enable_singlestep(struct kprobe_ctlblk *kcb,
175 struct pt_regs *regs,
176 unsigned long ip)
177{
178 struct per_regs per_kprobe;
179
180 /* Set up the PER control registers %cr9-%cr11 */
181 per_kprobe.control = PER_EVENT_IFETCH;
182 per_kprobe.start = ip;
183 per_kprobe.end = ip;
184
185 /* Save control regs and psw mask */
186 __ctl_store(kcb->kprobe_saved_ctl, 9, 11);
187 kcb->kprobe_saved_imask = regs->psw.mask &
188 (PSW_MASK_PER | PSW_MASK_IO | PSW_MASK_EXT);
189
190 /* Set PER control regs, turns on single step for the given address */
191 __ctl_load(per_kprobe, 9, 11);
192 regs->psw.mask |= PSW_MASK_PER;
193 regs->psw.mask &= ~(PSW_MASK_IO | PSW_MASK_EXT);
194 regs->psw.addr = ip | PSW_ADDR_AMODE;
195}
196
197static void __kprobes disable_singlestep(struct kprobe_ctlblk *kcb,
198 struct pt_regs *regs,
199 unsigned long ip)
200{
201 /* Restore control regs and psw mask, set new psw address */
202 __ctl_load(kcb->kprobe_saved_ctl, 9, 11);
203 regs->psw.mask &= ~PSW_MASK_PER;
204 regs->psw.mask |= kcb->kprobe_saved_imask;
205 regs->psw.addr = ip | PSW_ADDR_AMODE;
206}
207
208/*
209 * Activate a kprobe by storing its pointer to current_kprobe. The
210 * previous kprobe is stored in kcb->prev_kprobe. A stack of up to
211 * two kprobes can be active, see KPROBE_REENTER.
212 */
213static void __kprobes push_kprobe(struct kprobe_ctlblk *kcb, struct kprobe *p)
214{
215 kcb->prev_kprobe.kp = __get_cpu_var(current_kprobe);
216 kcb->prev_kprobe.status = kcb->kprobe_status;
217 __get_cpu_var(current_kprobe) = p;
218}
219
220/*
221 * Deactivate a kprobe by backing up to the previous state. If the
222 * current state is KPROBE_REENTER prev_kprobe.kp will be non-NULL,
223 * for any other state prev_kprobe.kp will be NULL.
224 */
225static void __kprobes pop_kprobe(struct kprobe_ctlblk *kcb)
226{
227 __get_cpu_var(current_kprobe) = kcb->prev_kprobe.kp;
228 kcb->kprobe_status = kcb->prev_kprobe.status;
229}
230
231void __kprobes arch_prepare_kretprobe(struct kretprobe_instance *ri,
232 struct pt_regs *regs)
233{
234 ri->ret_addr = (kprobe_opcode_t *) regs->gprs[14];
235
236 /* Replace the return addr with trampoline addr */
237 regs->gprs[14] = (unsigned long) &kretprobe_trampoline;
238}
239
240static void __kprobes kprobe_reenter_check(struct kprobe_ctlblk *kcb,
241 struct kprobe *p)
242{
243 switch (kcb->kprobe_status) {
244 case KPROBE_HIT_SSDONE:
245 case KPROBE_HIT_ACTIVE:
246 kprobes_inc_nmissed_count(p);
247 break;
248 case KPROBE_HIT_SS:
249 case KPROBE_REENTER:
250 default:
251 /*
252 * A kprobe on the code path to single step an instruction
253 * is a BUG. The code path resides in the .kprobes.text
254 * section and is executed with interrupts disabled.
255 */
256 printk(KERN_EMERG "Invalid kprobe detected at %p.\n", p->addr);
257 dump_kprobe(p);
258 BUG();
259 }
260}
261
262static int __kprobes kprobe_handler(struct pt_regs *regs)
263{
264 struct kprobe_ctlblk *kcb;
265 struct kprobe *p;
266
267 /*
268 * We want to disable preemption for the entire duration of kprobe
269 * processing. That includes the calls to the pre/post handlers
270 * and single stepping the kprobe instruction.
271 */
272 preempt_disable();
273 kcb = get_kprobe_ctlblk();
274 p = get_kprobe((void *)((regs->psw.addr & PSW_ADDR_INSN) - 2));
275
276 if (p) {
277 if (kprobe_running()) {
278 /*
279 * We have hit a kprobe while another is still
280 * active. This can happen in the pre and post
281 * handler. Single step the instruction of the
282 * new probe but do not call any handler function
283 * of this secondary kprobe.
284 * push_kprobe and pop_kprobe saves and restores
285 * the currently active kprobe.
286 */
287 kprobe_reenter_check(kcb, p);
288 push_kprobe(kcb, p);
289 kcb->kprobe_status = KPROBE_REENTER;
290 } else {
291 /*
292 * If we have no pre-handler or it returned 0, we
293 * continue with single stepping. If we have a
294 * pre-handler and it returned non-zero, it prepped
295 * for calling the break_handler below on re-entry
296 * for jprobe processing, so get out doing nothing
297 * more here.
298 */
299 push_kprobe(kcb, p);
300 kcb->kprobe_status = KPROBE_HIT_ACTIVE;
301 if (p->pre_handler && p->pre_handler(p, regs))
302 return 1;
303 kcb->kprobe_status = KPROBE_HIT_SS;
304 }
305 enable_singlestep(kcb, regs, (unsigned long) p->ainsn.insn);
306 return 1;
307 } else if (kprobe_running()) {
308 p = __get_cpu_var(current_kprobe);
309 if (p->break_handler && p->break_handler(p, regs)) {
310 /*
311 * Continuation after the jprobe completed and
312 * caused the jprobe_return trap. The jprobe
313 * break_handler "returns" to the original
314 * function that still has the kprobe breakpoint
315 * installed. We continue with single stepping.
316 */
317 kcb->kprobe_status = KPROBE_HIT_SS;
318 enable_singlestep(kcb, regs,
319 (unsigned long) p->ainsn.insn);
320 return 1;
321 } /* else:
322 * No kprobe at this address and the current kprobe
323 * has no break handler (no jprobe!). The kernel just
324 * exploded, let the standard trap handler pick up the
325 * pieces.
326 */
327 } /* else:
328 * No kprobe at this address and no active kprobe. The trap has
329 * not been caused by a kprobe breakpoint. The race of breakpoint
330 * vs. kprobe remove does not exist because on s390 as we use
331 * stop_machine to arm/disarm the breakpoints.
332 */
333 preempt_enable_no_resched();
334 return 0;
335}
336
337/*
338 * Function return probe trampoline:
339 * - init_kprobes() establishes a probepoint here
340 * - When the probed function returns, this probe
341 * causes the handlers to fire
342 */
343static void __used kretprobe_trampoline_holder(void)
344{
345 asm volatile(".global kretprobe_trampoline\n"
346 "kretprobe_trampoline: bcr 0,0\n");
347}
348
349/*
350 * Called when the probe at kretprobe trampoline is hit
351 */
352static int __kprobes trampoline_probe_handler(struct kprobe *p,
353 struct pt_regs *regs)
354{
355 struct kretprobe_instance *ri;
356 struct hlist_head *head, empty_rp;
357 struct hlist_node *node, *tmp;
358 unsigned long flags, orig_ret_address;
359 unsigned long trampoline_address;
360 kprobe_opcode_t *correct_ret_addr;
361
362 INIT_HLIST_HEAD(&empty_rp);
363 kretprobe_hash_lock(current, &head, &flags);
364
365 /*
366 * It is possible to have multiple instances associated with a given
367 * task either because an multiple functions in the call path
368 * have a return probe installed on them, and/or more than one return
369 * return probe was registered for a target function.
370 *
371 * We can handle this because:
372 * - instances are always inserted at the head of the list
373 * - when multiple return probes are registered for the same
374 * function, the first instance's ret_addr will point to the
375 * real return address, and all the rest will point to
376 * kretprobe_trampoline
377 */
378 ri = NULL;
379 orig_ret_address = 0;
380 correct_ret_addr = NULL;
381 trampoline_address = (unsigned long) &kretprobe_trampoline;
382 hlist_for_each_entry_safe(ri, node, tmp, head, hlist) {
383 if (ri->task != current)
384 /* another task is sharing our hash bucket */
385 continue;
386
387 orig_ret_address = (unsigned long) ri->ret_addr;
388
389 if (orig_ret_address != trampoline_address)
390 /*
391 * This is the real return address. Any other
392 * instances associated with this task are for
393 * other calls deeper on the call stack
394 */
395 break;
396 }
397
398 kretprobe_assert(ri, orig_ret_address, trampoline_address);
399
400 correct_ret_addr = ri->ret_addr;
401 hlist_for_each_entry_safe(ri, node, tmp, head, hlist) {
402 if (ri->task != current)
403 /* another task is sharing our hash bucket */
404 continue;
405
406 orig_ret_address = (unsigned long) ri->ret_addr;
407
408 if (ri->rp && ri->rp->handler) {
409 ri->ret_addr = correct_ret_addr;
410 ri->rp->handler(ri, regs);
411 }
412
413 recycle_rp_inst(ri, &empty_rp);
414
415 if (orig_ret_address != trampoline_address)
416 /*
417 * This is the real return address. Any other
418 * instances associated with this task are for
419 * other calls deeper on the call stack
420 */
421 break;
422 }
423
424 regs->psw.addr = orig_ret_address | PSW_ADDR_AMODE;
425
426 pop_kprobe(get_kprobe_ctlblk());
427 kretprobe_hash_unlock(current, &flags);
428 preempt_enable_no_resched();
429
430 hlist_for_each_entry_safe(ri, node, tmp, &empty_rp, hlist) {
431 hlist_del(&ri->hlist);
432 kfree(ri);
433 }
434 /*
435 * By returning a non-zero value, we are telling
436 * kprobe_handler() that we don't want the post_handler
437 * to run (and have re-enabled preemption)
438 */
439 return 1;
440}
441
442/*
443 * Called after single-stepping. p->addr is the address of the
444 * instruction whose first byte has been replaced by the "breakpoint"
445 * instruction. To avoid the SMP problems that can occur when we
446 * temporarily put back the original opcode to single-step, we
447 * single-stepped a copy of the instruction. The address of this
448 * copy is p->ainsn.insn.
449 */
450static void __kprobes resume_execution(struct kprobe *p, struct pt_regs *regs)
451{
452 struct kprobe_ctlblk *kcb = get_kprobe_ctlblk();
453 unsigned long ip = regs->psw.addr & PSW_ADDR_INSN;
454 int fixup = get_fixup_type(p->ainsn.insn);
455
456 if (fixup & FIXUP_PSW_NORMAL)
457 ip += (unsigned long) p->addr - (unsigned long) p->ainsn.insn;
458
459 if (fixup & FIXUP_BRANCH_NOT_TAKEN) {
460 int ilen = ((p->ainsn.insn[0] >> 14) + 3) & -2;
461 if (ip - (unsigned long) p->ainsn.insn == ilen)
462 ip = (unsigned long) p->addr + ilen;
463 }
464
465 if (fixup & FIXUP_RETURN_REGISTER) {
466 int reg = (p->ainsn.insn[0] & 0xf0) >> 4;
467 regs->gprs[reg] += (unsigned long) p->addr -
468 (unsigned long) p->ainsn.insn;
469 }
470
471 disable_singlestep(kcb, regs, ip);
472}
473
474static int __kprobes post_kprobe_handler(struct pt_regs *regs)
475{
476 struct kprobe_ctlblk *kcb = get_kprobe_ctlblk();
477 struct kprobe *p = kprobe_running();
478
479 if (!p)
480 return 0;
481
482 if (kcb->kprobe_status != KPROBE_REENTER && p->post_handler) {
483 kcb->kprobe_status = KPROBE_HIT_SSDONE;
484 p->post_handler(p, regs, 0);
485 }
486
487 resume_execution(p, regs);
488 pop_kprobe(kcb);
489 preempt_enable_no_resched();
490
491 /*
492 * if somebody else is singlestepping across a probe point, psw mask
493 * will have PER set, in which case, continue the remaining processing
494 * of do_single_step, as if this is not a probe hit.
495 */
496 if (regs->psw.mask & PSW_MASK_PER)
497 return 0;
498
499 return 1;
500}
501
502static int __kprobes kprobe_trap_handler(struct pt_regs *regs, int trapnr)
503{
504 struct kprobe_ctlblk *kcb = get_kprobe_ctlblk();
505 struct kprobe *p = kprobe_running();
506 const struct exception_table_entry *entry;
507
508 switch(kcb->kprobe_status) {
509 case KPROBE_SWAP_INST:
510 /* We are here because the instruction replacement failed */
511 return 0;
512 case KPROBE_HIT_SS:
513 case KPROBE_REENTER:
514 /*
515 * We are here because the instruction being single
516 * stepped caused a page fault. We reset the current
517 * kprobe and the nip points back to the probe address
518 * and allow the page fault handler to continue as a
519 * normal page fault.
520 */
521 disable_singlestep(kcb, regs, (unsigned long) p->addr);
522 pop_kprobe(kcb);
523 preempt_enable_no_resched();
524 break;
525 case KPROBE_HIT_ACTIVE:
526 case KPROBE_HIT_SSDONE:
527 /*
528 * We increment the nmissed count for accounting,
529 * we can also use npre/npostfault count for accouting
530 * these specific fault cases.
531 */
532 kprobes_inc_nmissed_count(p);
533
534 /*
535 * We come here because instructions in the pre/post
536 * handler caused the page_fault, this could happen
537 * if handler tries to access user space by
538 * copy_from_user(), get_user() etc. Let the
539 * user-specified handler try to fix it first.
540 */
541 if (p->fault_handler && p->fault_handler(p, regs, trapnr))
542 return 1;
543
544 /*
545 * In case the user-specified fault handler returned
546 * zero, try to fix up.
547 */
548 entry = search_exception_tables(regs->psw.addr & PSW_ADDR_INSN);
549 if (entry) {
550 regs->psw.addr = entry->fixup | PSW_ADDR_AMODE;
551 return 1;
552 }
553
554 /*
555 * fixup_exception() could not handle it,
556 * Let do_page_fault() fix it.
557 */
558 break;
559 default:
560 break;
561 }
562 return 0;
563}
564
565int __kprobes kprobe_fault_handler(struct pt_regs *regs, int trapnr)
566{
567 int ret;
568
569 if (regs->psw.mask & (PSW_MASK_IO | PSW_MASK_EXT))
570 local_irq_disable();
571 ret = kprobe_trap_handler(regs, trapnr);
572 if (regs->psw.mask & (PSW_MASK_IO | PSW_MASK_EXT))
573 local_irq_restore(regs->psw.mask & ~PSW_MASK_PER);
574 return ret;
575}
576
577/*
578 * Wrapper routine to for handling exceptions.
579 */
580int __kprobes kprobe_exceptions_notify(struct notifier_block *self,
581 unsigned long val, void *data)
582{
583 struct die_args *args = (struct die_args *) data;
584 struct pt_regs *regs = args->regs;
585 int ret = NOTIFY_DONE;
586
587 if (regs->psw.mask & (PSW_MASK_IO | PSW_MASK_EXT))
588 local_irq_disable();
589
590 switch (val) {
591 case DIE_BPT:
592 if (kprobe_handler(regs))
593 ret = NOTIFY_STOP;
594 break;
595 case DIE_SSTEP:
596 if (post_kprobe_handler(regs))
597 ret = NOTIFY_STOP;
598 break;
599 case DIE_TRAP:
600 if (!preemptible() && kprobe_running() &&
601 kprobe_trap_handler(regs, args->trapnr))
602 ret = NOTIFY_STOP;
603 break;
604 default:
605 break;
606 }
607
608 if (regs->psw.mask & (PSW_MASK_IO | PSW_MASK_EXT))
609 local_irq_restore(regs->psw.mask & ~PSW_MASK_PER);
610
611 return ret;
612}
613
614int __kprobes setjmp_pre_handler(struct kprobe *p, struct pt_regs *regs)
615{
616 struct jprobe *jp = container_of(p, struct jprobe, kp);
617 struct kprobe_ctlblk *kcb = get_kprobe_ctlblk();
618 unsigned long stack;
619
620 memcpy(&kcb->jprobe_saved_regs, regs, sizeof(struct pt_regs));
621
622 /* setup return addr to the jprobe handler routine */
623 regs->psw.addr = (unsigned long) jp->entry | PSW_ADDR_AMODE;
624 regs->psw.mask &= ~(PSW_MASK_IO | PSW_MASK_EXT);
625
626 /* r15 is the stack pointer */
627 stack = (unsigned long) regs->gprs[15];
628
629 memcpy(kcb->jprobes_stack, (void *) stack, MIN_STACK_SIZE(stack));
630 return 1;
631}
632
633void __kprobes jprobe_return(void)
634{
635 asm volatile(".word 0x0002");
636}
637
638static void __used __kprobes jprobe_return_end(void)
639{
640 asm volatile("bcr 0,0");
641}
642
643int __kprobes longjmp_break_handler(struct kprobe *p, struct pt_regs *regs)
644{
645 struct kprobe_ctlblk *kcb = get_kprobe_ctlblk();
646 unsigned long stack;
647
648 stack = (unsigned long) kcb->jprobe_saved_regs.gprs[15];
649
650 /* Put the regs back */
651 memcpy(regs, &kcb->jprobe_saved_regs, sizeof(struct pt_regs));
652 /* put the stack back */
653 memcpy((void *) stack, kcb->jprobes_stack, MIN_STACK_SIZE(stack));
654 preempt_enable_no_resched();
655 return 1;
656}
657
658static struct kprobe trampoline = {
659 .addr = (kprobe_opcode_t *) &kretprobe_trampoline,
660 .pre_handler = trampoline_probe_handler
661};
662
663int __init arch_init_kprobes(void)
664{
665 return register_kprobe(&trampoline);
666}
667
668int __kprobes arch_trampoline_kprobe(struct kprobe *p)
669{
670 return p->addr == (kprobe_opcode_t *) &kretprobe_trampoline;
671}