]>
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 | ||
059053a2 MH |
630 | /* kprobes with post_handler can not be optimized */ |
631 | if (p->post_handler) | |
afd66255 MH |
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 | 702 | /* Cancel unoptimizing for reusing */ |
82b0f70f | 703 | static int reuse_unused_kprobe(struct kprobe *ap) |
0490cd1f MH |
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); | |
4458515b | 713 | WARN_ON_ONCE(list_empty(&op->list)); |
0490cd1f MH |
714 | /* Enable the probe again */ |
715 | ap->flags &= ~KPROBE_FLAG_DISABLED; | |
716 | /* Optimize it again (remove from op->list) */ | |
23a926e5 MH |
717 | if (!kprobe_optready(ap)) |
718 | return -EINVAL; | |
82b0f70f | 719 | |
0490cd1f | 720 | optimize_kprobe(ap); |
82b0f70f | 721 | return 0; |
0490cd1f MH |
722 | } |
723 | ||
afd66255 | 724 | /* Remove optimized instructions */ |
55479f64 | 725 | static void kill_optimized_kprobe(struct kprobe *p) |
afd66255 MH |
726 | { |
727 | struct optimized_kprobe *op; | |
728 | ||
729 | op = container_of(p, struct optimized_kprobe, kp); | |
6274de49 MH |
730 | if (!list_empty(&op->list)) |
731 | /* Dequeue from the (un)optimization queue */ | |
afd66255 | 732 | list_del_init(&op->list); |
6274de49 | 733 | op->kp.flags &= ~KPROBE_FLAG_OPTIMIZED; |
7b959fc5 MH |
734 | |
735 | if (kprobe_unused(p)) { | |
736 | /* Enqueue if it is unused */ | |
737 | list_add(&op->list, &freeing_list); | |
738 | /* | |
739 | * Remove unused probes from the hash list. After waiting | |
740 | * for synchronization, this probe is reclaimed. | |
741 | * (reclaiming is done by do_free_cleaned_kprobes().) | |
742 | */ | |
743 | hlist_del_rcu(&op->kp.hlist); | |
744 | } | |
745 | ||
6274de49 | 746 | /* Don't touch the code, because it is already freed. */ |
afd66255 MH |
747 | arch_remove_optimized_kprobe(op); |
748 | } | |
749 | ||
a460246c MH |
750 | static inline |
751 | void __prepare_optimized_kprobe(struct optimized_kprobe *op, struct kprobe *p) | |
752 | { | |
753 | if (!kprobe_ftrace(p)) | |
754 | arch_prepare_optimized_kprobe(op, p); | |
755 | } | |
756 | ||
afd66255 | 757 | /* Try to prepare optimized instructions */ |
55479f64 | 758 | static void prepare_optimized_kprobe(struct kprobe *p) |
afd66255 MH |
759 | { |
760 | struct optimized_kprobe *op; | |
761 | ||
762 | op = container_of(p, struct optimized_kprobe, kp); | |
a460246c | 763 | __prepare_optimized_kprobe(op, p); |
afd66255 MH |
764 | } |
765 | ||
afd66255 | 766 | /* Allocate new optimized_kprobe and try to prepare optimized instructions */ |
55479f64 | 767 | static struct kprobe *alloc_aggr_kprobe(struct kprobe *p) |
afd66255 MH |
768 | { |
769 | struct optimized_kprobe *op; | |
770 | ||
771 | op = kzalloc(sizeof(struct optimized_kprobe), GFP_KERNEL); | |
772 | if (!op) | |
773 | return NULL; | |
774 | ||
775 | INIT_LIST_HEAD(&op->list); | |
776 | op->kp.addr = p->addr; | |
a460246c | 777 | __prepare_optimized_kprobe(op, p); |
afd66255 MH |
778 | |
779 | return &op->kp; | |
780 | } | |
781 | ||
55479f64 | 782 | static void init_aggr_kprobe(struct kprobe *ap, struct kprobe *p); |
afd66255 MH |
783 | |
784 | /* | |
785 | * Prepare an optimized_kprobe and optimize it | |
786 | * NOTE: p must be a normal registered kprobe | |
787 | */ | |
55479f64 | 788 | static void try_to_optimize_kprobe(struct kprobe *p) |
afd66255 MH |
789 | { |
790 | struct kprobe *ap; | |
791 | struct optimized_kprobe *op; | |
792 | ||
ae6aa16f MH |
793 | /* Impossible to optimize ftrace-based kprobe */ |
794 | if (kprobe_ftrace(p)) | |
795 | return; | |
796 | ||
25764288 | 797 | /* For preparing optimization, jump_label_text_reserved() is called */ |
2d1e38f5 | 798 | cpus_read_lock(); |
25764288 MH |
799 | jump_label_lock(); |
800 | mutex_lock(&text_mutex); | |
801 | ||
afd66255 MH |
802 | ap = alloc_aggr_kprobe(p); |
803 | if (!ap) | |
25764288 | 804 | goto out; |
afd66255 MH |
805 | |
806 | op = container_of(ap, struct optimized_kprobe, kp); | |
807 | if (!arch_prepared_optinsn(&op->optinsn)) { | |
808 | /* If failed to setup optimizing, fallback to kprobe */ | |
6274de49 MH |
809 | arch_remove_optimized_kprobe(op); |
810 | kfree(op); | |
25764288 | 811 | goto out; |
afd66255 MH |
812 | } |
813 | ||
814 | init_aggr_kprobe(ap, p); | |
25764288 MH |
815 | optimize_kprobe(ap); /* This just kicks optimizer thread */ |
816 | ||
817 | out: | |
818 | mutex_unlock(&text_mutex); | |
819 | jump_label_unlock(); | |
2d1e38f5 | 820 | cpus_read_unlock(); |
afd66255 MH |
821 | } |
822 | ||
b2be84df | 823 | #ifdef CONFIG_SYSCTL |
55479f64 | 824 | static void optimize_all_kprobes(void) |
b2be84df MH |
825 | { |
826 | struct hlist_head *head; | |
b2be84df MH |
827 | struct kprobe *p; |
828 | unsigned int i; | |
829 | ||
5c51543b | 830 | mutex_lock(&kprobe_mutex); |
b2be84df MH |
831 | /* If optimization is already allowed, just return */ |
832 | if (kprobes_allow_optimization) | |
5c51543b | 833 | goto out; |
b2be84df | 834 | |
2d1e38f5 | 835 | cpus_read_lock(); |
b2be84df | 836 | kprobes_allow_optimization = true; |
b2be84df MH |
837 | for (i = 0; i < KPROBE_TABLE_SIZE; i++) { |
838 | head = &kprobe_table[i]; | |
b67bfe0d | 839 | hlist_for_each_entry_rcu(p, head, hlist) |
b2be84df MH |
840 | if (!kprobe_disabled(p)) |
841 | optimize_kprobe(p); | |
842 | } | |
2d1e38f5 | 843 | cpus_read_unlock(); |
b2be84df | 844 | printk(KERN_INFO "Kprobes globally optimized\n"); |
5c51543b MH |
845 | out: |
846 | mutex_unlock(&kprobe_mutex); | |
b2be84df MH |
847 | } |
848 | ||
55479f64 | 849 | static void unoptimize_all_kprobes(void) |
b2be84df MH |
850 | { |
851 | struct hlist_head *head; | |
b2be84df MH |
852 | struct kprobe *p; |
853 | unsigned int i; | |
854 | ||
5c51543b | 855 | mutex_lock(&kprobe_mutex); |
b2be84df | 856 | /* If optimization is already prohibited, just return */ |
5c51543b MH |
857 | if (!kprobes_allow_optimization) { |
858 | mutex_unlock(&kprobe_mutex); | |
b2be84df | 859 | return; |
5c51543b | 860 | } |
b2be84df | 861 | |
2d1e38f5 | 862 | cpus_read_lock(); |
b2be84df | 863 | kprobes_allow_optimization = false; |
b2be84df MH |
864 | for (i = 0; i < KPROBE_TABLE_SIZE; i++) { |
865 | head = &kprobe_table[i]; | |
b67bfe0d | 866 | hlist_for_each_entry_rcu(p, head, hlist) { |
b2be84df | 867 | if (!kprobe_disabled(p)) |
6274de49 | 868 | unoptimize_kprobe(p, false); |
b2be84df MH |
869 | } |
870 | } | |
2d1e38f5 | 871 | cpus_read_unlock(); |
5c51543b MH |
872 | mutex_unlock(&kprobe_mutex); |
873 | ||
6274de49 MH |
874 | /* Wait for unoptimizing completion */ |
875 | wait_for_kprobe_optimizer(); | |
876 | printk(KERN_INFO "Kprobes globally unoptimized\n"); | |
b2be84df MH |
877 | } |
878 | ||
5c51543b | 879 | static DEFINE_MUTEX(kprobe_sysctl_mutex); |
b2be84df MH |
880 | int sysctl_kprobes_optimization; |
881 | int proc_kprobes_optimization_handler(struct ctl_table *table, int write, | |
882 | void __user *buffer, size_t *length, | |
883 | loff_t *ppos) | |
884 | { | |
885 | int ret; | |
886 | ||
5c51543b | 887 | mutex_lock(&kprobe_sysctl_mutex); |
b2be84df MH |
888 | sysctl_kprobes_optimization = kprobes_allow_optimization ? 1 : 0; |
889 | ret = proc_dointvec_minmax(table, write, buffer, length, ppos); | |
890 | ||
891 | if (sysctl_kprobes_optimization) | |
892 | optimize_all_kprobes(); | |
893 | else | |
894 | unoptimize_all_kprobes(); | |
5c51543b | 895 | mutex_unlock(&kprobe_sysctl_mutex); |
b2be84df MH |
896 | |
897 | return ret; | |
898 | } | |
899 | #endif /* CONFIG_SYSCTL */ | |
900 | ||
6274de49 | 901 | /* Put a breakpoint for a probe. Must be called with text_mutex locked */ |
55479f64 | 902 | static void __arm_kprobe(struct kprobe *p) |
afd66255 | 903 | { |
6d8e40a8 | 904 | struct kprobe *_p; |
afd66255 MH |
905 | |
906 | /* Check collision with other optimized kprobes */ | |
6d8e40a8 MH |
907 | _p = get_optimized_kprobe((unsigned long)p->addr); |
908 | if (unlikely(_p)) | |
6274de49 MH |
909 | /* Fallback to unoptimized kprobe */ |
910 | unoptimize_kprobe(_p, true); | |
afd66255 MH |
911 | |
912 | arch_arm_kprobe(p); | |
913 | optimize_kprobe(p); /* Try to optimize (add kprobe to a list) */ | |
914 | } | |
915 | ||
6274de49 | 916 | /* Remove the breakpoint of a probe. Must be called with text_mutex locked */ |
55479f64 | 917 | static void __disarm_kprobe(struct kprobe *p, bool reopt) |
afd66255 | 918 | { |
6d8e40a8 | 919 | struct kprobe *_p; |
afd66255 | 920 | |
69d54b91 WN |
921 | /* Try to unoptimize */ |
922 | unoptimize_kprobe(p, kprobes_all_disarmed); | |
afd66255 | 923 | |
6274de49 MH |
924 | if (!kprobe_queued(p)) { |
925 | arch_disarm_kprobe(p); | |
926 | /* If another kprobe was blocked, optimize it. */ | |
927 | _p = get_optimized_kprobe((unsigned long)p->addr); | |
928 | if (unlikely(_p) && reopt) | |
929 | optimize_kprobe(_p); | |
930 | } | |
931 | /* TODO: reoptimize others after unoptimized this probe */ | |
afd66255 MH |
932 | } |
933 | ||
934 | #else /* !CONFIG_OPTPROBES */ | |
935 | ||
936 | #define optimize_kprobe(p) do {} while (0) | |
6274de49 | 937 | #define unoptimize_kprobe(p, f) do {} while (0) |
afd66255 MH |
938 | #define kill_optimized_kprobe(p) do {} while (0) |
939 | #define prepare_optimized_kprobe(p) do {} while (0) | |
940 | #define try_to_optimize_kprobe(p) do {} while (0) | |
941 | #define __arm_kprobe(p) arch_arm_kprobe(p) | |
6274de49 MH |
942 | #define __disarm_kprobe(p, o) arch_disarm_kprobe(p) |
943 | #define kprobe_disarmed(p) kprobe_disabled(p) | |
944 | #define wait_for_kprobe_optimizer() do {} while (0) | |
afd66255 | 945 | |
82b0f70f | 946 | static int reuse_unused_kprobe(struct kprobe *ap) |
0490cd1f | 947 | { |
82b0f70f MH |
948 | /* |
949 | * If the optimized kprobe is NOT supported, the aggr kprobe is | |
950 | * released at the same time that the last aggregated kprobe is | |
951 | * unregistered. | |
952 | * Thus there should be no chance to reuse unused kprobe. | |
953 | */ | |
0490cd1f | 954 | printk(KERN_ERR "Error: There should be no unused kprobe here.\n"); |
82b0f70f | 955 | return -EINVAL; |
0490cd1f MH |
956 | } |
957 | ||
55479f64 | 958 | static void free_aggr_kprobe(struct kprobe *p) |
afd66255 | 959 | { |
6274de49 | 960 | arch_remove_kprobe(p); |
afd66255 MH |
961 | kfree(p); |
962 | } | |
963 | ||
55479f64 | 964 | static struct kprobe *alloc_aggr_kprobe(struct kprobe *p) |
afd66255 MH |
965 | { |
966 | return kzalloc(sizeof(struct kprobe), GFP_KERNEL); | |
967 | } | |
968 | #endif /* CONFIG_OPTPROBES */ | |
969 | ||
e7dbfe34 | 970 | #ifdef CONFIG_KPROBES_ON_FTRACE |
ae6aa16f | 971 | static struct ftrace_ops kprobe_ftrace_ops __read_mostly = { |
e5253896 | 972 | .func = kprobe_ftrace_handler, |
1d70be34 | 973 | .flags = FTRACE_OPS_FL_SAVE_REGS | FTRACE_OPS_FL_IPMODIFY, |
ae6aa16f MH |
974 | }; |
975 | static int kprobe_ftrace_enabled; | |
976 | ||
977 | /* Must ensure p->addr is really on ftrace */ | |
55479f64 | 978 | static int prepare_kprobe(struct kprobe *p) |
ae6aa16f MH |
979 | { |
980 | if (!kprobe_ftrace(p)) | |
981 | return arch_prepare_kprobe(p); | |
982 | ||
983 | return arch_prepare_kprobe_ftrace(p); | |
984 | } | |
985 | ||
986 | /* Caller must lock kprobe_mutex */ | |
12310e34 | 987 | static int arm_kprobe_ftrace(struct kprobe *p) |
ae6aa16f | 988 | { |
12310e34 | 989 | int ret = 0; |
ae6aa16f MH |
990 | |
991 | ret = ftrace_set_filter_ip(&kprobe_ftrace_ops, | |
992 | (unsigned long)p->addr, 0, 0); | |
12310e34 | 993 | if (ret) { |
4458515b MH |
994 | pr_debug("Failed to arm kprobe-ftrace at %pS (%d)\n", |
995 | p->addr, ret); | |
12310e34 JY |
996 | return ret; |
997 | } | |
998 | ||
999 | if (kprobe_ftrace_enabled == 0) { | |
ae6aa16f | 1000 | ret = register_ftrace_function(&kprobe_ftrace_ops); |
12310e34 JY |
1001 | if (ret) { |
1002 | pr_debug("Failed to init kprobe-ftrace (%d)\n", ret); | |
1003 | goto err_ftrace; | |
1004 | } | |
ae6aa16f | 1005 | } |
12310e34 JY |
1006 | |
1007 | kprobe_ftrace_enabled++; | |
1008 | return ret; | |
1009 | ||
1010 | err_ftrace: | |
1011 | /* | |
1012 | * Note: Since kprobe_ftrace_ops has IPMODIFY set, and ftrace requires a | |
1013 | * non-empty filter_hash for IPMODIFY ops, we're safe from an accidental | |
1014 | * empty filter_hash which would undesirably trace all functions. | |
1015 | */ | |
1016 | ftrace_set_filter_ip(&kprobe_ftrace_ops, (unsigned long)p->addr, 1, 0); | |
1017 | return ret; | |
ae6aa16f MH |
1018 | } |
1019 | ||
1020 | /* Caller must lock kprobe_mutex */ | |
297f9233 | 1021 | static int disarm_kprobe_ftrace(struct kprobe *p) |
ae6aa16f | 1022 | { |
297f9233 | 1023 | int ret = 0; |
ae6aa16f | 1024 | |
297f9233 | 1025 | if (kprobe_ftrace_enabled == 1) { |
ae6aa16f | 1026 | ret = unregister_ftrace_function(&kprobe_ftrace_ops); |
297f9233 JY |
1027 | if (WARN(ret < 0, "Failed to unregister kprobe-ftrace (%d)\n", ret)) |
1028 | return ret; | |
ae6aa16f | 1029 | } |
297f9233 JY |
1030 | |
1031 | kprobe_ftrace_enabled--; | |
1032 | ||
ae6aa16f MH |
1033 | ret = ftrace_set_filter_ip(&kprobe_ftrace_ops, |
1034 | (unsigned long)p->addr, 1, 0); | |
4458515b MH |
1035 | WARN_ONCE(ret < 0, "Failed to disarm kprobe-ftrace at %pS (%d)\n", |
1036 | p->addr, ret); | |
297f9233 | 1037 | return ret; |
ae6aa16f | 1038 | } |
e7dbfe34 | 1039 | #else /* !CONFIG_KPROBES_ON_FTRACE */ |
ae6aa16f | 1040 | #define prepare_kprobe(p) arch_prepare_kprobe(p) |
12310e34 | 1041 | #define arm_kprobe_ftrace(p) (-ENODEV) |
297f9233 | 1042 | #define disarm_kprobe_ftrace(p) (-ENODEV) |
ae6aa16f MH |
1043 | #endif |
1044 | ||
201517a7 | 1045 | /* Arm a kprobe with text_mutex */ |
12310e34 | 1046 | static int arm_kprobe(struct kprobe *kp) |
201517a7 | 1047 | { |
12310e34 JY |
1048 | if (unlikely(kprobe_ftrace(kp))) |
1049 | return arm_kprobe_ftrace(kp); | |
1050 | ||
2d1e38f5 | 1051 | cpus_read_lock(); |
201517a7 | 1052 | mutex_lock(&text_mutex); |
afd66255 | 1053 | __arm_kprobe(kp); |
201517a7 | 1054 | mutex_unlock(&text_mutex); |
2d1e38f5 | 1055 | cpus_read_unlock(); |
12310e34 JY |
1056 | |
1057 | return 0; | |
201517a7 MH |
1058 | } |
1059 | ||
1060 | /* Disarm a kprobe with text_mutex */ | |
297f9233 | 1061 | static int disarm_kprobe(struct kprobe *kp, bool reopt) |
201517a7 | 1062 | { |
297f9233 JY |
1063 | if (unlikely(kprobe_ftrace(kp))) |
1064 | return disarm_kprobe_ftrace(kp); | |
2d1e38f5 TG |
1065 | |
1066 | cpus_read_lock(); | |
201517a7 | 1067 | mutex_lock(&text_mutex); |
ae6aa16f | 1068 | __disarm_kprobe(kp, reopt); |
201517a7 | 1069 | mutex_unlock(&text_mutex); |
2d1e38f5 | 1070 | cpus_read_unlock(); |
297f9233 JY |
1071 | |
1072 | return 0; | |
201517a7 MH |
1073 | } |
1074 | ||
64f562c6 AM |
1075 | /* |
1076 | * Aggregate handlers for multiple kprobes support - these handlers | |
1077 | * take care of invoking the individual kprobe handlers on p->list | |
1078 | */ | |
820aede0 | 1079 | static int aggr_pre_handler(struct kprobe *p, struct pt_regs *regs) |
64f562c6 AM |
1080 | { |
1081 | struct kprobe *kp; | |
1082 | ||
3516a460 | 1083 | list_for_each_entry_rcu(kp, &p->list, list) { |
de5bd88d | 1084 | if (kp->pre_handler && likely(!kprobe_disabled(kp))) { |
e6584523 | 1085 | set_kprobe_instance(kp); |
8b0914ea PP |
1086 | if (kp->pre_handler(kp, regs)) |
1087 | return 1; | |
64f562c6 | 1088 | } |
e6584523 | 1089 | reset_kprobe_instance(); |
64f562c6 AM |
1090 | } |
1091 | return 0; | |
1092 | } | |
820aede0 | 1093 | NOKPROBE_SYMBOL(aggr_pre_handler); |
64f562c6 | 1094 | |
820aede0 MH |
1095 | static void aggr_post_handler(struct kprobe *p, struct pt_regs *regs, |
1096 | unsigned long flags) | |
64f562c6 AM |
1097 | { |
1098 | struct kprobe *kp; | |
1099 | ||
3516a460 | 1100 | list_for_each_entry_rcu(kp, &p->list, list) { |
de5bd88d | 1101 | if (kp->post_handler && likely(!kprobe_disabled(kp))) { |
e6584523 | 1102 | set_kprobe_instance(kp); |
64f562c6 | 1103 | kp->post_handler(kp, regs, flags); |
e6584523 | 1104 | reset_kprobe_instance(); |
64f562c6 AM |
1105 | } |
1106 | } | |
64f562c6 | 1107 | } |
820aede0 | 1108 | NOKPROBE_SYMBOL(aggr_post_handler); |
64f562c6 | 1109 | |
820aede0 MH |
1110 | static int aggr_fault_handler(struct kprobe *p, struct pt_regs *regs, |
1111 | int trapnr) | |
64f562c6 | 1112 | { |
b76834bc | 1113 | struct kprobe *cur = __this_cpu_read(kprobe_instance); |
e6584523 | 1114 | |
64f562c6 AM |
1115 | /* |
1116 | * if we faulted "during" the execution of a user specified | |
1117 | * probe handler, invoke just that probe's fault handler | |
1118 | */ | |
e6584523 AM |
1119 | if (cur && cur->fault_handler) { |
1120 | if (cur->fault_handler(cur, regs, trapnr)) | |
64f562c6 AM |
1121 | return 1; |
1122 | } | |
1123 | return 0; | |
1124 | } | |
820aede0 | 1125 | NOKPROBE_SYMBOL(aggr_fault_handler); |
64f562c6 | 1126 | |
bf8d5c52 | 1127 | /* Walks the list and increments nmissed count for multiprobe case */ |
820aede0 | 1128 | void kprobes_inc_nmissed_count(struct kprobe *p) |
bf8d5c52 KA |
1129 | { |
1130 | struct kprobe *kp; | |
afd66255 | 1131 | if (!kprobe_aggrprobe(p)) { |
bf8d5c52 KA |
1132 | p->nmissed++; |
1133 | } else { | |
1134 | list_for_each_entry_rcu(kp, &p->list, list) | |
1135 | kp->nmissed++; | |
1136 | } | |
1137 | return; | |
1138 | } | |
820aede0 | 1139 | NOKPROBE_SYMBOL(kprobes_inc_nmissed_count); |
bf8d5c52 | 1140 | |
820aede0 MH |
1141 | void recycle_rp_inst(struct kretprobe_instance *ri, |
1142 | struct hlist_head *head) | |
b94cce92 | 1143 | { |
ef53d9c5 S |
1144 | struct kretprobe *rp = ri->rp; |
1145 | ||
b94cce92 HN |
1146 | /* remove rp inst off the rprobe_inst_table */ |
1147 | hlist_del(&ri->hlist); | |
ef53d9c5 S |
1148 | INIT_HLIST_NODE(&ri->hlist); |
1149 | if (likely(rp)) { | |
ec484608 | 1150 | raw_spin_lock(&rp->lock); |
ef53d9c5 | 1151 | hlist_add_head(&ri->hlist, &rp->free_instances); |
ec484608 | 1152 | raw_spin_unlock(&rp->lock); |
b94cce92 HN |
1153 | } else |
1154 | /* Unregistering */ | |
99219a3f | 1155 | hlist_add_head(&ri->hlist, head); |
b94cce92 | 1156 | } |
820aede0 | 1157 | NOKPROBE_SYMBOL(recycle_rp_inst); |
b94cce92 | 1158 | |
820aede0 | 1159 | void kretprobe_hash_lock(struct task_struct *tsk, |
ef53d9c5 | 1160 | struct hlist_head **head, unsigned long *flags) |
635c17c2 | 1161 | __acquires(hlist_lock) |
ef53d9c5 S |
1162 | { |
1163 | unsigned long hash = hash_ptr(tsk, KPROBE_HASH_BITS); | |
ec484608 | 1164 | raw_spinlock_t *hlist_lock; |
ef53d9c5 S |
1165 | |
1166 | *head = &kretprobe_inst_table[hash]; | |
1167 | hlist_lock = kretprobe_table_lock_ptr(hash); | |
ec484608 | 1168 | raw_spin_lock_irqsave(hlist_lock, *flags); |
ef53d9c5 | 1169 | } |
820aede0 | 1170 | NOKPROBE_SYMBOL(kretprobe_hash_lock); |
ef53d9c5 | 1171 | |
820aede0 MH |
1172 | static void kretprobe_table_lock(unsigned long hash, |
1173 | unsigned long *flags) | |
635c17c2 | 1174 | __acquires(hlist_lock) |
b94cce92 | 1175 | { |
ec484608 TG |
1176 | raw_spinlock_t *hlist_lock = kretprobe_table_lock_ptr(hash); |
1177 | raw_spin_lock_irqsave(hlist_lock, *flags); | |
ef53d9c5 | 1178 | } |
820aede0 | 1179 | NOKPROBE_SYMBOL(kretprobe_table_lock); |
ef53d9c5 | 1180 | |
820aede0 MH |
1181 | void kretprobe_hash_unlock(struct task_struct *tsk, |
1182 | unsigned long *flags) | |
635c17c2 | 1183 | __releases(hlist_lock) |
ef53d9c5 S |
1184 | { |
1185 | unsigned long hash = hash_ptr(tsk, KPROBE_HASH_BITS); | |
ec484608 | 1186 | raw_spinlock_t *hlist_lock; |
ef53d9c5 S |
1187 | |
1188 | hlist_lock = kretprobe_table_lock_ptr(hash); | |
ec484608 | 1189 | raw_spin_unlock_irqrestore(hlist_lock, *flags); |
ef53d9c5 | 1190 | } |
820aede0 | 1191 | NOKPROBE_SYMBOL(kretprobe_hash_unlock); |
ef53d9c5 | 1192 | |
820aede0 MH |
1193 | static void kretprobe_table_unlock(unsigned long hash, |
1194 | unsigned long *flags) | |
635c17c2 | 1195 | __releases(hlist_lock) |
ef53d9c5 | 1196 | { |
ec484608 TG |
1197 | raw_spinlock_t *hlist_lock = kretprobe_table_lock_ptr(hash); |
1198 | raw_spin_unlock_irqrestore(hlist_lock, *flags); | |
b94cce92 | 1199 | } |
820aede0 | 1200 | NOKPROBE_SYMBOL(kretprobe_table_unlock); |
b94cce92 | 1201 | |
b94cce92 | 1202 | /* |
c6fd91f0 | 1203 | * This function is called from finish_task_switch when task tk becomes dead, |
1204 | * so that we can recycle any function-return probe instances associated | |
1205 | * with this task. These left over instances represent probed functions | |
1206 | * that have been called but will never return. | |
b94cce92 | 1207 | */ |
820aede0 | 1208 | void kprobe_flush_task(struct task_struct *tk) |
b94cce92 | 1209 | { |
62c27be0 | 1210 | struct kretprobe_instance *ri; |
99219a3f | 1211 | struct hlist_head *head, empty_rp; |
b67bfe0d | 1212 | struct hlist_node *tmp; |
ef53d9c5 | 1213 | unsigned long hash, flags = 0; |
802eae7c | 1214 | |
ef53d9c5 S |
1215 | if (unlikely(!kprobes_initialized)) |
1216 | /* Early boot. kretprobe_table_locks not yet initialized. */ | |
1217 | return; | |
1218 | ||
d496aab5 | 1219 | INIT_HLIST_HEAD(&empty_rp); |
ef53d9c5 S |
1220 | hash = hash_ptr(tk, KPROBE_HASH_BITS); |
1221 | head = &kretprobe_inst_table[hash]; | |
1222 | kretprobe_table_lock(hash, &flags); | |
b67bfe0d | 1223 | hlist_for_each_entry_safe(ri, tmp, head, hlist) { |
62c27be0 | 1224 | if (ri->task == tk) |
99219a3f | 1225 | recycle_rp_inst(ri, &empty_rp); |
62c27be0 | 1226 | } |
ef53d9c5 | 1227 | kretprobe_table_unlock(hash, &flags); |
b67bfe0d | 1228 | hlist_for_each_entry_safe(ri, tmp, &empty_rp, hlist) { |
99219a3f | 1229 | hlist_del(&ri->hlist); |
1230 | kfree(ri); | |
1231 | } | |
b94cce92 | 1232 | } |
820aede0 | 1233 | NOKPROBE_SYMBOL(kprobe_flush_task); |
b94cce92 | 1234 | |
b94cce92 HN |
1235 | static inline void free_rp_inst(struct kretprobe *rp) |
1236 | { | |
1237 | struct kretprobe_instance *ri; | |
b67bfe0d | 1238 | struct hlist_node *next; |
4c4308cb | 1239 | |
b67bfe0d | 1240 | hlist_for_each_entry_safe(ri, next, &rp->free_instances, hlist) { |
ef53d9c5 | 1241 | hlist_del(&ri->hlist); |
b94cce92 HN |
1242 | kfree(ri); |
1243 | } | |
1244 | } | |
1245 | ||
820aede0 | 1246 | static void cleanup_rp_inst(struct kretprobe *rp) |
4a296e07 | 1247 | { |
ef53d9c5 | 1248 | unsigned long flags, hash; |
4a296e07 | 1249 | struct kretprobe_instance *ri; |
b67bfe0d | 1250 | struct hlist_node *next; |
ef53d9c5 S |
1251 | struct hlist_head *head; |
1252 | ||
4a296e07 | 1253 | /* No race here */ |
ef53d9c5 S |
1254 | for (hash = 0; hash < KPROBE_TABLE_SIZE; hash++) { |
1255 | kretprobe_table_lock(hash, &flags); | |
1256 | head = &kretprobe_inst_table[hash]; | |
b67bfe0d | 1257 | hlist_for_each_entry_safe(ri, next, head, hlist) { |
ef53d9c5 S |
1258 | if (ri->rp == rp) |
1259 | ri->rp = NULL; | |
1260 | } | |
1261 | kretprobe_table_unlock(hash, &flags); | |
4a296e07 | 1262 | } |
4a296e07 MH |
1263 | free_rp_inst(rp); |
1264 | } | |
820aede0 | 1265 | NOKPROBE_SYMBOL(cleanup_rp_inst); |
4a296e07 | 1266 | |
059053a2 | 1267 | /* Add the new probe to ap->list */ |
55479f64 | 1268 | static int add_new_kprobe(struct kprobe *ap, struct kprobe *p) |
8b0914ea | 1269 | { |
de5bd88d | 1270 | BUG_ON(kprobe_gone(ap) || kprobe_gone(p)); |
afd66255 | 1271 | |
059053a2 | 1272 | if (p->post_handler) |
6274de49 | 1273 | unoptimize_kprobe(ap, true); /* Fall back to normal kprobe */ |
afd66255 | 1274 | |
059053a2 | 1275 | list_add_rcu(&p->list, &ap->list); |
b918e5e6 MH |
1276 | if (p->post_handler && !ap->post_handler) |
1277 | ap->post_handler = aggr_post_handler; | |
de5bd88d | 1278 | |
8b0914ea PP |
1279 | return 0; |
1280 | } | |
1281 | ||
64f562c6 AM |
1282 | /* |
1283 | * Fill in the required fields of the "manager kprobe". Replace the | |
1284 | * earlier kprobe in the hlist with the manager kprobe | |
1285 | */ | |
55479f64 | 1286 | static void init_aggr_kprobe(struct kprobe *ap, struct kprobe *p) |
64f562c6 | 1287 | { |
afd66255 | 1288 | /* Copy p's insn slot to ap */ |
8b0914ea | 1289 | copy_kprobe(p, ap); |
a9ad965e | 1290 | flush_insn_slot(ap); |
64f562c6 | 1291 | ap->addr = p->addr; |
afd66255 | 1292 | ap->flags = p->flags & ~KPROBE_FLAG_OPTIMIZED; |
64f562c6 | 1293 | ap->pre_handler = aggr_pre_handler; |
64f562c6 | 1294 | ap->fault_handler = aggr_fault_handler; |
e8386a0c MH |
1295 | /* We don't care the kprobe which has gone. */ |
1296 | if (p->post_handler && !kprobe_gone(p)) | |
36721656 | 1297 | ap->post_handler = aggr_post_handler; |
64f562c6 AM |
1298 | |
1299 | INIT_LIST_HEAD(&ap->list); | |
afd66255 | 1300 | INIT_HLIST_NODE(&ap->hlist); |
64f562c6 | 1301 | |
afd66255 | 1302 | list_add_rcu(&p->list, &ap->list); |
adad0f33 | 1303 | hlist_replace_rcu(&p->hlist, &ap->hlist); |
64f562c6 AM |
1304 | } |
1305 | ||
1306 | /* | |
1307 | * This is the second or subsequent kprobe at the address - handle | |
1308 | * the intricacies | |
64f562c6 | 1309 | */ |
55479f64 | 1310 | static int register_aggr_kprobe(struct kprobe *orig_p, struct kprobe *p) |
64f562c6 AM |
1311 | { |
1312 | int ret = 0; | |
6d8e40a8 | 1313 | struct kprobe *ap = orig_p; |
64f562c6 | 1314 | |
2d1e38f5 TG |
1315 | cpus_read_lock(); |
1316 | ||
25764288 MH |
1317 | /* For preparing optimization, jump_label_text_reserved() is called */ |
1318 | jump_label_lock(); | |
25764288 MH |
1319 | mutex_lock(&text_mutex); |
1320 | ||
6d8e40a8 MH |
1321 | if (!kprobe_aggrprobe(orig_p)) { |
1322 | /* If orig_p is not an aggr_kprobe, create new aggr_kprobe. */ | |
1323 | ap = alloc_aggr_kprobe(orig_p); | |
25764288 MH |
1324 | if (!ap) { |
1325 | ret = -ENOMEM; | |
1326 | goto out; | |
1327 | } | |
6d8e40a8 | 1328 | init_aggr_kprobe(ap, orig_p); |
82b0f70f | 1329 | } else if (kprobe_unused(ap)) { |
0490cd1f | 1330 | /* This probe is going to die. Rescue it */ |
82b0f70f MH |
1331 | ret = reuse_unused_kprobe(ap); |
1332 | if (ret) | |
1333 | goto out; | |
1334 | } | |
b918e5e6 MH |
1335 | |
1336 | if (kprobe_gone(ap)) { | |
e8386a0c MH |
1337 | /* |
1338 | * Attempting to insert new probe at the same location that | |
1339 | * had a probe in the module vaddr area which already | |
1340 | * freed. So, the instruction slot has already been | |
1341 | * released. We need a new slot for the new probe. | |
1342 | */ | |
b918e5e6 | 1343 | ret = arch_prepare_kprobe(ap); |
e8386a0c | 1344 | if (ret) |
b918e5e6 MH |
1345 | /* |
1346 | * Even if fail to allocate new slot, don't need to | |
1347 | * free aggr_probe. It will be used next time, or | |
1348 | * freed by unregister_kprobe. | |
1349 | */ | |
25764288 | 1350 | goto out; |
de5bd88d | 1351 | |
afd66255 MH |
1352 | /* Prepare optimized instructions if possible. */ |
1353 | prepare_optimized_kprobe(ap); | |
1354 | ||
e8386a0c | 1355 | /* |
de5bd88d MH |
1356 | * Clear gone flag to prevent allocating new slot again, and |
1357 | * set disabled flag because it is not armed yet. | |
e8386a0c | 1358 | */ |
de5bd88d MH |
1359 | ap->flags = (ap->flags & ~KPROBE_FLAG_GONE) |
1360 | | KPROBE_FLAG_DISABLED; | |
e8386a0c | 1361 | } |
b918e5e6 | 1362 | |
afd66255 | 1363 | /* Copy ap's insn slot to p */ |
b918e5e6 | 1364 | copy_kprobe(ap, p); |
25764288 MH |
1365 | ret = add_new_kprobe(ap, p); |
1366 | ||
1367 | out: | |
1368 | mutex_unlock(&text_mutex); | |
25764288 | 1369 | jump_label_unlock(); |
2d1e38f5 | 1370 | cpus_read_unlock(); |
25764288 MH |
1371 | |
1372 | if (ret == 0 && kprobe_disabled(ap) && !kprobe_disabled(p)) { | |
1373 | ap->flags &= ~KPROBE_FLAG_DISABLED; | |
12310e34 | 1374 | if (!kprobes_all_disarmed) { |
25764288 | 1375 | /* Arm the breakpoint again. */ |
12310e34 JY |
1376 | ret = arm_kprobe(ap); |
1377 | if (ret) { | |
1378 | ap->flags |= KPROBE_FLAG_DISABLED; | |
1379 | list_del_rcu(&p->list); | |
1380 | synchronize_sched(); | |
1381 | } | |
1382 | } | |
25764288 MH |
1383 | } |
1384 | return ret; | |
64f562c6 AM |
1385 | } |
1386 | ||
be8f2743 MH |
1387 | bool __weak arch_within_kprobe_blacklist(unsigned long addr) |
1388 | { | |
1389 | /* The __kprobes marked functions and entry code must not be probed */ | |
1390 | return addr >= (unsigned long)__kprobes_text_start && | |
1391 | addr < (unsigned long)__kprobes_text_end; | |
1392 | } | |
1393 | ||
e5779e8e | 1394 | bool within_kprobe_blacklist(unsigned long addr) |
d0aaff97 | 1395 | { |
376e2424 | 1396 | struct kprobe_blacklist_entry *ent; |
3d8d996e | 1397 | |
be8f2743 | 1398 | if (arch_within_kprobe_blacklist(addr)) |
376e2424 | 1399 | return true; |
3d8d996e SD |
1400 | /* |
1401 | * If there exists a kprobe_blacklist, verify and | |
1402 | * fail any probe registration in the prohibited area | |
1403 | */ | |
376e2424 MH |
1404 | list_for_each_entry(ent, &kprobe_blacklist, list) { |
1405 | if (addr >= ent->start_addr && addr < ent->end_addr) | |
1406 | return true; | |
3d8d996e | 1407 | } |
376e2424 MH |
1408 | |
1409 | return false; | |
d0aaff97 PP |
1410 | } |
1411 | ||
b2a5cd69 MH |
1412 | /* |
1413 | * If we have a symbol_name argument, look it up and add the offset field | |
1414 | * to it. This way, we can specify a relative address to a symbol. | |
bc81d48d MH |
1415 | * This returns encoded errors if it fails to look up symbol or invalid |
1416 | * combination of parameters. | |
b2a5cd69 | 1417 | */ |
1d585e70 NR |
1418 | static kprobe_opcode_t *_kprobe_addr(kprobe_opcode_t *addr, |
1419 | const char *symbol_name, unsigned int offset) | |
b2a5cd69 | 1420 | { |
1d585e70 | 1421 | if ((symbol_name && addr) || (!symbol_name && !addr)) |
bc81d48d MH |
1422 | goto invalid; |
1423 | ||
1d585e70 | 1424 | if (symbol_name) { |
7246f600 | 1425 | addr = kprobe_lookup_name(symbol_name, offset); |
bc81d48d MH |
1426 | if (!addr) |
1427 | return ERR_PTR(-ENOENT); | |
b2a5cd69 MH |
1428 | } |
1429 | ||
1d585e70 | 1430 | addr = (kprobe_opcode_t *)(((char *)addr) + offset); |
bc81d48d MH |
1431 | if (addr) |
1432 | return addr; | |
1433 | ||
1434 | invalid: | |
1435 | return ERR_PTR(-EINVAL); | |
b2a5cd69 MH |
1436 | } |
1437 | ||
1d585e70 NR |
1438 | static kprobe_opcode_t *kprobe_addr(struct kprobe *p) |
1439 | { | |
1440 | return _kprobe_addr(p->addr, p->symbol_name, p->offset); | |
1441 | } | |
1442 | ||
1f0ab409 | 1443 | /* Check passed kprobe is valid and return kprobe in kprobe_table. */ |
55479f64 | 1444 | static struct kprobe *__get_valid_kprobe(struct kprobe *p) |
1f0ab409 | 1445 | { |
6d8e40a8 | 1446 | struct kprobe *ap, *list_p; |
1f0ab409 | 1447 | |
6d8e40a8 MH |
1448 | ap = get_kprobe(p->addr); |
1449 | if (unlikely(!ap)) | |
1f0ab409 AM |
1450 | return NULL; |
1451 | ||
6d8e40a8 MH |
1452 | if (p != ap) { |
1453 | list_for_each_entry_rcu(list_p, &ap->list, list) | |
1f0ab409 AM |
1454 | if (list_p == p) |
1455 | /* kprobe p is a valid probe */ | |
1456 | goto valid; | |
1457 | return NULL; | |
1458 | } | |
1459 | valid: | |
6d8e40a8 | 1460 | return ap; |
1f0ab409 AM |
1461 | } |
1462 | ||
1463 | /* Return error if the kprobe is being re-registered */ | |
1464 | static inline int check_kprobe_rereg(struct kprobe *p) | |
1465 | { | |
1466 | int ret = 0; | |
1f0ab409 AM |
1467 | |
1468 | mutex_lock(&kprobe_mutex); | |
6d8e40a8 | 1469 | if (__get_valid_kprobe(p)) |
1f0ab409 AM |
1470 | ret = -EINVAL; |
1471 | mutex_unlock(&kprobe_mutex); | |
6d8e40a8 | 1472 | |
1f0ab409 AM |
1473 | return ret; |
1474 | } | |
1475 | ||
f7f242ff | 1476 | int __weak arch_check_ftrace_location(struct kprobe *p) |
1da177e4 | 1477 | { |
ae6aa16f MH |
1478 | unsigned long ftrace_addr; |
1479 | ||
ae6aa16f MH |
1480 | ftrace_addr = ftrace_location((unsigned long)p->addr); |
1481 | if (ftrace_addr) { | |
e7dbfe34 | 1482 | #ifdef CONFIG_KPROBES_ON_FTRACE |
ae6aa16f MH |
1483 | /* Given address is not on the instruction boundary */ |
1484 | if ((unsigned long)p->addr != ftrace_addr) | |
1485 | return -EILSEQ; | |
ae6aa16f | 1486 | p->flags |= KPROBE_FLAG_FTRACE; |
e7dbfe34 | 1487 | #else /* !CONFIG_KPROBES_ON_FTRACE */ |
ae6aa16f MH |
1488 | return -EINVAL; |
1489 | #endif | |
1490 | } | |
f7f242ff HC |
1491 | return 0; |
1492 | } | |
1493 | ||
1494 | static int check_kprobe_address_safe(struct kprobe *p, | |
1495 | struct module **probed_mod) | |
1496 | { | |
1497 | int ret; | |
1f0ab409 | 1498 | |
f7f242ff HC |
1499 | ret = arch_check_ftrace_location(p); |
1500 | if (ret) | |
1501 | return ret; | |
91bad2f8 | 1502 | jump_label_lock(); |
de31c3ca | 1503 | preempt_disable(); |
f7fa6ef0 MH |
1504 | |
1505 | /* Ensure it is not in reserved area nor out of text */ | |
ec30c5f3 | 1506 | if (!kernel_text_address((unsigned long) p->addr) || |
376e2424 | 1507 | within_kprobe_blacklist((unsigned long) p->addr) || |
f986a499 PN |
1508 | jump_label_text_reserved(p->addr, p->addr)) { |
1509 | ret = -EINVAL; | |
f7fa6ef0 | 1510 | goto out; |
f986a499 | 1511 | } |
b3e55c72 | 1512 | |
f7fa6ef0 MH |
1513 | /* Check if are we probing a module */ |
1514 | *probed_mod = __module_text_address((unsigned long) p->addr); | |
1515 | if (*probed_mod) { | |
6f716acd | 1516 | /* |
e8386a0c MH |
1517 | * We must hold a refcount of the probed module while updating |
1518 | * its code to prohibit unexpected unloading. | |
df019b1d | 1519 | */ |
f7fa6ef0 MH |
1520 | if (unlikely(!try_module_get(*probed_mod))) { |
1521 | ret = -ENOENT; | |
1522 | goto out; | |
1523 | } | |
de31c3ca | 1524 | |
f24659d9 MH |
1525 | /* |
1526 | * If the module freed .init.text, we couldn't insert | |
1527 | * kprobes in there. | |
1528 | */ | |
f7fa6ef0 MH |
1529 | if (within_module_init((unsigned long)p->addr, *probed_mod) && |
1530 | (*probed_mod)->state != MODULE_STATE_COMING) { | |
1531 | module_put(*probed_mod); | |
1532 | *probed_mod = NULL; | |
1533 | ret = -ENOENT; | |
f24659d9 | 1534 | } |
df019b1d | 1535 | } |
f7fa6ef0 | 1536 | out: |
a189d035 | 1537 | preempt_enable(); |
de31c3ca | 1538 | jump_label_unlock(); |
1da177e4 | 1539 | |
f7fa6ef0 MH |
1540 | return ret; |
1541 | } | |
1542 | ||
55479f64 | 1543 | int register_kprobe(struct kprobe *p) |
f7fa6ef0 MH |
1544 | { |
1545 | int ret; | |
1546 | struct kprobe *old_p; | |
1547 | struct module *probed_mod; | |
1548 | kprobe_opcode_t *addr; | |
1549 | ||
1550 | /* Adjust probe address from symbol */ | |
1551 | addr = kprobe_addr(p); | |
1552 | if (IS_ERR(addr)) | |
1553 | return PTR_ERR(addr); | |
1554 | p->addr = addr; | |
1555 | ||
1556 | ret = check_kprobe_rereg(p); | |
1557 | if (ret) | |
1558 | return ret; | |
1559 | ||
1560 | /* User can pass only KPROBE_FLAG_DISABLED to register_kprobe */ | |
1561 | p->flags &= KPROBE_FLAG_DISABLED; | |
3516a460 | 1562 | p->nmissed = 0; |
9861668f | 1563 | INIT_LIST_HEAD(&p->list); |
afd66255 | 1564 | |
f7fa6ef0 MH |
1565 | ret = check_kprobe_address_safe(p, &probed_mod); |
1566 | if (ret) | |
1567 | return ret; | |
1568 | ||
1569 | mutex_lock(&kprobe_mutex); | |
afd66255 | 1570 | |
64f562c6 AM |
1571 | old_p = get_kprobe(p->addr); |
1572 | if (old_p) { | |
afd66255 | 1573 | /* Since this may unoptimize old_p, locking text_mutex. */ |
64f562c6 | 1574 | ret = register_aggr_kprobe(old_p, p); |
1da177e4 LT |
1575 | goto out; |
1576 | } | |
1da177e4 | 1577 | |
2d1e38f5 TG |
1578 | cpus_read_lock(); |
1579 | /* Prevent text modification */ | |
1580 | mutex_lock(&text_mutex); | |
ae6aa16f | 1581 | ret = prepare_kprobe(p); |
25764288 | 1582 | mutex_unlock(&text_mutex); |
2d1e38f5 | 1583 | cpus_read_unlock(); |
6f716acd | 1584 | if (ret) |
afd66255 | 1585 | goto out; |
49a2a1b8 | 1586 | |
64f562c6 | 1587 | INIT_HLIST_NODE(&p->hlist); |
3516a460 | 1588 | hlist_add_head_rcu(&p->hlist, |
1da177e4 LT |
1589 | &kprobe_table[hash_ptr(p->addr, KPROBE_HASH_BITS)]); |
1590 | ||
12310e34 JY |
1591 | if (!kprobes_all_disarmed && !kprobe_disabled(p)) { |
1592 | ret = arm_kprobe(p); | |
1593 | if (ret) { | |
1594 | hlist_del_rcu(&p->hlist); | |
1595 | synchronize_sched(); | |
1596 | goto out; | |
1597 | } | |
1598 | } | |
afd66255 MH |
1599 | |
1600 | /* Try to optimize kprobe */ | |
1601 | try_to_optimize_kprobe(p); | |
1da177e4 | 1602 | out: |
7a7d1cf9 | 1603 | mutex_unlock(&kprobe_mutex); |
49a2a1b8 | 1604 | |
e8386a0c | 1605 | if (probed_mod) |
df019b1d | 1606 | module_put(probed_mod); |
e8386a0c | 1607 | |
1da177e4 LT |
1608 | return ret; |
1609 | } | |
99081ab5 | 1610 | EXPORT_SYMBOL_GPL(register_kprobe); |
1da177e4 | 1611 | |
6f0f1dd7 | 1612 | /* Check if all probes on the aggrprobe are disabled */ |
55479f64 | 1613 | static int aggr_kprobe_disabled(struct kprobe *ap) |
6f0f1dd7 MH |
1614 | { |
1615 | struct kprobe *kp; | |
1616 | ||
1617 | list_for_each_entry_rcu(kp, &ap->list, list) | |
1618 | if (!kprobe_disabled(kp)) | |
1619 | /* | |
1620 | * There is an active probe on the list. | |
1621 | * We can't disable this ap. | |
1622 | */ | |
1623 | return 0; | |
1624 | ||
1625 | return 1; | |
1626 | } | |
1627 | ||
1628 | /* Disable one kprobe: Make sure called under kprobe_mutex is locked */ | |
55479f64 | 1629 | static struct kprobe *__disable_kprobe(struct kprobe *p) |
6f0f1dd7 MH |
1630 | { |
1631 | struct kprobe *orig_p; | |
297f9233 | 1632 | int ret; |
6f0f1dd7 MH |
1633 | |
1634 | /* Get an original kprobe for return */ | |
1635 | orig_p = __get_valid_kprobe(p); | |
1636 | if (unlikely(orig_p == NULL)) | |
297f9233 | 1637 | return ERR_PTR(-EINVAL); |
6f0f1dd7 MH |
1638 | |
1639 | if (!kprobe_disabled(p)) { | |
1640 | /* Disable probe if it is a child probe */ | |
1641 | if (p != orig_p) | |
1642 | p->flags |= KPROBE_FLAG_DISABLED; | |
1643 | ||
1644 | /* Try to disarm and disable this/parent probe */ | |
1645 | if (p == orig_p || aggr_kprobe_disabled(orig_p)) { | |
69d54b91 WN |
1646 | /* |
1647 | * If kprobes_all_disarmed is set, orig_p | |
1648 | * should have already been disarmed, so | |
1649 | * skip unneed disarming process. | |
1650 | */ | |
297f9233 JY |
1651 | if (!kprobes_all_disarmed) { |
1652 | ret = disarm_kprobe(orig_p, true); | |
1653 | if (ret) { | |
1654 | p->flags &= ~KPROBE_FLAG_DISABLED; | |
1655 | return ERR_PTR(ret); | |
1656 | } | |
1657 | } | |
6f0f1dd7 MH |
1658 | orig_p->flags |= KPROBE_FLAG_DISABLED; |
1659 | } | |
1660 | } | |
1661 | ||
1662 | return orig_p; | |
1663 | } | |
1664 | ||
de5bd88d MH |
1665 | /* |
1666 | * Unregister a kprobe without a scheduler synchronization. | |
1667 | */ | |
55479f64 | 1668 | static int __unregister_kprobe_top(struct kprobe *p) |
de5bd88d | 1669 | { |
6d8e40a8 | 1670 | struct kprobe *ap, *list_p; |
de5bd88d | 1671 | |
6f0f1dd7 MH |
1672 | /* Disable kprobe. This will disarm it if needed. */ |
1673 | ap = __disable_kprobe(p); | |
297f9233 JY |
1674 | if (IS_ERR(ap)) |
1675 | return PTR_ERR(ap); | |
de5bd88d | 1676 | |
6f0f1dd7 | 1677 | if (ap == p) |
bf8f6e5b | 1678 | /* |
6f0f1dd7 MH |
1679 | * This probe is an independent(and non-optimized) kprobe |
1680 | * (not an aggrprobe). Remove from the hash list. | |
bf8f6e5b | 1681 | */ |
6f0f1dd7 MH |
1682 | goto disarmed; |
1683 | ||
1684 | /* Following process expects this probe is an aggrprobe */ | |
1685 | WARN_ON(!kprobe_aggrprobe(ap)); | |
1686 | ||
6274de49 MH |
1687 | if (list_is_singular(&ap->list) && kprobe_disarmed(ap)) |
1688 | /* | |
1689 | * !disarmed could be happen if the probe is under delayed | |
1690 | * unoptimizing. | |
1691 | */ | |
6f0f1dd7 MH |
1692 | goto disarmed; |
1693 | else { | |
1694 | /* If disabling probe has special handlers, update aggrprobe */ | |
e8386a0c | 1695 | if (p->post_handler && !kprobe_gone(p)) { |
6d8e40a8 | 1696 | list_for_each_entry_rcu(list_p, &ap->list, list) { |
9861668f MH |
1697 | if ((list_p != p) && (list_p->post_handler)) |
1698 | goto noclean; | |
1699 | } | |
6d8e40a8 | 1700 | ap->post_handler = NULL; |
9861668f MH |
1701 | } |
1702 | noclean: | |
6f0f1dd7 MH |
1703 | /* |
1704 | * Remove from the aggrprobe: this path will do nothing in | |
1705 | * __unregister_kprobe_bottom(). | |
1706 | */ | |
49a2a1b8 | 1707 | list_del_rcu(&p->list); |
6f0f1dd7 MH |
1708 | if (!kprobe_disabled(ap) && !kprobes_all_disarmed) |
1709 | /* | |
1710 | * Try to optimize this probe again, because post | |
1711 | * handler may have been changed. | |
1712 | */ | |
1713 | optimize_kprobe(ap); | |
49a2a1b8 | 1714 | } |
9861668f | 1715 | return 0; |
6f0f1dd7 MH |
1716 | |
1717 | disarmed: | |
6274de49 | 1718 | BUG_ON(!kprobe_disarmed(ap)); |
6f0f1dd7 MH |
1719 | hlist_del_rcu(&ap->hlist); |
1720 | return 0; | |
9861668f | 1721 | } |
3516a460 | 1722 | |
55479f64 | 1723 | static void __unregister_kprobe_bottom(struct kprobe *p) |
9861668f | 1724 | { |
6d8e40a8 | 1725 | struct kprobe *ap; |
b3e55c72 | 1726 | |
e8386a0c | 1727 | if (list_empty(&p->list)) |
6274de49 | 1728 | /* This is an independent kprobe */ |
0498b635 | 1729 | arch_remove_kprobe(p); |
e8386a0c | 1730 | else if (list_is_singular(&p->list)) { |
6274de49 | 1731 | /* This is the last child of an aggrprobe */ |
6d8e40a8 | 1732 | ap = list_entry(p->list.next, struct kprobe, list); |
e8386a0c | 1733 | list_del(&p->list); |
6d8e40a8 | 1734 | free_aggr_kprobe(ap); |
9861668f | 1735 | } |
6274de49 | 1736 | /* Otherwise, do nothing. */ |
9861668f MH |
1737 | } |
1738 | ||
55479f64 | 1739 | int register_kprobes(struct kprobe **kps, int num) |
9861668f MH |
1740 | { |
1741 | int i, ret = 0; | |
1742 | ||
1743 | if (num <= 0) | |
1744 | return -EINVAL; | |
1745 | for (i = 0; i < num; i++) { | |
49ad2fd7 | 1746 | ret = register_kprobe(kps[i]); |
67dddaad MH |
1747 | if (ret < 0) { |
1748 | if (i > 0) | |
1749 | unregister_kprobes(kps, i); | |
9861668f | 1750 | break; |
36721656 | 1751 | } |
49a2a1b8 | 1752 | } |
9861668f MH |
1753 | return ret; |
1754 | } | |
99081ab5 | 1755 | EXPORT_SYMBOL_GPL(register_kprobes); |
9861668f | 1756 | |
55479f64 | 1757 | void unregister_kprobe(struct kprobe *p) |
9861668f MH |
1758 | { |
1759 | unregister_kprobes(&p, 1); | |
1760 | } | |
99081ab5 | 1761 | EXPORT_SYMBOL_GPL(unregister_kprobe); |
9861668f | 1762 | |
55479f64 | 1763 | void unregister_kprobes(struct kprobe **kps, int num) |
9861668f MH |
1764 | { |
1765 | int i; | |
1766 | ||
1767 | if (num <= 0) | |
1768 | return; | |
1769 | mutex_lock(&kprobe_mutex); | |
1770 | for (i = 0; i < num; i++) | |
1771 | if (__unregister_kprobe_top(kps[i]) < 0) | |
1772 | kps[i]->addr = NULL; | |
1773 | mutex_unlock(&kprobe_mutex); | |
1774 | ||
1775 | synchronize_sched(); | |
1776 | for (i = 0; i < num; i++) | |
1777 | if (kps[i]->addr) | |
1778 | __unregister_kprobe_bottom(kps[i]); | |
1da177e4 | 1779 | } |
99081ab5 | 1780 | EXPORT_SYMBOL_GPL(unregister_kprobes); |
1da177e4 | 1781 | |
5f6bee34 NR |
1782 | int __weak kprobe_exceptions_notify(struct notifier_block *self, |
1783 | unsigned long val, void *data) | |
fc62d020 NR |
1784 | { |
1785 | return NOTIFY_DONE; | |
1786 | } | |
5f6bee34 | 1787 | NOKPROBE_SYMBOL(kprobe_exceptions_notify); |
fc62d020 | 1788 | |
1da177e4 | 1789 | static struct notifier_block kprobe_exceptions_nb = { |
3d5631e0 AK |
1790 | .notifier_call = kprobe_exceptions_notify, |
1791 | .priority = 0x7fffffff /* we need to be notified first */ | |
1792 | }; | |
1793 | ||
3d7e3382 ME |
1794 | unsigned long __weak arch_deref_entry_point(void *entry) |
1795 | { | |
1796 | return (unsigned long)entry; | |
1797 | } | |
1da177e4 | 1798 | |
9edddaa2 | 1799 | #ifdef CONFIG_KRETPROBES |
e65cefe8 AB |
1800 | /* |
1801 | * This kprobe pre_handler is registered with every kretprobe. When probe | |
1802 | * hits it will set up the return probe. | |
1803 | */ | |
820aede0 | 1804 | static int pre_handler_kretprobe(struct kprobe *p, struct pt_regs *regs) |
e65cefe8 AB |
1805 | { |
1806 | struct kretprobe *rp = container_of(p, struct kretprobe, kp); | |
ef53d9c5 S |
1807 | unsigned long hash, flags = 0; |
1808 | struct kretprobe_instance *ri; | |
e65cefe8 | 1809 | |
f96f5678 MH |
1810 | /* |
1811 | * To avoid deadlocks, prohibit return probing in NMI contexts, | |
1812 | * just skip the probe and increase the (inexact) 'nmissed' | |
1813 | * statistical counter, so that the user is informed that | |
1814 | * something happened: | |
1815 | */ | |
1816 | if (unlikely(in_nmi())) { | |
1817 | rp->nmissed++; | |
1818 | return 0; | |
1819 | } | |
1820 | ||
1821 | /* TODO: consider to only swap the RA after the last pre_handler fired */ | |
ef53d9c5 | 1822 | hash = hash_ptr(current, KPROBE_HASH_BITS); |
ec484608 | 1823 | raw_spin_lock_irqsave(&rp->lock, flags); |
4c4308cb | 1824 | if (!hlist_empty(&rp->free_instances)) { |
4c4308cb | 1825 | ri = hlist_entry(rp->free_instances.first, |
ef53d9c5 S |
1826 | struct kretprobe_instance, hlist); |
1827 | hlist_del(&ri->hlist); | |
ec484608 | 1828 | raw_spin_unlock_irqrestore(&rp->lock, flags); |
ef53d9c5 | 1829 | |
4c4308cb CH |
1830 | ri->rp = rp; |
1831 | ri->task = current; | |
f47cd9b5 | 1832 | |
55ca6140 JL |
1833 | if (rp->entry_handler && rp->entry_handler(ri, regs)) { |
1834 | raw_spin_lock_irqsave(&rp->lock, flags); | |
1835 | hlist_add_head(&ri->hlist, &rp->free_instances); | |
1836 | raw_spin_unlock_irqrestore(&rp->lock, flags); | |
f47cd9b5 | 1837 | return 0; |
55ca6140 | 1838 | } |
f47cd9b5 | 1839 | |
4c4308cb CH |
1840 | arch_prepare_kretprobe(ri, regs); |
1841 | ||
1842 | /* XXX(hch): why is there no hlist_move_head? */ | |
ef53d9c5 S |
1843 | INIT_HLIST_NODE(&ri->hlist); |
1844 | kretprobe_table_lock(hash, &flags); | |
1845 | hlist_add_head(&ri->hlist, &kretprobe_inst_table[hash]); | |
1846 | kretprobe_table_unlock(hash, &flags); | |
1847 | } else { | |
4c4308cb | 1848 | rp->nmissed++; |
ec484608 | 1849 | raw_spin_unlock_irqrestore(&rp->lock, flags); |
ef53d9c5 | 1850 | } |
e65cefe8 AB |
1851 | return 0; |
1852 | } | |
820aede0 | 1853 | NOKPROBE_SYMBOL(pre_handler_kretprobe); |
e65cefe8 | 1854 | |
659b957f | 1855 | bool __weak arch_kprobe_on_func_entry(unsigned long offset) |
90ec5e89 NR |
1856 | { |
1857 | return !offset; | |
1858 | } | |
1859 | ||
659b957f | 1860 | bool kprobe_on_func_entry(kprobe_opcode_t *addr, const char *sym, unsigned long offset) |
1d585e70 NR |
1861 | { |
1862 | kprobe_opcode_t *kp_addr = _kprobe_addr(addr, sym, offset); | |
1863 | ||
1864 | if (IS_ERR(kp_addr)) | |
1865 | return false; | |
1866 | ||
1867 | if (!kallsyms_lookup_size_offset((unsigned long)kp_addr, NULL, &offset) || | |
659b957f | 1868 | !arch_kprobe_on_func_entry(offset)) |
1d585e70 NR |
1869 | return false; |
1870 | ||
1871 | return true; | |
1872 | } | |
1873 | ||
55479f64 | 1874 | int register_kretprobe(struct kretprobe *rp) |
b94cce92 HN |
1875 | { |
1876 | int ret = 0; | |
1877 | struct kretprobe_instance *inst; | |
1878 | int i; | |
b2a5cd69 | 1879 | void *addr; |
90ec5e89 | 1880 | |
659b957f | 1881 | if (!kprobe_on_func_entry(rp->kp.addr, rp->kp.symbol_name, rp->kp.offset)) |
90ec5e89 | 1882 | return -EINVAL; |
f438d914 MH |
1883 | |
1884 | if (kretprobe_blacklist_size) { | |
b2a5cd69 | 1885 | addr = kprobe_addr(&rp->kp); |
bc81d48d MH |
1886 | if (IS_ERR(addr)) |
1887 | return PTR_ERR(addr); | |
f438d914 MH |
1888 | |
1889 | for (i = 0; kretprobe_blacklist[i].name != NULL; i++) { | |
1890 | if (kretprobe_blacklist[i].addr == addr) | |
1891 | return -EINVAL; | |
1892 | } | |
1893 | } | |
b94cce92 HN |
1894 | |
1895 | rp->kp.pre_handler = pre_handler_kretprobe; | |
7522a842 AM |
1896 | rp->kp.post_handler = NULL; |
1897 | rp->kp.fault_handler = NULL; | |
b94cce92 HN |
1898 | |
1899 | /* Pre-allocate memory for max kretprobe instances */ | |
1900 | if (rp->maxactive <= 0) { | |
1901 | #ifdef CONFIG_PREEMPT | |
c2ef6661 | 1902 | rp->maxactive = max_t(unsigned int, 10, 2*num_possible_cpus()); |
b94cce92 | 1903 | #else |
4dae560f | 1904 | rp->maxactive = num_possible_cpus(); |
b94cce92 HN |
1905 | #endif |
1906 | } | |
ec484608 | 1907 | raw_spin_lock_init(&rp->lock); |
b94cce92 HN |
1908 | INIT_HLIST_HEAD(&rp->free_instances); |
1909 | for (i = 0; i < rp->maxactive; i++) { | |
f47cd9b5 AS |
1910 | inst = kmalloc(sizeof(struct kretprobe_instance) + |
1911 | rp->data_size, GFP_KERNEL); | |
b94cce92 HN |
1912 | if (inst == NULL) { |
1913 | free_rp_inst(rp); | |
1914 | return -ENOMEM; | |
1915 | } | |
ef53d9c5 S |
1916 | INIT_HLIST_NODE(&inst->hlist); |
1917 | hlist_add_head(&inst->hlist, &rp->free_instances); | |
b94cce92 HN |
1918 | } |
1919 | ||
1920 | rp->nmissed = 0; | |
1921 | /* Establish function entry probe point */ | |
49ad2fd7 | 1922 | ret = register_kprobe(&rp->kp); |
4a296e07 | 1923 | if (ret != 0) |
b94cce92 HN |
1924 | free_rp_inst(rp); |
1925 | return ret; | |
1926 | } | |
99081ab5 | 1927 | EXPORT_SYMBOL_GPL(register_kretprobe); |
b94cce92 | 1928 | |
55479f64 | 1929 | int register_kretprobes(struct kretprobe **rps, int num) |
4a296e07 MH |
1930 | { |
1931 | int ret = 0, i; | |
1932 | ||
1933 | if (num <= 0) | |
1934 | return -EINVAL; | |
1935 | for (i = 0; i < num; i++) { | |
49ad2fd7 | 1936 | ret = register_kretprobe(rps[i]); |
67dddaad MH |
1937 | if (ret < 0) { |
1938 | if (i > 0) | |
1939 | unregister_kretprobes(rps, i); | |
4a296e07 MH |
1940 | break; |
1941 | } | |
1942 | } | |
1943 | return ret; | |
1944 | } | |
99081ab5 | 1945 | EXPORT_SYMBOL_GPL(register_kretprobes); |
4a296e07 | 1946 | |
55479f64 | 1947 | void unregister_kretprobe(struct kretprobe *rp) |
4a296e07 MH |
1948 | { |
1949 | unregister_kretprobes(&rp, 1); | |
1950 | } | |
99081ab5 | 1951 | EXPORT_SYMBOL_GPL(unregister_kretprobe); |
4a296e07 | 1952 | |
55479f64 | 1953 | void unregister_kretprobes(struct kretprobe **rps, int num) |
4a296e07 MH |
1954 | { |
1955 | int i; | |
1956 | ||
1957 | if (num <= 0) | |
1958 | return; | |
1959 | mutex_lock(&kprobe_mutex); | |
1960 | for (i = 0; i < num; i++) | |
1961 | if (__unregister_kprobe_top(&rps[i]->kp) < 0) | |
1962 | rps[i]->kp.addr = NULL; | |
1963 | mutex_unlock(&kprobe_mutex); | |
1964 | ||
1965 | synchronize_sched(); | |
1966 | for (i = 0; i < num; i++) { | |
1967 | if (rps[i]->kp.addr) { | |
1968 | __unregister_kprobe_bottom(&rps[i]->kp); | |
1969 | cleanup_rp_inst(rps[i]); | |
1970 | } | |
1971 | } | |
1972 | } | |
99081ab5 | 1973 | EXPORT_SYMBOL_GPL(unregister_kretprobes); |
4a296e07 | 1974 | |
9edddaa2 | 1975 | #else /* CONFIG_KRETPROBES */ |
55479f64 | 1976 | int register_kretprobe(struct kretprobe *rp) |
b94cce92 HN |
1977 | { |
1978 | return -ENOSYS; | |
1979 | } | |
99081ab5 | 1980 | EXPORT_SYMBOL_GPL(register_kretprobe); |
b94cce92 | 1981 | |
55479f64 | 1982 | int register_kretprobes(struct kretprobe **rps, int num) |
346fd59b | 1983 | { |
4a296e07 | 1984 | return -ENOSYS; |
346fd59b | 1985 | } |
99081ab5 MH |
1986 | EXPORT_SYMBOL_GPL(register_kretprobes); |
1987 | ||
55479f64 | 1988 | void unregister_kretprobe(struct kretprobe *rp) |
b94cce92 | 1989 | { |
4a296e07 | 1990 | } |
99081ab5 | 1991 | EXPORT_SYMBOL_GPL(unregister_kretprobe); |
b94cce92 | 1992 | |
55479f64 | 1993 | void unregister_kretprobes(struct kretprobe **rps, int num) |
4a296e07 MH |
1994 | { |
1995 | } | |
99081ab5 | 1996 | EXPORT_SYMBOL_GPL(unregister_kretprobes); |
4c4308cb | 1997 | |
820aede0 | 1998 | static int pre_handler_kretprobe(struct kprobe *p, struct pt_regs *regs) |
4a296e07 MH |
1999 | { |
2000 | return 0; | |
b94cce92 | 2001 | } |
820aede0 | 2002 | NOKPROBE_SYMBOL(pre_handler_kretprobe); |
b94cce92 | 2003 | |
4a296e07 MH |
2004 | #endif /* CONFIG_KRETPROBES */ |
2005 | ||
e8386a0c | 2006 | /* Set the kprobe gone and remove its instruction buffer. */ |
55479f64 | 2007 | static void kill_kprobe(struct kprobe *p) |
e8386a0c MH |
2008 | { |
2009 | struct kprobe *kp; | |
de5bd88d | 2010 | |
e8386a0c | 2011 | p->flags |= KPROBE_FLAG_GONE; |
afd66255 | 2012 | if (kprobe_aggrprobe(p)) { |
e8386a0c MH |
2013 | /* |
2014 | * If this is an aggr_kprobe, we have to list all the | |
2015 | * chained probes and mark them GONE. | |
2016 | */ | |
2017 | list_for_each_entry_rcu(kp, &p->list, list) | |
2018 | kp->flags |= KPROBE_FLAG_GONE; | |
2019 | p->post_handler = NULL; | |
afd66255 | 2020 | kill_optimized_kprobe(p); |
e8386a0c MH |
2021 | } |
2022 | /* | |
2023 | * Here, we can remove insn_slot safely, because no thread calls | |
2024 | * the original probed function (which will be freed soon) any more. | |
2025 | */ | |
2026 | arch_remove_kprobe(p); | |
2027 | } | |
2028 | ||
c0614829 | 2029 | /* Disable one kprobe */ |
55479f64 | 2030 | int disable_kprobe(struct kprobe *kp) |
c0614829 MH |
2031 | { |
2032 | int ret = 0; | |
297f9233 | 2033 | struct kprobe *p; |
c0614829 MH |
2034 | |
2035 | mutex_lock(&kprobe_mutex); | |
2036 | ||
6f0f1dd7 | 2037 | /* Disable this kprobe */ |
297f9233 JY |
2038 | p = __disable_kprobe(kp); |
2039 | if (IS_ERR(p)) | |
2040 | ret = PTR_ERR(p); | |
c0614829 | 2041 | |
c0614829 MH |
2042 | mutex_unlock(&kprobe_mutex); |
2043 | return ret; | |
2044 | } | |
2045 | EXPORT_SYMBOL_GPL(disable_kprobe); | |
2046 | ||
2047 | /* Enable one kprobe */ | |
55479f64 | 2048 | int enable_kprobe(struct kprobe *kp) |
c0614829 MH |
2049 | { |
2050 | int ret = 0; | |
2051 | struct kprobe *p; | |
2052 | ||
2053 | mutex_lock(&kprobe_mutex); | |
2054 | ||
2055 | /* Check whether specified probe is valid. */ | |
2056 | p = __get_valid_kprobe(kp); | |
2057 | if (unlikely(p == NULL)) { | |
2058 | ret = -EINVAL; | |
2059 | goto out; | |
2060 | } | |
2061 | ||
2062 | if (kprobe_gone(kp)) { | |
2063 | /* This kprobe has gone, we couldn't enable it. */ | |
2064 | ret = -EINVAL; | |
2065 | goto out; | |
2066 | } | |
2067 | ||
2068 | if (p != kp) | |
2069 | kp->flags &= ~KPROBE_FLAG_DISABLED; | |
2070 | ||
2071 | if (!kprobes_all_disarmed && kprobe_disabled(p)) { | |
2072 | p->flags &= ~KPROBE_FLAG_DISABLED; | |
12310e34 JY |
2073 | ret = arm_kprobe(p); |
2074 | if (ret) | |
2075 | p->flags |= KPROBE_FLAG_DISABLED; | |
c0614829 MH |
2076 | } |
2077 | out: | |
2078 | mutex_unlock(&kprobe_mutex); | |
2079 | return ret; | |
2080 | } | |
2081 | EXPORT_SYMBOL_GPL(enable_kprobe); | |
2082 | ||
4458515b | 2083 | /* Caller must NOT call this in usual path. This is only for critical case */ |
820aede0 | 2084 | void dump_kprobe(struct kprobe *kp) |
24851d24 | 2085 | { |
4458515b MH |
2086 | pr_err("Dumping kprobe:\n"); |
2087 | pr_err("Name: %s\nOffset: %x\nAddress: %pS\n", | |
2088 | kp->symbol_name, kp->offset, kp->addr); | |
24851d24 | 2089 | } |
820aede0 | 2090 | NOKPROBE_SYMBOL(dump_kprobe); |
24851d24 | 2091 | |
376e2424 MH |
2092 | /* |
2093 | * Lookup and populate the kprobe_blacklist. | |
2094 | * | |
2095 | * Unlike the kretprobe blacklist, we'll need to determine | |
2096 | * the range of addresses that belong to the said functions, | |
2097 | * since a kprobe need not necessarily be at the beginning | |
2098 | * of a function. | |
2099 | */ | |
2100 | static int __init populate_kprobe_blacklist(unsigned long *start, | |
2101 | unsigned long *end) | |
2102 | { | |
2103 | unsigned long *iter; | |
2104 | struct kprobe_blacklist_entry *ent; | |
d81b4253 | 2105 | unsigned long entry, offset = 0, size = 0; |
376e2424 MH |
2106 | |
2107 | for (iter = start; iter < end; iter++) { | |
d81b4253 MH |
2108 | entry = arch_deref_entry_point((void *)*iter); |
2109 | ||
2110 | if (!kernel_text_address(entry) || | |
4458515b | 2111 | !kallsyms_lookup_size_offset(entry, &size, &offset)) |
376e2424 | 2112 | continue; |
376e2424 MH |
2113 | |
2114 | ent = kmalloc(sizeof(*ent), GFP_KERNEL); | |
2115 | if (!ent) | |
2116 | return -ENOMEM; | |
d81b4253 MH |
2117 | ent->start_addr = entry; |
2118 | ent->end_addr = entry + size; | |
376e2424 MH |
2119 | INIT_LIST_HEAD(&ent->list); |
2120 | list_add_tail(&ent->list, &kprobe_blacklist); | |
2121 | } | |
2122 | return 0; | |
2123 | } | |
2124 | ||
e8386a0c | 2125 | /* Module notifier call back, checking kprobes on the module */ |
55479f64 MH |
2126 | static int kprobes_module_callback(struct notifier_block *nb, |
2127 | unsigned long val, void *data) | |
e8386a0c MH |
2128 | { |
2129 | struct module *mod = data; | |
2130 | struct hlist_head *head; | |
e8386a0c MH |
2131 | struct kprobe *p; |
2132 | unsigned int i; | |
f24659d9 | 2133 | int checkcore = (val == MODULE_STATE_GOING); |
e8386a0c | 2134 | |
f24659d9 | 2135 | if (val != MODULE_STATE_GOING && val != MODULE_STATE_LIVE) |
e8386a0c MH |
2136 | return NOTIFY_DONE; |
2137 | ||
2138 | /* | |
f24659d9 MH |
2139 | * When MODULE_STATE_GOING was notified, both of module .text and |
2140 | * .init.text sections would be freed. When MODULE_STATE_LIVE was | |
2141 | * notified, only .init.text section would be freed. We need to | |
2142 | * disable kprobes which have been inserted in the sections. | |
e8386a0c MH |
2143 | */ |
2144 | mutex_lock(&kprobe_mutex); | |
2145 | for (i = 0; i < KPROBE_TABLE_SIZE; i++) { | |
2146 | head = &kprobe_table[i]; | |
b67bfe0d | 2147 | hlist_for_each_entry_rcu(p, head, hlist) |
f24659d9 MH |
2148 | if (within_module_init((unsigned long)p->addr, mod) || |
2149 | (checkcore && | |
2150 | within_module_core((unsigned long)p->addr, mod))) { | |
e8386a0c MH |
2151 | /* |
2152 | * The vaddr this probe is installed will soon | |
2153 | * be vfreed buy not synced to disk. Hence, | |
2154 | * disarming the breakpoint isn't needed. | |
545a0281 SRV |
2155 | * |
2156 | * Note, this will also move any optimized probes | |
2157 | * that are pending to be removed from their | |
2158 | * corresponding lists to the freeing_list and | |
2159 | * will not be touched by the delayed | |
2160 | * kprobe_optimizer work handler. | |
e8386a0c MH |
2161 | */ |
2162 | kill_kprobe(p); | |
2163 | } | |
2164 | } | |
2165 | mutex_unlock(&kprobe_mutex); | |
2166 | return NOTIFY_DONE; | |
2167 | } | |
2168 | ||
2169 | static struct notifier_block kprobe_module_nb = { | |
2170 | .notifier_call = kprobes_module_callback, | |
2171 | .priority = 0 | |
2172 | }; | |
2173 | ||
376e2424 MH |
2174 | /* Markers of _kprobe_blacklist section */ |
2175 | extern unsigned long __start_kprobe_blacklist[]; | |
2176 | extern unsigned long __stop_kprobe_blacklist[]; | |
2177 | ||
1da177e4 LT |
2178 | static int __init init_kprobes(void) |
2179 | { | |
2180 | int i, err = 0; | |
2181 | ||
2182 | /* FIXME allocate the probe table, currently defined statically */ | |
2183 | /* initialize all list heads */ | |
b94cce92 | 2184 | for (i = 0; i < KPROBE_TABLE_SIZE; i++) { |
1da177e4 | 2185 | INIT_HLIST_HEAD(&kprobe_table[i]); |
b94cce92 | 2186 | INIT_HLIST_HEAD(&kretprobe_inst_table[i]); |
ec484608 | 2187 | raw_spin_lock_init(&(kretprobe_table_locks[i].lock)); |
b94cce92 | 2188 | } |
1da177e4 | 2189 | |
376e2424 MH |
2190 | err = populate_kprobe_blacklist(__start_kprobe_blacklist, |
2191 | __stop_kprobe_blacklist); | |
2192 | if (err) { | |
2193 | pr_err("kprobes: failed to populate blacklist: %d\n", err); | |
2194 | pr_err("Please take care of using kprobes.\n"); | |
3d8d996e SD |
2195 | } |
2196 | ||
f438d914 MH |
2197 | if (kretprobe_blacklist_size) { |
2198 | /* lookup the function address from its name */ | |
2199 | for (i = 0; kretprobe_blacklist[i].name != NULL; i++) { | |
49e0b465 | 2200 | kretprobe_blacklist[i].addr = |
290e3070 | 2201 | kprobe_lookup_name(kretprobe_blacklist[i].name, 0); |
f438d914 MH |
2202 | if (!kretprobe_blacklist[i].addr) |
2203 | printk("kretprobe: lookup failed: %s\n", | |
2204 | kretprobe_blacklist[i].name); | |
2205 | } | |
2206 | } | |
2207 | ||
b2be84df MH |
2208 | #if defined(CONFIG_OPTPROBES) |
2209 | #if defined(__ARCH_WANT_KPROBES_INSN_SLOT) | |
afd66255 MH |
2210 | /* Init kprobe_optinsn_slots */ |
2211 | kprobe_optinsn_slots.insn_size = MAX_OPTINSN_SIZE; | |
2212 | #endif | |
b2be84df MH |
2213 | /* By default, kprobes can be optimized */ |
2214 | kprobes_allow_optimization = true; | |
2215 | #endif | |
afd66255 | 2216 | |
e579abeb MH |
2217 | /* By default, kprobes are armed */ |
2218 | kprobes_all_disarmed = false; | |
bf8f6e5b | 2219 | |
6772926b | 2220 | err = arch_init_kprobes(); |
802eae7c RL |
2221 | if (!err) |
2222 | err = register_die_notifier(&kprobe_exceptions_nb); | |
e8386a0c MH |
2223 | if (!err) |
2224 | err = register_module_notifier(&kprobe_module_nb); | |
2225 | ||
ef53d9c5 | 2226 | kprobes_initialized = (err == 0); |
802eae7c | 2227 | |
8c1c9356 AM |
2228 | if (!err) |
2229 | init_test_probes(); | |
1da177e4 LT |
2230 | return err; |
2231 | } | |
2232 | ||
346fd59b | 2233 | #ifdef CONFIG_DEBUG_FS |
55479f64 | 2234 | static void report_probe(struct seq_file *pi, struct kprobe *p, |
afd66255 | 2235 | const char *sym, int offset, char *modname, struct kprobe *pp) |
346fd59b SD |
2236 | { |
2237 | char *kprobe_type; | |
81365a94 | 2238 | void *addr = p->addr; |
346fd59b SD |
2239 | |
2240 | if (p->pre_handler == pre_handler_kretprobe) | |
2241 | kprobe_type = "r"; | |
346fd59b SD |
2242 | else |
2243 | kprobe_type = "k"; | |
afd66255 | 2244 | |
81365a94 MH |
2245 | if (!kallsyms_show_value()) |
2246 | addr = NULL; | |
2247 | ||
346fd59b | 2248 | if (sym) |
81365a94 MH |
2249 | seq_printf(pi, "%px %s %s+0x%x %s ", |
2250 | addr, kprobe_type, sym, offset, | |
afd66255 | 2251 | (modname ? modname : " ")); |
81365a94 MH |
2252 | else /* try to use %pS */ |
2253 | seq_printf(pi, "%px %s %pS ", | |
2254 | addr, kprobe_type, p->addr); | |
afd66255 MH |
2255 | |
2256 | if (!pp) | |
2257 | pp = p; | |
ae6aa16f | 2258 | seq_printf(pi, "%s%s%s%s\n", |
afd66255 MH |
2259 | (kprobe_gone(p) ? "[GONE]" : ""), |
2260 | ((kprobe_disabled(p) && !kprobe_gone(p)) ? "[DISABLED]" : ""), | |
ae6aa16f MH |
2261 | (kprobe_optimized(pp) ? "[OPTIMIZED]" : ""), |
2262 | (kprobe_ftrace(pp) ? "[FTRACE]" : "")); | |
346fd59b SD |
2263 | } |
2264 | ||
55479f64 | 2265 | static void *kprobe_seq_start(struct seq_file *f, loff_t *pos) |
346fd59b SD |
2266 | { |
2267 | return (*pos < KPROBE_TABLE_SIZE) ? pos : NULL; | |
2268 | } | |
2269 | ||
55479f64 | 2270 | static void *kprobe_seq_next(struct seq_file *f, void *v, loff_t *pos) |
346fd59b SD |
2271 | { |
2272 | (*pos)++; | |
2273 | if (*pos >= KPROBE_TABLE_SIZE) | |
2274 | return NULL; | |
2275 | return pos; | |
2276 | } | |
2277 | ||
55479f64 | 2278 | static void kprobe_seq_stop(struct seq_file *f, void *v) |
346fd59b SD |
2279 | { |
2280 | /* Nothing to do */ | |
2281 | } | |
2282 | ||
55479f64 | 2283 | static int show_kprobe_addr(struct seq_file *pi, void *v) |
346fd59b SD |
2284 | { |
2285 | struct hlist_head *head; | |
346fd59b SD |
2286 | struct kprobe *p, *kp; |
2287 | const char *sym = NULL; | |
2288 | unsigned int i = *(loff_t *) v; | |
ffb45122 | 2289 | unsigned long offset = 0; |
ab767865 | 2290 | char *modname, namebuf[KSYM_NAME_LEN]; |
346fd59b SD |
2291 | |
2292 | head = &kprobe_table[i]; | |
2293 | preempt_disable(); | |
b67bfe0d | 2294 | hlist_for_each_entry_rcu(p, head, hlist) { |
ffb45122 | 2295 | sym = kallsyms_lookup((unsigned long)p->addr, NULL, |
346fd59b | 2296 | &offset, &modname, namebuf); |
afd66255 | 2297 | if (kprobe_aggrprobe(p)) { |
346fd59b | 2298 | list_for_each_entry_rcu(kp, &p->list, list) |
afd66255 | 2299 | report_probe(pi, kp, sym, offset, modname, p); |
346fd59b | 2300 | } else |
afd66255 | 2301 | report_probe(pi, p, sym, offset, modname, NULL); |
346fd59b SD |
2302 | } |
2303 | preempt_enable(); | |
2304 | return 0; | |
2305 | } | |
2306 | ||
88e9d34c | 2307 | static const struct seq_operations kprobes_seq_ops = { |
346fd59b SD |
2308 | .start = kprobe_seq_start, |
2309 | .next = kprobe_seq_next, | |
2310 | .stop = kprobe_seq_stop, | |
2311 | .show = show_kprobe_addr | |
2312 | }; | |
2313 | ||
55479f64 | 2314 | static int kprobes_open(struct inode *inode, struct file *filp) |
346fd59b SD |
2315 | { |
2316 | return seq_open(filp, &kprobes_seq_ops); | |
2317 | } | |
2318 | ||
828c0950 | 2319 | static const struct file_operations debugfs_kprobes_operations = { |
346fd59b SD |
2320 | .open = kprobes_open, |
2321 | .read = seq_read, | |
2322 | .llseek = seq_lseek, | |
2323 | .release = seq_release, | |
2324 | }; | |
2325 | ||
63724740 MH |
2326 | /* kprobes/blacklist -- shows which functions can not be probed */ |
2327 | static void *kprobe_blacklist_seq_start(struct seq_file *m, loff_t *pos) | |
2328 | { | |
2329 | return seq_list_start(&kprobe_blacklist, *pos); | |
2330 | } | |
2331 | ||
2332 | static void *kprobe_blacklist_seq_next(struct seq_file *m, void *v, loff_t *pos) | |
2333 | { | |
2334 | return seq_list_next(v, &kprobe_blacklist, pos); | |
2335 | } | |
2336 | ||
2337 | static int kprobe_blacklist_seq_show(struct seq_file *m, void *v) | |
2338 | { | |
2339 | struct kprobe_blacklist_entry *ent = | |
2340 | list_entry(v, struct kprobe_blacklist_entry, list); | |
2341 | ||
ffb9bd68 MH |
2342 | /* |
2343 | * If /proc/kallsyms is not showing kernel address, we won't | |
2344 | * show them here either. | |
2345 | */ | |
2346 | if (!kallsyms_show_value()) | |
2347 | seq_printf(m, "0x%px-0x%px\t%ps\n", NULL, NULL, | |
2348 | (void *)ent->start_addr); | |
2349 | else | |
2350 | seq_printf(m, "0x%px-0x%px\t%ps\n", (void *)ent->start_addr, | |
2351 | (void *)ent->end_addr, (void *)ent->start_addr); | |
63724740 MH |
2352 | return 0; |
2353 | } | |
2354 | ||
2355 | static const struct seq_operations kprobe_blacklist_seq_ops = { | |
2356 | .start = kprobe_blacklist_seq_start, | |
2357 | .next = kprobe_blacklist_seq_next, | |
2358 | .stop = kprobe_seq_stop, /* Reuse void function */ | |
2359 | .show = kprobe_blacklist_seq_show, | |
2360 | }; | |
2361 | ||
2362 | static int kprobe_blacklist_open(struct inode *inode, struct file *filp) | |
2363 | { | |
2364 | return seq_open(filp, &kprobe_blacklist_seq_ops); | |
2365 | } | |
2366 | ||
2367 | static const struct file_operations debugfs_kprobe_blacklist_ops = { | |
2368 | .open = kprobe_blacklist_open, | |
2369 | .read = seq_read, | |
2370 | .llseek = seq_lseek, | |
2371 | .release = seq_release, | |
2372 | }; | |
2373 | ||
12310e34 | 2374 | static int arm_all_kprobes(void) |
bf8f6e5b AM |
2375 | { |
2376 | struct hlist_head *head; | |
bf8f6e5b | 2377 | struct kprobe *p; |
12310e34 JY |
2378 | unsigned int i, total = 0, errors = 0; |
2379 | int err, ret = 0; | |
bf8f6e5b AM |
2380 | |
2381 | mutex_lock(&kprobe_mutex); | |
2382 | ||
e579abeb MH |
2383 | /* If kprobes are armed, just return */ |
2384 | if (!kprobes_all_disarmed) | |
bf8f6e5b AM |
2385 | goto already_enabled; |
2386 | ||
977ad481 WN |
2387 | /* |
2388 | * optimize_kprobe() called by arm_kprobe() checks | |
2389 | * kprobes_all_disarmed, so set kprobes_all_disarmed before | |
2390 | * arm_kprobe. | |
2391 | */ | |
2392 | kprobes_all_disarmed = false; | |
afd66255 | 2393 | /* Arming kprobes doesn't optimize kprobe itself */ |
bf8f6e5b AM |
2394 | for (i = 0; i < KPROBE_TABLE_SIZE; i++) { |
2395 | head = &kprobe_table[i]; | |
12310e34 JY |
2396 | /* Arm all kprobes on a best-effort basis */ |
2397 | hlist_for_each_entry_rcu(p, head, hlist) { | |
2398 | if (!kprobe_disabled(p)) { | |
2399 | err = arm_kprobe(p); | |
2400 | if (err) { | |
2401 | errors++; | |
2402 | ret = err; | |
2403 | } | |
2404 | total++; | |
2405 | } | |
2406 | } | |
bf8f6e5b AM |
2407 | } |
2408 | ||
12310e34 JY |
2409 | if (errors) |
2410 | pr_warn("Kprobes globally enabled, but failed to arm %d out of %d probes\n", | |
2411 | errors, total); | |
2412 | else | |
2413 | pr_info("Kprobes globally enabled\n"); | |
bf8f6e5b AM |
2414 | |
2415 | already_enabled: | |
2416 | mutex_unlock(&kprobe_mutex); | |
12310e34 | 2417 | return ret; |
bf8f6e5b AM |
2418 | } |
2419 | ||
297f9233 | 2420 | static int disarm_all_kprobes(void) |
bf8f6e5b AM |
2421 | { |
2422 | struct hlist_head *head; | |
bf8f6e5b | 2423 | struct kprobe *p; |
297f9233 JY |
2424 | unsigned int i, total = 0, errors = 0; |
2425 | int err, ret = 0; | |
bf8f6e5b AM |
2426 | |
2427 | mutex_lock(&kprobe_mutex); | |
2428 | ||
e579abeb | 2429 | /* If kprobes are already disarmed, just return */ |
6274de49 MH |
2430 | if (kprobes_all_disarmed) { |
2431 | mutex_unlock(&kprobe_mutex); | |
297f9233 | 2432 | return 0; |
6274de49 | 2433 | } |
bf8f6e5b | 2434 | |
e579abeb | 2435 | kprobes_all_disarmed = true; |
afd66255 | 2436 | |
bf8f6e5b AM |
2437 | for (i = 0; i < KPROBE_TABLE_SIZE; i++) { |
2438 | head = &kprobe_table[i]; | |
297f9233 | 2439 | /* Disarm all kprobes on a best-effort basis */ |
b67bfe0d | 2440 | hlist_for_each_entry_rcu(p, head, hlist) { |
297f9233 JY |
2441 | if (!arch_trampoline_kprobe(p) && !kprobe_disabled(p)) { |
2442 | err = disarm_kprobe(p, false); | |
2443 | if (err) { | |
2444 | errors++; | |
2445 | ret = err; | |
2446 | } | |
2447 | total++; | |
2448 | } | |
bf8f6e5b AM |
2449 | } |
2450 | } | |
297f9233 JY |
2451 | |
2452 | if (errors) | |
2453 | pr_warn("Kprobes globally disabled, but failed to disarm %d out of %d probes\n", | |
2454 | errors, total); | |
2455 | else | |
2456 | pr_info("Kprobes globally disabled\n"); | |
2457 | ||
bf8f6e5b | 2458 | mutex_unlock(&kprobe_mutex); |
bf8f6e5b | 2459 | |
6274de49 MH |
2460 | /* Wait for disarming all kprobes by optimizer */ |
2461 | wait_for_kprobe_optimizer(); | |
297f9233 JY |
2462 | |
2463 | return ret; | |
bf8f6e5b AM |
2464 | } |
2465 | ||
2466 | /* | |
2467 | * XXX: The debugfs bool file interface doesn't allow for callbacks | |
2468 | * when the bool state is switched. We can reuse that facility when | |
2469 | * available | |
2470 | */ | |
2471 | static ssize_t read_enabled_file_bool(struct file *file, | |
2472 | char __user *user_buf, size_t count, loff_t *ppos) | |
2473 | { | |
2474 | char buf[3]; | |
2475 | ||
e579abeb | 2476 | if (!kprobes_all_disarmed) |
bf8f6e5b AM |
2477 | buf[0] = '1'; |
2478 | else | |
2479 | buf[0] = '0'; | |
2480 | buf[1] = '\n'; | |
2481 | buf[2] = 0x00; | |
2482 | return simple_read_from_buffer(user_buf, count, ppos, buf, 2); | |
2483 | } | |
2484 | ||
2485 | static ssize_t write_enabled_file_bool(struct file *file, | |
2486 | const char __user *user_buf, size_t count, loff_t *ppos) | |
2487 | { | |
2488 | char buf[32]; | |
efeb156e | 2489 | size_t buf_size; |
12310e34 | 2490 | int ret = 0; |
bf8f6e5b AM |
2491 | |
2492 | buf_size = min(count, (sizeof(buf)-1)); | |
2493 | if (copy_from_user(buf, user_buf, buf_size)) | |
2494 | return -EFAULT; | |
2495 | ||
10fb46d5 | 2496 | buf[buf_size] = '\0'; |
bf8f6e5b AM |
2497 | switch (buf[0]) { |
2498 | case 'y': | |
2499 | case 'Y': | |
2500 | case '1': | |
12310e34 | 2501 | ret = arm_all_kprobes(); |
bf8f6e5b AM |
2502 | break; |
2503 | case 'n': | |
2504 | case 'N': | |
2505 | case '0': | |
297f9233 | 2506 | ret = disarm_all_kprobes(); |
bf8f6e5b | 2507 | break; |
10fb46d5 MK |
2508 | default: |
2509 | return -EINVAL; | |
bf8f6e5b AM |
2510 | } |
2511 | ||
12310e34 JY |
2512 | if (ret) |
2513 | return ret; | |
2514 | ||
bf8f6e5b AM |
2515 | return count; |
2516 | } | |
2517 | ||
828c0950 | 2518 | static const struct file_operations fops_kp = { |
bf8f6e5b AM |
2519 | .read = read_enabled_file_bool, |
2520 | .write = write_enabled_file_bool, | |
6038f373 | 2521 | .llseek = default_llseek, |
bf8f6e5b AM |
2522 | }; |
2523 | ||
55479f64 | 2524 | static int __init debugfs_kprobe_init(void) |
346fd59b SD |
2525 | { |
2526 | struct dentry *dir, *file; | |
bf8f6e5b | 2527 | unsigned int value = 1; |
346fd59b SD |
2528 | |
2529 | dir = debugfs_create_dir("kprobes", NULL); | |
2530 | if (!dir) | |
2531 | return -ENOMEM; | |
2532 | ||
f2a3ab36 | 2533 | file = debugfs_create_file("list", 0400, dir, NULL, |
346fd59b | 2534 | &debugfs_kprobes_operations); |
63724740 MH |
2535 | if (!file) |
2536 | goto error; | |
346fd59b | 2537 | |
bf8f6e5b AM |
2538 | file = debugfs_create_file("enabled", 0600, dir, |
2539 | &value, &fops_kp); | |
63724740 MH |
2540 | if (!file) |
2541 | goto error; | |
2542 | ||
f2a3ab36 | 2543 | file = debugfs_create_file("blacklist", 0400, dir, NULL, |
63724740 MH |
2544 | &debugfs_kprobe_blacklist_ops); |
2545 | if (!file) | |
2546 | goto error; | |
bf8f6e5b | 2547 | |
346fd59b | 2548 | return 0; |
63724740 MH |
2549 | |
2550 | error: | |
2551 | debugfs_remove(dir); | |
2552 | return -ENOMEM; | |
346fd59b SD |
2553 | } |
2554 | ||
2555 | late_initcall(debugfs_kprobe_init); | |
2556 | #endif /* CONFIG_DEBUG_FS */ | |
2557 | ||
2558 | module_init(init_kprobes); |