]>
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 | |
bfd45be0 | 52 | #include <asm/sections.h> |
1da177e4 LT |
53 | #include <asm/cacheflush.h> |
54 | #include <asm/errno.h> | |
7c0f6ba6 | 55 | #include <linux/uaccess.h> |
1da177e4 LT |
56 | |
57 | #define KPROBE_HASH_BITS 6 | |
58 | #define KPROBE_TABLE_SIZE (1 << KPROBE_HASH_BITS) | |
59 | ||
3a872d89 | 60 | |
ef53d9c5 | 61 | static int kprobes_initialized; |
1da177e4 | 62 | static struct hlist_head kprobe_table[KPROBE_TABLE_SIZE]; |
b94cce92 | 63 | static struct hlist_head kretprobe_inst_table[KPROBE_TABLE_SIZE]; |
1da177e4 | 64 | |
bf8f6e5b | 65 | /* NOTE: change this value only with kprobe_mutex held */ |
e579abeb | 66 | static bool kprobes_all_disarmed; |
bf8f6e5b | 67 | |
43948f50 MH |
68 | /* This protects kprobe_table and optimizing_list */ |
69 | static DEFINE_MUTEX(kprobe_mutex); | |
e6584523 | 70 | static DEFINE_PER_CPU(struct kprobe *, kprobe_instance) = NULL; |
ef53d9c5 | 71 | static struct { |
ec484608 | 72 | raw_spinlock_t lock ____cacheline_aligned_in_smp; |
ef53d9c5 S |
73 | } kretprobe_table_locks[KPROBE_TABLE_SIZE]; |
74 | ||
290e3070 NR |
75 | kprobe_opcode_t * __weak kprobe_lookup_name(const char *name, |
76 | unsigned int __unused) | |
49e0b465 NR |
77 | { |
78 | return ((kprobe_opcode_t *)(kallsyms_lookup_name(name))); | |
79 | } | |
80 | ||
ec484608 | 81 | static raw_spinlock_t *kretprobe_table_lock_ptr(unsigned long hash) |
ef53d9c5 S |
82 | { |
83 | return &(kretprobe_table_locks[hash].lock); | |
84 | } | |
1da177e4 | 85 | |
376e2424 MH |
86 | /* Blacklist -- list of struct kprobe_blacklist_entry */ |
87 | static LIST_HEAD(kprobe_blacklist); | |
3d8d996e | 88 | |
2d14e39d | 89 | #ifdef __ARCH_WANT_KPROBES_INSN_SLOT |
9ec4b1f3 AM |
90 | /* |
91 | * kprobe->ainsn.insn points to the copy of the instruction to be | |
92 | * single-stepped. x86_64, POWER4 and above have no-exec support and | |
93 | * stepping on the instruction on a vmalloced/kmalloced/data page | |
94 | * is a recipe for disaster | |
95 | */ | |
9ec4b1f3 | 96 | struct kprobe_insn_page { |
c5cb5a2d | 97 | struct list_head list; |
9ec4b1f3 | 98 | kprobe_opcode_t *insns; /* Page of instruction slots */ |
af96397d | 99 | struct kprobe_insn_cache *cache; |
9ec4b1f3 | 100 | int nused; |
b4c6c34a | 101 | int ngarbage; |
4610ee1d | 102 | char slot_used[]; |
9ec4b1f3 AM |
103 | }; |
104 | ||
4610ee1d MH |
105 | #define KPROBE_INSN_PAGE_SIZE(slots) \ |
106 | (offsetof(struct kprobe_insn_page, slot_used) + \ | |
107 | (sizeof(char) * (slots))) | |
108 | ||
4610ee1d MH |
109 | static int slots_per_page(struct kprobe_insn_cache *c) |
110 | { | |
111 | return PAGE_SIZE/(c->insn_size * sizeof(kprobe_opcode_t)); | |
112 | } | |
113 | ||
ab40c5c6 MH |
114 | enum kprobe_slot_state { |
115 | SLOT_CLEAN = 0, | |
116 | SLOT_DIRTY = 1, | |
117 | SLOT_USED = 2, | |
118 | }; | |
119 | ||
63fef14f | 120 | void __weak *alloc_insn_page(void) |
af96397d HC |
121 | { |
122 | return module_alloc(PAGE_SIZE); | |
123 | } | |
124 | ||
c93f5cf5 | 125 | void __weak free_insn_page(void *page) |
af96397d | 126 | { |
be1f221c | 127 | module_memfree(page); |
af96397d HC |
128 | } |
129 | ||
c802d64a HC |
130 | struct kprobe_insn_cache kprobe_insn_slots = { |
131 | .mutex = __MUTEX_INITIALIZER(kprobe_insn_slots.mutex), | |
af96397d HC |
132 | .alloc = alloc_insn_page, |
133 | .free = free_insn_page, | |
4610ee1d MH |
134 | .pages = LIST_HEAD_INIT(kprobe_insn_slots.pages), |
135 | .insn_size = MAX_INSN_SIZE, | |
136 | .nr_garbage = 0, | |
137 | }; | |
55479f64 | 138 | static int collect_garbage_slots(struct kprobe_insn_cache *c); |
b4c6c34a | 139 | |
9ec4b1f3 | 140 | /** |
12941560 | 141 | * __get_insn_slot() - Find a slot on an executable page for an instruction. |
9ec4b1f3 AM |
142 | * We allocate an executable page if there's no room on existing ones. |
143 | */ | |
55479f64 | 144 | kprobe_opcode_t *__get_insn_slot(struct kprobe_insn_cache *c) |
9ec4b1f3 AM |
145 | { |
146 | struct kprobe_insn_page *kip; | |
c802d64a | 147 | kprobe_opcode_t *slot = NULL; |
9ec4b1f3 | 148 | |
5b485629 | 149 | /* Since the slot array is not protected by rcu, we need a mutex */ |
c802d64a | 150 | mutex_lock(&c->mutex); |
6f716acd | 151 | retry: |
5b485629 MH |
152 | rcu_read_lock(); |
153 | list_for_each_entry_rcu(kip, &c->pages, list) { | |
4610ee1d | 154 | if (kip->nused < slots_per_page(c)) { |
9ec4b1f3 | 155 | int i; |
4610ee1d | 156 | for (i = 0; i < slots_per_page(c); i++) { |
ab40c5c6 MH |
157 | if (kip->slot_used[i] == SLOT_CLEAN) { |
158 | kip->slot_used[i] = SLOT_USED; | |
9ec4b1f3 | 159 | kip->nused++; |
c802d64a | 160 | slot = kip->insns + (i * c->insn_size); |
5b485629 | 161 | rcu_read_unlock(); |
c802d64a | 162 | goto out; |
9ec4b1f3 AM |
163 | } |
164 | } | |
4610ee1d MH |
165 | /* kip->nused is broken. Fix it. */ |
166 | kip->nused = slots_per_page(c); | |
167 | WARN_ON(1); | |
9ec4b1f3 AM |
168 | } |
169 | } | |
5b485629 | 170 | rcu_read_unlock(); |
9ec4b1f3 | 171 | |
b4c6c34a | 172 | /* If there are any garbage slots, collect it and try again. */ |
4610ee1d | 173 | if (c->nr_garbage && collect_garbage_slots(c) == 0) |
b4c6c34a | 174 | goto retry; |
4610ee1d MH |
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); | |
6f716acd | 178 | if (!kip) |
c802d64a | 179 | goto out; |
9ec4b1f3 AM |
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 | */ | |
af96397d | 186 | kip->insns = c->alloc(); |
9ec4b1f3 AM |
187 | if (!kip->insns) { |
188 | kfree(kip); | |
c802d64a | 189 | goto out; |
9ec4b1f3 | 190 | } |
c5cb5a2d | 191 | INIT_LIST_HEAD(&kip->list); |
4610ee1d | 192 | memset(kip->slot_used, SLOT_CLEAN, slots_per_page(c)); |
ab40c5c6 | 193 | kip->slot_used[0] = SLOT_USED; |
9ec4b1f3 | 194 | kip->nused = 1; |
b4c6c34a | 195 | kip->ngarbage = 0; |
af96397d | 196 | kip->cache = c; |
5b485629 | 197 | list_add_rcu(&kip->list, &c->pages); |
c802d64a HC |
198 | slot = kip->insns; |
199 | out: | |
200 | mutex_unlock(&c->mutex); | |
201 | return slot; | |
12941560 MH |
202 | } |
203 | ||
b4c6c34a | 204 | /* Return 1 if all garbages are collected, otherwise 0. */ |
55479f64 | 205 | static int collect_one_slot(struct kprobe_insn_page *kip, int idx) |
b4c6c34a | 206 | { |
ab40c5c6 | 207 | kip->slot_used[idx] = SLOT_CLEAN; |
b4c6c34a MH |
208 | kip->nused--; |
209 | if (kip->nused == 0) { | |
210 | /* | |
211 | * Page is no longer in use. Free it unless | |
212 | * it's the last one. We keep the last one | |
213 | * so as not to have to set it up again the | |
214 | * next time somebody inserts a probe. | |
215 | */ | |
4610ee1d | 216 | if (!list_is_singular(&kip->list)) { |
5b485629 MH |
217 | list_del_rcu(&kip->list); |
218 | synchronize_rcu(); | |
af96397d | 219 | kip->cache->free(kip->insns); |
b4c6c34a MH |
220 | kfree(kip); |
221 | } | |
222 | return 1; | |
223 | } | |
224 | return 0; | |
225 | } | |
226 | ||
55479f64 | 227 | static int collect_garbage_slots(struct kprobe_insn_cache *c) |
b4c6c34a | 228 | { |
c5cb5a2d | 229 | struct kprobe_insn_page *kip, *next; |
b4c6c34a | 230 | |
615d0ebb MH |
231 | /* Ensure no-one is interrupted on the garbages */ |
232 | synchronize_sched(); | |
b4c6c34a | 233 | |
4610ee1d | 234 | list_for_each_entry_safe(kip, next, &c->pages, list) { |
b4c6c34a | 235 | int i; |
b4c6c34a MH |
236 | if (kip->ngarbage == 0) |
237 | continue; | |
238 | kip->ngarbage = 0; /* we will collect all garbages */ | |
4610ee1d | 239 | for (i = 0; i < slots_per_page(c); i++) { |
5b485629 | 240 | if (kip->slot_used[i] == SLOT_DIRTY && collect_one_slot(kip, i)) |
b4c6c34a MH |
241 | break; |
242 | } | |
243 | } | |
4610ee1d | 244 | c->nr_garbage = 0; |
b4c6c34a MH |
245 | return 0; |
246 | } | |
247 | ||
55479f64 MH |
248 | void __free_insn_slot(struct kprobe_insn_cache *c, |
249 | kprobe_opcode_t *slot, int dirty) | |
9ec4b1f3 AM |
250 | { |
251 | struct kprobe_insn_page *kip; | |
5b485629 | 252 | long idx; |
9ec4b1f3 | 253 | |
c802d64a | 254 | mutex_lock(&c->mutex); |
5b485629 MH |
255 | rcu_read_lock(); |
256 | list_for_each_entry_rcu(kip, &c->pages, list) { | |
257 | idx = ((long)slot - (long)kip->insns) / | |
258 | (c->insn_size * sizeof(kprobe_opcode_t)); | |
259 | if (idx >= 0 && idx < slots_per_page(c)) | |
c802d64a | 260 | goto out; |
9ec4b1f3 | 261 | } |
5b485629 | 262 | /* Could not find this slot. */ |
4610ee1d | 263 | WARN_ON(1); |
5b485629 | 264 | kip = NULL; |
c802d64a | 265 | out: |
5b485629 MH |
266 | rcu_read_unlock(); |
267 | /* Mark and sweep: this may sleep */ | |
268 | if (kip) { | |
269 | /* Check double free */ | |
270 | WARN_ON(kip->slot_used[idx] != SLOT_USED); | |
271 | if (dirty) { | |
272 | kip->slot_used[idx] = SLOT_DIRTY; | |
273 | kip->ngarbage++; | |
274 | if (++c->nr_garbage > slots_per_page(c)) | |
275 | collect_garbage_slots(c); | |
276 | } else { | |
277 | collect_one_slot(kip, idx); | |
278 | } | |
279 | } | |
c802d64a | 280 | mutex_unlock(&c->mutex); |
4610ee1d | 281 | } |
6f716acd | 282 | |
5b485629 MH |
283 | /* |
284 | * Check given address is on the page of kprobe instruction slots. | |
285 | * This will be used for checking whether the address on a stack | |
286 | * is on a text area or not. | |
287 | */ | |
288 | bool __is_insn_slot_addr(struct kprobe_insn_cache *c, unsigned long addr) | |
289 | { | |
290 | struct kprobe_insn_page *kip; | |
291 | bool ret = false; | |
292 | ||
293 | rcu_read_lock(); | |
294 | list_for_each_entry_rcu(kip, &c->pages, list) { | |
295 | if (addr >= (unsigned long)kip->insns && | |
296 | addr < (unsigned long)kip->insns + PAGE_SIZE) { | |
297 | ret = true; | |
298 | break; | |
299 | } | |
300 | } | |
301 | rcu_read_unlock(); | |
302 | ||
303 | return ret; | |
304 | } | |
305 | ||
afd66255 MH |
306 | #ifdef CONFIG_OPTPROBES |
307 | /* For optimized_kprobe buffer */ | |
c802d64a HC |
308 | struct kprobe_insn_cache kprobe_optinsn_slots = { |
309 | .mutex = __MUTEX_INITIALIZER(kprobe_optinsn_slots.mutex), | |
af96397d HC |
310 | .alloc = alloc_insn_page, |
311 | .free = free_insn_page, | |
afd66255 MH |
312 | .pages = LIST_HEAD_INIT(kprobe_optinsn_slots.pages), |
313 | /* .insn_size is initialized later */ | |
314 | .nr_garbage = 0, | |
315 | }; | |
afd66255 | 316 | #endif |
2d14e39d | 317 | #endif |
9ec4b1f3 | 318 | |
e6584523 AM |
319 | /* We have preemption disabled.. so it is safe to use __ versions */ |
320 | static inline void set_kprobe_instance(struct kprobe *kp) | |
321 | { | |
b76834bc | 322 | __this_cpu_write(kprobe_instance, kp); |
e6584523 AM |
323 | } |
324 | ||
325 | static inline void reset_kprobe_instance(void) | |
326 | { | |
b76834bc | 327 | __this_cpu_write(kprobe_instance, NULL); |
e6584523 AM |
328 | } |
329 | ||
3516a460 AM |
330 | /* |
331 | * This routine is called either: | |
49a2a1b8 | 332 | * - under the kprobe_mutex - during kprobe_[un]register() |
3516a460 | 333 | * OR |
d217d545 | 334 | * - with preemption disabled - from arch/xxx/kernel/kprobes.c |
3516a460 | 335 | */ |
820aede0 | 336 | struct kprobe *get_kprobe(void *addr) |
1da177e4 LT |
337 | { |
338 | struct hlist_head *head; | |
3516a460 | 339 | struct kprobe *p; |
1da177e4 LT |
340 | |
341 | head = &kprobe_table[hash_ptr(addr, KPROBE_HASH_BITS)]; | |
b67bfe0d | 342 | hlist_for_each_entry_rcu(p, head, hlist) { |
1da177e4 LT |
343 | if (p->addr == addr) |
344 | return p; | |
345 | } | |
afd66255 | 346 | |
1da177e4 LT |
347 | return NULL; |
348 | } | |
820aede0 | 349 | NOKPROBE_SYMBOL(get_kprobe); |
1da177e4 | 350 | |
820aede0 | 351 | static int aggr_pre_handler(struct kprobe *p, struct pt_regs *regs); |
afd66255 MH |
352 | |
353 | /* Return true if the kprobe is an aggregator */ | |
354 | static inline int kprobe_aggrprobe(struct kprobe *p) | |
355 | { | |
356 | return p->pre_handler == aggr_pre_handler; | |
357 | } | |
358 | ||
6274de49 MH |
359 | /* Return true(!0) if the kprobe is unused */ |
360 | static inline int kprobe_unused(struct kprobe *p) | |
361 | { | |
362 | return kprobe_aggrprobe(p) && kprobe_disabled(p) && | |
363 | list_empty(&p->list); | |
364 | } | |
365 | ||
afd66255 MH |
366 | /* |
367 | * Keep all fields in the kprobe consistent | |
368 | */ | |
6d8e40a8 | 369 | static inline void copy_kprobe(struct kprobe *ap, struct kprobe *p) |
afd66255 | 370 | { |
6d8e40a8 MH |
371 | memcpy(&p->opcode, &ap->opcode, sizeof(kprobe_opcode_t)); |
372 | memcpy(&p->ainsn, &ap->ainsn, sizeof(struct arch_specific_insn)); | |
afd66255 MH |
373 | } |
374 | ||
375 | #ifdef CONFIG_OPTPROBES | |
b2be84df MH |
376 | /* NOTE: change this value only with kprobe_mutex held */ |
377 | static bool kprobes_allow_optimization; | |
378 | ||
afd66255 MH |
379 | /* |
380 | * Call all pre_handler on the list, but ignores its return value. | |
381 | * This must be called from arch-dep optimized caller. | |
382 | */ | |
820aede0 | 383 | void opt_pre_handler(struct kprobe *p, struct pt_regs *regs) |
afd66255 MH |
384 | { |
385 | struct kprobe *kp; | |
386 | ||
387 | list_for_each_entry_rcu(kp, &p->list, list) { | |
388 | if (kp->pre_handler && likely(!kprobe_disabled(kp))) { | |
389 | set_kprobe_instance(kp); | |
4f3a8714 | 390 | kp->pre_handler(kp, regs); |
afd66255 MH |
391 | } |
392 | reset_kprobe_instance(); | |
393 | } | |
394 | } | |
820aede0 | 395 | NOKPROBE_SYMBOL(opt_pre_handler); |
afd66255 | 396 | |
6274de49 | 397 | /* Free optimized instructions and optimized_kprobe */ |
55479f64 | 398 | static void free_aggr_kprobe(struct kprobe *p) |
6274de49 MH |
399 | { |
400 | struct optimized_kprobe *op; | |
401 | ||
402 | op = container_of(p, struct optimized_kprobe, kp); | |
403 | arch_remove_optimized_kprobe(op); | |
404 | arch_remove_kprobe(p); | |
405 | kfree(op); | |
406 | } | |
407 | ||
afd66255 MH |
408 | /* Return true(!0) if the kprobe is ready for optimization. */ |
409 | static inline int kprobe_optready(struct kprobe *p) | |
410 | { | |
411 | struct optimized_kprobe *op; | |
412 | ||
413 | if (kprobe_aggrprobe(p)) { | |
414 | op = container_of(p, struct optimized_kprobe, kp); | |
415 | return arch_prepared_optinsn(&op->optinsn); | |
416 | } | |
417 | ||
418 | return 0; | |
419 | } | |
420 | ||
6274de49 MH |
421 | /* Return true(!0) if the kprobe is disarmed. Note: p must be on hash list */ |
422 | static inline int kprobe_disarmed(struct kprobe *p) | |
423 | { | |
424 | struct optimized_kprobe *op; | |
425 | ||
426 | /* If kprobe is not aggr/opt probe, just return kprobe is disabled */ | |
427 | if (!kprobe_aggrprobe(p)) | |
428 | return kprobe_disabled(p); | |
429 | ||
430 | op = container_of(p, struct optimized_kprobe, kp); | |
431 | ||
432 | return kprobe_disabled(p) && list_empty(&op->list); | |
433 | } | |
434 | ||
435 | /* Return true(!0) if the probe is queued on (un)optimizing lists */ | |
55479f64 | 436 | static int kprobe_queued(struct kprobe *p) |
6274de49 MH |
437 | { |
438 | struct optimized_kprobe *op; | |
439 | ||
440 | if (kprobe_aggrprobe(p)) { | |
441 | op = container_of(p, struct optimized_kprobe, kp); | |
442 | if (!list_empty(&op->list)) | |
443 | return 1; | |
444 | } | |
445 | return 0; | |
446 | } | |
447 | ||
afd66255 MH |
448 | /* |
449 | * Return an optimized kprobe whose optimizing code replaces | |
450 | * instructions including addr (exclude breakpoint). | |
451 | */ | |
55479f64 | 452 | static struct kprobe *get_optimized_kprobe(unsigned long addr) |
afd66255 MH |
453 | { |
454 | int i; | |
455 | struct kprobe *p = NULL; | |
456 | struct optimized_kprobe *op; | |
457 | ||
458 | /* Don't check i == 0, since that is a breakpoint case. */ | |
459 | for (i = 1; !p && i < MAX_OPTIMIZED_LENGTH; i++) | |
460 | p = get_kprobe((void *)(addr - i)); | |
461 | ||
462 | if (p && kprobe_optready(p)) { | |
463 | op = container_of(p, struct optimized_kprobe, kp); | |
464 | if (arch_within_optimized_kprobe(op, addr)) | |
465 | return p; | |
466 | } | |
467 | ||
468 | return NULL; | |
469 | } | |
470 | ||
471 | /* Optimization staging list, protected by kprobe_mutex */ | |
472 | static LIST_HEAD(optimizing_list); | |
6274de49 | 473 | static LIST_HEAD(unoptimizing_list); |
7b959fc5 | 474 | static LIST_HEAD(freeing_list); |
afd66255 MH |
475 | |
476 | static void kprobe_optimizer(struct work_struct *work); | |
477 | static DECLARE_DELAYED_WORK(optimizing_work, kprobe_optimizer); | |
478 | #define OPTIMIZE_DELAY 5 | |
479 | ||
61f4e13f MH |
480 | /* |
481 | * Optimize (replace a breakpoint with a jump) kprobes listed on | |
482 | * optimizing_list. | |
483 | */ | |
55479f64 | 484 | static void do_optimize_kprobes(void) |
afd66255 | 485 | { |
afd66255 MH |
486 | /* |
487 | * The optimization/unoptimization refers online_cpus via | |
488 | * stop_machine() and cpu-hotplug modifies online_cpus. | |
489 | * And same time, text_mutex will be held in cpu-hotplug and here. | |
490 | * This combination can cause a deadlock (cpu-hotplug try to lock | |
491 | * text_mutex but stop_machine can not be done because online_cpus | |
492 | * has been changed) | |
2d1e38f5 | 493 | * To avoid this deadlock, caller must have locked cpu hotplug |
afd66255 MH |
494 | * for preventing cpu-hotplug outside of text_mutex locking. |
495 | */ | |
2d1e38f5 TG |
496 | lockdep_assert_cpus_held(); |
497 | ||
498 | /* Optimization never be done when disarmed */ | |
499 | if (kprobes_all_disarmed || !kprobes_allow_optimization || | |
500 | list_empty(&optimizing_list)) | |
501 | return; | |
502 | ||
afd66255 | 503 | mutex_lock(&text_mutex); |
cd7ebe22 | 504 | arch_optimize_kprobes(&optimizing_list); |
afd66255 | 505 | mutex_unlock(&text_mutex); |
61f4e13f MH |
506 | } |
507 | ||
6274de49 MH |
508 | /* |
509 | * Unoptimize (replace a jump with a breakpoint and remove the breakpoint | |
510 | * if need) kprobes listed on unoptimizing_list. | |
511 | */ | |
55479f64 | 512 | static void do_unoptimize_kprobes(void) |
6274de49 MH |
513 | { |
514 | struct optimized_kprobe *op, *tmp; | |
515 | ||
2d1e38f5 TG |
516 | /* See comment in do_optimize_kprobes() */ |
517 | lockdep_assert_cpus_held(); | |
518 | ||
6274de49 MH |
519 | /* Unoptimization must be done anytime */ |
520 | if (list_empty(&unoptimizing_list)) | |
521 | return; | |
522 | ||
6274de49 | 523 | mutex_lock(&text_mutex); |
7b959fc5 | 524 | arch_unoptimize_kprobes(&unoptimizing_list, &freeing_list); |
f984ba4e | 525 | /* Loop free_list for disarming */ |
7b959fc5 | 526 | list_for_each_entry_safe(op, tmp, &freeing_list, list) { |
6274de49 MH |
527 | /* Disarm probes if marked disabled */ |
528 | if (kprobe_disabled(&op->kp)) | |
529 | arch_disarm_kprobe(&op->kp); | |
530 | if (kprobe_unused(&op->kp)) { | |
531 | /* | |
532 | * Remove unused probes from hash list. After waiting | |
533 | * for synchronization, these probes are reclaimed. | |
534 | * (reclaiming is done by do_free_cleaned_kprobes.) | |
535 | */ | |
536 | hlist_del_rcu(&op->kp.hlist); | |
6274de49 MH |
537 | } else |
538 | list_del_init(&op->list); | |
539 | } | |
540 | mutex_unlock(&text_mutex); | |
6274de49 MH |
541 | } |
542 | ||
543 | /* Reclaim all kprobes on the free_list */ | |
55479f64 | 544 | static void do_free_cleaned_kprobes(void) |
6274de49 MH |
545 | { |
546 | struct optimized_kprobe *op, *tmp; | |
547 | ||
7b959fc5 | 548 | list_for_each_entry_safe(op, tmp, &freeing_list, list) { |
6274de49 MH |
549 | BUG_ON(!kprobe_unused(&op->kp)); |
550 | list_del_init(&op->list); | |
551 | free_aggr_kprobe(&op->kp); | |
552 | } | |
553 | } | |
554 | ||
555 | /* Start optimizer after OPTIMIZE_DELAY passed */ | |
55479f64 | 556 | static void kick_kprobe_optimizer(void) |
6274de49 | 557 | { |
ad72b3be | 558 | schedule_delayed_work(&optimizing_work, OPTIMIZE_DELAY); |
6274de49 MH |
559 | } |
560 | ||
61f4e13f | 561 | /* Kprobe jump optimizer */ |
55479f64 | 562 | static void kprobe_optimizer(struct work_struct *work) |
61f4e13f | 563 | { |
72ef3794 | 564 | mutex_lock(&kprobe_mutex); |
2d1e38f5 | 565 | cpus_read_lock(); |
61f4e13f MH |
566 | /* Lock modules while optimizing kprobes */ |
567 | mutex_lock(&module_mutex); | |
61f4e13f MH |
568 | |
569 | /* | |
6274de49 MH |
570 | * Step 1: Unoptimize kprobes and collect cleaned (unused and disarmed) |
571 | * kprobes before waiting for quiesence period. | |
572 | */ | |
7b959fc5 | 573 | do_unoptimize_kprobes(); |
6274de49 MH |
574 | |
575 | /* | |
a30b85df MH |
576 | * Step 2: Wait for quiesence period to ensure all potentially |
577 | * preempted tasks to have normally scheduled. Because optprobe | |
578 | * may modify multiple instructions, there is a chance that Nth | |
579 | * instruction is preempted. In that case, such tasks can return | |
580 | * to 2nd-Nth byte of jump instruction. This wait is for avoiding it. | |
581 | * Note that on non-preemptive kernel, this is transparently converted | |
582 | * to synchronoze_sched() to wait for all interrupts to have completed. | |
61f4e13f | 583 | */ |
a30b85df | 584 | synchronize_rcu_tasks(); |
61f4e13f | 585 | |
6274de49 | 586 | /* Step 3: Optimize kprobes after quiesence period */ |
61f4e13f | 587 | do_optimize_kprobes(); |
6274de49 MH |
588 | |
589 | /* Step 4: Free cleaned kprobes after quiesence period */ | |
7b959fc5 | 590 | do_free_cleaned_kprobes(); |
6274de49 | 591 | |
afd66255 | 592 | mutex_unlock(&module_mutex); |
2d1e38f5 | 593 | cpus_read_unlock(); |
72ef3794 | 594 | mutex_unlock(&kprobe_mutex); |
6274de49 | 595 | |
cd7ebe22 | 596 | /* Step 5: Kick optimizer again if needed */ |
f984ba4e | 597 | if (!list_empty(&optimizing_list) || !list_empty(&unoptimizing_list)) |
cd7ebe22 | 598 | kick_kprobe_optimizer(); |
6274de49 MH |
599 | } |
600 | ||
601 | /* Wait for completing optimization and unoptimization */ | |
30e7d894 | 602 | void wait_for_kprobe_optimizer(void) |
6274de49 | 603 | { |
ad72b3be TH |
604 | mutex_lock(&kprobe_mutex); |
605 | ||
606 | while (!list_empty(&optimizing_list) || !list_empty(&unoptimizing_list)) { | |
607 | mutex_unlock(&kprobe_mutex); | |
608 | ||
609 | /* this will also make optimizing_work execute immmediately */ | |
610 | flush_delayed_work(&optimizing_work); | |
611 | /* @optimizing_work might not have been queued yet, relax */ | |
612 | cpu_relax(); | |
613 | ||
614 | mutex_lock(&kprobe_mutex); | |
615 | } | |
616 | ||
617 | mutex_unlock(&kprobe_mutex); | |
afd66255 MH |
618 | } |
619 | ||
620 | /* Optimize kprobe if p is ready to be optimized */ | |
55479f64 | 621 | static void optimize_kprobe(struct kprobe *p) |
afd66255 MH |
622 | { |
623 | struct optimized_kprobe *op; | |
624 | ||
625 | /* Check if the kprobe is disabled or not ready for optimization. */ | |
b2be84df | 626 | if (!kprobe_optready(p) || !kprobes_allow_optimization || |
afd66255 MH |
627 | (kprobe_disabled(p) || kprobes_all_disarmed)) |
628 | return; | |
629 | ||
630 | /* Both of break_handler and post_handler are not supported. */ | |
631 | if (p->break_handler || p->post_handler) | |
632 | return; | |
633 | ||
634 | op = container_of(p, struct optimized_kprobe, kp); | |
635 | ||
636 | /* Check there is no other kprobes at the optimized instructions */ | |
637 | if (arch_check_optimized_kprobe(op) < 0) | |
638 | return; | |
639 | ||
640 | /* Check if it is already optimized. */ | |
641 | if (op->kp.flags & KPROBE_FLAG_OPTIMIZED) | |
642 | return; | |
afd66255 | 643 | op->kp.flags |= KPROBE_FLAG_OPTIMIZED; |
6274de49 MH |
644 | |
645 | if (!list_empty(&op->list)) | |
646 | /* This is under unoptimizing. Just dequeue the probe */ | |
647 | list_del_init(&op->list); | |
648 | else { | |
649 | list_add(&op->list, &optimizing_list); | |
650 | kick_kprobe_optimizer(); | |
651 | } | |
652 | } | |
653 | ||
654 | /* Short cut to direct unoptimizing */ | |
55479f64 | 655 | static void force_unoptimize_kprobe(struct optimized_kprobe *op) |
6274de49 | 656 | { |
2d1e38f5 | 657 | lockdep_assert_cpus_held(); |
6274de49 | 658 | arch_unoptimize_kprobe(op); |
6274de49 MH |
659 | if (kprobe_disabled(&op->kp)) |
660 | arch_disarm_kprobe(&op->kp); | |
afd66255 MH |
661 | } |
662 | ||
663 | /* Unoptimize a kprobe if p is optimized */ | |
55479f64 | 664 | static void unoptimize_kprobe(struct kprobe *p, bool force) |
afd66255 MH |
665 | { |
666 | struct optimized_kprobe *op; | |
667 | ||
6274de49 MH |
668 | if (!kprobe_aggrprobe(p) || kprobe_disarmed(p)) |
669 | return; /* This is not an optprobe nor optimized */ | |
670 | ||
671 | op = container_of(p, struct optimized_kprobe, kp); | |
672 | if (!kprobe_optimized(p)) { | |
673 | /* Unoptimized or unoptimizing case */ | |
674 | if (force && !list_empty(&op->list)) { | |
675 | /* | |
676 | * Only if this is unoptimizing kprobe and forced, | |
677 | * forcibly unoptimize it. (No need to unoptimize | |
678 | * unoptimized kprobe again :) | |
679 | */ | |
afd66255 | 680 | list_del_init(&op->list); |
6274de49 MH |
681 | force_unoptimize_kprobe(op); |
682 | } | |
683 | return; | |
684 | } | |
685 | ||
686 | op->kp.flags &= ~KPROBE_FLAG_OPTIMIZED; | |
687 | if (!list_empty(&op->list)) { | |
688 | /* Dequeue from the optimization queue */ | |
689 | list_del_init(&op->list); | |
690 | return; | |
691 | } | |
692 | /* Optimized kprobe case */ | |
693 | if (force) | |
694 | /* Forcibly update the code: this is a special case */ | |
695 | force_unoptimize_kprobe(op); | |
696 | else { | |
697 | list_add(&op->list, &unoptimizing_list); | |
698 | kick_kprobe_optimizer(); | |
afd66255 MH |
699 | } |
700 | } | |
701 | ||
0490cd1f MH |
702 | /* Cancel unoptimizing for reusing */ |
703 | static void reuse_unused_kprobe(struct kprobe *ap) | |
704 | { | |
705 | struct optimized_kprobe *op; | |
706 | ||
707 | BUG_ON(!kprobe_unused(ap)); | |
708 | /* | |
709 | * Unused kprobe MUST be on the way of delayed unoptimizing (means | |
710 | * there is still a relative jump) and disabled. | |
711 | */ | |
712 | op = container_of(ap, struct optimized_kprobe, kp); | |
713 | if (unlikely(list_empty(&op->list))) | |
714 | printk(KERN_WARNING "Warning: found a stray unused " | |
715 | "aggrprobe@%p\n", ap->addr); | |
716 | /* Enable the probe again */ | |
717 | ap->flags &= ~KPROBE_FLAG_DISABLED; | |
718 | /* Optimize it again (remove from op->list) */ | |
719 | BUG_ON(!kprobe_optready(ap)); | |
720 | optimize_kprobe(ap); | |
721 | } | |
722 | ||
afd66255 | 723 | /* Remove optimized instructions */ |
55479f64 | 724 | static void kill_optimized_kprobe(struct kprobe *p) |
afd66255 MH |
725 | { |
726 | struct optimized_kprobe *op; | |
727 | ||
728 | op = container_of(p, struct optimized_kprobe, kp); | |
6274de49 MH |
729 | if (!list_empty(&op->list)) |
730 | /* Dequeue from the (un)optimization queue */ | |
afd66255 | 731 | list_del_init(&op->list); |
6274de49 | 732 | op->kp.flags &= ~KPROBE_FLAG_OPTIMIZED; |
7b959fc5 MH |
733 | |
734 | if (kprobe_unused(p)) { | |
735 | /* Enqueue if it is unused */ | |
736 | list_add(&op->list, &freeing_list); | |
737 | /* | |
738 | * Remove unused probes from the hash list. After waiting | |
739 | * for synchronization, this probe is reclaimed. | |
740 | * (reclaiming is done by do_free_cleaned_kprobes().) | |
741 | */ | |
742 | hlist_del_rcu(&op->kp.hlist); | |
743 | } | |
744 | ||
6274de49 | 745 | /* Don't touch the code, because it is already freed. */ |
afd66255 MH |
746 | arch_remove_optimized_kprobe(op); |
747 | } | |
748 | ||
a460246c MH |
749 | static inline |
750 | void __prepare_optimized_kprobe(struct optimized_kprobe *op, struct kprobe *p) | |
751 | { | |
752 | if (!kprobe_ftrace(p)) | |
753 | arch_prepare_optimized_kprobe(op, p); | |
754 | } | |
755 | ||
afd66255 | 756 | /* Try to prepare optimized instructions */ |
55479f64 | 757 | static void prepare_optimized_kprobe(struct kprobe *p) |
afd66255 MH |
758 | { |
759 | struct optimized_kprobe *op; | |
760 | ||
761 | op = container_of(p, struct optimized_kprobe, kp); | |
a460246c | 762 | __prepare_optimized_kprobe(op, p); |
afd66255 MH |
763 | } |
764 | ||
afd66255 | 765 | /* Allocate new optimized_kprobe and try to prepare optimized instructions */ |
55479f64 | 766 | static struct kprobe *alloc_aggr_kprobe(struct kprobe *p) |
afd66255 MH |
767 | { |
768 | struct optimized_kprobe *op; | |
769 | ||
770 | op = kzalloc(sizeof(struct optimized_kprobe), GFP_KERNEL); | |
771 | if (!op) | |
772 | return NULL; | |
773 | ||
774 | INIT_LIST_HEAD(&op->list); | |
775 | op->kp.addr = p->addr; | |
a460246c | 776 | __prepare_optimized_kprobe(op, p); |
afd66255 MH |
777 | |
778 | return &op->kp; | |
779 | } | |
780 | ||
55479f64 | 781 | static void init_aggr_kprobe(struct kprobe *ap, struct kprobe *p); |
afd66255 MH |
782 | |
783 | /* | |
784 | * Prepare an optimized_kprobe and optimize it | |
785 | * NOTE: p must be a normal registered kprobe | |
786 | */ | |
55479f64 | 787 | static void try_to_optimize_kprobe(struct kprobe *p) |
afd66255 MH |
788 | { |
789 | struct kprobe *ap; | |
790 | struct optimized_kprobe *op; | |
791 | ||
ae6aa16f MH |
792 | /* Impossible to optimize ftrace-based kprobe */ |
793 | if (kprobe_ftrace(p)) | |
794 | return; | |
795 | ||
25764288 | 796 | /* For preparing optimization, jump_label_text_reserved() is called */ |
2d1e38f5 | 797 | cpus_read_lock(); |
25764288 MH |
798 | jump_label_lock(); |
799 | mutex_lock(&text_mutex); | |
800 | ||
afd66255 MH |
801 | ap = alloc_aggr_kprobe(p); |
802 | if (!ap) | |
25764288 | 803 | goto out; |
afd66255 MH |
804 | |
805 | op = container_of(ap, struct optimized_kprobe, kp); | |
806 | if (!arch_prepared_optinsn(&op->optinsn)) { | |
807 | /* If failed to setup optimizing, fallback to kprobe */ | |
6274de49 MH |
808 | arch_remove_optimized_kprobe(op); |
809 | kfree(op); | |
25764288 | 810 | goto out; |
afd66255 MH |
811 | } |
812 | ||
813 | init_aggr_kprobe(ap, p); | |
25764288 MH |
814 | optimize_kprobe(ap); /* This just kicks optimizer thread */ |
815 | ||
816 | out: | |
817 | mutex_unlock(&text_mutex); | |
818 | jump_label_unlock(); | |
2d1e38f5 | 819 | cpus_read_unlock(); |
afd66255 MH |
820 | } |
821 | ||
b2be84df | 822 | #ifdef CONFIG_SYSCTL |
55479f64 | 823 | static void optimize_all_kprobes(void) |
b2be84df MH |
824 | { |
825 | struct hlist_head *head; | |
b2be84df MH |
826 | struct kprobe *p; |
827 | unsigned int i; | |
828 | ||
5c51543b | 829 | mutex_lock(&kprobe_mutex); |
b2be84df MH |
830 | /* If optimization is already allowed, just return */ |
831 | if (kprobes_allow_optimization) | |
5c51543b | 832 | goto out; |
b2be84df | 833 | |
2d1e38f5 | 834 | cpus_read_lock(); |
b2be84df | 835 | kprobes_allow_optimization = true; |
b2be84df MH |
836 | for (i = 0; i < KPROBE_TABLE_SIZE; i++) { |
837 | head = &kprobe_table[i]; | |
b67bfe0d | 838 | hlist_for_each_entry_rcu(p, head, hlist) |
b2be84df MH |
839 | if (!kprobe_disabled(p)) |
840 | optimize_kprobe(p); | |
841 | } | |
2d1e38f5 | 842 | cpus_read_unlock(); |
b2be84df | 843 | printk(KERN_INFO "Kprobes globally optimized\n"); |
5c51543b MH |
844 | out: |
845 | mutex_unlock(&kprobe_mutex); | |
b2be84df MH |
846 | } |
847 | ||
55479f64 | 848 | static void unoptimize_all_kprobes(void) |
b2be84df MH |
849 | { |
850 | struct hlist_head *head; | |
b2be84df MH |
851 | struct kprobe *p; |
852 | unsigned int i; | |
853 | ||
5c51543b | 854 | mutex_lock(&kprobe_mutex); |
b2be84df | 855 | /* If optimization is already prohibited, just return */ |
5c51543b MH |
856 | if (!kprobes_allow_optimization) { |
857 | mutex_unlock(&kprobe_mutex); | |
b2be84df | 858 | return; |
5c51543b | 859 | } |
b2be84df | 860 | |
2d1e38f5 | 861 | cpus_read_lock(); |
b2be84df | 862 | kprobes_allow_optimization = false; |
b2be84df MH |
863 | for (i = 0; i < KPROBE_TABLE_SIZE; i++) { |
864 | head = &kprobe_table[i]; | |
b67bfe0d | 865 | hlist_for_each_entry_rcu(p, head, hlist) { |
b2be84df | 866 | if (!kprobe_disabled(p)) |
6274de49 | 867 | unoptimize_kprobe(p, false); |
b2be84df MH |
868 | } |
869 | } | |
2d1e38f5 | 870 | cpus_read_unlock(); |
5c51543b MH |
871 | mutex_unlock(&kprobe_mutex); |
872 | ||
6274de49 MH |
873 | /* Wait for unoptimizing completion */ |
874 | wait_for_kprobe_optimizer(); | |
875 | printk(KERN_INFO "Kprobes globally unoptimized\n"); | |
b2be84df MH |
876 | } |
877 | ||
5c51543b | 878 | static DEFINE_MUTEX(kprobe_sysctl_mutex); |
b2be84df MH |
879 | int sysctl_kprobes_optimization; |
880 | int proc_kprobes_optimization_handler(struct ctl_table *table, int write, | |
881 | void __user *buffer, size_t *length, | |
882 | loff_t *ppos) | |
883 | { | |
884 | int ret; | |
885 | ||
5c51543b | 886 | mutex_lock(&kprobe_sysctl_mutex); |
b2be84df MH |
887 | sysctl_kprobes_optimization = kprobes_allow_optimization ? 1 : 0; |
888 | ret = proc_dointvec_minmax(table, write, buffer, length, ppos); | |
889 | ||
890 | if (sysctl_kprobes_optimization) | |
891 | optimize_all_kprobes(); | |
892 | else | |
893 | unoptimize_all_kprobes(); | |
5c51543b | 894 | mutex_unlock(&kprobe_sysctl_mutex); |
b2be84df MH |
895 | |
896 | return ret; | |
897 | } | |
898 | #endif /* CONFIG_SYSCTL */ | |
899 | ||
6274de49 | 900 | /* Put a breakpoint for a probe. Must be called with text_mutex locked */ |
55479f64 | 901 | static void __arm_kprobe(struct kprobe *p) |
afd66255 | 902 | { |
6d8e40a8 | 903 | struct kprobe *_p; |
afd66255 MH |
904 | |
905 | /* Check collision with other optimized kprobes */ | |
6d8e40a8 MH |
906 | _p = get_optimized_kprobe((unsigned long)p->addr); |
907 | if (unlikely(_p)) | |
6274de49 MH |
908 | /* Fallback to unoptimized kprobe */ |
909 | unoptimize_kprobe(_p, true); | |
afd66255 MH |
910 | |
911 | arch_arm_kprobe(p); | |
912 | optimize_kprobe(p); /* Try to optimize (add kprobe to a list) */ | |
913 | } | |
914 | ||
6274de49 | 915 | /* Remove the breakpoint of a probe. Must be called with text_mutex locked */ |
55479f64 | 916 | static void __disarm_kprobe(struct kprobe *p, bool reopt) |
afd66255 | 917 | { |
6d8e40a8 | 918 | struct kprobe *_p; |
afd66255 | 919 | |
69d54b91 WN |
920 | /* Try to unoptimize */ |
921 | unoptimize_kprobe(p, kprobes_all_disarmed); | |
afd66255 | 922 | |
6274de49 MH |
923 | if (!kprobe_queued(p)) { |
924 | arch_disarm_kprobe(p); | |
925 | /* If another kprobe was blocked, optimize it. */ | |
926 | _p = get_optimized_kprobe((unsigned long)p->addr); | |
927 | if (unlikely(_p) && reopt) | |
928 | optimize_kprobe(_p); | |
929 | } | |
930 | /* TODO: reoptimize others after unoptimized this probe */ | |
afd66255 MH |
931 | } |
932 | ||
933 | #else /* !CONFIG_OPTPROBES */ | |
934 | ||
935 | #define optimize_kprobe(p) do {} while (0) | |
6274de49 | 936 | #define unoptimize_kprobe(p, f) do {} while (0) |
afd66255 MH |
937 | #define kill_optimized_kprobe(p) do {} while (0) |
938 | #define prepare_optimized_kprobe(p) do {} while (0) | |
939 | #define try_to_optimize_kprobe(p) do {} while (0) | |
940 | #define __arm_kprobe(p) arch_arm_kprobe(p) | |
6274de49 MH |
941 | #define __disarm_kprobe(p, o) arch_disarm_kprobe(p) |
942 | #define kprobe_disarmed(p) kprobe_disabled(p) | |
943 | #define wait_for_kprobe_optimizer() do {} while (0) | |
afd66255 | 944 | |
0490cd1f MH |
945 | /* There should be no unused kprobes can be reused without optimization */ |
946 | static void reuse_unused_kprobe(struct kprobe *ap) | |
947 | { | |
948 | printk(KERN_ERR "Error: There should be no unused kprobe here.\n"); | |
949 | BUG_ON(kprobe_unused(ap)); | |
950 | } | |
951 | ||
55479f64 | 952 | static void free_aggr_kprobe(struct kprobe *p) |
afd66255 | 953 | { |
6274de49 | 954 | arch_remove_kprobe(p); |
afd66255 MH |
955 | kfree(p); |
956 | } | |
957 | ||
55479f64 | 958 | static struct kprobe *alloc_aggr_kprobe(struct kprobe *p) |
afd66255 MH |
959 | { |
960 | return kzalloc(sizeof(struct kprobe), GFP_KERNEL); | |
961 | } | |
962 | #endif /* CONFIG_OPTPROBES */ | |
963 | ||
e7dbfe34 | 964 | #ifdef CONFIG_KPROBES_ON_FTRACE |
ae6aa16f | 965 | static struct ftrace_ops kprobe_ftrace_ops __read_mostly = { |
e5253896 | 966 | .func = kprobe_ftrace_handler, |
1d70be34 | 967 | .flags = FTRACE_OPS_FL_SAVE_REGS | FTRACE_OPS_FL_IPMODIFY, |
ae6aa16f MH |
968 | }; |
969 | static int kprobe_ftrace_enabled; | |
970 | ||
971 | /* Must ensure p->addr is really on ftrace */ | |
55479f64 | 972 | static int prepare_kprobe(struct kprobe *p) |
ae6aa16f MH |
973 | { |
974 | if (!kprobe_ftrace(p)) | |
975 | return arch_prepare_kprobe(p); | |
976 | ||
977 | return arch_prepare_kprobe_ftrace(p); | |
978 | } | |
979 | ||
980 | /* Caller must lock kprobe_mutex */ | |
12310e34 | 981 | static int arm_kprobe_ftrace(struct kprobe *p) |
ae6aa16f | 982 | { |
12310e34 | 983 | int ret = 0; |
ae6aa16f MH |
984 | |
985 | ret = ftrace_set_filter_ip(&kprobe_ftrace_ops, | |
986 | (unsigned long)p->addr, 0, 0); | |
12310e34 JY |
987 | if (ret) { |
988 | pr_debug("Failed to arm kprobe-ftrace at %p (%d)\n", p->addr, ret); | |
989 | return ret; | |
990 | } | |
991 | ||
992 | if (kprobe_ftrace_enabled == 0) { | |
ae6aa16f | 993 | ret = register_ftrace_function(&kprobe_ftrace_ops); |
12310e34 JY |
994 | if (ret) { |
995 | pr_debug("Failed to init kprobe-ftrace (%d)\n", ret); | |
996 | goto err_ftrace; | |
997 | } | |
ae6aa16f | 998 | } |
12310e34 JY |
999 | |
1000 | kprobe_ftrace_enabled++; | |
1001 | return ret; | |
1002 | ||
1003 | err_ftrace: | |
1004 | /* | |
1005 | * Note: Since kprobe_ftrace_ops has IPMODIFY set, and ftrace requires a | |
1006 | * non-empty filter_hash for IPMODIFY ops, we're safe from an accidental | |
1007 | * empty filter_hash which would undesirably trace all functions. | |
1008 | */ | |
1009 | ftrace_set_filter_ip(&kprobe_ftrace_ops, (unsigned long)p->addr, 1, 0); | |
1010 | return ret; | |
ae6aa16f MH |
1011 | } |
1012 | ||
1013 | /* Caller must lock kprobe_mutex */ | |
297f9233 | 1014 | static int disarm_kprobe_ftrace(struct kprobe *p) |
ae6aa16f | 1015 | { |
297f9233 | 1016 | int ret = 0; |
ae6aa16f | 1017 | |
297f9233 | 1018 | if (kprobe_ftrace_enabled == 1) { |
ae6aa16f | 1019 | ret = unregister_ftrace_function(&kprobe_ftrace_ops); |
297f9233 JY |
1020 | if (WARN(ret < 0, "Failed to unregister kprobe-ftrace (%d)\n", ret)) |
1021 | return ret; | |
ae6aa16f | 1022 | } |
297f9233 JY |
1023 | |
1024 | kprobe_ftrace_enabled--; | |
1025 | ||
ae6aa16f MH |
1026 | ret = ftrace_set_filter_ip(&kprobe_ftrace_ops, |
1027 | (unsigned long)p->addr, 1, 0); | |
1028 | WARN(ret < 0, "Failed to disarm kprobe-ftrace at %p (%d)\n", p->addr, ret); | |
297f9233 | 1029 | return ret; |
ae6aa16f | 1030 | } |
e7dbfe34 | 1031 | #else /* !CONFIG_KPROBES_ON_FTRACE */ |
ae6aa16f | 1032 | #define prepare_kprobe(p) arch_prepare_kprobe(p) |
12310e34 | 1033 | #define arm_kprobe_ftrace(p) (-ENODEV) |
297f9233 | 1034 | #define disarm_kprobe_ftrace(p) (-ENODEV) |
ae6aa16f MH |
1035 | #endif |
1036 | ||
201517a7 | 1037 | /* Arm a kprobe with text_mutex */ |
12310e34 | 1038 | static int arm_kprobe(struct kprobe *kp) |
201517a7 | 1039 | { |
12310e34 JY |
1040 | if (unlikely(kprobe_ftrace(kp))) |
1041 | return arm_kprobe_ftrace(kp); | |
1042 | ||
2d1e38f5 | 1043 | cpus_read_lock(); |
201517a7 | 1044 | mutex_lock(&text_mutex); |
afd66255 | 1045 | __arm_kprobe(kp); |
201517a7 | 1046 | mutex_unlock(&text_mutex); |
2d1e38f5 | 1047 | cpus_read_unlock(); |
12310e34 JY |
1048 | |
1049 | return 0; | |
201517a7 MH |
1050 | } |
1051 | ||
1052 | /* Disarm a kprobe with text_mutex */ | |
297f9233 | 1053 | static int disarm_kprobe(struct kprobe *kp, bool reopt) |
201517a7 | 1054 | { |
297f9233 JY |
1055 | if (unlikely(kprobe_ftrace(kp))) |
1056 | return disarm_kprobe_ftrace(kp); | |
2d1e38f5 TG |
1057 | |
1058 | cpus_read_lock(); | |
201517a7 | 1059 | mutex_lock(&text_mutex); |
ae6aa16f | 1060 | __disarm_kprobe(kp, reopt); |
201517a7 | 1061 | mutex_unlock(&text_mutex); |
2d1e38f5 | 1062 | cpus_read_unlock(); |
297f9233 JY |
1063 | |
1064 | return 0; | |
201517a7 MH |
1065 | } |
1066 | ||
64f562c6 AM |
1067 | /* |
1068 | * Aggregate handlers for multiple kprobes support - these handlers | |
1069 | * take care of invoking the individual kprobe handlers on p->list | |
1070 | */ | |
820aede0 | 1071 | static int aggr_pre_handler(struct kprobe *p, struct pt_regs *regs) |
64f562c6 AM |
1072 | { |
1073 | struct kprobe *kp; | |
1074 | ||
3516a460 | 1075 | list_for_each_entry_rcu(kp, &p->list, list) { |
de5bd88d | 1076 | if (kp->pre_handler && likely(!kprobe_disabled(kp))) { |
e6584523 | 1077 | set_kprobe_instance(kp); |
8b0914ea PP |
1078 | if (kp->pre_handler(kp, regs)) |
1079 | return 1; | |
64f562c6 | 1080 | } |
e6584523 | 1081 | reset_kprobe_instance(); |
64f562c6 AM |
1082 | } |
1083 | return 0; | |
1084 | } | |
820aede0 | 1085 | NOKPROBE_SYMBOL(aggr_pre_handler); |
64f562c6 | 1086 | |
820aede0 MH |
1087 | static void aggr_post_handler(struct kprobe *p, struct pt_regs *regs, |
1088 | unsigned long flags) | |
64f562c6 AM |
1089 | { |
1090 | struct kprobe *kp; | |
1091 | ||
3516a460 | 1092 | list_for_each_entry_rcu(kp, &p->list, list) { |
de5bd88d | 1093 | if (kp->post_handler && likely(!kprobe_disabled(kp))) { |
e6584523 | 1094 | set_kprobe_instance(kp); |
64f562c6 | 1095 | kp->post_handler(kp, regs, flags); |
e6584523 | 1096 | reset_kprobe_instance(); |
64f562c6 AM |
1097 | } |
1098 | } | |
64f562c6 | 1099 | } |
820aede0 | 1100 | NOKPROBE_SYMBOL(aggr_post_handler); |
64f562c6 | 1101 | |
820aede0 MH |
1102 | static int aggr_fault_handler(struct kprobe *p, struct pt_regs *regs, |
1103 | int trapnr) | |
64f562c6 | 1104 | { |
b76834bc | 1105 | struct kprobe *cur = __this_cpu_read(kprobe_instance); |
e6584523 | 1106 | |
64f562c6 AM |
1107 | /* |
1108 | * if we faulted "during" the execution of a user specified | |
1109 | * probe handler, invoke just that probe's fault handler | |
1110 | */ | |
e6584523 AM |
1111 | if (cur && cur->fault_handler) { |
1112 | if (cur->fault_handler(cur, regs, trapnr)) | |
64f562c6 AM |
1113 | return 1; |
1114 | } | |
1115 | return 0; | |
1116 | } | |
820aede0 | 1117 | NOKPROBE_SYMBOL(aggr_fault_handler); |
64f562c6 | 1118 | |
820aede0 | 1119 | static int aggr_break_handler(struct kprobe *p, struct pt_regs *regs) |
8b0914ea | 1120 | { |
b76834bc | 1121 | struct kprobe *cur = __this_cpu_read(kprobe_instance); |
e6584523 AM |
1122 | int ret = 0; |
1123 | ||
1124 | if (cur && cur->break_handler) { | |
1125 | if (cur->break_handler(cur, regs)) | |
1126 | ret = 1; | |
8b0914ea | 1127 | } |
e6584523 AM |
1128 | reset_kprobe_instance(); |
1129 | return ret; | |
8b0914ea | 1130 | } |
820aede0 | 1131 | NOKPROBE_SYMBOL(aggr_break_handler); |
8b0914ea | 1132 | |
bf8d5c52 | 1133 | /* Walks the list and increments nmissed count for multiprobe case */ |
820aede0 | 1134 | void kprobes_inc_nmissed_count(struct kprobe *p) |
bf8d5c52 KA |
1135 | { |
1136 | struct kprobe *kp; | |
afd66255 | 1137 | if (!kprobe_aggrprobe(p)) { |
bf8d5c52 KA |
1138 | p->nmissed++; |
1139 | } else { | |
1140 | list_for_each_entry_rcu(kp, &p->list, list) | |
1141 | kp->nmissed++; | |
1142 | } | |
1143 | return; | |
1144 | } | |
820aede0 | 1145 | NOKPROBE_SYMBOL(kprobes_inc_nmissed_count); |
bf8d5c52 | 1146 | |
820aede0 MH |
1147 | void recycle_rp_inst(struct kretprobe_instance *ri, |
1148 | struct hlist_head *head) | |
b94cce92 | 1149 | { |
ef53d9c5 S |
1150 | struct kretprobe *rp = ri->rp; |
1151 | ||
b94cce92 HN |
1152 | /* remove rp inst off the rprobe_inst_table */ |
1153 | hlist_del(&ri->hlist); | |
ef53d9c5 S |
1154 | INIT_HLIST_NODE(&ri->hlist); |
1155 | if (likely(rp)) { | |
ec484608 | 1156 | raw_spin_lock(&rp->lock); |
ef53d9c5 | 1157 | hlist_add_head(&ri->hlist, &rp->free_instances); |
ec484608 | 1158 | raw_spin_unlock(&rp->lock); |
b94cce92 HN |
1159 | } else |
1160 | /* Unregistering */ | |
99219a3f | 1161 | hlist_add_head(&ri->hlist, head); |
b94cce92 | 1162 | } |
820aede0 | 1163 | NOKPROBE_SYMBOL(recycle_rp_inst); |
b94cce92 | 1164 | |
820aede0 | 1165 | void kretprobe_hash_lock(struct task_struct *tsk, |
ef53d9c5 | 1166 | struct hlist_head **head, unsigned long *flags) |
635c17c2 | 1167 | __acquires(hlist_lock) |
ef53d9c5 S |
1168 | { |
1169 | unsigned long hash = hash_ptr(tsk, KPROBE_HASH_BITS); | |
ec484608 | 1170 | raw_spinlock_t *hlist_lock; |
ef53d9c5 S |
1171 | |
1172 | *head = &kretprobe_inst_table[hash]; | |
1173 | hlist_lock = kretprobe_table_lock_ptr(hash); | |
ec484608 | 1174 | raw_spin_lock_irqsave(hlist_lock, *flags); |
ef53d9c5 | 1175 | } |
820aede0 | 1176 | NOKPROBE_SYMBOL(kretprobe_hash_lock); |
ef53d9c5 | 1177 | |
820aede0 MH |
1178 | static void kretprobe_table_lock(unsigned long hash, |
1179 | unsigned long *flags) | |
635c17c2 | 1180 | __acquires(hlist_lock) |
b94cce92 | 1181 | { |
ec484608 TG |
1182 | raw_spinlock_t *hlist_lock = kretprobe_table_lock_ptr(hash); |
1183 | raw_spin_lock_irqsave(hlist_lock, *flags); | |
ef53d9c5 | 1184 | } |
820aede0 | 1185 | NOKPROBE_SYMBOL(kretprobe_table_lock); |
ef53d9c5 | 1186 | |
820aede0 MH |
1187 | void kretprobe_hash_unlock(struct task_struct *tsk, |
1188 | unsigned long *flags) | |
635c17c2 | 1189 | __releases(hlist_lock) |
ef53d9c5 S |
1190 | { |
1191 | unsigned long hash = hash_ptr(tsk, KPROBE_HASH_BITS); | |
ec484608 | 1192 | raw_spinlock_t *hlist_lock; |
ef53d9c5 S |
1193 | |
1194 | hlist_lock = kretprobe_table_lock_ptr(hash); | |
ec484608 | 1195 | raw_spin_unlock_irqrestore(hlist_lock, *flags); |
ef53d9c5 | 1196 | } |
820aede0 | 1197 | NOKPROBE_SYMBOL(kretprobe_hash_unlock); |
ef53d9c5 | 1198 | |
820aede0 MH |
1199 | static void kretprobe_table_unlock(unsigned long hash, |
1200 | unsigned long *flags) | |
635c17c2 | 1201 | __releases(hlist_lock) |
ef53d9c5 | 1202 | { |
ec484608 TG |
1203 | raw_spinlock_t *hlist_lock = kretprobe_table_lock_ptr(hash); |
1204 | raw_spin_unlock_irqrestore(hlist_lock, *flags); | |
b94cce92 | 1205 | } |
820aede0 | 1206 | NOKPROBE_SYMBOL(kretprobe_table_unlock); |
b94cce92 | 1207 | |
b94cce92 | 1208 | /* |
c6fd91f0 | 1209 | * This function is called from finish_task_switch when task tk becomes dead, |
1210 | * so that we can recycle any function-return probe instances associated | |
1211 | * with this task. These left over instances represent probed functions | |
1212 | * that have been called but will never return. | |
b94cce92 | 1213 | */ |
820aede0 | 1214 | void kprobe_flush_task(struct task_struct *tk) |
b94cce92 | 1215 | { |
62c27be0 | 1216 | struct kretprobe_instance *ri; |
99219a3f | 1217 | struct hlist_head *head, empty_rp; |
b67bfe0d | 1218 | struct hlist_node *tmp; |
ef53d9c5 | 1219 | unsigned long hash, flags = 0; |
802eae7c | 1220 | |
ef53d9c5 S |
1221 | if (unlikely(!kprobes_initialized)) |
1222 | /* Early boot. kretprobe_table_locks not yet initialized. */ | |
1223 | return; | |
1224 | ||
d496aab5 | 1225 | INIT_HLIST_HEAD(&empty_rp); |
ef53d9c5 S |
1226 | hash = hash_ptr(tk, KPROBE_HASH_BITS); |
1227 | head = &kretprobe_inst_table[hash]; | |
1228 | kretprobe_table_lock(hash, &flags); | |
b67bfe0d | 1229 | hlist_for_each_entry_safe(ri, tmp, head, hlist) { |
62c27be0 | 1230 | if (ri->task == tk) |
99219a3f | 1231 | recycle_rp_inst(ri, &empty_rp); |
62c27be0 | 1232 | } |
ef53d9c5 | 1233 | kretprobe_table_unlock(hash, &flags); |
b67bfe0d | 1234 | hlist_for_each_entry_safe(ri, tmp, &empty_rp, hlist) { |
99219a3f | 1235 | hlist_del(&ri->hlist); |
1236 | kfree(ri); | |
1237 | } | |
b94cce92 | 1238 | } |
820aede0 | 1239 | NOKPROBE_SYMBOL(kprobe_flush_task); |
b94cce92 | 1240 | |
b94cce92 HN |
1241 | static inline void free_rp_inst(struct kretprobe *rp) |
1242 | { | |
1243 | struct kretprobe_instance *ri; | |
b67bfe0d | 1244 | struct hlist_node *next; |
4c4308cb | 1245 | |
b67bfe0d | 1246 | hlist_for_each_entry_safe(ri, next, &rp->free_instances, hlist) { |
ef53d9c5 | 1247 | hlist_del(&ri->hlist); |
b94cce92 HN |
1248 | kfree(ri); |
1249 | } | |
1250 | } | |
1251 | ||
820aede0 | 1252 | static void cleanup_rp_inst(struct kretprobe *rp) |
4a296e07 | 1253 | { |
ef53d9c5 | 1254 | unsigned long flags, hash; |
4a296e07 | 1255 | struct kretprobe_instance *ri; |
b67bfe0d | 1256 | struct hlist_node *next; |
ef53d9c5 S |
1257 | struct hlist_head *head; |
1258 | ||
4a296e07 | 1259 | /* No race here */ |
ef53d9c5 S |
1260 | for (hash = 0; hash < KPROBE_TABLE_SIZE; hash++) { |
1261 | kretprobe_table_lock(hash, &flags); | |
1262 | head = &kretprobe_inst_table[hash]; | |
b67bfe0d | 1263 | hlist_for_each_entry_safe(ri, next, head, hlist) { |
ef53d9c5 S |
1264 | if (ri->rp == rp) |
1265 | ri->rp = NULL; | |
1266 | } | |
1267 | kretprobe_table_unlock(hash, &flags); | |
4a296e07 | 1268 | } |
4a296e07 MH |
1269 | free_rp_inst(rp); |
1270 | } | |
820aede0 | 1271 | NOKPROBE_SYMBOL(cleanup_rp_inst); |
4a296e07 | 1272 | |
8b0914ea | 1273 | /* |
b918e5e6 | 1274 | * Add the new probe to ap->list. Fail if this is the |
5a6cf77f | 1275 | * second break_handler at the address |
8b0914ea | 1276 | */ |
55479f64 | 1277 | static int add_new_kprobe(struct kprobe *ap, struct kprobe *p) |
8b0914ea | 1278 | { |
de5bd88d | 1279 | BUG_ON(kprobe_gone(ap) || kprobe_gone(p)); |
afd66255 MH |
1280 | |
1281 | if (p->break_handler || p->post_handler) | |
6274de49 | 1282 | unoptimize_kprobe(ap, true); /* Fall back to normal kprobe */ |
afd66255 | 1283 | |
8b0914ea | 1284 | if (p->break_handler) { |
b918e5e6 | 1285 | if (ap->break_handler) |
36721656 | 1286 | return -EEXIST; |
b918e5e6 MH |
1287 | list_add_tail_rcu(&p->list, &ap->list); |
1288 | ap->break_handler = aggr_break_handler; | |
8b0914ea | 1289 | } else |
b918e5e6 MH |
1290 | list_add_rcu(&p->list, &ap->list); |
1291 | if (p->post_handler && !ap->post_handler) | |
1292 | ap->post_handler = aggr_post_handler; | |
de5bd88d | 1293 | |
8b0914ea PP |
1294 | return 0; |
1295 | } | |
1296 | ||
64f562c6 AM |
1297 | /* |
1298 | * Fill in the required fields of the "manager kprobe". Replace the | |
1299 | * earlier kprobe in the hlist with the manager kprobe | |
1300 | */ | |
55479f64 | 1301 | static void init_aggr_kprobe(struct kprobe *ap, struct kprobe *p) |
64f562c6 | 1302 | { |
afd66255 | 1303 | /* Copy p's insn slot to ap */ |
8b0914ea | 1304 | copy_kprobe(p, ap); |
a9ad965e | 1305 | flush_insn_slot(ap); |
64f562c6 | 1306 | ap->addr = p->addr; |
afd66255 | 1307 | ap->flags = p->flags & ~KPROBE_FLAG_OPTIMIZED; |
64f562c6 | 1308 | ap->pre_handler = aggr_pre_handler; |
64f562c6 | 1309 | ap->fault_handler = aggr_fault_handler; |
e8386a0c MH |
1310 | /* We don't care the kprobe which has gone. */ |
1311 | if (p->post_handler && !kprobe_gone(p)) | |
36721656 | 1312 | ap->post_handler = aggr_post_handler; |
e8386a0c | 1313 | if (p->break_handler && !kprobe_gone(p)) |
36721656 | 1314 | ap->break_handler = aggr_break_handler; |
64f562c6 AM |
1315 | |
1316 | INIT_LIST_HEAD(&ap->list); | |
afd66255 | 1317 | INIT_HLIST_NODE(&ap->hlist); |
64f562c6 | 1318 | |
afd66255 | 1319 | list_add_rcu(&p->list, &ap->list); |
adad0f33 | 1320 | hlist_replace_rcu(&p->hlist, &ap->hlist); |
64f562c6 AM |
1321 | } |
1322 | ||
1323 | /* | |
1324 | * This is the second or subsequent kprobe at the address - handle | |
1325 | * the intricacies | |
64f562c6 | 1326 | */ |
55479f64 | 1327 | static int register_aggr_kprobe(struct kprobe *orig_p, struct kprobe *p) |
64f562c6 AM |
1328 | { |
1329 | int ret = 0; | |
6d8e40a8 | 1330 | struct kprobe *ap = orig_p; |
64f562c6 | 1331 | |
2d1e38f5 TG |
1332 | cpus_read_lock(); |
1333 | ||
25764288 MH |
1334 | /* For preparing optimization, jump_label_text_reserved() is called */ |
1335 | jump_label_lock(); | |
25764288 MH |
1336 | mutex_lock(&text_mutex); |
1337 | ||
6d8e40a8 MH |
1338 | if (!kprobe_aggrprobe(orig_p)) { |
1339 | /* If orig_p is not an aggr_kprobe, create new aggr_kprobe. */ | |
1340 | ap = alloc_aggr_kprobe(orig_p); | |
25764288 MH |
1341 | if (!ap) { |
1342 | ret = -ENOMEM; | |
1343 | goto out; | |
1344 | } | |
6d8e40a8 | 1345 | init_aggr_kprobe(ap, orig_p); |
6274de49 | 1346 | } else if (kprobe_unused(ap)) |
0490cd1f MH |
1347 | /* This probe is going to die. Rescue it */ |
1348 | reuse_unused_kprobe(ap); | |
b918e5e6 MH |
1349 | |
1350 | if (kprobe_gone(ap)) { | |
e8386a0c MH |
1351 | /* |
1352 | * Attempting to insert new probe at the same location that | |
1353 | * had a probe in the module vaddr area which already | |
1354 | * freed. So, the instruction slot has already been | |
1355 | * released. We need a new slot for the new probe. | |
1356 | */ | |
b918e5e6 | 1357 | ret = arch_prepare_kprobe(ap); |
e8386a0c | 1358 | if (ret) |
b918e5e6 MH |
1359 | /* |
1360 | * Even if fail to allocate new slot, don't need to | |
1361 | * free aggr_probe. It will be used next time, or | |
1362 | * freed by unregister_kprobe. | |
1363 | */ | |
25764288 | 1364 | goto out; |
de5bd88d | 1365 | |
afd66255 MH |
1366 | /* Prepare optimized instructions if possible. */ |
1367 | prepare_optimized_kprobe(ap); | |
1368 | ||
e8386a0c | 1369 | /* |
de5bd88d MH |
1370 | * Clear gone flag to prevent allocating new slot again, and |
1371 | * set disabled flag because it is not armed yet. | |
e8386a0c | 1372 | */ |
de5bd88d MH |
1373 | ap->flags = (ap->flags & ~KPROBE_FLAG_GONE) |
1374 | | KPROBE_FLAG_DISABLED; | |
e8386a0c | 1375 | } |
b918e5e6 | 1376 | |
afd66255 | 1377 | /* Copy ap's insn slot to p */ |
b918e5e6 | 1378 | copy_kprobe(ap, p); |
25764288 MH |
1379 | ret = add_new_kprobe(ap, p); |
1380 | ||
1381 | out: | |
1382 | mutex_unlock(&text_mutex); | |
25764288 | 1383 | jump_label_unlock(); |
2d1e38f5 | 1384 | cpus_read_unlock(); |
25764288 MH |
1385 | |
1386 | if (ret == 0 && kprobe_disabled(ap) && !kprobe_disabled(p)) { | |
1387 | ap->flags &= ~KPROBE_FLAG_DISABLED; | |
12310e34 | 1388 | if (!kprobes_all_disarmed) { |
25764288 | 1389 | /* Arm the breakpoint again. */ |
12310e34 JY |
1390 | ret = arm_kprobe(ap); |
1391 | if (ret) { | |
1392 | ap->flags |= KPROBE_FLAG_DISABLED; | |
1393 | list_del_rcu(&p->list); | |
1394 | synchronize_sched(); | |
1395 | } | |
1396 | } | |
25764288 MH |
1397 | } |
1398 | return ret; | |
64f562c6 AM |
1399 | } |
1400 | ||
be8f2743 MH |
1401 | bool __weak arch_within_kprobe_blacklist(unsigned long addr) |
1402 | { | |
1403 | /* The __kprobes marked functions and entry code must not be probed */ | |
1404 | return addr >= (unsigned long)__kprobes_text_start && | |
1405 | addr < (unsigned long)__kprobes_text_end; | |
1406 | } | |
1407 | ||
e5779e8e | 1408 | bool within_kprobe_blacklist(unsigned long addr) |
d0aaff97 | 1409 | { |
376e2424 | 1410 | struct kprobe_blacklist_entry *ent; |
3d8d996e | 1411 | |
be8f2743 | 1412 | if (arch_within_kprobe_blacklist(addr)) |
376e2424 | 1413 | return true; |
3d8d996e SD |
1414 | /* |
1415 | * If there exists a kprobe_blacklist, verify and | |
1416 | * fail any probe registration in the prohibited area | |
1417 | */ | |
376e2424 MH |
1418 | list_for_each_entry(ent, &kprobe_blacklist, list) { |
1419 | if (addr >= ent->start_addr && addr < ent->end_addr) | |
1420 | return true; | |
3d8d996e | 1421 | } |
376e2424 MH |
1422 | |
1423 | return false; | |
d0aaff97 PP |
1424 | } |
1425 | ||
b2a5cd69 MH |
1426 | /* |
1427 | * If we have a symbol_name argument, look it up and add the offset field | |
1428 | * to it. This way, we can specify a relative address to a symbol. | |
bc81d48d MH |
1429 | * This returns encoded errors if it fails to look up symbol or invalid |
1430 | * combination of parameters. | |
b2a5cd69 | 1431 | */ |
1d585e70 NR |
1432 | static kprobe_opcode_t *_kprobe_addr(kprobe_opcode_t *addr, |
1433 | const char *symbol_name, unsigned int offset) | |
b2a5cd69 | 1434 | { |
1d585e70 | 1435 | if ((symbol_name && addr) || (!symbol_name && !addr)) |
bc81d48d MH |
1436 | goto invalid; |
1437 | ||
1d585e70 | 1438 | if (symbol_name) { |
7246f600 | 1439 | addr = kprobe_lookup_name(symbol_name, offset); |
bc81d48d MH |
1440 | if (!addr) |
1441 | return ERR_PTR(-ENOENT); | |
b2a5cd69 MH |
1442 | } |
1443 | ||
1d585e70 | 1444 | addr = (kprobe_opcode_t *)(((char *)addr) + offset); |
bc81d48d MH |
1445 | if (addr) |
1446 | return addr; | |
1447 | ||
1448 | invalid: | |
1449 | return ERR_PTR(-EINVAL); | |
b2a5cd69 MH |
1450 | } |
1451 | ||
1d585e70 NR |
1452 | static kprobe_opcode_t *kprobe_addr(struct kprobe *p) |
1453 | { | |
1454 | return _kprobe_addr(p->addr, p->symbol_name, p->offset); | |
1455 | } | |
1456 | ||
1f0ab409 | 1457 | /* Check passed kprobe is valid and return kprobe in kprobe_table. */ |
55479f64 | 1458 | static struct kprobe *__get_valid_kprobe(struct kprobe *p) |
1f0ab409 | 1459 | { |
6d8e40a8 | 1460 | struct kprobe *ap, *list_p; |
1f0ab409 | 1461 | |
6d8e40a8 MH |
1462 | ap = get_kprobe(p->addr); |
1463 | if (unlikely(!ap)) | |
1f0ab409 AM |
1464 | return NULL; |
1465 | ||
6d8e40a8 MH |
1466 | if (p != ap) { |
1467 | list_for_each_entry_rcu(list_p, &ap->list, list) | |
1f0ab409 AM |
1468 | if (list_p == p) |
1469 | /* kprobe p is a valid probe */ | |
1470 | goto valid; | |
1471 | return NULL; | |
1472 | } | |
1473 | valid: | |
6d8e40a8 | 1474 | return ap; |
1f0ab409 AM |
1475 | } |
1476 | ||
1477 | /* Return error if the kprobe is being re-registered */ | |
1478 | static inline int check_kprobe_rereg(struct kprobe *p) | |
1479 | { | |
1480 | int ret = 0; | |
1f0ab409 AM |
1481 | |
1482 | mutex_lock(&kprobe_mutex); | |
6d8e40a8 | 1483 | if (__get_valid_kprobe(p)) |
1f0ab409 AM |
1484 | ret = -EINVAL; |
1485 | mutex_unlock(&kprobe_mutex); | |
6d8e40a8 | 1486 | |
1f0ab409 AM |
1487 | return ret; |
1488 | } | |
1489 | ||
f7f242ff | 1490 | int __weak arch_check_ftrace_location(struct kprobe *p) |
1da177e4 | 1491 | { |
ae6aa16f MH |
1492 | unsigned long ftrace_addr; |
1493 | ||
ae6aa16f MH |
1494 | ftrace_addr = ftrace_location((unsigned long)p->addr); |
1495 | if (ftrace_addr) { | |
e7dbfe34 | 1496 | #ifdef CONFIG_KPROBES_ON_FTRACE |
ae6aa16f MH |
1497 | /* Given address is not on the instruction boundary */ |
1498 | if ((unsigned long)p->addr != ftrace_addr) | |
1499 | return -EILSEQ; | |
ae6aa16f | 1500 | p->flags |= KPROBE_FLAG_FTRACE; |
e7dbfe34 | 1501 | #else /* !CONFIG_KPROBES_ON_FTRACE */ |
ae6aa16f MH |
1502 | return -EINVAL; |
1503 | #endif | |
1504 | } | |
f7f242ff HC |
1505 | return 0; |
1506 | } | |
1507 | ||
1508 | static int check_kprobe_address_safe(struct kprobe *p, | |
1509 | struct module **probed_mod) | |
1510 | { | |
1511 | int ret; | |
1f0ab409 | 1512 | |
f7f242ff HC |
1513 | ret = arch_check_ftrace_location(p); |
1514 | if (ret) | |
1515 | return ret; | |
91bad2f8 | 1516 | jump_label_lock(); |
de31c3ca | 1517 | preempt_disable(); |
f7fa6ef0 MH |
1518 | |
1519 | /* Ensure it is not in reserved area nor out of text */ | |
ec30c5f3 | 1520 | if (!kernel_text_address((unsigned long) p->addr) || |
376e2424 | 1521 | within_kprobe_blacklist((unsigned long) p->addr) || |
f986a499 PN |
1522 | jump_label_text_reserved(p->addr, p->addr)) { |
1523 | ret = -EINVAL; | |
f7fa6ef0 | 1524 | goto out; |
f986a499 | 1525 | } |
b3e55c72 | 1526 | |
f7fa6ef0 MH |
1527 | /* Check if are we probing a module */ |
1528 | *probed_mod = __module_text_address((unsigned long) p->addr); | |
1529 | if (*probed_mod) { | |
6f716acd | 1530 | /* |
e8386a0c MH |
1531 | * We must hold a refcount of the probed module while updating |
1532 | * its code to prohibit unexpected unloading. | |
df019b1d | 1533 | */ |
f7fa6ef0 MH |
1534 | if (unlikely(!try_module_get(*probed_mod))) { |
1535 | ret = -ENOENT; | |
1536 | goto out; | |
1537 | } | |
de31c3ca | 1538 | |
f24659d9 MH |
1539 | /* |
1540 | * If the module freed .init.text, we couldn't insert | |
1541 | * kprobes in there. | |
1542 | */ | |
f7fa6ef0 MH |
1543 | if (within_module_init((unsigned long)p->addr, *probed_mod) && |
1544 | (*probed_mod)->state != MODULE_STATE_COMING) { | |
1545 | module_put(*probed_mod); | |
1546 | *probed_mod = NULL; | |
1547 | ret = -ENOENT; | |
f24659d9 | 1548 | } |
df019b1d | 1549 | } |
f7fa6ef0 | 1550 | out: |
a189d035 | 1551 | preempt_enable(); |
de31c3ca | 1552 | jump_label_unlock(); |
1da177e4 | 1553 | |
f7fa6ef0 MH |
1554 | return ret; |
1555 | } | |
1556 | ||
55479f64 | 1557 | int register_kprobe(struct kprobe *p) |
f7fa6ef0 MH |
1558 | { |
1559 | int ret; | |
1560 | struct kprobe *old_p; | |
1561 | struct module *probed_mod; | |
1562 | kprobe_opcode_t *addr; | |
1563 | ||
1564 | /* Adjust probe address from symbol */ | |
1565 | addr = kprobe_addr(p); | |
1566 | if (IS_ERR(addr)) | |
1567 | return PTR_ERR(addr); | |
1568 | p->addr = addr; | |
1569 | ||
1570 | ret = check_kprobe_rereg(p); | |
1571 | if (ret) | |
1572 | return ret; | |
1573 | ||
1574 | /* User can pass only KPROBE_FLAG_DISABLED to register_kprobe */ | |
1575 | p->flags &= KPROBE_FLAG_DISABLED; | |
3516a460 | 1576 | p->nmissed = 0; |
9861668f | 1577 | INIT_LIST_HEAD(&p->list); |
afd66255 | 1578 | |
f7fa6ef0 MH |
1579 | ret = check_kprobe_address_safe(p, &probed_mod); |
1580 | if (ret) | |
1581 | return ret; | |
1582 | ||
1583 | mutex_lock(&kprobe_mutex); | |
afd66255 | 1584 | |
64f562c6 AM |
1585 | old_p = get_kprobe(p->addr); |
1586 | if (old_p) { | |
afd66255 | 1587 | /* Since this may unoptimize old_p, locking text_mutex. */ |
64f562c6 | 1588 | ret = register_aggr_kprobe(old_p, p); |
1da177e4 LT |
1589 | goto out; |
1590 | } | |
1da177e4 | 1591 | |
2d1e38f5 TG |
1592 | cpus_read_lock(); |
1593 | /* Prevent text modification */ | |
1594 | mutex_lock(&text_mutex); | |
ae6aa16f | 1595 | ret = prepare_kprobe(p); |
25764288 | 1596 | mutex_unlock(&text_mutex); |
2d1e38f5 | 1597 | cpus_read_unlock(); |
6f716acd | 1598 | if (ret) |
afd66255 | 1599 | goto out; |
49a2a1b8 | 1600 | |
64f562c6 | 1601 | INIT_HLIST_NODE(&p->hlist); |
3516a460 | 1602 | hlist_add_head_rcu(&p->hlist, |
1da177e4 LT |
1603 | &kprobe_table[hash_ptr(p->addr, KPROBE_HASH_BITS)]); |
1604 | ||
12310e34 JY |
1605 | if (!kprobes_all_disarmed && !kprobe_disabled(p)) { |
1606 | ret = arm_kprobe(p); | |
1607 | if (ret) { | |
1608 | hlist_del_rcu(&p->hlist); | |
1609 | synchronize_sched(); | |
1610 | goto out; | |
1611 | } | |
1612 | } | |
afd66255 MH |
1613 | |
1614 | /* Try to optimize kprobe */ | |
1615 | try_to_optimize_kprobe(p); | |
1da177e4 | 1616 | out: |
7a7d1cf9 | 1617 | mutex_unlock(&kprobe_mutex); |
49a2a1b8 | 1618 | |
e8386a0c | 1619 | if (probed_mod) |
df019b1d | 1620 | module_put(probed_mod); |
e8386a0c | 1621 | |
1da177e4 LT |
1622 | return ret; |
1623 | } | |
99081ab5 | 1624 | EXPORT_SYMBOL_GPL(register_kprobe); |
1da177e4 | 1625 | |
6f0f1dd7 | 1626 | /* Check if all probes on the aggrprobe are disabled */ |
55479f64 | 1627 | static int aggr_kprobe_disabled(struct kprobe *ap) |
6f0f1dd7 MH |
1628 | { |
1629 | struct kprobe *kp; | |
1630 | ||
1631 | list_for_each_entry_rcu(kp, &ap->list, list) | |
1632 | if (!kprobe_disabled(kp)) | |
1633 | /* | |
1634 | * There is an active probe on the list. | |
1635 | * We can't disable this ap. | |
1636 | */ | |
1637 | return 0; | |
1638 | ||
1639 | return 1; | |
1640 | } | |
1641 | ||
1642 | /* Disable one kprobe: Make sure called under kprobe_mutex is locked */ | |
55479f64 | 1643 | static struct kprobe *__disable_kprobe(struct kprobe *p) |
6f0f1dd7 MH |
1644 | { |
1645 | struct kprobe *orig_p; | |
297f9233 | 1646 | int ret; |
6f0f1dd7 MH |
1647 | |
1648 | /* Get an original kprobe for return */ | |
1649 | orig_p = __get_valid_kprobe(p); | |
1650 | if (unlikely(orig_p == NULL)) | |
297f9233 | 1651 | return ERR_PTR(-EINVAL); |
6f0f1dd7 MH |
1652 | |
1653 | if (!kprobe_disabled(p)) { | |
1654 | /* Disable probe if it is a child probe */ | |
1655 | if (p != orig_p) | |
1656 | p->flags |= KPROBE_FLAG_DISABLED; | |
1657 | ||
1658 | /* Try to disarm and disable this/parent probe */ | |
1659 | if (p == orig_p || aggr_kprobe_disabled(orig_p)) { | |
69d54b91 WN |
1660 | /* |
1661 | * If kprobes_all_disarmed is set, orig_p | |
1662 | * should have already been disarmed, so | |
1663 | * skip unneed disarming process. | |
1664 | */ | |
297f9233 JY |
1665 | if (!kprobes_all_disarmed) { |
1666 | ret = disarm_kprobe(orig_p, true); | |
1667 | if (ret) { | |
1668 | p->flags &= ~KPROBE_FLAG_DISABLED; | |
1669 | return ERR_PTR(ret); | |
1670 | } | |
1671 | } | |
6f0f1dd7 MH |
1672 | orig_p->flags |= KPROBE_FLAG_DISABLED; |
1673 | } | |
1674 | } | |
1675 | ||
1676 | return orig_p; | |
1677 | } | |
1678 | ||
de5bd88d MH |
1679 | /* |
1680 | * Unregister a kprobe without a scheduler synchronization. | |
1681 | */ | |
55479f64 | 1682 | static int __unregister_kprobe_top(struct kprobe *p) |
de5bd88d | 1683 | { |
6d8e40a8 | 1684 | struct kprobe *ap, *list_p; |
de5bd88d | 1685 | |
6f0f1dd7 MH |
1686 | /* Disable kprobe. This will disarm it if needed. */ |
1687 | ap = __disable_kprobe(p); | |
297f9233 JY |
1688 | if (IS_ERR(ap)) |
1689 | return PTR_ERR(ap); | |
de5bd88d | 1690 | |
6f0f1dd7 | 1691 | if (ap == p) |
bf8f6e5b | 1692 | /* |
6f0f1dd7 MH |
1693 | * This probe is an independent(and non-optimized) kprobe |
1694 | * (not an aggrprobe). Remove from the hash list. | |
bf8f6e5b | 1695 | */ |
6f0f1dd7 MH |
1696 | goto disarmed; |
1697 | ||
1698 | /* Following process expects this probe is an aggrprobe */ | |
1699 | WARN_ON(!kprobe_aggrprobe(ap)); | |
1700 | ||
6274de49 MH |
1701 | if (list_is_singular(&ap->list) && kprobe_disarmed(ap)) |
1702 | /* | |
1703 | * !disarmed could be happen if the probe is under delayed | |
1704 | * unoptimizing. | |
1705 | */ | |
6f0f1dd7 MH |
1706 | goto disarmed; |
1707 | else { | |
1708 | /* If disabling probe has special handlers, update aggrprobe */ | |
e8386a0c | 1709 | if (p->break_handler && !kprobe_gone(p)) |
6d8e40a8 | 1710 | ap->break_handler = NULL; |
e8386a0c | 1711 | if (p->post_handler && !kprobe_gone(p)) { |
6d8e40a8 | 1712 | list_for_each_entry_rcu(list_p, &ap->list, list) { |
9861668f MH |
1713 | if ((list_p != p) && (list_p->post_handler)) |
1714 | goto noclean; | |
1715 | } | |
6d8e40a8 | 1716 | ap->post_handler = NULL; |
9861668f MH |
1717 | } |
1718 | noclean: | |
6f0f1dd7 MH |
1719 | /* |
1720 | * Remove from the aggrprobe: this path will do nothing in | |
1721 | * __unregister_kprobe_bottom(). | |
1722 | */ | |
49a2a1b8 | 1723 | list_del_rcu(&p->list); |
6f0f1dd7 MH |
1724 | if (!kprobe_disabled(ap) && !kprobes_all_disarmed) |
1725 | /* | |
1726 | * Try to optimize this probe again, because post | |
1727 | * handler may have been changed. | |
1728 | */ | |
1729 | optimize_kprobe(ap); | |
49a2a1b8 | 1730 | } |
9861668f | 1731 | return 0; |
6f0f1dd7 MH |
1732 | |
1733 | disarmed: | |
6274de49 | 1734 | BUG_ON(!kprobe_disarmed(ap)); |
6f0f1dd7 MH |
1735 | hlist_del_rcu(&ap->hlist); |
1736 | return 0; | |
9861668f | 1737 | } |
3516a460 | 1738 | |
55479f64 | 1739 | static void __unregister_kprobe_bottom(struct kprobe *p) |
9861668f | 1740 | { |
6d8e40a8 | 1741 | struct kprobe *ap; |
b3e55c72 | 1742 | |
e8386a0c | 1743 | if (list_empty(&p->list)) |
6274de49 | 1744 | /* This is an independent kprobe */ |
0498b635 | 1745 | arch_remove_kprobe(p); |
e8386a0c | 1746 | else if (list_is_singular(&p->list)) { |
6274de49 | 1747 | /* This is the last child of an aggrprobe */ |
6d8e40a8 | 1748 | ap = list_entry(p->list.next, struct kprobe, list); |
e8386a0c | 1749 | list_del(&p->list); |
6d8e40a8 | 1750 | free_aggr_kprobe(ap); |
9861668f | 1751 | } |
6274de49 | 1752 | /* Otherwise, do nothing. */ |
9861668f MH |
1753 | } |
1754 | ||
55479f64 | 1755 | int register_kprobes(struct kprobe **kps, int num) |
9861668f MH |
1756 | { |
1757 | int i, ret = 0; | |
1758 | ||
1759 | if (num <= 0) | |
1760 | return -EINVAL; | |
1761 | for (i = 0; i < num; i++) { | |
49ad2fd7 | 1762 | ret = register_kprobe(kps[i]); |
67dddaad MH |
1763 | if (ret < 0) { |
1764 | if (i > 0) | |
1765 | unregister_kprobes(kps, i); | |
9861668f | 1766 | break; |
36721656 | 1767 | } |
49a2a1b8 | 1768 | } |
9861668f MH |
1769 | return ret; |
1770 | } | |
99081ab5 | 1771 | EXPORT_SYMBOL_GPL(register_kprobes); |
9861668f | 1772 | |
55479f64 | 1773 | void unregister_kprobe(struct kprobe *p) |
9861668f MH |
1774 | { |
1775 | unregister_kprobes(&p, 1); | |
1776 | } | |
99081ab5 | 1777 | EXPORT_SYMBOL_GPL(unregister_kprobe); |
9861668f | 1778 | |
55479f64 | 1779 | void unregister_kprobes(struct kprobe **kps, int num) |
9861668f MH |
1780 | { |
1781 | int i; | |
1782 | ||
1783 | if (num <= 0) | |
1784 | return; | |
1785 | mutex_lock(&kprobe_mutex); | |
1786 | for (i = 0; i < num; i++) | |
1787 | if (__unregister_kprobe_top(kps[i]) < 0) | |
1788 | kps[i]->addr = NULL; | |
1789 | mutex_unlock(&kprobe_mutex); | |
1790 | ||
1791 | synchronize_sched(); | |
1792 | for (i = 0; i < num; i++) | |
1793 | if (kps[i]->addr) | |
1794 | __unregister_kprobe_bottom(kps[i]); | |
1da177e4 | 1795 | } |
99081ab5 | 1796 | EXPORT_SYMBOL_GPL(unregister_kprobes); |
1da177e4 | 1797 | |
5f6bee34 NR |
1798 | int __weak kprobe_exceptions_notify(struct notifier_block *self, |
1799 | unsigned long val, void *data) | |
fc62d020 NR |
1800 | { |
1801 | return NOTIFY_DONE; | |
1802 | } | |
5f6bee34 | 1803 | NOKPROBE_SYMBOL(kprobe_exceptions_notify); |
fc62d020 | 1804 | |
1da177e4 | 1805 | static struct notifier_block kprobe_exceptions_nb = { |
3d5631e0 AK |
1806 | .notifier_call = kprobe_exceptions_notify, |
1807 | .priority = 0x7fffffff /* we need to be notified first */ | |
1808 | }; | |
1809 | ||
3d7e3382 ME |
1810 | unsigned long __weak arch_deref_entry_point(void *entry) |
1811 | { | |
1812 | return (unsigned long)entry; | |
1813 | } | |
1da177e4 | 1814 | |
9edddaa2 | 1815 | #ifdef CONFIG_KRETPROBES |
e65cefe8 AB |
1816 | /* |
1817 | * This kprobe pre_handler is registered with every kretprobe. When probe | |
1818 | * hits it will set up the return probe. | |
1819 | */ | |
820aede0 | 1820 | static int pre_handler_kretprobe(struct kprobe *p, struct pt_regs *regs) |
e65cefe8 AB |
1821 | { |
1822 | struct kretprobe *rp = container_of(p, struct kretprobe, kp); | |
ef53d9c5 S |
1823 | unsigned long hash, flags = 0; |
1824 | struct kretprobe_instance *ri; | |
e65cefe8 | 1825 | |
f96f5678 MH |
1826 | /* |
1827 | * To avoid deadlocks, prohibit return probing in NMI contexts, | |
1828 | * just skip the probe and increase the (inexact) 'nmissed' | |
1829 | * statistical counter, so that the user is informed that | |
1830 | * something happened: | |
1831 | */ | |
1832 | if (unlikely(in_nmi())) { | |
1833 | rp->nmissed++; | |
1834 | return 0; | |
1835 | } | |
1836 | ||
1837 | /* TODO: consider to only swap the RA after the last pre_handler fired */ | |
ef53d9c5 | 1838 | hash = hash_ptr(current, KPROBE_HASH_BITS); |
ec484608 | 1839 | raw_spin_lock_irqsave(&rp->lock, flags); |
4c4308cb | 1840 | if (!hlist_empty(&rp->free_instances)) { |
4c4308cb | 1841 | ri = hlist_entry(rp->free_instances.first, |
ef53d9c5 S |
1842 | struct kretprobe_instance, hlist); |
1843 | hlist_del(&ri->hlist); | |
ec484608 | 1844 | raw_spin_unlock_irqrestore(&rp->lock, flags); |
ef53d9c5 | 1845 | |
4c4308cb CH |
1846 | ri->rp = rp; |
1847 | ri->task = current; | |
f47cd9b5 | 1848 | |
55ca6140 JL |
1849 | if (rp->entry_handler && rp->entry_handler(ri, regs)) { |
1850 | raw_spin_lock_irqsave(&rp->lock, flags); | |
1851 | hlist_add_head(&ri->hlist, &rp->free_instances); | |
1852 | raw_spin_unlock_irqrestore(&rp->lock, flags); | |
f47cd9b5 | 1853 | return 0; |
55ca6140 | 1854 | } |
f47cd9b5 | 1855 | |
4c4308cb CH |
1856 | arch_prepare_kretprobe(ri, regs); |
1857 | ||
1858 | /* XXX(hch): why is there no hlist_move_head? */ | |
ef53d9c5 S |
1859 | INIT_HLIST_NODE(&ri->hlist); |
1860 | kretprobe_table_lock(hash, &flags); | |
1861 | hlist_add_head(&ri->hlist, &kretprobe_inst_table[hash]); | |
1862 | kretprobe_table_unlock(hash, &flags); | |
1863 | } else { | |
4c4308cb | 1864 | rp->nmissed++; |
ec484608 | 1865 | raw_spin_unlock_irqrestore(&rp->lock, flags); |
ef53d9c5 | 1866 | } |
e65cefe8 AB |
1867 | return 0; |
1868 | } | |
820aede0 | 1869 | NOKPROBE_SYMBOL(pre_handler_kretprobe); |
e65cefe8 | 1870 | |
659b957f | 1871 | bool __weak arch_kprobe_on_func_entry(unsigned long offset) |
90ec5e89 NR |
1872 | { |
1873 | return !offset; | |
1874 | } | |
1875 | ||
659b957f | 1876 | bool kprobe_on_func_entry(kprobe_opcode_t *addr, const char *sym, unsigned long offset) |
1d585e70 NR |
1877 | { |
1878 | kprobe_opcode_t *kp_addr = _kprobe_addr(addr, sym, offset); | |
1879 | ||
1880 | if (IS_ERR(kp_addr)) | |
1881 | return false; | |
1882 | ||
1883 | if (!kallsyms_lookup_size_offset((unsigned long)kp_addr, NULL, &offset) || | |
659b957f | 1884 | !arch_kprobe_on_func_entry(offset)) |
1d585e70 NR |
1885 | return false; |
1886 | ||
1887 | return true; | |
1888 | } | |
1889 | ||
55479f64 | 1890 | int register_kretprobe(struct kretprobe *rp) |
b94cce92 HN |
1891 | { |
1892 | int ret = 0; | |
1893 | struct kretprobe_instance *inst; | |
1894 | int i; | |
b2a5cd69 | 1895 | void *addr; |
90ec5e89 | 1896 | |
659b957f | 1897 | if (!kprobe_on_func_entry(rp->kp.addr, rp->kp.symbol_name, rp->kp.offset)) |
90ec5e89 | 1898 | return -EINVAL; |
f438d914 MH |
1899 | |
1900 | if (kretprobe_blacklist_size) { | |
b2a5cd69 | 1901 | addr = kprobe_addr(&rp->kp); |
bc81d48d MH |
1902 | if (IS_ERR(addr)) |
1903 | return PTR_ERR(addr); | |
f438d914 MH |
1904 | |
1905 | for (i = 0; kretprobe_blacklist[i].name != NULL; i++) { | |
1906 | if (kretprobe_blacklist[i].addr == addr) | |
1907 | return -EINVAL; | |
1908 | } | |
1909 | } | |
b94cce92 HN |
1910 | |
1911 | rp->kp.pre_handler = pre_handler_kretprobe; | |
7522a842 AM |
1912 | rp->kp.post_handler = NULL; |
1913 | rp->kp.fault_handler = NULL; | |
1914 | rp->kp.break_handler = NULL; | |
b94cce92 HN |
1915 | |
1916 | /* Pre-allocate memory for max kretprobe instances */ | |
1917 | if (rp->maxactive <= 0) { | |
1918 | #ifdef CONFIG_PREEMPT | |
c2ef6661 | 1919 | rp->maxactive = max_t(unsigned int, 10, 2*num_possible_cpus()); |
b94cce92 | 1920 | #else |
4dae560f | 1921 | rp->maxactive = num_possible_cpus(); |
b94cce92 HN |
1922 | #endif |
1923 | } | |
ec484608 | 1924 | raw_spin_lock_init(&rp->lock); |
b94cce92 HN |
1925 | INIT_HLIST_HEAD(&rp->free_instances); |
1926 | for (i = 0; i < rp->maxactive; i++) { | |
f47cd9b5 AS |
1927 | inst = kmalloc(sizeof(struct kretprobe_instance) + |
1928 | rp->data_size, GFP_KERNEL); | |
b94cce92 HN |
1929 | if (inst == NULL) { |
1930 | free_rp_inst(rp); | |
1931 | return -ENOMEM; | |
1932 | } | |
ef53d9c5 S |
1933 | INIT_HLIST_NODE(&inst->hlist); |
1934 | hlist_add_head(&inst->hlist, &rp->free_instances); | |
b94cce92 HN |
1935 | } |
1936 | ||
1937 | rp->nmissed = 0; | |
1938 | /* Establish function entry probe point */ | |
49ad2fd7 | 1939 | ret = register_kprobe(&rp->kp); |
4a296e07 | 1940 | if (ret != 0) |
b94cce92 HN |
1941 | free_rp_inst(rp); |
1942 | return ret; | |
1943 | } | |
99081ab5 | 1944 | EXPORT_SYMBOL_GPL(register_kretprobe); |
b94cce92 | 1945 | |
55479f64 | 1946 | int register_kretprobes(struct kretprobe **rps, int num) |
4a296e07 MH |
1947 | { |
1948 | int ret = 0, i; | |
1949 | ||
1950 | if (num <= 0) | |
1951 | return -EINVAL; | |
1952 | for (i = 0; i < num; i++) { | |
49ad2fd7 | 1953 | ret = register_kretprobe(rps[i]); |
67dddaad MH |
1954 | if (ret < 0) { |
1955 | if (i > 0) | |
1956 | unregister_kretprobes(rps, i); | |
4a296e07 MH |
1957 | break; |
1958 | } | |
1959 | } | |
1960 | return ret; | |
1961 | } | |
99081ab5 | 1962 | EXPORT_SYMBOL_GPL(register_kretprobes); |
4a296e07 | 1963 | |
55479f64 | 1964 | void unregister_kretprobe(struct kretprobe *rp) |
4a296e07 MH |
1965 | { |
1966 | unregister_kretprobes(&rp, 1); | |
1967 | } | |
99081ab5 | 1968 | EXPORT_SYMBOL_GPL(unregister_kretprobe); |
4a296e07 | 1969 | |
55479f64 | 1970 | void unregister_kretprobes(struct kretprobe **rps, int num) |
4a296e07 MH |
1971 | { |
1972 | int i; | |
1973 | ||
1974 | if (num <= 0) | |
1975 | return; | |
1976 | mutex_lock(&kprobe_mutex); | |
1977 | for (i = 0; i < num; i++) | |
1978 | if (__unregister_kprobe_top(&rps[i]->kp) < 0) | |
1979 | rps[i]->kp.addr = NULL; | |
1980 | mutex_unlock(&kprobe_mutex); | |
1981 | ||
1982 | synchronize_sched(); | |
1983 | for (i = 0; i < num; i++) { | |
1984 | if (rps[i]->kp.addr) { | |
1985 | __unregister_kprobe_bottom(&rps[i]->kp); | |
1986 | cleanup_rp_inst(rps[i]); | |
1987 | } | |
1988 | } | |
1989 | } | |
99081ab5 | 1990 | EXPORT_SYMBOL_GPL(unregister_kretprobes); |
4a296e07 | 1991 | |
9edddaa2 | 1992 | #else /* CONFIG_KRETPROBES */ |
55479f64 | 1993 | int register_kretprobe(struct kretprobe *rp) |
b94cce92 HN |
1994 | { |
1995 | return -ENOSYS; | |
1996 | } | |
99081ab5 | 1997 | EXPORT_SYMBOL_GPL(register_kretprobe); |
b94cce92 | 1998 | |
55479f64 | 1999 | int register_kretprobes(struct kretprobe **rps, int num) |
346fd59b | 2000 | { |
4a296e07 | 2001 | return -ENOSYS; |
346fd59b | 2002 | } |
99081ab5 MH |
2003 | EXPORT_SYMBOL_GPL(register_kretprobes); |
2004 | ||
55479f64 | 2005 | void unregister_kretprobe(struct kretprobe *rp) |
b94cce92 | 2006 | { |
4a296e07 | 2007 | } |
99081ab5 | 2008 | EXPORT_SYMBOL_GPL(unregister_kretprobe); |
b94cce92 | 2009 | |
55479f64 | 2010 | void unregister_kretprobes(struct kretprobe **rps, int num) |
4a296e07 MH |
2011 | { |
2012 | } | |
99081ab5 | 2013 | EXPORT_SYMBOL_GPL(unregister_kretprobes); |
4c4308cb | 2014 | |
820aede0 | 2015 | static int pre_handler_kretprobe(struct kprobe *p, struct pt_regs *regs) |
4a296e07 MH |
2016 | { |
2017 | return 0; | |
b94cce92 | 2018 | } |
820aede0 | 2019 | NOKPROBE_SYMBOL(pre_handler_kretprobe); |
b94cce92 | 2020 | |
4a296e07 MH |
2021 | #endif /* CONFIG_KRETPROBES */ |
2022 | ||
e8386a0c | 2023 | /* Set the kprobe gone and remove its instruction buffer. */ |
55479f64 | 2024 | static void kill_kprobe(struct kprobe *p) |
e8386a0c MH |
2025 | { |
2026 | struct kprobe *kp; | |
de5bd88d | 2027 | |
e8386a0c | 2028 | p->flags |= KPROBE_FLAG_GONE; |
afd66255 | 2029 | if (kprobe_aggrprobe(p)) { |
e8386a0c MH |
2030 | /* |
2031 | * If this is an aggr_kprobe, we have to list all the | |
2032 | * chained probes and mark them GONE. | |
2033 | */ | |
2034 | list_for_each_entry_rcu(kp, &p->list, list) | |
2035 | kp->flags |= KPROBE_FLAG_GONE; | |
2036 | p->post_handler = NULL; | |
2037 | p->break_handler = NULL; | |
afd66255 | 2038 | kill_optimized_kprobe(p); |
e8386a0c MH |
2039 | } |
2040 | /* | |
2041 | * Here, we can remove insn_slot safely, because no thread calls | |
2042 | * the original probed function (which will be freed soon) any more. | |
2043 | */ | |
2044 | arch_remove_kprobe(p); | |
2045 | } | |
2046 | ||
c0614829 | 2047 | /* Disable one kprobe */ |
55479f64 | 2048 | int disable_kprobe(struct kprobe *kp) |
c0614829 MH |
2049 | { |
2050 | int ret = 0; | |
297f9233 | 2051 | struct kprobe *p; |
c0614829 MH |
2052 | |
2053 | mutex_lock(&kprobe_mutex); | |
2054 | ||
6f0f1dd7 | 2055 | /* Disable this kprobe */ |
297f9233 JY |
2056 | p = __disable_kprobe(kp); |
2057 | if (IS_ERR(p)) | |
2058 | ret = PTR_ERR(p); | |
c0614829 | 2059 | |
c0614829 MH |
2060 | mutex_unlock(&kprobe_mutex); |
2061 | return ret; | |
2062 | } | |
2063 | EXPORT_SYMBOL_GPL(disable_kprobe); | |
2064 | ||
2065 | /* Enable one kprobe */ | |
55479f64 | 2066 | int enable_kprobe(struct kprobe *kp) |
c0614829 MH |
2067 | { |
2068 | int ret = 0; | |
2069 | struct kprobe *p; | |
2070 | ||
2071 | mutex_lock(&kprobe_mutex); | |
2072 | ||
2073 | /* Check whether specified probe is valid. */ | |
2074 | p = __get_valid_kprobe(kp); | |
2075 | if (unlikely(p == NULL)) { | |
2076 | ret = -EINVAL; | |
2077 | goto out; | |
2078 | } | |
2079 | ||
2080 | if (kprobe_gone(kp)) { | |
2081 | /* This kprobe has gone, we couldn't enable it. */ | |
2082 | ret = -EINVAL; | |
2083 | goto out; | |
2084 | } | |
2085 | ||
2086 | if (p != kp) | |
2087 | kp->flags &= ~KPROBE_FLAG_DISABLED; | |
2088 | ||
2089 | if (!kprobes_all_disarmed && kprobe_disabled(p)) { | |
2090 | p->flags &= ~KPROBE_FLAG_DISABLED; | |
12310e34 JY |
2091 | ret = arm_kprobe(p); |
2092 | if (ret) | |
2093 | p->flags |= KPROBE_FLAG_DISABLED; | |
c0614829 MH |
2094 | } |
2095 | out: | |
2096 | mutex_unlock(&kprobe_mutex); | |
2097 | return ret; | |
2098 | } | |
2099 | EXPORT_SYMBOL_GPL(enable_kprobe); | |
2100 | ||
820aede0 | 2101 | void dump_kprobe(struct kprobe *kp) |
24851d24 FW |
2102 | { |
2103 | printk(KERN_WARNING "Dumping kprobe:\n"); | |
2104 | printk(KERN_WARNING "Name: %s\nAddress: %p\nOffset: %x\n", | |
2105 | kp->symbol_name, kp->addr, kp->offset); | |
2106 | } | |
820aede0 | 2107 | NOKPROBE_SYMBOL(dump_kprobe); |
24851d24 | 2108 | |
376e2424 MH |
2109 | /* |
2110 | * Lookup and populate the kprobe_blacklist. | |
2111 | * | |
2112 | * Unlike the kretprobe blacklist, we'll need to determine | |
2113 | * the range of addresses that belong to the said functions, | |
2114 | * since a kprobe need not necessarily be at the beginning | |
2115 | * of a function. | |
2116 | */ | |
2117 | static int __init populate_kprobe_blacklist(unsigned long *start, | |
2118 | unsigned long *end) | |
2119 | { | |
2120 | unsigned long *iter; | |
2121 | struct kprobe_blacklist_entry *ent; | |
d81b4253 | 2122 | unsigned long entry, offset = 0, size = 0; |
376e2424 MH |
2123 | |
2124 | for (iter = start; iter < end; iter++) { | |
d81b4253 MH |
2125 | entry = arch_deref_entry_point((void *)*iter); |
2126 | ||
2127 | if (!kernel_text_address(entry) || | |
2128 | !kallsyms_lookup_size_offset(entry, &size, &offset)) { | |
2129 | pr_err("Failed to find blacklist at %p\n", | |
2130 | (void *)entry); | |
376e2424 MH |
2131 | continue; |
2132 | } | |
2133 | ||
2134 | ent = kmalloc(sizeof(*ent), GFP_KERNEL); | |
2135 | if (!ent) | |
2136 | return -ENOMEM; | |
d81b4253 MH |
2137 | ent->start_addr = entry; |
2138 | ent->end_addr = entry + size; | |
376e2424 MH |
2139 | INIT_LIST_HEAD(&ent->list); |
2140 | list_add_tail(&ent->list, &kprobe_blacklist); | |
2141 | } | |
2142 | return 0; | |
2143 | } | |
2144 | ||
e8386a0c | 2145 | /* Module notifier call back, checking kprobes on the module */ |
55479f64 MH |
2146 | static int kprobes_module_callback(struct notifier_block *nb, |
2147 | unsigned long val, void *data) | |
e8386a0c MH |
2148 | { |
2149 | struct module *mod = data; | |
2150 | struct hlist_head *head; | |
e8386a0c MH |
2151 | struct kprobe *p; |
2152 | unsigned int i; | |
f24659d9 | 2153 | int checkcore = (val == MODULE_STATE_GOING); |
e8386a0c | 2154 | |
f24659d9 | 2155 | if (val != MODULE_STATE_GOING && val != MODULE_STATE_LIVE) |
e8386a0c MH |
2156 | return NOTIFY_DONE; |
2157 | ||
2158 | /* | |
f24659d9 MH |
2159 | * When MODULE_STATE_GOING was notified, both of module .text and |
2160 | * .init.text sections would be freed. When MODULE_STATE_LIVE was | |
2161 | * notified, only .init.text section would be freed. We need to | |
2162 | * disable kprobes which have been inserted in the sections. | |
e8386a0c MH |
2163 | */ |
2164 | mutex_lock(&kprobe_mutex); | |
2165 | for (i = 0; i < KPROBE_TABLE_SIZE; i++) { | |
2166 | head = &kprobe_table[i]; | |
b67bfe0d | 2167 | hlist_for_each_entry_rcu(p, head, hlist) |
f24659d9 MH |
2168 | if (within_module_init((unsigned long)p->addr, mod) || |
2169 | (checkcore && | |
2170 | within_module_core((unsigned long)p->addr, mod))) { | |
e8386a0c MH |
2171 | /* |
2172 | * The vaddr this probe is installed will soon | |
2173 | * be vfreed buy not synced to disk. Hence, | |
2174 | * disarming the breakpoint isn't needed. | |
545a0281 SRV |
2175 | * |
2176 | * Note, this will also move any optimized probes | |
2177 | * that are pending to be removed from their | |
2178 | * corresponding lists to the freeing_list and | |
2179 | * will not be touched by the delayed | |
2180 | * kprobe_optimizer work handler. | |
e8386a0c MH |
2181 | */ |
2182 | kill_kprobe(p); | |
2183 | } | |
2184 | } | |
2185 | mutex_unlock(&kprobe_mutex); | |
2186 | return NOTIFY_DONE; | |
2187 | } | |
2188 | ||
2189 | static struct notifier_block kprobe_module_nb = { | |
2190 | .notifier_call = kprobes_module_callback, | |
2191 | .priority = 0 | |
2192 | }; | |
2193 | ||
376e2424 MH |
2194 | /* Markers of _kprobe_blacklist section */ |
2195 | extern unsigned long __start_kprobe_blacklist[]; | |
2196 | extern unsigned long __stop_kprobe_blacklist[]; | |
2197 | ||
1da177e4 LT |
2198 | static int __init init_kprobes(void) |
2199 | { | |
2200 | int i, err = 0; | |
2201 | ||
2202 | /* FIXME allocate the probe table, currently defined statically */ | |
2203 | /* initialize all list heads */ | |
b94cce92 | 2204 | for (i = 0; i < KPROBE_TABLE_SIZE; i++) { |
1da177e4 | 2205 | INIT_HLIST_HEAD(&kprobe_table[i]); |
b94cce92 | 2206 | INIT_HLIST_HEAD(&kretprobe_inst_table[i]); |
ec484608 | 2207 | raw_spin_lock_init(&(kretprobe_table_locks[i].lock)); |
b94cce92 | 2208 | } |
1da177e4 | 2209 | |
376e2424 MH |
2210 | err = populate_kprobe_blacklist(__start_kprobe_blacklist, |
2211 | __stop_kprobe_blacklist); | |
2212 | if (err) { | |
2213 | pr_err("kprobes: failed to populate blacklist: %d\n", err); | |
2214 | pr_err("Please take care of using kprobes.\n"); | |
3d8d996e SD |
2215 | } |
2216 | ||
f438d914 MH |
2217 | if (kretprobe_blacklist_size) { |
2218 | /* lookup the function address from its name */ | |
2219 | for (i = 0; kretprobe_blacklist[i].name != NULL; i++) { | |
49e0b465 | 2220 | kretprobe_blacklist[i].addr = |
290e3070 | 2221 | kprobe_lookup_name(kretprobe_blacklist[i].name, 0); |
f438d914 MH |
2222 | if (!kretprobe_blacklist[i].addr) |
2223 | printk("kretprobe: lookup failed: %s\n", | |
2224 | kretprobe_blacklist[i].name); | |
2225 | } | |
2226 | } | |
2227 | ||
b2be84df MH |
2228 | #if defined(CONFIG_OPTPROBES) |
2229 | #if defined(__ARCH_WANT_KPROBES_INSN_SLOT) | |
afd66255 MH |
2230 | /* Init kprobe_optinsn_slots */ |
2231 | kprobe_optinsn_slots.insn_size = MAX_OPTINSN_SIZE; | |
2232 | #endif | |
b2be84df MH |
2233 | /* By default, kprobes can be optimized */ |
2234 | kprobes_allow_optimization = true; | |
2235 | #endif | |
afd66255 | 2236 | |
e579abeb MH |
2237 | /* By default, kprobes are armed */ |
2238 | kprobes_all_disarmed = false; | |
bf8f6e5b | 2239 | |
6772926b | 2240 | err = arch_init_kprobes(); |
802eae7c RL |
2241 | if (!err) |
2242 | err = register_die_notifier(&kprobe_exceptions_nb); | |
e8386a0c MH |
2243 | if (!err) |
2244 | err = register_module_notifier(&kprobe_module_nb); | |
2245 | ||
ef53d9c5 | 2246 | kprobes_initialized = (err == 0); |
802eae7c | 2247 | |
8c1c9356 AM |
2248 | if (!err) |
2249 | init_test_probes(); | |
1da177e4 LT |
2250 | return err; |
2251 | } | |
2252 | ||
346fd59b | 2253 | #ifdef CONFIG_DEBUG_FS |
55479f64 | 2254 | static void report_probe(struct seq_file *pi, struct kprobe *p, |
afd66255 | 2255 | const char *sym, int offset, char *modname, struct kprobe *pp) |
346fd59b SD |
2256 | { |
2257 | char *kprobe_type; | |
2258 | ||
2259 | if (p->pre_handler == pre_handler_kretprobe) | |
2260 | kprobe_type = "r"; | |
346fd59b SD |
2261 | else |
2262 | kprobe_type = "k"; | |
afd66255 | 2263 | |
346fd59b | 2264 | if (sym) |
afd66255 | 2265 | seq_printf(pi, "%p %s %s+0x%x %s ", |
de5bd88d | 2266 | p->addr, kprobe_type, sym, offset, |
afd66255 | 2267 | (modname ? modname : " ")); |
346fd59b | 2268 | else |
afd66255 MH |
2269 | seq_printf(pi, "%p %s %p ", |
2270 | p->addr, kprobe_type, p->addr); | |
2271 | ||
2272 | if (!pp) | |
2273 | pp = p; | |
ae6aa16f | 2274 | seq_printf(pi, "%s%s%s%s\n", |
afd66255 MH |
2275 | (kprobe_gone(p) ? "[GONE]" : ""), |
2276 | ((kprobe_disabled(p) && !kprobe_gone(p)) ? "[DISABLED]" : ""), | |
ae6aa16f MH |
2277 | (kprobe_optimized(pp) ? "[OPTIMIZED]" : ""), |
2278 | (kprobe_ftrace(pp) ? "[FTRACE]" : "")); | |
346fd59b SD |
2279 | } |
2280 | ||
55479f64 | 2281 | static void *kprobe_seq_start(struct seq_file *f, loff_t *pos) |
346fd59b SD |
2282 | { |
2283 | return (*pos < KPROBE_TABLE_SIZE) ? pos : NULL; | |
2284 | } | |
2285 | ||
55479f64 | 2286 | static void *kprobe_seq_next(struct seq_file *f, void *v, loff_t *pos) |
346fd59b SD |
2287 | { |
2288 | (*pos)++; | |
2289 | if (*pos >= KPROBE_TABLE_SIZE) | |
2290 | return NULL; | |
2291 | return pos; | |
2292 | } | |
2293 | ||
55479f64 | 2294 | static void kprobe_seq_stop(struct seq_file *f, void *v) |
346fd59b SD |
2295 | { |
2296 | /* Nothing to do */ | |
2297 | } | |
2298 | ||
55479f64 | 2299 | static int show_kprobe_addr(struct seq_file *pi, void *v) |
346fd59b SD |
2300 | { |
2301 | struct hlist_head *head; | |
346fd59b SD |
2302 | struct kprobe *p, *kp; |
2303 | const char *sym = NULL; | |
2304 | unsigned int i = *(loff_t *) v; | |
ffb45122 | 2305 | unsigned long offset = 0; |
ab767865 | 2306 | char *modname, namebuf[KSYM_NAME_LEN]; |
346fd59b SD |
2307 | |
2308 | head = &kprobe_table[i]; | |
2309 | preempt_disable(); | |
b67bfe0d | 2310 | hlist_for_each_entry_rcu(p, head, hlist) { |
ffb45122 | 2311 | sym = kallsyms_lookup((unsigned long)p->addr, NULL, |
346fd59b | 2312 | &offset, &modname, namebuf); |
afd66255 | 2313 | if (kprobe_aggrprobe(p)) { |
346fd59b | 2314 | list_for_each_entry_rcu(kp, &p->list, list) |
afd66255 | 2315 | report_probe(pi, kp, sym, offset, modname, p); |
346fd59b | 2316 | } else |
afd66255 | 2317 | report_probe(pi, p, sym, offset, modname, NULL); |
346fd59b SD |
2318 | } |
2319 | preempt_enable(); | |
2320 | return 0; | |
2321 | } | |
2322 | ||
88e9d34c | 2323 | static const struct seq_operations kprobes_seq_ops = { |
346fd59b SD |
2324 | .start = kprobe_seq_start, |
2325 | .next = kprobe_seq_next, | |
2326 | .stop = kprobe_seq_stop, | |
2327 | .show = show_kprobe_addr | |
2328 | }; | |
2329 | ||
55479f64 | 2330 | static int kprobes_open(struct inode *inode, struct file *filp) |
346fd59b SD |
2331 | { |
2332 | return seq_open(filp, &kprobes_seq_ops); | |
2333 | } | |
2334 | ||
828c0950 | 2335 | static const struct file_operations debugfs_kprobes_operations = { |
346fd59b SD |
2336 | .open = kprobes_open, |
2337 | .read = seq_read, | |
2338 | .llseek = seq_lseek, | |
2339 | .release = seq_release, | |
2340 | }; | |
2341 | ||
63724740 MH |
2342 | /* kprobes/blacklist -- shows which functions can not be probed */ |
2343 | static void *kprobe_blacklist_seq_start(struct seq_file *m, loff_t *pos) | |
2344 | { | |
2345 | return seq_list_start(&kprobe_blacklist, *pos); | |
2346 | } | |
2347 | ||
2348 | static void *kprobe_blacklist_seq_next(struct seq_file *m, void *v, loff_t *pos) | |
2349 | { | |
2350 | return seq_list_next(v, &kprobe_blacklist, pos); | |
2351 | } | |
2352 | ||
2353 | static int kprobe_blacklist_seq_show(struct seq_file *m, void *v) | |
2354 | { | |
2355 | struct kprobe_blacklist_entry *ent = | |
2356 | list_entry(v, struct kprobe_blacklist_entry, list); | |
2357 | ||
bcbd385b | 2358 | seq_printf(m, "0x%px-0x%px\t%ps\n", (void *)ent->start_addr, |
63724740 MH |
2359 | (void *)ent->end_addr, (void *)ent->start_addr); |
2360 | return 0; | |
2361 | } | |
2362 | ||
2363 | static const struct seq_operations kprobe_blacklist_seq_ops = { | |
2364 | .start = kprobe_blacklist_seq_start, | |
2365 | .next = kprobe_blacklist_seq_next, | |
2366 | .stop = kprobe_seq_stop, /* Reuse void function */ | |
2367 | .show = kprobe_blacklist_seq_show, | |
2368 | }; | |
2369 | ||
2370 | static int kprobe_blacklist_open(struct inode *inode, struct file *filp) | |
2371 | { | |
2372 | return seq_open(filp, &kprobe_blacklist_seq_ops); | |
2373 | } | |
2374 | ||
2375 | static const struct file_operations debugfs_kprobe_blacklist_ops = { | |
2376 | .open = kprobe_blacklist_open, | |
2377 | .read = seq_read, | |
2378 | .llseek = seq_lseek, | |
2379 | .release = seq_release, | |
2380 | }; | |
2381 | ||
12310e34 | 2382 | static int arm_all_kprobes(void) |
bf8f6e5b AM |
2383 | { |
2384 | struct hlist_head *head; | |
bf8f6e5b | 2385 | struct kprobe *p; |
12310e34 JY |
2386 | unsigned int i, total = 0, errors = 0; |
2387 | int err, ret = 0; | |
bf8f6e5b AM |
2388 | |
2389 | mutex_lock(&kprobe_mutex); | |
2390 | ||
e579abeb MH |
2391 | /* If kprobes are armed, just return */ |
2392 | if (!kprobes_all_disarmed) | |
bf8f6e5b AM |
2393 | goto already_enabled; |
2394 | ||
977ad481 WN |
2395 | /* |
2396 | * optimize_kprobe() called by arm_kprobe() checks | |
2397 | * kprobes_all_disarmed, so set kprobes_all_disarmed before | |
2398 | * arm_kprobe. | |
2399 | */ | |
2400 | kprobes_all_disarmed = false; | |
afd66255 | 2401 | /* Arming kprobes doesn't optimize kprobe itself */ |
bf8f6e5b AM |
2402 | for (i = 0; i < KPROBE_TABLE_SIZE; i++) { |
2403 | head = &kprobe_table[i]; | |
12310e34 JY |
2404 | /* Arm all kprobes on a best-effort basis */ |
2405 | hlist_for_each_entry_rcu(p, head, hlist) { | |
2406 | if (!kprobe_disabled(p)) { | |
2407 | err = arm_kprobe(p); | |
2408 | if (err) { | |
2409 | errors++; | |
2410 | ret = err; | |
2411 | } | |
2412 | total++; | |
2413 | } | |
2414 | } | |
bf8f6e5b AM |
2415 | } |
2416 | ||
12310e34 JY |
2417 | if (errors) |
2418 | pr_warn("Kprobes globally enabled, but failed to arm %d out of %d probes\n", | |
2419 | errors, total); | |
2420 | else | |
2421 | pr_info("Kprobes globally enabled\n"); | |
bf8f6e5b AM |
2422 | |
2423 | already_enabled: | |
2424 | mutex_unlock(&kprobe_mutex); | |
12310e34 | 2425 | return ret; |
bf8f6e5b AM |
2426 | } |
2427 | ||
297f9233 | 2428 | static int disarm_all_kprobes(void) |
bf8f6e5b AM |
2429 | { |
2430 | struct hlist_head *head; | |
bf8f6e5b | 2431 | struct kprobe *p; |
297f9233 JY |
2432 | unsigned int i, total = 0, errors = 0; |
2433 | int err, ret = 0; | |
bf8f6e5b AM |
2434 | |
2435 | mutex_lock(&kprobe_mutex); | |
2436 | ||
e579abeb | 2437 | /* If kprobes are already disarmed, just return */ |
6274de49 MH |
2438 | if (kprobes_all_disarmed) { |
2439 | mutex_unlock(&kprobe_mutex); | |
297f9233 | 2440 | return 0; |
6274de49 | 2441 | } |
bf8f6e5b | 2442 | |
e579abeb | 2443 | kprobes_all_disarmed = true; |
afd66255 | 2444 | |
bf8f6e5b AM |
2445 | for (i = 0; i < KPROBE_TABLE_SIZE; i++) { |
2446 | head = &kprobe_table[i]; | |
297f9233 | 2447 | /* Disarm all kprobes on a best-effort basis */ |
b67bfe0d | 2448 | hlist_for_each_entry_rcu(p, head, hlist) { |
297f9233 JY |
2449 | if (!arch_trampoline_kprobe(p) && !kprobe_disabled(p)) { |
2450 | err = disarm_kprobe(p, false); | |
2451 | if (err) { | |
2452 | errors++; | |
2453 | ret = err; | |
2454 | } | |
2455 | total++; | |
2456 | } | |
bf8f6e5b AM |
2457 | } |
2458 | } | |
297f9233 JY |
2459 | |
2460 | if (errors) | |
2461 | pr_warn("Kprobes globally disabled, but failed to disarm %d out of %d probes\n", | |
2462 | errors, total); | |
2463 | else | |
2464 | pr_info("Kprobes globally disabled\n"); | |
2465 | ||
bf8f6e5b | 2466 | mutex_unlock(&kprobe_mutex); |
bf8f6e5b | 2467 | |
6274de49 MH |
2468 | /* Wait for disarming all kprobes by optimizer */ |
2469 | wait_for_kprobe_optimizer(); | |
297f9233 JY |
2470 | |
2471 | return ret; | |
bf8f6e5b AM |
2472 | } |
2473 | ||
2474 | /* | |
2475 | * XXX: The debugfs bool file interface doesn't allow for callbacks | |
2476 | * when the bool state is switched. We can reuse that facility when | |
2477 | * available | |
2478 | */ | |
2479 | static ssize_t read_enabled_file_bool(struct file *file, | |
2480 | char __user *user_buf, size_t count, loff_t *ppos) | |
2481 | { | |
2482 | char buf[3]; | |
2483 | ||
e579abeb | 2484 | if (!kprobes_all_disarmed) |
bf8f6e5b AM |
2485 | buf[0] = '1'; |
2486 | else | |
2487 | buf[0] = '0'; | |
2488 | buf[1] = '\n'; | |
2489 | buf[2] = 0x00; | |
2490 | return simple_read_from_buffer(user_buf, count, ppos, buf, 2); | |
2491 | } | |
2492 | ||
2493 | static ssize_t write_enabled_file_bool(struct file *file, | |
2494 | const char __user *user_buf, size_t count, loff_t *ppos) | |
2495 | { | |
2496 | char buf[32]; | |
efeb156e | 2497 | size_t buf_size; |
12310e34 | 2498 | int ret = 0; |
bf8f6e5b AM |
2499 | |
2500 | buf_size = min(count, (sizeof(buf)-1)); | |
2501 | if (copy_from_user(buf, user_buf, buf_size)) | |
2502 | return -EFAULT; | |
2503 | ||
10fb46d5 | 2504 | buf[buf_size] = '\0'; |
bf8f6e5b AM |
2505 | switch (buf[0]) { |
2506 | case 'y': | |
2507 | case 'Y': | |
2508 | case '1': | |
12310e34 | 2509 | ret = arm_all_kprobes(); |
bf8f6e5b AM |
2510 | break; |
2511 | case 'n': | |
2512 | case 'N': | |
2513 | case '0': | |
297f9233 | 2514 | ret = disarm_all_kprobes(); |
bf8f6e5b | 2515 | break; |
10fb46d5 MK |
2516 | default: |
2517 | return -EINVAL; | |
bf8f6e5b AM |
2518 | } |
2519 | ||
12310e34 JY |
2520 | if (ret) |
2521 | return ret; | |
2522 | ||
bf8f6e5b AM |
2523 | return count; |
2524 | } | |
2525 | ||
828c0950 | 2526 | static const struct file_operations fops_kp = { |
bf8f6e5b AM |
2527 | .read = read_enabled_file_bool, |
2528 | .write = write_enabled_file_bool, | |
6038f373 | 2529 | .llseek = default_llseek, |
bf8f6e5b AM |
2530 | }; |
2531 | ||
55479f64 | 2532 | static int __init debugfs_kprobe_init(void) |
346fd59b SD |
2533 | { |
2534 | struct dentry *dir, *file; | |
bf8f6e5b | 2535 | unsigned int value = 1; |
346fd59b SD |
2536 | |
2537 | dir = debugfs_create_dir("kprobes", NULL); | |
2538 | if (!dir) | |
2539 | return -ENOMEM; | |
2540 | ||
e3869792 | 2541 | file = debugfs_create_file("list", 0444, dir, NULL, |
346fd59b | 2542 | &debugfs_kprobes_operations); |
63724740 MH |
2543 | if (!file) |
2544 | goto error; | |
346fd59b | 2545 | |
bf8f6e5b AM |
2546 | file = debugfs_create_file("enabled", 0600, dir, |
2547 | &value, &fops_kp); | |
63724740 MH |
2548 | if (!file) |
2549 | goto error; | |
2550 | ||
2551 | file = debugfs_create_file("blacklist", 0444, dir, NULL, | |
2552 | &debugfs_kprobe_blacklist_ops); | |
2553 | if (!file) | |
2554 | goto error; | |
bf8f6e5b | 2555 | |
346fd59b | 2556 | return 0; |
63724740 MH |
2557 | |
2558 | error: | |
2559 | debugfs_remove(dir); | |
2560 | return -ENOMEM; | |
346fd59b SD |
2561 | } |
2562 | ||
2563 | late_initcall(debugfs_kprobe_init); | |
2564 | #endif /* CONFIG_DEBUG_FS */ | |
2565 | ||
2566 | module_init(init_kprobes); |