]>
Commit | Line | Data |
---|---|---|
1da177e4 LT |
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. | |
b94cce92 HN |
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. | |
1da177e4 LT |
33 | */ |
34 | #include <linux/kprobes.h> | |
1da177e4 LT |
35 | #include <linux/hash.h> |
36 | #include <linux/init.h> | |
4e57b681 | 37 | #include <linux/slab.h> |
e3869792 | 38 | #include <linux/stddef.h> |
9984de1a | 39 | #include <linux/export.h> |
9ec4b1f3 | 40 | #include <linux/moduleloader.h> |
3a872d89 | 41 | #include <linux/kallsyms.h> |
b4c6c34a | 42 | #include <linux/freezer.h> |
346fd59b SD |
43 | #include <linux/seq_file.h> |
44 | #include <linux/debugfs.h> | |
b2be84df | 45 | #include <linux/sysctl.h> |
1eeb66a1 | 46 | #include <linux/kdebug.h> |
4460fdad | 47 | #include <linux/memory.h> |
4554dbcb | 48 | #include <linux/ftrace.h> |
afd66255 | 49 | #include <linux/cpu.h> |
bf5438fc | 50 | #include <linux/jump_label.h> |
bf8f6e5b | 51 | |
d0aaff97 | 52 | #include <asm-generic/sections.h> |
1da177e4 LT |
53 | #include <asm/cacheflush.h> |
54 | #include <asm/errno.h> | |
bf8f6e5b | 55 | #include <asm/uaccess.h> |
1da177e4 LT |
56 | |
57 | #define KPROBE_HASH_BITS 6 | |
58 | #define KPROBE_TABLE_SIZE (1 << KPROBE_HASH_BITS) | |
59 | ||
3a872d89 AM |
60 | |
61 | /* | |
62 | * Some oddball architectures like 64bit powerpc have function descriptors | |
63 | * so this must be overridable. | |
64 | */ | |
65 | #ifndef kprobe_lookup_name | |
66 | #define kprobe_lookup_name(name, addr) \ | |
67 | addr = ((kprobe_opcode_t *)(kallsyms_lookup_name(name))) | |
68 | #endif | |
69 | ||
ef53d9c5 | 70 | static int kprobes_initialized; |
1da177e4 | 71 | static struct hlist_head kprobe_table[KPROBE_TABLE_SIZE]; |
b94cce92 | 72 | static struct hlist_head kretprobe_inst_table[KPROBE_TABLE_SIZE]; |
1da177e4 | 73 | |
bf8f6e5b | 74 | /* NOTE: change this value only with kprobe_mutex held */ |
e579abeb | 75 | static bool kprobes_all_disarmed; |
bf8f6e5b | 76 | |
43948f50 MH |
77 | /* This protects kprobe_table and optimizing_list */ |
78 | static DEFINE_MUTEX(kprobe_mutex); | |
e6584523 | 79 | static DEFINE_PER_CPU(struct kprobe *, kprobe_instance) = NULL; |
ef53d9c5 | 80 | static struct { |
ec484608 | 81 | raw_spinlock_t lock ____cacheline_aligned_in_smp; |
ef53d9c5 S |
82 | } kretprobe_table_locks[KPROBE_TABLE_SIZE]; |
83 | ||
ec484608 | 84 | static raw_spinlock_t *kretprobe_table_lock_ptr(unsigned long hash) |
ef53d9c5 S |
85 | { |
86 | return &(kretprobe_table_locks[hash].lock); | |
87 | } | |
1da177e4 | 88 | |
3d8d996e SD |
89 | /* |
90 | * Normally, functions that we'd want to prohibit kprobes in, are marked | |
91 | * __kprobes. But, there are cases where such functions already belong to | |
92 | * a different section (__sched for preempt_schedule) | |
93 | * | |
94 | * For such cases, we now have a blacklist | |
95 | */ | |
544304b2 | 96 | static struct kprobe_blackpoint kprobe_blacklist[] = { |
3d8d996e | 97 | {"preempt_schedule",}, |
65e234ec | 98 | {"native_get_debugreg",}, |
3d8d996e SD |
99 | {NULL} /* Terminator */ |
100 | }; | |
101 | ||
2d14e39d | 102 | #ifdef __ARCH_WANT_KPROBES_INSN_SLOT |
9ec4b1f3 AM |
103 | /* |
104 | * kprobe->ainsn.insn points to the copy of the instruction to be | |
105 | * single-stepped. x86_64, POWER4 and above have no-exec support and | |
106 | * stepping on the instruction on a vmalloced/kmalloced/data page | |
107 | * is a recipe for disaster | |
108 | */ | |
9ec4b1f3 | 109 | struct kprobe_insn_page { |
c5cb5a2d | 110 | struct list_head list; |
9ec4b1f3 | 111 | kprobe_opcode_t *insns; /* Page of instruction slots */ |
af96397d | 112 | struct kprobe_insn_cache *cache; |
9ec4b1f3 | 113 | int nused; |
b4c6c34a | 114 | int ngarbage; |
4610ee1d | 115 | char slot_used[]; |
9ec4b1f3 AM |
116 | }; |
117 | ||
4610ee1d MH |
118 | #define KPROBE_INSN_PAGE_SIZE(slots) \ |
119 | (offsetof(struct kprobe_insn_page, slot_used) + \ | |
120 | (sizeof(char) * (slots))) | |
121 | ||
4610ee1d MH |
122 | static int slots_per_page(struct kprobe_insn_cache *c) |
123 | { | |
124 | return PAGE_SIZE/(c->insn_size * sizeof(kprobe_opcode_t)); | |
125 | } | |
126 | ||
ab40c5c6 MH |
127 | enum kprobe_slot_state { |
128 | SLOT_CLEAN = 0, | |
129 | SLOT_DIRTY = 1, | |
130 | SLOT_USED = 2, | |
131 | }; | |
132 | ||
af96397d HC |
133 | static void *alloc_insn_page(void) |
134 | { | |
135 | return module_alloc(PAGE_SIZE); | |
136 | } | |
137 | ||
138 | static void free_insn_page(void *page) | |
139 | { | |
140 | module_free(NULL, page); | |
141 | } | |
142 | ||
c802d64a HC |
143 | struct kprobe_insn_cache kprobe_insn_slots = { |
144 | .mutex = __MUTEX_INITIALIZER(kprobe_insn_slots.mutex), | |
af96397d HC |
145 | .alloc = alloc_insn_page, |
146 | .free = free_insn_page, | |
4610ee1d MH |
147 | .pages = LIST_HEAD_INIT(kprobe_insn_slots.pages), |
148 | .insn_size = MAX_INSN_SIZE, | |
149 | .nr_garbage = 0, | |
150 | }; | |
151 | static int __kprobes collect_garbage_slots(struct kprobe_insn_cache *c); | |
b4c6c34a | 152 | |
9ec4b1f3 | 153 | /** |
12941560 | 154 | * __get_insn_slot() - Find a slot on an executable page for an instruction. |
9ec4b1f3 AM |
155 | * We allocate an executable page if there's no room on existing ones. |
156 | */ | |
c802d64a | 157 | kprobe_opcode_t __kprobes *__get_insn_slot(struct kprobe_insn_cache *c) |
9ec4b1f3 AM |
158 | { |
159 | struct kprobe_insn_page *kip; | |
c802d64a | 160 | kprobe_opcode_t *slot = NULL; |
9ec4b1f3 | 161 | |
c802d64a | 162 | mutex_lock(&c->mutex); |
6f716acd | 163 | retry: |
4610ee1d MH |
164 | list_for_each_entry(kip, &c->pages, list) { |
165 | if (kip->nused < slots_per_page(c)) { | |
9ec4b1f3 | 166 | int i; |
4610ee1d | 167 | for (i = 0; i < slots_per_page(c); i++) { |
ab40c5c6 MH |
168 | if (kip->slot_used[i] == SLOT_CLEAN) { |
169 | kip->slot_used[i] = SLOT_USED; | |
9ec4b1f3 | 170 | kip->nused++; |
c802d64a HC |
171 | slot = kip->insns + (i * c->insn_size); |
172 | goto out; | |
9ec4b1f3 AM |
173 | } |
174 | } | |
4610ee1d MH |
175 | /* kip->nused is broken. Fix it. */ |
176 | kip->nused = slots_per_page(c); | |
177 | WARN_ON(1); | |
9ec4b1f3 AM |
178 | } |
179 | } | |
180 | ||
b4c6c34a | 181 | /* If there are any garbage slots, collect it and try again. */ |
4610ee1d | 182 | if (c->nr_garbage && collect_garbage_slots(c) == 0) |
b4c6c34a | 183 | goto retry; |
4610ee1d MH |
184 | |
185 | /* All out of space. Need to allocate a new page. */ | |
186 | kip = kmalloc(KPROBE_INSN_PAGE_SIZE(slots_per_page(c)), GFP_KERNEL); | |
6f716acd | 187 | if (!kip) |
c802d64a | 188 | goto out; |
9ec4b1f3 AM |
189 | |
190 | /* | |
191 | * Use module_alloc so this page is within +/- 2GB of where the | |
192 | * kernel image and loaded module images reside. This is required | |
193 | * so x86_64 can correctly handle the %rip-relative fixups. | |
194 | */ | |
af96397d | 195 | kip->insns = c->alloc(); |
9ec4b1f3 AM |
196 | if (!kip->insns) { |
197 | kfree(kip); | |
c802d64a | 198 | goto out; |
9ec4b1f3 | 199 | } |
c5cb5a2d | 200 | INIT_LIST_HEAD(&kip->list); |
4610ee1d | 201 | memset(kip->slot_used, SLOT_CLEAN, slots_per_page(c)); |
ab40c5c6 | 202 | kip->slot_used[0] = SLOT_USED; |
9ec4b1f3 | 203 | kip->nused = 1; |
b4c6c34a | 204 | kip->ngarbage = 0; |
af96397d | 205 | kip->cache = c; |
4610ee1d | 206 | list_add(&kip->list, &c->pages); |
c802d64a HC |
207 | slot = kip->insns; |
208 | out: | |
209 | mutex_unlock(&c->mutex); | |
210 | return slot; | |
12941560 MH |
211 | } |
212 | ||
b4c6c34a MH |
213 | /* Return 1 if all garbages are collected, otherwise 0. */ |
214 | static int __kprobes collect_one_slot(struct kprobe_insn_page *kip, int idx) | |
215 | { | |
ab40c5c6 | 216 | kip->slot_used[idx] = SLOT_CLEAN; |
b4c6c34a MH |
217 | kip->nused--; |
218 | if (kip->nused == 0) { | |
219 | /* | |
220 | * Page is no longer in use. Free it unless | |
221 | * it's the last one. We keep the last one | |
222 | * so as not to have to set it up again the | |
223 | * next time somebody inserts a probe. | |
224 | */ | |
4610ee1d | 225 | if (!list_is_singular(&kip->list)) { |
c5cb5a2d | 226 | list_del(&kip->list); |
af96397d | 227 | kip->cache->free(kip->insns); |
b4c6c34a MH |
228 | kfree(kip); |
229 | } | |
230 | return 1; | |
231 | } | |
232 | return 0; | |
233 | } | |
234 | ||
4610ee1d | 235 | static int __kprobes collect_garbage_slots(struct kprobe_insn_cache *c) |
b4c6c34a | 236 | { |
c5cb5a2d | 237 | struct kprobe_insn_page *kip, *next; |
b4c6c34a | 238 | |
615d0ebb MH |
239 | /* Ensure no-one is interrupted on the garbages */ |
240 | synchronize_sched(); | |
b4c6c34a | 241 | |
4610ee1d | 242 | list_for_each_entry_safe(kip, next, &c->pages, list) { |
b4c6c34a | 243 | int i; |
b4c6c34a MH |
244 | if (kip->ngarbage == 0) |
245 | continue; | |
246 | kip->ngarbage = 0; /* we will collect all garbages */ | |
4610ee1d | 247 | for (i = 0; i < slots_per_page(c); i++) { |
ab40c5c6 | 248 | if (kip->slot_used[i] == SLOT_DIRTY && |
b4c6c34a MH |
249 | collect_one_slot(kip, i)) |
250 | break; | |
251 | } | |
252 | } | |
4610ee1d | 253 | c->nr_garbage = 0; |
b4c6c34a MH |
254 | return 0; |
255 | } | |
256 | ||
c802d64a HC |
257 | void __kprobes __free_insn_slot(struct kprobe_insn_cache *c, |
258 | kprobe_opcode_t *slot, int dirty) | |
9ec4b1f3 AM |
259 | { |
260 | struct kprobe_insn_page *kip; | |
9ec4b1f3 | 261 | |
c802d64a | 262 | mutex_lock(&c->mutex); |
4610ee1d | 263 | list_for_each_entry(kip, &c->pages, list) { |
83ff56f4 MH |
264 | long idx = ((long)slot - (long)kip->insns) / |
265 | (c->insn_size * sizeof(kprobe_opcode_t)); | |
4610ee1d MH |
266 | if (idx >= 0 && idx < slots_per_page(c)) { |
267 | WARN_ON(kip->slot_used[idx] != SLOT_USED); | |
b4c6c34a | 268 | if (dirty) { |
4610ee1d | 269 | kip->slot_used[idx] = SLOT_DIRTY; |
b4c6c34a | 270 | kip->ngarbage++; |
4610ee1d MH |
271 | if (++c->nr_garbage > slots_per_page(c)) |
272 | collect_garbage_slots(c); | |
c5cb5a2d | 273 | } else |
4610ee1d | 274 | collect_one_slot(kip, idx); |
c802d64a | 275 | goto out; |
9ec4b1f3 AM |
276 | } |
277 | } | |
4610ee1d MH |
278 | /* Could not free this slot. */ |
279 | WARN_ON(1); | |
c802d64a HC |
280 | out: |
281 | mutex_unlock(&c->mutex); | |
4610ee1d | 282 | } |
6f716acd | 283 | |
afd66255 MH |
284 | #ifdef CONFIG_OPTPROBES |
285 | /* For optimized_kprobe buffer */ | |
c802d64a HC |
286 | struct kprobe_insn_cache kprobe_optinsn_slots = { |
287 | .mutex = __MUTEX_INITIALIZER(kprobe_optinsn_slots.mutex), | |
af96397d HC |
288 | .alloc = alloc_insn_page, |
289 | .free = free_insn_page, | |
afd66255 MH |
290 | .pages = LIST_HEAD_INIT(kprobe_optinsn_slots.pages), |
291 | /* .insn_size is initialized later */ | |
292 | .nr_garbage = 0, | |
293 | }; | |
afd66255 | 294 | #endif |
2d14e39d | 295 | #endif |
9ec4b1f3 | 296 | |
e6584523 AM |
297 | /* We have preemption disabled.. so it is safe to use __ versions */ |
298 | static inline void set_kprobe_instance(struct kprobe *kp) | |
299 | { | |
b76834bc | 300 | __this_cpu_write(kprobe_instance, kp); |
e6584523 AM |
301 | } |
302 | ||
303 | static inline void reset_kprobe_instance(void) | |
304 | { | |
b76834bc | 305 | __this_cpu_write(kprobe_instance, NULL); |
e6584523 AM |
306 | } |
307 | ||
3516a460 AM |
308 | /* |
309 | * This routine is called either: | |
49a2a1b8 | 310 | * - under the kprobe_mutex - during kprobe_[un]register() |
3516a460 | 311 | * OR |
d217d545 | 312 | * - with preemption disabled - from arch/xxx/kernel/kprobes.c |
3516a460 | 313 | */ |
d0aaff97 | 314 | struct kprobe __kprobes *get_kprobe(void *addr) |
1da177e4 LT |
315 | { |
316 | struct hlist_head *head; | |
3516a460 | 317 | struct kprobe *p; |
1da177e4 LT |
318 | |
319 | head = &kprobe_table[hash_ptr(addr, KPROBE_HASH_BITS)]; | |
b67bfe0d | 320 | hlist_for_each_entry_rcu(p, head, hlist) { |
1da177e4 LT |
321 | if (p->addr == addr) |
322 | return p; | |
323 | } | |
afd66255 | 324 | |
1da177e4 LT |
325 | return NULL; |
326 | } | |
327 | ||
afd66255 MH |
328 | static int __kprobes aggr_pre_handler(struct kprobe *p, struct pt_regs *regs); |
329 | ||
330 | /* Return true if the kprobe is an aggregator */ | |
331 | static inline int kprobe_aggrprobe(struct kprobe *p) | |
332 | { | |
333 | return p->pre_handler == aggr_pre_handler; | |
334 | } | |
335 | ||
6274de49 MH |
336 | /* Return true(!0) if the kprobe is unused */ |
337 | static inline int kprobe_unused(struct kprobe *p) | |
338 | { | |
339 | return kprobe_aggrprobe(p) && kprobe_disabled(p) && | |
340 | list_empty(&p->list); | |
341 | } | |
342 | ||
afd66255 MH |
343 | /* |
344 | * Keep all fields in the kprobe consistent | |
345 | */ | |
6d8e40a8 | 346 | static inline void copy_kprobe(struct kprobe *ap, struct kprobe *p) |
afd66255 | 347 | { |
6d8e40a8 MH |
348 | memcpy(&p->opcode, &ap->opcode, sizeof(kprobe_opcode_t)); |
349 | memcpy(&p->ainsn, &ap->ainsn, sizeof(struct arch_specific_insn)); | |
afd66255 MH |
350 | } |
351 | ||
352 | #ifdef CONFIG_OPTPROBES | |
b2be84df MH |
353 | /* NOTE: change this value only with kprobe_mutex held */ |
354 | static bool kprobes_allow_optimization; | |
355 | ||
afd66255 MH |
356 | /* |
357 | * Call all pre_handler on the list, but ignores its return value. | |
358 | * This must be called from arch-dep optimized caller. | |
359 | */ | |
360 | void __kprobes opt_pre_handler(struct kprobe *p, struct pt_regs *regs) | |
361 | { | |
362 | struct kprobe *kp; | |
363 | ||
364 | list_for_each_entry_rcu(kp, &p->list, list) { | |
365 | if (kp->pre_handler && likely(!kprobe_disabled(kp))) { | |
366 | set_kprobe_instance(kp); | |
367 | kp->pre_handler(kp, regs); | |
368 | } | |
369 | reset_kprobe_instance(); | |
370 | } | |
371 | } | |
372 | ||
6274de49 MH |
373 | /* Free optimized instructions and optimized_kprobe */ |
374 | static __kprobes void free_aggr_kprobe(struct kprobe *p) | |
375 | { | |
376 | struct optimized_kprobe *op; | |
377 | ||
378 | op = container_of(p, struct optimized_kprobe, kp); | |
379 | arch_remove_optimized_kprobe(op); | |
380 | arch_remove_kprobe(p); | |
381 | kfree(op); | |
382 | } | |
383 | ||
afd66255 MH |
384 | /* Return true(!0) if the kprobe is ready for optimization. */ |
385 | static inline int kprobe_optready(struct kprobe *p) | |
386 | { | |
387 | struct optimized_kprobe *op; | |
388 | ||
389 | if (kprobe_aggrprobe(p)) { | |
390 | op = container_of(p, struct optimized_kprobe, kp); | |
391 | return arch_prepared_optinsn(&op->optinsn); | |
392 | } | |
393 | ||
394 | return 0; | |
395 | } | |
396 | ||
6274de49 MH |
397 | /* Return true(!0) if the kprobe is disarmed. Note: p must be on hash list */ |
398 | static inline int kprobe_disarmed(struct kprobe *p) | |
399 | { | |
400 | struct optimized_kprobe *op; | |
401 | ||
402 | /* If kprobe is not aggr/opt probe, just return kprobe is disabled */ | |
403 | if (!kprobe_aggrprobe(p)) | |
404 | return kprobe_disabled(p); | |
405 | ||
406 | op = container_of(p, struct optimized_kprobe, kp); | |
407 | ||
408 | return kprobe_disabled(p) && list_empty(&op->list); | |
409 | } | |
410 | ||
411 | /* Return true(!0) if the probe is queued on (un)optimizing lists */ | |
412 | static int __kprobes kprobe_queued(struct kprobe *p) | |
413 | { | |
414 | struct optimized_kprobe *op; | |
415 | ||
416 | if (kprobe_aggrprobe(p)) { | |
417 | op = container_of(p, struct optimized_kprobe, kp); | |
418 | if (!list_empty(&op->list)) | |
419 | return 1; | |
420 | } | |
421 | return 0; | |
422 | } | |
423 | ||
afd66255 MH |
424 | /* |
425 | * Return an optimized kprobe whose optimizing code replaces | |
426 | * instructions including addr (exclude breakpoint). | |
427 | */ | |
6376b229 | 428 | static struct kprobe *__kprobes get_optimized_kprobe(unsigned long addr) |
afd66255 MH |
429 | { |
430 | int i; | |
431 | struct kprobe *p = NULL; | |
432 | struct optimized_kprobe *op; | |
433 | ||
434 | /* Don't check i == 0, since that is a breakpoint case. */ | |
435 | for (i = 1; !p && i < MAX_OPTIMIZED_LENGTH; i++) | |
436 | p = get_kprobe((void *)(addr - i)); | |
437 | ||
438 | if (p && kprobe_optready(p)) { | |
439 | op = container_of(p, struct optimized_kprobe, kp); | |
440 | if (arch_within_optimized_kprobe(op, addr)) | |
441 | return p; | |
442 | } | |
443 | ||
444 | return NULL; | |
445 | } | |
446 | ||
447 | /* Optimization staging list, protected by kprobe_mutex */ | |
448 | static LIST_HEAD(optimizing_list); | |
6274de49 | 449 | static LIST_HEAD(unoptimizing_list); |
7b959fc5 | 450 | static LIST_HEAD(freeing_list); |
afd66255 MH |
451 | |
452 | static void kprobe_optimizer(struct work_struct *work); | |
453 | static DECLARE_DELAYED_WORK(optimizing_work, kprobe_optimizer); | |
454 | #define OPTIMIZE_DELAY 5 | |
455 | ||
61f4e13f MH |
456 | /* |
457 | * Optimize (replace a breakpoint with a jump) kprobes listed on | |
458 | * optimizing_list. | |
459 | */ | |
460 | static __kprobes void do_optimize_kprobes(void) | |
afd66255 | 461 | { |
6274de49 MH |
462 | /* Optimization never be done when disarmed */ |
463 | if (kprobes_all_disarmed || !kprobes_allow_optimization || | |
464 | list_empty(&optimizing_list)) | |
465 | return; | |
466 | ||
afd66255 MH |
467 | /* |
468 | * The optimization/unoptimization refers online_cpus via | |
469 | * stop_machine() and cpu-hotplug modifies online_cpus. | |
470 | * And same time, text_mutex will be held in cpu-hotplug and here. | |
471 | * This combination can cause a deadlock (cpu-hotplug try to lock | |
472 | * text_mutex but stop_machine can not be done because online_cpus | |
473 | * has been changed) | |
474 | * To avoid this deadlock, we need to call get_online_cpus() | |
475 | * for preventing cpu-hotplug outside of text_mutex locking. | |
476 | */ | |
477 | get_online_cpus(); | |
478 | mutex_lock(&text_mutex); | |
cd7ebe22 | 479 | arch_optimize_kprobes(&optimizing_list); |
afd66255 MH |
480 | mutex_unlock(&text_mutex); |
481 | put_online_cpus(); | |
61f4e13f MH |
482 | } |
483 | ||
6274de49 MH |
484 | /* |
485 | * Unoptimize (replace a jump with a breakpoint and remove the breakpoint | |
486 | * if need) kprobes listed on unoptimizing_list. | |
487 | */ | |
7b959fc5 | 488 | static __kprobes void do_unoptimize_kprobes(void) |
6274de49 MH |
489 | { |
490 | struct optimized_kprobe *op, *tmp; | |
491 | ||
492 | /* Unoptimization must be done anytime */ | |
493 | if (list_empty(&unoptimizing_list)) | |
494 | return; | |
495 | ||
496 | /* Ditto to do_optimize_kprobes */ | |
497 | get_online_cpus(); | |
498 | mutex_lock(&text_mutex); | |
7b959fc5 | 499 | arch_unoptimize_kprobes(&unoptimizing_list, &freeing_list); |
f984ba4e | 500 | /* Loop free_list for disarming */ |
7b959fc5 | 501 | list_for_each_entry_safe(op, tmp, &freeing_list, list) { |
6274de49 MH |
502 | /* Disarm probes if marked disabled */ |
503 | if (kprobe_disabled(&op->kp)) | |
504 | arch_disarm_kprobe(&op->kp); | |
505 | if (kprobe_unused(&op->kp)) { | |
506 | /* | |
507 | * Remove unused probes from hash list. After waiting | |
508 | * for synchronization, these probes are reclaimed. | |
509 | * (reclaiming is done by do_free_cleaned_kprobes.) | |
510 | */ | |
511 | hlist_del_rcu(&op->kp.hlist); | |
6274de49 MH |
512 | } else |
513 | list_del_init(&op->list); | |
514 | } | |
515 | mutex_unlock(&text_mutex); | |
516 | put_online_cpus(); | |
517 | } | |
518 | ||
519 | /* Reclaim all kprobes on the free_list */ | |
7b959fc5 | 520 | static __kprobes void do_free_cleaned_kprobes(void) |
6274de49 MH |
521 | { |
522 | struct optimized_kprobe *op, *tmp; | |
523 | ||
7b959fc5 | 524 | list_for_each_entry_safe(op, tmp, &freeing_list, list) { |
6274de49 MH |
525 | BUG_ON(!kprobe_unused(&op->kp)); |
526 | list_del_init(&op->list); | |
527 | free_aggr_kprobe(&op->kp); | |
528 | } | |
529 | } | |
530 | ||
531 | /* Start optimizer after OPTIMIZE_DELAY passed */ | |
532 | static __kprobes void kick_kprobe_optimizer(void) | |
533 | { | |
ad72b3be | 534 | schedule_delayed_work(&optimizing_work, OPTIMIZE_DELAY); |
6274de49 MH |
535 | } |
536 | ||
61f4e13f MH |
537 | /* Kprobe jump optimizer */ |
538 | static __kprobes void kprobe_optimizer(struct work_struct *work) | |
539 | { | |
72ef3794 | 540 | mutex_lock(&kprobe_mutex); |
61f4e13f MH |
541 | /* Lock modules while optimizing kprobes */ |
542 | mutex_lock(&module_mutex); | |
61f4e13f MH |
543 | |
544 | /* | |
6274de49 MH |
545 | * Step 1: Unoptimize kprobes and collect cleaned (unused and disarmed) |
546 | * kprobes before waiting for quiesence period. | |
547 | */ | |
7b959fc5 | 548 | do_unoptimize_kprobes(); |
6274de49 MH |
549 | |
550 | /* | |
551 | * Step 2: Wait for quiesence period to ensure all running interrupts | |
61f4e13f MH |
552 | * are done. Because optprobe may modify multiple instructions |
553 | * there is a chance that Nth instruction is interrupted. In that | |
554 | * case, running interrupt can return to 2nd-Nth byte of jump | |
555 | * instruction. This wait is for avoiding it. | |
556 | */ | |
557 | synchronize_sched(); | |
558 | ||
6274de49 | 559 | /* Step 3: Optimize kprobes after quiesence period */ |
61f4e13f | 560 | do_optimize_kprobes(); |
6274de49 MH |
561 | |
562 | /* Step 4: Free cleaned kprobes after quiesence period */ | |
7b959fc5 | 563 | do_free_cleaned_kprobes(); |
6274de49 | 564 | |
afd66255 | 565 | mutex_unlock(&module_mutex); |
72ef3794 | 566 | mutex_unlock(&kprobe_mutex); |
6274de49 | 567 | |
cd7ebe22 | 568 | /* Step 5: Kick optimizer again if needed */ |
f984ba4e | 569 | if (!list_empty(&optimizing_list) || !list_empty(&unoptimizing_list)) |
cd7ebe22 | 570 | kick_kprobe_optimizer(); |
6274de49 MH |
571 | } |
572 | ||
573 | /* Wait for completing optimization and unoptimization */ | |
574 | static __kprobes void wait_for_kprobe_optimizer(void) | |
575 | { | |
ad72b3be TH |
576 | mutex_lock(&kprobe_mutex); |
577 | ||
578 | while (!list_empty(&optimizing_list) || !list_empty(&unoptimizing_list)) { | |
579 | mutex_unlock(&kprobe_mutex); | |
580 | ||
581 | /* this will also make optimizing_work execute immmediately */ | |
582 | flush_delayed_work(&optimizing_work); | |
583 | /* @optimizing_work might not have been queued yet, relax */ | |
584 | cpu_relax(); | |
585 | ||
586 | mutex_lock(&kprobe_mutex); | |
587 | } | |
588 | ||
589 | mutex_unlock(&kprobe_mutex); | |
afd66255 MH |
590 | } |
591 | ||
592 | /* Optimize kprobe if p is ready to be optimized */ | |
593 | static __kprobes void optimize_kprobe(struct kprobe *p) | |
594 | { | |
595 | struct optimized_kprobe *op; | |
596 | ||
597 | /* Check if the kprobe is disabled or not ready for optimization. */ | |
b2be84df | 598 | if (!kprobe_optready(p) || !kprobes_allow_optimization || |
afd66255 MH |
599 | (kprobe_disabled(p) || kprobes_all_disarmed)) |
600 | return; | |
601 | ||
602 | /* Both of break_handler and post_handler are not supported. */ | |
603 | if (p->break_handler || p->post_handler) | |
604 | return; | |
605 | ||
606 | op = container_of(p, struct optimized_kprobe, kp); | |
607 | ||
608 | /* Check there is no other kprobes at the optimized instructions */ | |
609 | if (arch_check_optimized_kprobe(op) < 0) | |
610 | return; | |
611 | ||
612 | /* Check if it is already optimized. */ | |
613 | if (op->kp.flags & KPROBE_FLAG_OPTIMIZED) | |
614 | return; | |
afd66255 | 615 | op->kp.flags |= KPROBE_FLAG_OPTIMIZED; |
6274de49 MH |
616 | |
617 | if (!list_empty(&op->list)) | |
618 | /* This is under unoptimizing. Just dequeue the probe */ | |
619 | list_del_init(&op->list); | |
620 | else { | |
621 | list_add(&op->list, &optimizing_list); | |
622 | kick_kprobe_optimizer(); | |
623 | } | |
624 | } | |
625 | ||
626 | /* Short cut to direct unoptimizing */ | |
627 | static __kprobes void force_unoptimize_kprobe(struct optimized_kprobe *op) | |
628 | { | |
629 | get_online_cpus(); | |
630 | arch_unoptimize_kprobe(op); | |
631 | put_online_cpus(); | |
632 | if (kprobe_disabled(&op->kp)) | |
633 | arch_disarm_kprobe(&op->kp); | |
afd66255 MH |
634 | } |
635 | ||
636 | /* Unoptimize a kprobe if p is optimized */ | |
6274de49 | 637 | static __kprobes void unoptimize_kprobe(struct kprobe *p, bool force) |
afd66255 MH |
638 | { |
639 | struct optimized_kprobe *op; | |
640 | ||
6274de49 MH |
641 | if (!kprobe_aggrprobe(p) || kprobe_disarmed(p)) |
642 | return; /* This is not an optprobe nor optimized */ | |
643 | ||
644 | op = container_of(p, struct optimized_kprobe, kp); | |
645 | if (!kprobe_optimized(p)) { | |
646 | /* Unoptimized or unoptimizing case */ | |
647 | if (force && !list_empty(&op->list)) { | |
648 | /* | |
649 | * Only if this is unoptimizing kprobe and forced, | |
650 | * forcibly unoptimize it. (No need to unoptimize | |
651 | * unoptimized kprobe again :) | |
652 | */ | |
afd66255 | 653 | list_del_init(&op->list); |
6274de49 MH |
654 | force_unoptimize_kprobe(op); |
655 | } | |
656 | return; | |
657 | } | |
658 | ||
659 | op->kp.flags &= ~KPROBE_FLAG_OPTIMIZED; | |
660 | if (!list_empty(&op->list)) { | |
661 | /* Dequeue from the optimization queue */ | |
662 | list_del_init(&op->list); | |
663 | return; | |
664 | } | |
665 | /* Optimized kprobe case */ | |
666 | if (force) | |
667 | /* Forcibly update the code: this is a special case */ | |
668 | force_unoptimize_kprobe(op); | |
669 | else { | |
670 | list_add(&op->list, &unoptimizing_list); | |
671 | kick_kprobe_optimizer(); | |
afd66255 MH |
672 | } |
673 | } | |
674 | ||
0490cd1f MH |
675 | /* Cancel unoptimizing for reusing */ |
676 | static void reuse_unused_kprobe(struct kprobe *ap) | |
677 | { | |
678 | struct optimized_kprobe *op; | |
679 | ||
680 | BUG_ON(!kprobe_unused(ap)); | |
681 | /* | |
682 | * Unused kprobe MUST be on the way of delayed unoptimizing (means | |
683 | * there is still a relative jump) and disabled. | |
684 | */ | |
685 | op = container_of(ap, struct optimized_kprobe, kp); | |
686 | if (unlikely(list_empty(&op->list))) | |
687 | printk(KERN_WARNING "Warning: found a stray unused " | |
688 | "aggrprobe@%p\n", ap->addr); | |
689 | /* Enable the probe again */ | |
690 | ap->flags &= ~KPROBE_FLAG_DISABLED; | |
691 | /* Optimize it again (remove from op->list) */ | |
692 | BUG_ON(!kprobe_optready(ap)); | |
693 | optimize_kprobe(ap); | |
694 | } | |
695 | ||
afd66255 MH |
696 | /* Remove optimized instructions */ |
697 | static void __kprobes kill_optimized_kprobe(struct kprobe *p) | |
698 | { | |
699 | struct optimized_kprobe *op; | |
700 | ||
701 | op = container_of(p, struct optimized_kprobe, kp); | |
6274de49 MH |
702 | if (!list_empty(&op->list)) |
703 | /* Dequeue from the (un)optimization queue */ | |
afd66255 | 704 | list_del_init(&op->list); |
6274de49 | 705 | op->kp.flags &= ~KPROBE_FLAG_OPTIMIZED; |
7b959fc5 MH |
706 | |
707 | if (kprobe_unused(p)) { | |
708 | /* Enqueue if it is unused */ | |
709 | list_add(&op->list, &freeing_list); | |
710 | /* | |
711 | * Remove unused probes from the hash list. After waiting | |
712 | * for synchronization, this probe is reclaimed. | |
713 | * (reclaiming is done by do_free_cleaned_kprobes().) | |
714 | */ | |
715 | hlist_del_rcu(&op->kp.hlist); | |
716 | } | |
717 | ||
6274de49 | 718 | /* Don't touch the code, because it is already freed. */ |
afd66255 MH |
719 | arch_remove_optimized_kprobe(op); |
720 | } | |
721 | ||
722 | /* Try to prepare optimized instructions */ | |
723 | static __kprobes void prepare_optimized_kprobe(struct kprobe *p) | |
724 | { | |
725 | struct optimized_kprobe *op; | |
726 | ||
727 | op = container_of(p, struct optimized_kprobe, kp); | |
728 | arch_prepare_optimized_kprobe(op); | |
729 | } | |
730 | ||
afd66255 MH |
731 | /* Allocate new optimized_kprobe and try to prepare optimized instructions */ |
732 | static __kprobes struct kprobe *alloc_aggr_kprobe(struct kprobe *p) | |
733 | { | |
734 | struct optimized_kprobe *op; | |
735 | ||
736 | op = kzalloc(sizeof(struct optimized_kprobe), GFP_KERNEL); | |
737 | if (!op) | |
738 | return NULL; | |
739 | ||
740 | INIT_LIST_HEAD(&op->list); | |
741 | op->kp.addr = p->addr; | |
742 | arch_prepare_optimized_kprobe(op); | |
743 | ||
744 | return &op->kp; | |
745 | } | |
746 | ||
747 | static void __kprobes init_aggr_kprobe(struct kprobe *ap, struct kprobe *p); | |
748 | ||
749 | /* | |
750 | * Prepare an optimized_kprobe and optimize it | |
751 | * NOTE: p must be a normal registered kprobe | |
752 | */ | |
753 | static __kprobes void try_to_optimize_kprobe(struct kprobe *p) | |
754 | { | |
755 | struct kprobe *ap; | |
756 | struct optimized_kprobe *op; | |
757 | ||
ae6aa16f MH |
758 | /* Impossible to optimize ftrace-based kprobe */ |
759 | if (kprobe_ftrace(p)) | |
760 | return; | |
761 | ||
25764288 MH |
762 | /* For preparing optimization, jump_label_text_reserved() is called */ |
763 | jump_label_lock(); | |
764 | mutex_lock(&text_mutex); | |
765 | ||
afd66255 MH |
766 | ap = alloc_aggr_kprobe(p); |
767 | if (!ap) | |
25764288 | 768 | goto out; |
afd66255 MH |
769 | |
770 | op = container_of(ap, struct optimized_kprobe, kp); | |
771 | if (!arch_prepared_optinsn(&op->optinsn)) { | |
772 | /* If failed to setup optimizing, fallback to kprobe */ | |
6274de49 MH |
773 | arch_remove_optimized_kprobe(op); |
774 | kfree(op); | |
25764288 | 775 | goto out; |
afd66255 MH |
776 | } |
777 | ||
778 | init_aggr_kprobe(ap, p); | |
25764288 MH |
779 | optimize_kprobe(ap); /* This just kicks optimizer thread */ |
780 | ||
781 | out: | |
782 | mutex_unlock(&text_mutex); | |
783 | jump_label_unlock(); | |
afd66255 MH |
784 | } |
785 | ||
b2be84df MH |
786 | #ifdef CONFIG_SYSCTL |
787 | static void __kprobes optimize_all_kprobes(void) | |
788 | { | |
789 | struct hlist_head *head; | |
b2be84df MH |
790 | struct kprobe *p; |
791 | unsigned int i; | |
792 | ||
5c51543b | 793 | mutex_lock(&kprobe_mutex); |
b2be84df MH |
794 | /* If optimization is already allowed, just return */ |
795 | if (kprobes_allow_optimization) | |
5c51543b | 796 | goto out; |
b2be84df MH |
797 | |
798 | kprobes_allow_optimization = true; | |
b2be84df MH |
799 | for (i = 0; i < KPROBE_TABLE_SIZE; i++) { |
800 | head = &kprobe_table[i]; | |
b67bfe0d | 801 | hlist_for_each_entry_rcu(p, head, hlist) |
b2be84df MH |
802 | if (!kprobe_disabled(p)) |
803 | optimize_kprobe(p); | |
804 | } | |
b2be84df | 805 | printk(KERN_INFO "Kprobes globally optimized\n"); |
5c51543b MH |
806 | out: |
807 | mutex_unlock(&kprobe_mutex); | |
b2be84df MH |
808 | } |
809 | ||
810 | static void __kprobes unoptimize_all_kprobes(void) | |
811 | { | |
812 | struct hlist_head *head; | |
b2be84df MH |
813 | struct kprobe *p; |
814 | unsigned int i; | |
815 | ||
5c51543b | 816 | mutex_lock(&kprobe_mutex); |
b2be84df | 817 | /* If optimization is already prohibited, just return */ |
5c51543b MH |
818 | if (!kprobes_allow_optimization) { |
819 | mutex_unlock(&kprobe_mutex); | |
b2be84df | 820 | return; |
5c51543b | 821 | } |
b2be84df MH |
822 | |
823 | kprobes_allow_optimization = false; | |
b2be84df MH |
824 | for (i = 0; i < KPROBE_TABLE_SIZE; i++) { |
825 | head = &kprobe_table[i]; | |
b67bfe0d | 826 | hlist_for_each_entry_rcu(p, head, hlist) { |
b2be84df | 827 | if (!kprobe_disabled(p)) |
6274de49 | 828 | unoptimize_kprobe(p, false); |
b2be84df MH |
829 | } |
830 | } | |
5c51543b MH |
831 | mutex_unlock(&kprobe_mutex); |
832 | ||
6274de49 MH |
833 | /* Wait for unoptimizing completion */ |
834 | wait_for_kprobe_optimizer(); | |
835 | printk(KERN_INFO "Kprobes globally unoptimized\n"); | |
b2be84df MH |
836 | } |
837 | ||
5c51543b | 838 | static DEFINE_MUTEX(kprobe_sysctl_mutex); |
b2be84df MH |
839 | int sysctl_kprobes_optimization; |
840 | int proc_kprobes_optimization_handler(struct ctl_table *table, int write, | |
841 | void __user *buffer, size_t *length, | |
842 | loff_t *ppos) | |
843 | { | |
844 | int ret; | |
845 | ||
5c51543b | 846 | mutex_lock(&kprobe_sysctl_mutex); |
b2be84df MH |
847 | sysctl_kprobes_optimization = kprobes_allow_optimization ? 1 : 0; |
848 | ret = proc_dointvec_minmax(table, write, buffer, length, ppos); | |
849 | ||
850 | if (sysctl_kprobes_optimization) | |
851 | optimize_all_kprobes(); | |
852 | else | |
853 | unoptimize_all_kprobes(); | |
5c51543b | 854 | mutex_unlock(&kprobe_sysctl_mutex); |
b2be84df MH |
855 | |
856 | return ret; | |
857 | } | |
858 | #endif /* CONFIG_SYSCTL */ | |
859 | ||
6274de49 | 860 | /* Put a breakpoint for a probe. Must be called with text_mutex locked */ |
afd66255 MH |
861 | static void __kprobes __arm_kprobe(struct kprobe *p) |
862 | { | |
6d8e40a8 | 863 | struct kprobe *_p; |
afd66255 MH |
864 | |
865 | /* Check collision with other optimized kprobes */ | |
6d8e40a8 MH |
866 | _p = get_optimized_kprobe((unsigned long)p->addr); |
867 | if (unlikely(_p)) | |
6274de49 MH |
868 | /* Fallback to unoptimized kprobe */ |
869 | unoptimize_kprobe(_p, true); | |
afd66255 MH |
870 | |
871 | arch_arm_kprobe(p); | |
872 | optimize_kprobe(p); /* Try to optimize (add kprobe to a list) */ | |
873 | } | |
874 | ||
6274de49 MH |
875 | /* Remove the breakpoint of a probe. Must be called with text_mutex locked */ |
876 | static void __kprobes __disarm_kprobe(struct kprobe *p, bool reopt) | |
afd66255 | 877 | { |
6d8e40a8 | 878 | struct kprobe *_p; |
afd66255 | 879 | |
6274de49 | 880 | unoptimize_kprobe(p, false); /* Try to unoptimize */ |
afd66255 | 881 | |
6274de49 MH |
882 | if (!kprobe_queued(p)) { |
883 | arch_disarm_kprobe(p); | |
884 | /* If another kprobe was blocked, optimize it. */ | |
885 | _p = get_optimized_kprobe((unsigned long)p->addr); | |
886 | if (unlikely(_p) && reopt) | |
887 | optimize_kprobe(_p); | |
888 | } | |
889 | /* TODO: reoptimize others after unoptimized this probe */ | |
afd66255 MH |
890 | } |
891 | ||
892 | #else /* !CONFIG_OPTPROBES */ | |
893 | ||
894 | #define optimize_kprobe(p) do {} while (0) | |
6274de49 | 895 | #define unoptimize_kprobe(p, f) do {} while (0) |
afd66255 MH |
896 | #define kill_optimized_kprobe(p) do {} while (0) |
897 | #define prepare_optimized_kprobe(p) do {} while (0) | |
898 | #define try_to_optimize_kprobe(p) do {} while (0) | |
899 | #define __arm_kprobe(p) arch_arm_kprobe(p) | |
6274de49 MH |
900 | #define __disarm_kprobe(p, o) arch_disarm_kprobe(p) |
901 | #define kprobe_disarmed(p) kprobe_disabled(p) | |
902 | #define wait_for_kprobe_optimizer() do {} while (0) | |
afd66255 | 903 | |
0490cd1f MH |
904 | /* There should be no unused kprobes can be reused without optimization */ |
905 | static void reuse_unused_kprobe(struct kprobe *ap) | |
906 | { | |
907 | printk(KERN_ERR "Error: There should be no unused kprobe here.\n"); | |
908 | BUG_ON(kprobe_unused(ap)); | |
909 | } | |
910 | ||
afd66255 MH |
911 | static __kprobes void free_aggr_kprobe(struct kprobe *p) |
912 | { | |
6274de49 | 913 | arch_remove_kprobe(p); |
afd66255 MH |
914 | kfree(p); |
915 | } | |
916 | ||
917 | static __kprobes struct kprobe *alloc_aggr_kprobe(struct kprobe *p) | |
918 | { | |
919 | return kzalloc(sizeof(struct kprobe), GFP_KERNEL); | |
920 | } | |
921 | #endif /* CONFIG_OPTPROBES */ | |
922 | ||
e7dbfe34 | 923 | #ifdef CONFIG_KPROBES_ON_FTRACE |
ae6aa16f | 924 | static struct ftrace_ops kprobe_ftrace_ops __read_mostly = { |
e5253896 | 925 | .func = kprobe_ftrace_handler, |
ae6aa16f MH |
926 | .flags = FTRACE_OPS_FL_SAVE_REGS, |
927 | }; | |
928 | static int kprobe_ftrace_enabled; | |
929 | ||
930 | /* Must ensure p->addr is really on ftrace */ | |
931 | static int __kprobes prepare_kprobe(struct kprobe *p) | |
932 | { | |
933 | if (!kprobe_ftrace(p)) | |
934 | return arch_prepare_kprobe(p); | |
935 | ||
936 | return arch_prepare_kprobe_ftrace(p); | |
937 | } | |
938 | ||
939 | /* Caller must lock kprobe_mutex */ | |
940 | static void __kprobes arm_kprobe_ftrace(struct kprobe *p) | |
941 | { | |
942 | int ret; | |
943 | ||
944 | ret = ftrace_set_filter_ip(&kprobe_ftrace_ops, | |
945 | (unsigned long)p->addr, 0, 0); | |
946 | WARN(ret < 0, "Failed to arm kprobe-ftrace at %p (%d)\n", p->addr, ret); | |
947 | kprobe_ftrace_enabled++; | |
948 | if (kprobe_ftrace_enabled == 1) { | |
949 | ret = register_ftrace_function(&kprobe_ftrace_ops); | |
950 | WARN(ret < 0, "Failed to init kprobe-ftrace (%d)\n", ret); | |
951 | } | |
952 | } | |
953 | ||
954 | /* Caller must lock kprobe_mutex */ | |
955 | static void __kprobes disarm_kprobe_ftrace(struct kprobe *p) | |
956 | { | |
957 | int ret; | |
958 | ||
959 | kprobe_ftrace_enabled--; | |
960 | if (kprobe_ftrace_enabled == 0) { | |
961 | ret = unregister_ftrace_function(&kprobe_ftrace_ops); | |
962 | WARN(ret < 0, "Failed to init kprobe-ftrace (%d)\n", ret); | |
963 | } | |
964 | ret = ftrace_set_filter_ip(&kprobe_ftrace_ops, | |
965 | (unsigned long)p->addr, 1, 0); | |
966 | WARN(ret < 0, "Failed to disarm kprobe-ftrace at %p (%d)\n", p->addr, ret); | |
967 | } | |
e7dbfe34 | 968 | #else /* !CONFIG_KPROBES_ON_FTRACE */ |
ae6aa16f MH |
969 | #define prepare_kprobe(p) arch_prepare_kprobe(p) |
970 | #define arm_kprobe_ftrace(p) do {} while (0) | |
971 | #define disarm_kprobe_ftrace(p) do {} while (0) | |
972 | #endif | |
973 | ||
201517a7 MH |
974 | /* Arm a kprobe with text_mutex */ |
975 | static void __kprobes arm_kprobe(struct kprobe *kp) | |
976 | { | |
ae6aa16f MH |
977 | if (unlikely(kprobe_ftrace(kp))) { |
978 | arm_kprobe_ftrace(kp); | |
979 | return; | |
980 | } | |
afd66255 MH |
981 | /* |
982 | * Here, since __arm_kprobe() doesn't use stop_machine(), | |
983 | * this doesn't cause deadlock on text_mutex. So, we don't | |
984 | * need get_online_cpus(). | |
985 | */ | |
201517a7 | 986 | mutex_lock(&text_mutex); |
afd66255 | 987 | __arm_kprobe(kp); |
201517a7 MH |
988 | mutex_unlock(&text_mutex); |
989 | } | |
990 | ||
991 | /* Disarm a kprobe with text_mutex */ | |
ae6aa16f | 992 | static void __kprobes disarm_kprobe(struct kprobe *kp, bool reopt) |
201517a7 | 993 | { |
ae6aa16f MH |
994 | if (unlikely(kprobe_ftrace(kp))) { |
995 | disarm_kprobe_ftrace(kp); | |
996 | return; | |
997 | } | |
6274de49 | 998 | /* Ditto */ |
201517a7 | 999 | mutex_lock(&text_mutex); |
ae6aa16f | 1000 | __disarm_kprobe(kp, reopt); |
201517a7 MH |
1001 | mutex_unlock(&text_mutex); |
1002 | } | |
1003 | ||
64f562c6 AM |
1004 | /* |
1005 | * Aggregate handlers for multiple kprobes support - these handlers | |
1006 | * take care of invoking the individual kprobe handlers on p->list | |
1007 | */ | |
d0aaff97 | 1008 | static int __kprobes aggr_pre_handler(struct kprobe *p, struct pt_regs *regs) |
64f562c6 AM |
1009 | { |
1010 | struct kprobe *kp; | |
1011 | ||
3516a460 | 1012 | list_for_each_entry_rcu(kp, &p->list, list) { |
de5bd88d | 1013 | if (kp->pre_handler && likely(!kprobe_disabled(kp))) { |
e6584523 | 1014 | set_kprobe_instance(kp); |
8b0914ea PP |
1015 | if (kp->pre_handler(kp, regs)) |
1016 | return 1; | |
64f562c6 | 1017 | } |
e6584523 | 1018 | reset_kprobe_instance(); |
64f562c6 AM |
1019 | } |
1020 | return 0; | |
1021 | } | |
1022 | ||
d0aaff97 PP |
1023 | static void __kprobes aggr_post_handler(struct kprobe *p, struct pt_regs *regs, |
1024 | unsigned long flags) | |
64f562c6 AM |
1025 | { |
1026 | struct kprobe *kp; | |
1027 | ||
3516a460 | 1028 | list_for_each_entry_rcu(kp, &p->list, list) { |
de5bd88d | 1029 | if (kp->post_handler && likely(!kprobe_disabled(kp))) { |
e6584523 | 1030 | set_kprobe_instance(kp); |
64f562c6 | 1031 | kp->post_handler(kp, regs, flags); |
e6584523 | 1032 | reset_kprobe_instance(); |
64f562c6 AM |
1033 | } |
1034 | } | |
64f562c6 AM |
1035 | } |
1036 | ||
d0aaff97 PP |
1037 | static int __kprobes aggr_fault_handler(struct kprobe *p, struct pt_regs *regs, |
1038 | int trapnr) | |
64f562c6 | 1039 | { |
b76834bc | 1040 | struct kprobe *cur = __this_cpu_read(kprobe_instance); |
e6584523 | 1041 | |
64f562c6 AM |
1042 | /* |
1043 | * if we faulted "during" the execution of a user specified | |
1044 | * probe handler, invoke just that probe's fault handler | |
1045 | */ | |
e6584523 AM |
1046 | if (cur && cur->fault_handler) { |
1047 | if (cur->fault_handler(cur, regs, trapnr)) | |
64f562c6 AM |
1048 | return 1; |
1049 | } | |
1050 | return 0; | |
1051 | } | |
1052 | ||
d0aaff97 | 1053 | static int __kprobes aggr_break_handler(struct kprobe *p, struct pt_regs *regs) |
8b0914ea | 1054 | { |
b76834bc | 1055 | struct kprobe *cur = __this_cpu_read(kprobe_instance); |
e6584523 AM |
1056 | int ret = 0; |
1057 | ||
1058 | if (cur && cur->break_handler) { | |
1059 | if (cur->break_handler(cur, regs)) | |
1060 | ret = 1; | |
8b0914ea | 1061 | } |
e6584523 AM |
1062 | reset_kprobe_instance(); |
1063 | return ret; | |
8b0914ea PP |
1064 | } |
1065 | ||
bf8d5c52 KA |
1066 | /* Walks the list and increments nmissed count for multiprobe case */ |
1067 | void __kprobes kprobes_inc_nmissed_count(struct kprobe *p) | |
1068 | { | |
1069 | struct kprobe *kp; | |
afd66255 | 1070 | if (!kprobe_aggrprobe(p)) { |
bf8d5c52 KA |
1071 | p->nmissed++; |
1072 | } else { | |
1073 | list_for_each_entry_rcu(kp, &p->list, list) | |
1074 | kp->nmissed++; | |
1075 | } | |
1076 | return; | |
1077 | } | |
1078 | ||
99219a3f | 1079 | void __kprobes recycle_rp_inst(struct kretprobe_instance *ri, |
1080 | struct hlist_head *head) | |
b94cce92 | 1081 | { |
ef53d9c5 S |
1082 | struct kretprobe *rp = ri->rp; |
1083 | ||
b94cce92 HN |
1084 | /* remove rp inst off the rprobe_inst_table */ |
1085 | hlist_del(&ri->hlist); | |
ef53d9c5 S |
1086 | INIT_HLIST_NODE(&ri->hlist); |
1087 | if (likely(rp)) { | |
ec484608 | 1088 | raw_spin_lock(&rp->lock); |
ef53d9c5 | 1089 | hlist_add_head(&ri->hlist, &rp->free_instances); |
ec484608 | 1090 | raw_spin_unlock(&rp->lock); |
b94cce92 HN |
1091 | } else |
1092 | /* Unregistering */ | |
99219a3f | 1093 | hlist_add_head(&ri->hlist, head); |
b94cce92 HN |
1094 | } |
1095 | ||
017c39bd | 1096 | void __kprobes kretprobe_hash_lock(struct task_struct *tsk, |
ef53d9c5 | 1097 | struct hlist_head **head, unsigned long *flags) |
635c17c2 | 1098 | __acquires(hlist_lock) |
ef53d9c5 S |
1099 | { |
1100 | unsigned long hash = hash_ptr(tsk, KPROBE_HASH_BITS); | |
ec484608 | 1101 | raw_spinlock_t *hlist_lock; |
ef53d9c5 S |
1102 | |
1103 | *head = &kretprobe_inst_table[hash]; | |
1104 | hlist_lock = kretprobe_table_lock_ptr(hash); | |
ec484608 | 1105 | raw_spin_lock_irqsave(hlist_lock, *flags); |
ef53d9c5 S |
1106 | } |
1107 | ||
017c39bd MH |
1108 | static void __kprobes kretprobe_table_lock(unsigned long hash, |
1109 | unsigned long *flags) | |
635c17c2 | 1110 | __acquires(hlist_lock) |
b94cce92 | 1111 | { |
ec484608 TG |
1112 | raw_spinlock_t *hlist_lock = kretprobe_table_lock_ptr(hash); |
1113 | raw_spin_lock_irqsave(hlist_lock, *flags); | |
ef53d9c5 S |
1114 | } |
1115 | ||
017c39bd MH |
1116 | void __kprobes kretprobe_hash_unlock(struct task_struct *tsk, |
1117 | unsigned long *flags) | |
635c17c2 | 1118 | __releases(hlist_lock) |
ef53d9c5 S |
1119 | { |
1120 | unsigned long hash = hash_ptr(tsk, KPROBE_HASH_BITS); | |
ec484608 | 1121 | raw_spinlock_t *hlist_lock; |
ef53d9c5 S |
1122 | |
1123 | hlist_lock = kretprobe_table_lock_ptr(hash); | |
ec484608 | 1124 | raw_spin_unlock_irqrestore(hlist_lock, *flags); |
ef53d9c5 S |
1125 | } |
1126 | ||
6376b229 NK |
1127 | static void __kprobes kretprobe_table_unlock(unsigned long hash, |
1128 | unsigned long *flags) | |
635c17c2 | 1129 | __releases(hlist_lock) |
ef53d9c5 | 1130 | { |
ec484608 TG |
1131 | raw_spinlock_t *hlist_lock = kretprobe_table_lock_ptr(hash); |
1132 | raw_spin_unlock_irqrestore(hlist_lock, *flags); | |
b94cce92 HN |
1133 | } |
1134 | ||
b94cce92 | 1135 | /* |
c6fd91f0 | 1136 | * This function is called from finish_task_switch when task tk becomes dead, |
1137 | * so that we can recycle any function-return probe instances associated | |
1138 | * with this task. These left over instances represent probed functions | |
1139 | * that have been called but will never return. | |
b94cce92 | 1140 | */ |
d0aaff97 | 1141 | void __kprobes kprobe_flush_task(struct task_struct *tk) |
b94cce92 | 1142 | { |
62c27be0 | 1143 | struct kretprobe_instance *ri; |
99219a3f | 1144 | struct hlist_head *head, empty_rp; |
b67bfe0d | 1145 | struct hlist_node *tmp; |
ef53d9c5 | 1146 | unsigned long hash, flags = 0; |
802eae7c | 1147 | |
ef53d9c5 S |
1148 | if (unlikely(!kprobes_initialized)) |
1149 | /* Early boot. kretprobe_table_locks not yet initialized. */ | |
1150 | return; | |
1151 | ||
d496aab5 | 1152 | INIT_HLIST_HEAD(&empty_rp); |
ef53d9c5 S |
1153 | hash = hash_ptr(tk, KPROBE_HASH_BITS); |
1154 | head = &kretprobe_inst_table[hash]; | |
1155 | kretprobe_table_lock(hash, &flags); | |
b67bfe0d | 1156 | hlist_for_each_entry_safe(ri, tmp, head, hlist) { |
62c27be0 | 1157 | if (ri->task == tk) |
99219a3f | 1158 | recycle_rp_inst(ri, &empty_rp); |
62c27be0 | 1159 | } |
ef53d9c5 | 1160 | kretprobe_table_unlock(hash, &flags); |
b67bfe0d | 1161 | hlist_for_each_entry_safe(ri, tmp, &empty_rp, hlist) { |
99219a3f | 1162 | hlist_del(&ri->hlist); |
1163 | kfree(ri); | |
1164 | } | |
b94cce92 HN |
1165 | } |
1166 | ||
b94cce92 HN |
1167 | static inline void free_rp_inst(struct kretprobe *rp) |
1168 | { | |
1169 | struct kretprobe_instance *ri; | |
b67bfe0d | 1170 | struct hlist_node *next; |
4c4308cb | 1171 | |
b67bfe0d | 1172 | hlist_for_each_entry_safe(ri, next, &rp->free_instances, hlist) { |
ef53d9c5 | 1173 | hlist_del(&ri->hlist); |
b94cce92 HN |
1174 | kfree(ri); |
1175 | } | |
1176 | } | |
1177 | ||
4a296e07 MH |
1178 | static void __kprobes cleanup_rp_inst(struct kretprobe *rp) |
1179 | { | |
ef53d9c5 | 1180 | unsigned long flags, hash; |
4a296e07 | 1181 | struct kretprobe_instance *ri; |
b67bfe0d | 1182 | struct hlist_node *next; |
ef53d9c5 S |
1183 | struct hlist_head *head; |
1184 | ||
4a296e07 | 1185 | /* No race here */ |
ef53d9c5 S |
1186 | for (hash = 0; hash < KPROBE_TABLE_SIZE; hash++) { |
1187 | kretprobe_table_lock(hash, &flags); | |
1188 | head = &kretprobe_inst_table[hash]; | |
b67bfe0d | 1189 | hlist_for_each_entry_safe(ri, next, head, hlist) { |
ef53d9c5 S |
1190 | if (ri->rp == rp) |
1191 | ri->rp = NULL; | |
1192 | } | |
1193 | kretprobe_table_unlock(hash, &flags); | |
4a296e07 | 1194 | } |
4a296e07 MH |
1195 | free_rp_inst(rp); |
1196 | } | |
1197 | ||
8b0914ea | 1198 | /* |
b918e5e6 | 1199 | * Add the new probe to ap->list. Fail if this is the |
8b0914ea PP |
1200 | * second jprobe at the address - two jprobes can't coexist |
1201 | */ | |
b918e5e6 | 1202 | static int __kprobes add_new_kprobe(struct kprobe *ap, struct kprobe *p) |
8b0914ea | 1203 | { |
de5bd88d | 1204 | BUG_ON(kprobe_gone(ap) || kprobe_gone(p)); |
afd66255 MH |
1205 | |
1206 | if (p->break_handler || p->post_handler) | |
6274de49 | 1207 | unoptimize_kprobe(ap, true); /* Fall back to normal kprobe */ |
afd66255 | 1208 | |
8b0914ea | 1209 | if (p->break_handler) { |
b918e5e6 | 1210 | if (ap->break_handler) |
36721656 | 1211 | return -EEXIST; |
b918e5e6 MH |
1212 | list_add_tail_rcu(&p->list, &ap->list); |
1213 | ap->break_handler = aggr_break_handler; | |
8b0914ea | 1214 | } else |
b918e5e6 MH |
1215 | list_add_rcu(&p->list, &ap->list); |
1216 | if (p->post_handler && !ap->post_handler) | |
1217 | ap->post_handler = aggr_post_handler; | |
de5bd88d | 1218 | |
8b0914ea PP |
1219 | return 0; |
1220 | } | |
1221 | ||
64f562c6 AM |
1222 | /* |
1223 | * Fill in the required fields of the "manager kprobe". Replace the | |
1224 | * earlier kprobe in the hlist with the manager kprobe | |
1225 | */ | |
afd66255 | 1226 | static void __kprobes init_aggr_kprobe(struct kprobe *ap, struct kprobe *p) |
64f562c6 | 1227 | { |
afd66255 | 1228 | /* Copy p's insn slot to ap */ |
8b0914ea | 1229 | copy_kprobe(p, ap); |
a9ad965e | 1230 | flush_insn_slot(ap); |
64f562c6 | 1231 | ap->addr = p->addr; |
afd66255 | 1232 | ap->flags = p->flags & ~KPROBE_FLAG_OPTIMIZED; |
64f562c6 | 1233 | ap->pre_handler = aggr_pre_handler; |
64f562c6 | 1234 | ap->fault_handler = aggr_fault_handler; |
e8386a0c MH |
1235 | /* We don't care the kprobe which has gone. */ |
1236 | if (p->post_handler && !kprobe_gone(p)) | |
36721656 | 1237 | ap->post_handler = aggr_post_handler; |
e8386a0c | 1238 | if (p->break_handler && !kprobe_gone(p)) |
36721656 | 1239 | ap->break_handler = aggr_break_handler; |
64f562c6 AM |
1240 | |
1241 | INIT_LIST_HEAD(&ap->list); | |
afd66255 | 1242 | INIT_HLIST_NODE(&ap->hlist); |
64f562c6 | 1243 | |
afd66255 | 1244 | list_add_rcu(&p->list, &ap->list); |
adad0f33 | 1245 | hlist_replace_rcu(&p->hlist, &ap->hlist); |
64f562c6 AM |
1246 | } |
1247 | ||
1248 | /* | |
1249 | * This is the second or subsequent kprobe at the address - handle | |
1250 | * the intricacies | |
64f562c6 | 1251 | */ |
6d8e40a8 | 1252 | static int __kprobes register_aggr_kprobe(struct kprobe *orig_p, |
d0aaff97 | 1253 | struct kprobe *p) |
64f562c6 AM |
1254 | { |
1255 | int ret = 0; | |
6d8e40a8 | 1256 | struct kprobe *ap = orig_p; |
64f562c6 | 1257 | |
25764288 MH |
1258 | /* For preparing optimization, jump_label_text_reserved() is called */ |
1259 | jump_label_lock(); | |
1260 | /* | |
1261 | * Get online CPUs to avoid text_mutex deadlock.with stop machine, | |
1262 | * which is invoked by unoptimize_kprobe() in add_new_kprobe() | |
1263 | */ | |
1264 | get_online_cpus(); | |
1265 | mutex_lock(&text_mutex); | |
1266 | ||
6d8e40a8 MH |
1267 | if (!kprobe_aggrprobe(orig_p)) { |
1268 | /* If orig_p is not an aggr_kprobe, create new aggr_kprobe. */ | |
1269 | ap = alloc_aggr_kprobe(orig_p); | |
25764288 MH |
1270 | if (!ap) { |
1271 | ret = -ENOMEM; | |
1272 | goto out; | |
1273 | } | |
6d8e40a8 | 1274 | init_aggr_kprobe(ap, orig_p); |
6274de49 | 1275 | } else if (kprobe_unused(ap)) |
0490cd1f MH |
1276 | /* This probe is going to die. Rescue it */ |
1277 | reuse_unused_kprobe(ap); | |
b918e5e6 MH |
1278 | |
1279 | if (kprobe_gone(ap)) { | |
e8386a0c MH |
1280 | /* |
1281 | * Attempting to insert new probe at the same location that | |
1282 | * had a probe in the module vaddr area which already | |
1283 | * freed. So, the instruction slot has already been | |
1284 | * released. We need a new slot for the new probe. | |
1285 | */ | |
b918e5e6 | 1286 | ret = arch_prepare_kprobe(ap); |
e8386a0c | 1287 | if (ret) |
b918e5e6 MH |
1288 | /* |
1289 | * Even if fail to allocate new slot, don't need to | |
1290 | * free aggr_probe. It will be used next time, or | |
1291 | * freed by unregister_kprobe. | |
1292 | */ | |
25764288 | 1293 | goto out; |
de5bd88d | 1294 | |
afd66255 MH |
1295 | /* Prepare optimized instructions if possible. */ |
1296 | prepare_optimized_kprobe(ap); | |
1297 | ||
e8386a0c | 1298 | /* |
de5bd88d MH |
1299 | * Clear gone flag to prevent allocating new slot again, and |
1300 | * set disabled flag because it is not armed yet. | |
e8386a0c | 1301 | */ |
de5bd88d MH |
1302 | ap->flags = (ap->flags & ~KPROBE_FLAG_GONE) |
1303 | | KPROBE_FLAG_DISABLED; | |
e8386a0c | 1304 | } |
b918e5e6 | 1305 | |
afd66255 | 1306 | /* Copy ap's insn slot to p */ |
b918e5e6 | 1307 | copy_kprobe(ap, p); |
25764288 MH |
1308 | ret = add_new_kprobe(ap, p); |
1309 | ||
1310 | out: | |
1311 | mutex_unlock(&text_mutex); | |
1312 | put_online_cpus(); | |
1313 | jump_label_unlock(); | |
1314 | ||
1315 | if (ret == 0 && kprobe_disabled(ap) && !kprobe_disabled(p)) { | |
1316 | ap->flags &= ~KPROBE_FLAG_DISABLED; | |
1317 | if (!kprobes_all_disarmed) | |
1318 | /* Arm the breakpoint again. */ | |
1319 | arm_kprobe(ap); | |
1320 | } | |
1321 | return ret; | |
64f562c6 AM |
1322 | } |
1323 | ||
be8f2743 MH |
1324 | bool __weak arch_within_kprobe_blacklist(unsigned long addr) |
1325 | { | |
1326 | /* The __kprobes marked functions and entry code must not be probed */ | |
1327 | return addr >= (unsigned long)__kprobes_text_start && | |
1328 | addr < (unsigned long)__kprobes_text_end; | |
1329 | } | |
1330 | ||
d0aaff97 PP |
1331 | static int __kprobes in_kprobes_functions(unsigned long addr) |
1332 | { | |
3d8d996e SD |
1333 | struct kprobe_blackpoint *kb; |
1334 | ||
be8f2743 | 1335 | if (arch_within_kprobe_blacklist(addr)) |
d0aaff97 | 1336 | return -EINVAL; |
3d8d996e SD |
1337 | /* |
1338 | * If there exists a kprobe_blacklist, verify and | |
1339 | * fail any probe registration in the prohibited area | |
1340 | */ | |
1341 | for (kb = kprobe_blacklist; kb->name != NULL; kb++) { | |
1342 | if (kb->start_addr) { | |
1343 | if (addr >= kb->start_addr && | |
1344 | addr < (kb->start_addr + kb->range)) | |
1345 | return -EINVAL; | |
1346 | } | |
1347 | } | |
d0aaff97 PP |
1348 | return 0; |
1349 | } | |
1350 | ||
b2a5cd69 MH |
1351 | /* |
1352 | * If we have a symbol_name argument, look it up and add the offset field | |
1353 | * to it. This way, we can specify a relative address to a symbol. | |
bc81d48d MH |
1354 | * This returns encoded errors if it fails to look up symbol or invalid |
1355 | * combination of parameters. | |
b2a5cd69 MH |
1356 | */ |
1357 | static kprobe_opcode_t __kprobes *kprobe_addr(struct kprobe *p) | |
1358 | { | |
1359 | kprobe_opcode_t *addr = p->addr; | |
bc81d48d MH |
1360 | |
1361 | if ((p->symbol_name && p->addr) || | |
1362 | (!p->symbol_name && !p->addr)) | |
1363 | goto invalid; | |
1364 | ||
b2a5cd69 | 1365 | if (p->symbol_name) { |
b2a5cd69 | 1366 | kprobe_lookup_name(p->symbol_name, addr); |
bc81d48d MH |
1367 | if (!addr) |
1368 | return ERR_PTR(-ENOENT); | |
b2a5cd69 MH |
1369 | } |
1370 | ||
bc81d48d MH |
1371 | addr = (kprobe_opcode_t *)(((char *)addr) + p->offset); |
1372 | if (addr) | |
1373 | return addr; | |
1374 | ||
1375 | invalid: | |
1376 | return ERR_PTR(-EINVAL); | |
b2a5cd69 MH |
1377 | } |
1378 | ||
1f0ab409 AM |
1379 | /* Check passed kprobe is valid and return kprobe in kprobe_table. */ |
1380 | static struct kprobe * __kprobes __get_valid_kprobe(struct kprobe *p) | |
1381 | { | |
6d8e40a8 | 1382 | struct kprobe *ap, *list_p; |
1f0ab409 | 1383 | |
6d8e40a8 MH |
1384 | ap = get_kprobe(p->addr); |
1385 | if (unlikely(!ap)) | |
1f0ab409 AM |
1386 | return NULL; |
1387 | ||
6d8e40a8 MH |
1388 | if (p != ap) { |
1389 | list_for_each_entry_rcu(list_p, &ap->list, list) | |
1f0ab409 AM |
1390 | if (list_p == p) |
1391 | /* kprobe p is a valid probe */ | |
1392 | goto valid; | |
1393 | return NULL; | |
1394 | } | |
1395 | valid: | |
6d8e40a8 | 1396 | return ap; |
1f0ab409 AM |
1397 | } |
1398 | ||
1399 | /* Return error if the kprobe is being re-registered */ | |
1400 | static inline int check_kprobe_rereg(struct kprobe *p) | |
1401 | { | |
1402 | int ret = 0; | |
1f0ab409 AM |
1403 | |
1404 | mutex_lock(&kprobe_mutex); | |
6d8e40a8 | 1405 | if (__get_valid_kprobe(p)) |
1f0ab409 AM |
1406 | ret = -EINVAL; |
1407 | mutex_unlock(&kprobe_mutex); | |
6d8e40a8 | 1408 | |
1f0ab409 AM |
1409 | return ret; |
1410 | } | |
1411 | ||
f7fa6ef0 MH |
1412 | static __kprobes int check_kprobe_address_safe(struct kprobe *p, |
1413 | struct module **probed_mod) | |
1da177e4 LT |
1414 | { |
1415 | int ret = 0; | |
ae6aa16f MH |
1416 | unsigned long ftrace_addr; |
1417 | ||
1418 | /* | |
1419 | * If the address is located on a ftrace nop, set the | |
1420 | * breakpoint to the following instruction. | |
1421 | */ | |
1422 | ftrace_addr = ftrace_location((unsigned long)p->addr); | |
1423 | if (ftrace_addr) { | |
e7dbfe34 | 1424 | #ifdef CONFIG_KPROBES_ON_FTRACE |
ae6aa16f MH |
1425 | /* Given address is not on the instruction boundary */ |
1426 | if ((unsigned long)p->addr != ftrace_addr) | |
1427 | return -EILSEQ; | |
ae6aa16f | 1428 | p->flags |= KPROBE_FLAG_FTRACE; |
e7dbfe34 | 1429 | #else /* !CONFIG_KPROBES_ON_FTRACE */ |
ae6aa16f MH |
1430 | return -EINVAL; |
1431 | #endif | |
1432 | } | |
1f0ab409 | 1433 | |
91bad2f8 | 1434 | jump_label_lock(); |
de31c3ca | 1435 | preempt_disable(); |
f7fa6ef0 MH |
1436 | |
1437 | /* Ensure it is not in reserved area nor out of text */ | |
ec30c5f3 | 1438 | if (!kernel_text_address((unsigned long) p->addr) || |
4554dbcb | 1439 | in_kprobes_functions((unsigned long) p->addr) || |
f986a499 PN |
1440 | jump_label_text_reserved(p->addr, p->addr)) { |
1441 | ret = -EINVAL; | |
f7fa6ef0 | 1442 | goto out; |
f986a499 | 1443 | } |
b3e55c72 | 1444 | |
f7fa6ef0 MH |
1445 | /* Check if are we probing a module */ |
1446 | *probed_mod = __module_text_address((unsigned long) p->addr); | |
1447 | if (*probed_mod) { | |
6f716acd | 1448 | /* |
e8386a0c MH |
1449 | * We must hold a refcount of the probed module while updating |
1450 | * its code to prohibit unexpected unloading. | |
df019b1d | 1451 | */ |
f7fa6ef0 MH |
1452 | if (unlikely(!try_module_get(*probed_mod))) { |
1453 | ret = -ENOENT; | |
1454 | goto out; | |
1455 | } | |
de31c3ca | 1456 | |
f24659d9 MH |
1457 | /* |
1458 | * If the module freed .init.text, we couldn't insert | |
1459 | * kprobes in there. | |
1460 | */ | |
f7fa6ef0 MH |
1461 | if (within_module_init((unsigned long)p->addr, *probed_mod) && |
1462 | (*probed_mod)->state != MODULE_STATE_COMING) { | |
1463 | module_put(*probed_mod); | |
1464 | *probed_mod = NULL; | |
1465 | ret = -ENOENT; | |
f24659d9 | 1466 | } |
df019b1d | 1467 | } |
f7fa6ef0 | 1468 | out: |
a189d035 | 1469 | preempt_enable(); |
de31c3ca | 1470 | jump_label_unlock(); |
1da177e4 | 1471 | |
f7fa6ef0 MH |
1472 | return ret; |
1473 | } | |
1474 | ||
1475 | int __kprobes register_kprobe(struct kprobe *p) | |
1476 | { | |
1477 | int ret; | |
1478 | struct kprobe *old_p; | |
1479 | struct module *probed_mod; | |
1480 | kprobe_opcode_t *addr; | |
1481 | ||
1482 | /* Adjust probe address from symbol */ | |
1483 | addr = kprobe_addr(p); | |
1484 | if (IS_ERR(addr)) | |
1485 | return PTR_ERR(addr); | |
1486 | p->addr = addr; | |
1487 | ||
1488 | ret = check_kprobe_rereg(p); | |
1489 | if (ret) | |
1490 | return ret; | |
1491 | ||
1492 | /* User can pass only KPROBE_FLAG_DISABLED to register_kprobe */ | |
1493 | p->flags &= KPROBE_FLAG_DISABLED; | |
3516a460 | 1494 | p->nmissed = 0; |
9861668f | 1495 | INIT_LIST_HEAD(&p->list); |
afd66255 | 1496 | |
f7fa6ef0 MH |
1497 | ret = check_kprobe_address_safe(p, &probed_mod); |
1498 | if (ret) | |
1499 | return ret; | |
1500 | ||
1501 | mutex_lock(&kprobe_mutex); | |
afd66255 | 1502 | |
64f562c6 AM |
1503 | old_p = get_kprobe(p->addr); |
1504 | if (old_p) { | |
afd66255 | 1505 | /* Since this may unoptimize old_p, locking text_mutex. */ |
64f562c6 | 1506 | ret = register_aggr_kprobe(old_p, p); |
1da177e4 LT |
1507 | goto out; |
1508 | } | |
1da177e4 | 1509 | |
25764288 | 1510 | mutex_lock(&text_mutex); /* Avoiding text modification */ |
ae6aa16f | 1511 | ret = prepare_kprobe(p); |
25764288 | 1512 | mutex_unlock(&text_mutex); |
6f716acd | 1513 | if (ret) |
afd66255 | 1514 | goto out; |
49a2a1b8 | 1515 | |
64f562c6 | 1516 | INIT_HLIST_NODE(&p->hlist); |
3516a460 | 1517 | hlist_add_head_rcu(&p->hlist, |
1da177e4 LT |
1518 | &kprobe_table[hash_ptr(p->addr, KPROBE_HASH_BITS)]); |
1519 | ||
de5bd88d | 1520 | if (!kprobes_all_disarmed && !kprobe_disabled(p)) |
25764288 | 1521 | arm_kprobe(p); |
afd66255 MH |
1522 | |
1523 | /* Try to optimize kprobe */ | |
1524 | try_to_optimize_kprobe(p); | |
74a0b576 | 1525 | |
1da177e4 | 1526 | out: |
7a7d1cf9 | 1527 | mutex_unlock(&kprobe_mutex); |
49a2a1b8 | 1528 | |
e8386a0c | 1529 | if (probed_mod) |
df019b1d | 1530 | module_put(probed_mod); |
e8386a0c | 1531 | |
1da177e4 LT |
1532 | return ret; |
1533 | } | |
99081ab5 | 1534 | EXPORT_SYMBOL_GPL(register_kprobe); |
1da177e4 | 1535 | |
6f0f1dd7 MH |
1536 | /* Check if all probes on the aggrprobe are disabled */ |
1537 | static int __kprobes aggr_kprobe_disabled(struct kprobe *ap) | |
1538 | { | |
1539 | struct kprobe *kp; | |
1540 | ||
1541 | list_for_each_entry_rcu(kp, &ap->list, list) | |
1542 | if (!kprobe_disabled(kp)) | |
1543 | /* | |
1544 | * There is an active probe on the list. | |
1545 | * We can't disable this ap. | |
1546 | */ | |
1547 | return 0; | |
1548 | ||
1549 | return 1; | |
1550 | } | |
1551 | ||
1552 | /* Disable one kprobe: Make sure called under kprobe_mutex is locked */ | |
1553 | static struct kprobe *__kprobes __disable_kprobe(struct kprobe *p) | |
1554 | { | |
1555 | struct kprobe *orig_p; | |
1556 | ||
1557 | /* Get an original kprobe for return */ | |
1558 | orig_p = __get_valid_kprobe(p); | |
1559 | if (unlikely(orig_p == NULL)) | |
1560 | return NULL; | |
1561 | ||
1562 | if (!kprobe_disabled(p)) { | |
1563 | /* Disable probe if it is a child probe */ | |
1564 | if (p != orig_p) | |
1565 | p->flags |= KPROBE_FLAG_DISABLED; | |
1566 | ||
1567 | /* Try to disarm and disable this/parent probe */ | |
1568 | if (p == orig_p || aggr_kprobe_disabled(orig_p)) { | |
ae6aa16f | 1569 | disarm_kprobe(orig_p, true); |
6f0f1dd7 MH |
1570 | orig_p->flags |= KPROBE_FLAG_DISABLED; |
1571 | } | |
1572 | } | |
1573 | ||
1574 | return orig_p; | |
1575 | } | |
1576 | ||
de5bd88d MH |
1577 | /* |
1578 | * Unregister a kprobe without a scheduler synchronization. | |
1579 | */ | |
1580 | static int __kprobes __unregister_kprobe_top(struct kprobe *p) | |
1581 | { | |
6d8e40a8 | 1582 | struct kprobe *ap, *list_p; |
de5bd88d | 1583 | |
6f0f1dd7 MH |
1584 | /* Disable kprobe. This will disarm it if needed. */ |
1585 | ap = __disable_kprobe(p); | |
6d8e40a8 | 1586 | if (ap == NULL) |
de5bd88d MH |
1587 | return -EINVAL; |
1588 | ||
6f0f1dd7 | 1589 | if (ap == p) |
bf8f6e5b | 1590 | /* |
6f0f1dd7 MH |
1591 | * This probe is an independent(and non-optimized) kprobe |
1592 | * (not an aggrprobe). Remove from the hash list. | |
bf8f6e5b | 1593 | */ |
6f0f1dd7 MH |
1594 | goto disarmed; |
1595 | ||
1596 | /* Following process expects this probe is an aggrprobe */ | |
1597 | WARN_ON(!kprobe_aggrprobe(ap)); | |
1598 | ||
6274de49 MH |
1599 | if (list_is_singular(&ap->list) && kprobe_disarmed(ap)) |
1600 | /* | |
1601 | * !disarmed could be happen if the probe is under delayed | |
1602 | * unoptimizing. | |
1603 | */ | |
6f0f1dd7 MH |
1604 | goto disarmed; |
1605 | else { | |
1606 | /* If disabling probe has special handlers, update aggrprobe */ | |
e8386a0c | 1607 | if (p->break_handler && !kprobe_gone(p)) |
6d8e40a8 | 1608 | ap->break_handler = NULL; |
e8386a0c | 1609 | if (p->post_handler && !kprobe_gone(p)) { |
6d8e40a8 | 1610 | list_for_each_entry_rcu(list_p, &ap->list, list) { |
9861668f MH |
1611 | if ((list_p != p) && (list_p->post_handler)) |
1612 | goto noclean; | |
1613 | } | |
6d8e40a8 | 1614 | ap->post_handler = NULL; |
9861668f MH |
1615 | } |
1616 | noclean: | |
6f0f1dd7 MH |
1617 | /* |
1618 | * Remove from the aggrprobe: this path will do nothing in | |
1619 | * __unregister_kprobe_bottom(). | |
1620 | */ | |
49a2a1b8 | 1621 | list_del_rcu(&p->list); |
6f0f1dd7 MH |
1622 | if (!kprobe_disabled(ap) && !kprobes_all_disarmed) |
1623 | /* | |
1624 | * Try to optimize this probe again, because post | |
1625 | * handler may have been changed. | |
1626 | */ | |
1627 | optimize_kprobe(ap); | |
49a2a1b8 | 1628 | } |
9861668f | 1629 | return 0; |
6f0f1dd7 MH |
1630 | |
1631 | disarmed: | |
6274de49 | 1632 | BUG_ON(!kprobe_disarmed(ap)); |
6f0f1dd7 MH |
1633 | hlist_del_rcu(&ap->hlist); |
1634 | return 0; | |
9861668f | 1635 | } |
3516a460 | 1636 | |
9861668f MH |
1637 | static void __kprobes __unregister_kprobe_bottom(struct kprobe *p) |
1638 | { | |
6d8e40a8 | 1639 | struct kprobe *ap; |
b3e55c72 | 1640 | |
e8386a0c | 1641 | if (list_empty(&p->list)) |
6274de49 | 1642 | /* This is an independent kprobe */ |
0498b635 | 1643 | arch_remove_kprobe(p); |
e8386a0c | 1644 | else if (list_is_singular(&p->list)) { |
6274de49 | 1645 | /* This is the last child of an aggrprobe */ |
6d8e40a8 | 1646 | ap = list_entry(p->list.next, struct kprobe, list); |
e8386a0c | 1647 | list_del(&p->list); |
6d8e40a8 | 1648 | free_aggr_kprobe(ap); |
9861668f | 1649 | } |
6274de49 | 1650 | /* Otherwise, do nothing. */ |
9861668f MH |
1651 | } |
1652 | ||
49ad2fd7 | 1653 | int __kprobes register_kprobes(struct kprobe **kps, int num) |
9861668f MH |
1654 | { |
1655 | int i, ret = 0; | |
1656 | ||
1657 | if (num <= 0) | |
1658 | return -EINVAL; | |
1659 | for (i = 0; i < num; i++) { | |
49ad2fd7 | 1660 | ret = register_kprobe(kps[i]); |
67dddaad MH |
1661 | if (ret < 0) { |
1662 | if (i > 0) | |
1663 | unregister_kprobes(kps, i); | |
9861668f | 1664 | break; |
36721656 | 1665 | } |
49a2a1b8 | 1666 | } |
9861668f MH |
1667 | return ret; |
1668 | } | |
99081ab5 | 1669 | EXPORT_SYMBOL_GPL(register_kprobes); |
9861668f | 1670 | |
9861668f MH |
1671 | void __kprobes unregister_kprobe(struct kprobe *p) |
1672 | { | |
1673 | unregister_kprobes(&p, 1); | |
1674 | } | |
99081ab5 | 1675 | EXPORT_SYMBOL_GPL(unregister_kprobe); |
9861668f | 1676 | |
9861668f MH |
1677 | void __kprobes unregister_kprobes(struct kprobe **kps, int num) |
1678 | { | |
1679 | int i; | |
1680 | ||
1681 | if (num <= 0) | |
1682 | return; | |
1683 | mutex_lock(&kprobe_mutex); | |
1684 | for (i = 0; i < num; i++) | |
1685 | if (__unregister_kprobe_top(kps[i]) < 0) | |
1686 | kps[i]->addr = NULL; | |
1687 | mutex_unlock(&kprobe_mutex); | |
1688 | ||
1689 | synchronize_sched(); | |
1690 | for (i = 0; i < num; i++) | |
1691 | if (kps[i]->addr) | |
1692 | __unregister_kprobe_bottom(kps[i]); | |
1da177e4 | 1693 | } |
99081ab5 | 1694 | EXPORT_SYMBOL_GPL(unregister_kprobes); |
1da177e4 LT |
1695 | |
1696 | static struct notifier_block kprobe_exceptions_nb = { | |
3d5631e0 AK |
1697 | .notifier_call = kprobe_exceptions_notify, |
1698 | .priority = 0x7fffffff /* we need to be notified first */ | |
1699 | }; | |
1700 | ||
3d7e3382 ME |
1701 | unsigned long __weak arch_deref_entry_point(void *entry) |
1702 | { | |
1703 | return (unsigned long)entry; | |
1704 | } | |
1da177e4 | 1705 | |
49ad2fd7 | 1706 | int __kprobes register_jprobes(struct jprobe **jps, int num) |
1da177e4 | 1707 | { |
26b31c19 MH |
1708 | struct jprobe *jp; |
1709 | int ret = 0, i; | |
3d7e3382 | 1710 | |
26b31c19 | 1711 | if (num <= 0) |
3d7e3382 | 1712 | return -EINVAL; |
26b31c19 | 1713 | for (i = 0; i < num; i++) { |
05662bdb | 1714 | unsigned long addr, offset; |
26b31c19 MH |
1715 | jp = jps[i]; |
1716 | addr = arch_deref_entry_point(jp->entry); | |
1717 | ||
05662bdb NK |
1718 | /* Verify probepoint is a function entry point */ |
1719 | if (kallsyms_lookup_size_offset(addr, NULL, &offset) && | |
1720 | offset == 0) { | |
1721 | jp->kp.pre_handler = setjmp_pre_handler; | |
1722 | jp->kp.break_handler = longjmp_break_handler; | |
1723 | ret = register_kprobe(&jp->kp); | |
1724 | } else | |
1725 | ret = -EINVAL; | |
edbaadbe | 1726 | |
67dddaad MH |
1727 | if (ret < 0) { |
1728 | if (i > 0) | |
1729 | unregister_jprobes(jps, i); | |
26b31c19 MH |
1730 | break; |
1731 | } | |
1732 | } | |
1733 | return ret; | |
1734 | } | |
99081ab5 | 1735 | EXPORT_SYMBOL_GPL(register_jprobes); |
3d7e3382 | 1736 | |
26b31c19 MH |
1737 | int __kprobes register_jprobe(struct jprobe *jp) |
1738 | { | |
49ad2fd7 | 1739 | return register_jprobes(&jp, 1); |
1da177e4 | 1740 | } |
99081ab5 | 1741 | EXPORT_SYMBOL_GPL(register_jprobe); |
1da177e4 | 1742 | |
d0aaff97 | 1743 | void __kprobes unregister_jprobe(struct jprobe *jp) |
1da177e4 | 1744 | { |
26b31c19 MH |
1745 | unregister_jprobes(&jp, 1); |
1746 | } | |
99081ab5 | 1747 | EXPORT_SYMBOL_GPL(unregister_jprobe); |
26b31c19 | 1748 | |
26b31c19 MH |
1749 | void __kprobes unregister_jprobes(struct jprobe **jps, int num) |
1750 | { | |
1751 | int i; | |
1752 | ||
1753 | if (num <= 0) | |
1754 | return; | |
1755 | mutex_lock(&kprobe_mutex); | |
1756 | for (i = 0; i < num; i++) | |
1757 | if (__unregister_kprobe_top(&jps[i]->kp) < 0) | |
1758 | jps[i]->kp.addr = NULL; | |
1759 | mutex_unlock(&kprobe_mutex); | |
1760 | ||
1761 | synchronize_sched(); | |
1762 | for (i = 0; i < num; i++) { | |
1763 | if (jps[i]->kp.addr) | |
1764 | __unregister_kprobe_bottom(&jps[i]->kp); | |
1765 | } | |
1da177e4 | 1766 | } |
99081ab5 | 1767 | EXPORT_SYMBOL_GPL(unregister_jprobes); |
1da177e4 | 1768 | |
9edddaa2 | 1769 | #ifdef CONFIG_KRETPROBES |
e65cefe8 AB |
1770 | /* |
1771 | * This kprobe pre_handler is registered with every kretprobe. When probe | |
1772 | * hits it will set up the return probe. | |
1773 | */ | |
1774 | static int __kprobes pre_handler_kretprobe(struct kprobe *p, | |
1775 | struct pt_regs *regs) | |
1776 | { | |
1777 | struct kretprobe *rp = container_of(p, struct kretprobe, kp); | |
ef53d9c5 S |
1778 | unsigned long hash, flags = 0; |
1779 | struct kretprobe_instance *ri; | |
e65cefe8 AB |
1780 | |
1781 | /*TODO: consider to only swap the RA after the last pre_handler fired */ | |
ef53d9c5 | 1782 | hash = hash_ptr(current, KPROBE_HASH_BITS); |
ec484608 | 1783 | raw_spin_lock_irqsave(&rp->lock, flags); |
4c4308cb | 1784 | if (!hlist_empty(&rp->free_instances)) { |
4c4308cb | 1785 | ri = hlist_entry(rp->free_instances.first, |
ef53d9c5 S |
1786 | struct kretprobe_instance, hlist); |
1787 | hlist_del(&ri->hlist); | |
ec484608 | 1788 | raw_spin_unlock_irqrestore(&rp->lock, flags); |
ef53d9c5 | 1789 | |
4c4308cb CH |
1790 | ri->rp = rp; |
1791 | ri->task = current; | |
f47cd9b5 | 1792 | |
55ca6140 JL |
1793 | if (rp->entry_handler && rp->entry_handler(ri, regs)) { |
1794 | raw_spin_lock_irqsave(&rp->lock, flags); | |
1795 | hlist_add_head(&ri->hlist, &rp->free_instances); | |
1796 | raw_spin_unlock_irqrestore(&rp->lock, flags); | |
f47cd9b5 | 1797 | return 0; |
55ca6140 | 1798 | } |
f47cd9b5 | 1799 | |
4c4308cb CH |
1800 | arch_prepare_kretprobe(ri, regs); |
1801 | ||
1802 | /* XXX(hch): why is there no hlist_move_head? */ | |
ef53d9c5 S |
1803 | INIT_HLIST_NODE(&ri->hlist); |
1804 | kretprobe_table_lock(hash, &flags); | |
1805 | hlist_add_head(&ri->hlist, &kretprobe_inst_table[hash]); | |
1806 | kretprobe_table_unlock(hash, &flags); | |
1807 | } else { | |
4c4308cb | 1808 | rp->nmissed++; |
ec484608 | 1809 | raw_spin_unlock_irqrestore(&rp->lock, flags); |
ef53d9c5 | 1810 | } |
e65cefe8 AB |
1811 | return 0; |
1812 | } | |
1813 | ||
49ad2fd7 | 1814 | int __kprobes register_kretprobe(struct kretprobe *rp) |
b94cce92 HN |
1815 | { |
1816 | int ret = 0; | |
1817 | struct kretprobe_instance *inst; | |
1818 | int i; | |
b2a5cd69 | 1819 | void *addr; |
f438d914 MH |
1820 | |
1821 | if (kretprobe_blacklist_size) { | |
b2a5cd69 | 1822 | addr = kprobe_addr(&rp->kp); |
bc81d48d MH |
1823 | if (IS_ERR(addr)) |
1824 | return PTR_ERR(addr); | |
f438d914 MH |
1825 | |
1826 | for (i = 0; kretprobe_blacklist[i].name != NULL; i++) { | |
1827 | if (kretprobe_blacklist[i].addr == addr) | |
1828 | return -EINVAL; | |
1829 | } | |
1830 | } | |
b94cce92 HN |
1831 | |
1832 | rp->kp.pre_handler = pre_handler_kretprobe; | |
7522a842 AM |
1833 | rp->kp.post_handler = NULL; |
1834 | rp->kp.fault_handler = NULL; | |
1835 | rp->kp.break_handler = NULL; | |
b94cce92 HN |
1836 | |
1837 | /* Pre-allocate memory for max kretprobe instances */ | |
1838 | if (rp->maxactive <= 0) { | |
1839 | #ifdef CONFIG_PREEMPT | |
c2ef6661 | 1840 | rp->maxactive = max_t(unsigned int, 10, 2*num_possible_cpus()); |
b94cce92 | 1841 | #else |
4dae560f | 1842 | rp->maxactive = num_possible_cpus(); |
b94cce92 HN |
1843 | #endif |
1844 | } | |
ec484608 | 1845 | raw_spin_lock_init(&rp->lock); |
b94cce92 HN |
1846 | INIT_HLIST_HEAD(&rp->free_instances); |
1847 | for (i = 0; i < rp->maxactive; i++) { | |
f47cd9b5 AS |
1848 | inst = kmalloc(sizeof(struct kretprobe_instance) + |
1849 | rp->data_size, GFP_KERNEL); | |
b94cce92 HN |
1850 | if (inst == NULL) { |
1851 | free_rp_inst(rp); | |
1852 | return -ENOMEM; | |
1853 | } | |
ef53d9c5 S |
1854 | INIT_HLIST_NODE(&inst->hlist); |
1855 | hlist_add_head(&inst->hlist, &rp->free_instances); | |
b94cce92 HN |
1856 | } |
1857 | ||
1858 | rp->nmissed = 0; | |
1859 | /* Establish function entry probe point */ | |
49ad2fd7 | 1860 | ret = register_kprobe(&rp->kp); |
4a296e07 | 1861 | if (ret != 0) |
b94cce92 HN |
1862 | free_rp_inst(rp); |
1863 | return ret; | |
1864 | } | |
99081ab5 | 1865 | EXPORT_SYMBOL_GPL(register_kretprobe); |
b94cce92 | 1866 | |
49ad2fd7 | 1867 | int __kprobes register_kretprobes(struct kretprobe **rps, int num) |
4a296e07 MH |
1868 | { |
1869 | int ret = 0, i; | |
1870 | ||
1871 | if (num <= 0) | |
1872 | return -EINVAL; | |
1873 | for (i = 0; i < num; i++) { | |
49ad2fd7 | 1874 | ret = register_kretprobe(rps[i]); |
67dddaad MH |
1875 | if (ret < 0) { |
1876 | if (i > 0) | |
1877 | unregister_kretprobes(rps, i); | |
4a296e07 MH |
1878 | break; |
1879 | } | |
1880 | } | |
1881 | return ret; | |
1882 | } | |
99081ab5 | 1883 | EXPORT_SYMBOL_GPL(register_kretprobes); |
4a296e07 | 1884 | |
4a296e07 MH |
1885 | void __kprobes unregister_kretprobe(struct kretprobe *rp) |
1886 | { | |
1887 | unregister_kretprobes(&rp, 1); | |
1888 | } | |
99081ab5 | 1889 | EXPORT_SYMBOL_GPL(unregister_kretprobe); |
4a296e07 | 1890 | |
4a296e07 MH |
1891 | void __kprobes unregister_kretprobes(struct kretprobe **rps, int num) |
1892 | { | |
1893 | int i; | |
1894 | ||
1895 | if (num <= 0) | |
1896 | return; | |
1897 | mutex_lock(&kprobe_mutex); | |
1898 | for (i = 0; i < num; i++) | |
1899 | if (__unregister_kprobe_top(&rps[i]->kp) < 0) | |
1900 | rps[i]->kp.addr = NULL; | |
1901 | mutex_unlock(&kprobe_mutex); | |
1902 | ||
1903 | synchronize_sched(); | |
1904 | for (i = 0; i < num; i++) { | |
1905 | if (rps[i]->kp.addr) { | |
1906 | __unregister_kprobe_bottom(&rps[i]->kp); | |
1907 | cleanup_rp_inst(rps[i]); | |
1908 | } | |
1909 | } | |
1910 | } | |
99081ab5 | 1911 | EXPORT_SYMBOL_GPL(unregister_kretprobes); |
4a296e07 | 1912 | |
9edddaa2 | 1913 | #else /* CONFIG_KRETPROBES */ |
d0aaff97 | 1914 | int __kprobes register_kretprobe(struct kretprobe *rp) |
b94cce92 HN |
1915 | { |
1916 | return -ENOSYS; | |
1917 | } | |
99081ab5 | 1918 | EXPORT_SYMBOL_GPL(register_kretprobe); |
b94cce92 | 1919 | |
4a296e07 | 1920 | int __kprobes register_kretprobes(struct kretprobe **rps, int num) |
346fd59b | 1921 | { |
4a296e07 | 1922 | return -ENOSYS; |
346fd59b | 1923 | } |
99081ab5 MH |
1924 | EXPORT_SYMBOL_GPL(register_kretprobes); |
1925 | ||
d0aaff97 | 1926 | void __kprobes unregister_kretprobe(struct kretprobe *rp) |
b94cce92 | 1927 | { |
4a296e07 | 1928 | } |
99081ab5 | 1929 | EXPORT_SYMBOL_GPL(unregister_kretprobe); |
b94cce92 | 1930 | |
4a296e07 MH |
1931 | void __kprobes unregister_kretprobes(struct kretprobe **rps, int num) |
1932 | { | |
1933 | } | |
99081ab5 | 1934 | EXPORT_SYMBOL_GPL(unregister_kretprobes); |
4c4308cb | 1935 | |
4a296e07 MH |
1936 | static int __kprobes pre_handler_kretprobe(struct kprobe *p, |
1937 | struct pt_regs *regs) | |
1938 | { | |
1939 | return 0; | |
b94cce92 HN |
1940 | } |
1941 | ||
4a296e07 MH |
1942 | #endif /* CONFIG_KRETPROBES */ |
1943 | ||
e8386a0c MH |
1944 | /* Set the kprobe gone and remove its instruction buffer. */ |
1945 | static void __kprobes kill_kprobe(struct kprobe *p) | |
1946 | { | |
1947 | struct kprobe *kp; | |
de5bd88d | 1948 | |
e8386a0c | 1949 | p->flags |= KPROBE_FLAG_GONE; |
afd66255 | 1950 | if (kprobe_aggrprobe(p)) { |
e8386a0c MH |
1951 | /* |
1952 | * If this is an aggr_kprobe, we have to list all the | |
1953 | * chained probes and mark them GONE. | |
1954 | */ | |
1955 | list_for_each_entry_rcu(kp, &p->list, list) | |
1956 | kp->flags |= KPROBE_FLAG_GONE; | |
1957 | p->post_handler = NULL; | |
1958 | p->break_handler = NULL; | |
afd66255 | 1959 | kill_optimized_kprobe(p); |
e8386a0c MH |
1960 | } |
1961 | /* | |
1962 | * Here, we can remove insn_slot safely, because no thread calls | |
1963 | * the original probed function (which will be freed soon) any more. | |
1964 | */ | |
1965 | arch_remove_kprobe(p); | |
1966 | } | |
1967 | ||
c0614829 MH |
1968 | /* Disable one kprobe */ |
1969 | int __kprobes disable_kprobe(struct kprobe *kp) | |
1970 | { | |
1971 | int ret = 0; | |
c0614829 MH |
1972 | |
1973 | mutex_lock(&kprobe_mutex); | |
1974 | ||
6f0f1dd7 MH |
1975 | /* Disable this kprobe */ |
1976 | if (__disable_kprobe(kp) == NULL) | |
c0614829 | 1977 | ret = -EINVAL; |
c0614829 | 1978 | |
c0614829 MH |
1979 | mutex_unlock(&kprobe_mutex); |
1980 | return ret; | |
1981 | } | |
1982 | EXPORT_SYMBOL_GPL(disable_kprobe); | |
1983 | ||
1984 | /* Enable one kprobe */ | |
1985 | int __kprobes enable_kprobe(struct kprobe *kp) | |
1986 | { | |
1987 | int ret = 0; | |
1988 | struct kprobe *p; | |
1989 | ||
1990 | mutex_lock(&kprobe_mutex); | |
1991 | ||
1992 | /* Check whether specified probe is valid. */ | |
1993 | p = __get_valid_kprobe(kp); | |
1994 | if (unlikely(p == NULL)) { | |
1995 | ret = -EINVAL; | |
1996 | goto out; | |
1997 | } | |
1998 | ||
1999 | if (kprobe_gone(kp)) { | |
2000 | /* This kprobe has gone, we couldn't enable it. */ | |
2001 | ret = -EINVAL; | |
2002 | goto out; | |
2003 | } | |
2004 | ||
2005 | if (p != kp) | |
2006 | kp->flags &= ~KPROBE_FLAG_DISABLED; | |
2007 | ||
2008 | if (!kprobes_all_disarmed && kprobe_disabled(p)) { | |
2009 | p->flags &= ~KPROBE_FLAG_DISABLED; | |
2010 | arm_kprobe(p); | |
2011 | } | |
2012 | out: | |
2013 | mutex_unlock(&kprobe_mutex); | |
2014 | return ret; | |
2015 | } | |
2016 | EXPORT_SYMBOL_GPL(enable_kprobe); | |
2017 | ||
24851d24 FW |
2018 | void __kprobes dump_kprobe(struct kprobe *kp) |
2019 | { | |
2020 | printk(KERN_WARNING "Dumping kprobe:\n"); | |
2021 | printk(KERN_WARNING "Name: %s\nAddress: %p\nOffset: %x\n", | |
2022 | kp->symbol_name, kp->addr, kp->offset); | |
2023 | } | |
2024 | ||
e8386a0c MH |
2025 | /* Module notifier call back, checking kprobes on the module */ |
2026 | static int __kprobes kprobes_module_callback(struct notifier_block *nb, | |
2027 | unsigned long val, void *data) | |
2028 | { | |
2029 | struct module *mod = data; | |
2030 | struct hlist_head *head; | |
e8386a0c MH |
2031 | struct kprobe *p; |
2032 | unsigned int i; | |
f24659d9 | 2033 | int checkcore = (val == MODULE_STATE_GOING); |
e8386a0c | 2034 | |
f24659d9 | 2035 | if (val != MODULE_STATE_GOING && val != MODULE_STATE_LIVE) |
e8386a0c MH |
2036 | return NOTIFY_DONE; |
2037 | ||
2038 | /* | |
f24659d9 MH |
2039 | * When MODULE_STATE_GOING was notified, both of module .text and |
2040 | * .init.text sections would be freed. When MODULE_STATE_LIVE was | |
2041 | * notified, only .init.text section would be freed. We need to | |
2042 | * disable kprobes which have been inserted in the sections. | |
e8386a0c MH |
2043 | */ |
2044 | mutex_lock(&kprobe_mutex); | |
2045 | for (i = 0; i < KPROBE_TABLE_SIZE; i++) { | |
2046 | head = &kprobe_table[i]; | |
b67bfe0d | 2047 | hlist_for_each_entry_rcu(p, head, hlist) |
f24659d9 MH |
2048 | if (within_module_init((unsigned long)p->addr, mod) || |
2049 | (checkcore && | |
2050 | within_module_core((unsigned long)p->addr, mod))) { | |
e8386a0c MH |
2051 | /* |
2052 | * The vaddr this probe is installed will soon | |
2053 | * be vfreed buy not synced to disk. Hence, | |
2054 | * disarming the breakpoint isn't needed. | |
2055 | */ | |
2056 | kill_kprobe(p); | |
2057 | } | |
2058 | } | |
2059 | mutex_unlock(&kprobe_mutex); | |
2060 | return NOTIFY_DONE; | |
2061 | } | |
2062 | ||
2063 | static struct notifier_block kprobe_module_nb = { | |
2064 | .notifier_call = kprobes_module_callback, | |
2065 | .priority = 0 | |
2066 | }; | |
2067 | ||
1da177e4 LT |
2068 | static int __init init_kprobes(void) |
2069 | { | |
2070 | int i, err = 0; | |
3d8d996e | 2071 | unsigned long offset = 0, size = 0; |
ab767865 | 2072 | char *modname, namebuf[KSYM_NAME_LEN]; |
3d8d996e SD |
2073 | const char *symbol_name; |
2074 | void *addr; | |
2075 | struct kprobe_blackpoint *kb; | |
1da177e4 LT |
2076 | |
2077 | /* FIXME allocate the probe table, currently defined statically */ | |
2078 | /* initialize all list heads */ | |
b94cce92 | 2079 | for (i = 0; i < KPROBE_TABLE_SIZE; i++) { |
1da177e4 | 2080 | INIT_HLIST_HEAD(&kprobe_table[i]); |
b94cce92 | 2081 | INIT_HLIST_HEAD(&kretprobe_inst_table[i]); |
ec484608 | 2082 | raw_spin_lock_init(&(kretprobe_table_locks[i].lock)); |
b94cce92 | 2083 | } |
1da177e4 | 2084 | |
3d8d996e SD |
2085 | /* |
2086 | * Lookup and populate the kprobe_blacklist. | |
2087 | * | |
2088 | * Unlike the kretprobe blacklist, we'll need to determine | |
2089 | * the range of addresses that belong to the said functions, | |
2090 | * since a kprobe need not necessarily be at the beginning | |
2091 | * of a function. | |
2092 | */ | |
2093 | for (kb = kprobe_blacklist; kb->name != NULL; kb++) { | |
2094 | kprobe_lookup_name(kb->name, addr); | |
2095 | if (!addr) | |
2096 | continue; | |
2097 | ||
2098 | kb->start_addr = (unsigned long)addr; | |
2099 | symbol_name = kallsyms_lookup(kb->start_addr, | |
2100 | &size, &offset, &modname, namebuf); | |
2101 | if (!symbol_name) | |
2102 | kb->range = 0; | |
2103 | else | |
2104 | kb->range = size; | |
2105 | } | |
2106 | ||
f438d914 MH |
2107 | if (kretprobe_blacklist_size) { |
2108 | /* lookup the function address from its name */ | |
2109 | for (i = 0; kretprobe_blacklist[i].name != NULL; i++) { | |
2110 | kprobe_lookup_name(kretprobe_blacklist[i].name, | |
2111 | kretprobe_blacklist[i].addr); | |
2112 | if (!kretprobe_blacklist[i].addr) | |
2113 | printk("kretprobe: lookup failed: %s\n", | |
2114 | kretprobe_blacklist[i].name); | |
2115 | } | |
2116 | } | |
2117 | ||
b2be84df MH |
2118 | #if defined(CONFIG_OPTPROBES) |
2119 | #if defined(__ARCH_WANT_KPROBES_INSN_SLOT) | |
afd66255 MH |
2120 | /* Init kprobe_optinsn_slots */ |
2121 | kprobe_optinsn_slots.insn_size = MAX_OPTINSN_SIZE; | |
2122 | #endif | |
b2be84df MH |
2123 | /* By default, kprobes can be optimized */ |
2124 | kprobes_allow_optimization = true; | |
2125 | #endif | |
afd66255 | 2126 | |
e579abeb MH |
2127 | /* By default, kprobes are armed */ |
2128 | kprobes_all_disarmed = false; | |
bf8f6e5b | 2129 | |
6772926b | 2130 | err = arch_init_kprobes(); |
802eae7c RL |
2131 | if (!err) |
2132 | err = register_die_notifier(&kprobe_exceptions_nb); | |
e8386a0c MH |
2133 | if (!err) |
2134 | err = register_module_notifier(&kprobe_module_nb); | |
2135 | ||
ef53d9c5 | 2136 | kprobes_initialized = (err == 0); |
802eae7c | 2137 | |
8c1c9356 AM |
2138 | if (!err) |
2139 | init_test_probes(); | |
1da177e4 LT |
2140 | return err; |
2141 | } | |
2142 | ||
346fd59b SD |
2143 | #ifdef CONFIG_DEBUG_FS |
2144 | static void __kprobes report_probe(struct seq_file *pi, struct kprobe *p, | |
afd66255 | 2145 | const char *sym, int offset, char *modname, struct kprobe *pp) |
346fd59b SD |
2146 | { |
2147 | char *kprobe_type; | |
2148 | ||
2149 | if (p->pre_handler == pre_handler_kretprobe) | |
2150 | kprobe_type = "r"; | |
2151 | else if (p->pre_handler == setjmp_pre_handler) | |
2152 | kprobe_type = "j"; | |
2153 | else | |
2154 | kprobe_type = "k"; | |
afd66255 | 2155 | |
346fd59b | 2156 | if (sym) |
afd66255 | 2157 | seq_printf(pi, "%p %s %s+0x%x %s ", |
de5bd88d | 2158 | p->addr, kprobe_type, sym, offset, |
afd66255 | 2159 | (modname ? modname : " ")); |
346fd59b | 2160 | else |
afd66255 MH |
2161 | seq_printf(pi, "%p %s %p ", |
2162 | p->addr, kprobe_type, p->addr); | |
2163 | ||
2164 | if (!pp) | |
2165 | pp = p; | |
ae6aa16f | 2166 | seq_printf(pi, "%s%s%s%s\n", |
afd66255 MH |
2167 | (kprobe_gone(p) ? "[GONE]" : ""), |
2168 | ((kprobe_disabled(p) && !kprobe_gone(p)) ? "[DISABLED]" : ""), | |
ae6aa16f MH |
2169 | (kprobe_optimized(pp) ? "[OPTIMIZED]" : ""), |
2170 | (kprobe_ftrace(pp) ? "[FTRACE]" : "")); | |
346fd59b SD |
2171 | } |
2172 | ||
2173 | static void __kprobes *kprobe_seq_start(struct seq_file *f, loff_t *pos) | |
2174 | { | |
2175 | return (*pos < KPROBE_TABLE_SIZE) ? pos : NULL; | |
2176 | } | |
2177 | ||
2178 | static void __kprobes *kprobe_seq_next(struct seq_file *f, void *v, loff_t *pos) | |
2179 | { | |
2180 | (*pos)++; | |
2181 | if (*pos >= KPROBE_TABLE_SIZE) | |
2182 | return NULL; | |
2183 | return pos; | |
2184 | } | |
2185 | ||
2186 | static void __kprobes kprobe_seq_stop(struct seq_file *f, void *v) | |
2187 | { | |
2188 | /* Nothing to do */ | |
2189 | } | |
2190 | ||
2191 | static int __kprobes show_kprobe_addr(struct seq_file *pi, void *v) | |
2192 | { | |
2193 | struct hlist_head *head; | |
346fd59b SD |
2194 | struct kprobe *p, *kp; |
2195 | const char *sym = NULL; | |
2196 | unsigned int i = *(loff_t *) v; | |
ffb45122 | 2197 | unsigned long offset = 0; |
ab767865 | 2198 | char *modname, namebuf[KSYM_NAME_LEN]; |
346fd59b SD |
2199 | |
2200 | head = &kprobe_table[i]; | |
2201 | preempt_disable(); | |
b67bfe0d | 2202 | hlist_for_each_entry_rcu(p, head, hlist) { |
ffb45122 | 2203 | sym = kallsyms_lookup((unsigned long)p->addr, NULL, |
346fd59b | 2204 | &offset, &modname, namebuf); |
afd66255 | 2205 | if (kprobe_aggrprobe(p)) { |
346fd59b | 2206 | list_for_each_entry_rcu(kp, &p->list, list) |
afd66255 | 2207 | report_probe(pi, kp, sym, offset, modname, p); |
346fd59b | 2208 | } else |
afd66255 | 2209 | report_probe(pi, p, sym, offset, modname, NULL); |
346fd59b SD |
2210 | } |
2211 | preempt_enable(); | |
2212 | return 0; | |
2213 | } | |
2214 | ||
88e9d34c | 2215 | static const struct seq_operations kprobes_seq_ops = { |
346fd59b SD |
2216 | .start = kprobe_seq_start, |
2217 | .next = kprobe_seq_next, | |
2218 | .stop = kprobe_seq_stop, | |
2219 | .show = show_kprobe_addr | |
2220 | }; | |
2221 | ||
2222 | static int __kprobes kprobes_open(struct inode *inode, struct file *filp) | |
2223 | { | |
2224 | return seq_open(filp, &kprobes_seq_ops); | |
2225 | } | |
2226 | ||
828c0950 | 2227 | static const struct file_operations debugfs_kprobes_operations = { |
346fd59b SD |
2228 | .open = kprobes_open, |
2229 | .read = seq_read, | |
2230 | .llseek = seq_lseek, | |
2231 | .release = seq_release, | |
2232 | }; | |
2233 | ||
e579abeb | 2234 | static void __kprobes arm_all_kprobes(void) |
bf8f6e5b AM |
2235 | { |
2236 | struct hlist_head *head; | |
bf8f6e5b AM |
2237 | struct kprobe *p; |
2238 | unsigned int i; | |
2239 | ||
2240 | mutex_lock(&kprobe_mutex); | |
2241 | ||
e579abeb MH |
2242 | /* If kprobes are armed, just return */ |
2243 | if (!kprobes_all_disarmed) | |
bf8f6e5b AM |
2244 | goto already_enabled; |
2245 | ||
afd66255 | 2246 | /* Arming kprobes doesn't optimize kprobe itself */ |
bf8f6e5b AM |
2247 | for (i = 0; i < KPROBE_TABLE_SIZE; i++) { |
2248 | head = &kprobe_table[i]; | |
b67bfe0d | 2249 | hlist_for_each_entry_rcu(p, head, hlist) |
de5bd88d | 2250 | if (!kprobe_disabled(p)) |
ae6aa16f | 2251 | arm_kprobe(p); |
bf8f6e5b AM |
2252 | } |
2253 | ||
e579abeb | 2254 | kprobes_all_disarmed = false; |
bf8f6e5b AM |
2255 | printk(KERN_INFO "Kprobes globally enabled\n"); |
2256 | ||
2257 | already_enabled: | |
2258 | mutex_unlock(&kprobe_mutex); | |
2259 | return; | |
2260 | } | |
2261 | ||
e579abeb | 2262 | static void __kprobes disarm_all_kprobes(void) |
bf8f6e5b AM |
2263 | { |
2264 | struct hlist_head *head; | |
bf8f6e5b AM |
2265 | struct kprobe *p; |
2266 | unsigned int i; | |
2267 | ||
2268 | mutex_lock(&kprobe_mutex); | |
2269 | ||
e579abeb | 2270 | /* If kprobes are already disarmed, just return */ |
6274de49 MH |
2271 | if (kprobes_all_disarmed) { |
2272 | mutex_unlock(&kprobe_mutex); | |
2273 | return; | |
2274 | } | |
bf8f6e5b | 2275 | |
e579abeb | 2276 | kprobes_all_disarmed = true; |
bf8f6e5b | 2277 | printk(KERN_INFO "Kprobes globally disabled\n"); |
afd66255 | 2278 | |
bf8f6e5b AM |
2279 | for (i = 0; i < KPROBE_TABLE_SIZE; i++) { |
2280 | head = &kprobe_table[i]; | |
b67bfe0d | 2281 | hlist_for_each_entry_rcu(p, head, hlist) { |
de5bd88d | 2282 | if (!arch_trampoline_kprobe(p) && !kprobe_disabled(p)) |
ae6aa16f | 2283 | disarm_kprobe(p, false); |
bf8f6e5b AM |
2284 | } |
2285 | } | |
bf8f6e5b | 2286 | mutex_unlock(&kprobe_mutex); |
bf8f6e5b | 2287 | |
6274de49 MH |
2288 | /* Wait for disarming all kprobes by optimizer */ |
2289 | wait_for_kprobe_optimizer(); | |
bf8f6e5b AM |
2290 | } |
2291 | ||
2292 | /* | |
2293 | * XXX: The debugfs bool file interface doesn't allow for callbacks | |
2294 | * when the bool state is switched. We can reuse that facility when | |
2295 | * available | |
2296 | */ | |
2297 | static ssize_t read_enabled_file_bool(struct file *file, | |
2298 | char __user *user_buf, size_t count, loff_t *ppos) | |
2299 | { | |
2300 | char buf[3]; | |
2301 | ||
e579abeb | 2302 | if (!kprobes_all_disarmed) |
bf8f6e5b AM |
2303 | buf[0] = '1'; |
2304 | else | |
2305 | buf[0] = '0'; | |
2306 | buf[1] = '\n'; | |
2307 | buf[2] = 0x00; | |
2308 | return simple_read_from_buffer(user_buf, count, ppos, buf, 2); | |
2309 | } | |
2310 | ||
2311 | static ssize_t write_enabled_file_bool(struct file *file, | |
2312 | const char __user *user_buf, size_t count, loff_t *ppos) | |
2313 | { | |
2314 | char buf[32]; | |
efeb156e | 2315 | size_t buf_size; |
bf8f6e5b AM |
2316 | |
2317 | buf_size = min(count, (sizeof(buf)-1)); | |
2318 | if (copy_from_user(buf, user_buf, buf_size)) | |
2319 | return -EFAULT; | |
2320 | ||
10fb46d5 | 2321 | buf[buf_size] = '\0'; |
bf8f6e5b AM |
2322 | switch (buf[0]) { |
2323 | case 'y': | |
2324 | case 'Y': | |
2325 | case '1': | |
e579abeb | 2326 | arm_all_kprobes(); |
bf8f6e5b AM |
2327 | break; |
2328 | case 'n': | |
2329 | case 'N': | |
2330 | case '0': | |
e579abeb | 2331 | disarm_all_kprobes(); |
bf8f6e5b | 2332 | break; |
10fb46d5 MK |
2333 | default: |
2334 | return -EINVAL; | |
bf8f6e5b AM |
2335 | } |
2336 | ||
2337 | return count; | |
2338 | } | |
2339 | ||
828c0950 | 2340 | static const struct file_operations fops_kp = { |
bf8f6e5b AM |
2341 | .read = read_enabled_file_bool, |
2342 | .write = write_enabled_file_bool, | |
6038f373 | 2343 | .llseek = default_llseek, |
bf8f6e5b AM |
2344 | }; |
2345 | ||
346fd59b SD |
2346 | static int __kprobes debugfs_kprobe_init(void) |
2347 | { | |
2348 | struct dentry *dir, *file; | |
bf8f6e5b | 2349 | unsigned int value = 1; |
346fd59b SD |
2350 | |
2351 | dir = debugfs_create_dir("kprobes", NULL); | |
2352 | if (!dir) | |
2353 | return -ENOMEM; | |
2354 | ||
e3869792 | 2355 | file = debugfs_create_file("list", 0444, dir, NULL, |
346fd59b SD |
2356 | &debugfs_kprobes_operations); |
2357 | if (!file) { | |
2358 | debugfs_remove(dir); | |
2359 | return -ENOMEM; | |
2360 | } | |
2361 | ||
bf8f6e5b AM |
2362 | file = debugfs_create_file("enabled", 0600, dir, |
2363 | &value, &fops_kp); | |
2364 | if (!file) { | |
2365 | debugfs_remove(dir); | |
2366 | return -ENOMEM; | |
2367 | } | |
2368 | ||
346fd59b SD |
2369 | return 0; |
2370 | } | |
2371 | ||
2372 | late_initcall(debugfs_kprobe_init); | |
2373 | #endif /* CONFIG_DEBUG_FS */ | |
2374 | ||
2375 | module_init(init_kprobes); | |
1da177e4 | 2376 | |
99081ab5 | 2377 | /* defined in arch/.../kernel/kprobes.c */ |
1da177e4 | 2378 | EXPORT_SYMBOL_GPL(jprobe_return); |