1 /*
2 * Kernel Probes Jump Optimization (Optprobes)
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 * Copyright (C) Hitachi Ltd., 2012
20 */
21 #include <linux/kprobes.h>
22 #include <linux/ptrace.h>
23 #include <linux/string.h>
24 #include <linux/slab.h>
25 #include <linux/hardirq.h>
26 #include <linux/preempt.h>
27 #include <linux/extable.h>
28 #include <linux/kdebug.h>
29 #include <linux/kallsyms.h>
30 #include <linux/ftrace.h>
31 #include <linux/frame.h>
32
33 #include <asm/text-patching.h>
34 #include <asm/cacheflush.h>
35 #include <asm/desc.h>
36 #include <asm/pgtable.h>
37 #include <linux/uaccess.h>
38 #include <asm/alternative.h>
39 #include <asm/insn.h>
40 #include <asm/debugreg.h>
41 #include <asm/set_memory.h>
42 #include <asm/sections.h>
43 #include <asm/nospec-branch.h>
44
45 #include "common.h"
46
__recover_optprobed_insn(kprobe_opcode_t * buf,unsigned long addr)47 unsigned long __recover_optprobed_insn(kprobe_opcode_t *buf, unsigned long addr)
48 {
49 struct optimized_kprobe *op;
50 struct kprobe *kp;
51 long offs;
52 int i;
53
54 for (i = 0; i < RELATIVEJUMP_SIZE; i++) {
55 kp = get_kprobe((void *)addr - i);
56 /* This function only handles jump-optimized kprobe */
57 if (kp && kprobe_optimized(kp)) {
58 op = container_of(kp, struct optimized_kprobe, kp);
59 /* If op->list is not empty, op is under optimizing */
60 if (list_empty(&op->list))
61 goto found;
62 }
63 }
64
65 return addr;
66 found:
67 /*
68 * If the kprobe can be optimized, original bytes which can be
69 * overwritten by jump destination address. In this case, original
70 * bytes must be recovered from op->optinsn.copied_insn buffer.
71 */
72 if (probe_kernel_read(buf, (void *)addr,
73 MAX_INSN_SIZE * sizeof(kprobe_opcode_t)))
74 return 0UL;
75
76 if (addr == (unsigned long)kp->addr) {
77 buf[0] = kp->opcode;
78 memcpy(buf + 1, op->optinsn.copied_insn, RELATIVE_ADDR_SIZE);
79 } else {
80 offs = addr - (unsigned long)kp->addr - 1;
81 memcpy(buf, op->optinsn.copied_insn + offs, RELATIVE_ADDR_SIZE - offs);
82 }
83
84 return (unsigned long)buf;
85 }
86
87 /* Insert a move instruction which sets a pointer to eax/rdi (1st arg). */
synthesize_set_arg1(kprobe_opcode_t * addr,unsigned long val)88 static void synthesize_set_arg1(kprobe_opcode_t *addr, unsigned long val)
89 {
90 #ifdef CONFIG_X86_64
91 *addr++ = 0x48;
92 *addr++ = 0xbf;
93 #else
94 *addr++ = 0xb8;
95 #endif
96 *(unsigned long *)addr = val;
97 }
98
99 asm (
100 "optprobe_template_func:\n"
101 ".global optprobe_template_entry\n"
102 "optprobe_template_entry:\n"
103 #ifdef CONFIG_X86_64
104 /* We don't bother saving the ss register */
105 " pushq %rsp\n"
106 " pushfq\n"
107 SAVE_REGS_STRING
108 " movq %rsp, %rsi\n"
109 ".global optprobe_template_val\n"
110 "optprobe_template_val:\n"
111 ASM_NOP5
112 ASM_NOP5
113 ".global optprobe_template_call\n"
114 "optprobe_template_call:\n"
115 ASM_NOP5
116 /* Move flags to rsp */
117 " movq 144(%rsp), %rdx\n"
118 " movq %rdx, 152(%rsp)\n"
119 RESTORE_REGS_STRING
120 /* Skip flags entry */
121 " addq $8, %rsp\n"
122 " popfq\n"
123 #else /* CONFIG_X86_32 */
124 " pushf\n"
125 SAVE_REGS_STRING
126 " movl %esp, %edx\n"
127 ".global optprobe_template_val\n"
128 "optprobe_template_val:\n"
129 ASM_NOP5
130 ".global optprobe_template_call\n"
131 "optprobe_template_call:\n"
132 ASM_NOP5
133 RESTORE_REGS_STRING
134 " addl $4, %esp\n" /* skip cs */
135 " popf\n"
136 #endif
137 ".global optprobe_template_end\n"
138 "optprobe_template_end:\n"
139 ".type optprobe_template_func, @function\n"
140 ".size optprobe_template_func, .-optprobe_template_func\n");
141
142 void optprobe_template_func(void);
143 STACK_FRAME_NON_STANDARD(optprobe_template_func);
144 NOKPROBE_SYMBOL(optprobe_template_func);
145 NOKPROBE_SYMBOL(optprobe_template_entry);
146 NOKPROBE_SYMBOL(optprobe_template_val);
147 NOKPROBE_SYMBOL(optprobe_template_call);
148 NOKPROBE_SYMBOL(optprobe_template_end);
149
150 #define TMPL_MOVE_IDX \
151 ((long)&optprobe_template_val - (long)&optprobe_template_entry)
152 #define TMPL_CALL_IDX \
153 ((long)&optprobe_template_call - (long)&optprobe_template_entry)
154 #define TMPL_END_IDX \
155 ((long)&optprobe_template_end - (long)&optprobe_template_entry)
156
157 #define INT3_SIZE sizeof(kprobe_opcode_t)
158
159 /* Optimized kprobe call back function: called from optinsn */
160 static void
optimized_callback(struct optimized_kprobe * op,struct pt_regs * regs)161 optimized_callback(struct optimized_kprobe *op, struct pt_regs *regs)
162 {
163 struct kprobe_ctlblk *kcb = get_kprobe_ctlblk();
164 unsigned long flags;
165
166 /* This is possible if op is under delayed unoptimizing */
167 if (kprobe_disabled(&op->kp))
168 return;
169
170 local_irq_save(flags);
171 if (kprobe_running()) {
172 kprobes_inc_nmissed_count(&op->kp);
173 } else {
174 /* Save skipped registers */
175 #ifdef CONFIG_X86_64
176 regs->cs = __KERNEL_CS;
177 #else
178 regs->cs = __KERNEL_CS | get_kernel_rpl();
179 regs->gs = 0;
180 #endif
181 regs->ip = (unsigned long)op->kp.addr + INT3_SIZE;
182 regs->orig_ax = ~0UL;
183
184 __this_cpu_write(current_kprobe, &op->kp);
185 kcb->kprobe_status = KPROBE_HIT_ACTIVE;
186 opt_pre_handler(&op->kp, regs);
187 __this_cpu_write(current_kprobe, NULL);
188 }
189 local_irq_restore(flags);
190 }
191 NOKPROBE_SYMBOL(optimized_callback);
192
copy_optimized_instructions(u8 * dest,u8 * src)193 static int copy_optimized_instructions(u8 *dest, u8 *src)
194 {
195 struct insn insn;
196 int len = 0, ret;
197
198 while (len < RELATIVEJUMP_SIZE) {
199 ret = __copy_instruction(dest + len, src + len, &insn);
200 if (!ret || !can_boost(&insn, src + len))
201 return -EINVAL;
202 len += ret;
203 }
204 /* Check whether the address range is reserved */
205 if (ftrace_text_reserved(src, src + len - 1) ||
206 alternatives_text_reserved(src, src + len - 1) ||
207 jump_label_text_reserved(src, src + len - 1))
208 return -EBUSY;
209
210 return len;
211 }
212
213 /* Check whether insn is indirect jump */
__insn_is_indirect_jump(struct insn * insn)214 static int __insn_is_indirect_jump(struct insn *insn)
215 {
216 return ((insn->opcode.bytes[0] == 0xff &&
217 (X86_MODRM_REG(insn->modrm.value) & 6) == 4) || /* Jump */
218 insn->opcode.bytes[0] == 0xea); /* Segment based jump */
219 }
220
221 /* Check whether insn jumps into specified address range */
insn_jump_into_range(struct insn * insn,unsigned long start,int len)222 static int insn_jump_into_range(struct insn *insn, unsigned long start, int len)
223 {
224 unsigned long target = 0;
225
226 switch (insn->opcode.bytes[0]) {
227 case 0xe0: /* loopne */
228 case 0xe1: /* loope */
229 case 0xe2: /* loop */
230 case 0xe3: /* jcxz */
231 case 0xe9: /* near relative jump */
232 case 0xeb: /* short relative jump */
233 break;
234 case 0x0f:
235 if ((insn->opcode.bytes[1] & 0xf0) == 0x80) /* jcc near */
236 break;
237 return 0;
238 default:
239 if ((insn->opcode.bytes[0] & 0xf0) == 0x70) /* jcc short */
240 break;
241 return 0;
242 }
243 target = (unsigned long)insn->next_byte + insn->immediate.value;
244
245 return (start <= target && target <= start + len);
246 }
247
insn_is_indirect_jump(struct insn * insn)248 static int insn_is_indirect_jump(struct insn *insn)
249 {
250 int ret = __insn_is_indirect_jump(insn);
251
252 #ifdef CONFIG_RETPOLINE
253 /*
254 * Jump to x86_indirect_thunk_* is treated as an indirect jump.
255 * Note that even with CONFIG_RETPOLINE=y, the kernel compiled with
256 * older gcc may use indirect jump. So we add this check instead of
257 * replace indirect-jump check.
258 */
259 if (!ret)
260 ret = insn_jump_into_range(insn,
261 (unsigned long)__indirect_thunk_start,
262 (unsigned long)__indirect_thunk_end -
263 (unsigned long)__indirect_thunk_start);
264 #endif
265 return ret;
266 }
267
268 /* Decode whole function to ensure any instructions don't jump into target */
can_optimize(unsigned long paddr)269 static int can_optimize(unsigned long paddr)
270 {
271 unsigned long addr, size = 0, offset = 0;
272 struct insn insn;
273 kprobe_opcode_t buf[MAX_INSN_SIZE];
274
275 /* Lookup symbol including addr */
276 if (!kallsyms_lookup_size_offset(paddr, &size, &offset))
277 return 0;
278
279 /*
280 * Do not optimize in the entry code due to the unstable
281 * stack handling and registers setup.
282 */
283 if (((paddr >= (unsigned long)__entry_text_start) &&
284 (paddr < (unsigned long)__entry_text_end)) ||
285 ((paddr >= (unsigned long)__irqentry_text_start) &&
286 (paddr < (unsigned long)__irqentry_text_end)))
287 return 0;
288
289 /* Check there is enough space for a relative jump. */
290 if (size - offset < RELATIVEJUMP_SIZE)
291 return 0;
292
293 /* Decode instructions */
294 addr = paddr - offset;
295 while (addr < paddr - offset + size) { /* Decode until function end */
296 unsigned long recovered_insn;
297 if (search_exception_tables(addr))
298 /*
299 * Since some fixup code will jumps into this function,
300 * we can't optimize kprobe in this function.
301 */
302 return 0;
303 recovered_insn = recover_probed_instruction(buf, addr);
304 if (!recovered_insn)
305 return 0;
306 kernel_insn_init(&insn, (void *)recovered_insn, MAX_INSN_SIZE);
307 insn_get_length(&insn);
308 /* Another subsystem puts a breakpoint */
309 if (insn.opcode.bytes[0] == BREAKPOINT_INSTRUCTION)
310 return 0;
311 /* Recover address */
312 insn.kaddr = (void *)addr;
313 insn.next_byte = (void *)(addr + insn.length);
314 /* Check any instructions don't jump into target */
315 if (insn_is_indirect_jump(&insn) ||
316 insn_jump_into_range(&insn, paddr + INT3_SIZE,
317 RELATIVE_ADDR_SIZE))
318 return 0;
319 addr += insn.length;
320 }
321
322 return 1;
323 }
324
325 /* Check optimized_kprobe can actually be optimized. */
arch_check_optimized_kprobe(struct optimized_kprobe * op)326 int arch_check_optimized_kprobe(struct optimized_kprobe *op)
327 {
328 int i;
329 struct kprobe *p;
330
331 for (i = 1; i < op->optinsn.size; i++) {
332 p = get_kprobe(op->kp.addr + i);
333 if (p && !kprobe_disabled(p))
334 return -EEXIST;
335 }
336
337 return 0;
338 }
339
340 /* Check the addr is within the optimized instructions. */
arch_within_optimized_kprobe(struct optimized_kprobe * op,unsigned long addr)341 int arch_within_optimized_kprobe(struct optimized_kprobe *op,
342 unsigned long addr)
343 {
344 return ((unsigned long)op->kp.addr <= addr &&
345 (unsigned long)op->kp.addr + op->optinsn.size > addr);
346 }
347
348 /* Free optimized instruction slot */
349 static
__arch_remove_optimized_kprobe(struct optimized_kprobe * op,int dirty)350 void __arch_remove_optimized_kprobe(struct optimized_kprobe *op, int dirty)
351 {
352 if (op->optinsn.insn) {
353 free_optinsn_slot(op->optinsn.insn, dirty);
354 op->optinsn.insn = NULL;
355 op->optinsn.size = 0;
356 }
357 }
358
arch_remove_optimized_kprobe(struct optimized_kprobe * op)359 void arch_remove_optimized_kprobe(struct optimized_kprobe *op)
360 {
361 __arch_remove_optimized_kprobe(op, 1);
362 }
363
364 /*
365 * Copy replacing target instructions
366 * Target instructions MUST be relocatable (checked inside)
367 * This is called when new aggr(opt)probe is allocated or reused.
368 */
arch_prepare_optimized_kprobe(struct optimized_kprobe * op,struct kprobe * __unused)369 int arch_prepare_optimized_kprobe(struct optimized_kprobe *op,
370 struct kprobe *__unused)
371 {
372 u8 *buf;
373 int ret;
374 long rel;
375
376 if (!can_optimize((unsigned long)op->kp.addr))
377 return -EILSEQ;
378
379 op->optinsn.insn = get_optinsn_slot();
380 if (!op->optinsn.insn)
381 return -ENOMEM;
382
383 /*
384 * Verify if the address gap is in 2GB range, because this uses
385 * a relative jump.
386 */
387 rel = (long)op->optinsn.insn - (long)op->kp.addr + RELATIVEJUMP_SIZE;
388 if (abs(rel) > 0x7fffffff) {
389 __arch_remove_optimized_kprobe(op, 0);
390 return -ERANGE;
391 }
392
393 buf = (u8 *)op->optinsn.insn;
394 set_memory_rw((unsigned long)buf & PAGE_MASK, 1);
395
396 /* Copy instructions into the out-of-line buffer */
397 ret = copy_optimized_instructions(buf + TMPL_END_IDX, op->kp.addr);
398 if (ret < 0) {
399 __arch_remove_optimized_kprobe(op, 0);
400 return ret;
401 }
402 op->optinsn.size = ret;
403
404 /* Copy arch-dep-instance from template */
405 memcpy(buf, &optprobe_template_entry, TMPL_END_IDX);
406
407 /* Set probe information */
408 synthesize_set_arg1(buf + TMPL_MOVE_IDX, (unsigned long)op);
409
410 /* Set probe function call */
411 synthesize_relcall(buf + TMPL_CALL_IDX, optimized_callback);
412
413 /* Set returning jmp instruction at the tail of out-of-line buffer */
414 synthesize_reljump(buf + TMPL_END_IDX + op->optinsn.size,
415 (u8 *)op->kp.addr + op->optinsn.size);
416
417 set_memory_ro((unsigned long)buf & PAGE_MASK, 1);
418
419 flush_icache_range((unsigned long) buf,
420 (unsigned long) buf + TMPL_END_IDX +
421 op->optinsn.size + RELATIVEJUMP_SIZE);
422 return 0;
423 }
424
425 /*
426 * Replace breakpoints (int3) with relative jumps.
427 * Caller must call with locking kprobe_mutex and text_mutex.
428 */
arch_optimize_kprobes(struct list_head * oplist)429 void arch_optimize_kprobes(struct list_head *oplist)
430 {
431 struct optimized_kprobe *op, *tmp;
432 u8 insn_buf[RELATIVEJUMP_SIZE];
433
434 list_for_each_entry_safe(op, tmp, oplist, list) {
435 s32 rel = (s32)((long)op->optinsn.insn -
436 ((long)op->kp.addr + RELATIVEJUMP_SIZE));
437
438 WARN_ON(kprobe_disabled(&op->kp));
439
440 /* Backup instructions which will be replaced by jump address */
441 memcpy(op->optinsn.copied_insn, op->kp.addr + INT3_SIZE,
442 RELATIVE_ADDR_SIZE);
443
444 insn_buf[0] = RELATIVEJUMP_OPCODE;
445 *(s32 *)(&insn_buf[1]) = rel;
446
447 text_poke_bp(op->kp.addr, insn_buf, RELATIVEJUMP_SIZE,
448 op->optinsn.insn);
449
450 list_del_init(&op->list);
451 }
452 }
453
454 /* Replace a relative jump with a breakpoint (int3). */
arch_unoptimize_kprobe(struct optimized_kprobe * op)455 void arch_unoptimize_kprobe(struct optimized_kprobe *op)
456 {
457 u8 insn_buf[RELATIVEJUMP_SIZE];
458
459 /* Set int3 to first byte for kprobes */
460 insn_buf[0] = BREAKPOINT_INSTRUCTION;
461 memcpy(insn_buf + 1, op->optinsn.copied_insn, RELATIVE_ADDR_SIZE);
462 text_poke_bp(op->kp.addr, insn_buf, RELATIVEJUMP_SIZE,
463 op->optinsn.insn);
464 }
465
466 /*
467 * Recover original instructions and breakpoints from relative jumps.
468 * Caller must call with locking kprobe_mutex.
469 */
arch_unoptimize_kprobes(struct list_head * oplist,struct list_head * done_list)470 extern void arch_unoptimize_kprobes(struct list_head *oplist,
471 struct list_head *done_list)
472 {
473 struct optimized_kprobe *op, *tmp;
474
475 list_for_each_entry_safe(op, tmp, oplist, list) {
476 arch_unoptimize_kprobe(op);
477 list_move(&op->list, done_list);
478 }
479 }
480
setup_detour_execution(struct kprobe * p,struct pt_regs * regs,int reenter)481 int setup_detour_execution(struct kprobe *p, struct pt_regs *regs, int reenter)
482 {
483 struct optimized_kprobe *op;
484
485 if (p->flags & KPROBE_FLAG_OPTIMIZED) {
486 /* This kprobe is really able to run optimized path. */
487 op = container_of(p, struct optimized_kprobe, kp);
488 /* Detour through copied instructions */
489 regs->ip = (unsigned long)op->optinsn.insn + TMPL_END_IDX;
490 if (!reenter)
491 reset_current_kprobe();
492 preempt_enable_no_resched();
493 return 1;
494 }
495 return 0;
496 }
497 NOKPROBE_SYMBOL(setup_detour_execution);
498