]>
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 | 231 | /* Ensure no-one is interrupted on the garbages */ |
ae8b7ce7 | 232 | synchronize_rcu(); |
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 | 549 | list_del_init(&op->list); |
cbdd96f5 MH |
550 | if (WARN_ON_ONCE(!kprobe_unused(&op->kp))) { |
551 | /* | |
552 | * This must not happen, but if there is a kprobe | |
553 | * still in use, keep it on kprobes hash list. | |
554 | */ | |
555 | continue; | |
556 | } | |
6274de49 MH |
557 | free_aggr_kprobe(&op->kp); |
558 | } | |
559 | } | |
560 | ||
561 | /* Start optimizer after OPTIMIZE_DELAY passed */ | |
55479f64 | 562 | static void kick_kprobe_optimizer(void) |
6274de49 | 563 | { |
ad72b3be | 564 | schedule_delayed_work(&optimizing_work, OPTIMIZE_DELAY); |
6274de49 MH |
565 | } |
566 | ||
61f4e13f | 567 | /* Kprobe jump optimizer */ |
55479f64 | 568 | static void kprobe_optimizer(struct work_struct *work) |
61f4e13f | 569 | { |
72ef3794 | 570 | mutex_lock(&kprobe_mutex); |
2d1e38f5 | 571 | cpus_read_lock(); |
61f4e13f MH |
572 | /* Lock modules while optimizing kprobes */ |
573 | mutex_lock(&module_mutex); | |
61f4e13f MH |
574 | |
575 | /* | |
6274de49 MH |
576 | * Step 1: Unoptimize kprobes and collect cleaned (unused and disarmed) |
577 | * kprobes before waiting for quiesence period. | |
578 | */ | |
7b959fc5 | 579 | do_unoptimize_kprobes(); |
6274de49 MH |
580 | |
581 | /* | |
a30b85df MH |
582 | * Step 2: Wait for quiesence period to ensure all potentially |
583 | * preempted tasks to have normally scheduled. Because optprobe | |
584 | * may modify multiple instructions, there is a chance that Nth | |
585 | * instruction is preempted. In that case, such tasks can return | |
586 | * to 2nd-Nth byte of jump instruction. This wait is for avoiding it. | |
587 | * Note that on non-preemptive kernel, this is transparently converted | |
588 | * to synchronoze_sched() to wait for all interrupts to have completed. | |
61f4e13f | 589 | */ |
a30b85df | 590 | synchronize_rcu_tasks(); |
61f4e13f | 591 | |
6274de49 | 592 | /* Step 3: Optimize kprobes after quiesence period */ |
61f4e13f | 593 | do_optimize_kprobes(); |
6274de49 MH |
594 | |
595 | /* Step 4: Free cleaned kprobes after quiesence period */ | |
7b959fc5 | 596 | do_free_cleaned_kprobes(); |
6274de49 | 597 | |
afd66255 | 598 | mutex_unlock(&module_mutex); |
2d1e38f5 | 599 | cpus_read_unlock(); |
72ef3794 | 600 | mutex_unlock(&kprobe_mutex); |
6274de49 | 601 | |
cd7ebe22 | 602 | /* Step 5: Kick optimizer again if needed */ |
f984ba4e | 603 | if (!list_empty(&optimizing_list) || !list_empty(&unoptimizing_list)) |
cd7ebe22 | 604 | kick_kprobe_optimizer(); |
6274de49 MH |
605 | } |
606 | ||
607 | /* Wait for completing optimization and unoptimization */ | |
30e7d894 | 608 | void wait_for_kprobe_optimizer(void) |
6274de49 | 609 | { |
ad72b3be TH |
610 | mutex_lock(&kprobe_mutex); |
611 | ||
612 | while (!list_empty(&optimizing_list) || !list_empty(&unoptimizing_list)) { | |
613 | mutex_unlock(&kprobe_mutex); | |
614 | ||
615 | /* this will also make optimizing_work execute immmediately */ | |
616 | flush_delayed_work(&optimizing_work); | |
617 | /* @optimizing_work might not have been queued yet, relax */ | |
618 | cpu_relax(); | |
619 | ||
620 | mutex_lock(&kprobe_mutex); | |
621 | } | |
622 | ||
623 | mutex_unlock(&kprobe_mutex); | |
afd66255 MH |
624 | } |
625 | ||
626 | /* Optimize kprobe if p is ready to be optimized */ | |
55479f64 | 627 | static void optimize_kprobe(struct kprobe *p) |
afd66255 MH |
628 | { |
629 | struct optimized_kprobe *op; | |
630 | ||
631 | /* Check if the kprobe is disabled or not ready for optimization. */ | |
b2be84df | 632 | if (!kprobe_optready(p) || !kprobes_allow_optimization || |
afd66255 MH |
633 | (kprobe_disabled(p) || kprobes_all_disarmed)) |
634 | return; | |
635 | ||
059053a2 MH |
636 | /* kprobes with post_handler can not be optimized */ |
637 | if (p->post_handler) | |
afd66255 MH |
638 | return; |
639 | ||
640 | op = container_of(p, struct optimized_kprobe, kp); | |
641 | ||
642 | /* Check there is no other kprobes at the optimized instructions */ | |
643 | if (arch_check_optimized_kprobe(op) < 0) | |
644 | return; | |
645 | ||
646 | /* Check if it is already optimized. */ | |
647 | if (op->kp.flags & KPROBE_FLAG_OPTIMIZED) | |
648 | return; | |
afd66255 | 649 | op->kp.flags |= KPROBE_FLAG_OPTIMIZED; |
6274de49 MH |
650 | |
651 | if (!list_empty(&op->list)) | |
652 | /* This is under unoptimizing. Just dequeue the probe */ | |
653 | list_del_init(&op->list); | |
654 | else { | |
655 | list_add(&op->list, &optimizing_list); | |
656 | kick_kprobe_optimizer(); | |
657 | } | |
658 | } | |
659 | ||
660 | /* Short cut to direct unoptimizing */ | |
55479f64 | 661 | static void force_unoptimize_kprobe(struct optimized_kprobe *op) |
6274de49 | 662 | { |
2d1e38f5 | 663 | lockdep_assert_cpus_held(); |
6274de49 | 664 | arch_unoptimize_kprobe(op); |
6274de49 MH |
665 | if (kprobe_disabled(&op->kp)) |
666 | arch_disarm_kprobe(&op->kp); | |
afd66255 MH |
667 | } |
668 | ||
669 | /* Unoptimize a kprobe if p is optimized */ | |
55479f64 | 670 | static void unoptimize_kprobe(struct kprobe *p, bool force) |
afd66255 MH |
671 | { |
672 | struct optimized_kprobe *op; | |
673 | ||
6274de49 MH |
674 | if (!kprobe_aggrprobe(p) || kprobe_disarmed(p)) |
675 | return; /* This is not an optprobe nor optimized */ | |
676 | ||
677 | op = container_of(p, struct optimized_kprobe, kp); | |
678 | if (!kprobe_optimized(p)) { | |
679 | /* Unoptimized or unoptimizing case */ | |
680 | if (force && !list_empty(&op->list)) { | |
681 | /* | |
682 | * Only if this is unoptimizing kprobe and forced, | |
683 | * forcibly unoptimize it. (No need to unoptimize | |
684 | * unoptimized kprobe again :) | |
685 | */ | |
afd66255 | 686 | list_del_init(&op->list); |
6274de49 MH |
687 | force_unoptimize_kprobe(op); |
688 | } | |
689 | return; | |
690 | } | |
691 | ||
692 | op->kp.flags &= ~KPROBE_FLAG_OPTIMIZED; | |
693 | if (!list_empty(&op->list)) { | |
694 | /* Dequeue from the optimization queue */ | |
695 | list_del_init(&op->list); | |
696 | return; | |
697 | } | |
698 | /* Optimized kprobe case */ | |
699 | if (force) | |
700 | /* Forcibly update the code: this is a special case */ | |
701 | force_unoptimize_kprobe(op); | |
702 | else { | |
703 | list_add(&op->list, &unoptimizing_list); | |
704 | kick_kprobe_optimizer(); | |
afd66255 MH |
705 | } |
706 | } | |
707 | ||
0490cd1f | 708 | /* Cancel unoptimizing for reusing */ |
819319fc | 709 | static int reuse_unused_kprobe(struct kprobe *ap) |
0490cd1f MH |
710 | { |
711 | struct optimized_kprobe *op; | |
712 | ||
0490cd1f MH |
713 | /* |
714 | * Unused kprobe MUST be on the way of delayed unoptimizing (means | |
715 | * there is still a relative jump) and disabled. | |
716 | */ | |
717 | op = container_of(ap, struct optimized_kprobe, kp); | |
4458515b | 718 | WARN_ON_ONCE(list_empty(&op->list)); |
0490cd1f MH |
719 | /* Enable the probe again */ |
720 | ap->flags &= ~KPROBE_FLAG_DISABLED; | |
721 | /* Optimize it again (remove from op->list) */ | |
221315d8 MH |
722 | if (!kprobe_optready(ap)) |
723 | return -EINVAL; | |
819319fc | 724 | |
0490cd1f | 725 | optimize_kprobe(ap); |
819319fc | 726 | return 0; |
0490cd1f MH |
727 | } |
728 | ||
afd66255 | 729 | /* Remove optimized instructions */ |
55479f64 | 730 | static void kill_optimized_kprobe(struct kprobe *p) |
afd66255 MH |
731 | { |
732 | struct optimized_kprobe *op; | |
733 | ||
734 | op = container_of(p, struct optimized_kprobe, kp); | |
6274de49 MH |
735 | if (!list_empty(&op->list)) |
736 | /* Dequeue from the (un)optimization queue */ | |
afd66255 | 737 | list_del_init(&op->list); |
6274de49 | 738 | op->kp.flags &= ~KPROBE_FLAG_OPTIMIZED; |
7b959fc5 MH |
739 | |
740 | if (kprobe_unused(p)) { | |
741 | /* Enqueue if it is unused */ | |
742 | list_add(&op->list, &freeing_list); | |
743 | /* | |
744 | * Remove unused probes from the hash list. After waiting | |
745 | * for synchronization, this probe is reclaimed. | |
746 | * (reclaiming is done by do_free_cleaned_kprobes().) | |
747 | */ | |
748 | hlist_del_rcu(&op->kp.hlist); | |
749 | } | |
750 | ||
6274de49 | 751 | /* Don't touch the code, because it is already freed. */ |
afd66255 MH |
752 | arch_remove_optimized_kprobe(op); |
753 | } | |
754 | ||
a460246c MH |
755 | static inline |
756 | void __prepare_optimized_kprobe(struct optimized_kprobe *op, struct kprobe *p) | |
757 | { | |
758 | if (!kprobe_ftrace(p)) | |
759 | arch_prepare_optimized_kprobe(op, p); | |
760 | } | |
761 | ||
afd66255 | 762 | /* Try to prepare optimized instructions */ |
55479f64 | 763 | static void prepare_optimized_kprobe(struct kprobe *p) |
afd66255 MH |
764 | { |
765 | struct optimized_kprobe *op; | |
766 | ||
767 | op = container_of(p, struct optimized_kprobe, kp); | |
a460246c | 768 | __prepare_optimized_kprobe(op, p); |
afd66255 MH |
769 | } |
770 | ||
afd66255 | 771 | /* Allocate new optimized_kprobe and try to prepare optimized instructions */ |
55479f64 | 772 | static struct kprobe *alloc_aggr_kprobe(struct kprobe *p) |
afd66255 MH |
773 | { |
774 | struct optimized_kprobe *op; | |
775 | ||
776 | op = kzalloc(sizeof(struct optimized_kprobe), GFP_KERNEL); | |
777 | if (!op) | |
778 | return NULL; | |
779 | ||
780 | INIT_LIST_HEAD(&op->list); | |
781 | op->kp.addr = p->addr; | |
a460246c | 782 | __prepare_optimized_kprobe(op, p); |
afd66255 MH |
783 | |
784 | return &op->kp; | |
785 | } | |
786 | ||
55479f64 | 787 | static void init_aggr_kprobe(struct kprobe *ap, struct kprobe *p); |
afd66255 MH |
788 | |
789 | /* | |
790 | * Prepare an optimized_kprobe and optimize it | |
791 | * NOTE: p must be a normal registered kprobe | |
792 | */ | |
55479f64 | 793 | static void try_to_optimize_kprobe(struct kprobe *p) |
afd66255 MH |
794 | { |
795 | struct kprobe *ap; | |
796 | struct optimized_kprobe *op; | |
797 | ||
ae6aa16f MH |
798 | /* Impossible to optimize ftrace-based kprobe */ |
799 | if (kprobe_ftrace(p)) | |
800 | return; | |
801 | ||
25764288 | 802 | /* For preparing optimization, jump_label_text_reserved() is called */ |
2d1e38f5 | 803 | cpus_read_lock(); |
25764288 MH |
804 | jump_label_lock(); |
805 | mutex_lock(&text_mutex); | |
806 | ||
afd66255 MH |
807 | ap = alloc_aggr_kprobe(p); |
808 | if (!ap) | |
25764288 | 809 | goto out; |
afd66255 MH |
810 | |
811 | op = container_of(ap, struct optimized_kprobe, kp); | |
812 | if (!arch_prepared_optinsn(&op->optinsn)) { | |
813 | /* If failed to setup optimizing, fallback to kprobe */ | |
6274de49 MH |
814 | arch_remove_optimized_kprobe(op); |
815 | kfree(op); | |
25764288 | 816 | goto out; |
afd66255 MH |
817 | } |
818 | ||
819 | init_aggr_kprobe(ap, p); | |
25764288 MH |
820 | optimize_kprobe(ap); /* This just kicks optimizer thread */ |
821 | ||
822 | out: | |
823 | mutex_unlock(&text_mutex); | |
824 | jump_label_unlock(); | |
2d1e38f5 | 825 | cpus_read_unlock(); |
afd66255 MH |
826 | } |
827 | ||
b2be84df | 828 | #ifdef CONFIG_SYSCTL |
55479f64 | 829 | static void optimize_all_kprobes(void) |
b2be84df MH |
830 | { |
831 | struct hlist_head *head; | |
b2be84df MH |
832 | struct kprobe *p; |
833 | unsigned int i; | |
834 | ||
5c51543b | 835 | mutex_lock(&kprobe_mutex); |
b2be84df MH |
836 | /* If optimization is already allowed, just return */ |
837 | if (kprobes_allow_optimization) | |
5c51543b | 838 | goto out; |
b2be84df | 839 | |
2d1e38f5 | 840 | cpus_read_lock(); |
b2be84df | 841 | kprobes_allow_optimization = true; |
b2be84df MH |
842 | for (i = 0; i < KPROBE_TABLE_SIZE; i++) { |
843 | head = &kprobe_table[i]; | |
b67bfe0d | 844 | hlist_for_each_entry_rcu(p, head, hlist) |
b2be84df MH |
845 | if (!kprobe_disabled(p)) |
846 | optimize_kprobe(p); | |
847 | } | |
2d1e38f5 | 848 | cpus_read_unlock(); |
b2be84df | 849 | printk(KERN_INFO "Kprobes globally optimized\n"); |
5c51543b MH |
850 | out: |
851 | mutex_unlock(&kprobe_mutex); | |
b2be84df MH |
852 | } |
853 | ||
55479f64 | 854 | static void unoptimize_all_kprobes(void) |
b2be84df MH |
855 | { |
856 | struct hlist_head *head; | |
b2be84df MH |
857 | struct kprobe *p; |
858 | unsigned int i; | |
859 | ||
5c51543b | 860 | mutex_lock(&kprobe_mutex); |
b2be84df | 861 | /* If optimization is already prohibited, just return */ |
5c51543b MH |
862 | if (!kprobes_allow_optimization) { |
863 | mutex_unlock(&kprobe_mutex); | |
b2be84df | 864 | return; |
5c51543b | 865 | } |
b2be84df | 866 | |
2d1e38f5 | 867 | cpus_read_lock(); |
b2be84df | 868 | kprobes_allow_optimization = false; |
b2be84df MH |
869 | for (i = 0; i < KPROBE_TABLE_SIZE; i++) { |
870 | head = &kprobe_table[i]; | |
b67bfe0d | 871 | hlist_for_each_entry_rcu(p, head, hlist) { |
b2be84df | 872 | if (!kprobe_disabled(p)) |
6274de49 | 873 | unoptimize_kprobe(p, false); |
b2be84df MH |
874 | } |
875 | } | |
2d1e38f5 | 876 | cpus_read_unlock(); |
5c51543b MH |
877 | mutex_unlock(&kprobe_mutex); |
878 | ||
6274de49 MH |
879 | /* Wait for unoptimizing completion */ |
880 | wait_for_kprobe_optimizer(); | |
881 | printk(KERN_INFO "Kprobes globally unoptimized\n"); | |
b2be84df MH |
882 | } |
883 | ||
5c51543b | 884 | static DEFINE_MUTEX(kprobe_sysctl_mutex); |
b2be84df MH |
885 | int sysctl_kprobes_optimization; |
886 | int proc_kprobes_optimization_handler(struct ctl_table *table, int write, | |
887 | void __user *buffer, size_t *length, | |
888 | loff_t *ppos) | |
889 | { | |
890 | int ret; | |
891 | ||
5c51543b | 892 | mutex_lock(&kprobe_sysctl_mutex); |
b2be84df MH |
893 | sysctl_kprobes_optimization = kprobes_allow_optimization ? 1 : 0; |
894 | ret = proc_dointvec_minmax(table, write, buffer, length, ppos); | |
895 | ||
896 | if (sysctl_kprobes_optimization) | |
897 | optimize_all_kprobes(); | |
898 | else | |
899 | unoptimize_all_kprobes(); | |
5c51543b | 900 | mutex_unlock(&kprobe_sysctl_mutex); |
b2be84df MH |
901 | |
902 | return ret; | |
903 | } | |
904 | #endif /* CONFIG_SYSCTL */ | |
905 | ||
6274de49 | 906 | /* Put a breakpoint for a probe. Must be called with text_mutex locked */ |
55479f64 | 907 | static void __arm_kprobe(struct kprobe *p) |
afd66255 | 908 | { |
6d8e40a8 | 909 | struct kprobe *_p; |
afd66255 MH |
910 | |
911 | /* Check collision with other optimized kprobes */ | |
6d8e40a8 MH |
912 | _p = get_optimized_kprobe((unsigned long)p->addr); |
913 | if (unlikely(_p)) | |
6274de49 MH |
914 | /* Fallback to unoptimized kprobe */ |
915 | unoptimize_kprobe(_p, true); | |
afd66255 MH |
916 | |
917 | arch_arm_kprobe(p); | |
918 | optimize_kprobe(p); /* Try to optimize (add kprobe to a list) */ | |
919 | } | |
920 | ||
6274de49 | 921 | /* Remove the breakpoint of a probe. Must be called with text_mutex locked */ |
55479f64 | 922 | static void __disarm_kprobe(struct kprobe *p, bool reopt) |
afd66255 | 923 | { |
6d8e40a8 | 924 | struct kprobe *_p; |
afd66255 | 925 | |
69d54b91 WN |
926 | /* Try to unoptimize */ |
927 | unoptimize_kprobe(p, kprobes_all_disarmed); | |
afd66255 | 928 | |
6274de49 MH |
929 | if (!kprobe_queued(p)) { |
930 | arch_disarm_kprobe(p); | |
931 | /* If another kprobe was blocked, optimize it. */ | |
932 | _p = get_optimized_kprobe((unsigned long)p->addr); | |
933 | if (unlikely(_p) && reopt) | |
934 | optimize_kprobe(_p); | |
935 | } | |
936 | /* TODO: reoptimize others after unoptimized this probe */ | |
afd66255 MH |
937 | } |
938 | ||
939 | #else /* !CONFIG_OPTPROBES */ | |
940 | ||
941 | #define optimize_kprobe(p) do {} while (0) | |
6274de49 | 942 | #define unoptimize_kprobe(p, f) do {} while (0) |
afd66255 MH |
943 | #define kill_optimized_kprobe(p) do {} while (0) |
944 | #define prepare_optimized_kprobe(p) do {} while (0) | |
945 | #define try_to_optimize_kprobe(p) do {} while (0) | |
946 | #define __arm_kprobe(p) arch_arm_kprobe(p) | |
6274de49 MH |
947 | #define __disarm_kprobe(p, o) arch_disarm_kprobe(p) |
948 | #define kprobe_disarmed(p) kprobe_disabled(p) | |
949 | #define wait_for_kprobe_optimizer() do {} while (0) | |
afd66255 | 950 | |
819319fc | 951 | static int reuse_unused_kprobe(struct kprobe *ap) |
0490cd1f | 952 | { |
819319fc MH |
953 | /* |
954 | * If the optimized kprobe is NOT supported, the aggr kprobe is | |
955 | * released at the same time that the last aggregated kprobe is | |
956 | * unregistered. | |
957 | * Thus there should be no chance to reuse unused kprobe. | |
958 | */ | |
0490cd1f | 959 | printk(KERN_ERR "Error: There should be no unused kprobe here.\n"); |
819319fc | 960 | return -EINVAL; |
0490cd1f MH |
961 | } |
962 | ||
55479f64 | 963 | static void free_aggr_kprobe(struct kprobe *p) |
afd66255 | 964 | { |
6274de49 | 965 | arch_remove_kprobe(p); |
afd66255 MH |
966 | kfree(p); |
967 | } | |
968 | ||
55479f64 | 969 | static struct kprobe *alloc_aggr_kprobe(struct kprobe *p) |
afd66255 MH |
970 | { |
971 | return kzalloc(sizeof(struct kprobe), GFP_KERNEL); | |
972 | } | |
973 | #endif /* CONFIG_OPTPROBES */ | |
974 | ||
e7dbfe34 | 975 | #ifdef CONFIG_KPROBES_ON_FTRACE |
ae6aa16f | 976 | static struct ftrace_ops kprobe_ftrace_ops __read_mostly = { |
e5253896 | 977 | .func = kprobe_ftrace_handler, |
1d70be34 | 978 | .flags = FTRACE_OPS_FL_SAVE_REGS | FTRACE_OPS_FL_IPMODIFY, |
ae6aa16f MH |
979 | }; |
980 | static int kprobe_ftrace_enabled; | |
981 | ||
982 | /* Must ensure p->addr is really on ftrace */ | |
55479f64 | 983 | static int prepare_kprobe(struct kprobe *p) |
ae6aa16f MH |
984 | { |
985 | if (!kprobe_ftrace(p)) | |
986 | return arch_prepare_kprobe(p); | |
987 | ||
988 | return arch_prepare_kprobe_ftrace(p); | |
989 | } | |
990 | ||
991 | /* Caller must lock kprobe_mutex */ | |
12310e34 | 992 | static int arm_kprobe_ftrace(struct kprobe *p) |
ae6aa16f | 993 | { |
12310e34 | 994 | int ret = 0; |
ae6aa16f MH |
995 | |
996 | ret = ftrace_set_filter_ip(&kprobe_ftrace_ops, | |
997 | (unsigned long)p->addr, 0, 0); | |
12310e34 | 998 | if (ret) { |
4458515b MH |
999 | pr_debug("Failed to arm kprobe-ftrace at %pS (%d)\n", |
1000 | p->addr, ret); | |
12310e34 JY |
1001 | return ret; |
1002 | } | |
1003 | ||
1004 | if (kprobe_ftrace_enabled == 0) { | |
ae6aa16f | 1005 | ret = register_ftrace_function(&kprobe_ftrace_ops); |
12310e34 JY |
1006 | if (ret) { |
1007 | pr_debug("Failed to init kprobe-ftrace (%d)\n", ret); | |
1008 | goto err_ftrace; | |
1009 | } | |
ae6aa16f | 1010 | } |
12310e34 JY |
1011 | |
1012 | kprobe_ftrace_enabled++; | |
1013 | return ret; | |
1014 | ||
1015 | err_ftrace: | |
1016 | /* | |
1017 | * Note: Since kprobe_ftrace_ops has IPMODIFY set, and ftrace requires a | |
1018 | * non-empty filter_hash for IPMODIFY ops, we're safe from an accidental | |
1019 | * empty filter_hash which would undesirably trace all functions. | |
1020 | */ | |
1021 | ftrace_set_filter_ip(&kprobe_ftrace_ops, (unsigned long)p->addr, 1, 0); | |
1022 | return ret; | |
ae6aa16f MH |
1023 | } |
1024 | ||
1025 | /* Caller must lock kprobe_mutex */ | |
297f9233 | 1026 | static int disarm_kprobe_ftrace(struct kprobe *p) |
ae6aa16f | 1027 | { |
297f9233 | 1028 | int ret = 0; |
ae6aa16f | 1029 | |
297f9233 | 1030 | if (kprobe_ftrace_enabled == 1) { |
ae6aa16f | 1031 | ret = unregister_ftrace_function(&kprobe_ftrace_ops); |
297f9233 JY |
1032 | if (WARN(ret < 0, "Failed to unregister kprobe-ftrace (%d)\n", ret)) |
1033 | return ret; | |
ae6aa16f | 1034 | } |
297f9233 JY |
1035 | |
1036 | kprobe_ftrace_enabled--; | |
1037 | ||
ae6aa16f MH |
1038 | ret = ftrace_set_filter_ip(&kprobe_ftrace_ops, |
1039 | (unsigned long)p->addr, 1, 0); | |
4458515b MH |
1040 | WARN_ONCE(ret < 0, "Failed to disarm kprobe-ftrace at %pS (%d)\n", |
1041 | p->addr, ret); | |
297f9233 | 1042 | return ret; |
ae6aa16f | 1043 | } |
e7dbfe34 | 1044 | #else /* !CONFIG_KPROBES_ON_FTRACE */ |
ae6aa16f | 1045 | #define prepare_kprobe(p) arch_prepare_kprobe(p) |
12310e34 | 1046 | #define arm_kprobe_ftrace(p) (-ENODEV) |
297f9233 | 1047 | #define disarm_kprobe_ftrace(p) (-ENODEV) |
ae6aa16f MH |
1048 | #endif |
1049 | ||
201517a7 | 1050 | /* Arm a kprobe with text_mutex */ |
12310e34 | 1051 | static int arm_kprobe(struct kprobe *kp) |
201517a7 | 1052 | { |
12310e34 JY |
1053 | if (unlikely(kprobe_ftrace(kp))) |
1054 | return arm_kprobe_ftrace(kp); | |
1055 | ||
2d1e38f5 | 1056 | cpus_read_lock(); |
201517a7 | 1057 | mutex_lock(&text_mutex); |
afd66255 | 1058 | __arm_kprobe(kp); |
201517a7 | 1059 | mutex_unlock(&text_mutex); |
2d1e38f5 | 1060 | cpus_read_unlock(); |
12310e34 JY |
1061 | |
1062 | return 0; | |
201517a7 MH |
1063 | } |
1064 | ||
1065 | /* Disarm a kprobe with text_mutex */ | |
297f9233 | 1066 | static int disarm_kprobe(struct kprobe *kp, bool reopt) |
201517a7 | 1067 | { |
297f9233 JY |
1068 | if (unlikely(kprobe_ftrace(kp))) |
1069 | return disarm_kprobe_ftrace(kp); | |
2d1e38f5 TG |
1070 | |
1071 | cpus_read_lock(); | |
201517a7 | 1072 | mutex_lock(&text_mutex); |
ae6aa16f | 1073 | __disarm_kprobe(kp, reopt); |
201517a7 | 1074 | mutex_unlock(&text_mutex); |
2d1e38f5 | 1075 | cpus_read_unlock(); |
297f9233 JY |
1076 | |
1077 | return 0; | |
201517a7 MH |
1078 | } |
1079 | ||
64f562c6 AM |
1080 | /* |
1081 | * Aggregate handlers for multiple kprobes support - these handlers | |
1082 | * take care of invoking the individual kprobe handlers on p->list | |
1083 | */ | |
820aede0 | 1084 | static int aggr_pre_handler(struct kprobe *p, struct pt_regs *regs) |
64f562c6 AM |
1085 | { |
1086 | struct kprobe *kp; | |
1087 | ||
3516a460 | 1088 | list_for_each_entry_rcu(kp, &p->list, list) { |
de5bd88d | 1089 | if (kp->pre_handler && likely(!kprobe_disabled(kp))) { |
e6584523 | 1090 | set_kprobe_instance(kp); |
8b0914ea PP |
1091 | if (kp->pre_handler(kp, regs)) |
1092 | return 1; | |
64f562c6 | 1093 | } |
e6584523 | 1094 | reset_kprobe_instance(); |
64f562c6 AM |
1095 | } |
1096 | return 0; | |
1097 | } | |
820aede0 | 1098 | NOKPROBE_SYMBOL(aggr_pre_handler); |
64f562c6 | 1099 | |
820aede0 MH |
1100 | static void aggr_post_handler(struct kprobe *p, struct pt_regs *regs, |
1101 | unsigned long flags) | |
64f562c6 AM |
1102 | { |
1103 | struct kprobe *kp; | |
1104 | ||
3516a460 | 1105 | list_for_each_entry_rcu(kp, &p->list, list) { |
de5bd88d | 1106 | if (kp->post_handler && likely(!kprobe_disabled(kp))) { |
e6584523 | 1107 | set_kprobe_instance(kp); |
64f562c6 | 1108 | kp->post_handler(kp, regs, flags); |
e6584523 | 1109 | reset_kprobe_instance(); |
64f562c6 AM |
1110 | } |
1111 | } | |
64f562c6 | 1112 | } |
820aede0 | 1113 | NOKPROBE_SYMBOL(aggr_post_handler); |
64f562c6 | 1114 | |
820aede0 MH |
1115 | static int aggr_fault_handler(struct kprobe *p, struct pt_regs *regs, |
1116 | int trapnr) | |
64f562c6 | 1117 | { |
b76834bc | 1118 | struct kprobe *cur = __this_cpu_read(kprobe_instance); |
e6584523 | 1119 | |
64f562c6 AM |
1120 | /* |
1121 | * if we faulted "during" the execution of a user specified | |
1122 | * probe handler, invoke just that probe's fault handler | |
1123 | */ | |
e6584523 AM |
1124 | if (cur && cur->fault_handler) { |
1125 | if (cur->fault_handler(cur, regs, trapnr)) | |
64f562c6 AM |
1126 | return 1; |
1127 | } | |
1128 | return 0; | |
1129 | } | |
820aede0 | 1130 | NOKPROBE_SYMBOL(aggr_fault_handler); |
64f562c6 | 1131 | |
bf8d5c52 | 1132 | /* Walks the list and increments nmissed count for multiprobe case */ |
820aede0 | 1133 | void kprobes_inc_nmissed_count(struct kprobe *p) |
bf8d5c52 KA |
1134 | { |
1135 | struct kprobe *kp; | |
afd66255 | 1136 | if (!kprobe_aggrprobe(p)) { |
bf8d5c52 KA |
1137 | p->nmissed++; |
1138 | } else { | |
1139 | list_for_each_entry_rcu(kp, &p->list, list) | |
1140 | kp->nmissed++; | |
1141 | } | |
1142 | return; | |
1143 | } | |
820aede0 | 1144 | NOKPROBE_SYMBOL(kprobes_inc_nmissed_count); |
bf8d5c52 | 1145 | |
820aede0 MH |
1146 | void recycle_rp_inst(struct kretprobe_instance *ri, |
1147 | struct hlist_head *head) | |
b94cce92 | 1148 | { |
ef53d9c5 S |
1149 | struct kretprobe *rp = ri->rp; |
1150 | ||
b94cce92 HN |
1151 | /* remove rp inst off the rprobe_inst_table */ |
1152 | hlist_del(&ri->hlist); | |
ef53d9c5 S |
1153 | INIT_HLIST_NODE(&ri->hlist); |
1154 | if (likely(rp)) { | |
ec484608 | 1155 | raw_spin_lock(&rp->lock); |
ef53d9c5 | 1156 | hlist_add_head(&ri->hlist, &rp->free_instances); |
ec484608 | 1157 | raw_spin_unlock(&rp->lock); |
b94cce92 HN |
1158 | } else |
1159 | /* Unregistering */ | |
99219a3f | 1160 | hlist_add_head(&ri->hlist, head); |
b94cce92 | 1161 | } |
820aede0 | 1162 | NOKPROBE_SYMBOL(recycle_rp_inst); |
b94cce92 | 1163 | |
820aede0 | 1164 | void kretprobe_hash_lock(struct task_struct *tsk, |
ef53d9c5 | 1165 | struct hlist_head **head, unsigned long *flags) |
635c17c2 | 1166 | __acquires(hlist_lock) |
ef53d9c5 S |
1167 | { |
1168 | unsigned long hash = hash_ptr(tsk, KPROBE_HASH_BITS); | |
ec484608 | 1169 | raw_spinlock_t *hlist_lock; |
ef53d9c5 S |
1170 | |
1171 | *head = &kretprobe_inst_table[hash]; | |
1172 | hlist_lock = kretprobe_table_lock_ptr(hash); | |
ec484608 | 1173 | raw_spin_lock_irqsave(hlist_lock, *flags); |
ef53d9c5 | 1174 | } |
820aede0 | 1175 | NOKPROBE_SYMBOL(kretprobe_hash_lock); |
ef53d9c5 | 1176 | |
820aede0 MH |
1177 | static void kretprobe_table_lock(unsigned long hash, |
1178 | unsigned long *flags) | |
635c17c2 | 1179 | __acquires(hlist_lock) |
b94cce92 | 1180 | { |
ec484608 TG |
1181 | raw_spinlock_t *hlist_lock = kretprobe_table_lock_ptr(hash); |
1182 | raw_spin_lock_irqsave(hlist_lock, *flags); | |
ef53d9c5 | 1183 | } |
820aede0 | 1184 | NOKPROBE_SYMBOL(kretprobe_table_lock); |
ef53d9c5 | 1185 | |
820aede0 MH |
1186 | void kretprobe_hash_unlock(struct task_struct *tsk, |
1187 | unsigned long *flags) | |
635c17c2 | 1188 | __releases(hlist_lock) |
ef53d9c5 S |
1189 | { |
1190 | unsigned long hash = hash_ptr(tsk, KPROBE_HASH_BITS); | |
ec484608 | 1191 | raw_spinlock_t *hlist_lock; |
ef53d9c5 S |
1192 | |
1193 | hlist_lock = kretprobe_table_lock_ptr(hash); | |
ec484608 | 1194 | raw_spin_unlock_irqrestore(hlist_lock, *flags); |
ef53d9c5 | 1195 | } |
820aede0 | 1196 | NOKPROBE_SYMBOL(kretprobe_hash_unlock); |
ef53d9c5 | 1197 | |
820aede0 MH |
1198 | static void kretprobe_table_unlock(unsigned long hash, |
1199 | unsigned long *flags) | |
635c17c2 | 1200 | __releases(hlist_lock) |
ef53d9c5 | 1201 | { |
ec484608 TG |
1202 | raw_spinlock_t *hlist_lock = kretprobe_table_lock_ptr(hash); |
1203 | raw_spin_unlock_irqrestore(hlist_lock, *flags); | |
b94cce92 | 1204 | } |
820aede0 | 1205 | NOKPROBE_SYMBOL(kretprobe_table_unlock); |
b94cce92 | 1206 | |
b94cce92 | 1207 | /* |
c6fd91f0 | 1208 | * This function is called from finish_task_switch when task tk becomes dead, |
1209 | * so that we can recycle any function-return probe instances associated | |
1210 | * with this task. These left over instances represent probed functions | |
1211 | * that have been called but will never return. | |
b94cce92 | 1212 | */ |
820aede0 | 1213 | void kprobe_flush_task(struct task_struct *tk) |
b94cce92 | 1214 | { |
62c27be0 | 1215 | struct kretprobe_instance *ri; |
99219a3f | 1216 | struct hlist_head *head, empty_rp; |
b67bfe0d | 1217 | struct hlist_node *tmp; |
ef53d9c5 | 1218 | unsigned long hash, flags = 0; |
802eae7c | 1219 | |
ef53d9c5 S |
1220 | if (unlikely(!kprobes_initialized)) |
1221 | /* Early boot. kretprobe_table_locks not yet initialized. */ | |
1222 | return; | |
1223 | ||
d496aab5 | 1224 | INIT_HLIST_HEAD(&empty_rp); |
ef53d9c5 S |
1225 | hash = hash_ptr(tk, KPROBE_HASH_BITS); |
1226 | head = &kretprobe_inst_table[hash]; | |
1227 | kretprobe_table_lock(hash, &flags); | |
b67bfe0d | 1228 | hlist_for_each_entry_safe(ri, tmp, head, hlist) { |
62c27be0 | 1229 | if (ri->task == tk) |
99219a3f | 1230 | recycle_rp_inst(ri, &empty_rp); |
62c27be0 | 1231 | } |
ef53d9c5 | 1232 | kretprobe_table_unlock(hash, &flags); |
b67bfe0d | 1233 | hlist_for_each_entry_safe(ri, tmp, &empty_rp, hlist) { |
99219a3f | 1234 | hlist_del(&ri->hlist); |
1235 | kfree(ri); | |
1236 | } | |
b94cce92 | 1237 | } |
820aede0 | 1238 | NOKPROBE_SYMBOL(kprobe_flush_task); |
b94cce92 | 1239 | |
b94cce92 HN |
1240 | static inline void free_rp_inst(struct kretprobe *rp) |
1241 | { | |
1242 | struct kretprobe_instance *ri; | |
b67bfe0d | 1243 | struct hlist_node *next; |
4c4308cb | 1244 | |
b67bfe0d | 1245 | hlist_for_each_entry_safe(ri, next, &rp->free_instances, hlist) { |
ef53d9c5 | 1246 | hlist_del(&ri->hlist); |
b94cce92 HN |
1247 | kfree(ri); |
1248 | } | |
1249 | } | |
1250 | ||
820aede0 | 1251 | static void cleanup_rp_inst(struct kretprobe *rp) |
4a296e07 | 1252 | { |
ef53d9c5 | 1253 | unsigned long flags, hash; |
4a296e07 | 1254 | struct kretprobe_instance *ri; |
b67bfe0d | 1255 | struct hlist_node *next; |
ef53d9c5 S |
1256 | struct hlist_head *head; |
1257 | ||
4a296e07 | 1258 | /* No race here */ |
ef53d9c5 S |
1259 | for (hash = 0; hash < KPROBE_TABLE_SIZE; hash++) { |
1260 | kretprobe_table_lock(hash, &flags); | |
1261 | head = &kretprobe_inst_table[hash]; | |
b67bfe0d | 1262 | hlist_for_each_entry_safe(ri, next, head, hlist) { |
ef53d9c5 S |
1263 | if (ri->rp == rp) |
1264 | ri->rp = NULL; | |
1265 | } | |
1266 | kretprobe_table_unlock(hash, &flags); | |
4a296e07 | 1267 | } |
4a296e07 MH |
1268 | free_rp_inst(rp); |
1269 | } | |
820aede0 | 1270 | NOKPROBE_SYMBOL(cleanup_rp_inst); |
4a296e07 | 1271 | |
059053a2 | 1272 | /* Add the new probe to ap->list */ |
55479f64 | 1273 | static int add_new_kprobe(struct kprobe *ap, struct kprobe *p) |
8b0914ea | 1274 | { |
059053a2 | 1275 | if (p->post_handler) |
6274de49 | 1276 | unoptimize_kprobe(ap, true); /* Fall back to normal kprobe */ |
afd66255 | 1277 | |
059053a2 | 1278 | list_add_rcu(&p->list, &ap->list); |
b918e5e6 MH |
1279 | if (p->post_handler && !ap->post_handler) |
1280 | ap->post_handler = aggr_post_handler; | |
de5bd88d | 1281 | |
8b0914ea PP |
1282 | return 0; |
1283 | } | |
1284 | ||
64f562c6 AM |
1285 | /* |
1286 | * Fill in the required fields of the "manager kprobe". Replace the | |
1287 | * earlier kprobe in the hlist with the manager kprobe | |
1288 | */ | |
55479f64 | 1289 | static void init_aggr_kprobe(struct kprobe *ap, struct kprobe *p) |
64f562c6 | 1290 | { |
afd66255 | 1291 | /* Copy p's insn slot to ap */ |
8b0914ea | 1292 | copy_kprobe(p, ap); |
a9ad965e | 1293 | flush_insn_slot(ap); |
64f562c6 | 1294 | ap->addr = p->addr; |
afd66255 | 1295 | ap->flags = p->flags & ~KPROBE_FLAG_OPTIMIZED; |
64f562c6 | 1296 | ap->pre_handler = aggr_pre_handler; |
64f562c6 | 1297 | ap->fault_handler = aggr_fault_handler; |
e8386a0c MH |
1298 | /* We don't care the kprobe which has gone. */ |
1299 | if (p->post_handler && !kprobe_gone(p)) | |
36721656 | 1300 | ap->post_handler = aggr_post_handler; |
64f562c6 AM |
1301 | |
1302 | INIT_LIST_HEAD(&ap->list); | |
afd66255 | 1303 | INIT_HLIST_NODE(&ap->hlist); |
64f562c6 | 1304 | |
afd66255 | 1305 | list_add_rcu(&p->list, &ap->list); |
adad0f33 | 1306 | hlist_replace_rcu(&p->hlist, &ap->hlist); |
64f562c6 AM |
1307 | } |
1308 | ||
1309 | /* | |
1310 | * This is the second or subsequent kprobe at the address - handle | |
1311 | * the intricacies | |
64f562c6 | 1312 | */ |
55479f64 | 1313 | static int register_aggr_kprobe(struct kprobe *orig_p, struct kprobe *p) |
64f562c6 AM |
1314 | { |
1315 | int ret = 0; | |
6d8e40a8 | 1316 | struct kprobe *ap = orig_p; |
64f562c6 | 1317 | |
2d1e38f5 TG |
1318 | cpus_read_lock(); |
1319 | ||
25764288 MH |
1320 | /* For preparing optimization, jump_label_text_reserved() is called */ |
1321 | jump_label_lock(); | |
25764288 MH |
1322 | mutex_lock(&text_mutex); |
1323 | ||
6d8e40a8 MH |
1324 | if (!kprobe_aggrprobe(orig_p)) { |
1325 | /* If orig_p is not an aggr_kprobe, create new aggr_kprobe. */ | |
1326 | ap = alloc_aggr_kprobe(orig_p); | |
25764288 MH |
1327 | if (!ap) { |
1328 | ret = -ENOMEM; | |
1329 | goto out; | |
1330 | } | |
6d8e40a8 | 1331 | init_aggr_kprobe(ap, orig_p); |
819319fc | 1332 | } else if (kprobe_unused(ap)) { |
0490cd1f | 1333 | /* This probe is going to die. Rescue it */ |
819319fc MH |
1334 | ret = reuse_unused_kprobe(ap); |
1335 | if (ret) | |
1336 | goto out; | |
1337 | } | |
b918e5e6 MH |
1338 | |
1339 | if (kprobe_gone(ap)) { | |
e8386a0c MH |
1340 | /* |
1341 | * Attempting to insert new probe at the same location that | |
1342 | * had a probe in the module vaddr area which already | |
1343 | * freed. So, the instruction slot has already been | |
1344 | * released. We need a new slot for the new probe. | |
1345 | */ | |
b918e5e6 | 1346 | ret = arch_prepare_kprobe(ap); |
e8386a0c | 1347 | if (ret) |
b918e5e6 MH |
1348 | /* |
1349 | * Even if fail to allocate new slot, don't need to | |
1350 | * free aggr_probe. It will be used next time, or | |
1351 | * freed by unregister_kprobe. | |
1352 | */ | |
25764288 | 1353 | goto out; |
de5bd88d | 1354 | |
afd66255 MH |
1355 | /* Prepare optimized instructions if possible. */ |
1356 | prepare_optimized_kprobe(ap); | |
1357 | ||
e8386a0c | 1358 | /* |
de5bd88d MH |
1359 | * Clear gone flag to prevent allocating new slot again, and |
1360 | * set disabled flag because it is not armed yet. | |
e8386a0c | 1361 | */ |
de5bd88d MH |
1362 | ap->flags = (ap->flags & ~KPROBE_FLAG_GONE) |
1363 | | KPROBE_FLAG_DISABLED; | |
e8386a0c | 1364 | } |
b918e5e6 | 1365 | |
afd66255 | 1366 | /* Copy ap's insn slot to p */ |
b918e5e6 | 1367 | copy_kprobe(ap, p); |
25764288 MH |
1368 | ret = add_new_kprobe(ap, p); |
1369 | ||
1370 | out: | |
1371 | mutex_unlock(&text_mutex); | |
25764288 | 1372 | jump_label_unlock(); |
2d1e38f5 | 1373 | cpus_read_unlock(); |
25764288 MH |
1374 | |
1375 | if (ret == 0 && kprobe_disabled(ap) && !kprobe_disabled(p)) { | |
1376 | ap->flags &= ~KPROBE_FLAG_DISABLED; | |
12310e34 | 1377 | if (!kprobes_all_disarmed) { |
25764288 | 1378 | /* Arm the breakpoint again. */ |
12310e34 JY |
1379 | ret = arm_kprobe(ap); |
1380 | if (ret) { | |
1381 | ap->flags |= KPROBE_FLAG_DISABLED; | |
1382 | list_del_rcu(&p->list); | |
ae8b7ce7 | 1383 | synchronize_rcu(); |
12310e34 JY |
1384 | } |
1385 | } | |
25764288 MH |
1386 | } |
1387 | return ret; | |
64f562c6 AM |
1388 | } |
1389 | ||
be8f2743 MH |
1390 | bool __weak arch_within_kprobe_blacklist(unsigned long addr) |
1391 | { | |
1392 | /* The __kprobes marked functions and entry code must not be probed */ | |
1393 | return addr >= (unsigned long)__kprobes_text_start && | |
1394 | addr < (unsigned long)__kprobes_text_end; | |
1395 | } | |
1396 | ||
e5779e8e | 1397 | bool within_kprobe_blacklist(unsigned long addr) |
d0aaff97 | 1398 | { |
376e2424 | 1399 | struct kprobe_blacklist_entry *ent; |
3d8d996e | 1400 | |
be8f2743 | 1401 | if (arch_within_kprobe_blacklist(addr)) |
376e2424 | 1402 | return true; |
3d8d996e SD |
1403 | /* |
1404 | * If there exists a kprobe_blacklist, verify and | |
1405 | * fail any probe registration in the prohibited area | |
1406 | */ | |
376e2424 MH |
1407 | list_for_each_entry(ent, &kprobe_blacklist, list) { |
1408 | if (addr >= ent->start_addr && addr < ent->end_addr) | |
1409 | return true; | |
3d8d996e | 1410 | } |
376e2424 MH |
1411 | |
1412 | return false; | |
d0aaff97 PP |
1413 | } |
1414 | ||
b2a5cd69 MH |
1415 | /* |
1416 | * If we have a symbol_name argument, look it up and add the offset field | |
1417 | * to it. This way, we can specify a relative address to a symbol. | |
bc81d48d MH |
1418 | * This returns encoded errors if it fails to look up symbol or invalid |
1419 | * combination of parameters. | |
b2a5cd69 | 1420 | */ |
1d585e70 NR |
1421 | static kprobe_opcode_t *_kprobe_addr(kprobe_opcode_t *addr, |
1422 | const char *symbol_name, unsigned int offset) | |
b2a5cd69 | 1423 | { |
1d585e70 | 1424 | if ((symbol_name && addr) || (!symbol_name && !addr)) |
bc81d48d MH |
1425 | goto invalid; |
1426 | ||
1d585e70 | 1427 | if (symbol_name) { |
7246f600 | 1428 | addr = kprobe_lookup_name(symbol_name, offset); |
bc81d48d MH |
1429 | if (!addr) |
1430 | return ERR_PTR(-ENOENT); | |
b2a5cd69 MH |
1431 | } |
1432 | ||
1d585e70 | 1433 | addr = (kprobe_opcode_t *)(((char *)addr) + offset); |
bc81d48d MH |
1434 | if (addr) |
1435 | return addr; | |
1436 | ||
1437 | invalid: | |
1438 | return ERR_PTR(-EINVAL); | |
b2a5cd69 MH |
1439 | } |
1440 | ||
1d585e70 NR |
1441 | static kprobe_opcode_t *kprobe_addr(struct kprobe *p) |
1442 | { | |
1443 | return _kprobe_addr(p->addr, p->symbol_name, p->offset); | |
1444 | } | |
1445 | ||
1f0ab409 | 1446 | /* Check passed kprobe is valid and return kprobe in kprobe_table. */ |
55479f64 | 1447 | static struct kprobe *__get_valid_kprobe(struct kprobe *p) |
1f0ab409 | 1448 | { |
6d8e40a8 | 1449 | struct kprobe *ap, *list_p; |
1f0ab409 | 1450 | |
6d8e40a8 MH |
1451 | ap = get_kprobe(p->addr); |
1452 | if (unlikely(!ap)) | |
1f0ab409 AM |
1453 | return NULL; |
1454 | ||
6d8e40a8 MH |
1455 | if (p != ap) { |
1456 | list_for_each_entry_rcu(list_p, &ap->list, list) | |
1f0ab409 AM |
1457 | if (list_p == p) |
1458 | /* kprobe p is a valid probe */ | |
1459 | goto valid; | |
1460 | return NULL; | |
1461 | } | |
1462 | valid: | |
6d8e40a8 | 1463 | return ap; |
1f0ab409 AM |
1464 | } |
1465 | ||
1466 | /* Return error if the kprobe is being re-registered */ | |
1467 | static inline int check_kprobe_rereg(struct kprobe *p) | |
1468 | { | |
1469 | int ret = 0; | |
1f0ab409 AM |
1470 | |
1471 | mutex_lock(&kprobe_mutex); | |
6d8e40a8 | 1472 | if (__get_valid_kprobe(p)) |
1f0ab409 AM |
1473 | ret = -EINVAL; |
1474 | mutex_unlock(&kprobe_mutex); | |
6d8e40a8 | 1475 | |
1f0ab409 AM |
1476 | return ret; |
1477 | } | |
1478 | ||
f7f242ff | 1479 | int __weak arch_check_ftrace_location(struct kprobe *p) |
1da177e4 | 1480 | { |
ae6aa16f MH |
1481 | unsigned long ftrace_addr; |
1482 | ||
ae6aa16f MH |
1483 | ftrace_addr = ftrace_location((unsigned long)p->addr); |
1484 | if (ftrace_addr) { | |
e7dbfe34 | 1485 | #ifdef CONFIG_KPROBES_ON_FTRACE |
ae6aa16f MH |
1486 | /* Given address is not on the instruction boundary */ |
1487 | if ((unsigned long)p->addr != ftrace_addr) | |
1488 | return -EILSEQ; | |
ae6aa16f | 1489 | p->flags |= KPROBE_FLAG_FTRACE; |
e7dbfe34 | 1490 | #else /* !CONFIG_KPROBES_ON_FTRACE */ |
ae6aa16f MH |
1491 | return -EINVAL; |
1492 | #endif | |
1493 | } | |
f7f242ff HC |
1494 | return 0; |
1495 | } | |
1496 | ||
1497 | static int check_kprobe_address_safe(struct kprobe *p, | |
1498 | struct module **probed_mod) | |
1499 | { | |
1500 | int ret; | |
1f0ab409 | 1501 | |
f7f242ff HC |
1502 | ret = arch_check_ftrace_location(p); |
1503 | if (ret) | |
1504 | return ret; | |
91bad2f8 | 1505 | jump_label_lock(); |
de31c3ca | 1506 | preempt_disable(); |
f7fa6ef0 MH |
1507 | |
1508 | /* Ensure it is not in reserved area nor out of text */ | |
ec30c5f3 | 1509 | if (!kernel_text_address((unsigned long) p->addr) || |
376e2424 | 1510 | within_kprobe_blacklist((unsigned long) p->addr) || |
f986a499 PN |
1511 | jump_label_text_reserved(p->addr, p->addr)) { |
1512 | ret = -EINVAL; | |
f7fa6ef0 | 1513 | goto out; |
f986a499 | 1514 | } |
b3e55c72 | 1515 | |
f7fa6ef0 MH |
1516 | /* Check if are we probing a module */ |
1517 | *probed_mod = __module_text_address((unsigned long) p->addr); | |
1518 | if (*probed_mod) { | |
6f716acd | 1519 | /* |
e8386a0c MH |
1520 | * We must hold a refcount of the probed module while updating |
1521 | * its code to prohibit unexpected unloading. | |
df019b1d | 1522 | */ |
f7fa6ef0 MH |
1523 | if (unlikely(!try_module_get(*probed_mod))) { |
1524 | ret = -ENOENT; | |
1525 | goto out; | |
1526 | } | |
de31c3ca | 1527 | |
f24659d9 MH |
1528 | /* |
1529 | * If the module freed .init.text, we couldn't insert | |
1530 | * kprobes in there. | |
1531 | */ | |
f7fa6ef0 MH |
1532 | if (within_module_init((unsigned long)p->addr, *probed_mod) && |
1533 | (*probed_mod)->state != MODULE_STATE_COMING) { | |
1534 | module_put(*probed_mod); | |
1535 | *probed_mod = NULL; | |
1536 | ret = -ENOENT; | |
f24659d9 | 1537 | } |
df019b1d | 1538 | } |
f7fa6ef0 | 1539 | out: |
a189d035 | 1540 | preempt_enable(); |
de31c3ca | 1541 | jump_label_unlock(); |
1da177e4 | 1542 | |
f7fa6ef0 MH |
1543 | return ret; |
1544 | } | |
1545 | ||
55479f64 | 1546 | int register_kprobe(struct kprobe *p) |
f7fa6ef0 MH |
1547 | { |
1548 | int ret; | |
1549 | struct kprobe *old_p; | |
1550 | struct module *probed_mod; | |
1551 | kprobe_opcode_t *addr; | |
1552 | ||
13a1ddd0 DH |
1553 | if (kernel_is_locked_down("Use of kprobes")) |
1554 | return -EPERM; | |
1555 | ||
f7fa6ef0 MH |
1556 | /* Adjust probe address from symbol */ |
1557 | addr = kprobe_addr(p); | |
1558 | if (IS_ERR(addr)) | |
1559 | return PTR_ERR(addr); | |
1560 | p->addr = addr; | |
1561 | ||
1562 | ret = check_kprobe_rereg(p); | |
1563 | if (ret) | |
1564 | return ret; | |
1565 | ||
1566 | /* User can pass only KPROBE_FLAG_DISABLED to register_kprobe */ | |
1567 | p->flags &= KPROBE_FLAG_DISABLED; | |
3516a460 | 1568 | p->nmissed = 0; |
9861668f | 1569 | INIT_LIST_HEAD(&p->list); |
afd66255 | 1570 | |
f7fa6ef0 MH |
1571 | ret = check_kprobe_address_safe(p, &probed_mod); |
1572 | if (ret) | |
1573 | return ret; | |
1574 | ||
1575 | mutex_lock(&kprobe_mutex); | |
afd66255 | 1576 | |
64f562c6 AM |
1577 | old_p = get_kprobe(p->addr); |
1578 | if (old_p) { | |
afd66255 | 1579 | /* Since this may unoptimize old_p, locking text_mutex. */ |
64f562c6 | 1580 | ret = register_aggr_kprobe(old_p, p); |
1da177e4 LT |
1581 | goto out; |
1582 | } | |
1da177e4 | 1583 | |
2d1e38f5 TG |
1584 | cpus_read_lock(); |
1585 | /* Prevent text modification */ | |
1586 | mutex_lock(&text_mutex); | |
ae6aa16f | 1587 | ret = prepare_kprobe(p); |
25764288 | 1588 | mutex_unlock(&text_mutex); |
2d1e38f5 | 1589 | cpus_read_unlock(); |
6f716acd | 1590 | if (ret) |
afd66255 | 1591 | goto out; |
49a2a1b8 | 1592 | |
64f562c6 | 1593 | INIT_HLIST_NODE(&p->hlist); |
3516a460 | 1594 | hlist_add_head_rcu(&p->hlist, |
1da177e4 LT |
1595 | &kprobe_table[hash_ptr(p->addr, KPROBE_HASH_BITS)]); |
1596 | ||
12310e34 JY |
1597 | if (!kprobes_all_disarmed && !kprobe_disabled(p)) { |
1598 | ret = arm_kprobe(p); | |
1599 | if (ret) { | |
1600 | hlist_del_rcu(&p->hlist); | |
ae8b7ce7 | 1601 | synchronize_rcu(); |
12310e34 JY |
1602 | goto out; |
1603 | } | |
1604 | } | |
afd66255 MH |
1605 | |
1606 | /* Try to optimize kprobe */ | |
1607 | try_to_optimize_kprobe(p); | |
1da177e4 | 1608 | out: |
7a7d1cf9 | 1609 | mutex_unlock(&kprobe_mutex); |
49a2a1b8 | 1610 | |
e8386a0c | 1611 | if (probed_mod) |
df019b1d | 1612 | module_put(probed_mod); |
e8386a0c | 1613 | |
1da177e4 LT |
1614 | return ret; |
1615 | } | |
99081ab5 | 1616 | EXPORT_SYMBOL_GPL(register_kprobe); |
1da177e4 | 1617 | |
6f0f1dd7 | 1618 | /* Check if all probes on the aggrprobe are disabled */ |
55479f64 | 1619 | static int aggr_kprobe_disabled(struct kprobe *ap) |
6f0f1dd7 MH |
1620 | { |
1621 | struct kprobe *kp; | |
1622 | ||
1623 | list_for_each_entry_rcu(kp, &ap->list, list) | |
1624 | if (!kprobe_disabled(kp)) | |
1625 | /* | |
1626 | * There is an active probe on the list. | |
1627 | * We can't disable this ap. | |
1628 | */ | |
1629 | return 0; | |
1630 | ||
1631 | return 1; | |
1632 | } | |
1633 | ||
1634 | /* Disable one kprobe: Make sure called under kprobe_mutex is locked */ | |
55479f64 | 1635 | static struct kprobe *__disable_kprobe(struct kprobe *p) |
6f0f1dd7 MH |
1636 | { |
1637 | struct kprobe *orig_p; | |
297f9233 | 1638 | int ret; |
6f0f1dd7 MH |
1639 | |
1640 | /* Get an original kprobe for return */ | |
1641 | orig_p = __get_valid_kprobe(p); | |
1642 | if (unlikely(orig_p == NULL)) | |
297f9233 | 1643 | return ERR_PTR(-EINVAL); |
6f0f1dd7 MH |
1644 | |
1645 | if (!kprobe_disabled(p)) { | |
1646 | /* Disable probe if it is a child probe */ | |
1647 | if (p != orig_p) | |
1648 | p->flags |= KPROBE_FLAG_DISABLED; | |
1649 | ||
1650 | /* Try to disarm and disable this/parent probe */ | |
1651 | if (p == orig_p || aggr_kprobe_disabled(orig_p)) { | |
69d54b91 WN |
1652 | /* |
1653 | * If kprobes_all_disarmed is set, orig_p | |
1654 | * should have already been disarmed, so | |
1655 | * skip unneed disarming process. | |
1656 | */ | |
297f9233 JY |
1657 | if (!kprobes_all_disarmed) { |
1658 | ret = disarm_kprobe(orig_p, true); | |
1659 | if (ret) { | |
1660 | p->flags &= ~KPROBE_FLAG_DISABLED; | |
1661 | return ERR_PTR(ret); | |
1662 | } | |
1663 | } | |
6f0f1dd7 MH |
1664 | orig_p->flags |= KPROBE_FLAG_DISABLED; |
1665 | } | |
1666 | } | |
1667 | ||
1668 | return orig_p; | |
1669 | } | |
1670 | ||
de5bd88d MH |
1671 | /* |
1672 | * Unregister a kprobe without a scheduler synchronization. | |
1673 | */ | |
55479f64 | 1674 | static int __unregister_kprobe_top(struct kprobe *p) |
de5bd88d | 1675 | { |
6d8e40a8 | 1676 | struct kprobe *ap, *list_p; |
de5bd88d | 1677 | |
6f0f1dd7 MH |
1678 | /* Disable kprobe. This will disarm it if needed. */ |
1679 | ap = __disable_kprobe(p); | |
297f9233 JY |
1680 | if (IS_ERR(ap)) |
1681 | return PTR_ERR(ap); | |
de5bd88d | 1682 | |
6f0f1dd7 | 1683 | if (ap == p) |
bf8f6e5b | 1684 | /* |
6f0f1dd7 MH |
1685 | * This probe is an independent(and non-optimized) kprobe |
1686 | * (not an aggrprobe). Remove from the hash list. | |
bf8f6e5b | 1687 | */ |
6f0f1dd7 MH |
1688 | goto disarmed; |
1689 | ||
1690 | /* Following process expects this probe is an aggrprobe */ | |
1691 | WARN_ON(!kprobe_aggrprobe(ap)); | |
1692 | ||
6274de49 MH |
1693 | if (list_is_singular(&ap->list) && kprobe_disarmed(ap)) |
1694 | /* | |
1695 | * !disarmed could be happen if the probe is under delayed | |
1696 | * unoptimizing. | |
1697 | */ | |
6f0f1dd7 MH |
1698 | goto disarmed; |
1699 | else { | |
1700 | /* If disabling probe has special handlers, update aggrprobe */ | |
e8386a0c | 1701 | if (p->post_handler && !kprobe_gone(p)) { |
6d8e40a8 | 1702 | list_for_each_entry_rcu(list_p, &ap->list, list) { |
9861668f MH |
1703 | if ((list_p != p) && (list_p->post_handler)) |
1704 | goto noclean; | |
1705 | } | |
6d8e40a8 | 1706 | ap->post_handler = NULL; |
9861668f MH |
1707 | } |
1708 | noclean: | |
6f0f1dd7 MH |
1709 | /* |
1710 | * Remove from the aggrprobe: this path will do nothing in | |
1711 | * __unregister_kprobe_bottom(). | |
1712 | */ | |
49a2a1b8 | 1713 | list_del_rcu(&p->list); |
6f0f1dd7 MH |
1714 | if (!kprobe_disabled(ap) && !kprobes_all_disarmed) |
1715 | /* | |
1716 | * Try to optimize this probe again, because post | |
1717 | * handler may have been changed. | |
1718 | */ | |
1719 | optimize_kprobe(ap); | |
49a2a1b8 | 1720 | } |
9861668f | 1721 | return 0; |
6f0f1dd7 MH |
1722 | |
1723 | disarmed: | |
1724 | hlist_del_rcu(&ap->hlist); | |
1725 | return 0; | |
9861668f | 1726 | } |
3516a460 | 1727 | |
55479f64 | 1728 | static void __unregister_kprobe_bottom(struct kprobe *p) |
9861668f | 1729 | { |
6d8e40a8 | 1730 | struct kprobe *ap; |
b3e55c72 | 1731 | |
e8386a0c | 1732 | if (list_empty(&p->list)) |
6274de49 | 1733 | /* This is an independent kprobe */ |
0498b635 | 1734 | arch_remove_kprobe(p); |
e8386a0c | 1735 | else if (list_is_singular(&p->list)) { |
6274de49 | 1736 | /* This is the last child of an aggrprobe */ |
6d8e40a8 | 1737 | ap = list_entry(p->list.next, struct kprobe, list); |
e8386a0c | 1738 | list_del(&p->list); |
6d8e40a8 | 1739 | free_aggr_kprobe(ap); |
9861668f | 1740 | } |
6274de49 | 1741 | /* Otherwise, do nothing. */ |
9861668f MH |
1742 | } |
1743 | ||
55479f64 | 1744 | int register_kprobes(struct kprobe **kps, int num) |
9861668f MH |
1745 | { |
1746 | int i, ret = 0; | |
1747 | ||
1748 | if (num <= 0) | |
1749 | return -EINVAL; | |
1750 | for (i = 0; i < num; i++) { | |
49ad2fd7 | 1751 | ret = register_kprobe(kps[i]); |
67dddaad MH |
1752 | if (ret < 0) { |
1753 | if (i > 0) | |
1754 | unregister_kprobes(kps, i); | |
9861668f | 1755 | break; |
36721656 | 1756 | } |
49a2a1b8 | 1757 | } |
9861668f MH |
1758 | return ret; |
1759 | } | |
99081ab5 | 1760 | EXPORT_SYMBOL_GPL(register_kprobes); |
9861668f | 1761 | |
55479f64 | 1762 | void unregister_kprobe(struct kprobe *p) |
9861668f MH |
1763 | { |
1764 | unregister_kprobes(&p, 1); | |
1765 | } | |
99081ab5 | 1766 | EXPORT_SYMBOL_GPL(unregister_kprobe); |
9861668f | 1767 | |
55479f64 | 1768 | void unregister_kprobes(struct kprobe **kps, int num) |
9861668f MH |
1769 | { |
1770 | int i; | |
1771 | ||
1772 | if (num <= 0) | |
1773 | return; | |
1774 | mutex_lock(&kprobe_mutex); | |
1775 | for (i = 0; i < num; i++) | |
1776 | if (__unregister_kprobe_top(kps[i]) < 0) | |
1777 | kps[i]->addr = NULL; | |
1778 | mutex_unlock(&kprobe_mutex); | |
1779 | ||
ae8b7ce7 | 1780 | synchronize_rcu(); |
9861668f MH |
1781 | for (i = 0; i < num; i++) |
1782 | if (kps[i]->addr) | |
1783 | __unregister_kprobe_bottom(kps[i]); | |
1da177e4 | 1784 | } |
99081ab5 | 1785 | EXPORT_SYMBOL_GPL(unregister_kprobes); |
1da177e4 | 1786 | |
5f6bee34 NR |
1787 | int __weak kprobe_exceptions_notify(struct notifier_block *self, |
1788 | unsigned long val, void *data) | |
fc62d020 NR |
1789 | { |
1790 | return NOTIFY_DONE; | |
1791 | } | |
5f6bee34 | 1792 | NOKPROBE_SYMBOL(kprobe_exceptions_notify); |
fc62d020 | 1793 | |
1da177e4 | 1794 | static struct notifier_block kprobe_exceptions_nb = { |
3d5631e0 AK |
1795 | .notifier_call = kprobe_exceptions_notify, |
1796 | .priority = 0x7fffffff /* we need to be notified first */ | |
1797 | }; | |
1798 | ||
3d7e3382 ME |
1799 | unsigned long __weak arch_deref_entry_point(void *entry) |
1800 | { | |
1801 | return (unsigned long)entry; | |
1802 | } | |
1da177e4 | 1803 | |
9edddaa2 | 1804 | #ifdef CONFIG_KRETPROBES |
e65cefe8 AB |
1805 | /* |
1806 | * This kprobe pre_handler is registered with every kretprobe. When probe | |
1807 | * hits it will set up the return probe. | |
1808 | */ | |
820aede0 | 1809 | static int pre_handler_kretprobe(struct kprobe *p, struct pt_regs *regs) |
e65cefe8 AB |
1810 | { |
1811 | struct kretprobe *rp = container_of(p, struct kretprobe, kp); | |
ef53d9c5 S |
1812 | unsigned long hash, flags = 0; |
1813 | struct kretprobe_instance *ri; | |
e65cefe8 | 1814 | |
f96f5678 MH |
1815 | /* |
1816 | * To avoid deadlocks, prohibit return probing in NMI contexts, | |
1817 | * just skip the probe and increase the (inexact) 'nmissed' | |
1818 | * statistical counter, so that the user is informed that | |
1819 | * something happened: | |
1820 | */ | |
1821 | if (unlikely(in_nmi())) { | |
1822 | rp->nmissed++; | |
1823 | return 0; | |
1824 | } | |
1825 | ||
1826 | /* TODO: consider to only swap the RA after the last pre_handler fired */ | |
ef53d9c5 | 1827 | hash = hash_ptr(current, KPROBE_HASH_BITS); |
ec484608 | 1828 | raw_spin_lock_irqsave(&rp->lock, flags); |
4c4308cb | 1829 | if (!hlist_empty(&rp->free_instances)) { |
4c4308cb | 1830 | ri = hlist_entry(rp->free_instances.first, |
ef53d9c5 S |
1831 | struct kretprobe_instance, hlist); |
1832 | hlist_del(&ri->hlist); | |
ec484608 | 1833 | raw_spin_unlock_irqrestore(&rp->lock, flags); |
ef53d9c5 | 1834 | |
4c4308cb CH |
1835 | ri->rp = rp; |
1836 | ri->task = current; | |
f47cd9b5 | 1837 | |
55ca6140 JL |
1838 | if (rp->entry_handler && rp->entry_handler(ri, regs)) { |
1839 | raw_spin_lock_irqsave(&rp->lock, flags); | |
1840 | hlist_add_head(&ri->hlist, &rp->free_instances); | |
1841 | raw_spin_unlock_irqrestore(&rp->lock, flags); | |
f47cd9b5 | 1842 | return 0; |
55ca6140 | 1843 | } |
f47cd9b5 | 1844 | |
4c4308cb CH |
1845 | arch_prepare_kretprobe(ri, regs); |
1846 | ||
1847 | /* XXX(hch): why is there no hlist_move_head? */ | |
ef53d9c5 S |
1848 | INIT_HLIST_NODE(&ri->hlist); |
1849 | kretprobe_table_lock(hash, &flags); | |
1850 | hlist_add_head(&ri->hlist, &kretprobe_inst_table[hash]); | |
1851 | kretprobe_table_unlock(hash, &flags); | |
1852 | } else { | |
4c4308cb | 1853 | rp->nmissed++; |
ec484608 | 1854 | raw_spin_unlock_irqrestore(&rp->lock, flags); |
ef53d9c5 | 1855 | } |
e65cefe8 AB |
1856 | return 0; |
1857 | } | |
820aede0 | 1858 | NOKPROBE_SYMBOL(pre_handler_kretprobe); |
e65cefe8 | 1859 | |
659b957f | 1860 | bool __weak arch_kprobe_on_func_entry(unsigned long offset) |
90ec5e89 NR |
1861 | { |
1862 | return !offset; | |
1863 | } | |
1864 | ||
659b957f | 1865 | bool kprobe_on_func_entry(kprobe_opcode_t *addr, const char *sym, unsigned long offset) |
1d585e70 NR |
1866 | { |
1867 | kprobe_opcode_t *kp_addr = _kprobe_addr(addr, sym, offset); | |
1868 | ||
1869 | if (IS_ERR(kp_addr)) | |
1870 | return false; | |
1871 | ||
1872 | if (!kallsyms_lookup_size_offset((unsigned long)kp_addr, NULL, &offset) || | |
659b957f | 1873 | !arch_kprobe_on_func_entry(offset)) |
1d585e70 NR |
1874 | return false; |
1875 | ||
1876 | return true; | |
1877 | } | |
1878 | ||
55479f64 | 1879 | int register_kretprobe(struct kretprobe *rp) |
b94cce92 HN |
1880 | { |
1881 | int ret = 0; | |
1882 | struct kretprobe_instance *inst; | |
1883 | int i; | |
b2a5cd69 | 1884 | void *addr; |
90ec5e89 | 1885 | |
659b957f | 1886 | if (!kprobe_on_func_entry(rp->kp.addr, rp->kp.symbol_name, rp->kp.offset)) |
90ec5e89 | 1887 | return -EINVAL; |
f438d914 MH |
1888 | |
1889 | if (kretprobe_blacklist_size) { | |
b2a5cd69 | 1890 | addr = kprobe_addr(&rp->kp); |
bc81d48d MH |
1891 | if (IS_ERR(addr)) |
1892 | return PTR_ERR(addr); | |
f438d914 MH |
1893 | |
1894 | for (i = 0; kretprobe_blacklist[i].name != NULL; i++) { | |
1895 | if (kretprobe_blacklist[i].addr == addr) | |
1896 | return -EINVAL; | |
1897 | } | |
1898 | } | |
b94cce92 HN |
1899 | |
1900 | rp->kp.pre_handler = pre_handler_kretprobe; | |
7522a842 AM |
1901 | rp->kp.post_handler = NULL; |
1902 | rp->kp.fault_handler = NULL; | |
b94cce92 HN |
1903 | |
1904 | /* Pre-allocate memory for max kretprobe instances */ | |
1905 | if (rp->maxactive <= 0) { | |
1906 | #ifdef CONFIG_PREEMPT | |
c2ef6661 | 1907 | rp->maxactive = max_t(unsigned int, 10, 2*num_possible_cpus()); |
b94cce92 | 1908 | #else |
4dae560f | 1909 | rp->maxactive = num_possible_cpus(); |
b94cce92 HN |
1910 | #endif |
1911 | } | |
ec484608 | 1912 | raw_spin_lock_init(&rp->lock); |
b94cce92 HN |
1913 | INIT_HLIST_HEAD(&rp->free_instances); |
1914 | for (i = 0; i < rp->maxactive; i++) { | |
f47cd9b5 AS |
1915 | inst = kmalloc(sizeof(struct kretprobe_instance) + |
1916 | rp->data_size, GFP_KERNEL); | |
b94cce92 HN |
1917 | if (inst == NULL) { |
1918 | free_rp_inst(rp); | |
1919 | return -ENOMEM; | |
1920 | } | |
ef53d9c5 S |
1921 | INIT_HLIST_NODE(&inst->hlist); |
1922 | hlist_add_head(&inst->hlist, &rp->free_instances); | |
b94cce92 HN |
1923 | } |
1924 | ||
1925 | rp->nmissed = 0; | |
1926 | /* Establish function entry probe point */ | |
49ad2fd7 | 1927 | ret = register_kprobe(&rp->kp); |
4a296e07 | 1928 | if (ret != 0) |
b94cce92 HN |
1929 | free_rp_inst(rp); |
1930 | return ret; | |
1931 | } | |
99081ab5 | 1932 | EXPORT_SYMBOL_GPL(register_kretprobe); |
b94cce92 | 1933 | |
55479f64 | 1934 | int register_kretprobes(struct kretprobe **rps, int num) |
4a296e07 MH |
1935 | { |
1936 | int ret = 0, i; | |
1937 | ||
1938 | if (num <= 0) | |
1939 | return -EINVAL; | |
1940 | for (i = 0; i < num; i++) { | |
49ad2fd7 | 1941 | ret = register_kretprobe(rps[i]); |
67dddaad MH |
1942 | if (ret < 0) { |
1943 | if (i > 0) | |
1944 | unregister_kretprobes(rps, i); | |
4a296e07 MH |
1945 | break; |
1946 | } | |
1947 | } | |
1948 | return ret; | |
1949 | } | |
99081ab5 | 1950 | EXPORT_SYMBOL_GPL(register_kretprobes); |
4a296e07 | 1951 | |
55479f64 | 1952 | void unregister_kretprobe(struct kretprobe *rp) |
4a296e07 MH |
1953 | { |
1954 | unregister_kretprobes(&rp, 1); | |
1955 | } | |
99081ab5 | 1956 | EXPORT_SYMBOL_GPL(unregister_kretprobe); |
4a296e07 | 1957 | |
55479f64 | 1958 | void unregister_kretprobes(struct kretprobe **rps, int num) |
4a296e07 MH |
1959 | { |
1960 | int i; | |
1961 | ||
1962 | if (num <= 0) | |
1963 | return; | |
1964 | mutex_lock(&kprobe_mutex); | |
1965 | for (i = 0; i < num; i++) | |
1966 | if (__unregister_kprobe_top(&rps[i]->kp) < 0) | |
1967 | rps[i]->kp.addr = NULL; | |
1968 | mutex_unlock(&kprobe_mutex); | |
1969 | ||
ae8b7ce7 | 1970 | synchronize_rcu(); |
4a296e07 MH |
1971 | for (i = 0; i < num; i++) { |
1972 | if (rps[i]->kp.addr) { | |
1973 | __unregister_kprobe_bottom(&rps[i]->kp); | |
1974 | cleanup_rp_inst(rps[i]); | |
1975 | } | |
1976 | } | |
1977 | } | |
99081ab5 | 1978 | EXPORT_SYMBOL_GPL(unregister_kretprobes); |
4a296e07 | 1979 | |
9edddaa2 | 1980 | #else /* CONFIG_KRETPROBES */ |
55479f64 | 1981 | int register_kretprobe(struct kretprobe *rp) |
b94cce92 HN |
1982 | { |
1983 | return -ENOSYS; | |
1984 | } | |
99081ab5 | 1985 | EXPORT_SYMBOL_GPL(register_kretprobe); |
b94cce92 | 1986 | |
55479f64 | 1987 | int register_kretprobes(struct kretprobe **rps, int num) |
346fd59b | 1988 | { |
4a296e07 | 1989 | return -ENOSYS; |
346fd59b | 1990 | } |
99081ab5 MH |
1991 | EXPORT_SYMBOL_GPL(register_kretprobes); |
1992 | ||
55479f64 | 1993 | void unregister_kretprobe(struct kretprobe *rp) |
b94cce92 | 1994 | { |
4a296e07 | 1995 | } |
99081ab5 | 1996 | EXPORT_SYMBOL_GPL(unregister_kretprobe); |
b94cce92 | 1997 | |
55479f64 | 1998 | void unregister_kretprobes(struct kretprobe **rps, int num) |
4a296e07 MH |
1999 | { |
2000 | } | |
99081ab5 | 2001 | EXPORT_SYMBOL_GPL(unregister_kretprobes); |
4c4308cb | 2002 | |
820aede0 | 2003 | static int pre_handler_kretprobe(struct kprobe *p, struct pt_regs *regs) |
4a296e07 MH |
2004 | { |
2005 | return 0; | |
b94cce92 | 2006 | } |
820aede0 | 2007 | NOKPROBE_SYMBOL(pre_handler_kretprobe); |
b94cce92 | 2008 | |
4a296e07 MH |
2009 | #endif /* CONFIG_KRETPROBES */ |
2010 | ||
e8386a0c | 2011 | /* Set the kprobe gone and remove its instruction buffer. */ |
55479f64 | 2012 | static void kill_kprobe(struct kprobe *p) |
e8386a0c MH |
2013 | { |
2014 | struct kprobe *kp; | |
de5bd88d | 2015 | |
e8386a0c | 2016 | p->flags |= KPROBE_FLAG_GONE; |
afd66255 | 2017 | if (kprobe_aggrprobe(p)) { |
e8386a0c MH |
2018 | /* |
2019 | * If this is an aggr_kprobe, we have to list all the | |
2020 | * chained probes and mark them GONE. | |
2021 | */ | |
2022 | list_for_each_entry_rcu(kp, &p->list, list) | |
2023 | kp->flags |= KPROBE_FLAG_GONE; | |
2024 | p->post_handler = NULL; | |
afd66255 | 2025 | kill_optimized_kprobe(p); |
e8386a0c MH |
2026 | } |
2027 | /* | |
2028 | * Here, we can remove insn_slot safely, because no thread calls | |
2029 | * the original probed function (which will be freed soon) any more. | |
2030 | */ | |
2031 | arch_remove_kprobe(p); | |
2032 | } | |
2033 | ||
c0614829 | 2034 | /* Disable one kprobe */ |
55479f64 | 2035 | int disable_kprobe(struct kprobe *kp) |
c0614829 MH |
2036 | { |
2037 | int ret = 0; | |
297f9233 | 2038 | struct kprobe *p; |
c0614829 MH |
2039 | |
2040 | mutex_lock(&kprobe_mutex); | |
2041 | ||
6f0f1dd7 | 2042 | /* Disable this kprobe */ |
297f9233 JY |
2043 | p = __disable_kprobe(kp); |
2044 | if (IS_ERR(p)) | |
2045 | ret = PTR_ERR(p); | |
c0614829 | 2046 | |
c0614829 MH |
2047 | mutex_unlock(&kprobe_mutex); |
2048 | return ret; | |
2049 | } | |
2050 | EXPORT_SYMBOL_GPL(disable_kprobe); | |
2051 | ||
2052 | /* Enable one kprobe */ | |
55479f64 | 2053 | int enable_kprobe(struct kprobe *kp) |
c0614829 MH |
2054 | { |
2055 | int ret = 0; | |
2056 | struct kprobe *p; | |
2057 | ||
2058 | mutex_lock(&kprobe_mutex); | |
2059 | ||
2060 | /* Check whether specified probe is valid. */ | |
2061 | p = __get_valid_kprobe(kp); | |
2062 | if (unlikely(p == NULL)) { | |
2063 | ret = -EINVAL; | |
2064 | goto out; | |
2065 | } | |
2066 | ||
2067 | if (kprobe_gone(kp)) { | |
2068 | /* This kprobe has gone, we couldn't enable it. */ | |
2069 | ret = -EINVAL; | |
2070 | goto out; | |
2071 | } | |
2072 | ||
2073 | if (p != kp) | |
2074 | kp->flags &= ~KPROBE_FLAG_DISABLED; | |
2075 | ||
2076 | if (!kprobes_all_disarmed && kprobe_disabled(p)) { | |
2077 | p->flags &= ~KPROBE_FLAG_DISABLED; | |
12310e34 JY |
2078 | ret = arm_kprobe(p); |
2079 | if (ret) | |
2080 | p->flags |= KPROBE_FLAG_DISABLED; | |
c0614829 MH |
2081 | } |
2082 | out: | |
2083 | mutex_unlock(&kprobe_mutex); | |
2084 | return ret; | |
2085 | } | |
2086 | EXPORT_SYMBOL_GPL(enable_kprobe); | |
2087 | ||
4458515b | 2088 | /* Caller must NOT call this in usual path. This is only for critical case */ |
820aede0 | 2089 | void dump_kprobe(struct kprobe *kp) |
24851d24 | 2090 | { |
4458515b MH |
2091 | pr_err("Dumping kprobe:\n"); |
2092 | pr_err("Name: %s\nOffset: %x\nAddress: %pS\n", | |
2093 | kp->symbol_name, kp->offset, kp->addr); | |
24851d24 | 2094 | } |
820aede0 | 2095 | NOKPROBE_SYMBOL(dump_kprobe); |
24851d24 | 2096 | |
fb1a59fa MH |
2097 | int kprobe_add_ksym_blacklist(unsigned long entry) |
2098 | { | |
2099 | struct kprobe_blacklist_entry *ent; | |
2100 | unsigned long offset = 0, size = 0; | |
2101 | ||
2102 | if (!kernel_text_address(entry) || | |
2103 | !kallsyms_lookup_size_offset(entry, &size, &offset)) | |
2104 | return -EINVAL; | |
2105 | ||
2106 | ent = kmalloc(sizeof(*ent), GFP_KERNEL); | |
2107 | if (!ent) | |
2108 | return -ENOMEM; | |
2109 | ent->start_addr = entry; | |
2110 | ent->end_addr = entry + size; | |
2111 | INIT_LIST_HEAD(&ent->list); | |
2112 | list_add_tail(&ent->list, &kprobe_blacklist); | |
2113 | ||
2114 | return (int)size; | |
2115 | } | |
2116 | ||
2117 | /* Add all symbols in given area into kprobe blacklist */ | |
2118 | int kprobe_add_area_blacklist(unsigned long start, unsigned long end) | |
2119 | { | |
2120 | unsigned long entry; | |
2121 | int ret = 0; | |
2122 | ||
2123 | for (entry = start; entry < end; entry += ret) { | |
2124 | ret = kprobe_add_ksym_blacklist(entry); | |
2125 | if (ret < 0) | |
2126 | return ret; | |
2127 | if (ret == 0) /* In case of alias symbol */ | |
2128 | ret = 1; | |
2129 | } | |
2130 | return 0; | |
2131 | } | |
2132 | ||
2133 | int __init __weak arch_populate_kprobe_blacklist(void) | |
2134 | { | |
2135 | return 0; | |
2136 | } | |
2137 | ||
376e2424 MH |
2138 | /* |
2139 | * Lookup and populate the kprobe_blacklist. | |
2140 | * | |
2141 | * Unlike the kretprobe blacklist, we'll need to determine | |
2142 | * the range of addresses that belong to the said functions, | |
2143 | * since a kprobe need not necessarily be at the beginning | |
2144 | * of a function. | |
2145 | */ | |
2146 | static int __init populate_kprobe_blacklist(unsigned long *start, | |
2147 | unsigned long *end) | |
2148 | { | |
fb1a59fa | 2149 | unsigned long entry; |
376e2424 | 2150 | unsigned long *iter; |
fb1a59fa | 2151 | int ret; |
376e2424 MH |
2152 | |
2153 | for (iter = start; iter < end; iter++) { | |
d81b4253 | 2154 | entry = arch_deref_entry_point((void *)*iter); |
fb1a59fa MH |
2155 | ret = kprobe_add_ksym_blacklist(entry); |
2156 | if (ret == -EINVAL) | |
376e2424 | 2157 | continue; |
fb1a59fa MH |
2158 | if (ret < 0) |
2159 | return ret; | |
376e2424 | 2160 | } |
fb1a59fa MH |
2161 | |
2162 | /* Symbols in __kprobes_text are blacklisted */ | |
2163 | ret = kprobe_add_area_blacklist((unsigned long)__kprobes_text_start, | |
2164 | (unsigned long)__kprobes_text_end); | |
2165 | ||
2166 | return ret ? : arch_populate_kprobe_blacklist(); | |
376e2424 MH |
2167 | } |
2168 | ||
e8386a0c | 2169 | /* Module notifier call back, checking kprobes on the module */ |
55479f64 MH |
2170 | static int kprobes_module_callback(struct notifier_block *nb, |
2171 | unsigned long val, void *data) | |
e8386a0c MH |
2172 | { |
2173 | struct module *mod = data; | |
2174 | struct hlist_head *head; | |
e8386a0c MH |
2175 | struct kprobe *p; |
2176 | unsigned int i; | |
f24659d9 | 2177 | int checkcore = (val == MODULE_STATE_GOING); |
e8386a0c | 2178 | |
f24659d9 | 2179 | if (val != MODULE_STATE_GOING && val != MODULE_STATE_LIVE) |
e8386a0c MH |
2180 | return NOTIFY_DONE; |
2181 | ||
2182 | /* | |
f24659d9 MH |
2183 | * When MODULE_STATE_GOING was notified, both of module .text and |
2184 | * .init.text sections would be freed. When MODULE_STATE_LIVE was | |
2185 | * notified, only .init.text section would be freed. We need to | |
2186 | * disable kprobes which have been inserted in the sections. | |
e8386a0c MH |
2187 | */ |
2188 | mutex_lock(&kprobe_mutex); | |
2189 | for (i = 0; i < KPROBE_TABLE_SIZE; i++) { | |
2190 | head = &kprobe_table[i]; | |
b67bfe0d | 2191 | hlist_for_each_entry_rcu(p, head, hlist) |
f24659d9 MH |
2192 | if (within_module_init((unsigned long)p->addr, mod) || |
2193 | (checkcore && | |
2194 | within_module_core((unsigned long)p->addr, mod))) { | |
e8386a0c MH |
2195 | /* |
2196 | * The vaddr this probe is installed will soon | |
2197 | * be vfreed buy not synced to disk. Hence, | |
2198 | * disarming the breakpoint isn't needed. | |
545a0281 SRV |
2199 | * |
2200 | * Note, this will also move any optimized probes | |
2201 | * that are pending to be removed from their | |
2202 | * corresponding lists to the freeing_list and | |
2203 | * will not be touched by the delayed | |
2204 | * kprobe_optimizer work handler. | |
e8386a0c MH |
2205 | */ |
2206 | kill_kprobe(p); | |
2207 | } | |
2208 | } | |
2209 | mutex_unlock(&kprobe_mutex); | |
2210 | return NOTIFY_DONE; | |
2211 | } | |
2212 | ||
2213 | static struct notifier_block kprobe_module_nb = { | |
2214 | .notifier_call = kprobes_module_callback, | |
2215 | .priority = 0 | |
2216 | }; | |
2217 | ||
376e2424 MH |
2218 | /* Markers of _kprobe_blacklist section */ |
2219 | extern unsigned long __start_kprobe_blacklist[]; | |
2220 | extern unsigned long __stop_kprobe_blacklist[]; | |
2221 | ||
1da177e4 LT |
2222 | static int __init init_kprobes(void) |
2223 | { | |
2224 | int i, err = 0; | |
2225 | ||
2226 | /* FIXME allocate the probe table, currently defined statically */ | |
2227 | /* initialize all list heads */ | |
b94cce92 | 2228 | for (i = 0; i < KPROBE_TABLE_SIZE; i++) { |
1da177e4 | 2229 | INIT_HLIST_HEAD(&kprobe_table[i]); |
b94cce92 | 2230 | INIT_HLIST_HEAD(&kretprobe_inst_table[i]); |
ec484608 | 2231 | raw_spin_lock_init(&(kretprobe_table_locks[i].lock)); |
b94cce92 | 2232 | } |
1da177e4 | 2233 | |
376e2424 MH |
2234 | err = populate_kprobe_blacklist(__start_kprobe_blacklist, |
2235 | __stop_kprobe_blacklist); | |
2236 | if (err) { | |
2237 | pr_err("kprobes: failed to populate blacklist: %d\n", err); | |
2238 | pr_err("Please take care of using kprobes.\n"); | |
3d8d996e SD |
2239 | } |
2240 | ||
f438d914 MH |
2241 | if (kretprobe_blacklist_size) { |
2242 | /* lookup the function address from its name */ | |
2243 | for (i = 0; kretprobe_blacklist[i].name != NULL; i++) { | |
49e0b465 | 2244 | kretprobe_blacklist[i].addr = |
290e3070 | 2245 | kprobe_lookup_name(kretprobe_blacklist[i].name, 0); |
f438d914 MH |
2246 | if (!kretprobe_blacklist[i].addr) |
2247 | printk("kretprobe: lookup failed: %s\n", | |
2248 | kretprobe_blacklist[i].name); | |
2249 | } | |
2250 | } | |
2251 | ||
b2be84df MH |
2252 | #if defined(CONFIG_OPTPROBES) |
2253 | #if defined(__ARCH_WANT_KPROBES_INSN_SLOT) | |
afd66255 MH |
2254 | /* Init kprobe_optinsn_slots */ |
2255 | kprobe_optinsn_slots.insn_size = MAX_OPTINSN_SIZE; | |
2256 | #endif | |
b2be84df MH |
2257 | /* By default, kprobes can be optimized */ |
2258 | kprobes_allow_optimization = true; | |
2259 | #endif | |
afd66255 | 2260 | |
e579abeb MH |
2261 | /* By default, kprobes are armed */ |
2262 | kprobes_all_disarmed = false; | |
bf8f6e5b | 2263 | |
6772926b | 2264 | err = arch_init_kprobes(); |
802eae7c RL |
2265 | if (!err) |
2266 | err = register_die_notifier(&kprobe_exceptions_nb); | |
e8386a0c MH |
2267 | if (!err) |
2268 | err = register_module_notifier(&kprobe_module_nb); | |
2269 | ||
ef53d9c5 | 2270 | kprobes_initialized = (err == 0); |
802eae7c | 2271 | |
8c1c9356 AM |
2272 | if (!err) |
2273 | init_test_probes(); | |
1da177e4 LT |
2274 | return err; |
2275 | } | |
2276 | ||
346fd59b | 2277 | #ifdef CONFIG_DEBUG_FS |
55479f64 | 2278 | static void report_probe(struct seq_file *pi, struct kprobe *p, |
afd66255 | 2279 | const char *sym, int offset, char *modname, struct kprobe *pp) |
346fd59b SD |
2280 | { |
2281 | char *kprobe_type; | |
81365a94 | 2282 | void *addr = p->addr; |
346fd59b SD |
2283 | |
2284 | if (p->pre_handler == pre_handler_kretprobe) | |
2285 | kprobe_type = "r"; | |
346fd59b SD |
2286 | else |
2287 | kprobe_type = "k"; | |
afd66255 | 2288 | |
81365a94 MH |
2289 | if (!kallsyms_show_value()) |
2290 | addr = NULL; | |
2291 | ||
346fd59b | 2292 | if (sym) |
81365a94 MH |
2293 | seq_printf(pi, "%px %s %s+0x%x %s ", |
2294 | addr, kprobe_type, sym, offset, | |
afd66255 | 2295 | (modname ? modname : " ")); |
81365a94 MH |
2296 | else /* try to use %pS */ |
2297 | seq_printf(pi, "%px %s %pS ", | |
2298 | addr, kprobe_type, p->addr); | |
afd66255 MH |
2299 | |
2300 | if (!pp) | |
2301 | pp = p; | |
ae6aa16f | 2302 | seq_printf(pi, "%s%s%s%s\n", |
afd66255 MH |
2303 | (kprobe_gone(p) ? "[GONE]" : ""), |
2304 | ((kprobe_disabled(p) && !kprobe_gone(p)) ? "[DISABLED]" : ""), | |
ae6aa16f MH |
2305 | (kprobe_optimized(pp) ? "[OPTIMIZED]" : ""), |
2306 | (kprobe_ftrace(pp) ? "[FTRACE]" : "")); | |
346fd59b SD |
2307 | } |
2308 | ||
55479f64 | 2309 | static void *kprobe_seq_start(struct seq_file *f, loff_t *pos) |
346fd59b SD |
2310 | { |
2311 | return (*pos < KPROBE_TABLE_SIZE) ? pos : NULL; | |
2312 | } | |
2313 | ||
55479f64 | 2314 | static void *kprobe_seq_next(struct seq_file *f, void *v, loff_t *pos) |
346fd59b SD |
2315 | { |
2316 | (*pos)++; | |
2317 | if (*pos >= KPROBE_TABLE_SIZE) | |
2318 | return NULL; | |
2319 | return pos; | |
2320 | } | |
2321 | ||
55479f64 | 2322 | static void kprobe_seq_stop(struct seq_file *f, void *v) |
346fd59b SD |
2323 | { |
2324 | /* Nothing to do */ | |
2325 | } | |
2326 | ||
55479f64 | 2327 | static int show_kprobe_addr(struct seq_file *pi, void *v) |
346fd59b SD |
2328 | { |
2329 | struct hlist_head *head; | |
346fd59b SD |
2330 | struct kprobe *p, *kp; |
2331 | const char *sym = NULL; | |
2332 | unsigned int i = *(loff_t *) v; | |
ffb45122 | 2333 | unsigned long offset = 0; |
ab767865 | 2334 | char *modname, namebuf[KSYM_NAME_LEN]; |
346fd59b SD |
2335 | |
2336 | head = &kprobe_table[i]; | |
2337 | preempt_disable(); | |
b67bfe0d | 2338 | hlist_for_each_entry_rcu(p, head, hlist) { |
ffb45122 | 2339 | sym = kallsyms_lookup((unsigned long)p->addr, NULL, |
346fd59b | 2340 | &offset, &modname, namebuf); |
afd66255 | 2341 | if (kprobe_aggrprobe(p)) { |
346fd59b | 2342 | list_for_each_entry_rcu(kp, &p->list, list) |
afd66255 | 2343 | report_probe(pi, kp, sym, offset, modname, p); |
346fd59b | 2344 | } else |
afd66255 | 2345 | report_probe(pi, p, sym, offset, modname, NULL); |
346fd59b SD |
2346 | } |
2347 | preempt_enable(); | |
2348 | return 0; | |
2349 | } | |
2350 | ||
88e9d34c | 2351 | static const struct seq_operations kprobes_seq_ops = { |
346fd59b SD |
2352 | .start = kprobe_seq_start, |
2353 | .next = kprobe_seq_next, | |
2354 | .stop = kprobe_seq_stop, | |
2355 | .show = show_kprobe_addr | |
2356 | }; | |
2357 | ||
55479f64 | 2358 | static int kprobes_open(struct inode *inode, struct file *filp) |
346fd59b SD |
2359 | { |
2360 | return seq_open(filp, &kprobes_seq_ops); | |
2361 | } | |
2362 | ||
828c0950 | 2363 | static const struct file_operations debugfs_kprobes_operations = { |
346fd59b SD |
2364 | .open = kprobes_open, |
2365 | .read = seq_read, | |
2366 | .llseek = seq_lseek, | |
2367 | .release = seq_release, | |
2368 | }; | |
2369 | ||
63724740 MH |
2370 | /* kprobes/blacklist -- shows which functions can not be probed */ |
2371 | static void *kprobe_blacklist_seq_start(struct seq_file *m, loff_t *pos) | |
2372 | { | |
2373 | return seq_list_start(&kprobe_blacklist, *pos); | |
2374 | } | |
2375 | ||
2376 | static void *kprobe_blacklist_seq_next(struct seq_file *m, void *v, loff_t *pos) | |
2377 | { | |
2378 | return seq_list_next(v, &kprobe_blacklist, pos); | |
2379 | } | |
2380 | ||
2381 | static int kprobe_blacklist_seq_show(struct seq_file *m, void *v) | |
2382 | { | |
2383 | struct kprobe_blacklist_entry *ent = | |
2384 | list_entry(v, struct kprobe_blacklist_entry, list); | |
2385 | ||
ffb9bd68 MH |
2386 | /* |
2387 | * If /proc/kallsyms is not showing kernel address, we won't | |
2388 | * show them here either. | |
2389 | */ | |
2390 | if (!kallsyms_show_value()) | |
2391 | seq_printf(m, "0x%px-0x%px\t%ps\n", NULL, NULL, | |
2392 | (void *)ent->start_addr); | |
2393 | else | |
2394 | seq_printf(m, "0x%px-0x%px\t%ps\n", (void *)ent->start_addr, | |
2395 | (void *)ent->end_addr, (void *)ent->start_addr); | |
63724740 MH |
2396 | return 0; |
2397 | } | |
2398 | ||
2399 | static const struct seq_operations kprobe_blacklist_seq_ops = { | |
2400 | .start = kprobe_blacklist_seq_start, | |
2401 | .next = kprobe_blacklist_seq_next, | |
2402 | .stop = kprobe_seq_stop, /* Reuse void function */ | |
2403 | .show = kprobe_blacklist_seq_show, | |
2404 | }; | |
2405 | ||
2406 | static int kprobe_blacklist_open(struct inode *inode, struct file *filp) | |
2407 | { | |
2408 | return seq_open(filp, &kprobe_blacklist_seq_ops); | |
2409 | } | |
2410 | ||
2411 | static const struct file_operations debugfs_kprobe_blacklist_ops = { | |
2412 | .open = kprobe_blacklist_open, | |
2413 | .read = seq_read, | |
2414 | .llseek = seq_lseek, | |
2415 | .release = seq_release, | |
2416 | }; | |
2417 | ||
12310e34 | 2418 | static int arm_all_kprobes(void) |
bf8f6e5b AM |
2419 | { |
2420 | struct hlist_head *head; | |
bf8f6e5b | 2421 | struct kprobe *p; |
12310e34 JY |
2422 | unsigned int i, total = 0, errors = 0; |
2423 | int err, ret = 0; | |
bf8f6e5b AM |
2424 | |
2425 | mutex_lock(&kprobe_mutex); | |
2426 | ||
e579abeb MH |
2427 | /* If kprobes are armed, just return */ |
2428 | if (!kprobes_all_disarmed) | |
bf8f6e5b AM |
2429 | goto already_enabled; |
2430 | ||
977ad481 WN |
2431 | /* |
2432 | * optimize_kprobe() called by arm_kprobe() checks | |
2433 | * kprobes_all_disarmed, so set kprobes_all_disarmed before | |
2434 | * arm_kprobe. | |
2435 | */ | |
2436 | kprobes_all_disarmed = false; | |
afd66255 | 2437 | /* Arming kprobes doesn't optimize kprobe itself */ |
bf8f6e5b AM |
2438 | for (i = 0; i < KPROBE_TABLE_SIZE; i++) { |
2439 | head = &kprobe_table[i]; | |
12310e34 JY |
2440 | /* Arm all kprobes on a best-effort basis */ |
2441 | hlist_for_each_entry_rcu(p, head, hlist) { | |
2442 | if (!kprobe_disabled(p)) { | |
2443 | err = arm_kprobe(p); | |
2444 | if (err) { | |
2445 | errors++; | |
2446 | ret = err; | |
2447 | } | |
2448 | total++; | |
2449 | } | |
2450 | } | |
bf8f6e5b AM |
2451 | } |
2452 | ||
12310e34 JY |
2453 | if (errors) |
2454 | pr_warn("Kprobes globally enabled, but failed to arm %d out of %d probes\n", | |
2455 | errors, total); | |
2456 | else | |
2457 | pr_info("Kprobes globally enabled\n"); | |
bf8f6e5b AM |
2458 | |
2459 | already_enabled: | |
2460 | mutex_unlock(&kprobe_mutex); | |
12310e34 | 2461 | return ret; |
bf8f6e5b AM |
2462 | } |
2463 | ||
297f9233 | 2464 | static int disarm_all_kprobes(void) |
bf8f6e5b AM |
2465 | { |
2466 | struct hlist_head *head; | |
bf8f6e5b | 2467 | struct kprobe *p; |
297f9233 JY |
2468 | unsigned int i, total = 0, errors = 0; |
2469 | int err, ret = 0; | |
bf8f6e5b AM |
2470 | |
2471 | mutex_lock(&kprobe_mutex); | |
2472 | ||
e579abeb | 2473 | /* If kprobes are already disarmed, just return */ |
6274de49 MH |
2474 | if (kprobes_all_disarmed) { |
2475 | mutex_unlock(&kprobe_mutex); | |
297f9233 | 2476 | return 0; |
6274de49 | 2477 | } |
bf8f6e5b | 2478 | |
e579abeb | 2479 | kprobes_all_disarmed = true; |
afd66255 | 2480 | |
bf8f6e5b AM |
2481 | for (i = 0; i < KPROBE_TABLE_SIZE; i++) { |
2482 | head = &kprobe_table[i]; | |
297f9233 | 2483 | /* Disarm all kprobes on a best-effort basis */ |
b67bfe0d | 2484 | hlist_for_each_entry_rcu(p, head, hlist) { |
297f9233 JY |
2485 | if (!arch_trampoline_kprobe(p) && !kprobe_disabled(p)) { |
2486 | err = disarm_kprobe(p, false); | |
2487 | if (err) { | |
2488 | errors++; | |
2489 | ret = err; | |
2490 | } | |
2491 | total++; | |
2492 | } | |
bf8f6e5b AM |
2493 | } |
2494 | } | |
297f9233 JY |
2495 | |
2496 | if (errors) | |
2497 | pr_warn("Kprobes globally disabled, but failed to disarm %d out of %d probes\n", | |
2498 | errors, total); | |
2499 | else | |
2500 | pr_info("Kprobes globally disabled\n"); | |
2501 | ||
bf8f6e5b | 2502 | mutex_unlock(&kprobe_mutex); |
bf8f6e5b | 2503 | |
6274de49 MH |
2504 | /* Wait for disarming all kprobes by optimizer */ |
2505 | wait_for_kprobe_optimizer(); | |
297f9233 JY |
2506 | |
2507 | return ret; | |
bf8f6e5b AM |
2508 | } |
2509 | ||
2510 | /* | |
2511 | * XXX: The debugfs bool file interface doesn't allow for callbacks | |
2512 | * when the bool state is switched. We can reuse that facility when | |
2513 | * available | |
2514 | */ | |
2515 | static ssize_t read_enabled_file_bool(struct file *file, | |
2516 | char __user *user_buf, size_t count, loff_t *ppos) | |
2517 | { | |
2518 | char buf[3]; | |
2519 | ||
e579abeb | 2520 | if (!kprobes_all_disarmed) |
bf8f6e5b AM |
2521 | buf[0] = '1'; |
2522 | else | |
2523 | buf[0] = '0'; | |
2524 | buf[1] = '\n'; | |
2525 | buf[2] = 0x00; | |
2526 | return simple_read_from_buffer(user_buf, count, ppos, buf, 2); | |
2527 | } | |
2528 | ||
2529 | static ssize_t write_enabled_file_bool(struct file *file, | |
2530 | const char __user *user_buf, size_t count, loff_t *ppos) | |
2531 | { | |
2532 | char buf[32]; | |
efeb156e | 2533 | size_t buf_size; |
12310e34 | 2534 | int ret = 0; |
bf8f6e5b AM |
2535 | |
2536 | buf_size = min(count, (sizeof(buf)-1)); | |
2537 | if (copy_from_user(buf, user_buf, buf_size)) | |
2538 | return -EFAULT; | |
2539 | ||
10fb46d5 | 2540 | buf[buf_size] = '\0'; |
bf8f6e5b AM |
2541 | switch (buf[0]) { |
2542 | case 'y': | |
2543 | case 'Y': | |
2544 | case '1': | |
12310e34 | 2545 | ret = arm_all_kprobes(); |
bf8f6e5b AM |
2546 | break; |
2547 | case 'n': | |
2548 | case 'N': | |
2549 | case '0': | |
297f9233 | 2550 | ret = disarm_all_kprobes(); |
bf8f6e5b | 2551 | break; |
10fb46d5 MK |
2552 | default: |
2553 | return -EINVAL; | |
bf8f6e5b AM |
2554 | } |
2555 | ||
12310e34 JY |
2556 | if (ret) |
2557 | return ret; | |
2558 | ||
bf8f6e5b AM |
2559 | return count; |
2560 | } | |
2561 | ||
828c0950 | 2562 | static const struct file_operations fops_kp = { |
bf8f6e5b AM |
2563 | .read = read_enabled_file_bool, |
2564 | .write = write_enabled_file_bool, | |
6038f373 | 2565 | .llseek = default_llseek, |
bf8f6e5b AM |
2566 | }; |
2567 | ||
55479f64 | 2568 | static int __init debugfs_kprobe_init(void) |
346fd59b SD |
2569 | { |
2570 | struct dentry *dir, *file; | |
bf8f6e5b | 2571 | unsigned int value = 1; |
346fd59b SD |
2572 | |
2573 | dir = debugfs_create_dir("kprobes", NULL); | |
2574 | if (!dir) | |
2575 | return -ENOMEM; | |
2576 | ||
f2a3ab36 | 2577 | file = debugfs_create_file("list", 0400, dir, NULL, |
346fd59b | 2578 | &debugfs_kprobes_operations); |
63724740 MH |
2579 | if (!file) |
2580 | goto error; | |
346fd59b | 2581 | |
bf8f6e5b AM |
2582 | file = debugfs_create_file("enabled", 0600, dir, |
2583 | &value, &fops_kp); | |
63724740 MH |
2584 | if (!file) |
2585 | goto error; | |
2586 | ||
f2a3ab36 | 2587 | file = debugfs_create_file("blacklist", 0400, dir, NULL, |
63724740 MH |
2588 | &debugfs_kprobe_blacklist_ops); |
2589 | if (!file) | |
2590 | goto error; | |
bf8f6e5b | 2591 | |
346fd59b | 2592 | return 0; |
63724740 MH |
2593 | |
2594 | error: | |
2595 | debugfs_remove(dir); | |
2596 | return -ENOMEM; | |
346fd59b SD |
2597 | } |
2598 | ||
2599 | late_initcall(debugfs_kprobe_init); | |
2600 | #endif /* CONFIG_DEBUG_FS */ | |
2601 | ||
2602 | module_init(init_kprobes); |