]>
Commit | Line | Data |
---|---|---|
1a59d1b8 | 1 | // SPDX-License-Identifier: GPL-2.0-or-later |
1da177e4 LT |
2 | /* |
3 | * Kernel Probes (KProbes) | |
4 | * kernel/kprobes.c | |
5 | * | |
1da177e4 LT |
6 | * Copyright (C) IBM Corporation, 2002, 2004 |
7 | * | |
8 | * 2002-Oct Created by Vamsi Krishna S <vamsi_krishna@in.ibm.com> Kernel | |
9 | * Probes initial implementation (includes suggestions from | |
10 | * Rusty Russell). | |
11 | * 2004-Aug Updated by Prasanna S Panchamukhi <prasanna@in.ibm.com> with | |
12 | * hlists and exceptions notifier as suggested by Andi Kleen. | |
13 | * 2004-July Suparna Bhattacharya <suparna@in.ibm.com> added jumper probes | |
14 | * interface to access function arguments. | |
15 | * 2004-Sep Prasanna S Panchamukhi <prasanna@in.ibm.com> Changed Kprobes | |
16 | * exceptions notifier to be first on the priority list. | |
b94cce92 HN |
17 | * 2005-May Hien Nguyen <hien@us.ibm.com>, Jim Keniston |
18 | * <jkenisto@us.ibm.com> and Prasanna S Panchamukhi | |
19 | * <prasanna@in.ibm.com> added function-return probes. | |
1da177e4 LT |
20 | */ |
21 | #include <linux/kprobes.h> | |
1da177e4 LT |
22 | #include <linux/hash.h> |
23 | #include <linux/init.h> | |
4e57b681 | 24 | #include <linux/slab.h> |
e3869792 | 25 | #include <linux/stddef.h> |
9984de1a | 26 | #include <linux/export.h> |
9ec4b1f3 | 27 | #include <linux/moduleloader.h> |
3a872d89 | 28 | #include <linux/kallsyms.h> |
b4c6c34a | 29 | #include <linux/freezer.h> |
346fd59b SD |
30 | #include <linux/seq_file.h> |
31 | #include <linux/debugfs.h> | |
b2be84df | 32 | #include <linux/sysctl.h> |
1eeb66a1 | 33 | #include <linux/kdebug.h> |
4460fdad | 34 | #include <linux/memory.h> |
4554dbcb | 35 | #include <linux/ftrace.h> |
afd66255 | 36 | #include <linux/cpu.h> |
bf5438fc | 37 | #include <linux/jump_label.h> |
69e49088 | 38 | #include <linux/perf_event.h> |
bf8f6e5b | 39 | |
bfd45be0 | 40 | #include <asm/sections.h> |
1da177e4 LT |
41 | #include <asm/cacheflush.h> |
42 | #include <asm/errno.h> | |
7c0f6ba6 | 43 | #include <linux/uaccess.h> |
1da177e4 LT |
44 | |
45 | #define KPROBE_HASH_BITS 6 | |
46 | #define KPROBE_TABLE_SIZE (1 << KPROBE_HASH_BITS) | |
47 | ||
3a872d89 | 48 | |
ef53d9c5 | 49 | static int kprobes_initialized; |
7e6a71d8 MH |
50 | /* kprobe_table can be accessed by |
51 | * - Normal hlist traversal and RCU add/del under kprobe_mutex is held. | |
52 | * Or | |
53 | * - RCU hlist traversal under disabling preempt (breakpoint handlers) | |
54 | */ | |
1da177e4 LT |
55 | static struct hlist_head kprobe_table[KPROBE_TABLE_SIZE]; |
56 | ||
bf8f6e5b | 57 | /* NOTE: change this value only with kprobe_mutex held */ |
e579abeb | 58 | static bool kprobes_all_disarmed; |
bf8f6e5b | 59 | |
43948f50 MH |
60 | /* This protects kprobe_table and optimizing_list */ |
61 | static DEFINE_MUTEX(kprobe_mutex); | |
e6584523 | 62 | static DEFINE_PER_CPU(struct kprobe *, kprobe_instance) = NULL; |
ef53d9c5 | 63 | |
290e3070 NR |
64 | kprobe_opcode_t * __weak kprobe_lookup_name(const char *name, |
65 | unsigned int __unused) | |
49e0b465 NR |
66 | { |
67 | return ((kprobe_opcode_t *)(kallsyms_lookup_name(name))); | |
68 | } | |
69 | ||
376e2424 MH |
70 | /* Blacklist -- list of struct kprobe_blacklist_entry */ |
71 | static LIST_HEAD(kprobe_blacklist); | |
3d8d996e | 72 | |
2d14e39d | 73 | #ifdef __ARCH_WANT_KPROBES_INSN_SLOT |
9ec4b1f3 AM |
74 | /* |
75 | * kprobe->ainsn.insn points to the copy of the instruction to be | |
76 | * single-stepped. x86_64, POWER4 and above have no-exec support and | |
77 | * stepping on the instruction on a vmalloced/kmalloced/data page | |
78 | * is a recipe for disaster | |
79 | */ | |
9ec4b1f3 | 80 | struct kprobe_insn_page { |
c5cb5a2d | 81 | struct list_head list; |
9ec4b1f3 | 82 | kprobe_opcode_t *insns; /* Page of instruction slots */ |
af96397d | 83 | struct kprobe_insn_cache *cache; |
9ec4b1f3 | 84 | int nused; |
b4c6c34a | 85 | int ngarbage; |
4610ee1d | 86 | char slot_used[]; |
9ec4b1f3 AM |
87 | }; |
88 | ||
4610ee1d MH |
89 | #define KPROBE_INSN_PAGE_SIZE(slots) \ |
90 | (offsetof(struct kprobe_insn_page, slot_used) + \ | |
91 | (sizeof(char) * (slots))) | |
92 | ||
4610ee1d MH |
93 | static int slots_per_page(struct kprobe_insn_cache *c) |
94 | { | |
95 | return PAGE_SIZE/(c->insn_size * sizeof(kprobe_opcode_t)); | |
96 | } | |
97 | ||
ab40c5c6 MH |
98 | enum kprobe_slot_state { |
99 | SLOT_CLEAN = 0, | |
100 | SLOT_DIRTY = 1, | |
101 | SLOT_USED = 2, | |
102 | }; | |
103 | ||
63fef14f | 104 | void __weak *alloc_insn_page(void) |
af96397d HC |
105 | { |
106 | return module_alloc(PAGE_SIZE); | |
107 | } | |
108 | ||
c93f5cf5 | 109 | void __weak free_insn_page(void *page) |
af96397d | 110 | { |
be1f221c | 111 | module_memfree(page); |
af96397d HC |
112 | } |
113 | ||
c802d64a HC |
114 | struct kprobe_insn_cache kprobe_insn_slots = { |
115 | .mutex = __MUTEX_INITIALIZER(kprobe_insn_slots.mutex), | |
af96397d HC |
116 | .alloc = alloc_insn_page, |
117 | .free = free_insn_page, | |
d002b8bc | 118 | .sym = KPROBE_INSN_PAGE_SYM, |
4610ee1d MH |
119 | .pages = LIST_HEAD_INIT(kprobe_insn_slots.pages), |
120 | .insn_size = MAX_INSN_SIZE, | |
121 | .nr_garbage = 0, | |
122 | }; | |
55479f64 | 123 | static int collect_garbage_slots(struct kprobe_insn_cache *c); |
b4c6c34a | 124 | |
9ec4b1f3 | 125 | /** |
12941560 | 126 | * __get_insn_slot() - Find a slot on an executable page for an instruction. |
9ec4b1f3 AM |
127 | * We allocate an executable page if there's no room on existing ones. |
128 | */ | |
55479f64 | 129 | kprobe_opcode_t *__get_insn_slot(struct kprobe_insn_cache *c) |
9ec4b1f3 AM |
130 | { |
131 | struct kprobe_insn_page *kip; | |
c802d64a | 132 | kprobe_opcode_t *slot = NULL; |
9ec4b1f3 | 133 | |
5b485629 | 134 | /* Since the slot array is not protected by rcu, we need a mutex */ |
c802d64a | 135 | mutex_lock(&c->mutex); |
6f716acd | 136 | retry: |
5b485629 MH |
137 | rcu_read_lock(); |
138 | list_for_each_entry_rcu(kip, &c->pages, list) { | |
4610ee1d | 139 | if (kip->nused < slots_per_page(c)) { |
9ec4b1f3 | 140 | int i; |
4610ee1d | 141 | for (i = 0; i < slots_per_page(c); i++) { |
ab40c5c6 MH |
142 | if (kip->slot_used[i] == SLOT_CLEAN) { |
143 | kip->slot_used[i] = SLOT_USED; | |
9ec4b1f3 | 144 | kip->nused++; |
c802d64a | 145 | slot = kip->insns + (i * c->insn_size); |
5b485629 | 146 | rcu_read_unlock(); |
c802d64a | 147 | goto out; |
9ec4b1f3 AM |
148 | } |
149 | } | |
4610ee1d MH |
150 | /* kip->nused is broken. Fix it. */ |
151 | kip->nused = slots_per_page(c); | |
152 | WARN_ON(1); | |
9ec4b1f3 AM |
153 | } |
154 | } | |
5b485629 | 155 | rcu_read_unlock(); |
9ec4b1f3 | 156 | |
b4c6c34a | 157 | /* If there are any garbage slots, collect it and try again. */ |
4610ee1d | 158 | if (c->nr_garbage && collect_garbage_slots(c) == 0) |
b4c6c34a | 159 | goto retry; |
4610ee1d MH |
160 | |
161 | /* All out of space. Need to allocate a new page. */ | |
162 | kip = kmalloc(KPROBE_INSN_PAGE_SIZE(slots_per_page(c)), GFP_KERNEL); | |
6f716acd | 163 | if (!kip) |
c802d64a | 164 | goto out; |
9ec4b1f3 AM |
165 | |
166 | /* | |
167 | * Use module_alloc so this page is within +/- 2GB of where the | |
168 | * kernel image and loaded module images reside. This is required | |
169 | * so x86_64 can correctly handle the %rip-relative fixups. | |
170 | */ | |
af96397d | 171 | kip->insns = c->alloc(); |
9ec4b1f3 AM |
172 | if (!kip->insns) { |
173 | kfree(kip); | |
c802d64a | 174 | goto out; |
9ec4b1f3 | 175 | } |
c5cb5a2d | 176 | INIT_LIST_HEAD(&kip->list); |
4610ee1d | 177 | memset(kip->slot_used, SLOT_CLEAN, slots_per_page(c)); |
ab40c5c6 | 178 | kip->slot_used[0] = SLOT_USED; |
9ec4b1f3 | 179 | kip->nused = 1; |
b4c6c34a | 180 | kip->ngarbage = 0; |
af96397d | 181 | kip->cache = c; |
5b485629 | 182 | list_add_rcu(&kip->list, &c->pages); |
c802d64a | 183 | slot = kip->insns; |
69e49088 AH |
184 | |
185 | /* Record the perf ksymbol register event after adding the page */ | |
186 | perf_event_ksymbol(PERF_RECORD_KSYMBOL_TYPE_OOL, (unsigned long)kip->insns, | |
187 | PAGE_SIZE, false, c->sym); | |
c802d64a HC |
188 | out: |
189 | mutex_unlock(&c->mutex); | |
190 | return slot; | |
12941560 MH |
191 | } |
192 | ||
b4c6c34a | 193 | /* Return 1 if all garbages are collected, otherwise 0. */ |
55479f64 | 194 | static int collect_one_slot(struct kprobe_insn_page *kip, int idx) |
b4c6c34a | 195 | { |
ab40c5c6 | 196 | kip->slot_used[idx] = SLOT_CLEAN; |
b4c6c34a MH |
197 | kip->nused--; |
198 | if (kip->nused == 0) { | |
199 | /* | |
200 | * Page is no longer in use. Free it unless | |
201 | * it's the last one. We keep the last one | |
202 | * so as not to have to set it up again the | |
203 | * next time somebody inserts a probe. | |
204 | */ | |
4610ee1d | 205 | if (!list_is_singular(&kip->list)) { |
69e49088 AH |
206 | /* |
207 | * Record perf ksymbol unregister event before removing | |
208 | * the page. | |
209 | */ | |
210 | perf_event_ksymbol(PERF_RECORD_KSYMBOL_TYPE_OOL, | |
211 | (unsigned long)kip->insns, PAGE_SIZE, true, | |
212 | kip->cache->sym); | |
5b485629 MH |
213 | list_del_rcu(&kip->list); |
214 | synchronize_rcu(); | |
af96397d | 215 | kip->cache->free(kip->insns); |
b4c6c34a MH |
216 | kfree(kip); |
217 | } | |
218 | return 1; | |
219 | } | |
220 | return 0; | |
221 | } | |
222 | ||
55479f64 | 223 | static int collect_garbage_slots(struct kprobe_insn_cache *c) |
b4c6c34a | 224 | { |
c5cb5a2d | 225 | struct kprobe_insn_page *kip, *next; |
b4c6c34a | 226 | |
615d0ebb | 227 | /* Ensure no-one is interrupted on the garbages */ |
ae8b7ce7 | 228 | synchronize_rcu(); |
b4c6c34a | 229 | |
4610ee1d | 230 | list_for_each_entry_safe(kip, next, &c->pages, list) { |
b4c6c34a | 231 | int i; |
b4c6c34a MH |
232 | if (kip->ngarbage == 0) |
233 | continue; | |
234 | kip->ngarbage = 0; /* we will collect all garbages */ | |
4610ee1d | 235 | for (i = 0; i < slots_per_page(c); i++) { |
5b485629 | 236 | if (kip->slot_used[i] == SLOT_DIRTY && collect_one_slot(kip, i)) |
b4c6c34a MH |
237 | break; |
238 | } | |
239 | } | |
4610ee1d | 240 | c->nr_garbage = 0; |
b4c6c34a MH |
241 | return 0; |
242 | } | |
243 | ||
55479f64 MH |
244 | void __free_insn_slot(struct kprobe_insn_cache *c, |
245 | kprobe_opcode_t *slot, int dirty) | |
9ec4b1f3 AM |
246 | { |
247 | struct kprobe_insn_page *kip; | |
5b485629 | 248 | long idx; |
9ec4b1f3 | 249 | |
c802d64a | 250 | mutex_lock(&c->mutex); |
5b485629 MH |
251 | rcu_read_lock(); |
252 | list_for_each_entry_rcu(kip, &c->pages, list) { | |
253 | idx = ((long)slot - (long)kip->insns) / | |
254 | (c->insn_size * sizeof(kprobe_opcode_t)); | |
255 | if (idx >= 0 && idx < slots_per_page(c)) | |
c802d64a | 256 | goto out; |
9ec4b1f3 | 257 | } |
5b485629 | 258 | /* Could not find this slot. */ |
4610ee1d | 259 | WARN_ON(1); |
5b485629 | 260 | kip = NULL; |
c802d64a | 261 | out: |
5b485629 MH |
262 | rcu_read_unlock(); |
263 | /* Mark and sweep: this may sleep */ | |
264 | if (kip) { | |
265 | /* Check double free */ | |
266 | WARN_ON(kip->slot_used[idx] != SLOT_USED); | |
267 | if (dirty) { | |
268 | kip->slot_used[idx] = SLOT_DIRTY; | |
269 | kip->ngarbage++; | |
270 | if (++c->nr_garbage > slots_per_page(c)) | |
271 | collect_garbage_slots(c); | |
272 | } else { | |
273 | collect_one_slot(kip, idx); | |
274 | } | |
275 | } | |
c802d64a | 276 | mutex_unlock(&c->mutex); |
4610ee1d | 277 | } |
6f716acd | 278 | |
5b485629 MH |
279 | /* |
280 | * Check given address is on the page of kprobe instruction slots. | |
281 | * This will be used for checking whether the address on a stack | |
282 | * is on a text area or not. | |
283 | */ | |
284 | bool __is_insn_slot_addr(struct kprobe_insn_cache *c, unsigned long addr) | |
285 | { | |
286 | struct kprobe_insn_page *kip; | |
287 | bool ret = false; | |
288 | ||
289 | rcu_read_lock(); | |
290 | list_for_each_entry_rcu(kip, &c->pages, list) { | |
291 | if (addr >= (unsigned long)kip->insns && | |
292 | addr < (unsigned long)kip->insns + PAGE_SIZE) { | |
293 | ret = true; | |
294 | break; | |
295 | } | |
296 | } | |
297 | rcu_read_unlock(); | |
298 | ||
299 | return ret; | |
300 | } | |
301 | ||
d002b8bc AH |
302 | int kprobe_cache_get_kallsym(struct kprobe_insn_cache *c, unsigned int *symnum, |
303 | unsigned long *value, char *type, char *sym) | |
304 | { | |
305 | struct kprobe_insn_page *kip; | |
306 | int ret = -ERANGE; | |
307 | ||
308 | rcu_read_lock(); | |
309 | list_for_each_entry_rcu(kip, &c->pages, list) { | |
310 | if ((*symnum)--) | |
311 | continue; | |
312 | strlcpy(sym, c->sym, KSYM_NAME_LEN); | |
313 | *type = 't'; | |
314 | *value = (unsigned long)kip->insns; | |
315 | ret = 0; | |
316 | break; | |
317 | } | |
318 | rcu_read_unlock(); | |
319 | ||
320 | return ret; | |
321 | } | |
322 | ||
afd66255 MH |
323 | #ifdef CONFIG_OPTPROBES |
324 | /* For optimized_kprobe buffer */ | |
c802d64a HC |
325 | struct kprobe_insn_cache kprobe_optinsn_slots = { |
326 | .mutex = __MUTEX_INITIALIZER(kprobe_optinsn_slots.mutex), | |
af96397d HC |
327 | .alloc = alloc_insn_page, |
328 | .free = free_insn_page, | |
d002b8bc | 329 | .sym = KPROBE_OPTINSN_PAGE_SYM, |
afd66255 MH |
330 | .pages = LIST_HEAD_INIT(kprobe_optinsn_slots.pages), |
331 | /* .insn_size is initialized later */ | |
332 | .nr_garbage = 0, | |
333 | }; | |
afd66255 | 334 | #endif |
2d14e39d | 335 | #endif |
9ec4b1f3 | 336 | |
e6584523 AM |
337 | /* We have preemption disabled.. so it is safe to use __ versions */ |
338 | static inline void set_kprobe_instance(struct kprobe *kp) | |
339 | { | |
b76834bc | 340 | __this_cpu_write(kprobe_instance, kp); |
e6584523 AM |
341 | } |
342 | ||
343 | static inline void reset_kprobe_instance(void) | |
344 | { | |
b76834bc | 345 | __this_cpu_write(kprobe_instance, NULL); |
e6584523 AM |
346 | } |
347 | ||
3516a460 AM |
348 | /* |
349 | * This routine is called either: | |
49a2a1b8 | 350 | * - under the kprobe_mutex - during kprobe_[un]register() |
3516a460 | 351 | * OR |
d217d545 | 352 | * - with preemption disabled - from arch/xxx/kernel/kprobes.c |
3516a460 | 353 | */ |
820aede0 | 354 | struct kprobe *get_kprobe(void *addr) |
1da177e4 LT |
355 | { |
356 | struct hlist_head *head; | |
3516a460 | 357 | struct kprobe *p; |
1da177e4 LT |
358 | |
359 | head = &kprobe_table[hash_ptr(addr, KPROBE_HASH_BITS)]; | |
6743ad43 MH |
360 | hlist_for_each_entry_rcu(p, head, hlist, |
361 | lockdep_is_held(&kprobe_mutex)) { | |
1da177e4 LT |
362 | if (p->addr == addr) |
363 | return p; | |
364 | } | |
afd66255 | 365 | |
1da177e4 LT |
366 | return NULL; |
367 | } | |
820aede0 | 368 | NOKPROBE_SYMBOL(get_kprobe); |
1da177e4 | 369 | |
820aede0 | 370 | static int aggr_pre_handler(struct kprobe *p, struct pt_regs *regs); |
afd66255 MH |
371 | |
372 | /* Return true if the kprobe is an aggregator */ | |
373 | static inline int kprobe_aggrprobe(struct kprobe *p) | |
374 | { | |
375 | return p->pre_handler == aggr_pre_handler; | |
376 | } | |
377 | ||
6274de49 MH |
378 | /* Return true(!0) if the kprobe is unused */ |
379 | static inline int kprobe_unused(struct kprobe *p) | |
380 | { | |
381 | return kprobe_aggrprobe(p) && kprobe_disabled(p) && | |
382 | list_empty(&p->list); | |
383 | } | |
384 | ||
afd66255 MH |
385 | /* |
386 | * Keep all fields in the kprobe consistent | |
387 | */ | |
6d8e40a8 | 388 | static inline void copy_kprobe(struct kprobe *ap, struct kprobe *p) |
afd66255 | 389 | { |
6d8e40a8 MH |
390 | memcpy(&p->opcode, &ap->opcode, sizeof(kprobe_opcode_t)); |
391 | memcpy(&p->ainsn, &ap->ainsn, sizeof(struct arch_specific_insn)); | |
afd66255 MH |
392 | } |
393 | ||
394 | #ifdef CONFIG_OPTPROBES | |
b2be84df MH |
395 | /* NOTE: change this value only with kprobe_mutex held */ |
396 | static bool kprobes_allow_optimization; | |
397 | ||
afd66255 MH |
398 | /* |
399 | * Call all pre_handler on the list, but ignores its return value. | |
400 | * This must be called from arch-dep optimized caller. | |
401 | */ | |
820aede0 | 402 | void opt_pre_handler(struct kprobe *p, struct pt_regs *regs) |
afd66255 MH |
403 | { |
404 | struct kprobe *kp; | |
405 | ||
406 | list_for_each_entry_rcu(kp, &p->list, list) { | |
407 | if (kp->pre_handler && likely(!kprobe_disabled(kp))) { | |
408 | set_kprobe_instance(kp); | |
4f3a8714 | 409 | kp->pre_handler(kp, regs); |
afd66255 MH |
410 | } |
411 | reset_kprobe_instance(); | |
412 | } | |
413 | } | |
820aede0 | 414 | NOKPROBE_SYMBOL(opt_pre_handler); |
afd66255 | 415 | |
6274de49 | 416 | /* Free optimized instructions and optimized_kprobe */ |
55479f64 | 417 | static void free_aggr_kprobe(struct kprobe *p) |
6274de49 MH |
418 | { |
419 | struct optimized_kprobe *op; | |
420 | ||
421 | op = container_of(p, struct optimized_kprobe, kp); | |
422 | arch_remove_optimized_kprobe(op); | |
423 | arch_remove_kprobe(p); | |
424 | kfree(op); | |
425 | } | |
426 | ||
afd66255 MH |
427 | /* Return true(!0) if the kprobe is ready for optimization. */ |
428 | static inline int kprobe_optready(struct kprobe *p) | |
429 | { | |
430 | struct optimized_kprobe *op; | |
431 | ||
432 | if (kprobe_aggrprobe(p)) { | |
433 | op = container_of(p, struct optimized_kprobe, kp); | |
434 | return arch_prepared_optinsn(&op->optinsn); | |
435 | } | |
436 | ||
437 | return 0; | |
438 | } | |
439 | ||
6274de49 MH |
440 | /* Return true(!0) if the kprobe is disarmed. Note: p must be on hash list */ |
441 | static inline int kprobe_disarmed(struct kprobe *p) | |
442 | { | |
443 | struct optimized_kprobe *op; | |
444 | ||
445 | /* If kprobe is not aggr/opt probe, just return kprobe is disabled */ | |
446 | if (!kprobe_aggrprobe(p)) | |
447 | return kprobe_disabled(p); | |
448 | ||
449 | op = container_of(p, struct optimized_kprobe, kp); | |
450 | ||
451 | return kprobe_disabled(p) && list_empty(&op->list); | |
452 | } | |
453 | ||
454 | /* Return true(!0) if the probe is queued on (un)optimizing lists */ | |
55479f64 | 455 | static int kprobe_queued(struct kprobe *p) |
6274de49 MH |
456 | { |
457 | struct optimized_kprobe *op; | |
458 | ||
459 | if (kprobe_aggrprobe(p)) { | |
460 | op = container_of(p, struct optimized_kprobe, kp); | |
461 | if (!list_empty(&op->list)) | |
462 | return 1; | |
463 | } | |
464 | return 0; | |
465 | } | |
466 | ||
afd66255 MH |
467 | /* |
468 | * Return an optimized kprobe whose optimizing code replaces | |
469 | * instructions including addr (exclude breakpoint). | |
470 | */ | |
55479f64 | 471 | static struct kprobe *get_optimized_kprobe(unsigned long addr) |
afd66255 MH |
472 | { |
473 | int i; | |
474 | struct kprobe *p = NULL; | |
475 | struct optimized_kprobe *op; | |
476 | ||
477 | /* Don't check i == 0, since that is a breakpoint case. */ | |
478 | for (i = 1; !p && i < MAX_OPTIMIZED_LENGTH; i++) | |
479 | p = get_kprobe((void *)(addr - i)); | |
480 | ||
481 | if (p && kprobe_optready(p)) { | |
482 | op = container_of(p, struct optimized_kprobe, kp); | |
483 | if (arch_within_optimized_kprobe(op, addr)) | |
484 | return p; | |
485 | } | |
486 | ||
487 | return NULL; | |
488 | } | |
489 | ||
490 | /* Optimization staging list, protected by kprobe_mutex */ | |
491 | static LIST_HEAD(optimizing_list); | |
6274de49 | 492 | static LIST_HEAD(unoptimizing_list); |
7b959fc5 | 493 | static LIST_HEAD(freeing_list); |
afd66255 MH |
494 | |
495 | static void kprobe_optimizer(struct work_struct *work); | |
496 | static DECLARE_DELAYED_WORK(optimizing_work, kprobe_optimizer); | |
497 | #define OPTIMIZE_DELAY 5 | |
498 | ||
61f4e13f MH |
499 | /* |
500 | * Optimize (replace a breakpoint with a jump) kprobes listed on | |
501 | * optimizing_list. | |
502 | */ | |
55479f64 | 503 | static void do_optimize_kprobes(void) |
afd66255 | 504 | { |
f1c6ece2 | 505 | lockdep_assert_held(&text_mutex); |
afd66255 MH |
506 | /* |
507 | * The optimization/unoptimization refers online_cpus via | |
508 | * stop_machine() and cpu-hotplug modifies online_cpus. | |
509 | * And same time, text_mutex will be held in cpu-hotplug and here. | |
510 | * This combination can cause a deadlock (cpu-hotplug try to lock | |
511 | * text_mutex but stop_machine can not be done because online_cpus | |
512 | * has been changed) | |
2d1e38f5 | 513 | * To avoid this deadlock, caller must have locked cpu hotplug |
afd66255 MH |
514 | * for preventing cpu-hotplug outside of text_mutex locking. |
515 | */ | |
2d1e38f5 TG |
516 | lockdep_assert_cpus_held(); |
517 | ||
518 | /* Optimization never be done when disarmed */ | |
519 | if (kprobes_all_disarmed || !kprobes_allow_optimization || | |
520 | list_empty(&optimizing_list)) | |
521 | return; | |
522 | ||
cd7ebe22 | 523 | arch_optimize_kprobes(&optimizing_list); |
61f4e13f MH |
524 | } |
525 | ||
6274de49 MH |
526 | /* |
527 | * Unoptimize (replace a jump with a breakpoint and remove the breakpoint | |
528 | * if need) kprobes listed on unoptimizing_list. | |
529 | */ | |
55479f64 | 530 | static void do_unoptimize_kprobes(void) |
6274de49 MH |
531 | { |
532 | struct optimized_kprobe *op, *tmp; | |
533 | ||
f1c6ece2 | 534 | lockdep_assert_held(&text_mutex); |
2d1e38f5 TG |
535 | /* See comment in do_optimize_kprobes() */ |
536 | lockdep_assert_cpus_held(); | |
537 | ||
6274de49 MH |
538 | /* Unoptimization must be done anytime */ |
539 | if (list_empty(&unoptimizing_list)) | |
540 | return; | |
541 | ||
7b959fc5 | 542 | arch_unoptimize_kprobes(&unoptimizing_list, &freeing_list); |
f984ba4e | 543 | /* Loop free_list for disarming */ |
7b959fc5 | 544 | list_for_each_entry_safe(op, tmp, &freeing_list, list) { |
f66c0447 MH |
545 | /* Switching from detour code to origin */ |
546 | op->kp.flags &= ~KPROBE_FLAG_OPTIMIZED; | |
6274de49 MH |
547 | /* Disarm probes if marked disabled */ |
548 | if (kprobe_disabled(&op->kp)) | |
549 | arch_disarm_kprobe(&op->kp); | |
550 | if (kprobe_unused(&op->kp)) { | |
551 | /* | |
552 | * Remove unused probes from hash list. After waiting | |
553 | * for synchronization, these probes are reclaimed. | |
554 | * (reclaiming is done by do_free_cleaned_kprobes.) | |
555 | */ | |
556 | hlist_del_rcu(&op->kp.hlist); | |
6274de49 MH |
557 | } else |
558 | list_del_init(&op->list); | |
559 | } | |
6274de49 MH |
560 | } |
561 | ||
562 | /* Reclaim all kprobes on the free_list */ | |
55479f64 | 563 | static void do_free_cleaned_kprobes(void) |
6274de49 MH |
564 | { |
565 | struct optimized_kprobe *op, *tmp; | |
566 | ||
7b959fc5 | 567 | list_for_each_entry_safe(op, tmp, &freeing_list, list) { |
6274de49 | 568 | list_del_init(&op->list); |
cbdd96f5 MH |
569 | if (WARN_ON_ONCE(!kprobe_unused(&op->kp))) { |
570 | /* | |
571 | * This must not happen, but if there is a kprobe | |
572 | * still in use, keep it on kprobes hash list. | |
573 | */ | |
574 | continue; | |
575 | } | |
6274de49 MH |
576 | free_aggr_kprobe(&op->kp); |
577 | } | |
578 | } | |
579 | ||
580 | /* Start optimizer after OPTIMIZE_DELAY passed */ | |
55479f64 | 581 | static void kick_kprobe_optimizer(void) |
6274de49 | 582 | { |
ad72b3be | 583 | schedule_delayed_work(&optimizing_work, OPTIMIZE_DELAY); |
6274de49 MH |
584 | } |
585 | ||
61f4e13f | 586 | /* Kprobe jump optimizer */ |
55479f64 | 587 | static void kprobe_optimizer(struct work_struct *work) |
61f4e13f | 588 | { |
72ef3794 | 589 | mutex_lock(&kprobe_mutex); |
2d1e38f5 | 590 | cpus_read_lock(); |
f1c6ece2 | 591 | mutex_lock(&text_mutex); |
61f4e13f MH |
592 | |
593 | /* | |
6274de49 MH |
594 | * Step 1: Unoptimize kprobes and collect cleaned (unused and disarmed) |
595 | * kprobes before waiting for quiesence period. | |
596 | */ | |
7b959fc5 | 597 | do_unoptimize_kprobes(); |
6274de49 MH |
598 | |
599 | /* | |
a30b85df MH |
600 | * Step 2: Wait for quiesence period to ensure all potentially |
601 | * preempted tasks to have normally scheduled. Because optprobe | |
602 | * may modify multiple instructions, there is a chance that Nth | |
603 | * instruction is preempted. In that case, such tasks can return | |
604 | * to 2nd-Nth byte of jump instruction. This wait is for avoiding it. | |
605 | * Note that on non-preemptive kernel, this is transparently converted | |
606 | * to synchronoze_sched() to wait for all interrupts to have completed. | |
61f4e13f | 607 | */ |
a30b85df | 608 | synchronize_rcu_tasks(); |
61f4e13f | 609 | |
6274de49 | 610 | /* Step 3: Optimize kprobes after quiesence period */ |
61f4e13f | 611 | do_optimize_kprobes(); |
6274de49 MH |
612 | |
613 | /* Step 4: Free cleaned kprobes after quiesence period */ | |
7b959fc5 | 614 | do_free_cleaned_kprobes(); |
6274de49 | 615 | |
f1c6ece2 | 616 | mutex_unlock(&text_mutex); |
2d1e38f5 | 617 | cpus_read_unlock(); |
6274de49 | 618 | |
cd7ebe22 | 619 | /* Step 5: Kick optimizer again if needed */ |
f984ba4e | 620 | if (!list_empty(&optimizing_list) || !list_empty(&unoptimizing_list)) |
cd7ebe22 | 621 | kick_kprobe_optimizer(); |
1a0aa991 MH |
622 | |
623 | mutex_unlock(&kprobe_mutex); | |
6274de49 MH |
624 | } |
625 | ||
626 | /* Wait for completing optimization and unoptimization */ | |
30e7d894 | 627 | void wait_for_kprobe_optimizer(void) |
6274de49 | 628 | { |
ad72b3be TH |
629 | mutex_lock(&kprobe_mutex); |
630 | ||
631 | while (!list_empty(&optimizing_list) || !list_empty(&unoptimizing_list)) { | |
632 | mutex_unlock(&kprobe_mutex); | |
633 | ||
634 | /* this will also make optimizing_work execute immmediately */ | |
635 | flush_delayed_work(&optimizing_work); | |
636 | /* @optimizing_work might not have been queued yet, relax */ | |
637 | cpu_relax(); | |
638 | ||
639 | mutex_lock(&kprobe_mutex); | |
640 | } | |
641 | ||
642 | mutex_unlock(&kprobe_mutex); | |
afd66255 MH |
643 | } |
644 | ||
e4add247 MH |
645 | static bool optprobe_queued_unopt(struct optimized_kprobe *op) |
646 | { | |
647 | struct optimized_kprobe *_op; | |
648 | ||
649 | list_for_each_entry(_op, &unoptimizing_list, list) { | |
650 | if (op == _op) | |
651 | return true; | |
652 | } | |
653 | ||
654 | return false; | |
655 | } | |
656 | ||
afd66255 | 657 | /* Optimize kprobe if p is ready to be optimized */ |
55479f64 | 658 | static void optimize_kprobe(struct kprobe *p) |
afd66255 MH |
659 | { |
660 | struct optimized_kprobe *op; | |
661 | ||
662 | /* Check if the kprobe is disabled or not ready for optimization. */ | |
b2be84df | 663 | if (!kprobe_optready(p) || !kprobes_allow_optimization || |
afd66255 MH |
664 | (kprobe_disabled(p) || kprobes_all_disarmed)) |
665 | return; | |
666 | ||
059053a2 MH |
667 | /* kprobes with post_handler can not be optimized */ |
668 | if (p->post_handler) | |
afd66255 MH |
669 | return; |
670 | ||
671 | op = container_of(p, struct optimized_kprobe, kp); | |
672 | ||
673 | /* Check there is no other kprobes at the optimized instructions */ | |
674 | if (arch_check_optimized_kprobe(op) < 0) | |
675 | return; | |
676 | ||
677 | /* Check if it is already optimized. */ | |
e4add247 MH |
678 | if (op->kp.flags & KPROBE_FLAG_OPTIMIZED) { |
679 | if (optprobe_queued_unopt(op)) { | |
680 | /* This is under unoptimizing. Just dequeue the probe */ | |
681 | list_del_init(&op->list); | |
682 | } | |
afd66255 | 683 | return; |
e4add247 | 684 | } |
afd66255 | 685 | op->kp.flags |= KPROBE_FLAG_OPTIMIZED; |
6274de49 | 686 | |
e4add247 MH |
687 | /* On unoptimizing/optimizing_list, op must have OPTIMIZED flag */ |
688 | if (WARN_ON_ONCE(!list_empty(&op->list))) | |
689 | return; | |
690 | ||
691 | list_add(&op->list, &optimizing_list); | |
692 | kick_kprobe_optimizer(); | |
6274de49 MH |
693 | } |
694 | ||
695 | /* Short cut to direct unoptimizing */ | |
55479f64 | 696 | static void force_unoptimize_kprobe(struct optimized_kprobe *op) |
6274de49 | 697 | { |
2d1e38f5 | 698 | lockdep_assert_cpus_held(); |
6274de49 | 699 | arch_unoptimize_kprobe(op); |
f66c0447 | 700 | op->kp.flags &= ~KPROBE_FLAG_OPTIMIZED; |
afd66255 MH |
701 | } |
702 | ||
703 | /* Unoptimize a kprobe if p is optimized */ | |
55479f64 | 704 | static void unoptimize_kprobe(struct kprobe *p, bool force) |
afd66255 MH |
705 | { |
706 | struct optimized_kprobe *op; | |
707 | ||
6274de49 MH |
708 | if (!kprobe_aggrprobe(p) || kprobe_disarmed(p)) |
709 | return; /* This is not an optprobe nor optimized */ | |
710 | ||
711 | op = container_of(p, struct optimized_kprobe, kp); | |
e4add247 | 712 | if (!kprobe_optimized(p)) |
6274de49 | 713 | return; |
6274de49 | 714 | |
6274de49 | 715 | if (!list_empty(&op->list)) { |
e4add247 MH |
716 | if (optprobe_queued_unopt(op)) { |
717 | /* Queued in unoptimizing queue */ | |
718 | if (force) { | |
719 | /* | |
720 | * Forcibly unoptimize the kprobe here, and queue it | |
721 | * in the freeing list for release afterwards. | |
722 | */ | |
723 | force_unoptimize_kprobe(op); | |
724 | list_move(&op->list, &freeing_list); | |
725 | } | |
726 | } else { | |
727 | /* Dequeue from the optimizing queue */ | |
728 | list_del_init(&op->list); | |
729 | op->kp.flags &= ~KPROBE_FLAG_OPTIMIZED; | |
730 | } | |
6274de49 MH |
731 | return; |
732 | } | |
e4add247 | 733 | |
6274de49 | 734 | /* Optimized kprobe case */ |
e4add247 | 735 | if (force) { |
6274de49 MH |
736 | /* Forcibly update the code: this is a special case */ |
737 | force_unoptimize_kprobe(op); | |
e4add247 | 738 | } else { |
6274de49 MH |
739 | list_add(&op->list, &unoptimizing_list); |
740 | kick_kprobe_optimizer(); | |
afd66255 MH |
741 | } |
742 | } | |
743 | ||
0490cd1f | 744 | /* Cancel unoptimizing for reusing */ |
819319fc | 745 | static int reuse_unused_kprobe(struct kprobe *ap) |
0490cd1f MH |
746 | { |
747 | struct optimized_kprobe *op; | |
748 | ||
0490cd1f MH |
749 | /* |
750 | * Unused kprobe MUST be on the way of delayed unoptimizing (means | |
751 | * there is still a relative jump) and disabled. | |
752 | */ | |
753 | op = container_of(ap, struct optimized_kprobe, kp); | |
4458515b | 754 | WARN_ON_ONCE(list_empty(&op->list)); |
0490cd1f MH |
755 | /* Enable the probe again */ |
756 | ap->flags &= ~KPROBE_FLAG_DISABLED; | |
757 | /* Optimize it again (remove from op->list) */ | |
5f843ed4 MH |
758 | if (!kprobe_optready(ap)) |
759 | return -EINVAL; | |
819319fc | 760 | |
0490cd1f | 761 | optimize_kprobe(ap); |
819319fc | 762 | return 0; |
0490cd1f MH |
763 | } |
764 | ||
afd66255 | 765 | /* Remove optimized instructions */ |
55479f64 | 766 | static void kill_optimized_kprobe(struct kprobe *p) |
afd66255 MH |
767 | { |
768 | struct optimized_kprobe *op; | |
769 | ||
770 | op = container_of(p, struct optimized_kprobe, kp); | |
6274de49 MH |
771 | if (!list_empty(&op->list)) |
772 | /* Dequeue from the (un)optimization queue */ | |
afd66255 | 773 | list_del_init(&op->list); |
6274de49 | 774 | op->kp.flags &= ~KPROBE_FLAG_OPTIMIZED; |
7b959fc5 MH |
775 | |
776 | if (kprobe_unused(p)) { | |
777 | /* Enqueue if it is unused */ | |
778 | list_add(&op->list, &freeing_list); | |
779 | /* | |
780 | * Remove unused probes from the hash list. After waiting | |
781 | * for synchronization, this probe is reclaimed. | |
782 | * (reclaiming is done by do_free_cleaned_kprobes().) | |
783 | */ | |
784 | hlist_del_rcu(&op->kp.hlist); | |
785 | } | |
786 | ||
6274de49 | 787 | /* Don't touch the code, because it is already freed. */ |
afd66255 MH |
788 | arch_remove_optimized_kprobe(op); |
789 | } | |
790 | ||
a460246c MH |
791 | static inline |
792 | void __prepare_optimized_kprobe(struct optimized_kprobe *op, struct kprobe *p) | |
793 | { | |
794 | if (!kprobe_ftrace(p)) | |
795 | arch_prepare_optimized_kprobe(op, p); | |
796 | } | |
797 | ||
afd66255 | 798 | /* Try to prepare optimized instructions */ |
55479f64 | 799 | static void prepare_optimized_kprobe(struct kprobe *p) |
afd66255 MH |
800 | { |
801 | struct optimized_kprobe *op; | |
802 | ||
803 | op = container_of(p, struct optimized_kprobe, kp); | |
a460246c | 804 | __prepare_optimized_kprobe(op, p); |
afd66255 MH |
805 | } |
806 | ||
afd66255 | 807 | /* Allocate new optimized_kprobe and try to prepare optimized instructions */ |
55479f64 | 808 | static struct kprobe *alloc_aggr_kprobe(struct kprobe *p) |
afd66255 MH |
809 | { |
810 | struct optimized_kprobe *op; | |
811 | ||
812 | op = kzalloc(sizeof(struct optimized_kprobe), GFP_KERNEL); | |
813 | if (!op) | |
814 | return NULL; | |
815 | ||
816 | INIT_LIST_HEAD(&op->list); | |
817 | op->kp.addr = p->addr; | |
a460246c | 818 | __prepare_optimized_kprobe(op, p); |
afd66255 MH |
819 | |
820 | return &op->kp; | |
821 | } | |
822 | ||
55479f64 | 823 | static void init_aggr_kprobe(struct kprobe *ap, struct kprobe *p); |
afd66255 MH |
824 | |
825 | /* | |
826 | * Prepare an optimized_kprobe and optimize it | |
827 | * NOTE: p must be a normal registered kprobe | |
828 | */ | |
55479f64 | 829 | static void try_to_optimize_kprobe(struct kprobe *p) |
afd66255 MH |
830 | { |
831 | struct kprobe *ap; | |
832 | struct optimized_kprobe *op; | |
833 | ||
ae6aa16f MH |
834 | /* Impossible to optimize ftrace-based kprobe */ |
835 | if (kprobe_ftrace(p)) | |
836 | return; | |
837 | ||
25764288 | 838 | /* For preparing optimization, jump_label_text_reserved() is called */ |
2d1e38f5 | 839 | cpus_read_lock(); |
25764288 MH |
840 | jump_label_lock(); |
841 | mutex_lock(&text_mutex); | |
842 | ||
afd66255 MH |
843 | ap = alloc_aggr_kprobe(p); |
844 | if (!ap) | |
25764288 | 845 | goto out; |
afd66255 MH |
846 | |
847 | op = container_of(ap, struct optimized_kprobe, kp); | |
848 | if (!arch_prepared_optinsn(&op->optinsn)) { | |
849 | /* If failed to setup optimizing, fallback to kprobe */ | |
6274de49 MH |
850 | arch_remove_optimized_kprobe(op); |
851 | kfree(op); | |
25764288 | 852 | goto out; |
afd66255 MH |
853 | } |
854 | ||
855 | init_aggr_kprobe(ap, p); | |
25764288 MH |
856 | optimize_kprobe(ap); /* This just kicks optimizer thread */ |
857 | ||
858 | out: | |
859 | mutex_unlock(&text_mutex); | |
860 | jump_label_unlock(); | |
2d1e38f5 | 861 | cpus_read_unlock(); |
afd66255 MH |
862 | } |
863 | ||
b2be84df | 864 | #ifdef CONFIG_SYSCTL |
55479f64 | 865 | static void optimize_all_kprobes(void) |
b2be84df MH |
866 | { |
867 | struct hlist_head *head; | |
b2be84df MH |
868 | struct kprobe *p; |
869 | unsigned int i; | |
870 | ||
5c51543b | 871 | mutex_lock(&kprobe_mutex); |
b2be84df MH |
872 | /* If optimization is already allowed, just return */ |
873 | if (kprobes_allow_optimization) | |
5c51543b | 874 | goto out; |
b2be84df | 875 | |
2d1e38f5 | 876 | cpus_read_lock(); |
b2be84df | 877 | kprobes_allow_optimization = true; |
b2be84df MH |
878 | for (i = 0; i < KPROBE_TABLE_SIZE; i++) { |
879 | head = &kprobe_table[i]; | |
7e6a71d8 | 880 | hlist_for_each_entry(p, head, hlist) |
b2be84df MH |
881 | if (!kprobe_disabled(p)) |
882 | optimize_kprobe(p); | |
883 | } | |
2d1e38f5 | 884 | cpus_read_unlock(); |
b2be84df | 885 | printk(KERN_INFO "Kprobes globally optimized\n"); |
5c51543b MH |
886 | out: |
887 | mutex_unlock(&kprobe_mutex); | |
b2be84df MH |
888 | } |
889 | ||
55479f64 | 890 | static void unoptimize_all_kprobes(void) |
b2be84df MH |
891 | { |
892 | struct hlist_head *head; | |
b2be84df MH |
893 | struct kprobe *p; |
894 | unsigned int i; | |
895 | ||
5c51543b | 896 | mutex_lock(&kprobe_mutex); |
b2be84df | 897 | /* If optimization is already prohibited, just return */ |
5c51543b MH |
898 | if (!kprobes_allow_optimization) { |
899 | mutex_unlock(&kprobe_mutex); | |
b2be84df | 900 | return; |
5c51543b | 901 | } |
b2be84df | 902 | |
2d1e38f5 | 903 | cpus_read_lock(); |
b2be84df | 904 | kprobes_allow_optimization = false; |
b2be84df MH |
905 | for (i = 0; i < KPROBE_TABLE_SIZE; i++) { |
906 | head = &kprobe_table[i]; | |
7e6a71d8 | 907 | hlist_for_each_entry(p, head, hlist) { |
b2be84df | 908 | if (!kprobe_disabled(p)) |
6274de49 | 909 | unoptimize_kprobe(p, false); |
b2be84df MH |
910 | } |
911 | } | |
2d1e38f5 | 912 | cpus_read_unlock(); |
5c51543b MH |
913 | mutex_unlock(&kprobe_mutex); |
914 | ||
6274de49 MH |
915 | /* Wait for unoptimizing completion */ |
916 | wait_for_kprobe_optimizer(); | |
917 | printk(KERN_INFO "Kprobes globally unoptimized\n"); | |
b2be84df MH |
918 | } |
919 | ||
5c51543b | 920 | static DEFINE_MUTEX(kprobe_sysctl_mutex); |
b2be84df MH |
921 | int sysctl_kprobes_optimization; |
922 | int proc_kprobes_optimization_handler(struct ctl_table *table, int write, | |
32927393 | 923 | void *buffer, size_t *length, |
b2be84df MH |
924 | loff_t *ppos) |
925 | { | |
926 | int ret; | |
927 | ||
5c51543b | 928 | mutex_lock(&kprobe_sysctl_mutex); |
b2be84df MH |
929 | sysctl_kprobes_optimization = kprobes_allow_optimization ? 1 : 0; |
930 | ret = proc_dointvec_minmax(table, write, buffer, length, ppos); | |
931 | ||
932 | if (sysctl_kprobes_optimization) | |
933 | optimize_all_kprobes(); | |
934 | else | |
935 | unoptimize_all_kprobes(); | |
5c51543b | 936 | mutex_unlock(&kprobe_sysctl_mutex); |
b2be84df MH |
937 | |
938 | return ret; | |
939 | } | |
940 | #endif /* CONFIG_SYSCTL */ | |
941 | ||
6274de49 | 942 | /* Put a breakpoint for a probe. Must be called with text_mutex locked */ |
55479f64 | 943 | static void __arm_kprobe(struct kprobe *p) |
afd66255 | 944 | { |
6d8e40a8 | 945 | struct kprobe *_p; |
afd66255 MH |
946 | |
947 | /* Check collision with other optimized kprobes */ | |
6d8e40a8 MH |
948 | _p = get_optimized_kprobe((unsigned long)p->addr); |
949 | if (unlikely(_p)) | |
6274de49 MH |
950 | /* Fallback to unoptimized kprobe */ |
951 | unoptimize_kprobe(_p, true); | |
afd66255 MH |
952 | |
953 | arch_arm_kprobe(p); | |
954 | optimize_kprobe(p); /* Try to optimize (add kprobe to a list) */ | |
955 | } | |
956 | ||
6274de49 | 957 | /* Remove the breakpoint of a probe. Must be called with text_mutex locked */ |
55479f64 | 958 | static void __disarm_kprobe(struct kprobe *p, bool reopt) |
afd66255 | 959 | { |
6d8e40a8 | 960 | struct kprobe *_p; |
afd66255 | 961 | |
69d54b91 WN |
962 | /* Try to unoptimize */ |
963 | unoptimize_kprobe(p, kprobes_all_disarmed); | |
afd66255 | 964 | |
6274de49 MH |
965 | if (!kprobe_queued(p)) { |
966 | arch_disarm_kprobe(p); | |
967 | /* If another kprobe was blocked, optimize it. */ | |
968 | _p = get_optimized_kprobe((unsigned long)p->addr); | |
969 | if (unlikely(_p) && reopt) | |
970 | optimize_kprobe(_p); | |
971 | } | |
972 | /* TODO: reoptimize others after unoptimized this probe */ | |
afd66255 MH |
973 | } |
974 | ||
975 | #else /* !CONFIG_OPTPROBES */ | |
976 | ||
977 | #define optimize_kprobe(p) do {} while (0) | |
6274de49 | 978 | #define unoptimize_kprobe(p, f) do {} while (0) |
afd66255 MH |
979 | #define kill_optimized_kprobe(p) do {} while (0) |
980 | #define prepare_optimized_kprobe(p) do {} while (0) | |
981 | #define try_to_optimize_kprobe(p) do {} while (0) | |
982 | #define __arm_kprobe(p) arch_arm_kprobe(p) | |
6274de49 MH |
983 | #define __disarm_kprobe(p, o) arch_disarm_kprobe(p) |
984 | #define kprobe_disarmed(p) kprobe_disabled(p) | |
985 | #define wait_for_kprobe_optimizer() do {} while (0) | |
afd66255 | 986 | |
819319fc | 987 | static int reuse_unused_kprobe(struct kprobe *ap) |
0490cd1f | 988 | { |
819319fc MH |
989 | /* |
990 | * If the optimized kprobe is NOT supported, the aggr kprobe is | |
991 | * released at the same time that the last aggregated kprobe is | |
992 | * unregistered. | |
993 | * Thus there should be no chance to reuse unused kprobe. | |
994 | */ | |
0490cd1f | 995 | printk(KERN_ERR "Error: There should be no unused kprobe here.\n"); |
819319fc | 996 | return -EINVAL; |
0490cd1f MH |
997 | } |
998 | ||
55479f64 | 999 | static void free_aggr_kprobe(struct kprobe *p) |
afd66255 | 1000 | { |
6274de49 | 1001 | arch_remove_kprobe(p); |
afd66255 MH |
1002 | kfree(p); |
1003 | } | |
1004 | ||
55479f64 | 1005 | static struct kprobe *alloc_aggr_kprobe(struct kprobe *p) |
afd66255 MH |
1006 | { |
1007 | return kzalloc(sizeof(struct kprobe), GFP_KERNEL); | |
1008 | } | |
1009 | #endif /* CONFIG_OPTPROBES */ | |
1010 | ||
e7dbfe34 | 1011 | #ifdef CONFIG_KPROBES_ON_FTRACE |
ae6aa16f | 1012 | static struct ftrace_ops kprobe_ftrace_ops __read_mostly = { |
0bc11ed5 MH |
1013 | .func = kprobe_ftrace_handler, |
1014 | .flags = FTRACE_OPS_FL_SAVE_REGS, | |
1015 | }; | |
1016 | ||
1017 | static struct ftrace_ops kprobe_ipmodify_ops __read_mostly = { | |
e5253896 | 1018 | .func = kprobe_ftrace_handler, |
1d70be34 | 1019 | .flags = FTRACE_OPS_FL_SAVE_REGS | FTRACE_OPS_FL_IPMODIFY, |
ae6aa16f | 1020 | }; |
0bc11ed5 MH |
1021 | |
1022 | static int kprobe_ipmodify_enabled; | |
ae6aa16f MH |
1023 | static int kprobe_ftrace_enabled; |
1024 | ||
1025 | /* Must ensure p->addr is really on ftrace */ | |
55479f64 | 1026 | static int prepare_kprobe(struct kprobe *p) |
ae6aa16f MH |
1027 | { |
1028 | if (!kprobe_ftrace(p)) | |
1029 | return arch_prepare_kprobe(p); | |
1030 | ||
1031 | return arch_prepare_kprobe_ftrace(p); | |
1032 | } | |
1033 | ||
1034 | /* Caller must lock kprobe_mutex */ | |
0bc11ed5 MH |
1035 | static int __arm_kprobe_ftrace(struct kprobe *p, struct ftrace_ops *ops, |
1036 | int *cnt) | |
ae6aa16f | 1037 | { |
12310e34 | 1038 | int ret = 0; |
ae6aa16f | 1039 | |
0bc11ed5 | 1040 | ret = ftrace_set_filter_ip(ops, (unsigned long)p->addr, 0, 0); |
12310e34 | 1041 | if (ret) { |
4458515b MH |
1042 | pr_debug("Failed to arm kprobe-ftrace at %pS (%d)\n", |
1043 | p->addr, ret); | |
12310e34 JY |
1044 | return ret; |
1045 | } | |
1046 | ||
0bc11ed5 MH |
1047 | if (*cnt == 0) { |
1048 | ret = register_ftrace_function(ops); | |
12310e34 JY |
1049 | if (ret) { |
1050 | pr_debug("Failed to init kprobe-ftrace (%d)\n", ret); | |
1051 | goto err_ftrace; | |
1052 | } | |
ae6aa16f | 1053 | } |
12310e34 | 1054 | |
0bc11ed5 | 1055 | (*cnt)++; |
12310e34 JY |
1056 | return ret; |
1057 | ||
1058 | err_ftrace: | |
1059 | /* | |
0bc11ed5 MH |
1060 | * At this point, sinec ops is not registered, we should be sefe from |
1061 | * registering empty filter. | |
12310e34 | 1062 | */ |
0bc11ed5 | 1063 | ftrace_set_filter_ip(ops, (unsigned long)p->addr, 1, 0); |
12310e34 | 1064 | return ret; |
ae6aa16f MH |
1065 | } |
1066 | ||
0bc11ed5 MH |
1067 | static int arm_kprobe_ftrace(struct kprobe *p) |
1068 | { | |
1069 | bool ipmodify = (p->post_handler != NULL); | |
1070 | ||
1071 | return __arm_kprobe_ftrace(p, | |
1072 | ipmodify ? &kprobe_ipmodify_ops : &kprobe_ftrace_ops, | |
1073 | ipmodify ? &kprobe_ipmodify_enabled : &kprobe_ftrace_enabled); | |
1074 | } | |
1075 | ||
ae6aa16f | 1076 | /* Caller must lock kprobe_mutex */ |
0bc11ed5 MH |
1077 | static int __disarm_kprobe_ftrace(struct kprobe *p, struct ftrace_ops *ops, |
1078 | int *cnt) | |
ae6aa16f | 1079 | { |
297f9233 | 1080 | int ret = 0; |
ae6aa16f | 1081 | |
0bc11ed5 MH |
1082 | if (*cnt == 1) { |
1083 | ret = unregister_ftrace_function(ops); | |
297f9233 JY |
1084 | if (WARN(ret < 0, "Failed to unregister kprobe-ftrace (%d)\n", ret)) |
1085 | return ret; | |
ae6aa16f | 1086 | } |
297f9233 | 1087 | |
0bc11ed5 | 1088 | (*cnt)--; |
297f9233 | 1089 | |
0bc11ed5 | 1090 | ret = ftrace_set_filter_ip(ops, (unsigned long)p->addr, 1, 0); |
4458515b MH |
1091 | WARN_ONCE(ret < 0, "Failed to disarm kprobe-ftrace at %pS (%d)\n", |
1092 | p->addr, ret); | |
297f9233 | 1093 | return ret; |
ae6aa16f | 1094 | } |
0bc11ed5 MH |
1095 | |
1096 | static int disarm_kprobe_ftrace(struct kprobe *p) | |
1097 | { | |
1098 | bool ipmodify = (p->post_handler != NULL); | |
1099 | ||
1100 | return __disarm_kprobe_ftrace(p, | |
1101 | ipmodify ? &kprobe_ipmodify_ops : &kprobe_ftrace_ops, | |
1102 | ipmodify ? &kprobe_ipmodify_enabled : &kprobe_ftrace_enabled); | |
1103 | } | |
e7dbfe34 | 1104 | #else /* !CONFIG_KPROBES_ON_FTRACE */ |
10de795a MS |
1105 | static inline int prepare_kprobe(struct kprobe *p) |
1106 | { | |
1107 | return arch_prepare_kprobe(p); | |
1108 | } | |
1109 | ||
1110 | static inline int arm_kprobe_ftrace(struct kprobe *p) | |
1111 | { | |
1112 | return -ENODEV; | |
1113 | } | |
1114 | ||
1115 | static inline int disarm_kprobe_ftrace(struct kprobe *p) | |
1116 | { | |
1117 | return -ENODEV; | |
1118 | } | |
ae6aa16f MH |
1119 | #endif |
1120 | ||
201517a7 | 1121 | /* Arm a kprobe with text_mutex */ |
12310e34 | 1122 | static int arm_kprobe(struct kprobe *kp) |
201517a7 | 1123 | { |
12310e34 JY |
1124 | if (unlikely(kprobe_ftrace(kp))) |
1125 | return arm_kprobe_ftrace(kp); | |
1126 | ||
2d1e38f5 | 1127 | cpus_read_lock(); |
201517a7 | 1128 | mutex_lock(&text_mutex); |
afd66255 | 1129 | __arm_kprobe(kp); |
201517a7 | 1130 | mutex_unlock(&text_mutex); |
2d1e38f5 | 1131 | cpus_read_unlock(); |
12310e34 JY |
1132 | |
1133 | return 0; | |
201517a7 MH |
1134 | } |
1135 | ||
1136 | /* Disarm a kprobe with text_mutex */ | |
297f9233 | 1137 | static int disarm_kprobe(struct kprobe *kp, bool reopt) |
201517a7 | 1138 | { |
297f9233 JY |
1139 | if (unlikely(kprobe_ftrace(kp))) |
1140 | return disarm_kprobe_ftrace(kp); | |
2d1e38f5 TG |
1141 | |
1142 | cpus_read_lock(); | |
201517a7 | 1143 | mutex_lock(&text_mutex); |
ae6aa16f | 1144 | __disarm_kprobe(kp, reopt); |
201517a7 | 1145 | mutex_unlock(&text_mutex); |
2d1e38f5 | 1146 | cpus_read_unlock(); |
297f9233 JY |
1147 | |
1148 | return 0; | |
201517a7 MH |
1149 | } |
1150 | ||
64f562c6 AM |
1151 | /* |
1152 | * Aggregate handlers for multiple kprobes support - these handlers | |
1153 | * take care of invoking the individual kprobe handlers on p->list | |
1154 | */ | |
820aede0 | 1155 | static int aggr_pre_handler(struct kprobe *p, struct pt_regs *regs) |
64f562c6 AM |
1156 | { |
1157 | struct kprobe *kp; | |
1158 | ||
3516a460 | 1159 | list_for_each_entry_rcu(kp, &p->list, list) { |
de5bd88d | 1160 | if (kp->pre_handler && likely(!kprobe_disabled(kp))) { |
e6584523 | 1161 | set_kprobe_instance(kp); |
8b0914ea PP |
1162 | if (kp->pre_handler(kp, regs)) |
1163 | return 1; | |
64f562c6 | 1164 | } |
e6584523 | 1165 | reset_kprobe_instance(); |
64f562c6 AM |
1166 | } |
1167 | return 0; | |
1168 | } | |
820aede0 | 1169 | NOKPROBE_SYMBOL(aggr_pre_handler); |
64f562c6 | 1170 | |
820aede0 MH |
1171 | static void aggr_post_handler(struct kprobe *p, struct pt_regs *regs, |
1172 | unsigned long flags) | |
64f562c6 AM |
1173 | { |
1174 | struct kprobe *kp; | |
1175 | ||
3516a460 | 1176 | list_for_each_entry_rcu(kp, &p->list, list) { |
de5bd88d | 1177 | if (kp->post_handler && likely(!kprobe_disabled(kp))) { |
e6584523 | 1178 | set_kprobe_instance(kp); |
64f562c6 | 1179 | kp->post_handler(kp, regs, flags); |
e6584523 | 1180 | reset_kprobe_instance(); |
64f562c6 AM |
1181 | } |
1182 | } | |
64f562c6 | 1183 | } |
820aede0 | 1184 | NOKPROBE_SYMBOL(aggr_post_handler); |
64f562c6 | 1185 | |
820aede0 MH |
1186 | static int aggr_fault_handler(struct kprobe *p, struct pt_regs *regs, |
1187 | int trapnr) | |
64f562c6 | 1188 | { |
b76834bc | 1189 | struct kprobe *cur = __this_cpu_read(kprobe_instance); |
e6584523 | 1190 | |
64f562c6 AM |
1191 | /* |
1192 | * if we faulted "during" the execution of a user specified | |
1193 | * probe handler, invoke just that probe's fault handler | |
1194 | */ | |
e6584523 AM |
1195 | if (cur && cur->fault_handler) { |
1196 | if (cur->fault_handler(cur, regs, trapnr)) | |
64f562c6 AM |
1197 | return 1; |
1198 | } | |
1199 | return 0; | |
1200 | } | |
820aede0 | 1201 | NOKPROBE_SYMBOL(aggr_fault_handler); |
64f562c6 | 1202 | |
bf8d5c52 | 1203 | /* Walks the list and increments nmissed count for multiprobe case */ |
820aede0 | 1204 | void kprobes_inc_nmissed_count(struct kprobe *p) |
bf8d5c52 KA |
1205 | { |
1206 | struct kprobe *kp; | |
afd66255 | 1207 | if (!kprobe_aggrprobe(p)) { |
bf8d5c52 KA |
1208 | p->nmissed++; |
1209 | } else { | |
1210 | list_for_each_entry_rcu(kp, &p->list, list) | |
1211 | kp->nmissed++; | |
1212 | } | |
1213 | return; | |
1214 | } | |
820aede0 | 1215 | NOKPROBE_SYMBOL(kprobes_inc_nmissed_count); |
bf8d5c52 | 1216 | |
d741bf41 PZ |
1217 | static void free_rp_inst_rcu(struct rcu_head *head) |
1218 | { | |
1219 | struct kretprobe_instance *ri = container_of(head, struct kretprobe_instance, rcu); | |
1220 | ||
1221 | if (refcount_dec_and_test(&ri->rph->ref)) | |
1222 | kfree(ri->rph); | |
1223 | kfree(ri); | |
1224 | } | |
1225 | NOKPROBE_SYMBOL(free_rp_inst_rcu); | |
1226 | ||
b3388178 | 1227 | static void recycle_rp_inst(struct kretprobe_instance *ri) |
b94cce92 | 1228 | { |
d741bf41 | 1229 | struct kretprobe *rp = get_kretprobe(ri); |
ef53d9c5 | 1230 | |
ef53d9c5 | 1231 | if (likely(rp)) { |
6e426e0f | 1232 | freelist_add(&ri->freelist, &rp->freelist); |
b94cce92 | 1233 | } else |
d741bf41 | 1234 | call_rcu(&ri->rcu, free_rp_inst_rcu); |
b94cce92 | 1235 | } |
820aede0 | 1236 | NOKPROBE_SYMBOL(recycle_rp_inst); |
b94cce92 | 1237 | |
319f0ce2 | 1238 | static struct kprobe kprobe_busy = { |
9b38cc70 JO |
1239 | .addr = (void *) get_kprobe, |
1240 | }; | |
1241 | ||
1242 | void kprobe_busy_begin(void) | |
1243 | { | |
1244 | struct kprobe_ctlblk *kcb; | |
1245 | ||
1246 | preempt_disable(); | |
1247 | __this_cpu_write(current_kprobe, &kprobe_busy); | |
1248 | kcb = get_kprobe_ctlblk(); | |
1249 | kcb->kprobe_status = KPROBE_HIT_ACTIVE; | |
1250 | } | |
1251 | ||
1252 | void kprobe_busy_end(void) | |
1253 | { | |
1254 | __this_cpu_write(current_kprobe, NULL); | |
1255 | preempt_enable(); | |
1256 | } | |
1257 | ||
b94cce92 | 1258 | /* |
c6fd91f0 | 1259 | * This function is called from finish_task_switch when task tk becomes dead, |
1260 | * so that we can recycle any function-return probe instances associated | |
1261 | * with this task. These left over instances represent probed functions | |
1262 | * that have been called but will never return. | |
b94cce92 | 1263 | */ |
820aede0 | 1264 | void kprobe_flush_task(struct task_struct *tk) |
b94cce92 | 1265 | { |
62c27be0 | 1266 | struct kretprobe_instance *ri; |
d741bf41 | 1267 | struct llist_node *node; |
802eae7c | 1268 | |
d741bf41 | 1269 | /* Early boot, not yet initialized. */ |
ef53d9c5 | 1270 | if (unlikely(!kprobes_initialized)) |
ef53d9c5 S |
1271 | return; |
1272 | ||
9b38cc70 JO |
1273 | kprobe_busy_begin(); |
1274 | ||
d741bf41 PZ |
1275 | node = __llist_del_all(&tk->kretprobe_instances); |
1276 | while (node) { | |
1277 | ri = container_of(node, struct kretprobe_instance, llist); | |
1278 | node = node->next; | |
1279 | ||
1280 | recycle_rp_inst(ri); | |
62c27be0 | 1281 | } |
9b38cc70 JO |
1282 | |
1283 | kprobe_busy_end(); | |
b94cce92 | 1284 | } |
820aede0 | 1285 | NOKPROBE_SYMBOL(kprobe_flush_task); |
b94cce92 | 1286 | |
b94cce92 HN |
1287 | static inline void free_rp_inst(struct kretprobe *rp) |
1288 | { | |
1289 | struct kretprobe_instance *ri; | |
6e426e0f | 1290 | struct freelist_node *node; |
d741bf41 | 1291 | int count = 0; |
4c4308cb | 1292 | |
6e426e0f PZ |
1293 | node = rp->freelist.head; |
1294 | while (node) { | |
1295 | ri = container_of(node, struct kretprobe_instance, freelist); | |
1296 | node = node->next; | |
1297 | ||
b94cce92 | 1298 | kfree(ri); |
d741bf41 | 1299 | count++; |
b94cce92 | 1300 | } |
ef53d9c5 | 1301 | |
d741bf41 PZ |
1302 | if (refcount_sub_and_test(count, &rp->rph->ref)) { |
1303 | kfree(rp->rph); | |
1304 | rp->rph = NULL; | |
4a296e07 | 1305 | } |
4a296e07 MH |
1306 | } |
1307 | ||
059053a2 | 1308 | /* Add the new probe to ap->list */ |
55479f64 | 1309 | static int add_new_kprobe(struct kprobe *ap, struct kprobe *p) |
8b0914ea | 1310 | { |
059053a2 | 1311 | if (p->post_handler) |
6274de49 | 1312 | unoptimize_kprobe(ap, true); /* Fall back to normal kprobe */ |
afd66255 | 1313 | |
059053a2 | 1314 | list_add_rcu(&p->list, &ap->list); |
b918e5e6 MH |
1315 | if (p->post_handler && !ap->post_handler) |
1316 | ap->post_handler = aggr_post_handler; | |
de5bd88d | 1317 | |
8b0914ea PP |
1318 | return 0; |
1319 | } | |
1320 | ||
64f562c6 AM |
1321 | /* |
1322 | * Fill in the required fields of the "manager kprobe". Replace the | |
1323 | * earlier kprobe in the hlist with the manager kprobe | |
1324 | */ | |
55479f64 | 1325 | static void init_aggr_kprobe(struct kprobe *ap, struct kprobe *p) |
64f562c6 | 1326 | { |
afd66255 | 1327 | /* Copy p's insn slot to ap */ |
8b0914ea | 1328 | copy_kprobe(p, ap); |
a9ad965e | 1329 | flush_insn_slot(ap); |
64f562c6 | 1330 | ap->addr = p->addr; |
afd66255 | 1331 | ap->flags = p->flags & ~KPROBE_FLAG_OPTIMIZED; |
64f562c6 | 1332 | ap->pre_handler = aggr_pre_handler; |
64f562c6 | 1333 | ap->fault_handler = aggr_fault_handler; |
e8386a0c MH |
1334 | /* We don't care the kprobe which has gone. */ |
1335 | if (p->post_handler && !kprobe_gone(p)) | |
36721656 | 1336 | ap->post_handler = aggr_post_handler; |
64f562c6 AM |
1337 | |
1338 | INIT_LIST_HEAD(&ap->list); | |
afd66255 | 1339 | INIT_HLIST_NODE(&ap->hlist); |
64f562c6 | 1340 | |
afd66255 | 1341 | list_add_rcu(&p->list, &ap->list); |
adad0f33 | 1342 | hlist_replace_rcu(&p->hlist, &ap->hlist); |
64f562c6 AM |
1343 | } |
1344 | ||
1345 | /* | |
1346 | * This is the second or subsequent kprobe at the address - handle | |
1347 | * the intricacies | |
64f562c6 | 1348 | */ |
55479f64 | 1349 | static int register_aggr_kprobe(struct kprobe *orig_p, struct kprobe *p) |
64f562c6 AM |
1350 | { |
1351 | int ret = 0; | |
6d8e40a8 | 1352 | struct kprobe *ap = orig_p; |
64f562c6 | 1353 | |
2d1e38f5 TG |
1354 | cpus_read_lock(); |
1355 | ||
25764288 MH |
1356 | /* For preparing optimization, jump_label_text_reserved() is called */ |
1357 | jump_label_lock(); | |
25764288 MH |
1358 | mutex_lock(&text_mutex); |
1359 | ||
6d8e40a8 MH |
1360 | if (!kprobe_aggrprobe(orig_p)) { |
1361 | /* If orig_p is not an aggr_kprobe, create new aggr_kprobe. */ | |
1362 | ap = alloc_aggr_kprobe(orig_p); | |
25764288 MH |
1363 | if (!ap) { |
1364 | ret = -ENOMEM; | |
1365 | goto out; | |
1366 | } | |
6d8e40a8 | 1367 | init_aggr_kprobe(ap, orig_p); |
819319fc | 1368 | } else if (kprobe_unused(ap)) { |
0490cd1f | 1369 | /* This probe is going to die. Rescue it */ |
819319fc MH |
1370 | ret = reuse_unused_kprobe(ap); |
1371 | if (ret) | |
1372 | goto out; | |
1373 | } | |
b918e5e6 MH |
1374 | |
1375 | if (kprobe_gone(ap)) { | |
e8386a0c MH |
1376 | /* |
1377 | * Attempting to insert new probe at the same location that | |
1378 | * had a probe in the module vaddr area which already | |
1379 | * freed. So, the instruction slot has already been | |
1380 | * released. We need a new slot for the new probe. | |
1381 | */ | |
b918e5e6 | 1382 | ret = arch_prepare_kprobe(ap); |
e8386a0c | 1383 | if (ret) |
b918e5e6 MH |
1384 | /* |
1385 | * Even if fail to allocate new slot, don't need to | |
1386 | * free aggr_probe. It will be used next time, or | |
1387 | * freed by unregister_kprobe. | |
1388 | */ | |
25764288 | 1389 | goto out; |
de5bd88d | 1390 | |
afd66255 MH |
1391 | /* Prepare optimized instructions if possible. */ |
1392 | prepare_optimized_kprobe(ap); | |
1393 | ||
e8386a0c | 1394 | /* |
de5bd88d MH |
1395 | * Clear gone flag to prevent allocating new slot again, and |
1396 | * set disabled flag because it is not armed yet. | |
e8386a0c | 1397 | */ |
de5bd88d MH |
1398 | ap->flags = (ap->flags & ~KPROBE_FLAG_GONE) |
1399 | | KPROBE_FLAG_DISABLED; | |
e8386a0c | 1400 | } |
b918e5e6 | 1401 | |
afd66255 | 1402 | /* Copy ap's insn slot to p */ |
b918e5e6 | 1403 | copy_kprobe(ap, p); |
25764288 MH |
1404 | ret = add_new_kprobe(ap, p); |
1405 | ||
1406 | out: | |
1407 | mutex_unlock(&text_mutex); | |
25764288 | 1408 | jump_label_unlock(); |
2d1e38f5 | 1409 | cpus_read_unlock(); |
25764288 MH |
1410 | |
1411 | if (ret == 0 && kprobe_disabled(ap) && !kprobe_disabled(p)) { | |
1412 | ap->flags &= ~KPROBE_FLAG_DISABLED; | |
12310e34 | 1413 | if (!kprobes_all_disarmed) { |
25764288 | 1414 | /* Arm the breakpoint again. */ |
12310e34 JY |
1415 | ret = arm_kprobe(ap); |
1416 | if (ret) { | |
1417 | ap->flags |= KPROBE_FLAG_DISABLED; | |
1418 | list_del_rcu(&p->list); | |
ae8b7ce7 | 1419 | synchronize_rcu(); |
12310e34 JY |
1420 | } |
1421 | } | |
25764288 MH |
1422 | } |
1423 | return ret; | |
64f562c6 AM |
1424 | } |
1425 | ||
be8f2743 MH |
1426 | bool __weak arch_within_kprobe_blacklist(unsigned long addr) |
1427 | { | |
1428 | /* The __kprobes marked functions and entry code must not be probed */ | |
1429 | return addr >= (unsigned long)__kprobes_text_start && | |
1430 | addr < (unsigned long)__kprobes_text_end; | |
1431 | } | |
1432 | ||
6143c6fb | 1433 | static bool __within_kprobe_blacklist(unsigned long addr) |
d0aaff97 | 1434 | { |
376e2424 | 1435 | struct kprobe_blacklist_entry *ent; |
3d8d996e | 1436 | |
be8f2743 | 1437 | if (arch_within_kprobe_blacklist(addr)) |
376e2424 | 1438 | return true; |
3d8d996e SD |
1439 | /* |
1440 | * If there exists a kprobe_blacklist, verify and | |
1441 | * fail any probe registration in the prohibited area | |
1442 | */ | |
376e2424 MH |
1443 | list_for_each_entry(ent, &kprobe_blacklist, list) { |
1444 | if (addr >= ent->start_addr && addr < ent->end_addr) | |
1445 | return true; | |
3d8d996e | 1446 | } |
6143c6fb MH |
1447 | return false; |
1448 | } | |
376e2424 | 1449 | |
6143c6fb MH |
1450 | bool within_kprobe_blacklist(unsigned long addr) |
1451 | { | |
1452 | char symname[KSYM_NAME_LEN], *p; | |
1453 | ||
1454 | if (__within_kprobe_blacklist(addr)) | |
1455 | return true; | |
1456 | ||
1457 | /* Check if the address is on a suffixed-symbol */ | |
1458 | if (!lookup_symbol_name(addr, symname)) { | |
1459 | p = strchr(symname, '.'); | |
1460 | if (!p) | |
1461 | return false; | |
1462 | *p = '\0'; | |
1463 | addr = (unsigned long)kprobe_lookup_name(symname, 0); | |
1464 | if (addr) | |
1465 | return __within_kprobe_blacklist(addr); | |
1466 | } | |
376e2424 | 1467 | return false; |
d0aaff97 PP |
1468 | } |
1469 | ||
b2a5cd69 MH |
1470 | /* |
1471 | * If we have a symbol_name argument, look it up and add the offset field | |
1472 | * to it. This way, we can specify a relative address to a symbol. | |
bc81d48d MH |
1473 | * This returns encoded errors if it fails to look up symbol or invalid |
1474 | * combination of parameters. | |
b2a5cd69 | 1475 | */ |
1d585e70 NR |
1476 | static kprobe_opcode_t *_kprobe_addr(kprobe_opcode_t *addr, |
1477 | const char *symbol_name, unsigned int offset) | |
b2a5cd69 | 1478 | { |
1d585e70 | 1479 | if ((symbol_name && addr) || (!symbol_name && !addr)) |
bc81d48d MH |
1480 | goto invalid; |
1481 | ||
1d585e70 | 1482 | if (symbol_name) { |
7246f600 | 1483 | addr = kprobe_lookup_name(symbol_name, offset); |
bc81d48d MH |
1484 | if (!addr) |
1485 | return ERR_PTR(-ENOENT); | |
b2a5cd69 MH |
1486 | } |
1487 | ||
1d585e70 | 1488 | addr = (kprobe_opcode_t *)(((char *)addr) + offset); |
bc81d48d MH |
1489 | if (addr) |
1490 | return addr; | |
1491 | ||
1492 | invalid: | |
1493 | return ERR_PTR(-EINVAL); | |
b2a5cd69 MH |
1494 | } |
1495 | ||
1d585e70 NR |
1496 | static kprobe_opcode_t *kprobe_addr(struct kprobe *p) |
1497 | { | |
1498 | return _kprobe_addr(p->addr, p->symbol_name, p->offset); | |
1499 | } | |
1500 | ||
1f0ab409 | 1501 | /* Check passed kprobe is valid and return kprobe in kprobe_table. */ |
55479f64 | 1502 | static struct kprobe *__get_valid_kprobe(struct kprobe *p) |
1f0ab409 | 1503 | { |
6d8e40a8 | 1504 | struct kprobe *ap, *list_p; |
1f0ab409 | 1505 | |
7e6a71d8 MH |
1506 | lockdep_assert_held(&kprobe_mutex); |
1507 | ||
6d8e40a8 MH |
1508 | ap = get_kprobe(p->addr); |
1509 | if (unlikely(!ap)) | |
1f0ab409 AM |
1510 | return NULL; |
1511 | ||
6d8e40a8 | 1512 | if (p != ap) { |
7e6a71d8 | 1513 | list_for_each_entry(list_p, &ap->list, list) |
1f0ab409 AM |
1514 | if (list_p == p) |
1515 | /* kprobe p is a valid probe */ | |
1516 | goto valid; | |
1517 | return NULL; | |
1518 | } | |
1519 | valid: | |
6d8e40a8 | 1520 | return ap; |
1f0ab409 AM |
1521 | } |
1522 | ||
1523 | /* Return error if the kprobe is being re-registered */ | |
1524 | static inline int check_kprobe_rereg(struct kprobe *p) | |
1525 | { | |
1526 | int ret = 0; | |
1f0ab409 AM |
1527 | |
1528 | mutex_lock(&kprobe_mutex); | |
6d8e40a8 | 1529 | if (__get_valid_kprobe(p)) |
1f0ab409 AM |
1530 | ret = -EINVAL; |
1531 | mutex_unlock(&kprobe_mutex); | |
6d8e40a8 | 1532 | |
1f0ab409 AM |
1533 | return ret; |
1534 | } | |
1535 | ||
f7f242ff | 1536 | int __weak arch_check_ftrace_location(struct kprobe *p) |
1da177e4 | 1537 | { |
ae6aa16f MH |
1538 | unsigned long ftrace_addr; |
1539 | ||
ae6aa16f MH |
1540 | ftrace_addr = ftrace_location((unsigned long)p->addr); |
1541 | if (ftrace_addr) { | |
e7dbfe34 | 1542 | #ifdef CONFIG_KPROBES_ON_FTRACE |
ae6aa16f MH |
1543 | /* Given address is not on the instruction boundary */ |
1544 | if ((unsigned long)p->addr != ftrace_addr) | |
1545 | return -EILSEQ; | |
ae6aa16f | 1546 | p->flags |= KPROBE_FLAG_FTRACE; |
e7dbfe34 | 1547 | #else /* !CONFIG_KPROBES_ON_FTRACE */ |
ae6aa16f MH |
1548 | return -EINVAL; |
1549 | #endif | |
1550 | } | |
f7f242ff HC |
1551 | return 0; |
1552 | } | |
1553 | ||
1554 | static int check_kprobe_address_safe(struct kprobe *p, | |
1555 | struct module **probed_mod) | |
1556 | { | |
1557 | int ret; | |
1f0ab409 | 1558 | |
f7f242ff HC |
1559 | ret = arch_check_ftrace_location(p); |
1560 | if (ret) | |
1561 | return ret; | |
91bad2f8 | 1562 | jump_label_lock(); |
de31c3ca | 1563 | preempt_disable(); |
f7fa6ef0 MH |
1564 | |
1565 | /* Ensure it is not in reserved area nor out of text */ | |
ec30c5f3 | 1566 | if (!kernel_text_address((unsigned long) p->addr) || |
376e2424 | 1567 | within_kprobe_blacklist((unsigned long) p->addr) || |
e336b402 MH |
1568 | jump_label_text_reserved(p->addr, p->addr) || |
1569 | find_bug((unsigned long)p->addr)) { | |
f986a499 | 1570 | ret = -EINVAL; |
f7fa6ef0 | 1571 | goto out; |
f986a499 | 1572 | } |
b3e55c72 | 1573 | |
f7fa6ef0 MH |
1574 | /* Check if are we probing a module */ |
1575 | *probed_mod = __module_text_address((unsigned long) p->addr); | |
1576 | if (*probed_mod) { | |
6f716acd | 1577 | /* |
e8386a0c MH |
1578 | * We must hold a refcount of the probed module while updating |
1579 | * its code to prohibit unexpected unloading. | |
df019b1d | 1580 | */ |
f7fa6ef0 MH |
1581 | if (unlikely(!try_module_get(*probed_mod))) { |
1582 | ret = -ENOENT; | |
1583 | goto out; | |
1584 | } | |
de31c3ca | 1585 | |
f24659d9 MH |
1586 | /* |
1587 | * If the module freed .init.text, we couldn't insert | |
1588 | * kprobes in there. | |
1589 | */ | |
f7fa6ef0 MH |
1590 | if (within_module_init((unsigned long)p->addr, *probed_mod) && |
1591 | (*probed_mod)->state != MODULE_STATE_COMING) { | |
1592 | module_put(*probed_mod); | |
1593 | *probed_mod = NULL; | |
1594 | ret = -ENOENT; | |
f24659d9 | 1595 | } |
df019b1d | 1596 | } |
f7fa6ef0 | 1597 | out: |
a189d035 | 1598 | preempt_enable(); |
de31c3ca | 1599 | jump_label_unlock(); |
1da177e4 | 1600 | |
f7fa6ef0 MH |
1601 | return ret; |
1602 | } | |
1603 | ||
55479f64 | 1604 | int register_kprobe(struct kprobe *p) |
f7fa6ef0 MH |
1605 | { |
1606 | int ret; | |
1607 | struct kprobe *old_p; | |
1608 | struct module *probed_mod; | |
1609 | kprobe_opcode_t *addr; | |
1610 | ||
1611 | /* Adjust probe address from symbol */ | |
1612 | addr = kprobe_addr(p); | |
1613 | if (IS_ERR(addr)) | |
1614 | return PTR_ERR(addr); | |
1615 | p->addr = addr; | |
1616 | ||
1617 | ret = check_kprobe_rereg(p); | |
1618 | if (ret) | |
1619 | return ret; | |
1620 | ||
1621 | /* User can pass only KPROBE_FLAG_DISABLED to register_kprobe */ | |
1622 | p->flags &= KPROBE_FLAG_DISABLED; | |
3516a460 | 1623 | p->nmissed = 0; |
9861668f | 1624 | INIT_LIST_HEAD(&p->list); |
afd66255 | 1625 | |
f7fa6ef0 MH |
1626 | ret = check_kprobe_address_safe(p, &probed_mod); |
1627 | if (ret) | |
1628 | return ret; | |
1629 | ||
1630 | mutex_lock(&kprobe_mutex); | |
afd66255 | 1631 | |
64f562c6 AM |
1632 | old_p = get_kprobe(p->addr); |
1633 | if (old_p) { | |
afd66255 | 1634 | /* Since this may unoptimize old_p, locking text_mutex. */ |
64f562c6 | 1635 | ret = register_aggr_kprobe(old_p, p); |
1da177e4 LT |
1636 | goto out; |
1637 | } | |
1da177e4 | 1638 | |
2d1e38f5 TG |
1639 | cpus_read_lock(); |
1640 | /* Prevent text modification */ | |
1641 | mutex_lock(&text_mutex); | |
ae6aa16f | 1642 | ret = prepare_kprobe(p); |
25764288 | 1643 | mutex_unlock(&text_mutex); |
2d1e38f5 | 1644 | cpus_read_unlock(); |
6f716acd | 1645 | if (ret) |
afd66255 | 1646 | goto out; |
49a2a1b8 | 1647 | |
64f562c6 | 1648 | INIT_HLIST_NODE(&p->hlist); |
3516a460 | 1649 | hlist_add_head_rcu(&p->hlist, |
1da177e4 LT |
1650 | &kprobe_table[hash_ptr(p->addr, KPROBE_HASH_BITS)]); |
1651 | ||
12310e34 JY |
1652 | if (!kprobes_all_disarmed && !kprobe_disabled(p)) { |
1653 | ret = arm_kprobe(p); | |
1654 | if (ret) { | |
1655 | hlist_del_rcu(&p->hlist); | |
ae8b7ce7 | 1656 | synchronize_rcu(); |
12310e34 JY |
1657 | goto out; |
1658 | } | |
1659 | } | |
afd66255 MH |
1660 | |
1661 | /* Try to optimize kprobe */ | |
1662 | try_to_optimize_kprobe(p); | |
1da177e4 | 1663 | out: |
7a7d1cf9 | 1664 | mutex_unlock(&kprobe_mutex); |
49a2a1b8 | 1665 | |
e8386a0c | 1666 | if (probed_mod) |
df019b1d | 1667 | module_put(probed_mod); |
e8386a0c | 1668 | |
1da177e4 LT |
1669 | return ret; |
1670 | } | |
99081ab5 | 1671 | EXPORT_SYMBOL_GPL(register_kprobe); |
1da177e4 | 1672 | |
6f0f1dd7 | 1673 | /* Check if all probes on the aggrprobe are disabled */ |
55479f64 | 1674 | static int aggr_kprobe_disabled(struct kprobe *ap) |
6f0f1dd7 MH |
1675 | { |
1676 | struct kprobe *kp; | |
1677 | ||
7e6a71d8 MH |
1678 | lockdep_assert_held(&kprobe_mutex); |
1679 | ||
1680 | list_for_each_entry(kp, &ap->list, list) | |
6f0f1dd7 MH |
1681 | if (!kprobe_disabled(kp)) |
1682 | /* | |
1683 | * There is an active probe on the list. | |
1684 | * We can't disable this ap. | |
1685 | */ | |
1686 | return 0; | |
1687 | ||
1688 | return 1; | |
1689 | } | |
1690 | ||
1691 | /* Disable one kprobe: Make sure called under kprobe_mutex is locked */ | |
55479f64 | 1692 | static struct kprobe *__disable_kprobe(struct kprobe *p) |
6f0f1dd7 MH |
1693 | { |
1694 | struct kprobe *orig_p; | |
297f9233 | 1695 | int ret; |
6f0f1dd7 MH |
1696 | |
1697 | /* Get an original kprobe for return */ | |
1698 | orig_p = __get_valid_kprobe(p); | |
1699 | if (unlikely(orig_p == NULL)) | |
297f9233 | 1700 | return ERR_PTR(-EINVAL); |
6f0f1dd7 MH |
1701 | |
1702 | if (!kprobe_disabled(p)) { | |
1703 | /* Disable probe if it is a child probe */ | |
1704 | if (p != orig_p) | |
1705 | p->flags |= KPROBE_FLAG_DISABLED; | |
1706 | ||
1707 | /* Try to disarm and disable this/parent probe */ | |
1708 | if (p == orig_p || aggr_kprobe_disabled(orig_p)) { | |
69d54b91 WN |
1709 | /* |
1710 | * If kprobes_all_disarmed is set, orig_p | |
1711 | * should have already been disarmed, so | |
1712 | * skip unneed disarming process. | |
1713 | */ | |
297f9233 JY |
1714 | if (!kprobes_all_disarmed) { |
1715 | ret = disarm_kprobe(orig_p, true); | |
1716 | if (ret) { | |
1717 | p->flags &= ~KPROBE_FLAG_DISABLED; | |
1718 | return ERR_PTR(ret); | |
1719 | } | |
1720 | } | |
6f0f1dd7 MH |
1721 | orig_p->flags |= KPROBE_FLAG_DISABLED; |
1722 | } | |
1723 | } | |
1724 | ||
1725 | return orig_p; | |
1726 | } | |
1727 | ||
de5bd88d MH |
1728 | /* |
1729 | * Unregister a kprobe without a scheduler synchronization. | |
1730 | */ | |
55479f64 | 1731 | static int __unregister_kprobe_top(struct kprobe *p) |
de5bd88d | 1732 | { |
6d8e40a8 | 1733 | struct kprobe *ap, *list_p; |
de5bd88d | 1734 | |
6f0f1dd7 MH |
1735 | /* Disable kprobe. This will disarm it if needed. */ |
1736 | ap = __disable_kprobe(p); | |
297f9233 JY |
1737 | if (IS_ERR(ap)) |
1738 | return PTR_ERR(ap); | |
de5bd88d | 1739 | |
6f0f1dd7 | 1740 | if (ap == p) |
bf8f6e5b | 1741 | /* |
6f0f1dd7 MH |
1742 | * This probe is an independent(and non-optimized) kprobe |
1743 | * (not an aggrprobe). Remove from the hash list. | |
bf8f6e5b | 1744 | */ |
6f0f1dd7 MH |
1745 | goto disarmed; |
1746 | ||
1747 | /* Following process expects this probe is an aggrprobe */ | |
1748 | WARN_ON(!kprobe_aggrprobe(ap)); | |
1749 | ||
6274de49 MH |
1750 | if (list_is_singular(&ap->list) && kprobe_disarmed(ap)) |
1751 | /* | |
1752 | * !disarmed could be happen if the probe is under delayed | |
1753 | * unoptimizing. | |
1754 | */ | |
6f0f1dd7 MH |
1755 | goto disarmed; |
1756 | else { | |
1757 | /* If disabling probe has special handlers, update aggrprobe */ | |
e8386a0c | 1758 | if (p->post_handler && !kprobe_gone(p)) { |
7e6a71d8 | 1759 | list_for_each_entry(list_p, &ap->list, list) { |
9861668f MH |
1760 | if ((list_p != p) && (list_p->post_handler)) |
1761 | goto noclean; | |
1762 | } | |
6d8e40a8 | 1763 | ap->post_handler = NULL; |
9861668f MH |
1764 | } |
1765 | noclean: | |
6f0f1dd7 MH |
1766 | /* |
1767 | * Remove from the aggrprobe: this path will do nothing in | |
1768 | * __unregister_kprobe_bottom(). | |
1769 | */ | |
49a2a1b8 | 1770 | list_del_rcu(&p->list); |
6f0f1dd7 MH |
1771 | if (!kprobe_disabled(ap) && !kprobes_all_disarmed) |
1772 | /* | |
1773 | * Try to optimize this probe again, because post | |
1774 | * handler may have been changed. | |
1775 | */ | |
1776 | optimize_kprobe(ap); | |
49a2a1b8 | 1777 | } |
9861668f | 1778 | return 0; |
6f0f1dd7 MH |
1779 | |
1780 | disarmed: | |
1781 | hlist_del_rcu(&ap->hlist); | |
1782 | return 0; | |
9861668f | 1783 | } |
3516a460 | 1784 | |
55479f64 | 1785 | static void __unregister_kprobe_bottom(struct kprobe *p) |
9861668f | 1786 | { |
6d8e40a8 | 1787 | struct kprobe *ap; |
b3e55c72 | 1788 | |
e8386a0c | 1789 | if (list_empty(&p->list)) |
6274de49 | 1790 | /* This is an independent kprobe */ |
0498b635 | 1791 | arch_remove_kprobe(p); |
e8386a0c | 1792 | else if (list_is_singular(&p->list)) { |
6274de49 | 1793 | /* This is the last child of an aggrprobe */ |
6d8e40a8 | 1794 | ap = list_entry(p->list.next, struct kprobe, list); |
e8386a0c | 1795 | list_del(&p->list); |
6d8e40a8 | 1796 | free_aggr_kprobe(ap); |
9861668f | 1797 | } |
6274de49 | 1798 | /* Otherwise, do nothing. */ |
9861668f MH |
1799 | } |
1800 | ||
55479f64 | 1801 | int register_kprobes(struct kprobe **kps, int num) |
9861668f MH |
1802 | { |
1803 | int i, ret = 0; | |
1804 | ||
1805 | if (num <= 0) | |
1806 | return -EINVAL; | |
1807 | for (i = 0; i < num; i++) { | |
49ad2fd7 | 1808 | ret = register_kprobe(kps[i]); |
67dddaad MH |
1809 | if (ret < 0) { |
1810 | if (i > 0) | |
1811 | unregister_kprobes(kps, i); | |
9861668f | 1812 | break; |
36721656 | 1813 | } |
49a2a1b8 | 1814 | } |
9861668f MH |
1815 | return ret; |
1816 | } | |
99081ab5 | 1817 | EXPORT_SYMBOL_GPL(register_kprobes); |
9861668f | 1818 | |
55479f64 | 1819 | void unregister_kprobe(struct kprobe *p) |
9861668f MH |
1820 | { |
1821 | unregister_kprobes(&p, 1); | |
1822 | } | |
99081ab5 | 1823 | EXPORT_SYMBOL_GPL(unregister_kprobe); |
9861668f | 1824 | |
55479f64 | 1825 | void unregister_kprobes(struct kprobe **kps, int num) |
9861668f MH |
1826 | { |
1827 | int i; | |
1828 | ||
1829 | if (num <= 0) | |
1830 | return; | |
1831 | mutex_lock(&kprobe_mutex); | |
1832 | for (i = 0; i < num; i++) | |
1833 | if (__unregister_kprobe_top(kps[i]) < 0) | |
1834 | kps[i]->addr = NULL; | |
1835 | mutex_unlock(&kprobe_mutex); | |
1836 | ||
ae8b7ce7 | 1837 | synchronize_rcu(); |
9861668f MH |
1838 | for (i = 0; i < num; i++) |
1839 | if (kps[i]->addr) | |
1840 | __unregister_kprobe_bottom(kps[i]); | |
1da177e4 | 1841 | } |
99081ab5 | 1842 | EXPORT_SYMBOL_GPL(unregister_kprobes); |
1da177e4 | 1843 | |
5f6bee34 NR |
1844 | int __weak kprobe_exceptions_notify(struct notifier_block *self, |
1845 | unsigned long val, void *data) | |
fc62d020 NR |
1846 | { |
1847 | return NOTIFY_DONE; | |
1848 | } | |
5f6bee34 | 1849 | NOKPROBE_SYMBOL(kprobe_exceptions_notify); |
fc62d020 | 1850 | |
1da177e4 | 1851 | static struct notifier_block kprobe_exceptions_nb = { |
3d5631e0 AK |
1852 | .notifier_call = kprobe_exceptions_notify, |
1853 | .priority = 0x7fffffff /* we need to be notified first */ | |
1854 | }; | |
1855 | ||
3d7e3382 ME |
1856 | unsigned long __weak arch_deref_entry_point(void *entry) |
1857 | { | |
1858 | return (unsigned long)entry; | |
1859 | } | |
1da177e4 | 1860 | |
9edddaa2 | 1861 | #ifdef CONFIG_KRETPROBES |
66ada2cc MH |
1862 | |
1863 | unsigned long __kretprobe_trampoline_handler(struct pt_regs *regs, | |
1864 | void *trampoline_address, | |
1865 | void *frame_pointer) | |
1866 | { | |
66ada2cc | 1867 | kprobe_opcode_t *correct_ret_addr = NULL; |
d741bf41 PZ |
1868 | struct kretprobe_instance *ri = NULL; |
1869 | struct llist_node *first, *node; | |
1870 | struct kretprobe *rp; | |
66ada2cc | 1871 | |
d741bf41 PZ |
1872 | /* Find all nodes for this frame. */ |
1873 | first = node = current->kretprobe_instances.first; | |
1874 | while (node) { | |
1875 | ri = container_of(node, struct kretprobe_instance, llist); | |
66ada2cc | 1876 | |
d741bf41 | 1877 | BUG_ON(ri->fp != frame_pointer); |
66ada2cc | 1878 | |
d741bf41 PZ |
1879 | if (ri->ret_addr != trampoline_address) { |
1880 | correct_ret_addr = ri->ret_addr; | |
66ada2cc MH |
1881 | /* |
1882 | * This is the real return address. Any other | |
1883 | * instances associated with this task are for | |
1884 | * other calls deeper on the call stack | |
1885 | */ | |
d741bf41 PZ |
1886 | goto found; |
1887 | } | |
1888 | ||
1889 | node = node->next; | |
66ada2cc | 1890 | } |
d741bf41 PZ |
1891 | pr_err("Oops! Kretprobe fails to find correct return address.\n"); |
1892 | BUG_ON(1); | |
66ada2cc | 1893 | |
d741bf41 PZ |
1894 | found: |
1895 | /* Unlink all nodes for this frame. */ | |
1896 | current->kretprobe_instances.first = node->next; | |
1897 | node->next = NULL; | |
66ada2cc | 1898 | |
d741bf41 PZ |
1899 | /* Run them.. */ |
1900 | while (first) { | |
1901 | ri = container_of(first, struct kretprobe_instance, llist); | |
1902 | first = first->next; | |
66ada2cc | 1903 | |
d741bf41 PZ |
1904 | rp = get_kretprobe(ri); |
1905 | if (rp && rp->handler) { | |
66ada2cc MH |
1906 | struct kprobe *prev = kprobe_running(); |
1907 | ||
d741bf41 | 1908 | __this_cpu_write(current_kprobe, &rp->kp); |
66ada2cc | 1909 | ri->ret_addr = correct_ret_addr; |
d741bf41 | 1910 | rp->handler(ri, regs); |
66ada2cc MH |
1911 | __this_cpu_write(current_kprobe, prev); |
1912 | } | |
1913 | ||
b3388178 | 1914 | recycle_rp_inst(ri); |
66ada2cc MH |
1915 | } |
1916 | ||
66ada2cc MH |
1917 | return (unsigned long)correct_ret_addr; |
1918 | } | |
1919 | NOKPROBE_SYMBOL(__kretprobe_trampoline_handler) | |
1920 | ||
e65cefe8 AB |
1921 | /* |
1922 | * This kprobe pre_handler is registered with every kretprobe. When probe | |
1923 | * hits it will set up the return probe. | |
1924 | */ | |
820aede0 | 1925 | static int pre_handler_kretprobe(struct kprobe *p, struct pt_regs *regs) |
e65cefe8 AB |
1926 | { |
1927 | struct kretprobe *rp = container_of(p, struct kretprobe, kp); | |
ef53d9c5 | 1928 | struct kretprobe_instance *ri; |
6e426e0f | 1929 | struct freelist_node *fn; |
e65cefe8 | 1930 | |
6e426e0f PZ |
1931 | fn = freelist_try_get(&rp->freelist); |
1932 | if (!fn) { | |
1933 | rp->nmissed++; | |
1934 | return 0; | |
1935 | } | |
4c4308cb | 1936 | |
6e426e0f | 1937 | ri = container_of(fn, struct kretprobe_instance, freelist); |
d741bf41 | 1938 | |
6e426e0f PZ |
1939 | if (rp->entry_handler && rp->entry_handler(ri, regs)) { |
1940 | freelist_add(&ri->freelist, &rp->freelist); | |
1941 | return 0; | |
ef53d9c5 | 1942 | } |
6e426e0f PZ |
1943 | |
1944 | arch_prepare_kretprobe(ri, regs); | |
1945 | ||
1946 | __llist_add(&ri->llist, ¤t->kretprobe_instances); | |
1947 | ||
e65cefe8 AB |
1948 | return 0; |
1949 | } | |
820aede0 | 1950 | NOKPROBE_SYMBOL(pre_handler_kretprobe); |
e65cefe8 | 1951 | |
659b957f | 1952 | bool __weak arch_kprobe_on_func_entry(unsigned long offset) |
90ec5e89 NR |
1953 | { |
1954 | return !offset; | |
1955 | } | |
1956 | ||
97c753e6 MH |
1957 | /** |
1958 | * kprobe_on_func_entry() -- check whether given address is function entry | |
1959 | * @addr: Target address | |
1960 | * @sym: Target symbol name | |
1961 | * @offset: The offset from the symbol or the address | |
1962 | * | |
1963 | * This checks whether the given @addr+@offset or @sym+@offset is on the | |
1964 | * function entry address or not. | |
1965 | * This returns 0 if it is the function entry, or -EINVAL if it is not. | |
1966 | * And also it returns -ENOENT if it fails the symbol or address lookup. | |
1967 | * Caller must pass @addr or @sym (either one must be NULL), or this | |
1968 | * returns -EINVAL. | |
1969 | */ | |
1970 | int kprobe_on_func_entry(kprobe_opcode_t *addr, const char *sym, unsigned long offset) | |
1d585e70 NR |
1971 | { |
1972 | kprobe_opcode_t *kp_addr = _kprobe_addr(addr, sym, offset); | |
1973 | ||
1974 | if (IS_ERR(kp_addr)) | |
97c753e6 | 1975 | return PTR_ERR(kp_addr); |
1d585e70 | 1976 | |
97c753e6 MH |
1977 | if (!kallsyms_lookup_size_offset((unsigned long)kp_addr, NULL, &offset)) |
1978 | return -ENOENT; | |
1d585e70 | 1979 | |
97c753e6 MH |
1980 | if (!arch_kprobe_on_func_entry(offset)) |
1981 | return -EINVAL; | |
1982 | ||
1983 | return 0; | |
1d585e70 NR |
1984 | } |
1985 | ||
55479f64 | 1986 | int register_kretprobe(struct kretprobe *rp) |
b94cce92 | 1987 | { |
97c753e6 | 1988 | int ret; |
b94cce92 HN |
1989 | struct kretprobe_instance *inst; |
1990 | int i; | |
b2a5cd69 | 1991 | void *addr; |
90ec5e89 | 1992 | |
97c753e6 MH |
1993 | ret = kprobe_on_func_entry(rp->kp.addr, rp->kp.symbol_name, rp->kp.offset); |
1994 | if (ret) | |
1995 | return ret; | |
f438d914 | 1996 | |
0188b878 WS |
1997 | /* If only rp->kp.addr is specified, check reregistering kprobes */ |
1998 | if (rp->kp.addr && check_kprobe_rereg(&rp->kp)) | |
1999 | return -EINVAL; | |
2000 | ||
f438d914 | 2001 | if (kretprobe_blacklist_size) { |
b2a5cd69 | 2002 | addr = kprobe_addr(&rp->kp); |
bc81d48d MH |
2003 | if (IS_ERR(addr)) |
2004 | return PTR_ERR(addr); | |
f438d914 MH |
2005 | |
2006 | for (i = 0; kretprobe_blacklist[i].name != NULL; i++) { | |
2007 | if (kretprobe_blacklist[i].addr == addr) | |
2008 | return -EINVAL; | |
2009 | } | |
2010 | } | |
b94cce92 HN |
2011 | |
2012 | rp->kp.pre_handler = pre_handler_kretprobe; | |
7522a842 AM |
2013 | rp->kp.post_handler = NULL; |
2014 | rp->kp.fault_handler = NULL; | |
b94cce92 HN |
2015 | |
2016 | /* Pre-allocate memory for max kretprobe instances */ | |
2017 | if (rp->maxactive <= 0) { | |
92616606 | 2018 | #ifdef CONFIG_PREEMPTION |
c2ef6661 | 2019 | rp->maxactive = max_t(unsigned int, 10, 2*num_possible_cpus()); |
b94cce92 | 2020 | #else |
4dae560f | 2021 | rp->maxactive = num_possible_cpus(); |
b94cce92 HN |
2022 | #endif |
2023 | } | |
6e426e0f | 2024 | rp->freelist.head = NULL; |
d741bf41 PZ |
2025 | rp->rph = kzalloc(sizeof(struct kretprobe_holder), GFP_KERNEL); |
2026 | if (!rp->rph) | |
2027 | return -ENOMEM; | |
2028 | ||
2029 | rp->rph->rp = rp; | |
b94cce92 | 2030 | for (i = 0; i < rp->maxactive; i++) { |
d741bf41 | 2031 | inst = kzalloc(sizeof(struct kretprobe_instance) + |
f47cd9b5 | 2032 | rp->data_size, GFP_KERNEL); |
b94cce92 | 2033 | if (inst == NULL) { |
d741bf41 | 2034 | refcount_set(&rp->rph->ref, i); |
b94cce92 HN |
2035 | free_rp_inst(rp); |
2036 | return -ENOMEM; | |
2037 | } | |
d741bf41 | 2038 | inst->rph = rp->rph; |
6e426e0f | 2039 | freelist_add(&inst->freelist, &rp->freelist); |
b94cce92 | 2040 | } |
d741bf41 | 2041 | refcount_set(&rp->rph->ref, i); |
b94cce92 HN |
2042 | |
2043 | rp->nmissed = 0; | |
2044 | /* Establish function entry probe point */ | |
49ad2fd7 | 2045 | ret = register_kprobe(&rp->kp); |
4a296e07 | 2046 | if (ret != 0) |
b94cce92 HN |
2047 | free_rp_inst(rp); |
2048 | return ret; | |
2049 | } | |
99081ab5 | 2050 | EXPORT_SYMBOL_GPL(register_kretprobe); |
b94cce92 | 2051 | |
55479f64 | 2052 | int register_kretprobes(struct kretprobe **rps, int num) |
4a296e07 MH |
2053 | { |
2054 | int ret = 0, i; | |
2055 | ||
2056 | if (num <= 0) | |
2057 | return -EINVAL; | |
2058 | for (i = 0; i < num; i++) { | |
49ad2fd7 | 2059 | ret = register_kretprobe(rps[i]); |
67dddaad MH |
2060 | if (ret < 0) { |
2061 | if (i > 0) | |
2062 | unregister_kretprobes(rps, i); | |
4a296e07 MH |
2063 | break; |
2064 | } | |
2065 | } | |
2066 | return ret; | |
2067 | } | |
99081ab5 | 2068 | EXPORT_SYMBOL_GPL(register_kretprobes); |
4a296e07 | 2069 | |
55479f64 | 2070 | void unregister_kretprobe(struct kretprobe *rp) |
4a296e07 MH |
2071 | { |
2072 | unregister_kretprobes(&rp, 1); | |
2073 | } | |
99081ab5 | 2074 | EXPORT_SYMBOL_GPL(unregister_kretprobe); |
4a296e07 | 2075 | |
55479f64 | 2076 | void unregister_kretprobes(struct kretprobe **rps, int num) |
4a296e07 MH |
2077 | { |
2078 | int i; | |
2079 | ||
2080 | if (num <= 0) | |
2081 | return; | |
2082 | mutex_lock(&kprobe_mutex); | |
d741bf41 | 2083 | for (i = 0; i < num; i++) { |
4a296e07 MH |
2084 | if (__unregister_kprobe_top(&rps[i]->kp) < 0) |
2085 | rps[i]->kp.addr = NULL; | |
d741bf41 PZ |
2086 | rps[i]->rph->rp = NULL; |
2087 | } | |
4a296e07 MH |
2088 | mutex_unlock(&kprobe_mutex); |
2089 | ||
ae8b7ce7 | 2090 | synchronize_rcu(); |
4a296e07 MH |
2091 | for (i = 0; i < num; i++) { |
2092 | if (rps[i]->kp.addr) { | |
2093 | __unregister_kprobe_bottom(&rps[i]->kp); | |
d741bf41 | 2094 | free_rp_inst(rps[i]); |
4a296e07 MH |
2095 | } |
2096 | } | |
2097 | } | |
99081ab5 | 2098 | EXPORT_SYMBOL_GPL(unregister_kretprobes); |
4a296e07 | 2099 | |
9edddaa2 | 2100 | #else /* CONFIG_KRETPROBES */ |
55479f64 | 2101 | int register_kretprobe(struct kretprobe *rp) |
b94cce92 HN |
2102 | { |
2103 | return -ENOSYS; | |
2104 | } | |
99081ab5 | 2105 | EXPORT_SYMBOL_GPL(register_kretprobe); |
b94cce92 | 2106 | |
55479f64 | 2107 | int register_kretprobes(struct kretprobe **rps, int num) |
346fd59b | 2108 | { |
4a296e07 | 2109 | return -ENOSYS; |
346fd59b | 2110 | } |
99081ab5 MH |
2111 | EXPORT_SYMBOL_GPL(register_kretprobes); |
2112 | ||
55479f64 | 2113 | void unregister_kretprobe(struct kretprobe *rp) |
b94cce92 | 2114 | { |
4a296e07 | 2115 | } |
99081ab5 | 2116 | EXPORT_SYMBOL_GPL(unregister_kretprobe); |
b94cce92 | 2117 | |
55479f64 | 2118 | void unregister_kretprobes(struct kretprobe **rps, int num) |
4a296e07 MH |
2119 | { |
2120 | } | |
99081ab5 | 2121 | EXPORT_SYMBOL_GPL(unregister_kretprobes); |
4c4308cb | 2122 | |
820aede0 | 2123 | static int pre_handler_kretprobe(struct kprobe *p, struct pt_regs *regs) |
4a296e07 MH |
2124 | { |
2125 | return 0; | |
b94cce92 | 2126 | } |
820aede0 | 2127 | NOKPROBE_SYMBOL(pre_handler_kretprobe); |
b94cce92 | 2128 | |
4a296e07 MH |
2129 | #endif /* CONFIG_KRETPROBES */ |
2130 | ||
e8386a0c | 2131 | /* Set the kprobe gone and remove its instruction buffer. */ |
55479f64 | 2132 | static void kill_kprobe(struct kprobe *p) |
e8386a0c MH |
2133 | { |
2134 | struct kprobe *kp; | |
de5bd88d | 2135 | |
7e6a71d8 MH |
2136 | lockdep_assert_held(&kprobe_mutex); |
2137 | ||
e8386a0c | 2138 | p->flags |= KPROBE_FLAG_GONE; |
afd66255 | 2139 | if (kprobe_aggrprobe(p)) { |
e8386a0c MH |
2140 | /* |
2141 | * If this is an aggr_kprobe, we have to list all the | |
2142 | * chained probes and mark them GONE. | |
2143 | */ | |
7e6a71d8 | 2144 | list_for_each_entry(kp, &p->list, list) |
e8386a0c MH |
2145 | kp->flags |= KPROBE_FLAG_GONE; |
2146 | p->post_handler = NULL; | |
afd66255 | 2147 | kill_optimized_kprobe(p); |
e8386a0c MH |
2148 | } |
2149 | /* | |
2150 | * Here, we can remove insn_slot safely, because no thread calls | |
2151 | * the original probed function (which will be freed soon) any more. | |
2152 | */ | |
2153 | arch_remove_kprobe(p); | |
0cb2f137 MS |
2154 | |
2155 | /* | |
2156 | * The module is going away. We should disarm the kprobe which | |
bcb53209 MH |
2157 | * is using ftrace, because ftrace framework is still available at |
2158 | * MODULE_STATE_GOING notification. | |
0cb2f137 | 2159 | */ |
bcb53209 | 2160 | if (kprobe_ftrace(p) && !kprobe_disabled(p) && !kprobes_all_disarmed) |
0cb2f137 | 2161 | disarm_kprobe_ftrace(p); |
e8386a0c MH |
2162 | } |
2163 | ||
c0614829 | 2164 | /* Disable one kprobe */ |
55479f64 | 2165 | int disable_kprobe(struct kprobe *kp) |
c0614829 MH |
2166 | { |
2167 | int ret = 0; | |
297f9233 | 2168 | struct kprobe *p; |
c0614829 MH |
2169 | |
2170 | mutex_lock(&kprobe_mutex); | |
2171 | ||
6f0f1dd7 | 2172 | /* Disable this kprobe */ |
297f9233 JY |
2173 | p = __disable_kprobe(kp); |
2174 | if (IS_ERR(p)) | |
2175 | ret = PTR_ERR(p); | |
c0614829 | 2176 | |
c0614829 MH |
2177 | mutex_unlock(&kprobe_mutex); |
2178 | return ret; | |
2179 | } | |
2180 | EXPORT_SYMBOL_GPL(disable_kprobe); | |
2181 | ||
2182 | /* Enable one kprobe */ | |
55479f64 | 2183 | int enable_kprobe(struct kprobe *kp) |
c0614829 MH |
2184 | { |
2185 | int ret = 0; | |
2186 | struct kprobe *p; | |
2187 | ||
2188 | mutex_lock(&kprobe_mutex); | |
2189 | ||
2190 | /* Check whether specified probe is valid. */ | |
2191 | p = __get_valid_kprobe(kp); | |
2192 | if (unlikely(p == NULL)) { | |
2193 | ret = -EINVAL; | |
2194 | goto out; | |
2195 | } | |
2196 | ||
2197 | if (kprobe_gone(kp)) { | |
2198 | /* This kprobe has gone, we couldn't enable it. */ | |
2199 | ret = -EINVAL; | |
2200 | goto out; | |
2201 | } | |
2202 | ||
2203 | if (p != kp) | |
2204 | kp->flags &= ~KPROBE_FLAG_DISABLED; | |
2205 | ||
2206 | if (!kprobes_all_disarmed && kprobe_disabled(p)) { | |
2207 | p->flags &= ~KPROBE_FLAG_DISABLED; | |
12310e34 JY |
2208 | ret = arm_kprobe(p); |
2209 | if (ret) | |
2210 | p->flags |= KPROBE_FLAG_DISABLED; | |
c0614829 MH |
2211 | } |
2212 | out: | |
2213 | mutex_unlock(&kprobe_mutex); | |
2214 | return ret; | |
2215 | } | |
2216 | EXPORT_SYMBOL_GPL(enable_kprobe); | |
2217 | ||
4458515b | 2218 | /* Caller must NOT call this in usual path. This is only for critical case */ |
820aede0 | 2219 | void dump_kprobe(struct kprobe *kp) |
24851d24 | 2220 | { |
4458515b MH |
2221 | pr_err("Dumping kprobe:\n"); |
2222 | pr_err("Name: %s\nOffset: %x\nAddress: %pS\n", | |
2223 | kp->symbol_name, kp->offset, kp->addr); | |
24851d24 | 2224 | } |
820aede0 | 2225 | NOKPROBE_SYMBOL(dump_kprobe); |
24851d24 | 2226 | |
fb1a59fa MH |
2227 | int kprobe_add_ksym_blacklist(unsigned long entry) |
2228 | { | |
2229 | struct kprobe_blacklist_entry *ent; | |
2230 | unsigned long offset = 0, size = 0; | |
2231 | ||
2232 | if (!kernel_text_address(entry) || | |
2233 | !kallsyms_lookup_size_offset(entry, &size, &offset)) | |
2234 | return -EINVAL; | |
2235 | ||
2236 | ent = kmalloc(sizeof(*ent), GFP_KERNEL); | |
2237 | if (!ent) | |
2238 | return -ENOMEM; | |
2239 | ent->start_addr = entry; | |
2240 | ent->end_addr = entry + size; | |
2241 | INIT_LIST_HEAD(&ent->list); | |
2242 | list_add_tail(&ent->list, &kprobe_blacklist); | |
2243 | ||
2244 | return (int)size; | |
2245 | } | |
2246 | ||
2247 | /* Add all symbols in given area into kprobe blacklist */ | |
2248 | int kprobe_add_area_blacklist(unsigned long start, unsigned long end) | |
2249 | { | |
2250 | unsigned long entry; | |
2251 | int ret = 0; | |
2252 | ||
2253 | for (entry = start; entry < end; entry += ret) { | |
2254 | ret = kprobe_add_ksym_blacklist(entry); | |
2255 | if (ret < 0) | |
2256 | return ret; | |
2257 | if (ret == 0) /* In case of alias symbol */ | |
2258 | ret = 1; | |
2259 | } | |
2260 | return 0; | |
2261 | } | |
2262 | ||
1e6769b0 MH |
2263 | /* Remove all symbols in given area from kprobe blacklist */ |
2264 | static void kprobe_remove_area_blacklist(unsigned long start, unsigned long end) | |
2265 | { | |
2266 | struct kprobe_blacklist_entry *ent, *n; | |
2267 | ||
2268 | list_for_each_entry_safe(ent, n, &kprobe_blacklist, list) { | |
2269 | if (ent->start_addr < start || ent->start_addr >= end) | |
2270 | continue; | |
2271 | list_del(&ent->list); | |
2272 | kfree(ent); | |
2273 | } | |
2274 | } | |
2275 | ||
16db6264 MH |
2276 | static void kprobe_remove_ksym_blacklist(unsigned long entry) |
2277 | { | |
2278 | kprobe_remove_area_blacklist(entry, entry + 1); | |
2279 | } | |
2280 | ||
d002b8bc AH |
2281 | int __weak arch_kprobe_get_kallsym(unsigned int *symnum, unsigned long *value, |
2282 | char *type, char *sym) | |
2283 | { | |
2284 | return -ERANGE; | |
2285 | } | |
2286 | ||
2287 | int kprobe_get_kallsym(unsigned int symnum, unsigned long *value, char *type, | |
2288 | char *sym) | |
2289 | { | |
2290 | #ifdef __ARCH_WANT_KPROBES_INSN_SLOT | |
2291 | if (!kprobe_cache_get_kallsym(&kprobe_insn_slots, &symnum, value, type, sym)) | |
2292 | return 0; | |
2293 | #ifdef CONFIG_OPTPROBES | |
2294 | if (!kprobe_cache_get_kallsym(&kprobe_optinsn_slots, &symnum, value, type, sym)) | |
2295 | return 0; | |
2296 | #endif | |
2297 | #endif | |
2298 | if (!arch_kprobe_get_kallsym(&symnum, value, type, sym)) | |
2299 | return 0; | |
2300 | return -ERANGE; | |
2301 | } | |
2302 | ||
fb1a59fa MH |
2303 | int __init __weak arch_populate_kprobe_blacklist(void) |
2304 | { | |
2305 | return 0; | |
2306 | } | |
2307 | ||
376e2424 MH |
2308 | /* |
2309 | * Lookup and populate the kprobe_blacklist. | |
2310 | * | |
2311 | * Unlike the kretprobe blacklist, we'll need to determine | |
2312 | * the range of addresses that belong to the said functions, | |
2313 | * since a kprobe need not necessarily be at the beginning | |
2314 | * of a function. | |
2315 | */ | |
2316 | static int __init populate_kprobe_blacklist(unsigned long *start, | |
2317 | unsigned long *end) | |
2318 | { | |
fb1a59fa | 2319 | unsigned long entry; |
376e2424 | 2320 | unsigned long *iter; |
fb1a59fa | 2321 | int ret; |
376e2424 MH |
2322 | |
2323 | for (iter = start; iter < end; iter++) { | |
d81b4253 | 2324 | entry = arch_deref_entry_point((void *)*iter); |
fb1a59fa MH |
2325 | ret = kprobe_add_ksym_blacklist(entry); |
2326 | if (ret == -EINVAL) | |
376e2424 | 2327 | continue; |
fb1a59fa MH |
2328 | if (ret < 0) |
2329 | return ret; | |
376e2424 | 2330 | } |
fb1a59fa MH |
2331 | |
2332 | /* Symbols in __kprobes_text are blacklisted */ | |
2333 | ret = kprobe_add_area_blacklist((unsigned long)__kprobes_text_start, | |
2334 | (unsigned long)__kprobes_text_end); | |
66e9b071 TG |
2335 | if (ret) |
2336 | return ret; | |
2337 | ||
2338 | /* Symbols in noinstr section are blacklisted */ | |
2339 | ret = kprobe_add_area_blacklist((unsigned long)__noinstr_text_start, | |
2340 | (unsigned long)__noinstr_text_end); | |
fb1a59fa MH |
2341 | |
2342 | return ret ? : arch_populate_kprobe_blacklist(); | |
376e2424 MH |
2343 | } |
2344 | ||
1e6769b0 MH |
2345 | static void add_module_kprobe_blacklist(struct module *mod) |
2346 | { | |
2347 | unsigned long start, end; | |
16db6264 MH |
2348 | int i; |
2349 | ||
2350 | if (mod->kprobe_blacklist) { | |
2351 | for (i = 0; i < mod->num_kprobe_blacklist; i++) | |
2352 | kprobe_add_ksym_blacklist(mod->kprobe_blacklist[i]); | |
2353 | } | |
1e6769b0 MH |
2354 | |
2355 | start = (unsigned long)mod->kprobes_text_start; | |
2356 | if (start) { | |
2357 | end = start + mod->kprobes_text_size; | |
2358 | kprobe_add_area_blacklist(start, end); | |
2359 | } | |
66e9b071 TG |
2360 | |
2361 | start = (unsigned long)mod->noinstr_text_start; | |
2362 | if (start) { | |
2363 | end = start + mod->noinstr_text_size; | |
2364 | kprobe_add_area_blacklist(start, end); | |
2365 | } | |
1e6769b0 MH |
2366 | } |
2367 | ||
2368 | static void remove_module_kprobe_blacklist(struct module *mod) | |
2369 | { | |
2370 | unsigned long start, end; | |
16db6264 MH |
2371 | int i; |
2372 | ||
2373 | if (mod->kprobe_blacklist) { | |
2374 | for (i = 0; i < mod->num_kprobe_blacklist; i++) | |
2375 | kprobe_remove_ksym_blacklist(mod->kprobe_blacklist[i]); | |
2376 | } | |
1e6769b0 MH |
2377 | |
2378 | start = (unsigned long)mod->kprobes_text_start; | |
2379 | if (start) { | |
2380 | end = start + mod->kprobes_text_size; | |
2381 | kprobe_remove_area_blacklist(start, end); | |
2382 | } | |
66e9b071 TG |
2383 | |
2384 | start = (unsigned long)mod->noinstr_text_start; | |
2385 | if (start) { | |
2386 | end = start + mod->noinstr_text_size; | |
2387 | kprobe_remove_area_blacklist(start, end); | |
2388 | } | |
1e6769b0 MH |
2389 | } |
2390 | ||
e8386a0c | 2391 | /* Module notifier call back, checking kprobes on the module */ |
55479f64 MH |
2392 | static int kprobes_module_callback(struct notifier_block *nb, |
2393 | unsigned long val, void *data) | |
e8386a0c MH |
2394 | { |
2395 | struct module *mod = data; | |
2396 | struct hlist_head *head; | |
e8386a0c MH |
2397 | struct kprobe *p; |
2398 | unsigned int i; | |
f24659d9 | 2399 | int checkcore = (val == MODULE_STATE_GOING); |
e8386a0c | 2400 | |
1e6769b0 MH |
2401 | if (val == MODULE_STATE_COMING) { |
2402 | mutex_lock(&kprobe_mutex); | |
2403 | add_module_kprobe_blacklist(mod); | |
2404 | mutex_unlock(&kprobe_mutex); | |
2405 | } | |
f24659d9 | 2406 | if (val != MODULE_STATE_GOING && val != MODULE_STATE_LIVE) |
e8386a0c MH |
2407 | return NOTIFY_DONE; |
2408 | ||
2409 | /* | |
f24659d9 MH |
2410 | * When MODULE_STATE_GOING was notified, both of module .text and |
2411 | * .init.text sections would be freed. When MODULE_STATE_LIVE was | |
2412 | * notified, only .init.text section would be freed. We need to | |
2413 | * disable kprobes which have been inserted in the sections. | |
e8386a0c MH |
2414 | */ |
2415 | mutex_lock(&kprobe_mutex); | |
2416 | for (i = 0; i < KPROBE_TABLE_SIZE; i++) { | |
2417 | head = &kprobe_table[i]; | |
7e6a71d8 | 2418 | hlist_for_each_entry(p, head, hlist) |
f24659d9 MH |
2419 | if (within_module_init((unsigned long)p->addr, mod) || |
2420 | (checkcore && | |
2421 | within_module_core((unsigned long)p->addr, mod))) { | |
e8386a0c MH |
2422 | /* |
2423 | * The vaddr this probe is installed will soon | |
2424 | * be vfreed buy not synced to disk. Hence, | |
2425 | * disarming the breakpoint isn't needed. | |
545a0281 SRV |
2426 | * |
2427 | * Note, this will also move any optimized probes | |
2428 | * that are pending to be removed from their | |
2429 | * corresponding lists to the freeing_list and | |
2430 | * will not be touched by the delayed | |
2431 | * kprobe_optimizer work handler. | |
e8386a0c MH |
2432 | */ |
2433 | kill_kprobe(p); | |
2434 | } | |
2435 | } | |
1e6769b0 MH |
2436 | if (val == MODULE_STATE_GOING) |
2437 | remove_module_kprobe_blacklist(mod); | |
e8386a0c MH |
2438 | mutex_unlock(&kprobe_mutex); |
2439 | return NOTIFY_DONE; | |
2440 | } | |
2441 | ||
2442 | static struct notifier_block kprobe_module_nb = { | |
2443 | .notifier_call = kprobes_module_callback, | |
2444 | .priority = 0 | |
2445 | }; | |
2446 | ||
376e2424 MH |
2447 | /* Markers of _kprobe_blacklist section */ |
2448 | extern unsigned long __start_kprobe_blacklist[]; | |
2449 | extern unsigned long __stop_kprobe_blacklist[]; | |
2450 | ||
82d083ab MH |
2451 | void kprobe_free_init_mem(void) |
2452 | { | |
2453 | void *start = (void *)(&__init_begin); | |
2454 | void *end = (void *)(&__init_end); | |
2455 | struct hlist_head *head; | |
2456 | struct kprobe *p; | |
2457 | int i; | |
2458 | ||
2459 | mutex_lock(&kprobe_mutex); | |
2460 | ||
2461 | /* Kill all kprobes on initmem */ | |
2462 | for (i = 0; i < KPROBE_TABLE_SIZE; i++) { | |
2463 | head = &kprobe_table[i]; | |
2464 | hlist_for_each_entry(p, head, hlist) { | |
2465 | if (start <= (void *)p->addr && (void *)p->addr < end) | |
2466 | kill_kprobe(p); | |
2467 | } | |
2468 | } | |
2469 | ||
2470 | mutex_unlock(&kprobe_mutex); | |
2471 | } | |
2472 | ||
1da177e4 LT |
2473 | static int __init init_kprobes(void) |
2474 | { | |
2475 | int i, err = 0; | |
2476 | ||
2477 | /* FIXME allocate the probe table, currently defined statically */ | |
2478 | /* initialize all list heads */ | |
d741bf41 | 2479 | for (i = 0; i < KPROBE_TABLE_SIZE; i++) |
1da177e4 LT |
2480 | INIT_HLIST_HEAD(&kprobe_table[i]); |
2481 | ||
376e2424 MH |
2482 | err = populate_kprobe_blacklist(__start_kprobe_blacklist, |
2483 | __stop_kprobe_blacklist); | |
2484 | if (err) { | |
2485 | pr_err("kprobes: failed to populate blacklist: %d\n", err); | |
2486 | pr_err("Please take care of using kprobes.\n"); | |
3d8d996e SD |
2487 | } |
2488 | ||
f438d914 MH |
2489 | if (kretprobe_blacklist_size) { |
2490 | /* lookup the function address from its name */ | |
2491 | for (i = 0; kretprobe_blacklist[i].name != NULL; i++) { | |
49e0b465 | 2492 | kretprobe_blacklist[i].addr = |
290e3070 | 2493 | kprobe_lookup_name(kretprobe_blacklist[i].name, 0); |
f438d914 MH |
2494 | if (!kretprobe_blacklist[i].addr) |
2495 | printk("kretprobe: lookup failed: %s\n", | |
2496 | kretprobe_blacklist[i].name); | |
2497 | } | |
2498 | } | |
2499 | ||
b2be84df MH |
2500 | #if defined(CONFIG_OPTPROBES) |
2501 | #if defined(__ARCH_WANT_KPROBES_INSN_SLOT) | |
afd66255 MH |
2502 | /* Init kprobe_optinsn_slots */ |
2503 | kprobe_optinsn_slots.insn_size = MAX_OPTINSN_SIZE; | |
2504 | #endif | |
b2be84df MH |
2505 | /* By default, kprobes can be optimized */ |
2506 | kprobes_allow_optimization = true; | |
2507 | #endif | |
afd66255 | 2508 | |
e579abeb MH |
2509 | /* By default, kprobes are armed */ |
2510 | kprobes_all_disarmed = false; | |
bf8f6e5b | 2511 | |
6772926b | 2512 | err = arch_init_kprobes(); |
802eae7c RL |
2513 | if (!err) |
2514 | err = register_die_notifier(&kprobe_exceptions_nb); | |
e8386a0c MH |
2515 | if (!err) |
2516 | err = register_module_notifier(&kprobe_module_nb); | |
2517 | ||
ef53d9c5 | 2518 | kprobes_initialized = (err == 0); |
802eae7c | 2519 | |
8c1c9356 AM |
2520 | if (!err) |
2521 | init_test_probes(); | |
1da177e4 LT |
2522 | return err; |
2523 | } | |
36dadef2 | 2524 | early_initcall(init_kprobes); |
1da177e4 | 2525 | |
346fd59b | 2526 | #ifdef CONFIG_DEBUG_FS |
55479f64 | 2527 | static void report_probe(struct seq_file *pi, struct kprobe *p, |
afd66255 | 2528 | const char *sym, int offset, char *modname, struct kprobe *pp) |
346fd59b SD |
2529 | { |
2530 | char *kprobe_type; | |
81365a94 | 2531 | void *addr = p->addr; |
346fd59b SD |
2532 | |
2533 | if (p->pre_handler == pre_handler_kretprobe) | |
2534 | kprobe_type = "r"; | |
346fd59b SD |
2535 | else |
2536 | kprobe_type = "k"; | |
afd66255 | 2537 | |
60f7bb66 | 2538 | if (!kallsyms_show_value(pi->file->f_cred)) |
81365a94 MH |
2539 | addr = NULL; |
2540 | ||
346fd59b | 2541 | if (sym) |
81365a94 MH |
2542 | seq_printf(pi, "%px %s %s+0x%x %s ", |
2543 | addr, kprobe_type, sym, offset, | |
afd66255 | 2544 | (modname ? modname : " ")); |
81365a94 MH |
2545 | else /* try to use %pS */ |
2546 | seq_printf(pi, "%px %s %pS ", | |
2547 | addr, kprobe_type, p->addr); | |
afd66255 MH |
2548 | |
2549 | if (!pp) | |
2550 | pp = p; | |
ae6aa16f | 2551 | seq_printf(pi, "%s%s%s%s\n", |
afd66255 MH |
2552 | (kprobe_gone(p) ? "[GONE]" : ""), |
2553 | ((kprobe_disabled(p) && !kprobe_gone(p)) ? "[DISABLED]" : ""), | |
ae6aa16f MH |
2554 | (kprobe_optimized(pp) ? "[OPTIMIZED]" : ""), |
2555 | (kprobe_ftrace(pp) ? "[FTRACE]" : "")); | |
346fd59b SD |
2556 | } |
2557 | ||
55479f64 | 2558 | static void *kprobe_seq_start(struct seq_file *f, loff_t *pos) |
346fd59b SD |
2559 | { |
2560 | return (*pos < KPROBE_TABLE_SIZE) ? pos : NULL; | |
2561 | } | |
2562 | ||
55479f64 | 2563 | static void *kprobe_seq_next(struct seq_file *f, void *v, loff_t *pos) |
346fd59b SD |
2564 | { |
2565 | (*pos)++; | |
2566 | if (*pos >= KPROBE_TABLE_SIZE) | |
2567 | return NULL; | |
2568 | return pos; | |
2569 | } | |
2570 | ||
55479f64 | 2571 | static void kprobe_seq_stop(struct seq_file *f, void *v) |
346fd59b SD |
2572 | { |
2573 | /* Nothing to do */ | |
2574 | } | |
2575 | ||
55479f64 | 2576 | static int show_kprobe_addr(struct seq_file *pi, void *v) |
346fd59b SD |
2577 | { |
2578 | struct hlist_head *head; | |
346fd59b SD |
2579 | struct kprobe *p, *kp; |
2580 | const char *sym = NULL; | |
2581 | unsigned int i = *(loff_t *) v; | |
ffb45122 | 2582 | unsigned long offset = 0; |
ab767865 | 2583 | char *modname, namebuf[KSYM_NAME_LEN]; |
346fd59b SD |
2584 | |
2585 | head = &kprobe_table[i]; | |
2586 | preempt_disable(); | |
b67bfe0d | 2587 | hlist_for_each_entry_rcu(p, head, hlist) { |
ffb45122 | 2588 | sym = kallsyms_lookup((unsigned long)p->addr, NULL, |
346fd59b | 2589 | &offset, &modname, namebuf); |
afd66255 | 2590 | if (kprobe_aggrprobe(p)) { |
346fd59b | 2591 | list_for_each_entry_rcu(kp, &p->list, list) |
afd66255 | 2592 | report_probe(pi, kp, sym, offset, modname, p); |
346fd59b | 2593 | } else |
afd66255 | 2594 | report_probe(pi, p, sym, offset, modname, NULL); |
346fd59b SD |
2595 | } |
2596 | preempt_enable(); | |
2597 | return 0; | |
2598 | } | |
2599 | ||
eac2cece | 2600 | static const struct seq_operations kprobes_sops = { |
346fd59b SD |
2601 | .start = kprobe_seq_start, |
2602 | .next = kprobe_seq_next, | |
2603 | .stop = kprobe_seq_stop, | |
2604 | .show = show_kprobe_addr | |
2605 | }; | |
2606 | ||
eac2cece | 2607 | DEFINE_SEQ_ATTRIBUTE(kprobes); |
346fd59b | 2608 | |
63724740 MH |
2609 | /* kprobes/blacklist -- shows which functions can not be probed */ |
2610 | static void *kprobe_blacklist_seq_start(struct seq_file *m, loff_t *pos) | |
2611 | { | |
4fdd8887 | 2612 | mutex_lock(&kprobe_mutex); |
63724740 MH |
2613 | return seq_list_start(&kprobe_blacklist, *pos); |
2614 | } | |
2615 | ||
2616 | static void *kprobe_blacklist_seq_next(struct seq_file *m, void *v, loff_t *pos) | |
2617 | { | |
2618 | return seq_list_next(v, &kprobe_blacklist, pos); | |
2619 | } | |
2620 | ||
2621 | static int kprobe_blacklist_seq_show(struct seq_file *m, void *v) | |
2622 | { | |
2623 | struct kprobe_blacklist_entry *ent = | |
2624 | list_entry(v, struct kprobe_blacklist_entry, list); | |
2625 | ||
ffb9bd68 MH |
2626 | /* |
2627 | * If /proc/kallsyms is not showing kernel address, we won't | |
2628 | * show them here either. | |
2629 | */ | |
60f7bb66 | 2630 | if (!kallsyms_show_value(m->file->f_cred)) |
ffb9bd68 MH |
2631 | seq_printf(m, "0x%px-0x%px\t%ps\n", NULL, NULL, |
2632 | (void *)ent->start_addr); | |
2633 | else | |
2634 | seq_printf(m, "0x%px-0x%px\t%ps\n", (void *)ent->start_addr, | |
2635 | (void *)ent->end_addr, (void *)ent->start_addr); | |
63724740 MH |
2636 | return 0; |
2637 | } | |
2638 | ||
4fdd8887 MH |
2639 | static void kprobe_blacklist_seq_stop(struct seq_file *f, void *v) |
2640 | { | |
2641 | mutex_unlock(&kprobe_mutex); | |
2642 | } | |
2643 | ||
eac2cece | 2644 | static const struct seq_operations kprobe_blacklist_sops = { |
63724740 MH |
2645 | .start = kprobe_blacklist_seq_start, |
2646 | .next = kprobe_blacklist_seq_next, | |
4fdd8887 | 2647 | .stop = kprobe_blacklist_seq_stop, |
63724740 MH |
2648 | .show = kprobe_blacklist_seq_show, |
2649 | }; | |
eac2cece | 2650 | DEFINE_SEQ_ATTRIBUTE(kprobe_blacklist); |
63724740 | 2651 | |
12310e34 | 2652 | static int arm_all_kprobes(void) |
bf8f6e5b AM |
2653 | { |
2654 | struct hlist_head *head; | |
bf8f6e5b | 2655 | struct kprobe *p; |
12310e34 JY |
2656 | unsigned int i, total = 0, errors = 0; |
2657 | int err, ret = 0; | |
bf8f6e5b AM |
2658 | |
2659 | mutex_lock(&kprobe_mutex); | |
2660 | ||
e579abeb MH |
2661 | /* If kprobes are armed, just return */ |
2662 | if (!kprobes_all_disarmed) | |
bf8f6e5b AM |
2663 | goto already_enabled; |
2664 | ||
977ad481 WN |
2665 | /* |
2666 | * optimize_kprobe() called by arm_kprobe() checks | |
2667 | * kprobes_all_disarmed, so set kprobes_all_disarmed before | |
2668 | * arm_kprobe. | |
2669 | */ | |
2670 | kprobes_all_disarmed = false; | |
afd66255 | 2671 | /* Arming kprobes doesn't optimize kprobe itself */ |
bf8f6e5b AM |
2672 | for (i = 0; i < KPROBE_TABLE_SIZE; i++) { |
2673 | head = &kprobe_table[i]; | |
12310e34 | 2674 | /* Arm all kprobes on a best-effort basis */ |
7e6a71d8 | 2675 | hlist_for_each_entry(p, head, hlist) { |
12310e34 JY |
2676 | if (!kprobe_disabled(p)) { |
2677 | err = arm_kprobe(p); | |
2678 | if (err) { | |
2679 | errors++; | |
2680 | ret = err; | |
2681 | } | |
2682 | total++; | |
2683 | } | |
2684 | } | |
bf8f6e5b AM |
2685 | } |
2686 | ||
12310e34 JY |
2687 | if (errors) |
2688 | pr_warn("Kprobes globally enabled, but failed to arm %d out of %d probes\n", | |
2689 | errors, total); | |
2690 | else | |
2691 | pr_info("Kprobes globally enabled\n"); | |
bf8f6e5b AM |
2692 | |
2693 | already_enabled: | |
2694 | mutex_unlock(&kprobe_mutex); | |
12310e34 | 2695 | return ret; |
bf8f6e5b AM |
2696 | } |
2697 | ||
297f9233 | 2698 | static int disarm_all_kprobes(void) |
bf8f6e5b AM |
2699 | { |
2700 | struct hlist_head *head; | |
bf8f6e5b | 2701 | struct kprobe *p; |
297f9233 JY |
2702 | unsigned int i, total = 0, errors = 0; |
2703 | int err, ret = 0; | |
bf8f6e5b AM |
2704 | |
2705 | mutex_lock(&kprobe_mutex); | |
2706 | ||
e579abeb | 2707 | /* If kprobes are already disarmed, just return */ |
6274de49 MH |
2708 | if (kprobes_all_disarmed) { |
2709 | mutex_unlock(&kprobe_mutex); | |
297f9233 | 2710 | return 0; |
6274de49 | 2711 | } |
bf8f6e5b | 2712 | |
e579abeb | 2713 | kprobes_all_disarmed = true; |
afd66255 | 2714 | |
bf8f6e5b AM |
2715 | for (i = 0; i < KPROBE_TABLE_SIZE; i++) { |
2716 | head = &kprobe_table[i]; | |
297f9233 | 2717 | /* Disarm all kprobes on a best-effort basis */ |
7e6a71d8 | 2718 | hlist_for_each_entry(p, head, hlist) { |
297f9233 JY |
2719 | if (!arch_trampoline_kprobe(p) && !kprobe_disabled(p)) { |
2720 | err = disarm_kprobe(p, false); | |
2721 | if (err) { | |
2722 | errors++; | |
2723 | ret = err; | |
2724 | } | |
2725 | total++; | |
2726 | } | |
bf8f6e5b AM |
2727 | } |
2728 | } | |
297f9233 JY |
2729 | |
2730 | if (errors) | |
2731 | pr_warn("Kprobes globally disabled, but failed to disarm %d out of %d probes\n", | |
2732 | errors, total); | |
2733 | else | |
2734 | pr_info("Kprobes globally disabled\n"); | |
2735 | ||
bf8f6e5b | 2736 | mutex_unlock(&kprobe_mutex); |
bf8f6e5b | 2737 | |
6274de49 MH |
2738 | /* Wait for disarming all kprobes by optimizer */ |
2739 | wait_for_kprobe_optimizer(); | |
297f9233 JY |
2740 | |
2741 | return ret; | |
bf8f6e5b AM |
2742 | } |
2743 | ||
2744 | /* | |
2745 | * XXX: The debugfs bool file interface doesn't allow for callbacks | |
2746 | * when the bool state is switched. We can reuse that facility when | |
2747 | * available | |
2748 | */ | |
2749 | static ssize_t read_enabled_file_bool(struct file *file, | |
2750 | char __user *user_buf, size_t count, loff_t *ppos) | |
2751 | { | |
2752 | char buf[3]; | |
2753 | ||
e579abeb | 2754 | if (!kprobes_all_disarmed) |
bf8f6e5b AM |
2755 | buf[0] = '1'; |
2756 | else | |
2757 | buf[0] = '0'; | |
2758 | buf[1] = '\n'; | |
2759 | buf[2] = 0x00; | |
2760 | return simple_read_from_buffer(user_buf, count, ppos, buf, 2); | |
2761 | } | |
2762 | ||
2763 | static ssize_t write_enabled_file_bool(struct file *file, | |
2764 | const char __user *user_buf, size_t count, loff_t *ppos) | |
2765 | { | |
2766 | char buf[32]; | |
efeb156e | 2767 | size_t buf_size; |
12310e34 | 2768 | int ret = 0; |
bf8f6e5b AM |
2769 | |
2770 | buf_size = min(count, (sizeof(buf)-1)); | |
2771 | if (copy_from_user(buf, user_buf, buf_size)) | |
2772 | return -EFAULT; | |
2773 | ||
10fb46d5 | 2774 | buf[buf_size] = '\0'; |
bf8f6e5b AM |
2775 | switch (buf[0]) { |
2776 | case 'y': | |
2777 | case 'Y': | |
2778 | case '1': | |
12310e34 | 2779 | ret = arm_all_kprobes(); |
bf8f6e5b AM |
2780 | break; |
2781 | case 'n': | |
2782 | case 'N': | |
2783 | case '0': | |
297f9233 | 2784 | ret = disarm_all_kprobes(); |
bf8f6e5b | 2785 | break; |
10fb46d5 MK |
2786 | default: |
2787 | return -EINVAL; | |
bf8f6e5b AM |
2788 | } |
2789 | ||
12310e34 JY |
2790 | if (ret) |
2791 | return ret; | |
2792 | ||
bf8f6e5b AM |
2793 | return count; |
2794 | } | |
2795 | ||
828c0950 | 2796 | static const struct file_operations fops_kp = { |
bf8f6e5b AM |
2797 | .read = read_enabled_file_bool, |
2798 | .write = write_enabled_file_bool, | |
6038f373 | 2799 | .llseek = default_llseek, |
bf8f6e5b AM |
2800 | }; |
2801 | ||
55479f64 | 2802 | static int __init debugfs_kprobe_init(void) |
346fd59b | 2803 | { |
8c0fd1fa | 2804 | struct dentry *dir; |
bf8f6e5b | 2805 | unsigned int value = 1; |
346fd59b SD |
2806 | |
2807 | dir = debugfs_create_dir("kprobes", NULL); | |
346fd59b | 2808 | |
eac2cece | 2809 | debugfs_create_file("list", 0400, dir, NULL, &kprobes_fops); |
346fd59b | 2810 | |
8c0fd1fa | 2811 | debugfs_create_file("enabled", 0600, dir, &value, &fops_kp); |
63724740 | 2812 | |
8c0fd1fa | 2813 | debugfs_create_file("blacklist", 0400, dir, NULL, |
eac2cece | 2814 | &kprobe_blacklist_fops); |
bf8f6e5b | 2815 | |
346fd59b SD |
2816 | return 0; |
2817 | } | |
2818 | ||
2819 | late_initcall(debugfs_kprobe_init); | |
2820 | #endif /* CONFIG_DEBUG_FS */ |