Linux Audio

Check our new training course

Loading...
v4.17
  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(dmainsn);
 31
 32static void *alloc_dmainsn_page(void)
 33{
 34	void *page;
 35
 36	page = (void *) __get_free_page(GFP_KERNEL | GFP_DMA);
 37	if (page)
 38		set_memory_x((unsigned long) page, 1);
 
 39	return page;
 40}
 41
 42static void free_dmainsn_page(void *page)
 
 
 
 
 
 
 
 43{
 44	set_memory_nx((unsigned long) page, 1);
 45	free_page((unsigned long)page);
 46}
 47
 48struct kprobe_insn_cache kprobe_dmainsn_slots = {
 49	.mutex = __MUTEX_INITIALIZER(kprobe_dmainsn_slots.mutex),
 50	.alloc = alloc_dmainsn_page,
 51	.free = free_dmainsn_page,
 52	.pages = LIST_HEAD_INIT(kprobe_dmainsn_slots.pages),
 53	.insn_size = MAX_INSN_SIZE,
 54};
 55
 56static void copy_instruction(struct kprobe *p)
 57{
 58	unsigned long ip = (unsigned long) p->addr;
 59	s64 disp, new_disp;
 60	u64 addr, new_addr;
 
 61
 62	if (ftrace_location(ip) == ip) {
 
 
 
 63		/*
 64		 * If kprobes patches the instruction that is morphed by
 65		 * ftrace make sure that kprobes always sees the branch
 66		 * "jg .+24" that skips the mcount block or the "brcl 0,0"
 67		 * in case of hotpatch.
 
 68		 */
 69		ftrace_generate_nop_insn((struct ftrace_insn *)p->ainsn.insn);
 70		p->ainsn.is_ftrace_insn = 1;
 71	} else
 72		memcpy(p->ainsn.insn, p->addr, insn_length(*p->addr >> 8));
 73	p->opcode = p->ainsn.insn[0];
 74	if (!probe_is_insn_relative_long(p->ainsn.insn))
 75		return;
 76	/*
 77	 * For pc-relative instructions in RIL-b or RIL-c format patch the
 78	 * RI2 displacement field. We have already made sure that the insn
 79	 * slot for the patched instruction is within the same 2GB area
 80	 * as the original instruction (either kernel image or module area).
 81	 * Therefore the new displacement will always fit.
 82	 */
 83	disp = *(s32 *)&p->ainsn.insn[1];
 84	addr = (u64)(unsigned long)p->addr;
 85	new_addr = (u64)(unsigned long)p->ainsn.insn;
 86	new_disp = ((addr + (disp * 2)) - new_addr) / 2;
 87	*(s32 *)&p->ainsn.insn[1] = new_disp;
 88}
 89NOKPROBE_SYMBOL(copy_instruction);
 90
 91static inline int is_kernel_addr(void *addr)
 92{
 93	return addr < (void *)_end;
 94}
 95
 96static 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_addr(p->addr))
105		p->ainsn.insn = get_dmainsn_slot();
106	else if (is_module_addr(p->addr))
107		p->ainsn.insn = get_insn_slot();
108	return p->ainsn.insn ? 0 : -ENOMEM;
109}
110NOKPROBE_SYMBOL(s390_get_insn_slot);
111
112static void s390_free_insn_slot(struct kprobe *p)
113{
114	if (!p->ainsn.insn)
115		return;
116	if (is_kernel_addr(p->addr))
117		free_dmainsn_slot(p->ainsn.insn, 0);
118	else
119		free_insn_slot(p->ainsn.insn, 0);
120	p->ainsn.insn = NULL;
121}
122NOKPROBE_SYMBOL(s390_free_insn_slot);
123
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
124int arch_prepare_kprobe(struct kprobe *p)
125{
126	if ((unsigned long) p->addr & 0x01)
127		return -EINVAL;
128	/* Make sure the probe isn't going on a difficult instruction */
129	if (probe_is_prohibited_opcode(p->addr))
130		return -EINVAL;
131	if (s390_get_insn_slot(p))
132		return -ENOMEM;
133	copy_instruction(p);
134	return 0;
135}
136NOKPROBE_SYMBOL(arch_prepare_kprobe);
137
138int arch_check_ftrace_location(struct kprobe *p)
139{
140	return 0;
141}
142
143struct swap_insn_args {
144	struct kprobe *p;
145	unsigned int arm_kprobe : 1;
146};
147
148static int swap_instruction(void *data)
149{
150	struct swap_insn_args *args = data;
151	struct ftrace_insn new_insn, *insn;
152	struct kprobe *p = args->p;
153	size_t len;
154
155	new_insn.opc = args->arm_kprobe ? BREAKPOINT_INSTRUCTION : p->opcode;
156	len = sizeof(new_insn.opc);
157	if (!p->ainsn.is_ftrace_insn)
158		goto skip_ftrace;
159	len = sizeof(new_insn);
160	insn = (struct ftrace_insn *) p->addr;
161	if (args->arm_kprobe) {
162		if (is_ftrace_nop(insn))
163			new_insn.disp = KPROBE_ON_FTRACE_NOP;
164		else
165			new_insn.disp = KPROBE_ON_FTRACE_CALL;
166	} else {
167		ftrace_generate_call_insn(&new_insn, (unsigned long)p->addr);
168		if (insn->disp == KPROBE_ON_FTRACE_NOP)
169			ftrace_generate_nop_insn(&new_insn);
170	}
171skip_ftrace:
172	s390_kernel_write(p->addr, &new_insn, len);
173	return 0;
174}
175NOKPROBE_SYMBOL(swap_instruction);
176
177void arch_arm_kprobe(struct kprobe *p)
178{
179	struct swap_insn_args args = {.p = p, .arm_kprobe = 1};
180
181	stop_machine_cpuslocked(swap_instruction, &args, NULL);
182}
183NOKPROBE_SYMBOL(arch_arm_kprobe);
184
185void arch_disarm_kprobe(struct kprobe *p)
186{
187	struct swap_insn_args args = {.p = p, .arm_kprobe = 0};
188
189	stop_machine_cpuslocked(swap_instruction, &args, NULL);
190}
191NOKPROBE_SYMBOL(arch_disarm_kprobe);
192
193void arch_remove_kprobe(struct kprobe *p)
194{
195	s390_free_insn_slot(p);
196}
197NOKPROBE_SYMBOL(arch_remove_kprobe);
198
199static void enable_singlestep(struct kprobe_ctlblk *kcb,
200			      struct pt_regs *regs,
201			      unsigned long ip)
202{
203	struct per_regs per_kprobe;
 
 
 
 
 
 
 
204
205	/* Set up the PER control registers %cr9-%cr11 */
206	per_kprobe.control = PER_EVENT_IFETCH;
207	per_kprobe.start = ip;
208	per_kprobe.end = ip;
209
210	/* Save control regs and psw mask */
211	__ctl_store(kcb->kprobe_saved_ctl, 9, 11);
212	kcb->kprobe_saved_imask = regs->psw.mask &
213		(PSW_MASK_PER | PSW_MASK_IO | PSW_MASK_EXT);
214
215	/* Set PER control regs, turns on single step for the given address */
216	__ctl_load(per_kprobe, 9, 11);
217	regs->psw.mask |= PSW_MASK_PER;
218	regs->psw.mask &= ~(PSW_MASK_IO | PSW_MASK_EXT);
219	regs->psw.addr = ip;
220}
221NOKPROBE_SYMBOL(enable_singlestep);
222
223static void disable_singlestep(struct kprobe_ctlblk *kcb,
224			       struct pt_regs *regs,
225			       unsigned long ip)
226{
227	/* Restore control regs and psw mask, set new psw address */
228	__ctl_load(kcb->kprobe_saved_ctl, 9, 11);
229	regs->psw.mask &= ~PSW_MASK_PER;
230	regs->psw.mask |= kcb->kprobe_saved_imask;
231	regs->psw.addr = ip;
232}
233NOKPROBE_SYMBOL(disable_singlestep);
234
235/*
236 * Activate a kprobe by storing its pointer to current_kprobe. The
237 * previous kprobe is stored in kcb->prev_kprobe. A stack of up to
238 * two kprobes can be active, see KPROBE_REENTER.
239 */
240static void push_kprobe(struct kprobe_ctlblk *kcb, struct kprobe *p)
241{
242	kcb->prev_kprobe.kp = __this_cpu_read(current_kprobe);
243	kcb->prev_kprobe.status = kcb->kprobe_status;
244	__this_cpu_write(current_kprobe, p);
245}
246NOKPROBE_SYMBOL(push_kprobe);
247
248/*
249 * Deactivate a kprobe by backing up to the previous state. If the
250 * current state is KPROBE_REENTER prev_kprobe.kp will be non-NULL,
251 * for any other state prev_kprobe.kp will be NULL.
252 */
253static void pop_kprobe(struct kprobe_ctlblk *kcb)
254{
255	__this_cpu_write(current_kprobe, kcb->prev_kprobe.kp);
256	kcb->kprobe_status = kcb->prev_kprobe.status;
 
257}
258NOKPROBE_SYMBOL(pop_kprobe);
259
260void arch_prepare_kretprobe(struct kretprobe_instance *ri, struct pt_regs *regs)
261{
262	ri->ret_addr = (kprobe_opcode_t *) regs->gprs[14];
263
264	/* Replace the return addr with trampoline addr */
265	regs->gprs[14] = (unsigned long) &kretprobe_trampoline;
266}
267NOKPROBE_SYMBOL(arch_prepare_kretprobe);
268
269static void kprobe_reenter_check(struct kprobe_ctlblk *kcb, struct kprobe *p)
270{
271	switch (kcb->kprobe_status) {
272	case KPROBE_HIT_SSDONE:
273	case KPROBE_HIT_ACTIVE:
274		kprobes_inc_nmissed_count(p);
275		break;
276	case KPROBE_HIT_SS:
277	case KPROBE_REENTER:
278	default:
279		/*
280		 * A kprobe on the code path to single step an instruction
281		 * is a BUG. The code path resides in the .kprobes.text
282		 * section and is executed with interrupts disabled.
283		 */
284		pr_err("Invalid kprobe detected.\n");
285		dump_kprobe(p);
286		BUG();
287	}
288}
289NOKPROBE_SYMBOL(kprobe_reenter_check);
290
291static int kprobe_handler(struct pt_regs *regs)
292{
293	struct kprobe_ctlblk *kcb;
294	struct kprobe *p;
295
296	/*
297	 * We want to disable preemption for the entire duration of kprobe
298	 * processing. That includes the calls to the pre/post handlers
299	 * and single stepping the kprobe instruction.
300	 */
301	preempt_disable();
302	kcb = get_kprobe_ctlblk();
303	p = get_kprobe((void *)(regs->psw.addr - 2));
304
305	if (p) {
306		if (kprobe_running()) {
307			/*
308			 * We have hit a kprobe while another is still
309			 * active. This can happen in the pre and post
310			 * handler. Single step the instruction of the
311			 * new probe but do not call any handler function
312			 * of this secondary kprobe.
313			 * push_kprobe and pop_kprobe saves and restores
314			 * the currently active kprobe.
315			 */
316			kprobe_reenter_check(kcb, p);
317			push_kprobe(kcb, p);
318			kcb->kprobe_status = KPROBE_REENTER;
319		} else {
320			/*
321			 * If we have no pre-handler or it returned 0, we
322			 * continue with single stepping. If we have a
323			 * pre-handler and it returned non-zero, it prepped
324			 * for calling the break_handler below on re-entry
325			 * for jprobe processing, so get out doing nothing
326			 * 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				return 1;
 
332			kcb->kprobe_status = KPROBE_HIT_SS;
333		}
334		enable_singlestep(kcb, regs, (unsigned long) p->ainsn.insn);
335		return 1;
336	} else if (kprobe_running()) {
337		p = __this_cpu_read(current_kprobe);
338		if (p->break_handler && p->break_handler(p, regs)) {
339			/*
340			 * Continuation after the jprobe completed and
341			 * caused the jprobe_return trap. The jprobe
342			 * break_handler "returns" to the original
343			 * function that still has the kprobe breakpoint
344			 * installed. We continue with single stepping.
345			 */
346			kcb->kprobe_status = KPROBE_HIT_SS;
347			enable_singlestep(kcb, regs,
348					  (unsigned long) p->ainsn.insn);
349			return 1;
350		} /* else:
351		   * No kprobe at this address and the current kprobe
352		   * has no break handler (no jprobe!). The kernel just
353		   * exploded, let the standard trap handler pick up the
354		   * pieces.
355		   */
356	} /* else:
357	   * No kprobe at this address and no active kprobe. The trap has
358	   * not been caused by a kprobe breakpoint. The race of breakpoint
359	   * vs. kprobe remove does not exist because on s390 as we use
360	   * stop_machine to arm/disarm the breakpoints.
361	   */
362	preempt_enable_no_resched();
363	return 0;
364}
365NOKPROBE_SYMBOL(kprobe_handler);
366
367/*
368 * Function return probe trampoline:
369 *	- init_kprobes() establishes a probepoint here
370 *	- When the probed function returns, this probe
371 *		causes the handlers to fire
372 */
373static void __used kretprobe_trampoline_holder(void)
374{
375	asm volatile(".global kretprobe_trampoline\n"
376		     "kretprobe_trampoline: bcr 0,0\n");
377}
378
379/*
380 * Called when the probe at kretprobe trampoline is hit
381 */
382static int trampoline_probe_handler(struct kprobe *p, struct pt_regs *regs)
383{
384	struct kretprobe_instance *ri;
385	struct hlist_head *head, empty_rp;
386	struct hlist_node *tmp;
387	unsigned long flags, orig_ret_address;
388	unsigned long trampoline_address;
389	kprobe_opcode_t *correct_ret_addr;
390
391	INIT_HLIST_HEAD(&empty_rp);
392	kretprobe_hash_lock(current, &head, &flags);
393
394	/*
395	 * It is possible to have multiple instances associated with a given
396	 * task either because an multiple functions in the call path
397	 * have a return probe installed on them, and/or more than one return
398	 * return probe was registered for a target function.
399	 *
400	 * We can handle this because:
401	 *     - instances are always inserted at the head of the list
402	 *     - when multiple return probes are registered for the same
403	 *	 function, the first instance's ret_addr will point to the
404	 *	 real return address, and all the rest will point to
405	 *	 kretprobe_trampoline
406	 */
407	ri = NULL;
408	orig_ret_address = 0;
409	correct_ret_addr = NULL;
410	trampoline_address = (unsigned long) &kretprobe_trampoline;
411	hlist_for_each_entry_safe(ri, tmp, head, hlist) {
412		if (ri->task != current)
413			/* another task is sharing our hash bucket */
414			continue;
415
416		orig_ret_address = (unsigned long) ri->ret_addr;
417
418		if (orig_ret_address != trampoline_address)
419			/*
420			 * This is the real return address. Any other
421			 * instances associated with this task are for
422			 * other calls deeper on the call stack
423			 */
424			break;
425	}
426
427	kretprobe_assert(ri, orig_ret_address, trampoline_address);
428
429	correct_ret_addr = ri->ret_addr;
430	hlist_for_each_entry_safe(ri, tmp, head, hlist) {
431		if (ri->task != current)
432			/* another task is sharing our hash bucket */
433			continue;
434
435		orig_ret_address = (unsigned long) ri->ret_addr;
436
437		if (ri->rp && ri->rp->handler) {
438			ri->ret_addr = correct_ret_addr;
439			ri->rp->handler(ri, regs);
440		}
441
442		recycle_rp_inst(ri, &empty_rp);
443
444		if (orig_ret_address != trampoline_address)
445			/*
446			 * This is the real return address. Any other
447			 * instances associated with this task are for
448			 * other calls deeper on the call stack
449			 */
450			break;
451	}
452
453	regs->psw.addr = orig_ret_address;
454
455	pop_kprobe(get_kprobe_ctlblk());
456	kretprobe_hash_unlock(current, &flags);
457	preempt_enable_no_resched();
458
459	hlist_for_each_entry_safe(ri, tmp, &empty_rp, hlist) {
460		hlist_del(&ri->hlist);
461		kfree(ri);
462	}
463	/*
464	 * By returning a non-zero value, we are telling
465	 * kprobe_handler() that we don't want the post_handler
466	 * to run (and have re-enabled preemption)
467	 */
468	return 1;
469}
470NOKPROBE_SYMBOL(trampoline_probe_handler);
471
472/*
473 * Called after single-stepping.  p->addr is the address of the
474 * instruction whose first byte has been replaced by the "breakpoint"
475 * instruction.  To avoid the SMP problems that can occur when we
476 * temporarily put back the original opcode to single-step, we
477 * single-stepped a copy of the instruction.  The address of this
478 * copy is p->ainsn.insn.
479 */
480static void resume_execution(struct kprobe *p, struct pt_regs *regs)
481{
482	struct kprobe_ctlblk *kcb = get_kprobe_ctlblk();
483	unsigned long ip = regs->psw.addr;
484	int fixup = probe_get_fixup_type(p->ainsn.insn);
485
486	/* Check if the kprobes location is an enabled ftrace caller */
487	if (p->ainsn.is_ftrace_insn) {
488		struct ftrace_insn *insn = (struct ftrace_insn *) p->addr;
489		struct ftrace_insn call_insn;
490
491		ftrace_generate_call_insn(&call_insn, (unsigned long) p->addr);
492		/*
493		 * A kprobe on an enabled ftrace call site actually single
494		 * stepped an unconditional branch (ftrace nop equivalent).
495		 * Now we need to fixup things and pretend that a brasl r0,...
496		 * was executed instead.
497		 */
498		if (insn->disp == KPROBE_ON_FTRACE_CALL) {
499			ip += call_insn.disp * 2 - MCOUNT_INSN_SIZE;
500			regs->gprs[0] = (unsigned long)p->addr + sizeof(*insn);
501		}
502	}
503
504	if (fixup & FIXUP_PSW_NORMAL)
505		ip += (unsigned long) p->addr - (unsigned long) p->ainsn.insn;
506
507	if (fixup & FIXUP_BRANCH_NOT_TAKEN) {
508		int ilen = insn_length(p->ainsn.insn[0] >> 8);
509		if (ip - (unsigned long) p->ainsn.insn == ilen)
510			ip = (unsigned long) p->addr + ilen;
511	}
512
513	if (fixup & FIXUP_RETURN_REGISTER) {
514		int reg = (p->ainsn.insn[0] & 0xf0) >> 4;
515		regs->gprs[reg] += (unsigned long) p->addr -
516				   (unsigned long) p->ainsn.insn;
517	}
518
519	disable_singlestep(kcb, regs, ip);
520}
521NOKPROBE_SYMBOL(resume_execution);
522
523static int post_kprobe_handler(struct pt_regs *regs)
524{
525	struct kprobe_ctlblk *kcb = get_kprobe_ctlblk();
526	struct kprobe *p = kprobe_running();
527
528	if (!p)
529		return 0;
530
 
531	if (kcb->kprobe_status != KPROBE_REENTER && p->post_handler) {
532		kcb->kprobe_status = KPROBE_HIT_SSDONE;
533		p->post_handler(p, regs, 0);
534	}
535
536	resume_execution(p, regs);
537	pop_kprobe(kcb);
538	preempt_enable_no_resched();
539
540	/*
541	 * if somebody else is singlestepping across a probe point, psw mask
542	 * will have PER set, in which case, continue the remaining processing
543	 * of do_single_step, as if this is not a probe hit.
544	 */
545	if (regs->psw.mask & PSW_MASK_PER)
546		return 0;
547
548	return 1;
549}
550NOKPROBE_SYMBOL(post_kprobe_handler);
551
552static int kprobe_trap_handler(struct pt_regs *regs, int trapnr)
553{
554	struct kprobe_ctlblk *kcb = get_kprobe_ctlblk();
555	struct kprobe *p = kprobe_running();
556	const struct exception_table_entry *entry;
557
558	switch(kcb->kprobe_status) {
559	case KPROBE_HIT_SS:
560	case KPROBE_REENTER:
561		/*
562		 * We are here because the instruction being single
563		 * stepped caused a page fault. We reset the current
564		 * kprobe and the nip points back to the probe address
565		 * and allow the page fault handler to continue as a
566		 * normal page fault.
567		 */
568		disable_singlestep(kcb, regs, (unsigned long) p->addr);
569		pop_kprobe(kcb);
570		preempt_enable_no_resched();
571		break;
572	case KPROBE_HIT_ACTIVE:
573	case KPROBE_HIT_SSDONE:
574		/*
575		 * We increment the nmissed count for accounting,
576		 * we can also use npre/npostfault count for accounting
577		 * these specific fault cases.
578		 */
579		kprobes_inc_nmissed_count(p);
580
581		/*
582		 * We come here because instructions in the pre/post
583		 * handler caused the page_fault, this could happen
584		 * if handler tries to access user space by
585		 * copy_from_user(), get_user() etc. Let the
586		 * user-specified handler try to fix it first.
587		 */
588		if (p->fault_handler && p->fault_handler(p, regs, trapnr))
589			return 1;
590
591		/*
592		 * In case the user-specified fault handler returned
593		 * zero, try to fix up.
594		 */
595		entry = search_exception_tables(regs->psw.addr);
596		if (entry) {
597			regs->psw.addr = extable_fixup(entry);
598			return 1;
599		}
600
601		/*
602		 * fixup_exception() could not handle it,
603		 * Let do_page_fault() fix it.
604		 */
605		break;
606	default:
607		break;
608	}
609	return 0;
610}
611NOKPROBE_SYMBOL(kprobe_trap_handler);
612
613int kprobe_fault_handler(struct pt_regs *regs, int trapnr)
614{
615	int ret;
616
617	if (regs->psw.mask & (PSW_MASK_IO | PSW_MASK_EXT))
618		local_irq_disable();
619	ret = kprobe_trap_handler(regs, trapnr);
620	if (regs->psw.mask & (PSW_MASK_IO | PSW_MASK_EXT))
621		local_irq_restore(regs->psw.mask & ~PSW_MASK_PER);
622	return ret;
623}
624NOKPROBE_SYMBOL(kprobe_fault_handler);
625
626/*
627 * Wrapper routine to for handling exceptions.
628 */
629int kprobe_exceptions_notify(struct notifier_block *self,
630			     unsigned long val, void *data)
631{
632	struct die_args *args = (struct die_args *) data;
633	struct pt_regs *regs = args->regs;
634	int ret = NOTIFY_DONE;
635
636	if (regs->psw.mask & (PSW_MASK_IO | PSW_MASK_EXT))
637		local_irq_disable();
638
639	switch (val) {
640	case DIE_BPT:
641		if (kprobe_handler(regs))
642			ret = NOTIFY_STOP;
643		break;
644	case DIE_SSTEP:
645		if (post_kprobe_handler(regs))
646			ret = NOTIFY_STOP;
647		break;
648	case DIE_TRAP:
649		if (!preemptible() && kprobe_running() &&
650		    kprobe_trap_handler(regs, args->trapnr))
651			ret = NOTIFY_STOP;
652		break;
653	default:
654		break;
655	}
656
657	if (regs->psw.mask & (PSW_MASK_IO | PSW_MASK_EXT))
658		local_irq_restore(regs->psw.mask & ~PSW_MASK_PER);
659
660	return ret;
661}
662NOKPROBE_SYMBOL(kprobe_exceptions_notify);
663
664int setjmp_pre_handler(struct kprobe *p, struct pt_regs *regs)
665{
666	struct jprobe *jp = container_of(p, struct jprobe, kp);
667	struct kprobe_ctlblk *kcb = get_kprobe_ctlblk();
668	unsigned long stack;
669
670	memcpy(&kcb->jprobe_saved_regs, regs, sizeof(struct pt_regs));
671
672	/* setup return addr to the jprobe handler routine */
673	regs->psw.addr = (unsigned long) jp->entry;
674	regs->psw.mask &= ~(PSW_MASK_IO | PSW_MASK_EXT);
675
676	/* r15 is the stack pointer */
677	stack = (unsigned long) regs->gprs[15];
678
679	memcpy(kcb->jprobes_stack, (void *) stack, MIN_STACK_SIZE(stack));
680
681	/*
682	 * jprobes use jprobe_return() which skips the normal return
683	 * path of the function, and this messes up the accounting of the
684	 * function graph tracer to get messed up.
685	 *
686	 * Pause function graph tracing while performing the jprobe function.
687	 */
688	pause_graph_tracing();
689	return 1;
690}
691NOKPROBE_SYMBOL(setjmp_pre_handler);
692
693void jprobe_return(void)
694{
695	asm volatile(".word 0x0002");
696}
697NOKPROBE_SYMBOL(jprobe_return);
698
699int longjmp_break_handler(struct kprobe *p, struct pt_regs *regs)
700{
701	struct kprobe_ctlblk *kcb = get_kprobe_ctlblk();
702	unsigned long stack;
703
704	/* It's OK to start function graph tracing again */
705	unpause_graph_tracing();
706
707	stack = (unsigned long) kcb->jprobe_saved_regs.gprs[15];
708
709	/* Put the regs back */
710	memcpy(regs, &kcb->jprobe_saved_regs, sizeof(struct pt_regs));
711	/* put the stack back */
712	memcpy((void *) stack, kcb->jprobes_stack, MIN_STACK_SIZE(stack));
713	preempt_enable_no_resched();
714	return 1;
715}
716NOKPROBE_SYMBOL(longjmp_break_handler);
717
718static struct kprobe trampoline = {
719	.addr = (kprobe_opcode_t *) &kretprobe_trampoline,
720	.pre_handler = trampoline_probe_handler
721};
722
723int __init arch_init_kprobes(void)
724{
725	return register_kprobe(&trampoline);
726}
727
728int arch_trampoline_kprobe(struct kprobe *p)
729{
730	return p->addr == (kprobe_opcode_t *) &kretprobe_trampoline;
731}
732NOKPROBE_SYMBOL(arch_trampoline_kprobe);
v6.9.4
  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
 30DEFINE_PER_CPU(struct kprobe *, current_kprobe);
 31DEFINE_PER_CPU(struct kprobe_ctlblk, kprobe_ctlblk);
 32
 33struct kretprobe_blackpoint kretprobe_blacklist[] = { };
 34
 35static int insn_page_in_use;
 36
 37void *alloc_insn_page(void)
 38{
 39	void *page;
 40
 41	page = module_alloc(PAGE_SIZE);
 42	if (!page)
 43		return NULL;
 44	set_memory_rox((unsigned long)page, 1);
 45	return page;
 46}
 47
 48static 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
 55static void free_s390_insn_page(void *page)
 56{
 57	xchg(&insn_page_in_use, 0);
 
 58}
 59
 60struct 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
 68static 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}
 94NOKPROBE_SYMBOL(copy_instruction);
 95
 
 
 
 
 
 96static 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}
110NOKPROBE_SYMBOL(s390_get_insn_slot);
111
112static 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}
122NOKPROBE_SYMBOL(s390_free_insn_slot);
123
124/* Check if paddr is at an instruction boundary */
125static 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
170int 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}
182NOKPROBE_SYMBOL(arch_prepare_kprobe);
183
 
 
 
 
 
184struct swap_insn_args {
185	struct kprobe *p;
186	unsigned int arm_kprobe : 1;
187};
188
189static 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}
199NOKPROBE_SYMBOL(swap_instruction);
200
201void 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}
207NOKPROBE_SYMBOL(arch_arm_kprobe);
208
209void 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}
215NOKPROBE_SYMBOL(arch_disarm_kprobe);
216
217void arch_remove_kprobe(struct kprobe *p)
218{
219	s390_free_insn_slot(p);
220}
221NOKPROBE_SYMBOL(arch_remove_kprobe);
222
223static void enable_singlestep(struct kprobe_ctlblk *kcb,
224			      struct pt_regs *regs,
225			      unsigned long ip)
226{
227	union {
228		struct ctlreg regs[3];
229		struct {
230			struct ctlreg control;
231			struct ctlreg start;
232			struct ctlreg end;
233		};
234	} per_kprobe;
235
236	/* Set up the PER control registers %cr9-%cr11 */
237	per_kprobe.control.val = PER_EVENT_IFETCH;
238	per_kprobe.start.val = ip;
239	per_kprobe.end.val = ip;
240
241	/* Save control regs and psw mask */
242	__local_ctl_store(9, 11, kcb->kprobe_saved_ctl);
243	kcb->kprobe_saved_imask = regs->psw.mask &
244		(PSW_MASK_PER | PSW_MASK_IO | PSW_MASK_EXT);
245
246	/* Set PER control regs, turns on single step for the given address */
247	__local_ctl_load(9, 11, per_kprobe.regs);
248	regs->psw.mask |= PSW_MASK_PER;
249	regs->psw.mask &= ~(PSW_MASK_IO | PSW_MASK_EXT);
250	regs->psw.addr = ip;
251}
252NOKPROBE_SYMBOL(enable_singlestep);
253
254static void disable_singlestep(struct kprobe_ctlblk *kcb,
255			       struct pt_regs *regs,
256			       unsigned long ip)
257{
258	/* Restore control regs and psw mask, set new psw address */
259	__local_ctl_load(9, 11, kcb->kprobe_saved_ctl);
260	regs->psw.mask &= ~PSW_MASK_PER;
261	regs->psw.mask |= kcb->kprobe_saved_imask;
262	regs->psw.addr = ip;
263}
264NOKPROBE_SYMBOL(disable_singlestep);
265
266/*
267 * Activate a kprobe by storing its pointer to current_kprobe. The
268 * previous kprobe is stored in kcb->prev_kprobe. A stack of up to
269 * two kprobes can be active, see KPROBE_REENTER.
270 */
271static void push_kprobe(struct kprobe_ctlblk *kcb, struct kprobe *p)
272{
273	kcb->prev_kprobe.kp = __this_cpu_read(current_kprobe);
274	kcb->prev_kprobe.status = kcb->kprobe_status;
275	__this_cpu_write(current_kprobe, p);
276}
277NOKPROBE_SYMBOL(push_kprobe);
278
279/*
280 * Deactivate a kprobe by backing up to the previous state. If the
281 * current state is KPROBE_REENTER prev_kprobe.kp will be non-NULL,
282 * for any other state prev_kprobe.kp will be NULL.
283 */
284static void pop_kprobe(struct kprobe_ctlblk *kcb)
285{
286	__this_cpu_write(current_kprobe, kcb->prev_kprobe.kp);
287	kcb->kprobe_status = kcb->prev_kprobe.status;
288	kcb->prev_kprobe.kp = NULL;
289}
290NOKPROBE_SYMBOL(pop_kprobe);
291
 
 
 
 
 
 
 
 
 
292static void kprobe_reenter_check(struct kprobe_ctlblk *kcb, struct kprobe *p)
293{
294	switch (kcb->kprobe_status) {
295	case KPROBE_HIT_SSDONE:
296	case KPROBE_HIT_ACTIVE:
297		kprobes_inc_nmissed_count(p);
298		break;
299	case KPROBE_HIT_SS:
300	case KPROBE_REENTER:
301	default:
302		/*
303		 * A kprobe on the code path to single step an instruction
304		 * is a BUG. The code path resides in the .kprobes.text
305		 * section and is executed with interrupts disabled.
306		 */
307		pr_err("Failed to recover from reentered kprobes.\n");
308		dump_kprobe(p);
309		BUG();
310	}
311}
312NOKPROBE_SYMBOL(kprobe_reenter_check);
313
314static int kprobe_handler(struct pt_regs *regs)
315{
316	struct kprobe_ctlblk *kcb;
317	struct kprobe *p;
318
319	/*
320	 * We want to disable preemption for the entire duration of kprobe
321	 * processing. That includes the calls to the pre/post handlers
322	 * and single stepping the kprobe instruction.
323	 */
324	preempt_disable();
325	kcb = get_kprobe_ctlblk();
326	p = get_kprobe((void *)(regs->psw.addr - 2));
327
328	if (p) {
329		if (kprobe_running()) {
330			/*
331			 * We have hit a kprobe while another is still
332			 * active. This can happen in the pre and post
333			 * handler. Single step the instruction of the
334			 * new probe but do not call any handler function
335			 * of this secondary kprobe.
336			 * push_kprobe and pop_kprobe saves and restores
337			 * the currently active kprobe.
338			 */
339			kprobe_reenter_check(kcb, p);
340			push_kprobe(kcb, p);
341			kcb->kprobe_status = KPROBE_REENTER;
342		} else {
343			/*
344			 * If we have no pre-handler or it returned 0, we
345			 * continue with single stepping. If we have a
346			 * pre-handler and it returned non-zero, it prepped
347			 * for changing execution path, so get out doing
348			 * nothing more here.
 
349			 */
350			push_kprobe(kcb, p);
351			kcb->kprobe_status = KPROBE_HIT_ACTIVE;
352			if (p->pre_handler && p->pre_handler(p, regs)) {
353				pop_kprobe(kcb);
354				preempt_enable_no_resched();
355				return 1;
356			}
357			kcb->kprobe_status = KPROBE_HIT_SS;
358		}
359		enable_singlestep(kcb, regs, (unsigned long) p->ainsn.insn);
360		return 1;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
361	} /* else:
362	   * No kprobe at this address and no active kprobe. The trap has
363	   * not been caused by a kprobe breakpoint. The race of breakpoint
364	   * vs. kprobe remove does not exist because on s390 as we use
365	   * stop_machine to arm/disarm the breakpoints.
366	   */
367	preempt_enable_no_resched();
368	return 0;
369}
370NOKPROBE_SYMBOL(kprobe_handler);
371
372/*
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
373 * Called after single-stepping.  p->addr is the address of the
374 * instruction whose first byte has been replaced by the "breakpoint"
375 * instruction.  To avoid the SMP problems that can occur when we
376 * temporarily put back the original opcode to single-step, we
377 * single-stepped a copy of the instruction.  The address of this
378 * copy is p->ainsn.insn.
379 */
380static void resume_execution(struct kprobe *p, struct pt_regs *regs)
381{
382	struct kprobe_ctlblk *kcb = get_kprobe_ctlblk();
383	unsigned long ip = regs->psw.addr;
384	int fixup = probe_get_fixup_type(p->ainsn.insn);
385
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
386	if (fixup & FIXUP_PSW_NORMAL)
387		ip += (unsigned long) p->addr - (unsigned long) p->ainsn.insn;
388
389	if (fixup & FIXUP_BRANCH_NOT_TAKEN) {
390		int ilen = insn_length(p->ainsn.insn[0] >> 8);
391		if (ip - (unsigned long) p->ainsn.insn == ilen)
392			ip = (unsigned long) p->addr + ilen;
393	}
394
395	if (fixup & FIXUP_RETURN_REGISTER) {
396		int reg = (p->ainsn.insn[0] & 0xf0) >> 4;
397		regs->gprs[reg] += (unsigned long) p->addr -
398				   (unsigned long) p->ainsn.insn;
399	}
400
401	disable_singlestep(kcb, regs, ip);
402}
403NOKPROBE_SYMBOL(resume_execution);
404
405static int post_kprobe_handler(struct pt_regs *regs)
406{
407	struct kprobe_ctlblk *kcb = get_kprobe_ctlblk();
408	struct kprobe *p = kprobe_running();
409
410	if (!p)
411		return 0;
412
413	resume_execution(p, regs);
414	if (kcb->kprobe_status != KPROBE_REENTER && p->post_handler) {
415		kcb->kprobe_status = KPROBE_HIT_SSDONE;
416		p->post_handler(p, regs, 0);
417	}
 
 
418	pop_kprobe(kcb);
419	preempt_enable_no_resched();
420
421	/*
422	 * if somebody else is singlestepping across a probe point, psw mask
423	 * will have PER set, in which case, continue the remaining processing
424	 * of do_single_step, as if this is not a probe hit.
425	 */
426	if (regs->psw.mask & PSW_MASK_PER)
427		return 0;
428
429	return 1;
430}
431NOKPROBE_SYMBOL(post_kprobe_handler);
432
433static int kprobe_trap_handler(struct pt_regs *regs, int trapnr)
434{
435	struct kprobe_ctlblk *kcb = get_kprobe_ctlblk();
436	struct kprobe *p = kprobe_running();
 
437
438	switch(kcb->kprobe_status) {
439	case KPROBE_HIT_SS:
440	case KPROBE_REENTER:
441		/*
442		 * We are here because the instruction being single
443		 * stepped caused a page fault. We reset the current
444		 * kprobe and the nip points back to the probe address
445		 * and allow the page fault handler to continue as a
446		 * normal page fault.
447		 */
448		disable_singlestep(kcb, regs, (unsigned long) p->addr);
449		pop_kprobe(kcb);
450		preempt_enable_no_resched();
451		break;
452	case KPROBE_HIT_ACTIVE:
453	case KPROBE_HIT_SSDONE:
454		/*
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
455		 * In case the user-specified fault handler returned
456		 * zero, try to fix up.
457		 */
458		if (fixup_exception(regs))
 
 
459			return 1;
 
 
460		/*
461		 * fixup_exception() could not handle it,
462		 * Let do_page_fault() fix it.
463		 */
464		break;
465	default:
466		break;
467	}
468	return 0;
469}
470NOKPROBE_SYMBOL(kprobe_trap_handler);
471
472int kprobe_fault_handler(struct pt_regs *regs, int trapnr)
473{
474	int ret;
475
476	if (regs->psw.mask & (PSW_MASK_IO | PSW_MASK_EXT))
477		local_irq_disable();
478	ret = kprobe_trap_handler(regs, trapnr);
479	if (regs->psw.mask & (PSW_MASK_IO | PSW_MASK_EXT))
480		local_irq_restore(regs->psw.mask & ~PSW_MASK_PER);
481	return ret;
482}
483NOKPROBE_SYMBOL(kprobe_fault_handler);
484
485/*
486 * Wrapper routine to for handling exceptions.
487 */
488int kprobe_exceptions_notify(struct notifier_block *self,
489			     unsigned long val, void *data)
490{
491	struct die_args *args = (struct die_args *) data;
492	struct pt_regs *regs = args->regs;
493	int ret = NOTIFY_DONE;
494
495	if (regs->psw.mask & (PSW_MASK_IO | PSW_MASK_EXT))
496		local_irq_disable();
497
498	switch (val) {
499	case DIE_BPT:
500		if (kprobe_handler(regs))
501			ret = NOTIFY_STOP;
502		break;
503	case DIE_SSTEP:
504		if (post_kprobe_handler(regs))
505			ret = NOTIFY_STOP;
506		break;
507	case DIE_TRAP:
508		if (!preemptible() && kprobe_running() &&
509		    kprobe_trap_handler(regs, args->trapnr))
510			ret = NOTIFY_STOP;
511		break;
512	default:
513		break;
514	}
515
516	if (regs->psw.mask & (PSW_MASK_IO | PSW_MASK_EXT))
517		local_irq_restore(regs->psw.mask & ~PSW_MASK_PER);
518
519	return ret;
520}
521NOKPROBE_SYMBOL(kprobe_exceptions_notify);
522
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
523int __init arch_init_kprobes(void)
524{
525	return 0;
526}
527
528int arch_trampoline_kprobe(struct kprobe *p)
529{
530	return 0;
531}
532NOKPROBE_SYMBOL(arch_trampoline_kprobe);