kprobes.c

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