2 * Kernel Probes Jump Optimization (Optprobes)
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.
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.
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.
18 * Copyright (C) IBM Corporation, 2002, 2004
19 * Copyright (C) Hitachi Ltd., 2012
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>
33 #include <asm/text-patching.h>
34 #include <asm/cacheflush.h>
36 #include <asm/pgtable.h>
37 #include <asm/uaccess.h>
38 #include <asm/alternative.h>
40 #include <asm/debugreg.h>
41 #include <asm/nospec-branch.h>
42 #include <asm/sections.h>
46 unsigned long __recover_optprobed_insn(kprobe_opcode_t *buf, unsigned long addr)
48 struct optimized_kprobe *op;
53 for (i = 0; i < RELATIVEJUMP_SIZE; i++) {
54 kp = get_kprobe((void *)addr - i);
55 /* This function only handles jump-optimized kprobe */
56 if (kp && kprobe_optimized(kp)) {
57 op = container_of(kp, struct optimized_kprobe, kp);
58 /* If op->list is not empty, op is under optimizing */
59 if (list_empty(&op->list))
67 * If the kprobe can be optimized, original bytes which can be
68 * overwritten by jump destination address. In this case, original
69 * bytes must be recovered from op->optinsn.copied_insn buffer.
71 memcpy(buf, (void *)addr, MAX_INSN_SIZE * sizeof(kprobe_opcode_t));
72 if (addr == (unsigned long)kp->addr) {
74 memcpy(buf + 1, op->optinsn.copied_insn, RELATIVE_ADDR_SIZE);
76 offs = addr - (unsigned long)kp->addr - 1;
77 memcpy(buf, op->optinsn.copied_insn + offs, RELATIVE_ADDR_SIZE - offs);
80 return (unsigned long)buf;
83 /* Insert a move instruction which sets a pointer to eax/rdi (1st arg). */
84 static void synthesize_set_arg1(kprobe_opcode_t *addr, unsigned long val)
92 *(unsigned long *)addr = val;
96 "optprobe_template_func:\n"
97 ".global optprobe_template_entry\n"
98 "optprobe_template_entry:\n"
100 /* We don't bother saving the ss register */
105 ".global optprobe_template_val\n"
106 "optprobe_template_val:\n"
109 ".global optprobe_template_call\n"
110 "optprobe_template_call:\n"
112 /* Move flags to rsp */
113 " movq 144(%rsp), %rdx\n"
114 " movq %rdx, 152(%rsp)\n"
116 /* Skip flags entry */
119 #else /* CONFIG_X86_32 */
123 ".global optprobe_template_val\n"
124 "optprobe_template_val:\n"
126 ".global optprobe_template_call\n"
127 "optprobe_template_call:\n"
130 " addl $4, %esp\n" /* skip cs */
133 ".global optprobe_template_end\n"
134 "optprobe_template_end:\n"
135 ".type optprobe_template_func, @function\n"
136 ".size optprobe_template_func, .-optprobe_template_func\n");
138 void optprobe_template_func(void);
139 STACK_FRAME_NON_STANDARD(optprobe_template_func);
141 #define TMPL_MOVE_IDX \
142 ((long)&optprobe_template_val - (long)&optprobe_template_entry)
143 #define TMPL_CALL_IDX \
144 ((long)&optprobe_template_call - (long)&optprobe_template_entry)
145 #define TMPL_END_IDX \
146 ((long)&optprobe_template_end - (long)&optprobe_template_entry)
148 #define INT3_SIZE sizeof(kprobe_opcode_t)
150 /* Optimized kprobe call back function: called from optinsn */
152 optimized_callback(struct optimized_kprobe *op, struct pt_regs *regs)
154 struct kprobe_ctlblk *kcb = get_kprobe_ctlblk();
157 /* This is possible if op is under delayed unoptimizing */
158 if (kprobe_disabled(&op->kp))
161 local_irq_save(flags);
162 if (kprobe_running()) {
163 kprobes_inc_nmissed_count(&op->kp);
165 /* Save skipped registers */
167 regs->cs = __KERNEL_CS;
169 regs->cs = __KERNEL_CS | get_kernel_rpl();
172 regs->ip = (unsigned long)op->kp.addr + INT3_SIZE;
173 regs->orig_ax = ~0UL;
175 __this_cpu_write(current_kprobe, &op->kp);
176 kcb->kprobe_status = KPROBE_HIT_ACTIVE;
177 opt_pre_handler(&op->kp, regs);
178 __this_cpu_write(current_kprobe, NULL);
180 local_irq_restore(flags);
182 NOKPROBE_SYMBOL(optimized_callback);
184 static int copy_optimized_instructions(u8 *dest, u8 *src)
188 while (len < RELATIVEJUMP_SIZE) {
189 ret = __copy_instruction(dest + len, src + len);
190 if (!ret || !can_boost(dest + len, src + len))
194 /* Check whether the address range is reserved */
195 if (ftrace_text_reserved(src, src + len - 1) ||
196 alternatives_text_reserved(src, src + len - 1) ||
197 jump_label_text_reserved(src, src + len - 1))
203 /* Check whether insn is indirect jump */
204 static int __insn_is_indirect_jump(struct insn *insn)
206 return ((insn->opcode.bytes[0] == 0xff &&
207 (X86_MODRM_REG(insn->modrm.value) & 6) == 4) || /* Jump */
208 insn->opcode.bytes[0] == 0xea); /* Segment based jump */
211 /* Check whether insn jumps into specified address range */
212 static int insn_jump_into_range(struct insn *insn, unsigned long start, int len)
214 unsigned long target = 0;
216 switch (insn->opcode.bytes[0]) {
217 case 0xe0: /* loopne */
218 case 0xe1: /* loope */
219 case 0xe2: /* loop */
220 case 0xe3: /* jcxz */
221 case 0xe9: /* near relative jump */
222 case 0xeb: /* short relative jump */
225 if ((insn->opcode.bytes[1] & 0xf0) == 0x80) /* jcc near */
229 if ((insn->opcode.bytes[0] & 0xf0) == 0x70) /* jcc short */
233 target = (unsigned long)insn->next_byte + insn->immediate.value;
235 return (start <= target && target <= start + len);
238 static int insn_is_indirect_jump(struct insn *insn)
240 int ret = __insn_is_indirect_jump(insn);
242 #ifdef CONFIG_RETPOLINE
244 * Jump to x86_indirect_thunk_* is treated as an indirect jump.
245 * Note that even with CONFIG_RETPOLINE=y, the kernel compiled with
246 * older gcc may use indirect jump. So we add this check instead of
247 * replace indirect-jump check.
250 ret = insn_jump_into_range(insn,
251 (unsigned long)__indirect_thunk_start,
252 (unsigned long)__indirect_thunk_end -
253 (unsigned long)__indirect_thunk_start);
258 /* Decode whole function to ensure any instructions don't jump into target */
259 static int can_optimize(unsigned long paddr)
261 unsigned long addr, size = 0, offset = 0;
263 kprobe_opcode_t buf[MAX_INSN_SIZE];
265 /* Lookup symbol including addr */
266 if (!kallsyms_lookup_size_offset(paddr, &size, &offset))
270 * Do not optimize in the entry code due to the unstable
273 if ((paddr >= (unsigned long)__entry_text_start) &&
274 (paddr < (unsigned long)__entry_text_end))
277 /* Check there is enough space for a relative jump. */
278 if (size - offset < RELATIVEJUMP_SIZE)
281 /* Decode instructions */
282 addr = paddr - offset;
283 while (addr < paddr - offset + size) { /* Decode until function end */
284 unsigned long recovered_insn;
285 if (search_exception_tables(addr))
287 * Since some fixup code will jumps into this function,
288 * we can't optimize kprobe in this function.
291 recovered_insn = recover_probed_instruction(buf, addr);
294 kernel_insn_init(&insn, (void *)recovered_insn, MAX_INSN_SIZE);
295 insn_get_length(&insn);
296 /* Another subsystem puts a breakpoint */
297 if (insn.opcode.bytes[0] == BREAKPOINT_INSTRUCTION)
299 /* Recover address */
300 insn.kaddr = (void *)addr;
301 insn.next_byte = (void *)(addr + insn.length);
302 /* Check any instructions don't jump into target */
303 if (insn_is_indirect_jump(&insn) ||
304 insn_jump_into_range(&insn, paddr + INT3_SIZE,
313 /* Check optimized_kprobe can actually be optimized. */
314 int arch_check_optimized_kprobe(struct optimized_kprobe *op)
319 for (i = 1; i < op->optinsn.size; i++) {
320 p = get_kprobe(op->kp.addr + i);
321 if (p && !kprobe_disabled(p))
328 /* Check the addr is within the optimized instructions. */
329 int arch_within_optimized_kprobe(struct optimized_kprobe *op,
332 return ((unsigned long)op->kp.addr <= addr &&
333 (unsigned long)op->kp.addr + op->optinsn.size > addr);
336 /* Free optimized instruction slot */
338 void __arch_remove_optimized_kprobe(struct optimized_kprobe *op, int dirty)
340 if (op->optinsn.insn) {
341 free_optinsn_slot(op->optinsn.insn, dirty);
342 op->optinsn.insn = NULL;
343 op->optinsn.size = 0;
347 void arch_remove_optimized_kprobe(struct optimized_kprobe *op)
349 __arch_remove_optimized_kprobe(op, 1);
353 * Copy replacing target instructions
354 * Target instructions MUST be relocatable (checked inside)
355 * This is called when new aggr(opt)probe is allocated or reused.
357 int arch_prepare_optimized_kprobe(struct optimized_kprobe *op,
358 struct kprobe *__unused)
364 if (!can_optimize((unsigned long)op->kp.addr))
367 op->optinsn.insn = get_optinsn_slot();
368 if (!op->optinsn.insn)
372 * Verify if the address gap is in 2GB range, because this uses
375 rel = (long)op->optinsn.insn - (long)op->kp.addr + RELATIVEJUMP_SIZE;
376 if (abs(rel) > 0x7fffffff) {
377 __arch_remove_optimized_kprobe(op, 0);
381 buf = (u8 *)op->optinsn.insn;
382 set_memory_rw((unsigned long)buf & PAGE_MASK, 1);
384 /* Copy instructions into the out-of-line buffer */
385 ret = copy_optimized_instructions(buf + TMPL_END_IDX, op->kp.addr);
387 __arch_remove_optimized_kprobe(op, 0);
390 op->optinsn.size = ret;
392 /* Copy arch-dep-instance from template */
393 memcpy(buf, &optprobe_template_entry, TMPL_END_IDX);
395 /* Set probe information */
396 synthesize_set_arg1(buf + TMPL_MOVE_IDX, (unsigned long)op);
398 /* Set probe function call */
399 synthesize_relcall(buf + TMPL_CALL_IDX, optimized_callback);
401 /* Set returning jmp instruction at the tail of out-of-line buffer */
402 synthesize_reljump(buf + TMPL_END_IDX + op->optinsn.size,
403 (u8 *)op->kp.addr + op->optinsn.size);
405 set_memory_ro((unsigned long)buf & PAGE_MASK, 1);
407 flush_icache_range((unsigned long) buf,
408 (unsigned long) buf + TMPL_END_IDX +
409 op->optinsn.size + RELATIVEJUMP_SIZE);
414 * Replace breakpoints (int3) with relative jumps.
415 * Caller must call with locking kprobe_mutex and text_mutex.
417 void arch_optimize_kprobes(struct list_head *oplist)
419 struct optimized_kprobe *op, *tmp;
420 u8 insn_buf[RELATIVEJUMP_SIZE];
422 list_for_each_entry_safe(op, tmp, oplist, list) {
423 s32 rel = (s32)((long)op->optinsn.insn -
424 ((long)op->kp.addr + RELATIVEJUMP_SIZE));
426 WARN_ON(kprobe_disabled(&op->kp));
428 /* Backup instructions which will be replaced by jump address */
429 memcpy(op->optinsn.copied_insn, op->kp.addr + INT3_SIZE,
432 insn_buf[0] = RELATIVEJUMP_OPCODE;
433 *(s32 *)(&insn_buf[1]) = rel;
435 text_poke_bp(op->kp.addr, insn_buf, RELATIVEJUMP_SIZE,
438 list_del_init(&op->list);
442 /* Replace a relative jump with a breakpoint (int3). */
443 void arch_unoptimize_kprobe(struct optimized_kprobe *op)
445 u8 insn_buf[RELATIVEJUMP_SIZE];
447 /* Set int3 to first byte for kprobes */
448 insn_buf[0] = BREAKPOINT_INSTRUCTION;
449 memcpy(insn_buf + 1, op->optinsn.copied_insn, RELATIVE_ADDR_SIZE);
450 text_poke_bp(op->kp.addr, insn_buf, RELATIVEJUMP_SIZE,
455 * Recover original instructions and breakpoints from relative jumps.
456 * Caller must call with locking kprobe_mutex.
458 extern void arch_unoptimize_kprobes(struct list_head *oplist,
459 struct list_head *done_list)
461 struct optimized_kprobe *op, *tmp;
463 list_for_each_entry_safe(op, tmp, oplist, list) {
464 arch_unoptimize_kprobe(op);
465 list_move(&op->list, done_list);
469 int setup_detour_execution(struct kprobe *p, struct pt_regs *regs, int reenter)
471 struct optimized_kprobe *op;
473 if (p->flags & KPROBE_FLAG_OPTIMIZED) {
474 /* This kprobe is really able to run optimized path. */
475 op = container_of(p, struct optimized_kprobe, kp);
476 /* Detour through copied instructions */
477 regs->ip = (unsigned long)op->optinsn.insn + TMPL_END_IDX;
479 reset_current_kprobe();
480 preempt_enable_no_resched();
485 NOKPROBE_SYMBOL(setup_detour_execution);