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/module.h>
28 #include <linux/kdebug.h>
29 #include <linux/kallsyms.h>
30 #include <linux/ftrace.h>
32 #include <asm/cacheflush.h>
34 #include <asm/pgtable.h>
35 #include <asm/uaccess.h>
36 #include <asm/alternative.h>
38 #include <asm/debugreg.h>
39 #include <asm/nospec-branch.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 memcpy(buf, (void *)addr, MAX_INSN_SIZE * sizeof(kprobe_opcode_t));
69 if (addr == (unsigned long)kp->addr) {
71 memcpy(buf + 1, op->optinsn.copied_insn, RELATIVE_ADDR_SIZE);
73 offs = addr - (unsigned long)kp->addr - 1;
74 memcpy(buf, op->optinsn.copied_insn + offs, RELATIVE_ADDR_SIZE - offs);
77 return (unsigned long)buf;
80 /* Insert a move instruction which sets a pointer to eax/rdi (1st arg). */
81 static void synthesize_set_arg1(kprobe_opcode_t *addr, unsigned long val)
89 *(unsigned long *)addr = val;
93 ".global optprobe_template_entry\n"
94 "optprobe_template_entry:\n"
96 /* We don't bother saving the ss register */
101 ".global optprobe_template_val\n"
102 "optprobe_template_val:\n"
105 ".global optprobe_template_call\n"
106 "optprobe_template_call:\n"
108 /* Move flags to rsp */
109 " movq 144(%rsp), %rdx\n"
110 " movq %rdx, 152(%rsp)\n"
112 /* Skip flags entry */
115 #else /* CONFIG_X86_32 */
119 ".global optprobe_template_val\n"
120 "optprobe_template_val:\n"
122 ".global optprobe_template_call\n"
123 "optprobe_template_call:\n"
126 " addl $4, %esp\n" /* skip cs */
129 ".global optprobe_template_end\n"
130 "optprobe_template_end:\n");
132 #define TMPL_MOVE_IDX \
133 ((long)&optprobe_template_val - (long)&optprobe_template_entry)
134 #define TMPL_CALL_IDX \
135 ((long)&optprobe_template_call - (long)&optprobe_template_entry)
136 #define TMPL_END_IDX \
137 ((long)&optprobe_template_end - (long)&optprobe_template_entry)
139 #define INT3_SIZE sizeof(kprobe_opcode_t)
141 /* Optimized kprobe call back function: called from optinsn */
143 optimized_callback(struct optimized_kprobe *op, struct pt_regs *regs)
145 struct kprobe_ctlblk *kcb = get_kprobe_ctlblk();
148 /* This is possible if op is under delayed unoptimizing */
149 if (kprobe_disabled(&op->kp))
152 local_irq_save(flags);
153 if (kprobe_running()) {
154 kprobes_inc_nmissed_count(&op->kp);
156 /* Save skipped registers */
158 regs->cs = __KERNEL_CS;
160 regs->cs = __KERNEL_CS | get_kernel_rpl();
163 regs->ip = (unsigned long)op->kp.addr + INT3_SIZE;
164 regs->orig_ax = ~0UL;
166 __this_cpu_write(current_kprobe, &op->kp);
167 kcb->kprobe_status = KPROBE_HIT_ACTIVE;
168 opt_pre_handler(&op->kp, regs);
169 __this_cpu_write(current_kprobe, NULL);
171 local_irq_restore(flags);
173 NOKPROBE_SYMBOL(optimized_callback);
175 static int copy_optimized_instructions(u8 *dest, u8 *src)
179 while (len < RELATIVEJUMP_SIZE) {
180 ret = __copy_instruction(dest + len, src + len);
181 if (!ret || !can_boost(dest + len, src + len))
185 /* Check whether the address range is reserved */
186 if (ftrace_text_reserved(src, src + len - 1) ||
187 alternatives_text_reserved(src, src + len - 1) ||
188 jump_label_text_reserved(src, src + len - 1))
194 /* Check whether insn is indirect jump */
195 static int __insn_is_indirect_jump(struct insn *insn)
197 return ((insn->opcode.bytes[0] == 0xff &&
198 (X86_MODRM_REG(insn->modrm.value) & 6) == 4) || /* Jump */
199 insn->opcode.bytes[0] == 0xea); /* Segment based jump */
202 /* Check whether insn jumps into specified address range */
203 static int insn_jump_into_range(struct insn *insn, unsigned long start, int len)
205 unsigned long target = 0;
207 switch (insn->opcode.bytes[0]) {
208 case 0xe0: /* loopne */
209 case 0xe1: /* loope */
210 case 0xe2: /* loop */
211 case 0xe3: /* jcxz */
212 case 0xe9: /* near relative jump */
213 case 0xeb: /* short relative jump */
216 if ((insn->opcode.bytes[1] & 0xf0) == 0x80) /* jcc near */
220 if ((insn->opcode.bytes[0] & 0xf0) == 0x70) /* jcc short */
224 target = (unsigned long)insn->next_byte + insn->immediate.value;
226 return (start <= target && target <= start + len);
229 static int insn_is_indirect_jump(struct insn *insn)
231 int ret = __insn_is_indirect_jump(insn);
233 #ifdef CONFIG_RETPOLINE
235 * Jump to x86_indirect_thunk_* is treated as an indirect jump.
236 * Note that even with CONFIG_RETPOLINE=y, the kernel compiled with
237 * older gcc may use indirect jump. So we add this check instead of
238 * replace indirect-jump check.
241 ret = insn_jump_into_range(insn,
242 (unsigned long)__indirect_thunk_start,
243 (unsigned long)__indirect_thunk_end -
244 (unsigned long)__indirect_thunk_start);
249 /* Decode whole function to ensure any instructions don't jump into target */
250 static int can_optimize(unsigned long paddr)
252 unsigned long addr, size = 0, offset = 0;
254 kprobe_opcode_t buf[MAX_INSN_SIZE];
256 /* Lookup symbol including addr */
257 if (!kallsyms_lookup_size_offset(paddr, &size, &offset))
261 * Do not optimize in the entry code due to the unstable
264 if ((paddr >= (unsigned long)__entry_text_start) &&
265 (paddr < (unsigned long)__entry_text_end))
268 /* Check there is enough space for a relative jump. */
269 if (size - offset < RELATIVEJUMP_SIZE)
272 /* Decode instructions */
273 addr = paddr - offset;
274 while (addr < paddr - offset + size) { /* Decode until function end */
275 unsigned long recovered_insn;
276 if (search_exception_tables(addr))
278 * Since some fixup code will jumps into this function,
279 * we can't optimize kprobe in this function.
282 recovered_insn = recover_probed_instruction(buf, addr);
285 kernel_insn_init(&insn, (void *)recovered_insn, MAX_INSN_SIZE);
286 insn_get_length(&insn);
287 /* Another subsystem puts a breakpoint */
288 if (insn.opcode.bytes[0] == BREAKPOINT_INSTRUCTION)
290 /* Recover address */
291 insn.kaddr = (void *)addr;
292 insn.next_byte = (void *)(addr + insn.length);
293 /* Check any instructions don't jump into target */
294 if (insn_is_indirect_jump(&insn) ||
295 insn_jump_into_range(&insn, paddr + INT3_SIZE,
304 /* Check optimized_kprobe can actually be optimized. */
305 int arch_check_optimized_kprobe(struct optimized_kprobe *op)
310 for (i = 1; i < op->optinsn.size; i++) {
311 p = get_kprobe(op->kp.addr + i);
312 if (p && !kprobe_disabled(p))
319 /* Check the addr is within the optimized instructions. */
320 int arch_within_optimized_kprobe(struct optimized_kprobe *op,
323 return ((unsigned long)op->kp.addr <= addr &&
324 (unsigned long)op->kp.addr + op->optinsn.size > addr);
327 /* Free optimized instruction slot */
329 void __arch_remove_optimized_kprobe(struct optimized_kprobe *op, int dirty)
331 if (op->optinsn.insn) {
332 free_optinsn_slot(op->optinsn.insn, dirty);
333 op->optinsn.insn = NULL;
334 op->optinsn.size = 0;
338 void arch_remove_optimized_kprobe(struct optimized_kprobe *op)
340 __arch_remove_optimized_kprobe(op, 1);
344 * Copy replacing target instructions
345 * Target instructions MUST be relocatable (checked inside)
346 * This is called when new aggr(opt)probe is allocated or reused.
348 int arch_prepare_optimized_kprobe(struct optimized_kprobe *op,
349 struct kprobe *__unused)
355 if (!can_optimize((unsigned long)op->kp.addr))
358 op->optinsn.insn = get_optinsn_slot();
359 if (!op->optinsn.insn)
363 * Verify if the address gap is in 2GB range, because this uses
366 rel = (long)op->optinsn.insn - (long)op->kp.addr + RELATIVEJUMP_SIZE;
367 if (abs(rel) > 0x7fffffff) {
368 __arch_remove_optimized_kprobe(op, 0);
372 buf = (u8 *)op->optinsn.insn;
373 set_memory_rw((unsigned long)buf & PAGE_MASK, 1);
375 /* Copy instructions into the out-of-line buffer */
376 ret = copy_optimized_instructions(buf + TMPL_END_IDX, op->kp.addr);
378 __arch_remove_optimized_kprobe(op, 0);
381 op->optinsn.size = ret;
383 /* Copy arch-dep-instance from template */
384 memcpy(buf, &optprobe_template_entry, TMPL_END_IDX);
386 /* Set probe information */
387 synthesize_set_arg1(buf + TMPL_MOVE_IDX, (unsigned long)op);
389 /* Set probe function call */
390 synthesize_relcall(buf + TMPL_CALL_IDX, optimized_callback);
392 /* Set returning jmp instruction at the tail of out-of-line buffer */
393 synthesize_reljump(buf + TMPL_END_IDX + op->optinsn.size,
394 (u8 *)op->kp.addr + op->optinsn.size);
396 set_memory_ro((unsigned long)buf & PAGE_MASK, 1);
398 flush_icache_range((unsigned long) buf,
399 (unsigned long) buf + TMPL_END_IDX +
400 op->optinsn.size + RELATIVEJUMP_SIZE);
405 * Replace breakpoints (int3) with relative jumps.
406 * Caller must call with locking kprobe_mutex and text_mutex.
408 void arch_optimize_kprobes(struct list_head *oplist)
410 struct optimized_kprobe *op, *tmp;
411 u8 insn_buf[RELATIVEJUMP_SIZE];
413 list_for_each_entry_safe(op, tmp, oplist, list) {
414 s32 rel = (s32)((long)op->optinsn.insn -
415 ((long)op->kp.addr + RELATIVEJUMP_SIZE));
417 WARN_ON(kprobe_disabled(&op->kp));
419 /* Backup instructions which will be replaced by jump address */
420 memcpy(op->optinsn.copied_insn, op->kp.addr + INT3_SIZE,
423 insn_buf[0] = RELATIVEJUMP_OPCODE;
424 *(s32 *)(&insn_buf[1]) = rel;
426 text_poke_bp(op->kp.addr, insn_buf, RELATIVEJUMP_SIZE,
429 list_del_init(&op->list);
433 /* Replace a relative jump with a breakpoint (int3). */
434 void arch_unoptimize_kprobe(struct optimized_kprobe *op)
436 u8 insn_buf[RELATIVEJUMP_SIZE];
438 /* Set int3 to first byte for kprobes */
439 insn_buf[0] = BREAKPOINT_INSTRUCTION;
440 memcpy(insn_buf + 1, op->optinsn.copied_insn, RELATIVE_ADDR_SIZE);
441 text_poke_bp(op->kp.addr, insn_buf, RELATIVEJUMP_SIZE,
446 * Recover original instructions and breakpoints from relative jumps.
447 * Caller must call with locking kprobe_mutex.
449 extern void arch_unoptimize_kprobes(struct list_head *oplist,
450 struct list_head *done_list)
452 struct optimized_kprobe *op, *tmp;
454 list_for_each_entry_safe(op, tmp, oplist, list) {
455 arch_unoptimize_kprobe(op);
456 list_move(&op->list, done_list);
460 int setup_detour_execution(struct kprobe *p, struct pt_regs *regs, int reenter)
462 struct optimized_kprobe *op;
464 if (p->flags & KPROBE_FLAG_OPTIMIZED) {
465 /* This kprobe is really able to run optimized path. */
466 op = container_of(p, struct optimized_kprobe, kp);
467 /* Detour through copied instructions */
468 regs->ip = (unsigned long)op->optinsn.insn + TMPL_END_IDX;
470 reset_current_kprobe();
471 preempt_enable_no_resched();
476 NOKPROBE_SYMBOL(setup_detour_execution);