]>
Commit | Line | Data |
---|---|---|
1 | /* | |
2 | * Kernel Probes (KProbes) | |
3 | * kernel/kprobes.c | |
4 | * | |
5 | * This program is free software; you can redistribute it and/or modify | |
6 | * it under the terms of the GNU General Public License as published by | |
7 | * the Free Software Foundation; either version 2 of the License, or | |
8 | * (at your option) any later version. | |
9 | * | |
10 | * This program is distributed in the hope that it will be useful, | |
11 | * but WITHOUT ANY WARRANTY; without even the implied warranty of | |
12 | * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the | |
13 | * GNU General Public License for more details. | |
14 | * | |
15 | * You should have received a copy of the GNU General Public License | |
16 | * along with this program; if not, write to the Free Software | |
17 | * Foundation, Inc., 59 Temple Place - Suite 330, Boston, MA 02111-1307, USA. | |
18 | * | |
19 | * Copyright (C) IBM Corporation, 2002, 2004 | |
20 | * | |
21 | * 2002-Oct Created by Vamsi Krishna S <vamsi_krishna@in.ibm.com> Kernel | |
22 | * Probes initial implementation (includes suggestions from | |
23 | * Rusty Russell). | |
24 | * 2004-Aug Updated by Prasanna S Panchamukhi <prasanna@in.ibm.com> with | |
25 | * hlists and exceptions notifier as suggested by Andi Kleen. | |
26 | * 2004-July Suparna Bhattacharya <suparna@in.ibm.com> added jumper probes | |
27 | * interface to access function arguments. | |
28 | * 2004-Sep Prasanna S Panchamukhi <prasanna@in.ibm.com> Changed Kprobes | |
29 | * exceptions notifier to be first on the priority list. | |
30 | * 2005-May Hien Nguyen <hien@us.ibm.com>, Jim Keniston | |
31 | * <jkenisto@us.ibm.com> and Prasanna S Panchamukhi | |
32 | * <prasanna@in.ibm.com> added function-return probes. | |
33 | */ | |
34 | #include <linux/kprobes.h> | |
35 | #include <linux/hash.h> | |
36 | #include <linux/init.h> | |
37 | #include <linux/slab.h> | |
38 | #include <linux/stddef.h> | |
39 | #include <linux/module.h> | |
40 | #include <linux/moduleloader.h> | |
41 | #include <linux/kallsyms.h> | |
42 | #include <linux/freezer.h> | |
43 | #include <linux/seq_file.h> | |
44 | #include <linux/debugfs.h> | |
45 | #include <linux/sysctl.h> | |
46 | #include <linux/kdebug.h> | |
47 | #include <linux/memory.h> | |
48 | #include <linux/ftrace.h> | |
49 | #include <linux/cpu.h> | |
50 | ||
51 | #include <asm-generic/sections.h> | |
52 | #include <asm/cacheflush.h> | |
53 | #include <asm/errno.h> | |
54 | #include <asm/uaccess.h> | |
55 | ||
56 | #define KPROBE_HASH_BITS 6 | |
57 | #define KPROBE_TABLE_SIZE (1 << KPROBE_HASH_BITS) | |
58 | ||
59 | ||
60 | /* | |
61 | * Some oddball architectures like 64bit powerpc have function descriptors | |
62 | * so this must be overridable. | |
63 | */ | |
64 | #ifndef kprobe_lookup_name | |
65 | #define kprobe_lookup_name(name, addr) \ | |
66 | addr = ((kprobe_opcode_t *)(kallsyms_lookup_name(name))) | |
67 | #endif | |
68 | ||
69 | static int kprobes_initialized; | |
70 | static struct hlist_head kprobe_table[KPROBE_TABLE_SIZE]; | |
71 | static struct hlist_head kretprobe_inst_table[KPROBE_TABLE_SIZE]; | |
72 | ||
73 | /* NOTE: change this value only with kprobe_mutex held */ | |
74 | static bool kprobes_all_disarmed; | |
75 | ||
76 | static DEFINE_MUTEX(kprobe_mutex); /* Protects kprobe_table */ | |
77 | static DEFINE_PER_CPU(struct kprobe *, kprobe_instance) = NULL; | |
78 | static struct { | |
79 | spinlock_t lock ____cacheline_aligned_in_smp; | |
80 | } kretprobe_table_locks[KPROBE_TABLE_SIZE]; | |
81 | ||
82 | static spinlock_t *kretprobe_table_lock_ptr(unsigned long hash) | |
83 | { | |
84 | return &(kretprobe_table_locks[hash].lock); | |
85 | } | |
86 | ||
87 | /* | |
88 | * Normally, functions that we'd want to prohibit kprobes in, are marked | |
89 | * __kprobes. But, there are cases where such functions already belong to | |
90 | * a different section (__sched for preempt_schedule) | |
91 | * | |
92 | * For such cases, we now have a blacklist | |
93 | */ | |
94 | static struct kprobe_blackpoint kprobe_blacklist[] = { | |
95 | {"preempt_schedule",}, | |
96 | {"native_get_debugreg",}, | |
97 | {"irq_entries_start",}, | |
98 | {"common_interrupt",}, | |
99 | {"mcount",}, /* mcount can be called from everywhere */ | |
100 | {NULL} /* Terminator */ | |
101 | }; | |
102 | ||
103 | #ifdef __ARCH_WANT_KPROBES_INSN_SLOT | |
104 | /* | |
105 | * kprobe->ainsn.insn points to the copy of the instruction to be | |
106 | * single-stepped. x86_64, POWER4 and above have no-exec support and | |
107 | * stepping on the instruction on a vmalloced/kmalloced/data page | |
108 | * is a recipe for disaster | |
109 | */ | |
110 | struct kprobe_insn_page { | |
111 | struct list_head list; | |
112 | kprobe_opcode_t *insns; /* Page of instruction slots */ | |
113 | int nused; | |
114 | int ngarbage; | |
115 | char slot_used[]; | |
116 | }; | |
117 | ||
118 | #define KPROBE_INSN_PAGE_SIZE(slots) \ | |
119 | (offsetof(struct kprobe_insn_page, slot_used) + \ | |
120 | (sizeof(char) * (slots))) | |
121 | ||
122 | struct kprobe_insn_cache { | |
123 | struct list_head pages; /* list of kprobe_insn_page */ | |
124 | size_t insn_size; /* size of instruction slot */ | |
125 | int nr_garbage; | |
126 | }; | |
127 | ||
128 | static int slots_per_page(struct kprobe_insn_cache *c) | |
129 | { | |
130 | return PAGE_SIZE/(c->insn_size * sizeof(kprobe_opcode_t)); | |
131 | } | |
132 | ||
133 | enum kprobe_slot_state { | |
134 | SLOT_CLEAN = 0, | |
135 | SLOT_DIRTY = 1, | |
136 | SLOT_USED = 2, | |
137 | }; | |
138 | ||
139 | static DEFINE_MUTEX(kprobe_insn_mutex); /* Protects kprobe_insn_slots */ | |
140 | static struct kprobe_insn_cache kprobe_insn_slots = { | |
141 | .pages = LIST_HEAD_INIT(kprobe_insn_slots.pages), | |
142 | .insn_size = MAX_INSN_SIZE, | |
143 | .nr_garbage = 0, | |
144 | }; | |
145 | static int __kprobes collect_garbage_slots(struct kprobe_insn_cache *c); | |
146 | ||
147 | /** | |
148 | * __get_insn_slot() - Find a slot on an executable page for an instruction. | |
149 | * We allocate an executable page if there's no room on existing ones. | |
150 | */ | |
151 | static kprobe_opcode_t __kprobes *__get_insn_slot(struct kprobe_insn_cache *c) | |
152 | { | |
153 | struct kprobe_insn_page *kip; | |
154 | ||
155 | retry: | |
156 | list_for_each_entry(kip, &c->pages, list) { | |
157 | if (kip->nused < slots_per_page(c)) { | |
158 | int i; | |
159 | for (i = 0; i < slots_per_page(c); i++) { | |
160 | if (kip->slot_used[i] == SLOT_CLEAN) { | |
161 | kip->slot_used[i] = SLOT_USED; | |
162 | kip->nused++; | |
163 | return kip->insns + (i * c->insn_size); | |
164 | } | |
165 | } | |
166 | /* kip->nused is broken. Fix it. */ | |
167 | kip->nused = slots_per_page(c); | |
168 | WARN_ON(1); | |
169 | } | |
170 | } | |
171 | ||
172 | /* If there are any garbage slots, collect it and try again. */ | |
173 | if (c->nr_garbage && collect_garbage_slots(c) == 0) | |
174 | goto retry; | |
175 | ||
176 | /* All out of space. Need to allocate a new page. */ | |
177 | kip = kmalloc(KPROBE_INSN_PAGE_SIZE(slots_per_page(c)), GFP_KERNEL); | |
178 | if (!kip) | |
179 | return NULL; | |
180 | ||
181 | /* | |
182 | * Use module_alloc so this page is within +/- 2GB of where the | |
183 | * kernel image and loaded module images reside. This is required | |
184 | * so x86_64 can correctly handle the %rip-relative fixups. | |
185 | */ | |
186 | kip->insns = module_alloc(PAGE_SIZE); | |
187 | if (!kip->insns) { | |
188 | kfree(kip); | |
189 | return NULL; | |
190 | } | |
191 | INIT_LIST_HEAD(&kip->list); | |
192 | memset(kip->slot_used, SLOT_CLEAN, slots_per_page(c)); | |
193 | kip->slot_used[0] = SLOT_USED; | |
194 | kip->nused = 1; | |
195 | kip->ngarbage = 0; | |
196 | list_add(&kip->list, &c->pages); | |
197 | return kip->insns; | |
198 | } | |
199 | ||
200 | ||
201 | kprobe_opcode_t __kprobes *get_insn_slot(void) | |
202 | { | |
203 | kprobe_opcode_t *ret = NULL; | |
204 | ||
205 | mutex_lock(&kprobe_insn_mutex); | |
206 | ret = __get_insn_slot(&kprobe_insn_slots); | |
207 | mutex_unlock(&kprobe_insn_mutex); | |
208 | ||
209 | return ret; | |
210 | } | |
211 | ||
212 | /* Return 1 if all garbages are collected, otherwise 0. */ | |
213 | static int __kprobes collect_one_slot(struct kprobe_insn_page *kip, int idx) | |
214 | { | |
215 | kip->slot_used[idx] = SLOT_CLEAN; | |
216 | kip->nused--; | |
217 | if (kip->nused == 0) { | |
218 | /* | |
219 | * Page is no longer in use. Free it unless | |
220 | * it's the last one. We keep the last one | |
221 | * so as not to have to set it up again the | |
222 | * next time somebody inserts a probe. | |
223 | */ | |
224 | if (!list_is_singular(&kip->list)) { | |
225 | list_del(&kip->list); | |
226 | module_free(NULL, kip->insns); | |
227 | kfree(kip); | |
228 | } | |
229 | return 1; | |
230 | } | |
231 | return 0; | |
232 | } | |
233 | ||
234 | static int __kprobes collect_garbage_slots(struct kprobe_insn_cache *c) | |
235 | { | |
236 | struct kprobe_insn_page *kip, *next; | |
237 | ||
238 | /* Ensure no-one is interrupted on the garbages */ | |
239 | synchronize_sched(); | |
240 | ||
241 | list_for_each_entry_safe(kip, next, &c->pages, list) { | |
242 | int i; | |
243 | if (kip->ngarbage == 0) | |
244 | continue; | |
245 | kip->ngarbage = 0; /* we will collect all garbages */ | |
246 | for (i = 0; i < slots_per_page(c); i++) { | |
247 | if (kip->slot_used[i] == SLOT_DIRTY && | |
248 | collect_one_slot(kip, i)) | |
249 | break; | |
250 | } | |
251 | } | |
252 | c->nr_garbage = 0; | |
253 | return 0; | |
254 | } | |
255 | ||
256 | static void __kprobes __free_insn_slot(struct kprobe_insn_cache *c, | |
257 | kprobe_opcode_t *slot, int dirty) | |
258 | { | |
259 | struct kprobe_insn_page *kip; | |
260 | ||
261 | list_for_each_entry(kip, &c->pages, list) { | |
262 | long idx = ((long)slot - (long)kip->insns) / | |
263 | (c->insn_size * sizeof(kprobe_opcode_t)); | |
264 | if (idx >= 0 && idx < slots_per_page(c)) { | |
265 | WARN_ON(kip->slot_used[idx] != SLOT_USED); | |
266 | if (dirty) { | |
267 | kip->slot_used[idx] = SLOT_DIRTY; | |
268 | kip->ngarbage++; | |
269 | if (++c->nr_garbage > slots_per_page(c)) | |
270 | collect_garbage_slots(c); | |
271 | } else | |
272 | collect_one_slot(kip, idx); | |
273 | return; | |
274 | } | |
275 | } | |
276 | /* Could not free this slot. */ | |
277 | WARN_ON(1); | |
278 | } | |
279 | ||
280 | void __kprobes free_insn_slot(kprobe_opcode_t * slot, int dirty) | |
281 | { | |
282 | mutex_lock(&kprobe_insn_mutex); | |
283 | __free_insn_slot(&kprobe_insn_slots, slot, dirty); | |
284 | mutex_unlock(&kprobe_insn_mutex); | |
285 | } | |
286 | #ifdef CONFIG_OPTPROBES | |
287 | /* For optimized_kprobe buffer */ | |
288 | static DEFINE_MUTEX(kprobe_optinsn_mutex); /* Protects kprobe_optinsn_slots */ | |
289 | static struct kprobe_insn_cache kprobe_optinsn_slots = { | |
290 | .pages = LIST_HEAD_INIT(kprobe_optinsn_slots.pages), | |
291 | /* .insn_size is initialized later */ | |
292 | .nr_garbage = 0, | |
293 | }; | |
294 | /* Get a slot for optimized_kprobe buffer */ | |
295 | kprobe_opcode_t __kprobes *get_optinsn_slot(void) | |
296 | { | |
297 | kprobe_opcode_t *ret = NULL; | |
298 | ||
299 | mutex_lock(&kprobe_optinsn_mutex); | |
300 | ret = __get_insn_slot(&kprobe_optinsn_slots); | |
301 | mutex_unlock(&kprobe_optinsn_mutex); | |
302 | ||
303 | return ret; | |
304 | } | |
305 | ||
306 | void __kprobes free_optinsn_slot(kprobe_opcode_t * slot, int dirty) | |
307 | { | |
308 | mutex_lock(&kprobe_optinsn_mutex); | |
309 | __free_insn_slot(&kprobe_optinsn_slots, slot, dirty); | |
310 | mutex_unlock(&kprobe_optinsn_mutex); | |
311 | } | |
312 | #endif | |
313 | #endif | |
314 | ||
315 | /* We have preemption disabled.. so it is safe to use __ versions */ | |
316 | static inline void set_kprobe_instance(struct kprobe *kp) | |
317 | { | |
318 | __get_cpu_var(kprobe_instance) = kp; | |
319 | } | |
320 | ||
321 | static inline void reset_kprobe_instance(void) | |
322 | { | |
323 | __get_cpu_var(kprobe_instance) = NULL; | |
324 | } | |
325 | ||
326 | /* | |
327 | * This routine is called either: | |
328 | * - under the kprobe_mutex - during kprobe_[un]register() | |
329 | * OR | |
330 | * - with preemption disabled - from arch/xxx/kernel/kprobes.c | |
331 | */ | |
332 | struct kprobe __kprobes *get_kprobe(void *addr) | |
333 | { | |
334 | struct hlist_head *head; | |
335 | struct hlist_node *node; | |
336 | struct kprobe *p; | |
337 | ||
338 | head = &kprobe_table[hash_ptr(addr, KPROBE_HASH_BITS)]; | |
339 | hlist_for_each_entry_rcu(p, node, head, hlist) { | |
340 | if (p->addr == addr) | |
341 | return p; | |
342 | } | |
343 | ||
344 | return NULL; | |
345 | } | |
346 | ||
347 | static int __kprobes aggr_pre_handler(struct kprobe *p, struct pt_regs *regs); | |
348 | ||
349 | /* Return true if the kprobe is an aggregator */ | |
350 | static inline int kprobe_aggrprobe(struct kprobe *p) | |
351 | { | |
352 | return p->pre_handler == aggr_pre_handler; | |
353 | } | |
354 | ||
355 | /* | |
356 | * Keep all fields in the kprobe consistent | |
357 | */ | |
358 | static inline void copy_kprobe(struct kprobe *old_p, struct kprobe *p) | |
359 | { | |
360 | memcpy(&p->opcode, &old_p->opcode, sizeof(kprobe_opcode_t)); | |
361 | memcpy(&p->ainsn, &old_p->ainsn, sizeof(struct arch_specific_insn)); | |
362 | } | |
363 | ||
364 | #ifdef CONFIG_OPTPROBES | |
365 | /* NOTE: change this value only with kprobe_mutex held */ | |
366 | static bool kprobes_allow_optimization; | |
367 | ||
368 | /* | |
369 | * Call all pre_handler on the list, but ignores its return value. | |
370 | * This must be called from arch-dep optimized caller. | |
371 | */ | |
372 | void __kprobes opt_pre_handler(struct kprobe *p, struct pt_regs *regs) | |
373 | { | |
374 | struct kprobe *kp; | |
375 | ||
376 | list_for_each_entry_rcu(kp, &p->list, list) { | |
377 | if (kp->pre_handler && likely(!kprobe_disabled(kp))) { | |
378 | set_kprobe_instance(kp); | |
379 | kp->pre_handler(kp, regs); | |
380 | } | |
381 | reset_kprobe_instance(); | |
382 | } | |
383 | } | |
384 | ||
385 | /* Return true(!0) if the kprobe is ready for optimization. */ | |
386 | static inline int kprobe_optready(struct kprobe *p) | |
387 | { | |
388 | struct optimized_kprobe *op; | |
389 | ||
390 | if (kprobe_aggrprobe(p)) { | |
391 | op = container_of(p, struct optimized_kprobe, kp); | |
392 | return arch_prepared_optinsn(&op->optinsn); | |
393 | } | |
394 | ||
395 | return 0; | |
396 | } | |
397 | ||
398 | /* | |
399 | * Return an optimized kprobe whose optimizing code replaces | |
400 | * instructions including addr (exclude breakpoint). | |
401 | */ | |
402 | static struct kprobe *__kprobes get_optimized_kprobe(unsigned long addr) | |
403 | { | |
404 | int i; | |
405 | struct kprobe *p = NULL; | |
406 | struct optimized_kprobe *op; | |
407 | ||
408 | /* Don't check i == 0, since that is a breakpoint case. */ | |
409 | for (i = 1; !p && i < MAX_OPTIMIZED_LENGTH; i++) | |
410 | p = get_kprobe((void *)(addr - i)); | |
411 | ||
412 | if (p && kprobe_optready(p)) { | |
413 | op = container_of(p, struct optimized_kprobe, kp); | |
414 | if (arch_within_optimized_kprobe(op, addr)) | |
415 | return p; | |
416 | } | |
417 | ||
418 | return NULL; | |
419 | } | |
420 | ||
421 | /* Optimization staging list, protected by kprobe_mutex */ | |
422 | static LIST_HEAD(optimizing_list); | |
423 | ||
424 | static void kprobe_optimizer(struct work_struct *work); | |
425 | static DECLARE_DELAYED_WORK(optimizing_work, kprobe_optimizer); | |
426 | #define OPTIMIZE_DELAY 5 | |
427 | ||
428 | /* Kprobe jump optimizer */ | |
429 | static __kprobes void kprobe_optimizer(struct work_struct *work) | |
430 | { | |
431 | struct optimized_kprobe *op, *tmp; | |
432 | ||
433 | /* Lock modules while optimizing kprobes */ | |
434 | mutex_lock(&module_mutex); | |
435 | mutex_lock(&kprobe_mutex); | |
436 | if (kprobes_all_disarmed || !kprobes_allow_optimization) | |
437 | goto end; | |
438 | ||
439 | /* | |
440 | * Wait for quiesence period to ensure all running interrupts | |
441 | * are done. Because optprobe may modify multiple instructions | |
442 | * there is a chance that Nth instruction is interrupted. In that | |
443 | * case, running interrupt can return to 2nd-Nth byte of jump | |
444 | * instruction. This wait is for avoiding it. | |
445 | */ | |
446 | synchronize_sched(); | |
447 | ||
448 | /* | |
449 | * The optimization/unoptimization refers online_cpus via | |
450 | * stop_machine() and cpu-hotplug modifies online_cpus. | |
451 | * And same time, text_mutex will be held in cpu-hotplug and here. | |
452 | * This combination can cause a deadlock (cpu-hotplug try to lock | |
453 | * text_mutex but stop_machine can not be done because online_cpus | |
454 | * has been changed) | |
455 | * To avoid this deadlock, we need to call get_online_cpus() | |
456 | * for preventing cpu-hotplug outside of text_mutex locking. | |
457 | */ | |
458 | get_online_cpus(); | |
459 | mutex_lock(&text_mutex); | |
460 | list_for_each_entry_safe(op, tmp, &optimizing_list, list) { | |
461 | WARN_ON(kprobe_disabled(&op->kp)); | |
462 | if (arch_optimize_kprobe(op) < 0) | |
463 | op->kp.flags &= ~KPROBE_FLAG_OPTIMIZED; | |
464 | list_del_init(&op->list); | |
465 | } | |
466 | mutex_unlock(&text_mutex); | |
467 | put_online_cpus(); | |
468 | end: | |
469 | mutex_unlock(&kprobe_mutex); | |
470 | mutex_unlock(&module_mutex); | |
471 | } | |
472 | ||
473 | /* Optimize kprobe if p is ready to be optimized */ | |
474 | static __kprobes void optimize_kprobe(struct kprobe *p) | |
475 | { | |
476 | struct optimized_kprobe *op; | |
477 | ||
478 | /* Check if the kprobe is disabled or not ready for optimization. */ | |
479 | if (!kprobe_optready(p) || !kprobes_allow_optimization || | |
480 | (kprobe_disabled(p) || kprobes_all_disarmed)) | |
481 | return; | |
482 | ||
483 | /* Both of break_handler and post_handler are not supported. */ | |
484 | if (p->break_handler || p->post_handler) | |
485 | return; | |
486 | ||
487 | op = container_of(p, struct optimized_kprobe, kp); | |
488 | ||
489 | /* Check there is no other kprobes at the optimized instructions */ | |
490 | if (arch_check_optimized_kprobe(op) < 0) | |
491 | return; | |
492 | ||
493 | /* Check if it is already optimized. */ | |
494 | if (op->kp.flags & KPROBE_FLAG_OPTIMIZED) | |
495 | return; | |
496 | ||
497 | op->kp.flags |= KPROBE_FLAG_OPTIMIZED; | |
498 | list_add(&op->list, &optimizing_list); | |
499 | if (!delayed_work_pending(&optimizing_work)) | |
500 | schedule_delayed_work(&optimizing_work, OPTIMIZE_DELAY); | |
501 | } | |
502 | ||
503 | /* Unoptimize a kprobe if p is optimized */ | |
504 | static __kprobes void unoptimize_kprobe(struct kprobe *p) | |
505 | { | |
506 | struct optimized_kprobe *op; | |
507 | ||
508 | if ((p->flags & KPROBE_FLAG_OPTIMIZED) && kprobe_aggrprobe(p)) { | |
509 | op = container_of(p, struct optimized_kprobe, kp); | |
510 | if (!list_empty(&op->list)) | |
511 | /* Dequeue from the optimization queue */ | |
512 | list_del_init(&op->list); | |
513 | else | |
514 | /* Replace jump with break */ | |
515 | arch_unoptimize_kprobe(op); | |
516 | op->kp.flags &= ~KPROBE_FLAG_OPTIMIZED; | |
517 | } | |
518 | } | |
519 | ||
520 | /* Remove optimized instructions */ | |
521 | static void __kprobes kill_optimized_kprobe(struct kprobe *p) | |
522 | { | |
523 | struct optimized_kprobe *op; | |
524 | ||
525 | op = container_of(p, struct optimized_kprobe, kp); | |
526 | if (!list_empty(&op->list)) { | |
527 | /* Dequeue from the optimization queue */ | |
528 | list_del_init(&op->list); | |
529 | op->kp.flags &= ~KPROBE_FLAG_OPTIMIZED; | |
530 | } | |
531 | /* Don't unoptimize, because the target code will be freed. */ | |
532 | arch_remove_optimized_kprobe(op); | |
533 | } | |
534 | ||
535 | /* Try to prepare optimized instructions */ | |
536 | static __kprobes void prepare_optimized_kprobe(struct kprobe *p) | |
537 | { | |
538 | struct optimized_kprobe *op; | |
539 | ||
540 | op = container_of(p, struct optimized_kprobe, kp); | |
541 | arch_prepare_optimized_kprobe(op); | |
542 | } | |
543 | ||
544 | /* Free optimized instructions and optimized_kprobe */ | |
545 | static __kprobes void free_aggr_kprobe(struct kprobe *p) | |
546 | { | |
547 | struct optimized_kprobe *op; | |
548 | ||
549 | op = container_of(p, struct optimized_kprobe, kp); | |
550 | arch_remove_optimized_kprobe(op); | |
551 | kfree(op); | |
552 | } | |
553 | ||
554 | /* Allocate new optimized_kprobe and try to prepare optimized instructions */ | |
555 | static __kprobes struct kprobe *alloc_aggr_kprobe(struct kprobe *p) | |
556 | { | |
557 | struct optimized_kprobe *op; | |
558 | ||
559 | op = kzalloc(sizeof(struct optimized_kprobe), GFP_KERNEL); | |
560 | if (!op) | |
561 | return NULL; | |
562 | ||
563 | INIT_LIST_HEAD(&op->list); | |
564 | op->kp.addr = p->addr; | |
565 | arch_prepare_optimized_kprobe(op); | |
566 | ||
567 | return &op->kp; | |
568 | } | |
569 | ||
570 | static void __kprobes init_aggr_kprobe(struct kprobe *ap, struct kprobe *p); | |
571 | ||
572 | /* | |
573 | * Prepare an optimized_kprobe and optimize it | |
574 | * NOTE: p must be a normal registered kprobe | |
575 | */ | |
576 | static __kprobes void try_to_optimize_kprobe(struct kprobe *p) | |
577 | { | |
578 | struct kprobe *ap; | |
579 | struct optimized_kprobe *op; | |
580 | ||
581 | ap = alloc_aggr_kprobe(p); | |
582 | if (!ap) | |
583 | return; | |
584 | ||
585 | op = container_of(ap, struct optimized_kprobe, kp); | |
586 | if (!arch_prepared_optinsn(&op->optinsn)) { | |
587 | /* If failed to setup optimizing, fallback to kprobe */ | |
588 | free_aggr_kprobe(ap); | |
589 | return; | |
590 | } | |
591 | ||
592 | init_aggr_kprobe(ap, p); | |
593 | optimize_kprobe(ap); | |
594 | } | |
595 | ||
596 | #ifdef CONFIG_SYSCTL | |
597 | static void __kprobes optimize_all_kprobes(void) | |
598 | { | |
599 | struct hlist_head *head; | |
600 | struct hlist_node *node; | |
601 | struct kprobe *p; | |
602 | unsigned int i; | |
603 | ||
604 | /* If optimization is already allowed, just return */ | |
605 | if (kprobes_allow_optimization) | |
606 | return; | |
607 | ||
608 | kprobes_allow_optimization = true; | |
609 | mutex_lock(&text_mutex); | |
610 | for (i = 0; i < KPROBE_TABLE_SIZE; i++) { | |
611 | head = &kprobe_table[i]; | |
612 | hlist_for_each_entry_rcu(p, node, head, hlist) | |
613 | if (!kprobe_disabled(p)) | |
614 | optimize_kprobe(p); | |
615 | } | |
616 | mutex_unlock(&text_mutex); | |
617 | printk(KERN_INFO "Kprobes globally optimized\n"); | |
618 | } | |
619 | ||
620 | static void __kprobes unoptimize_all_kprobes(void) | |
621 | { | |
622 | struct hlist_head *head; | |
623 | struct hlist_node *node; | |
624 | struct kprobe *p; | |
625 | unsigned int i; | |
626 | ||
627 | /* If optimization is already prohibited, just return */ | |
628 | if (!kprobes_allow_optimization) | |
629 | return; | |
630 | ||
631 | kprobes_allow_optimization = false; | |
632 | printk(KERN_INFO "Kprobes globally unoptimized\n"); | |
633 | get_online_cpus(); /* For avoiding text_mutex deadlock */ | |
634 | mutex_lock(&text_mutex); | |
635 | for (i = 0; i < KPROBE_TABLE_SIZE; i++) { | |
636 | head = &kprobe_table[i]; | |
637 | hlist_for_each_entry_rcu(p, node, head, hlist) { | |
638 | if (!kprobe_disabled(p)) | |
639 | unoptimize_kprobe(p); | |
640 | } | |
641 | } | |
642 | ||
643 | mutex_unlock(&text_mutex); | |
644 | put_online_cpus(); | |
645 | /* Allow all currently running kprobes to complete */ | |
646 | synchronize_sched(); | |
647 | } | |
648 | ||
649 | int sysctl_kprobes_optimization; | |
650 | int proc_kprobes_optimization_handler(struct ctl_table *table, int write, | |
651 | void __user *buffer, size_t *length, | |
652 | loff_t *ppos) | |
653 | { | |
654 | int ret; | |
655 | ||
656 | mutex_lock(&kprobe_mutex); | |
657 | sysctl_kprobes_optimization = kprobes_allow_optimization ? 1 : 0; | |
658 | ret = proc_dointvec_minmax(table, write, buffer, length, ppos); | |
659 | ||
660 | if (sysctl_kprobes_optimization) | |
661 | optimize_all_kprobes(); | |
662 | else | |
663 | unoptimize_all_kprobes(); | |
664 | mutex_unlock(&kprobe_mutex); | |
665 | ||
666 | return ret; | |
667 | } | |
668 | #endif /* CONFIG_SYSCTL */ | |
669 | ||
670 | static void __kprobes __arm_kprobe(struct kprobe *p) | |
671 | { | |
672 | struct kprobe *old_p; | |
673 | ||
674 | /* Check collision with other optimized kprobes */ | |
675 | old_p = get_optimized_kprobe((unsigned long)p->addr); | |
676 | if (unlikely(old_p)) | |
677 | unoptimize_kprobe(old_p); /* Fallback to unoptimized kprobe */ | |
678 | ||
679 | arch_arm_kprobe(p); | |
680 | optimize_kprobe(p); /* Try to optimize (add kprobe to a list) */ | |
681 | } | |
682 | ||
683 | static void __kprobes __disarm_kprobe(struct kprobe *p) | |
684 | { | |
685 | struct kprobe *old_p; | |
686 | ||
687 | unoptimize_kprobe(p); /* Try to unoptimize */ | |
688 | arch_disarm_kprobe(p); | |
689 | ||
690 | /* If another kprobe was blocked, optimize it. */ | |
691 | old_p = get_optimized_kprobe((unsigned long)p->addr); | |
692 | if (unlikely(old_p)) | |
693 | optimize_kprobe(old_p); | |
694 | } | |
695 | ||
696 | #else /* !CONFIG_OPTPROBES */ | |
697 | ||
698 | #define optimize_kprobe(p) do {} while (0) | |
699 | #define unoptimize_kprobe(p) do {} while (0) | |
700 | #define kill_optimized_kprobe(p) do {} while (0) | |
701 | #define prepare_optimized_kprobe(p) do {} while (0) | |
702 | #define try_to_optimize_kprobe(p) do {} while (0) | |
703 | #define __arm_kprobe(p) arch_arm_kprobe(p) | |
704 | #define __disarm_kprobe(p) arch_disarm_kprobe(p) | |
705 | ||
706 | static __kprobes void free_aggr_kprobe(struct kprobe *p) | |
707 | { | |
708 | kfree(p); | |
709 | } | |
710 | ||
711 | static __kprobes struct kprobe *alloc_aggr_kprobe(struct kprobe *p) | |
712 | { | |
713 | return kzalloc(sizeof(struct kprobe), GFP_KERNEL); | |
714 | } | |
715 | #endif /* CONFIG_OPTPROBES */ | |
716 | ||
717 | /* Arm a kprobe with text_mutex */ | |
718 | static void __kprobes arm_kprobe(struct kprobe *kp) | |
719 | { | |
720 | /* | |
721 | * Here, since __arm_kprobe() doesn't use stop_machine(), | |
722 | * this doesn't cause deadlock on text_mutex. So, we don't | |
723 | * need get_online_cpus(). | |
724 | */ | |
725 | mutex_lock(&text_mutex); | |
726 | __arm_kprobe(kp); | |
727 | mutex_unlock(&text_mutex); | |
728 | } | |
729 | ||
730 | /* Disarm a kprobe with text_mutex */ | |
731 | static void __kprobes disarm_kprobe(struct kprobe *kp) | |
732 | { | |
733 | get_online_cpus(); /* For avoiding text_mutex deadlock */ | |
734 | mutex_lock(&text_mutex); | |
735 | __disarm_kprobe(kp); | |
736 | mutex_unlock(&text_mutex); | |
737 | put_online_cpus(); | |
738 | } | |
739 | ||
740 | /* | |
741 | * Aggregate handlers for multiple kprobes support - these handlers | |
742 | * take care of invoking the individual kprobe handlers on p->list | |
743 | */ | |
744 | static int __kprobes aggr_pre_handler(struct kprobe *p, struct pt_regs *regs) | |
745 | { | |
746 | struct kprobe *kp; | |
747 | ||
748 | list_for_each_entry_rcu(kp, &p->list, list) { | |
749 | if (kp->pre_handler && likely(!kprobe_disabled(kp))) { | |
750 | set_kprobe_instance(kp); | |
751 | if (kp->pre_handler(kp, regs)) | |
752 | return 1; | |
753 | } | |
754 | reset_kprobe_instance(); | |
755 | } | |
756 | return 0; | |
757 | } | |
758 | ||
759 | static void __kprobes aggr_post_handler(struct kprobe *p, struct pt_regs *regs, | |
760 | unsigned long flags) | |
761 | { | |
762 | struct kprobe *kp; | |
763 | ||
764 | list_for_each_entry_rcu(kp, &p->list, list) { | |
765 | if (kp->post_handler && likely(!kprobe_disabled(kp))) { | |
766 | set_kprobe_instance(kp); | |
767 | kp->post_handler(kp, regs, flags); | |
768 | reset_kprobe_instance(); | |
769 | } | |
770 | } | |
771 | } | |
772 | ||
773 | static int __kprobes aggr_fault_handler(struct kprobe *p, struct pt_regs *regs, | |
774 | int trapnr) | |
775 | { | |
776 | struct kprobe *cur = __get_cpu_var(kprobe_instance); | |
777 | ||
778 | /* | |
779 | * if we faulted "during" the execution of a user specified | |
780 | * probe handler, invoke just that probe's fault handler | |
781 | */ | |
782 | if (cur && cur->fault_handler) { | |
783 | if (cur->fault_handler(cur, regs, trapnr)) | |
784 | return 1; | |
785 | } | |
786 | return 0; | |
787 | } | |
788 | ||
789 | static int __kprobes aggr_break_handler(struct kprobe *p, struct pt_regs *regs) | |
790 | { | |
791 | struct kprobe *cur = __get_cpu_var(kprobe_instance); | |
792 | int ret = 0; | |
793 | ||
794 | if (cur && cur->break_handler) { | |
795 | if (cur->break_handler(cur, regs)) | |
796 | ret = 1; | |
797 | } | |
798 | reset_kprobe_instance(); | |
799 | return ret; | |
800 | } | |
801 | ||
802 | /* Walks the list and increments nmissed count for multiprobe case */ | |
803 | void __kprobes kprobes_inc_nmissed_count(struct kprobe *p) | |
804 | { | |
805 | struct kprobe *kp; | |
806 | if (!kprobe_aggrprobe(p)) { | |
807 | p->nmissed++; | |
808 | } else { | |
809 | list_for_each_entry_rcu(kp, &p->list, list) | |
810 | kp->nmissed++; | |
811 | } | |
812 | return; | |
813 | } | |
814 | ||
815 | void __kprobes recycle_rp_inst(struct kretprobe_instance *ri, | |
816 | struct hlist_head *head) | |
817 | { | |
818 | struct kretprobe *rp = ri->rp; | |
819 | ||
820 | /* remove rp inst off the rprobe_inst_table */ | |
821 | hlist_del(&ri->hlist); | |
822 | INIT_HLIST_NODE(&ri->hlist); | |
823 | if (likely(rp)) { | |
824 | spin_lock(&rp->lock); | |
825 | hlist_add_head(&ri->hlist, &rp->free_instances); | |
826 | spin_unlock(&rp->lock); | |
827 | } else | |
828 | /* Unregistering */ | |
829 | hlist_add_head(&ri->hlist, head); | |
830 | } | |
831 | ||
832 | void __kprobes kretprobe_hash_lock(struct task_struct *tsk, | |
833 | struct hlist_head **head, unsigned long *flags) | |
834 | { | |
835 | unsigned long hash = hash_ptr(tsk, KPROBE_HASH_BITS); | |
836 | spinlock_t *hlist_lock; | |
837 | ||
838 | *head = &kretprobe_inst_table[hash]; | |
839 | hlist_lock = kretprobe_table_lock_ptr(hash); | |
840 | spin_lock_irqsave(hlist_lock, *flags); | |
841 | } | |
842 | ||
843 | static void __kprobes kretprobe_table_lock(unsigned long hash, | |
844 | unsigned long *flags) | |
845 | { | |
846 | spinlock_t *hlist_lock = kretprobe_table_lock_ptr(hash); | |
847 | spin_lock_irqsave(hlist_lock, *flags); | |
848 | } | |
849 | ||
850 | void __kprobes kretprobe_hash_unlock(struct task_struct *tsk, | |
851 | unsigned long *flags) | |
852 | { | |
853 | unsigned long hash = hash_ptr(tsk, KPROBE_HASH_BITS); | |
854 | spinlock_t *hlist_lock; | |
855 | ||
856 | hlist_lock = kretprobe_table_lock_ptr(hash); | |
857 | spin_unlock_irqrestore(hlist_lock, *flags); | |
858 | } | |
859 | ||
860 | static void __kprobes kretprobe_table_unlock(unsigned long hash, | |
861 | unsigned long *flags) | |
862 | { | |
863 | spinlock_t *hlist_lock = kretprobe_table_lock_ptr(hash); | |
864 | spin_unlock_irqrestore(hlist_lock, *flags); | |
865 | } | |
866 | ||
867 | /* | |
868 | * This function is called from finish_task_switch when task tk becomes dead, | |
869 | * so that we can recycle any function-return probe instances associated | |
870 | * with this task. These left over instances represent probed functions | |
871 | * that have been called but will never return. | |
872 | */ | |
873 | void __kprobes kprobe_flush_task(struct task_struct *tk) | |
874 | { | |
875 | struct kretprobe_instance *ri; | |
876 | struct hlist_head *head, empty_rp; | |
877 | struct hlist_node *node, *tmp; | |
878 | unsigned long hash, flags = 0; | |
879 | ||
880 | if (unlikely(!kprobes_initialized)) | |
881 | /* Early boot. kretprobe_table_locks not yet initialized. */ | |
882 | return; | |
883 | ||
884 | hash = hash_ptr(tk, KPROBE_HASH_BITS); | |
885 | head = &kretprobe_inst_table[hash]; | |
886 | kretprobe_table_lock(hash, &flags); | |
887 | hlist_for_each_entry_safe(ri, node, tmp, head, hlist) { | |
888 | if (ri->task == tk) | |
889 | recycle_rp_inst(ri, &empty_rp); | |
890 | } | |
891 | kretprobe_table_unlock(hash, &flags); | |
892 | INIT_HLIST_HEAD(&empty_rp); | |
893 | hlist_for_each_entry_safe(ri, node, tmp, &empty_rp, hlist) { | |
894 | hlist_del(&ri->hlist); | |
895 | kfree(ri); | |
896 | } | |
897 | } | |
898 | ||
899 | static inline void free_rp_inst(struct kretprobe *rp) | |
900 | { | |
901 | struct kretprobe_instance *ri; | |
902 | struct hlist_node *pos, *next; | |
903 | ||
904 | hlist_for_each_entry_safe(ri, pos, next, &rp->free_instances, hlist) { | |
905 | hlist_del(&ri->hlist); | |
906 | kfree(ri); | |
907 | } | |
908 | } | |
909 | ||
910 | static void __kprobes cleanup_rp_inst(struct kretprobe *rp) | |
911 | { | |
912 | unsigned long flags, hash; | |
913 | struct kretprobe_instance *ri; | |
914 | struct hlist_node *pos, *next; | |
915 | struct hlist_head *head; | |
916 | ||
917 | /* No race here */ | |
918 | for (hash = 0; hash < KPROBE_TABLE_SIZE; hash++) { | |
919 | kretprobe_table_lock(hash, &flags); | |
920 | head = &kretprobe_inst_table[hash]; | |
921 | hlist_for_each_entry_safe(ri, pos, next, head, hlist) { | |
922 | if (ri->rp == rp) | |
923 | ri->rp = NULL; | |
924 | } | |
925 | kretprobe_table_unlock(hash, &flags); | |
926 | } | |
927 | free_rp_inst(rp); | |
928 | } | |
929 | ||
930 | /* | |
931 | * Add the new probe to ap->list. Fail if this is the | |
932 | * second jprobe at the address - two jprobes can't coexist | |
933 | */ | |
934 | static int __kprobes add_new_kprobe(struct kprobe *ap, struct kprobe *p) | |
935 | { | |
936 | BUG_ON(kprobe_gone(ap) || kprobe_gone(p)); | |
937 | ||
938 | if (p->break_handler || p->post_handler) | |
939 | unoptimize_kprobe(ap); /* Fall back to normal kprobe */ | |
940 | ||
941 | if (p->break_handler) { | |
942 | if (ap->break_handler) | |
943 | return -EEXIST; | |
944 | list_add_tail_rcu(&p->list, &ap->list); | |
945 | ap->break_handler = aggr_break_handler; | |
946 | } else | |
947 | list_add_rcu(&p->list, &ap->list); | |
948 | if (p->post_handler && !ap->post_handler) | |
949 | ap->post_handler = aggr_post_handler; | |
950 | ||
951 | if (kprobe_disabled(ap) && !kprobe_disabled(p)) { | |
952 | ap->flags &= ~KPROBE_FLAG_DISABLED; | |
953 | if (!kprobes_all_disarmed) | |
954 | /* Arm the breakpoint again. */ | |
955 | __arm_kprobe(ap); | |
956 | } | |
957 | return 0; | |
958 | } | |
959 | ||
960 | /* | |
961 | * Fill in the required fields of the "manager kprobe". Replace the | |
962 | * earlier kprobe in the hlist with the manager kprobe | |
963 | */ | |
964 | static void __kprobes init_aggr_kprobe(struct kprobe *ap, struct kprobe *p) | |
965 | { | |
966 | /* Copy p's insn slot to ap */ | |
967 | copy_kprobe(p, ap); | |
968 | flush_insn_slot(ap); | |
969 | ap->addr = p->addr; | |
970 | ap->flags = p->flags & ~KPROBE_FLAG_OPTIMIZED; | |
971 | ap->pre_handler = aggr_pre_handler; | |
972 | ap->fault_handler = aggr_fault_handler; | |
973 | /* We don't care the kprobe which has gone. */ | |
974 | if (p->post_handler && !kprobe_gone(p)) | |
975 | ap->post_handler = aggr_post_handler; | |
976 | if (p->break_handler && !kprobe_gone(p)) | |
977 | ap->break_handler = aggr_break_handler; | |
978 | ||
979 | INIT_LIST_HEAD(&ap->list); | |
980 | INIT_HLIST_NODE(&ap->hlist); | |
981 | ||
982 | list_add_rcu(&p->list, &ap->list); | |
983 | hlist_replace_rcu(&p->hlist, &ap->hlist); | |
984 | } | |
985 | ||
986 | /* | |
987 | * This is the second or subsequent kprobe at the address - handle | |
988 | * the intricacies | |
989 | */ | |
990 | static int __kprobes register_aggr_kprobe(struct kprobe *old_p, | |
991 | struct kprobe *p) | |
992 | { | |
993 | int ret = 0; | |
994 | struct kprobe *ap = old_p; | |
995 | ||
996 | if (!kprobe_aggrprobe(old_p)) { | |
997 | /* If old_p is not an aggr_kprobe, create new aggr_kprobe. */ | |
998 | ap = alloc_aggr_kprobe(old_p); | |
999 | if (!ap) | |
1000 | return -ENOMEM; | |
1001 | init_aggr_kprobe(ap, old_p); | |
1002 | } | |
1003 | ||
1004 | if (kprobe_gone(ap)) { | |
1005 | /* | |
1006 | * Attempting to insert new probe at the same location that | |
1007 | * had a probe in the module vaddr area which already | |
1008 | * freed. So, the instruction slot has already been | |
1009 | * released. We need a new slot for the new probe. | |
1010 | */ | |
1011 | ret = arch_prepare_kprobe(ap); | |
1012 | if (ret) | |
1013 | /* | |
1014 | * Even if fail to allocate new slot, don't need to | |
1015 | * free aggr_probe. It will be used next time, or | |
1016 | * freed by unregister_kprobe. | |
1017 | */ | |
1018 | return ret; | |
1019 | ||
1020 | /* Prepare optimized instructions if possible. */ | |
1021 | prepare_optimized_kprobe(ap); | |
1022 | ||
1023 | /* | |
1024 | * Clear gone flag to prevent allocating new slot again, and | |
1025 | * set disabled flag because it is not armed yet. | |
1026 | */ | |
1027 | ap->flags = (ap->flags & ~KPROBE_FLAG_GONE) | |
1028 | | KPROBE_FLAG_DISABLED; | |
1029 | } | |
1030 | ||
1031 | /* Copy ap's insn slot to p */ | |
1032 | copy_kprobe(ap, p); | |
1033 | return add_new_kprobe(ap, p); | |
1034 | } | |
1035 | ||
1036 | /* Try to disable aggr_kprobe, and return 1 if succeeded.*/ | |
1037 | static int __kprobes try_to_disable_aggr_kprobe(struct kprobe *p) | |
1038 | { | |
1039 | struct kprobe *kp; | |
1040 | ||
1041 | list_for_each_entry_rcu(kp, &p->list, list) { | |
1042 | if (!kprobe_disabled(kp)) | |
1043 | /* | |
1044 | * There is an active probe on the list. | |
1045 | * We can't disable aggr_kprobe. | |
1046 | */ | |
1047 | return 0; | |
1048 | } | |
1049 | p->flags |= KPROBE_FLAG_DISABLED; | |
1050 | return 1; | |
1051 | } | |
1052 | ||
1053 | static int __kprobes in_kprobes_functions(unsigned long addr) | |
1054 | { | |
1055 | struct kprobe_blackpoint *kb; | |
1056 | ||
1057 | if (addr >= (unsigned long)__kprobes_text_start && | |
1058 | addr < (unsigned long)__kprobes_text_end) | |
1059 | return -EINVAL; | |
1060 | /* | |
1061 | * If there exists a kprobe_blacklist, verify and | |
1062 | * fail any probe registration in the prohibited area | |
1063 | */ | |
1064 | for (kb = kprobe_blacklist; kb->name != NULL; kb++) { | |
1065 | if (kb->start_addr) { | |
1066 | if (addr >= kb->start_addr && | |
1067 | addr < (kb->start_addr + kb->range)) | |
1068 | return -EINVAL; | |
1069 | } | |
1070 | } | |
1071 | return 0; | |
1072 | } | |
1073 | ||
1074 | /* | |
1075 | * If we have a symbol_name argument, look it up and add the offset field | |
1076 | * to it. This way, we can specify a relative address to a symbol. | |
1077 | */ | |
1078 | static kprobe_opcode_t __kprobes *kprobe_addr(struct kprobe *p) | |
1079 | { | |
1080 | kprobe_opcode_t *addr = p->addr; | |
1081 | if (p->symbol_name) { | |
1082 | if (addr) | |
1083 | return NULL; | |
1084 | kprobe_lookup_name(p->symbol_name, addr); | |
1085 | } | |
1086 | ||
1087 | if (!addr) | |
1088 | return NULL; | |
1089 | return (kprobe_opcode_t *)(((char *)addr) + p->offset); | |
1090 | } | |
1091 | ||
1092 | /* Check passed kprobe is valid and return kprobe in kprobe_table. */ | |
1093 | static struct kprobe * __kprobes __get_valid_kprobe(struct kprobe *p) | |
1094 | { | |
1095 | struct kprobe *old_p, *list_p; | |
1096 | ||
1097 | old_p = get_kprobe(p->addr); | |
1098 | if (unlikely(!old_p)) | |
1099 | return NULL; | |
1100 | ||
1101 | if (p != old_p) { | |
1102 | list_for_each_entry_rcu(list_p, &old_p->list, list) | |
1103 | if (list_p == p) | |
1104 | /* kprobe p is a valid probe */ | |
1105 | goto valid; | |
1106 | return NULL; | |
1107 | } | |
1108 | valid: | |
1109 | return old_p; | |
1110 | } | |
1111 | ||
1112 | /* Return error if the kprobe is being re-registered */ | |
1113 | static inline int check_kprobe_rereg(struct kprobe *p) | |
1114 | { | |
1115 | int ret = 0; | |
1116 | struct kprobe *old_p; | |
1117 | ||
1118 | mutex_lock(&kprobe_mutex); | |
1119 | old_p = __get_valid_kprobe(p); | |
1120 | if (old_p) | |
1121 | ret = -EINVAL; | |
1122 | mutex_unlock(&kprobe_mutex); | |
1123 | return ret; | |
1124 | } | |
1125 | ||
1126 | int __kprobes register_kprobe(struct kprobe *p) | |
1127 | { | |
1128 | int ret = 0; | |
1129 | struct kprobe *old_p; | |
1130 | struct module *probed_mod; | |
1131 | kprobe_opcode_t *addr; | |
1132 | ||
1133 | addr = kprobe_addr(p); | |
1134 | if (!addr) | |
1135 | return -EINVAL; | |
1136 | p->addr = addr; | |
1137 | ||
1138 | ret = check_kprobe_rereg(p); | |
1139 | if (ret) | |
1140 | return ret; | |
1141 | ||
1142 | preempt_disable(); | |
1143 | if (!kernel_text_address((unsigned long) p->addr) || | |
1144 | in_kprobes_functions((unsigned long) p->addr) || | |
1145 | ftrace_text_reserved(p->addr, p->addr)) { | |
1146 | preempt_enable(); | |
1147 | return -EINVAL; | |
1148 | } | |
1149 | ||
1150 | /* User can pass only KPROBE_FLAG_DISABLED to register_kprobe */ | |
1151 | p->flags &= KPROBE_FLAG_DISABLED; | |
1152 | ||
1153 | /* | |
1154 | * Check if are we probing a module. | |
1155 | */ | |
1156 | probed_mod = __module_text_address((unsigned long) p->addr); | |
1157 | if (probed_mod) { | |
1158 | /* | |
1159 | * We must hold a refcount of the probed module while updating | |
1160 | * its code to prohibit unexpected unloading. | |
1161 | */ | |
1162 | if (unlikely(!try_module_get(probed_mod))) { | |
1163 | preempt_enable(); | |
1164 | return -EINVAL; | |
1165 | } | |
1166 | /* | |
1167 | * If the module freed .init.text, we couldn't insert | |
1168 | * kprobes in there. | |
1169 | */ | |
1170 | if (within_module_init((unsigned long)p->addr, probed_mod) && | |
1171 | probed_mod->state != MODULE_STATE_COMING) { | |
1172 | module_put(probed_mod); | |
1173 | preempt_enable(); | |
1174 | return -EINVAL; | |
1175 | } | |
1176 | } | |
1177 | preempt_enable(); | |
1178 | ||
1179 | p->nmissed = 0; | |
1180 | INIT_LIST_HEAD(&p->list); | |
1181 | mutex_lock(&kprobe_mutex); | |
1182 | ||
1183 | get_online_cpus(); /* For avoiding text_mutex deadlock. */ | |
1184 | mutex_lock(&text_mutex); | |
1185 | ||
1186 | old_p = get_kprobe(p->addr); | |
1187 | if (old_p) { | |
1188 | /* Since this may unoptimize old_p, locking text_mutex. */ | |
1189 | ret = register_aggr_kprobe(old_p, p); | |
1190 | goto out; | |
1191 | } | |
1192 | ||
1193 | ret = arch_prepare_kprobe(p); | |
1194 | if (ret) | |
1195 | goto out; | |
1196 | ||
1197 | INIT_HLIST_NODE(&p->hlist); | |
1198 | hlist_add_head_rcu(&p->hlist, | |
1199 | &kprobe_table[hash_ptr(p->addr, KPROBE_HASH_BITS)]); | |
1200 | ||
1201 | if (!kprobes_all_disarmed && !kprobe_disabled(p)) | |
1202 | __arm_kprobe(p); | |
1203 | ||
1204 | /* Try to optimize kprobe */ | |
1205 | try_to_optimize_kprobe(p); | |
1206 | ||
1207 | out: | |
1208 | mutex_unlock(&text_mutex); | |
1209 | put_online_cpus(); | |
1210 | mutex_unlock(&kprobe_mutex); | |
1211 | ||
1212 | if (probed_mod) | |
1213 | module_put(probed_mod); | |
1214 | ||
1215 | return ret; | |
1216 | } | |
1217 | EXPORT_SYMBOL_GPL(register_kprobe); | |
1218 | ||
1219 | /* | |
1220 | * Unregister a kprobe without a scheduler synchronization. | |
1221 | */ | |
1222 | static int __kprobes __unregister_kprobe_top(struct kprobe *p) | |
1223 | { | |
1224 | struct kprobe *old_p, *list_p; | |
1225 | ||
1226 | old_p = __get_valid_kprobe(p); | |
1227 | if (old_p == NULL) | |
1228 | return -EINVAL; | |
1229 | ||
1230 | if (old_p == p || | |
1231 | (kprobe_aggrprobe(old_p) && | |
1232 | list_is_singular(&old_p->list))) { | |
1233 | /* | |
1234 | * Only probe on the hash list. Disarm only if kprobes are | |
1235 | * enabled and not gone - otherwise, the breakpoint would | |
1236 | * already have been removed. We save on flushing icache. | |
1237 | */ | |
1238 | if (!kprobes_all_disarmed && !kprobe_disabled(old_p)) | |
1239 | disarm_kprobe(old_p); | |
1240 | hlist_del_rcu(&old_p->hlist); | |
1241 | } else { | |
1242 | if (p->break_handler && !kprobe_gone(p)) | |
1243 | old_p->break_handler = NULL; | |
1244 | if (p->post_handler && !kprobe_gone(p)) { | |
1245 | list_for_each_entry_rcu(list_p, &old_p->list, list) { | |
1246 | if ((list_p != p) && (list_p->post_handler)) | |
1247 | goto noclean; | |
1248 | } | |
1249 | old_p->post_handler = NULL; | |
1250 | } | |
1251 | noclean: | |
1252 | list_del_rcu(&p->list); | |
1253 | if (!kprobe_disabled(old_p)) { | |
1254 | try_to_disable_aggr_kprobe(old_p); | |
1255 | if (!kprobes_all_disarmed) { | |
1256 | if (kprobe_disabled(old_p)) | |
1257 | disarm_kprobe(old_p); | |
1258 | else | |
1259 | /* Try to optimize this probe again */ | |
1260 | optimize_kprobe(old_p); | |
1261 | } | |
1262 | } | |
1263 | } | |
1264 | return 0; | |
1265 | } | |
1266 | ||
1267 | static void __kprobes __unregister_kprobe_bottom(struct kprobe *p) | |
1268 | { | |
1269 | struct kprobe *old_p; | |
1270 | ||
1271 | if (list_empty(&p->list)) | |
1272 | arch_remove_kprobe(p); | |
1273 | else if (list_is_singular(&p->list)) { | |
1274 | /* "p" is the last child of an aggr_kprobe */ | |
1275 | old_p = list_entry(p->list.next, struct kprobe, list); | |
1276 | list_del(&p->list); | |
1277 | arch_remove_kprobe(old_p); | |
1278 | free_aggr_kprobe(old_p); | |
1279 | } | |
1280 | } | |
1281 | ||
1282 | int __kprobes register_kprobes(struct kprobe **kps, int num) | |
1283 | { | |
1284 | int i, ret = 0; | |
1285 | ||
1286 | if (num <= 0) | |
1287 | return -EINVAL; | |
1288 | for (i = 0; i < num; i++) { | |
1289 | ret = register_kprobe(kps[i]); | |
1290 | if (ret < 0) { | |
1291 | if (i > 0) | |
1292 | unregister_kprobes(kps, i); | |
1293 | break; | |
1294 | } | |
1295 | } | |
1296 | return ret; | |
1297 | } | |
1298 | EXPORT_SYMBOL_GPL(register_kprobes); | |
1299 | ||
1300 | void __kprobes unregister_kprobe(struct kprobe *p) | |
1301 | { | |
1302 | unregister_kprobes(&p, 1); | |
1303 | } | |
1304 | EXPORT_SYMBOL_GPL(unregister_kprobe); | |
1305 | ||
1306 | void __kprobes unregister_kprobes(struct kprobe **kps, int num) | |
1307 | { | |
1308 | int i; | |
1309 | ||
1310 | if (num <= 0) | |
1311 | return; | |
1312 | mutex_lock(&kprobe_mutex); | |
1313 | for (i = 0; i < num; i++) | |
1314 | if (__unregister_kprobe_top(kps[i]) < 0) | |
1315 | kps[i]->addr = NULL; | |
1316 | mutex_unlock(&kprobe_mutex); | |
1317 | ||
1318 | synchronize_sched(); | |
1319 | for (i = 0; i < num; i++) | |
1320 | if (kps[i]->addr) | |
1321 | __unregister_kprobe_bottom(kps[i]); | |
1322 | } | |
1323 | EXPORT_SYMBOL_GPL(unregister_kprobes); | |
1324 | ||
1325 | static struct notifier_block kprobe_exceptions_nb = { | |
1326 | .notifier_call = kprobe_exceptions_notify, | |
1327 | .priority = 0x7fffffff /* we need to be notified first */ | |
1328 | }; | |
1329 | ||
1330 | unsigned long __weak arch_deref_entry_point(void *entry) | |
1331 | { | |
1332 | return (unsigned long)entry; | |
1333 | } | |
1334 | ||
1335 | int __kprobes register_jprobes(struct jprobe **jps, int num) | |
1336 | { | |
1337 | struct jprobe *jp; | |
1338 | int ret = 0, i; | |
1339 | ||
1340 | if (num <= 0) | |
1341 | return -EINVAL; | |
1342 | for (i = 0; i < num; i++) { | |
1343 | unsigned long addr, offset; | |
1344 | jp = jps[i]; | |
1345 | addr = arch_deref_entry_point(jp->entry); | |
1346 | ||
1347 | /* Verify probepoint is a function entry point */ | |
1348 | if (kallsyms_lookup_size_offset(addr, NULL, &offset) && | |
1349 | offset == 0) { | |
1350 | jp->kp.pre_handler = setjmp_pre_handler; | |
1351 | jp->kp.break_handler = longjmp_break_handler; | |
1352 | ret = register_kprobe(&jp->kp); | |
1353 | } else | |
1354 | ret = -EINVAL; | |
1355 | ||
1356 | if (ret < 0) { | |
1357 | if (i > 0) | |
1358 | unregister_jprobes(jps, i); | |
1359 | break; | |
1360 | } | |
1361 | } | |
1362 | return ret; | |
1363 | } | |
1364 | EXPORT_SYMBOL_GPL(register_jprobes); | |
1365 | ||
1366 | int __kprobes register_jprobe(struct jprobe *jp) | |
1367 | { | |
1368 | return register_jprobes(&jp, 1); | |
1369 | } | |
1370 | EXPORT_SYMBOL_GPL(register_jprobe); | |
1371 | ||
1372 | void __kprobes unregister_jprobe(struct jprobe *jp) | |
1373 | { | |
1374 | unregister_jprobes(&jp, 1); | |
1375 | } | |
1376 | EXPORT_SYMBOL_GPL(unregister_jprobe); | |
1377 | ||
1378 | void __kprobes unregister_jprobes(struct jprobe **jps, int num) | |
1379 | { | |
1380 | int i; | |
1381 | ||
1382 | if (num <= 0) | |
1383 | return; | |
1384 | mutex_lock(&kprobe_mutex); | |
1385 | for (i = 0; i < num; i++) | |
1386 | if (__unregister_kprobe_top(&jps[i]->kp) < 0) | |
1387 | jps[i]->kp.addr = NULL; | |
1388 | mutex_unlock(&kprobe_mutex); | |
1389 | ||
1390 | synchronize_sched(); | |
1391 | for (i = 0; i < num; i++) { | |
1392 | if (jps[i]->kp.addr) | |
1393 | __unregister_kprobe_bottom(&jps[i]->kp); | |
1394 | } | |
1395 | } | |
1396 | EXPORT_SYMBOL_GPL(unregister_jprobes); | |
1397 | ||
1398 | #ifdef CONFIG_KRETPROBES | |
1399 | /* | |
1400 | * This kprobe pre_handler is registered with every kretprobe. When probe | |
1401 | * hits it will set up the return probe. | |
1402 | */ | |
1403 | static int __kprobes pre_handler_kretprobe(struct kprobe *p, | |
1404 | struct pt_regs *regs) | |
1405 | { | |
1406 | struct kretprobe *rp = container_of(p, struct kretprobe, kp); | |
1407 | unsigned long hash, flags = 0; | |
1408 | struct kretprobe_instance *ri; | |
1409 | ||
1410 | /*TODO: consider to only swap the RA after the last pre_handler fired */ | |
1411 | hash = hash_ptr(current, KPROBE_HASH_BITS); | |
1412 | spin_lock_irqsave(&rp->lock, flags); | |
1413 | if (!hlist_empty(&rp->free_instances)) { | |
1414 | ri = hlist_entry(rp->free_instances.first, | |
1415 | struct kretprobe_instance, hlist); | |
1416 | hlist_del(&ri->hlist); | |
1417 | spin_unlock_irqrestore(&rp->lock, flags); | |
1418 | ||
1419 | ri->rp = rp; | |
1420 | ri->task = current; | |
1421 | ||
1422 | if (rp->entry_handler && rp->entry_handler(ri, regs)) | |
1423 | return 0; | |
1424 | ||
1425 | arch_prepare_kretprobe(ri, regs); | |
1426 | ||
1427 | /* XXX(hch): why is there no hlist_move_head? */ | |
1428 | INIT_HLIST_NODE(&ri->hlist); | |
1429 | kretprobe_table_lock(hash, &flags); | |
1430 | hlist_add_head(&ri->hlist, &kretprobe_inst_table[hash]); | |
1431 | kretprobe_table_unlock(hash, &flags); | |
1432 | } else { | |
1433 | rp->nmissed++; | |
1434 | spin_unlock_irqrestore(&rp->lock, flags); | |
1435 | } | |
1436 | return 0; | |
1437 | } | |
1438 | ||
1439 | int __kprobes register_kretprobe(struct kretprobe *rp) | |
1440 | { | |
1441 | int ret = 0; | |
1442 | struct kretprobe_instance *inst; | |
1443 | int i; | |
1444 | void *addr; | |
1445 | ||
1446 | if (kretprobe_blacklist_size) { | |
1447 | addr = kprobe_addr(&rp->kp); | |
1448 | if (!addr) | |
1449 | return -EINVAL; | |
1450 | ||
1451 | for (i = 0; kretprobe_blacklist[i].name != NULL; i++) { | |
1452 | if (kretprobe_blacklist[i].addr == addr) | |
1453 | return -EINVAL; | |
1454 | } | |
1455 | } | |
1456 | ||
1457 | rp->kp.pre_handler = pre_handler_kretprobe; | |
1458 | rp->kp.post_handler = NULL; | |
1459 | rp->kp.fault_handler = NULL; | |
1460 | rp->kp.break_handler = NULL; | |
1461 | ||
1462 | /* Pre-allocate memory for max kretprobe instances */ | |
1463 | if (rp->maxactive <= 0) { | |
1464 | #ifdef CONFIG_PREEMPT | |
1465 | rp->maxactive = max_t(unsigned int, 10, 2*num_possible_cpus()); | |
1466 | #else | |
1467 | rp->maxactive = num_possible_cpus(); | |
1468 | #endif | |
1469 | } | |
1470 | spin_lock_init(&rp->lock); | |
1471 | INIT_HLIST_HEAD(&rp->free_instances); | |
1472 | for (i = 0; i < rp->maxactive; i++) { | |
1473 | inst = kmalloc(sizeof(struct kretprobe_instance) + | |
1474 | rp->data_size, GFP_KERNEL); | |
1475 | if (inst == NULL) { | |
1476 | free_rp_inst(rp); | |
1477 | return -ENOMEM; | |
1478 | } | |
1479 | INIT_HLIST_NODE(&inst->hlist); | |
1480 | hlist_add_head(&inst->hlist, &rp->free_instances); | |
1481 | } | |
1482 | ||
1483 | rp->nmissed = 0; | |
1484 | /* Establish function entry probe point */ | |
1485 | ret = register_kprobe(&rp->kp); | |
1486 | if (ret != 0) | |
1487 | free_rp_inst(rp); | |
1488 | return ret; | |
1489 | } | |
1490 | EXPORT_SYMBOL_GPL(register_kretprobe); | |
1491 | ||
1492 | int __kprobes register_kretprobes(struct kretprobe **rps, int num) | |
1493 | { | |
1494 | int ret = 0, i; | |
1495 | ||
1496 | if (num <= 0) | |
1497 | return -EINVAL; | |
1498 | for (i = 0; i < num; i++) { | |
1499 | ret = register_kretprobe(rps[i]); | |
1500 | if (ret < 0) { | |
1501 | if (i > 0) | |
1502 | unregister_kretprobes(rps, i); | |
1503 | break; | |
1504 | } | |
1505 | } | |
1506 | return ret; | |
1507 | } | |
1508 | EXPORT_SYMBOL_GPL(register_kretprobes); | |
1509 | ||
1510 | void __kprobes unregister_kretprobe(struct kretprobe *rp) | |
1511 | { | |
1512 | unregister_kretprobes(&rp, 1); | |
1513 | } | |
1514 | EXPORT_SYMBOL_GPL(unregister_kretprobe); | |
1515 | ||
1516 | void __kprobes unregister_kretprobes(struct kretprobe **rps, int num) | |
1517 | { | |
1518 | int i; | |
1519 | ||
1520 | if (num <= 0) | |
1521 | return; | |
1522 | mutex_lock(&kprobe_mutex); | |
1523 | for (i = 0; i < num; i++) | |
1524 | if (__unregister_kprobe_top(&rps[i]->kp) < 0) | |
1525 | rps[i]->kp.addr = NULL; | |
1526 | mutex_unlock(&kprobe_mutex); | |
1527 | ||
1528 | synchronize_sched(); | |
1529 | for (i = 0; i < num; i++) { | |
1530 | if (rps[i]->kp.addr) { | |
1531 | __unregister_kprobe_bottom(&rps[i]->kp); | |
1532 | cleanup_rp_inst(rps[i]); | |
1533 | } | |
1534 | } | |
1535 | } | |
1536 | EXPORT_SYMBOL_GPL(unregister_kretprobes); | |
1537 | ||
1538 | #else /* CONFIG_KRETPROBES */ | |
1539 | int __kprobes register_kretprobe(struct kretprobe *rp) | |
1540 | { | |
1541 | return -ENOSYS; | |
1542 | } | |
1543 | EXPORT_SYMBOL_GPL(register_kretprobe); | |
1544 | ||
1545 | int __kprobes register_kretprobes(struct kretprobe **rps, int num) | |
1546 | { | |
1547 | return -ENOSYS; | |
1548 | } | |
1549 | EXPORT_SYMBOL_GPL(register_kretprobes); | |
1550 | ||
1551 | void __kprobes unregister_kretprobe(struct kretprobe *rp) | |
1552 | { | |
1553 | } | |
1554 | EXPORT_SYMBOL_GPL(unregister_kretprobe); | |
1555 | ||
1556 | void __kprobes unregister_kretprobes(struct kretprobe **rps, int num) | |
1557 | { | |
1558 | } | |
1559 | EXPORT_SYMBOL_GPL(unregister_kretprobes); | |
1560 | ||
1561 | static int __kprobes pre_handler_kretprobe(struct kprobe *p, | |
1562 | struct pt_regs *regs) | |
1563 | { | |
1564 | return 0; | |
1565 | } | |
1566 | ||
1567 | #endif /* CONFIG_KRETPROBES */ | |
1568 | ||
1569 | /* Set the kprobe gone and remove its instruction buffer. */ | |
1570 | static void __kprobes kill_kprobe(struct kprobe *p) | |
1571 | { | |
1572 | struct kprobe *kp; | |
1573 | ||
1574 | p->flags |= KPROBE_FLAG_GONE; | |
1575 | if (kprobe_aggrprobe(p)) { | |
1576 | /* | |
1577 | * If this is an aggr_kprobe, we have to list all the | |
1578 | * chained probes and mark them GONE. | |
1579 | */ | |
1580 | list_for_each_entry_rcu(kp, &p->list, list) | |
1581 | kp->flags |= KPROBE_FLAG_GONE; | |
1582 | p->post_handler = NULL; | |
1583 | p->break_handler = NULL; | |
1584 | kill_optimized_kprobe(p); | |
1585 | } | |
1586 | /* | |
1587 | * Here, we can remove insn_slot safely, because no thread calls | |
1588 | * the original probed function (which will be freed soon) any more. | |
1589 | */ | |
1590 | arch_remove_kprobe(p); | |
1591 | } | |
1592 | ||
1593 | /* Disable one kprobe */ | |
1594 | int __kprobes disable_kprobe(struct kprobe *kp) | |
1595 | { | |
1596 | int ret = 0; | |
1597 | struct kprobe *p; | |
1598 | ||
1599 | mutex_lock(&kprobe_mutex); | |
1600 | ||
1601 | /* Check whether specified probe is valid. */ | |
1602 | p = __get_valid_kprobe(kp); | |
1603 | if (unlikely(p == NULL)) { | |
1604 | ret = -EINVAL; | |
1605 | goto out; | |
1606 | } | |
1607 | ||
1608 | /* If the probe is already disabled (or gone), just return */ | |
1609 | if (kprobe_disabled(kp)) | |
1610 | goto out; | |
1611 | ||
1612 | kp->flags |= KPROBE_FLAG_DISABLED; | |
1613 | if (p != kp) | |
1614 | /* When kp != p, p is always enabled. */ | |
1615 | try_to_disable_aggr_kprobe(p); | |
1616 | ||
1617 | if (!kprobes_all_disarmed && kprobe_disabled(p)) | |
1618 | disarm_kprobe(p); | |
1619 | out: | |
1620 | mutex_unlock(&kprobe_mutex); | |
1621 | return ret; | |
1622 | } | |
1623 | EXPORT_SYMBOL_GPL(disable_kprobe); | |
1624 | ||
1625 | /* Enable one kprobe */ | |
1626 | int __kprobes enable_kprobe(struct kprobe *kp) | |
1627 | { | |
1628 | int ret = 0; | |
1629 | struct kprobe *p; | |
1630 | ||
1631 | mutex_lock(&kprobe_mutex); | |
1632 | ||
1633 | /* Check whether specified probe is valid. */ | |
1634 | p = __get_valid_kprobe(kp); | |
1635 | if (unlikely(p == NULL)) { | |
1636 | ret = -EINVAL; | |
1637 | goto out; | |
1638 | } | |
1639 | ||
1640 | if (kprobe_gone(kp)) { | |
1641 | /* This kprobe has gone, we couldn't enable it. */ | |
1642 | ret = -EINVAL; | |
1643 | goto out; | |
1644 | } | |
1645 | ||
1646 | if (p != kp) | |
1647 | kp->flags &= ~KPROBE_FLAG_DISABLED; | |
1648 | ||
1649 | if (!kprobes_all_disarmed && kprobe_disabled(p)) { | |
1650 | p->flags &= ~KPROBE_FLAG_DISABLED; | |
1651 | arm_kprobe(p); | |
1652 | } | |
1653 | out: | |
1654 | mutex_unlock(&kprobe_mutex); | |
1655 | return ret; | |
1656 | } | |
1657 | EXPORT_SYMBOL_GPL(enable_kprobe); | |
1658 | ||
1659 | void __kprobes dump_kprobe(struct kprobe *kp) | |
1660 | { | |
1661 | printk(KERN_WARNING "Dumping kprobe:\n"); | |
1662 | printk(KERN_WARNING "Name: %s\nAddress: %p\nOffset: %x\n", | |
1663 | kp->symbol_name, kp->addr, kp->offset); | |
1664 | } | |
1665 | ||
1666 | /* Module notifier call back, checking kprobes on the module */ | |
1667 | static int __kprobes kprobes_module_callback(struct notifier_block *nb, | |
1668 | unsigned long val, void *data) | |
1669 | { | |
1670 | struct module *mod = data; | |
1671 | struct hlist_head *head; | |
1672 | struct hlist_node *node; | |
1673 | struct kprobe *p; | |
1674 | unsigned int i; | |
1675 | int checkcore = (val == MODULE_STATE_GOING); | |
1676 | ||
1677 | if (val != MODULE_STATE_GOING && val != MODULE_STATE_LIVE) | |
1678 | return NOTIFY_DONE; | |
1679 | ||
1680 | /* | |
1681 | * When MODULE_STATE_GOING was notified, both of module .text and | |
1682 | * .init.text sections would be freed. When MODULE_STATE_LIVE was | |
1683 | * notified, only .init.text section would be freed. We need to | |
1684 | * disable kprobes which have been inserted in the sections. | |
1685 | */ | |
1686 | mutex_lock(&kprobe_mutex); | |
1687 | for (i = 0; i < KPROBE_TABLE_SIZE; i++) { | |
1688 | head = &kprobe_table[i]; | |
1689 | hlist_for_each_entry_rcu(p, node, head, hlist) | |
1690 | if (within_module_init((unsigned long)p->addr, mod) || | |
1691 | (checkcore && | |
1692 | within_module_core((unsigned long)p->addr, mod))) { | |
1693 | /* | |
1694 | * The vaddr this probe is installed will soon | |
1695 | * be vfreed buy not synced to disk. Hence, | |
1696 | * disarming the breakpoint isn't needed. | |
1697 | */ | |
1698 | kill_kprobe(p); | |
1699 | } | |
1700 | } | |
1701 | mutex_unlock(&kprobe_mutex); | |
1702 | return NOTIFY_DONE; | |
1703 | } | |
1704 | ||
1705 | static struct notifier_block kprobe_module_nb = { | |
1706 | .notifier_call = kprobes_module_callback, | |
1707 | .priority = 0 | |
1708 | }; | |
1709 | ||
1710 | static int __init init_kprobes(void) | |
1711 | { | |
1712 | int i, err = 0; | |
1713 | unsigned long offset = 0, size = 0; | |
1714 | char *modname, namebuf[128]; | |
1715 | const char *symbol_name; | |
1716 | void *addr; | |
1717 | struct kprobe_blackpoint *kb; | |
1718 | ||
1719 | /* FIXME allocate the probe table, currently defined statically */ | |
1720 | /* initialize all list heads */ | |
1721 | for (i = 0; i < KPROBE_TABLE_SIZE; i++) { | |
1722 | INIT_HLIST_HEAD(&kprobe_table[i]); | |
1723 | INIT_HLIST_HEAD(&kretprobe_inst_table[i]); | |
1724 | spin_lock_init(&(kretprobe_table_locks[i].lock)); | |
1725 | } | |
1726 | ||
1727 | /* | |
1728 | * Lookup and populate the kprobe_blacklist. | |
1729 | * | |
1730 | * Unlike the kretprobe blacklist, we'll need to determine | |
1731 | * the range of addresses that belong to the said functions, | |
1732 | * since a kprobe need not necessarily be at the beginning | |
1733 | * of a function. | |
1734 | */ | |
1735 | for (kb = kprobe_blacklist; kb->name != NULL; kb++) { | |
1736 | kprobe_lookup_name(kb->name, addr); | |
1737 | if (!addr) | |
1738 | continue; | |
1739 | ||
1740 | kb->start_addr = (unsigned long)addr; | |
1741 | symbol_name = kallsyms_lookup(kb->start_addr, | |
1742 | &size, &offset, &modname, namebuf); | |
1743 | if (!symbol_name) | |
1744 | kb->range = 0; | |
1745 | else | |
1746 | kb->range = size; | |
1747 | } | |
1748 | ||
1749 | if (kretprobe_blacklist_size) { | |
1750 | /* lookup the function address from its name */ | |
1751 | for (i = 0; kretprobe_blacklist[i].name != NULL; i++) { | |
1752 | kprobe_lookup_name(kretprobe_blacklist[i].name, | |
1753 | kretprobe_blacklist[i].addr); | |
1754 | if (!kretprobe_blacklist[i].addr) | |
1755 | printk("kretprobe: lookup failed: %s\n", | |
1756 | kretprobe_blacklist[i].name); | |
1757 | } | |
1758 | } | |
1759 | ||
1760 | #if defined(CONFIG_OPTPROBES) | |
1761 | #if defined(__ARCH_WANT_KPROBES_INSN_SLOT) | |
1762 | /* Init kprobe_optinsn_slots */ | |
1763 | kprobe_optinsn_slots.insn_size = MAX_OPTINSN_SIZE; | |
1764 | #endif | |
1765 | /* By default, kprobes can be optimized */ | |
1766 | kprobes_allow_optimization = true; | |
1767 | #endif | |
1768 | ||
1769 | /* By default, kprobes are armed */ | |
1770 | kprobes_all_disarmed = false; | |
1771 | ||
1772 | err = arch_init_kprobes(); | |
1773 | if (!err) | |
1774 | err = register_die_notifier(&kprobe_exceptions_nb); | |
1775 | if (!err) | |
1776 | err = register_module_notifier(&kprobe_module_nb); | |
1777 | ||
1778 | kprobes_initialized = (err == 0); | |
1779 | ||
1780 | if (!err) | |
1781 | init_test_probes(); | |
1782 | return err; | |
1783 | } | |
1784 | ||
1785 | #ifdef CONFIG_DEBUG_FS | |
1786 | static void __kprobes report_probe(struct seq_file *pi, struct kprobe *p, | |
1787 | const char *sym, int offset, char *modname, struct kprobe *pp) | |
1788 | { | |
1789 | char *kprobe_type; | |
1790 | ||
1791 | if (p->pre_handler == pre_handler_kretprobe) | |
1792 | kprobe_type = "r"; | |
1793 | else if (p->pre_handler == setjmp_pre_handler) | |
1794 | kprobe_type = "j"; | |
1795 | else | |
1796 | kprobe_type = "k"; | |
1797 | ||
1798 | if (sym) | |
1799 | seq_printf(pi, "%p %s %s+0x%x %s ", | |
1800 | p->addr, kprobe_type, sym, offset, | |
1801 | (modname ? modname : " ")); | |
1802 | else | |
1803 | seq_printf(pi, "%p %s %p ", | |
1804 | p->addr, kprobe_type, p->addr); | |
1805 | ||
1806 | if (!pp) | |
1807 | pp = p; | |
1808 | seq_printf(pi, "%s%s%s\n", | |
1809 | (kprobe_gone(p) ? "[GONE]" : ""), | |
1810 | ((kprobe_disabled(p) && !kprobe_gone(p)) ? "[DISABLED]" : ""), | |
1811 | (kprobe_optimized(pp) ? "[OPTIMIZED]" : "")); | |
1812 | } | |
1813 | ||
1814 | static void __kprobes *kprobe_seq_start(struct seq_file *f, loff_t *pos) | |
1815 | { | |
1816 | return (*pos < KPROBE_TABLE_SIZE) ? pos : NULL; | |
1817 | } | |
1818 | ||
1819 | static void __kprobes *kprobe_seq_next(struct seq_file *f, void *v, loff_t *pos) | |
1820 | { | |
1821 | (*pos)++; | |
1822 | if (*pos >= KPROBE_TABLE_SIZE) | |
1823 | return NULL; | |
1824 | return pos; | |
1825 | } | |
1826 | ||
1827 | static void __kprobes kprobe_seq_stop(struct seq_file *f, void *v) | |
1828 | { | |
1829 | /* Nothing to do */ | |
1830 | } | |
1831 | ||
1832 | static int __kprobes show_kprobe_addr(struct seq_file *pi, void *v) | |
1833 | { | |
1834 | struct hlist_head *head; | |
1835 | struct hlist_node *node; | |
1836 | struct kprobe *p, *kp; | |
1837 | const char *sym = NULL; | |
1838 | unsigned int i = *(loff_t *) v; | |
1839 | unsigned long offset = 0; | |
1840 | char *modname, namebuf[128]; | |
1841 | ||
1842 | head = &kprobe_table[i]; | |
1843 | preempt_disable(); | |
1844 | hlist_for_each_entry_rcu(p, node, head, hlist) { | |
1845 | sym = kallsyms_lookup((unsigned long)p->addr, NULL, | |
1846 | &offset, &modname, namebuf); | |
1847 | if (kprobe_aggrprobe(p)) { | |
1848 | list_for_each_entry_rcu(kp, &p->list, list) | |
1849 | report_probe(pi, kp, sym, offset, modname, p); | |
1850 | } else | |
1851 | report_probe(pi, p, sym, offset, modname, NULL); | |
1852 | } | |
1853 | preempt_enable(); | |
1854 | return 0; | |
1855 | } | |
1856 | ||
1857 | static const struct seq_operations kprobes_seq_ops = { | |
1858 | .start = kprobe_seq_start, | |
1859 | .next = kprobe_seq_next, | |
1860 | .stop = kprobe_seq_stop, | |
1861 | .show = show_kprobe_addr | |
1862 | }; | |
1863 | ||
1864 | static int __kprobes kprobes_open(struct inode *inode, struct file *filp) | |
1865 | { | |
1866 | return seq_open(filp, &kprobes_seq_ops); | |
1867 | } | |
1868 | ||
1869 | static const struct file_operations debugfs_kprobes_operations = { | |
1870 | .open = kprobes_open, | |
1871 | .read = seq_read, | |
1872 | .llseek = seq_lseek, | |
1873 | .release = seq_release, | |
1874 | }; | |
1875 | ||
1876 | static void __kprobes arm_all_kprobes(void) | |
1877 | { | |
1878 | struct hlist_head *head; | |
1879 | struct hlist_node *node; | |
1880 | struct kprobe *p; | |
1881 | unsigned int i; | |
1882 | ||
1883 | mutex_lock(&kprobe_mutex); | |
1884 | ||
1885 | /* If kprobes are armed, just return */ | |
1886 | if (!kprobes_all_disarmed) | |
1887 | goto already_enabled; | |
1888 | ||
1889 | /* Arming kprobes doesn't optimize kprobe itself */ | |
1890 | mutex_lock(&text_mutex); | |
1891 | for (i = 0; i < KPROBE_TABLE_SIZE; i++) { | |
1892 | head = &kprobe_table[i]; | |
1893 | hlist_for_each_entry_rcu(p, node, head, hlist) | |
1894 | if (!kprobe_disabled(p)) | |
1895 | __arm_kprobe(p); | |
1896 | } | |
1897 | mutex_unlock(&text_mutex); | |
1898 | ||
1899 | kprobes_all_disarmed = false; | |
1900 | printk(KERN_INFO "Kprobes globally enabled\n"); | |
1901 | ||
1902 | already_enabled: | |
1903 | mutex_unlock(&kprobe_mutex); | |
1904 | return; | |
1905 | } | |
1906 | ||
1907 | static void __kprobes disarm_all_kprobes(void) | |
1908 | { | |
1909 | struct hlist_head *head; | |
1910 | struct hlist_node *node; | |
1911 | struct kprobe *p; | |
1912 | unsigned int i; | |
1913 | ||
1914 | mutex_lock(&kprobe_mutex); | |
1915 | ||
1916 | /* If kprobes are already disarmed, just return */ | |
1917 | if (kprobes_all_disarmed) | |
1918 | goto already_disabled; | |
1919 | ||
1920 | kprobes_all_disarmed = true; | |
1921 | printk(KERN_INFO "Kprobes globally disabled\n"); | |
1922 | ||
1923 | /* | |
1924 | * Here we call get_online_cpus() for avoiding text_mutex deadlock, | |
1925 | * because disarming may also unoptimize kprobes. | |
1926 | */ | |
1927 | get_online_cpus(); | |
1928 | mutex_lock(&text_mutex); | |
1929 | for (i = 0; i < KPROBE_TABLE_SIZE; i++) { | |
1930 | head = &kprobe_table[i]; | |
1931 | hlist_for_each_entry_rcu(p, node, head, hlist) { | |
1932 | if (!arch_trampoline_kprobe(p) && !kprobe_disabled(p)) | |
1933 | __disarm_kprobe(p); | |
1934 | } | |
1935 | } | |
1936 | ||
1937 | mutex_unlock(&text_mutex); | |
1938 | put_online_cpus(); | |
1939 | mutex_unlock(&kprobe_mutex); | |
1940 | /* Allow all currently running kprobes to complete */ | |
1941 | synchronize_sched(); | |
1942 | return; | |
1943 | ||
1944 | already_disabled: | |
1945 | mutex_unlock(&kprobe_mutex); | |
1946 | return; | |
1947 | } | |
1948 | ||
1949 | /* | |
1950 | * XXX: The debugfs bool file interface doesn't allow for callbacks | |
1951 | * when the bool state is switched. We can reuse that facility when | |
1952 | * available | |
1953 | */ | |
1954 | static ssize_t read_enabled_file_bool(struct file *file, | |
1955 | char __user *user_buf, size_t count, loff_t *ppos) | |
1956 | { | |
1957 | char buf[3]; | |
1958 | ||
1959 | if (!kprobes_all_disarmed) | |
1960 | buf[0] = '1'; | |
1961 | else | |
1962 | buf[0] = '0'; | |
1963 | buf[1] = '\n'; | |
1964 | buf[2] = 0x00; | |
1965 | return simple_read_from_buffer(user_buf, count, ppos, buf, 2); | |
1966 | } | |
1967 | ||
1968 | static ssize_t write_enabled_file_bool(struct file *file, | |
1969 | const char __user *user_buf, size_t count, loff_t *ppos) | |
1970 | { | |
1971 | char buf[32]; | |
1972 | int buf_size; | |
1973 | ||
1974 | buf_size = min(count, (sizeof(buf)-1)); | |
1975 | if (copy_from_user(buf, user_buf, buf_size)) | |
1976 | return -EFAULT; | |
1977 | ||
1978 | switch (buf[0]) { | |
1979 | case 'y': | |
1980 | case 'Y': | |
1981 | case '1': | |
1982 | arm_all_kprobes(); | |
1983 | break; | |
1984 | case 'n': | |
1985 | case 'N': | |
1986 | case '0': | |
1987 | disarm_all_kprobes(); | |
1988 | break; | |
1989 | } | |
1990 | ||
1991 | return count; | |
1992 | } | |
1993 | ||
1994 | static const struct file_operations fops_kp = { | |
1995 | .read = read_enabled_file_bool, | |
1996 | .write = write_enabled_file_bool, | |
1997 | }; | |
1998 | ||
1999 | static int __kprobes debugfs_kprobe_init(void) | |
2000 | { | |
2001 | struct dentry *dir, *file; | |
2002 | unsigned int value = 1; | |
2003 | ||
2004 | dir = debugfs_create_dir("kprobes", NULL); | |
2005 | if (!dir) | |
2006 | return -ENOMEM; | |
2007 | ||
2008 | file = debugfs_create_file("list", 0444, dir, NULL, | |
2009 | &debugfs_kprobes_operations); | |
2010 | if (!file) { | |
2011 | debugfs_remove(dir); | |
2012 | return -ENOMEM; | |
2013 | } | |
2014 | ||
2015 | file = debugfs_create_file("enabled", 0600, dir, | |
2016 | &value, &fops_kp); | |
2017 | if (!file) { | |
2018 | debugfs_remove(dir); | |
2019 | return -ENOMEM; | |
2020 | } | |
2021 | ||
2022 | return 0; | |
2023 | } | |
2024 | ||
2025 | late_initcall(debugfs_kprobe_init); | |
2026 | #endif /* CONFIG_DEBUG_FS */ | |
2027 | ||
2028 | module_init(init_kprobes); | |
2029 | ||
2030 | /* defined in arch/.../kernel/kprobes.c */ | |
2031 | EXPORT_SYMBOL_GPL(jprobe_return); |