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>
32 #include <asm/text-patching.h>
33 #include <asm/cacheflush.h>
35 #include <asm/pgtable.h>
36 #include <linux/uaccess.h>
37 #include <asm/alternative.h>
39 #include <asm/debugreg.h>
43 unsigned long __recover_optprobed_insn(kprobe_opcode_t *buf, unsigned long addr)
45 struct optimized_kprobe *op;
50 for (i = 0; i < RELATIVEJUMP_SIZE; i++) {
51 kp = get_kprobe((void *)addr - i);
52 /* This function only handles jump-optimized kprobe */
53 if (kp && kprobe_optimized(kp)) {
54 op = container_of(kp, struct optimized_kprobe, kp);
55 /* If op->list is not empty, op is under optimizing */
56 if (list_empty(&op->list))
64 * If the kprobe can be optimized, original bytes which can be
65 * overwritten by jump destination address. In this case, original
66 * bytes must be recovered from op->optinsn.copied_insn buffer.
68 if (probe_kernel_read(buf, (void *)addr,
69 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 ".global optprobe_template_entry\n"
97 "optprobe_template_entry:\n"
99 /* We don't bother saving the ss register */
104 ".global optprobe_template_val\n"
105 "optprobe_template_val:\n"
108 ".global optprobe_template_call\n"
109 "optprobe_template_call:\n"
111 /* Move flags to rsp */
112 " movq 144(%rsp), %rdx\n"
113 " movq %rdx, 152(%rsp)\n"
115 /* Skip flags entry */
118 #else /* CONFIG_X86_32 */
122 ".global optprobe_template_val\n"
123 "optprobe_template_val:\n"
125 ".global optprobe_template_call\n"
126 "optprobe_template_call:\n"
129 " addl $4, %esp\n" /* skip cs */
132 ".global optprobe_template_end\n"
133 "optprobe_template_end:\n");
135 #define TMPL_MOVE_IDX \
136 ((long)&optprobe_template_val - (long)&optprobe_template_entry)
137 #define TMPL_CALL_IDX \
138 ((long)&optprobe_template_call - (long)&optprobe_template_entry)
139 #define TMPL_END_IDX \
140 ((long)&optprobe_template_end - (long)&optprobe_template_entry)
142 #define INT3_SIZE sizeof(kprobe_opcode_t)
144 /* Optimized kprobe call back function: called from optinsn */
146 optimized_callback(struct optimized_kprobe *op, struct pt_regs *regs)
148 struct kprobe_ctlblk *kcb = get_kprobe_ctlblk();
151 /* This is possible if op is under delayed unoptimizing */
152 if (kprobe_disabled(&op->kp))
155 local_irq_save(flags);
156 if (kprobe_running()) {
157 kprobes_inc_nmissed_count(&op->kp);
159 /* Save skipped registers */
161 regs->cs = __KERNEL_CS;
163 regs->cs = __KERNEL_CS | get_kernel_rpl();
166 regs->ip = (unsigned long)op->kp.addr + INT3_SIZE;
167 regs->orig_ax = ~0UL;
169 __this_cpu_write(current_kprobe, &op->kp);
170 kcb->kprobe_status = KPROBE_HIT_ACTIVE;
171 opt_pre_handler(&op->kp, regs);
172 __this_cpu_write(current_kprobe, NULL);
174 local_irq_restore(flags);
176 NOKPROBE_SYMBOL(optimized_callback);
178 static int copy_optimized_instructions(u8 *dest, u8 *src)
183 while (len < RELATIVEJUMP_SIZE) {
184 ret = __copy_instruction(dest + len, src + len, &insn);
185 if (!ret || !can_boost(&insn, src + len))
189 /* Check whether the address range is reserved */
190 if (ftrace_text_reserved(src, src + len - 1) ||
191 alternatives_text_reserved(src, src + len - 1) ||
192 jump_label_text_reserved(src, src + len - 1))
198 /* Check whether insn is indirect jump */
199 static int insn_is_indirect_jump(struct insn *insn)
201 return ((insn->opcode.bytes[0] == 0xff &&
202 (X86_MODRM_REG(insn->modrm.value) & 6) == 4) || /* Jump */
203 insn->opcode.bytes[0] == 0xea); /* Segment based jump */
206 /* Check whether insn jumps into specified address range */
207 static int insn_jump_into_range(struct insn *insn, unsigned long start, int len)
209 unsigned long target = 0;
211 switch (insn->opcode.bytes[0]) {
212 case 0xe0: /* loopne */
213 case 0xe1: /* loope */
214 case 0xe2: /* loop */
215 case 0xe3: /* jcxz */
216 case 0xe9: /* near relative jump */
217 case 0xeb: /* short relative jump */
220 if ((insn->opcode.bytes[1] & 0xf0) == 0x80) /* jcc near */
224 if ((insn->opcode.bytes[0] & 0xf0) == 0x70) /* jcc short */
228 target = (unsigned long)insn->next_byte + insn->immediate.value;
230 return (start <= target && target <= start + len);
233 /* Decode whole function to ensure any instructions don't jump into target */
234 static int can_optimize(unsigned long paddr)
236 unsigned long addr, size = 0, offset = 0;
238 kprobe_opcode_t buf[MAX_INSN_SIZE];
240 /* Lookup symbol including addr */
241 if (!kallsyms_lookup_size_offset(paddr, &size, &offset))
245 * Do not optimize in the entry code due to the unstable
248 if ((paddr >= (unsigned long)__entry_text_start) &&
249 (paddr < (unsigned long)__entry_text_end))
252 /* Check there is enough space for a relative jump. */
253 if (size - offset < RELATIVEJUMP_SIZE)
256 /* Decode instructions */
257 addr = paddr - offset;
258 while (addr < paddr - offset + size) { /* Decode until function end */
259 unsigned long recovered_insn;
260 if (search_exception_tables(addr))
262 * Since some fixup code will jumps into this function,
263 * we can't optimize kprobe in this function.
266 recovered_insn = recover_probed_instruction(buf, addr);
269 kernel_insn_init(&insn, (void *)recovered_insn, MAX_INSN_SIZE);
270 insn_get_length(&insn);
271 /* Another subsystem puts a breakpoint */
272 if (insn.opcode.bytes[0] == BREAKPOINT_INSTRUCTION)
274 /* Recover address */
275 insn.kaddr = (void *)addr;
276 insn.next_byte = (void *)(addr + insn.length);
277 /* Check any instructions don't jump into target */
278 if (insn_is_indirect_jump(&insn) ||
279 insn_jump_into_range(&insn, paddr + INT3_SIZE,
288 /* Check optimized_kprobe can actually be optimized. */
289 int arch_check_optimized_kprobe(struct optimized_kprobe *op)
294 for (i = 1; i < op->optinsn.size; i++) {
295 p = get_kprobe(op->kp.addr + i);
296 if (p && !kprobe_disabled(p))
303 /* Check the addr is within the optimized instructions. */
304 int arch_within_optimized_kprobe(struct optimized_kprobe *op,
307 return ((unsigned long)op->kp.addr <= addr &&
308 (unsigned long)op->kp.addr + op->optinsn.size > addr);
311 /* Free optimized instruction slot */
313 void __arch_remove_optimized_kprobe(struct optimized_kprobe *op, int dirty)
315 if (op->optinsn.insn) {
316 free_optinsn_slot(op->optinsn.insn, dirty);
317 op->optinsn.insn = NULL;
318 op->optinsn.size = 0;
322 void arch_remove_optimized_kprobe(struct optimized_kprobe *op)
324 __arch_remove_optimized_kprobe(op, 1);
328 * Copy replacing target instructions
329 * Target instructions MUST be relocatable (checked inside)
330 * This is called when new aggr(opt)probe is allocated or reused.
332 int arch_prepare_optimized_kprobe(struct optimized_kprobe *op,
333 struct kprobe *__unused)
339 if (!can_optimize((unsigned long)op->kp.addr))
342 op->optinsn.insn = get_optinsn_slot();
343 if (!op->optinsn.insn)
347 * Verify if the address gap is in 2GB range, because this uses
350 rel = (long)op->optinsn.insn - (long)op->kp.addr + RELATIVEJUMP_SIZE;
351 if (abs(rel) > 0x7fffffff) {
352 __arch_remove_optimized_kprobe(op, 0);
356 buf = (u8 *)op->optinsn.insn;
357 set_memory_rw((unsigned long)buf & PAGE_MASK, 1);
359 /* Copy instructions into the out-of-line buffer */
360 ret = copy_optimized_instructions(buf + TMPL_END_IDX, op->kp.addr);
362 __arch_remove_optimized_kprobe(op, 0);
365 op->optinsn.size = ret;
367 /* Copy arch-dep-instance from template */
368 memcpy(buf, &optprobe_template_entry, TMPL_END_IDX);
370 /* Set probe information */
371 synthesize_set_arg1(buf + TMPL_MOVE_IDX, (unsigned long)op);
373 /* Set probe function call */
374 synthesize_relcall(buf + TMPL_CALL_IDX, optimized_callback);
376 /* Set returning jmp instruction at the tail of out-of-line buffer */
377 synthesize_reljump(buf + TMPL_END_IDX + op->optinsn.size,
378 (u8 *)op->kp.addr + op->optinsn.size);
380 set_memory_ro((unsigned long)buf & PAGE_MASK, 1);
382 flush_icache_range((unsigned long) buf,
383 (unsigned long) buf + TMPL_END_IDX +
384 op->optinsn.size + RELATIVEJUMP_SIZE);
389 * Replace breakpoints (int3) with relative jumps.
390 * Caller must call with locking kprobe_mutex and text_mutex.
392 void arch_optimize_kprobes(struct list_head *oplist)
394 struct optimized_kprobe *op, *tmp;
395 u8 insn_buf[RELATIVEJUMP_SIZE];
397 list_for_each_entry_safe(op, tmp, oplist, list) {
398 s32 rel = (s32)((long)op->optinsn.insn -
399 ((long)op->kp.addr + RELATIVEJUMP_SIZE));
401 WARN_ON(kprobe_disabled(&op->kp));
403 /* Backup instructions which will be replaced by jump address */
404 memcpy(op->optinsn.copied_insn, op->kp.addr + INT3_SIZE,
407 insn_buf[0] = RELATIVEJUMP_OPCODE;
408 *(s32 *)(&insn_buf[1]) = rel;
410 text_poke_bp(op->kp.addr, insn_buf, RELATIVEJUMP_SIZE,
413 list_del_init(&op->list);
417 /* Replace a relative jump with a breakpoint (int3). */
418 void arch_unoptimize_kprobe(struct optimized_kprobe *op)
420 u8 insn_buf[RELATIVEJUMP_SIZE];
422 /* Set int3 to first byte for kprobes */
423 insn_buf[0] = BREAKPOINT_INSTRUCTION;
424 memcpy(insn_buf + 1, op->optinsn.copied_insn, RELATIVE_ADDR_SIZE);
425 text_poke_bp(op->kp.addr, insn_buf, RELATIVEJUMP_SIZE,
430 * Recover original instructions and breakpoints from relative jumps.
431 * Caller must call with locking kprobe_mutex.
433 extern void arch_unoptimize_kprobes(struct list_head *oplist,
434 struct list_head *done_list)
436 struct optimized_kprobe *op, *tmp;
438 list_for_each_entry_safe(op, tmp, oplist, list) {
439 arch_unoptimize_kprobe(op);
440 list_move(&op->list, done_list);
444 int setup_detour_execution(struct kprobe *p, struct pt_regs *regs, int reenter)
446 struct optimized_kprobe *op;
448 if (p->flags & KPROBE_FLAG_OPTIMIZED) {
449 /* This kprobe is really able to run optimized path. */
450 op = container_of(p, struct optimized_kprobe, kp);
451 /* Detour through copied instructions */
452 regs->ip = (unsigned long)op->optinsn.insn + TMPL_END_IDX;
454 reset_current_kprobe();
455 preempt_enable_no_resched();
460 NOKPROBE_SYMBOL(setup_detour_execution);