Linux Audio

Check our new training course

Loading...
v4.17
 
  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, 2004
 19 *
 20 * 2002-Oct	Created by Vamsi Krishna S <vamsi_krishna@in.ibm.com> Kernel
 21 *		Probes initial implementation ( includes contributions from
 22 *		Rusty Russell).
 23 * 2004-July	Suparna Bhattacharya <suparna@in.ibm.com> added jumper probes
 24 *		interface to access function arguments.
 25 * 2004-Nov	Ananth N Mavinakayanahalli <ananth@in.ibm.com> kprobes port
 26 *		for PPC64
 27 */
 28
 29#include <linux/kprobes.h>
 30#include <linux/ptrace.h>
 31#include <linux/preempt.h>
 32#include <linux/extable.h>
 33#include <linux/kdebug.h>
 34#include <linux/slab.h>
 35#include <asm/code-patching.h>
 36#include <asm/cacheflush.h>
 37#include <asm/sstep.h>
 38#include <asm/sections.h>
 39#include <linux/uaccess.h>
 40
 41DEFINE_PER_CPU(struct kprobe *, current_kprobe) = NULL;
 42DEFINE_PER_CPU(struct kprobe_ctlblk, kprobe_ctlblk);
 43
 44struct kretprobe_blackpoint kretprobe_blacklist[] = {{NULL, NULL}};
 45
 46bool arch_within_kprobe_blacklist(unsigned long addr)
 47{
 48	return  (addr >= (unsigned long)__kprobes_text_start &&
 49		 addr < (unsigned long)__kprobes_text_end) ||
 50		(addr >= (unsigned long)_stext &&
 51		 addr < (unsigned long)__head_end);
 52}
 53
 54kprobe_opcode_t *kprobe_lookup_name(const char *name, unsigned int offset)
 55{
 56	kprobe_opcode_t *addr = NULL;
 57
 58#ifdef PPC64_ELF_ABI_v2
 59	/* PPC64 ABIv2 needs local entry point */
 60	addr = (kprobe_opcode_t *)kallsyms_lookup_name(name);
 61	if (addr && !offset) {
 62#ifdef CONFIG_KPROBES_ON_FTRACE
 63		unsigned long faddr;
 64		/*
 65		 * Per livepatch.h, ftrace location is always within the first
 66		 * 16 bytes of a function on powerpc with -mprofile-kernel.
 67		 */
 68		faddr = ftrace_location_range((unsigned long)addr,
 69					      (unsigned long)addr + 16);
 70		if (faddr)
 71			addr = (kprobe_opcode_t *)faddr;
 72		else
 73#endif
 74			addr = (kprobe_opcode_t *)ppc_function_entry(addr);
 75	}
 76#elif defined(PPC64_ELF_ABI_v1)
 77	/*
 78	 * 64bit powerpc ABIv1 uses function descriptors:
 79	 * - Check for the dot variant of the symbol first.
 80	 * - If that fails, try looking up the symbol provided.
 81	 *
 82	 * This ensures we always get to the actual symbol and not
 83	 * the descriptor.
 84	 *
 85	 * Also handle <module:symbol> format.
 86	 */
 87	char dot_name[MODULE_NAME_LEN + 1 + KSYM_NAME_LEN];
 88	bool dot_appended = false;
 89	const char *c;
 90	ssize_t ret = 0;
 91	int len = 0;
 92
 93	if ((c = strnchr(name, MODULE_NAME_LEN, ':')) != NULL) {
 94		c++;
 95		len = c - name;
 96		memcpy(dot_name, name, len);
 97	} else
 98		c = name;
 99
100	if (*c != '\0' && *c != '.') {
101		dot_name[len++] = '.';
102		dot_appended = true;
103	}
104	ret = strscpy(dot_name + len, c, KSYM_NAME_LEN);
105	if (ret > 0)
106		addr = (kprobe_opcode_t *)kallsyms_lookup_name(dot_name);
107
108	/* Fallback to the original non-dot symbol lookup */
109	if (!addr && dot_appended)
110		addr = (kprobe_opcode_t *)kallsyms_lookup_name(name);
111#else
112	addr = (kprobe_opcode_t *)kallsyms_lookup_name(name);
113#endif
114
115	return addr;
116}
117
118int arch_prepare_kprobe(struct kprobe *p)
119{
120	int ret = 0;
121	kprobe_opcode_t insn = *p->addr;
122
123	if ((unsigned long)p->addr & 0x03) {
124		printk("Attempt to register kprobe at an unaligned address\n");
125		ret = -EINVAL;
126	} else if (IS_MTMSRD(insn) || IS_RFID(insn) || IS_RFI(insn)) {
127		printk("Cannot register a kprobe on rfi/rfid or mtmsr[d]\n");
128		ret = -EINVAL;
129	}
130
131	/* insn must be on a special executable page on ppc64.  This is
132	 * not explicitly required on ppc32 (right now), but it doesn't hurt */
133	if (!ret) {
134		p->ainsn.insn = get_insn_slot();
135		if (!p->ainsn.insn)
136			ret = -ENOMEM;
137	}
138
139	if (!ret) {
140		memcpy(p->ainsn.insn, p->addr,
141				MAX_INSN_SIZE * sizeof(kprobe_opcode_t));
142		p->opcode = *p->addr;
143		flush_icache_range((unsigned long)p->ainsn.insn,
144			(unsigned long)p->ainsn.insn + sizeof(kprobe_opcode_t));
145	}
146
147	p->ainsn.boostable = 0;
148	return ret;
149}
150NOKPROBE_SYMBOL(arch_prepare_kprobe);
151
152void arch_arm_kprobe(struct kprobe *p)
153{
154	patch_instruction(p->addr, BREAKPOINT_INSTRUCTION);
155}
156NOKPROBE_SYMBOL(arch_arm_kprobe);
157
158void arch_disarm_kprobe(struct kprobe *p)
159{
160	patch_instruction(p->addr, p->opcode);
161}
162NOKPROBE_SYMBOL(arch_disarm_kprobe);
163
164void arch_remove_kprobe(struct kprobe *p)
165{
166	if (p->ainsn.insn) {
167		free_insn_slot(p->ainsn.insn, 0);
168		p->ainsn.insn = NULL;
169	}
170}
171NOKPROBE_SYMBOL(arch_remove_kprobe);
172
173static nokprobe_inline void prepare_singlestep(struct kprobe *p, struct pt_regs *regs)
174{
175	enable_single_step(regs);
176
177	/*
178	 * On powerpc we should single step on the original
179	 * instruction even if the probed insn is a trap
180	 * variant as values in regs could play a part in
181	 * if the trap is taken or not
182	 */
183	regs->nip = (unsigned long)p->ainsn.insn;
184}
185
186static nokprobe_inline void save_previous_kprobe(struct kprobe_ctlblk *kcb)
187{
188	kcb->prev_kprobe.kp = kprobe_running();
189	kcb->prev_kprobe.status = kcb->kprobe_status;
190	kcb->prev_kprobe.saved_msr = kcb->kprobe_saved_msr;
191}
192
193static nokprobe_inline void restore_previous_kprobe(struct kprobe_ctlblk *kcb)
194{
195	__this_cpu_write(current_kprobe, kcb->prev_kprobe.kp);
196	kcb->kprobe_status = kcb->prev_kprobe.status;
197	kcb->kprobe_saved_msr = kcb->prev_kprobe.saved_msr;
198}
199
200static nokprobe_inline void set_current_kprobe(struct kprobe *p, struct pt_regs *regs,
201				struct kprobe_ctlblk *kcb)
202{
203	__this_cpu_write(current_kprobe, p);
204	kcb->kprobe_saved_msr = regs->msr;
205}
206
207bool arch_kprobe_on_func_entry(unsigned long offset)
208{
209#ifdef PPC64_ELF_ABI_v2
210#ifdef CONFIG_KPROBES_ON_FTRACE
211	return offset <= 16;
212#else
213	return offset <= 8;
214#endif
215#else
216	return !offset;
217#endif
218}
219
220void arch_prepare_kretprobe(struct kretprobe_instance *ri, struct pt_regs *regs)
221{
222	ri->ret_addr = (kprobe_opcode_t *)regs->link;
223
224	/* Replace the return addr with trampoline addr */
225	regs->link = (unsigned long)kretprobe_trampoline;
226}
227NOKPROBE_SYMBOL(arch_prepare_kretprobe);
228
229static int try_to_emulate(struct kprobe *p, struct pt_regs *regs)
230{
231	int ret;
232	unsigned int insn = *p->ainsn.insn;
233
234	/* regs->nip is also adjusted if emulate_step returns 1 */
235	ret = emulate_step(regs, insn);
236	if (ret > 0) {
237		/*
238		 * Once this instruction has been boosted
239		 * successfully, set the boostable flag
240		 */
241		if (unlikely(p->ainsn.boostable == 0))
242			p->ainsn.boostable = 1;
243	} else if (ret < 0) {
244		/*
245		 * We don't allow kprobes on mtmsr(d)/rfi(d), etc.
246		 * So, we should never get here... but, its still
247		 * good to catch them, just in case...
248		 */
249		printk("Can't step on instruction %x\n", insn);
250		BUG();
251	} else {
252		/*
253		 * If we haven't previously emulated this instruction, then it
254		 * can't be boosted. Note it down so we don't try to do so again.
255		 *
256		 * If, however, we had emulated this instruction in the past,
257		 * then this is just an error with the current run (for
258		 * instance, exceptions due to a load/store). We return 0 so
259		 * that this is now single-stepped, but continue to try
260		 * emulating it in subsequent probe hits.
261		 */
262		if (unlikely(p->ainsn.boostable != 1))
263			p->ainsn.boostable = -1;
264	}
265
266	return ret;
267}
268NOKPROBE_SYMBOL(try_to_emulate);
269
270int kprobe_handler(struct pt_regs *regs)
271{
272	struct kprobe *p;
273	int ret = 0;
274	unsigned int *addr = (unsigned int *)regs->nip;
275	struct kprobe_ctlblk *kcb;
276
277	if (user_mode(regs))
278		return 0;
279
280	/*
281	 * We don't want to be preempted for the entire
282	 * duration of kprobe processing
283	 */
284	preempt_disable();
285	kcb = get_kprobe_ctlblk();
286
287	/* Check we're not actually recursing */
288	if (kprobe_running()) {
289		p = get_kprobe(addr);
290		if (p) {
291			kprobe_opcode_t insn = *p->ainsn.insn;
292			if (kcb->kprobe_status == KPROBE_HIT_SS &&
293					is_trap(insn)) {
294				/* Turn off 'trace' bits */
295				regs->msr &= ~MSR_SINGLESTEP;
296				regs->msr |= kcb->kprobe_saved_msr;
297				goto no_kprobe;
298			}
299			/* We have reentered the kprobe_handler(), since
300			 * another probe was hit while within the handler.
301			 * We here save the original kprobes variables and
302			 * just single step on the instruction of the new probe
303			 * without calling any user handlers.
304			 */
305			save_previous_kprobe(kcb);
306			set_current_kprobe(p, regs, kcb);
307			kprobes_inc_nmissed_count(p);
308			kcb->kprobe_status = KPROBE_REENTER;
309			if (p->ainsn.boostable >= 0) {
310				ret = try_to_emulate(p, regs);
311
312				if (ret > 0) {
313					restore_previous_kprobe(kcb);
314					preempt_enable_no_resched();
315					return 1;
316				}
317			}
318			prepare_singlestep(p, regs);
319			return 1;
320		} else {
321			if (*addr != BREAKPOINT_INSTRUCTION) {
322				/* If trap variant, then it belongs not to us */
323				kprobe_opcode_t cur_insn = *addr;
324				if (is_trap(cur_insn))
325		       			goto no_kprobe;
326				/* The breakpoint instruction was removed by
327				 * another cpu right after we hit, no further
328				 * handling of this interrupt is appropriate
329				 */
330				ret = 1;
331				goto no_kprobe;
332			}
333			p = __this_cpu_read(current_kprobe);
334			if (p->break_handler && p->break_handler(p, regs)) {
335				if (!skip_singlestep(p, regs, kcb))
336					goto ss_probe;
337				ret = 1;
338			}
339		}
340		goto no_kprobe;
341	}
342
343	p = get_kprobe(addr);
344	if (!p) {
345		if (*addr != BREAKPOINT_INSTRUCTION) {
346			/*
347			 * PowerPC has multiple variants of the "trap"
348			 * instruction. If the current instruction is a
349			 * trap variant, it could belong to someone else
350			 */
351			kprobe_opcode_t cur_insn = *addr;
352			if (is_trap(cur_insn))
353		       		goto no_kprobe;
354			/*
355			 * The breakpoint instruction was removed right
356			 * after we hit it.  Another cpu has removed
357			 * either a probepoint or a debugger breakpoint
358			 * at this address.  In either case, no further
359			 * handling of this interrupt is appropriate.
360			 */
361			ret = 1;
362		}
363		/* Not one of ours: let kernel handle it */
364		goto no_kprobe;
365	}
366
367	kcb->kprobe_status = KPROBE_HIT_ACTIVE;
368	set_current_kprobe(p, regs, kcb);
369	if (p->pre_handler && p->pre_handler(p, regs))
370		/* handler has already set things up, so skip ss setup */
 
 
371		return 1;
 
372
373ss_probe:
374	if (p->ainsn.boostable >= 0) {
375		ret = try_to_emulate(p, regs);
376
377		if (ret > 0) {
378			if (p->post_handler)
379				p->post_handler(p, regs, 0);
380
381			kcb->kprobe_status = KPROBE_HIT_SSDONE;
382			reset_current_kprobe();
383			preempt_enable_no_resched();
384			return 1;
385		}
386	}
387	prepare_singlestep(p, regs);
388	kcb->kprobe_status = KPROBE_HIT_SS;
389	return 1;
390
391no_kprobe:
392	preempt_enable_no_resched();
393	return ret;
394}
395NOKPROBE_SYMBOL(kprobe_handler);
396
397/*
398 * Function return probe trampoline:
399 * 	- init_kprobes() establishes a probepoint here
400 * 	- When the probed function returns, this probe
401 * 		causes the handlers to fire
402 */
403asm(".global kretprobe_trampoline\n"
404	".type kretprobe_trampoline, @function\n"
405	"kretprobe_trampoline:\n"
406	"nop\n"
407	"blr\n"
408	".size kretprobe_trampoline, .-kretprobe_trampoline\n");
409
410/*
411 * Called when the probe at kretprobe trampoline is hit
412 */
413static int trampoline_probe_handler(struct kprobe *p, struct pt_regs *regs)
414{
415	struct kretprobe_instance *ri = NULL;
416	struct hlist_head *head, empty_rp;
417	struct hlist_node *tmp;
418	unsigned long flags, orig_ret_address = 0;
419	unsigned long trampoline_address =(unsigned long)&kretprobe_trampoline;
420
421	INIT_HLIST_HEAD(&empty_rp);
422	kretprobe_hash_lock(current, &head, &flags);
423
424	/*
425	 * It is possible to have multiple instances associated with a given
426	 * task either because an multiple functions in the call path
427	 * have a return probe installed on them, and/or more than one return
428	 * return probe was registered for a target function.
429	 *
430	 * We can handle this because:
431	 *     - instances are always inserted at the head of the list
432	 *     - when multiple return probes are registered for the same
433	 *       function, the first instance's ret_addr will point to the
434	 *       real return address, and all the rest will point to
435	 *       kretprobe_trampoline
436	 */
437	hlist_for_each_entry_safe(ri, tmp, head, hlist) {
438		if (ri->task != current)
439			/* another task is sharing our hash bucket */
440			continue;
441
442		if (ri->rp && ri->rp->handler)
443			ri->rp->handler(ri, regs);
444
445		orig_ret_address = (unsigned long)ri->ret_addr;
446		recycle_rp_inst(ri, &empty_rp);
447
448		if (orig_ret_address != trampoline_address)
449			/*
450			 * This is the real return address. Any other
451			 * instances associated with this task are for
452			 * other calls deeper on the call stack
453			 */
454			break;
455	}
456
457	kretprobe_assert(ri, orig_ret_address, trampoline_address);
458
459	/*
460	 * We get here through one of two paths:
461	 * 1. by taking a trap -> kprobe_handler() -> here
462	 * 2. by optprobe branch -> optimized_callback() -> opt_pre_handler() -> here
463	 *
464	 * When going back through (1), we need regs->nip to be setup properly
465	 * as it is used to determine the return address from the trap.
466	 * For (2), since nip is not honoured with optprobes, we instead setup
467	 * the link register properly so that the subsequent 'blr' in
468	 * kretprobe_trampoline jumps back to the right instruction.
469	 *
470	 * For nip, we should set the address to the previous instruction since
471	 * we end up emulating it in kprobe_handler(), which increments the nip
472	 * again.
473	 */
474	regs->nip = orig_ret_address - 4;
475	regs->link = orig_ret_address;
476
477	kretprobe_hash_unlock(current, &flags);
478
479	hlist_for_each_entry_safe(ri, tmp, &empty_rp, hlist) {
480		hlist_del(&ri->hlist);
481		kfree(ri);
482	}
483
484	return 0;
485}
486NOKPROBE_SYMBOL(trampoline_probe_handler);
487
488/*
489 * Called after single-stepping.  p->addr is the address of the
490 * instruction whose first byte has been replaced by the "breakpoint"
491 * instruction.  To avoid the SMP problems that can occur when we
492 * temporarily put back the original opcode to single-step, we
493 * single-stepped a copy of the instruction.  The address of this
494 * copy is p->ainsn.insn.
495 */
496int kprobe_post_handler(struct pt_regs *regs)
497{
498	struct kprobe *cur = kprobe_running();
499	struct kprobe_ctlblk *kcb = get_kprobe_ctlblk();
500
501	if (!cur || user_mode(regs))
502		return 0;
503
504	/* make sure we got here for instruction we have a kprobe on */
505	if (((unsigned long)cur->ainsn.insn + 4) != regs->nip)
506		return 0;
507
508	if ((kcb->kprobe_status != KPROBE_REENTER) && cur->post_handler) {
509		kcb->kprobe_status = KPROBE_HIT_SSDONE;
510		cur->post_handler(cur, regs, 0);
511	}
512
513	/* Adjust nip to after the single-stepped instruction */
514	regs->nip = (unsigned long)cur->addr + 4;
515	regs->msr |= kcb->kprobe_saved_msr;
516
517	/*Restore back the original saved kprobes variables and continue. */
518	if (kcb->kprobe_status == KPROBE_REENTER) {
519		restore_previous_kprobe(kcb);
520		goto out;
521	}
522	reset_current_kprobe();
523out:
524	preempt_enable_no_resched();
525
526	/*
527	 * if somebody else is singlestepping across a probe point, msr
528	 * will have DE/SE set, in which case, continue the remaining processing
529	 * of do_debug, as if this is not a probe hit.
530	 */
531	if (regs->msr & MSR_SINGLESTEP)
532		return 0;
533
534	return 1;
535}
536NOKPROBE_SYMBOL(kprobe_post_handler);
537
538int kprobe_fault_handler(struct pt_regs *regs, int trapnr)
539{
540	struct kprobe *cur = kprobe_running();
541	struct kprobe_ctlblk *kcb = get_kprobe_ctlblk();
542	const struct exception_table_entry *entry;
543
544	switch(kcb->kprobe_status) {
545	case KPROBE_HIT_SS:
546	case KPROBE_REENTER:
547		/*
548		 * We are here because the instruction being single
549		 * stepped caused a page fault. We reset the current
550		 * kprobe and the nip points back to the probe address
551		 * and allow the page fault handler to continue as a
552		 * normal page fault.
553		 */
554		regs->nip = (unsigned long)cur->addr;
555		regs->msr &= ~MSR_SINGLESTEP; /* Turn off 'trace' bits */
556		regs->msr |= kcb->kprobe_saved_msr;
557		if (kcb->kprobe_status == KPROBE_REENTER)
558			restore_previous_kprobe(kcb);
559		else
560			reset_current_kprobe();
561		preempt_enable_no_resched();
562		break;
563	case KPROBE_HIT_ACTIVE:
564	case KPROBE_HIT_SSDONE:
565		/*
566		 * We increment the nmissed count for accounting,
567		 * we can also use npre/npostfault count for accounting
568		 * these specific fault cases.
569		 */
570		kprobes_inc_nmissed_count(cur);
571
572		/*
573		 * We come here because instructions in the pre/post
574		 * handler caused the page_fault, this could happen
575		 * if handler tries to access user space by
576		 * copy_from_user(), get_user() etc. Let the
577		 * user-specified handler try to fix it first.
578		 */
579		if (cur->fault_handler && cur->fault_handler(cur, regs, trapnr))
580			return 1;
581
582		/*
583		 * In case the user-specified fault handler returned
584		 * zero, try to fix up.
585		 */
586		if ((entry = search_exception_tables(regs->nip)) != NULL) {
587			regs->nip = extable_fixup(entry);
588			return 1;
589		}
590
591		/*
592		 * fixup_exception() could not handle it,
593		 * Let do_page_fault() fix it.
594		 */
595		break;
596	default:
597		break;
598	}
599	return 0;
600}
601NOKPROBE_SYMBOL(kprobe_fault_handler);
602
603unsigned long arch_deref_entry_point(void *entry)
604{
605#ifdef PPC64_ELF_ABI_v1
606	if (!kernel_text_address((unsigned long)entry))
607		return ppc_global_function_entry(entry);
608	else
609#endif
610		return (unsigned long)entry;
611}
612NOKPROBE_SYMBOL(arch_deref_entry_point);
613
614int 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
619	memcpy(&kcb->jprobe_saved_regs, regs, sizeof(struct pt_regs));
620
621	/* setup return addr to the jprobe handler routine */
622	regs->nip = arch_deref_entry_point(jp->entry);
623#ifdef PPC64_ELF_ABI_v2
624	regs->gpr[12] = (unsigned long)jp->entry;
625#elif defined(PPC64_ELF_ABI_v1)
626	regs->gpr[2] = (unsigned long)(((func_descr_t *)jp->entry)->toc);
627#endif
628
629	/*
630	 * jprobes use jprobe_return() which skips the normal return
631	 * path of the function, and this messes up the accounting of the
632	 * function graph tracer.
633	 *
634	 * Pause function graph tracing while performing the jprobe function.
635	 */
636	pause_graph_tracing();
637
638	return 1;
639}
640NOKPROBE_SYMBOL(setjmp_pre_handler);
641
642void __used jprobe_return(void)
643{
644	asm volatile("jprobe_return_trap:\n"
645		     "trap\n"
646		     ::: "memory");
647}
648NOKPROBE_SYMBOL(jprobe_return);
649
650int longjmp_break_handler(struct kprobe *p, struct pt_regs *regs)
651{
652	struct kprobe_ctlblk *kcb = get_kprobe_ctlblk();
653
654	if (regs->nip != ppc_kallsyms_lookup_name("jprobe_return_trap")) {
655		pr_debug("longjmp_break_handler NIP (0x%lx) does not match jprobe_return_trap (0x%lx)\n",
656				regs->nip, ppc_kallsyms_lookup_name("jprobe_return_trap"));
657		return 0;
658	}
659
660	memcpy(regs, &kcb->jprobe_saved_regs, sizeof(struct pt_regs));
661	/* It's OK to start function graph tracing again */
662	unpause_graph_tracing();
663	preempt_enable_no_resched();
664	return 1;
665}
666NOKPROBE_SYMBOL(longjmp_break_handler);
667
668static struct kprobe trampoline_p = {
669	.addr = (kprobe_opcode_t *) &kretprobe_trampoline,
670	.pre_handler = trampoline_probe_handler
671};
672
673int __init arch_init_kprobes(void)
674{
675	return register_kprobe(&trampoline_p);
676}
677
678int arch_trampoline_kprobe(struct kprobe *p)
679{
680	if (p->addr == (kprobe_opcode_t *)&kretprobe_trampoline)
681		return 1;
682
683	return 0;
684}
685NOKPROBE_SYMBOL(arch_trampoline_kprobe);
v5.4
  1// SPDX-License-Identifier: GPL-2.0-or-later
  2/*
  3 *  Kernel Probes (KProbes)
  4 *
 
 
 
 
 
 
 
 
 
 
 
 
 
 
  5 * Copyright (C) IBM Corporation, 2002, 2004
  6 *
  7 * 2002-Oct	Created by Vamsi Krishna S <vamsi_krishna@in.ibm.com> Kernel
  8 *		Probes initial implementation ( includes contributions from
  9 *		Rusty Russell).
 10 * 2004-July	Suparna Bhattacharya <suparna@in.ibm.com> added jumper probes
 11 *		interface to access function arguments.
 12 * 2004-Nov	Ananth N Mavinakayanahalli <ananth@in.ibm.com> kprobes port
 13 *		for PPC64
 14 */
 15
 16#include <linux/kprobes.h>
 17#include <linux/ptrace.h>
 18#include <linux/preempt.h>
 19#include <linux/extable.h>
 20#include <linux/kdebug.h>
 21#include <linux/slab.h>
 22#include <asm/code-patching.h>
 23#include <asm/cacheflush.h>
 24#include <asm/sstep.h>
 25#include <asm/sections.h>
 26#include <linux/uaccess.h>
 27
 28DEFINE_PER_CPU(struct kprobe *, current_kprobe) = NULL;
 29DEFINE_PER_CPU(struct kprobe_ctlblk, kprobe_ctlblk);
 30
 31struct kretprobe_blackpoint kretprobe_blacklist[] = {{NULL, NULL}};
 32
 33bool arch_within_kprobe_blacklist(unsigned long addr)
 34{
 35	return  (addr >= (unsigned long)__kprobes_text_start &&
 36		 addr < (unsigned long)__kprobes_text_end) ||
 37		(addr >= (unsigned long)_stext &&
 38		 addr < (unsigned long)__head_end);
 39}
 40
 41kprobe_opcode_t *kprobe_lookup_name(const char *name, unsigned int offset)
 42{
 43	kprobe_opcode_t *addr = NULL;
 44
 45#ifdef PPC64_ELF_ABI_v2
 46	/* PPC64 ABIv2 needs local entry point */
 47	addr = (kprobe_opcode_t *)kallsyms_lookup_name(name);
 48	if (addr && !offset) {
 49#ifdef CONFIG_KPROBES_ON_FTRACE
 50		unsigned long faddr;
 51		/*
 52		 * Per livepatch.h, ftrace location is always within the first
 53		 * 16 bytes of a function on powerpc with -mprofile-kernel.
 54		 */
 55		faddr = ftrace_location_range((unsigned long)addr,
 56					      (unsigned long)addr + 16);
 57		if (faddr)
 58			addr = (kprobe_opcode_t *)faddr;
 59		else
 60#endif
 61			addr = (kprobe_opcode_t *)ppc_function_entry(addr);
 62	}
 63#elif defined(PPC64_ELF_ABI_v1)
 64	/*
 65	 * 64bit powerpc ABIv1 uses function descriptors:
 66	 * - Check for the dot variant of the symbol first.
 67	 * - If that fails, try looking up the symbol provided.
 68	 *
 69	 * This ensures we always get to the actual symbol and not
 70	 * the descriptor.
 71	 *
 72	 * Also handle <module:symbol> format.
 73	 */
 74	char dot_name[MODULE_NAME_LEN + 1 + KSYM_NAME_LEN];
 75	bool dot_appended = false;
 76	const char *c;
 77	ssize_t ret = 0;
 78	int len = 0;
 79
 80	if ((c = strnchr(name, MODULE_NAME_LEN, ':')) != NULL) {
 81		c++;
 82		len = c - name;
 83		memcpy(dot_name, name, len);
 84	} else
 85		c = name;
 86
 87	if (*c != '\0' && *c != '.') {
 88		dot_name[len++] = '.';
 89		dot_appended = true;
 90	}
 91	ret = strscpy(dot_name + len, c, KSYM_NAME_LEN);
 92	if (ret > 0)
 93		addr = (kprobe_opcode_t *)kallsyms_lookup_name(dot_name);
 94
 95	/* Fallback to the original non-dot symbol lookup */
 96	if (!addr && dot_appended)
 97		addr = (kprobe_opcode_t *)kallsyms_lookup_name(name);
 98#else
 99	addr = (kprobe_opcode_t *)kallsyms_lookup_name(name);
100#endif
101
102	return addr;
103}
104
105int arch_prepare_kprobe(struct kprobe *p)
106{
107	int ret = 0;
108	kprobe_opcode_t insn = *p->addr;
109
110	if ((unsigned long)p->addr & 0x03) {
111		printk("Attempt to register kprobe at an unaligned address\n");
112		ret = -EINVAL;
113	} else if (IS_MTMSRD(insn) || IS_RFID(insn) || IS_RFI(insn)) {
114		printk("Cannot register a kprobe on rfi/rfid or mtmsr[d]\n");
115		ret = -EINVAL;
116	}
117
118	/* insn must be on a special executable page on ppc64.  This is
119	 * not explicitly required on ppc32 (right now), but it doesn't hurt */
120	if (!ret) {
121		p->ainsn.insn = get_insn_slot();
122		if (!p->ainsn.insn)
123			ret = -ENOMEM;
124	}
125
126	if (!ret) {
127		memcpy(p->ainsn.insn, p->addr,
128				MAX_INSN_SIZE * sizeof(kprobe_opcode_t));
129		p->opcode = *p->addr;
130		flush_icache_range((unsigned long)p->ainsn.insn,
131			(unsigned long)p->ainsn.insn + sizeof(kprobe_opcode_t));
132	}
133
134	p->ainsn.boostable = 0;
135	return ret;
136}
137NOKPROBE_SYMBOL(arch_prepare_kprobe);
138
139void arch_arm_kprobe(struct kprobe *p)
140{
141	patch_instruction(p->addr, BREAKPOINT_INSTRUCTION);
142}
143NOKPROBE_SYMBOL(arch_arm_kprobe);
144
145void arch_disarm_kprobe(struct kprobe *p)
146{
147	patch_instruction(p->addr, p->opcode);
148}
149NOKPROBE_SYMBOL(arch_disarm_kprobe);
150
151void arch_remove_kprobe(struct kprobe *p)
152{
153	if (p->ainsn.insn) {
154		free_insn_slot(p->ainsn.insn, 0);
155		p->ainsn.insn = NULL;
156	}
157}
158NOKPROBE_SYMBOL(arch_remove_kprobe);
159
160static nokprobe_inline void prepare_singlestep(struct kprobe *p, struct pt_regs *regs)
161{
162	enable_single_step(regs);
163
164	/*
165	 * On powerpc we should single step on the original
166	 * instruction even if the probed insn is a trap
167	 * variant as values in regs could play a part in
168	 * if the trap is taken or not
169	 */
170	regs->nip = (unsigned long)p->ainsn.insn;
171}
172
173static nokprobe_inline void save_previous_kprobe(struct kprobe_ctlblk *kcb)
174{
175	kcb->prev_kprobe.kp = kprobe_running();
176	kcb->prev_kprobe.status = kcb->kprobe_status;
177	kcb->prev_kprobe.saved_msr = kcb->kprobe_saved_msr;
178}
179
180static nokprobe_inline void restore_previous_kprobe(struct kprobe_ctlblk *kcb)
181{
182	__this_cpu_write(current_kprobe, kcb->prev_kprobe.kp);
183	kcb->kprobe_status = kcb->prev_kprobe.status;
184	kcb->kprobe_saved_msr = kcb->prev_kprobe.saved_msr;
185}
186
187static nokprobe_inline void set_current_kprobe(struct kprobe *p, struct pt_regs *regs,
188				struct kprobe_ctlblk *kcb)
189{
190	__this_cpu_write(current_kprobe, p);
191	kcb->kprobe_saved_msr = regs->msr;
192}
193
194bool arch_kprobe_on_func_entry(unsigned long offset)
195{
196#ifdef PPC64_ELF_ABI_v2
197#ifdef CONFIG_KPROBES_ON_FTRACE
198	return offset <= 16;
199#else
200	return offset <= 8;
201#endif
202#else
203	return !offset;
204#endif
205}
206
207void arch_prepare_kretprobe(struct kretprobe_instance *ri, struct pt_regs *regs)
208{
209	ri->ret_addr = (kprobe_opcode_t *)regs->link;
210
211	/* Replace the return addr with trampoline addr */
212	regs->link = (unsigned long)kretprobe_trampoline;
213}
214NOKPROBE_SYMBOL(arch_prepare_kretprobe);
215
216static int try_to_emulate(struct kprobe *p, struct pt_regs *regs)
217{
218	int ret;
219	unsigned int insn = *p->ainsn.insn;
220
221	/* regs->nip is also adjusted if emulate_step returns 1 */
222	ret = emulate_step(regs, insn);
223	if (ret > 0) {
224		/*
225		 * Once this instruction has been boosted
226		 * successfully, set the boostable flag
227		 */
228		if (unlikely(p->ainsn.boostable == 0))
229			p->ainsn.boostable = 1;
230	} else if (ret < 0) {
231		/*
232		 * We don't allow kprobes on mtmsr(d)/rfi(d), etc.
233		 * So, we should never get here... but, its still
234		 * good to catch them, just in case...
235		 */
236		printk("Can't step on instruction %x\n", insn);
237		BUG();
238	} else {
239		/*
240		 * If we haven't previously emulated this instruction, then it
241		 * can't be boosted. Note it down so we don't try to do so again.
242		 *
243		 * If, however, we had emulated this instruction in the past,
244		 * then this is just an error with the current run (for
245		 * instance, exceptions due to a load/store). We return 0 so
246		 * that this is now single-stepped, but continue to try
247		 * emulating it in subsequent probe hits.
248		 */
249		if (unlikely(p->ainsn.boostable != 1))
250			p->ainsn.boostable = -1;
251	}
252
253	return ret;
254}
255NOKPROBE_SYMBOL(try_to_emulate);
256
257int kprobe_handler(struct pt_regs *regs)
258{
259	struct kprobe *p;
260	int ret = 0;
261	unsigned int *addr = (unsigned int *)regs->nip;
262	struct kprobe_ctlblk *kcb;
263
264	if (user_mode(regs))
265		return 0;
266
267	/*
268	 * We don't want to be preempted for the entire
269	 * duration of kprobe processing
270	 */
271	preempt_disable();
272	kcb = get_kprobe_ctlblk();
273
274	/* Check we're not actually recursing */
275	if (kprobe_running()) {
276		p = get_kprobe(addr);
277		if (p) {
278			kprobe_opcode_t insn = *p->ainsn.insn;
279			if (kcb->kprobe_status == KPROBE_HIT_SS &&
280					is_trap(insn)) {
281				/* Turn off 'trace' bits */
282				regs->msr &= ~MSR_SINGLESTEP;
283				regs->msr |= kcb->kprobe_saved_msr;
284				goto no_kprobe;
285			}
286			/* We have reentered the kprobe_handler(), since
287			 * another probe was hit while within the handler.
288			 * We here save the original kprobes variables and
289			 * just single step on the instruction of the new probe
290			 * without calling any user handlers.
291			 */
292			save_previous_kprobe(kcb);
293			set_current_kprobe(p, regs, kcb);
294			kprobes_inc_nmissed_count(p);
295			kcb->kprobe_status = KPROBE_REENTER;
296			if (p->ainsn.boostable >= 0) {
297				ret = try_to_emulate(p, regs);
298
299				if (ret > 0) {
300					restore_previous_kprobe(kcb);
301					preempt_enable_no_resched();
302					return 1;
303				}
304			}
305			prepare_singlestep(p, regs);
306			return 1;
307		} else if (*addr != BREAKPOINT_INSTRUCTION) {
308			/* If trap variant, then it belongs not to us */
309			kprobe_opcode_t cur_insn = *addr;
310
311			if (is_trap(cur_insn))
 
 
 
 
 
 
312				goto no_kprobe;
313			/* The breakpoint instruction was removed by
314			 * another cpu right after we hit, no further
315			 * handling of this interrupt is appropriate
316			 */
317			ret = 1;
 
 
318		}
319		goto no_kprobe;
320	}
321
322	p = get_kprobe(addr);
323	if (!p) {
324		if (*addr != BREAKPOINT_INSTRUCTION) {
325			/*
326			 * PowerPC has multiple variants of the "trap"
327			 * instruction. If the current instruction is a
328			 * trap variant, it could belong to someone else
329			 */
330			kprobe_opcode_t cur_insn = *addr;
331			if (is_trap(cur_insn))
332				goto no_kprobe;
333			/*
334			 * The breakpoint instruction was removed right
335			 * after we hit it.  Another cpu has removed
336			 * either a probepoint or a debugger breakpoint
337			 * at this address.  In either case, no further
338			 * handling of this interrupt is appropriate.
339			 */
340			ret = 1;
341		}
342		/* Not one of ours: let kernel handle it */
343		goto no_kprobe;
344	}
345
346	kcb->kprobe_status = KPROBE_HIT_ACTIVE;
347	set_current_kprobe(p, regs, kcb);
348	if (p->pre_handler && p->pre_handler(p, regs)) {
349		/* handler changed execution path, so skip ss setup */
350		reset_current_kprobe();
351		preempt_enable_no_resched();
352		return 1;
353	}
354
 
355	if (p->ainsn.boostable >= 0) {
356		ret = try_to_emulate(p, regs);
357
358		if (ret > 0) {
359			if (p->post_handler)
360				p->post_handler(p, regs, 0);
361
362			kcb->kprobe_status = KPROBE_HIT_SSDONE;
363			reset_current_kprobe();
364			preempt_enable_no_resched();
365			return 1;
366		}
367	}
368	prepare_singlestep(p, regs);
369	kcb->kprobe_status = KPROBE_HIT_SS;
370	return 1;
371
372no_kprobe:
373	preempt_enable_no_resched();
374	return ret;
375}
376NOKPROBE_SYMBOL(kprobe_handler);
377
378/*
379 * Function return probe trampoline:
380 * 	- init_kprobes() establishes a probepoint here
381 * 	- When the probed function returns, this probe
382 * 		causes the handlers to fire
383 */
384asm(".global kretprobe_trampoline\n"
385	".type kretprobe_trampoline, @function\n"
386	"kretprobe_trampoline:\n"
387	"nop\n"
388	"blr\n"
389	".size kretprobe_trampoline, .-kretprobe_trampoline\n");
390
391/*
392 * Called when the probe at kretprobe trampoline is hit
393 */
394static int trampoline_probe_handler(struct kprobe *p, struct pt_regs *regs)
395{
396	struct kretprobe_instance *ri = NULL;
397	struct hlist_head *head, empty_rp;
398	struct hlist_node *tmp;
399	unsigned long flags, orig_ret_address = 0;
400	unsigned long trampoline_address =(unsigned long)&kretprobe_trampoline;
401
402	INIT_HLIST_HEAD(&empty_rp);
403	kretprobe_hash_lock(current, &head, &flags);
404
405	/*
406	 * It is possible to have multiple instances associated with a given
407	 * task either because an multiple functions in the call path
408	 * have a return probe installed on them, and/or more than one return
409	 * return probe was registered for a target function.
410	 *
411	 * We can handle this because:
412	 *     - instances are always inserted at the head of the list
413	 *     - when multiple return probes are registered for the same
414	 *       function, the first instance's ret_addr will point to the
415	 *       real return address, and all the rest will point to
416	 *       kretprobe_trampoline
417	 */
418	hlist_for_each_entry_safe(ri, tmp, head, hlist) {
419		if (ri->task != current)
420			/* another task is sharing our hash bucket */
421			continue;
422
423		if (ri->rp && ri->rp->handler)
424			ri->rp->handler(ri, regs);
425
426		orig_ret_address = (unsigned long)ri->ret_addr;
427		recycle_rp_inst(ri, &empty_rp);
428
429		if (orig_ret_address != trampoline_address)
430			/*
431			 * This is the real return address. Any other
432			 * instances associated with this task are for
433			 * other calls deeper on the call stack
434			 */
435			break;
436	}
437
438	kretprobe_assert(ri, orig_ret_address, trampoline_address);
439
440	/*
441	 * We get here through one of two paths:
442	 * 1. by taking a trap -> kprobe_handler() -> here
443	 * 2. by optprobe branch -> optimized_callback() -> opt_pre_handler() -> here
444	 *
445	 * When going back through (1), we need regs->nip to be setup properly
446	 * as it is used to determine the return address from the trap.
447	 * For (2), since nip is not honoured with optprobes, we instead setup
448	 * the link register properly so that the subsequent 'blr' in
449	 * kretprobe_trampoline jumps back to the right instruction.
450	 *
451	 * For nip, we should set the address to the previous instruction since
452	 * we end up emulating it in kprobe_handler(), which increments the nip
453	 * again.
454	 */
455	regs->nip = orig_ret_address - 4;
456	regs->link = orig_ret_address;
457
458	kretprobe_hash_unlock(current, &flags);
459
460	hlist_for_each_entry_safe(ri, tmp, &empty_rp, hlist) {
461		hlist_del(&ri->hlist);
462		kfree(ri);
463	}
464
465	return 0;
466}
467NOKPROBE_SYMBOL(trampoline_probe_handler);
468
469/*
470 * Called after single-stepping.  p->addr is the address of the
471 * instruction whose first byte has been replaced by the "breakpoint"
472 * instruction.  To avoid the SMP problems that can occur when we
473 * temporarily put back the original opcode to single-step, we
474 * single-stepped a copy of the instruction.  The address of this
475 * copy is p->ainsn.insn.
476 */
477int kprobe_post_handler(struct pt_regs *regs)
478{
479	struct kprobe *cur = kprobe_running();
480	struct kprobe_ctlblk *kcb = get_kprobe_ctlblk();
481
482	if (!cur || user_mode(regs))
483		return 0;
484
485	/* make sure we got here for instruction we have a kprobe on */
486	if (((unsigned long)cur->ainsn.insn + 4) != regs->nip)
487		return 0;
488
489	if ((kcb->kprobe_status != KPROBE_REENTER) && cur->post_handler) {
490		kcb->kprobe_status = KPROBE_HIT_SSDONE;
491		cur->post_handler(cur, regs, 0);
492	}
493
494	/* Adjust nip to after the single-stepped instruction */
495	regs->nip = (unsigned long)cur->addr + 4;
496	regs->msr |= kcb->kprobe_saved_msr;
497
498	/*Restore back the original saved kprobes variables and continue. */
499	if (kcb->kprobe_status == KPROBE_REENTER) {
500		restore_previous_kprobe(kcb);
501		goto out;
502	}
503	reset_current_kprobe();
504out:
505	preempt_enable_no_resched();
506
507	/*
508	 * if somebody else is singlestepping across a probe point, msr
509	 * will have DE/SE set, in which case, continue the remaining processing
510	 * of do_debug, as if this is not a probe hit.
511	 */
512	if (regs->msr & MSR_SINGLESTEP)
513		return 0;
514
515	return 1;
516}
517NOKPROBE_SYMBOL(kprobe_post_handler);
518
519int kprobe_fault_handler(struct pt_regs *regs, int trapnr)
520{
521	struct kprobe *cur = kprobe_running();
522	struct kprobe_ctlblk *kcb = get_kprobe_ctlblk();
523	const struct exception_table_entry *entry;
524
525	switch(kcb->kprobe_status) {
526	case KPROBE_HIT_SS:
527	case KPROBE_REENTER:
528		/*
529		 * We are here because the instruction being single
530		 * stepped caused a page fault. We reset the current
531		 * kprobe and the nip points back to the probe address
532		 * and allow the page fault handler to continue as a
533		 * normal page fault.
534		 */
535		regs->nip = (unsigned long)cur->addr;
536		regs->msr &= ~MSR_SINGLESTEP; /* Turn off 'trace' bits */
537		regs->msr |= kcb->kprobe_saved_msr;
538		if (kcb->kprobe_status == KPROBE_REENTER)
539			restore_previous_kprobe(kcb);
540		else
541			reset_current_kprobe();
542		preempt_enable_no_resched();
543		break;
544	case KPROBE_HIT_ACTIVE:
545	case KPROBE_HIT_SSDONE:
546		/*
547		 * We increment the nmissed count for accounting,
548		 * we can also use npre/npostfault count for accounting
549		 * these specific fault cases.
550		 */
551		kprobes_inc_nmissed_count(cur);
552
553		/*
554		 * We come here because instructions in the pre/post
555		 * handler caused the page_fault, this could happen
556		 * if handler tries to access user space by
557		 * copy_from_user(), get_user() etc. Let the
558		 * user-specified handler try to fix it first.
559		 */
560		if (cur->fault_handler && cur->fault_handler(cur, regs, trapnr))
561			return 1;
562
563		/*
564		 * In case the user-specified fault handler returned
565		 * zero, try to fix up.
566		 */
567		if ((entry = search_exception_tables(regs->nip)) != NULL) {
568			regs->nip = extable_fixup(entry);
569			return 1;
570		}
571
572		/*
573		 * fixup_exception() could not handle it,
574		 * Let do_page_fault() fix it.
575		 */
576		break;
577	default:
578		break;
579	}
580	return 0;
581}
582NOKPROBE_SYMBOL(kprobe_fault_handler);
583
584unsigned long arch_deref_entry_point(void *entry)
585{
586#ifdef PPC64_ELF_ABI_v1
587	if (!kernel_text_address((unsigned long)entry))
588		return ppc_global_function_entry(entry);
589	else
590#endif
591		return (unsigned long)entry;
592}
593NOKPROBE_SYMBOL(arch_deref_entry_point);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
594
595static struct kprobe trampoline_p = {
596	.addr = (kprobe_opcode_t *) &kretprobe_trampoline,
597	.pre_handler = trampoline_probe_handler
598};
599
600int __init arch_init_kprobes(void)
601{
602	return register_kprobe(&trampoline_p);
603}
604
605int arch_trampoline_kprobe(struct kprobe *p)
606{
607	if (p->addr == (kprobe_opcode_t *)&kretprobe_trampoline)
608		return 1;
609
610	return 0;
611}
612NOKPROBE_SYMBOL(arch_trampoline_kprobe);