]>
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> |
fa68bd09 | 38 | #include <linux/static_call.h> |
69e49088 | 39 | #include <linux/perf_event.h> |
bf8f6e5b | 40 | |
bfd45be0 | 41 | #include <asm/sections.h> |
1da177e4 LT |
42 | #include <asm/cacheflush.h> |
43 | #include <asm/errno.h> | |
7c0f6ba6 | 44 | #include <linux/uaccess.h> |
1da177e4 LT |
45 | |
46 | #define KPROBE_HASH_BITS 6 | |
47 | #define KPROBE_TABLE_SIZE (1 << KPROBE_HASH_BITS) | |
48 | ||
3a872d89 | 49 | |
ef53d9c5 | 50 | static int kprobes_initialized; |
7e6a71d8 MH |
51 | /* kprobe_table can be accessed by |
52 | * - Normal hlist traversal and RCU add/del under kprobe_mutex is held. | |
53 | * Or | |
54 | * - RCU hlist traversal under disabling preempt (breakpoint handlers) | |
55 | */ | |
1da177e4 LT |
56 | static struct hlist_head kprobe_table[KPROBE_TABLE_SIZE]; |
57 | ||
bf8f6e5b | 58 | /* NOTE: change this value only with kprobe_mutex held */ |
e579abeb | 59 | static bool kprobes_all_disarmed; |
bf8f6e5b | 60 | |
43948f50 MH |
61 | /* This protects kprobe_table and optimizing_list */ |
62 | static DEFINE_MUTEX(kprobe_mutex); | |
e6584523 | 63 | static DEFINE_PER_CPU(struct kprobe *, kprobe_instance) = NULL; |
ef53d9c5 | 64 | |
290e3070 NR |
65 | kprobe_opcode_t * __weak kprobe_lookup_name(const char *name, |
66 | unsigned int __unused) | |
49e0b465 NR |
67 | { |
68 | return ((kprobe_opcode_t *)(kallsyms_lookup_name(name))); | |
69 | } | |
70 | ||
376e2424 MH |
71 | /* Blacklist -- list of struct kprobe_blacklist_entry */ |
72 | static LIST_HEAD(kprobe_blacklist); | |
3d8d996e | 73 | |
2d14e39d | 74 | #ifdef __ARCH_WANT_KPROBES_INSN_SLOT |
9ec4b1f3 AM |
75 | /* |
76 | * kprobe->ainsn.insn points to the copy of the instruction to be | |
77 | * single-stepped. x86_64, POWER4 and above have no-exec support and | |
78 | * stepping on the instruction on a vmalloced/kmalloced/data page | |
79 | * is a recipe for disaster | |
80 | */ | |
9ec4b1f3 | 81 | struct kprobe_insn_page { |
c5cb5a2d | 82 | struct list_head list; |
9ec4b1f3 | 83 | kprobe_opcode_t *insns; /* Page of instruction slots */ |
af96397d | 84 | struct kprobe_insn_cache *cache; |
9ec4b1f3 | 85 | int nused; |
b4c6c34a | 86 | int ngarbage; |
4610ee1d | 87 | char slot_used[]; |
9ec4b1f3 AM |
88 | }; |
89 | ||
4610ee1d MH |
90 | #define KPROBE_INSN_PAGE_SIZE(slots) \ |
91 | (offsetof(struct kprobe_insn_page, slot_used) + \ | |
92 | (sizeof(char) * (slots))) | |
93 | ||
4610ee1d MH |
94 | static int slots_per_page(struct kprobe_insn_cache *c) |
95 | { | |
96 | return PAGE_SIZE/(c->insn_size * sizeof(kprobe_opcode_t)); | |
97 | } | |
98 | ||
ab40c5c6 MH |
99 | enum kprobe_slot_state { |
100 | SLOT_CLEAN = 0, | |
101 | SLOT_DIRTY = 1, | |
102 | SLOT_USED = 2, | |
103 | }; | |
104 | ||
63fef14f | 105 | void __weak *alloc_insn_page(void) |
af96397d HC |
106 | { |
107 | return module_alloc(PAGE_SIZE); | |
108 | } | |
109 | ||
66ce7514 | 110 | static void free_insn_page(void *page) |
af96397d | 111 | { |
be1f221c | 112 | module_memfree(page); |
af96397d HC |
113 | } |
114 | ||
c802d64a HC |
115 | struct kprobe_insn_cache kprobe_insn_slots = { |
116 | .mutex = __MUTEX_INITIALIZER(kprobe_insn_slots.mutex), | |
af96397d HC |
117 | .alloc = alloc_insn_page, |
118 | .free = free_insn_page, | |
d002b8bc | 119 | .sym = KPROBE_INSN_PAGE_SYM, |
4610ee1d MH |
120 | .pages = LIST_HEAD_INIT(kprobe_insn_slots.pages), |
121 | .insn_size = MAX_INSN_SIZE, | |
122 | .nr_garbage = 0, | |
123 | }; | |
55479f64 | 124 | static int collect_garbage_slots(struct kprobe_insn_cache *c); |
b4c6c34a | 125 | |
9ec4b1f3 | 126 | /** |
12941560 | 127 | * __get_insn_slot() - Find a slot on an executable page for an instruction. |
9ec4b1f3 AM |
128 | * We allocate an executable page if there's no room on existing ones. |
129 | */ | |
55479f64 | 130 | kprobe_opcode_t *__get_insn_slot(struct kprobe_insn_cache *c) |
9ec4b1f3 AM |
131 | { |
132 | struct kprobe_insn_page *kip; | |
c802d64a | 133 | kprobe_opcode_t *slot = NULL; |
9ec4b1f3 | 134 | |
5b485629 | 135 | /* Since the slot array is not protected by rcu, we need a mutex */ |
c802d64a | 136 | mutex_lock(&c->mutex); |
6f716acd | 137 | retry: |
5b485629 MH |
138 | rcu_read_lock(); |
139 | list_for_each_entry_rcu(kip, &c->pages, list) { | |
4610ee1d | 140 | if (kip->nused < slots_per_page(c)) { |
9ec4b1f3 | 141 | int i; |
4610ee1d | 142 | for (i = 0; i < slots_per_page(c); i++) { |
ab40c5c6 MH |
143 | if (kip->slot_used[i] == SLOT_CLEAN) { |
144 | kip->slot_used[i] = SLOT_USED; | |
9ec4b1f3 | 145 | kip->nused++; |
c802d64a | 146 | slot = kip->insns + (i * c->insn_size); |
5b485629 | 147 | rcu_read_unlock(); |
c802d64a | 148 | goto out; |
9ec4b1f3 AM |
149 | } |
150 | } | |
4610ee1d MH |
151 | /* kip->nused is broken. Fix it. */ |
152 | kip->nused = slots_per_page(c); | |
153 | WARN_ON(1); | |
9ec4b1f3 AM |
154 | } |
155 | } | |
5b485629 | 156 | rcu_read_unlock(); |
9ec4b1f3 | 157 | |
b4c6c34a | 158 | /* If there are any garbage slots, collect it and try again. */ |
4610ee1d | 159 | if (c->nr_garbage && collect_garbage_slots(c) == 0) |
b4c6c34a | 160 | goto retry; |
4610ee1d MH |
161 | |
162 | /* All out of space. Need to allocate a new page. */ | |
163 | kip = kmalloc(KPROBE_INSN_PAGE_SIZE(slots_per_page(c)), GFP_KERNEL); | |
6f716acd | 164 | if (!kip) |
c802d64a | 165 | goto out; |
9ec4b1f3 AM |
166 | |
167 | /* | |
168 | * Use module_alloc so this page is within +/- 2GB of where the | |
169 | * kernel image and loaded module images reside. This is required | |
170 | * so x86_64 can correctly handle the %rip-relative fixups. | |
171 | */ | |
af96397d | 172 | kip->insns = c->alloc(); |
9ec4b1f3 AM |
173 | if (!kip->insns) { |
174 | kfree(kip); | |
c802d64a | 175 | goto out; |
9ec4b1f3 | 176 | } |
c5cb5a2d | 177 | INIT_LIST_HEAD(&kip->list); |
4610ee1d | 178 | memset(kip->slot_used, SLOT_CLEAN, slots_per_page(c)); |
ab40c5c6 | 179 | kip->slot_used[0] = SLOT_USED; |
9ec4b1f3 | 180 | kip->nused = 1; |
b4c6c34a | 181 | kip->ngarbage = 0; |
af96397d | 182 | kip->cache = c; |
5b485629 | 183 | list_add_rcu(&kip->list, &c->pages); |
c802d64a | 184 | slot = kip->insns; |
69e49088 AH |
185 | |
186 | /* Record the perf ksymbol register event after adding the page */ | |
187 | perf_event_ksymbol(PERF_RECORD_KSYMBOL_TYPE_OOL, (unsigned long)kip->insns, | |
188 | PAGE_SIZE, false, c->sym); | |
c802d64a HC |
189 | out: |
190 | mutex_unlock(&c->mutex); | |
191 | return slot; | |
12941560 MH |
192 | } |
193 | ||
b4c6c34a | 194 | /* Return 1 if all garbages are collected, otherwise 0. */ |
55479f64 | 195 | static int collect_one_slot(struct kprobe_insn_page *kip, int idx) |
b4c6c34a | 196 | { |
ab40c5c6 | 197 | kip->slot_used[idx] = SLOT_CLEAN; |
b4c6c34a MH |
198 | kip->nused--; |
199 | if (kip->nused == 0) { | |
200 | /* | |
201 | * Page is no longer in use. Free it unless | |
202 | * it's the last one. We keep the last one | |
203 | * so as not to have to set it up again the | |
204 | * next time somebody inserts a probe. | |
205 | */ | |
4610ee1d | 206 | if (!list_is_singular(&kip->list)) { |
69e49088 AH |
207 | /* |
208 | * Record perf ksymbol unregister event before removing | |
209 | * the page. | |
210 | */ | |
211 | perf_event_ksymbol(PERF_RECORD_KSYMBOL_TYPE_OOL, | |
212 | (unsigned long)kip->insns, PAGE_SIZE, true, | |
213 | kip->cache->sym); | |
5b485629 MH |
214 | list_del_rcu(&kip->list); |
215 | synchronize_rcu(); | |
af96397d | 216 | kip->cache->free(kip->insns); |
b4c6c34a MH |
217 | kfree(kip); |
218 | } | |
219 | return 1; | |
220 | } | |
221 | return 0; | |
222 | } | |
223 | ||
55479f64 | 224 | static int collect_garbage_slots(struct kprobe_insn_cache *c) |
b4c6c34a | 225 | { |
c5cb5a2d | 226 | struct kprobe_insn_page *kip, *next; |
b4c6c34a | 227 | |
615d0ebb | 228 | /* Ensure no-one is interrupted on the garbages */ |
ae8b7ce7 | 229 | synchronize_rcu(); |
b4c6c34a | 230 | |
4610ee1d | 231 | list_for_each_entry_safe(kip, next, &c->pages, list) { |
b4c6c34a | 232 | int i; |
b4c6c34a MH |
233 | if (kip->ngarbage == 0) |
234 | continue; | |
235 | kip->ngarbage = 0; /* we will collect all garbages */ | |
4610ee1d | 236 | for (i = 0; i < slots_per_page(c); i++) { |
5b485629 | 237 | if (kip->slot_used[i] == SLOT_DIRTY && collect_one_slot(kip, i)) |
b4c6c34a MH |
238 | break; |
239 | } | |
240 | } | |
4610ee1d | 241 | c->nr_garbage = 0; |
b4c6c34a MH |
242 | return 0; |
243 | } | |
244 | ||
55479f64 MH |
245 | void __free_insn_slot(struct kprobe_insn_cache *c, |
246 | kprobe_opcode_t *slot, int dirty) | |
9ec4b1f3 AM |
247 | { |
248 | struct kprobe_insn_page *kip; | |
5b485629 | 249 | long idx; |
9ec4b1f3 | 250 | |
c802d64a | 251 | mutex_lock(&c->mutex); |
5b485629 MH |
252 | rcu_read_lock(); |
253 | list_for_each_entry_rcu(kip, &c->pages, list) { | |
254 | idx = ((long)slot - (long)kip->insns) / | |
255 | (c->insn_size * sizeof(kprobe_opcode_t)); | |
256 | if (idx >= 0 && idx < slots_per_page(c)) | |
c802d64a | 257 | goto out; |
9ec4b1f3 | 258 | } |
5b485629 | 259 | /* Could not find this slot. */ |
4610ee1d | 260 | WARN_ON(1); |
5b485629 | 261 | kip = NULL; |
c802d64a | 262 | out: |
5b485629 MH |
263 | rcu_read_unlock(); |
264 | /* Mark and sweep: this may sleep */ | |
265 | if (kip) { | |
266 | /* Check double free */ | |
267 | WARN_ON(kip->slot_used[idx] != SLOT_USED); | |
268 | if (dirty) { | |
269 | kip->slot_used[idx] = SLOT_DIRTY; | |
270 | kip->ngarbage++; | |
271 | if (++c->nr_garbage > slots_per_page(c)) | |
272 | collect_garbage_slots(c); | |
273 | } else { | |
274 | collect_one_slot(kip, idx); | |
275 | } | |
276 | } | |
c802d64a | 277 | mutex_unlock(&c->mutex); |
4610ee1d | 278 | } |
6f716acd | 279 | |
5b485629 MH |
280 | /* |
281 | * Check given address is on the page of kprobe instruction slots. | |
282 | * This will be used for checking whether the address on a stack | |
283 | * is on a text area or not. | |
284 | */ | |
285 | bool __is_insn_slot_addr(struct kprobe_insn_cache *c, unsigned long addr) | |
286 | { | |
287 | struct kprobe_insn_page *kip; | |
288 | bool ret = false; | |
289 | ||
290 | rcu_read_lock(); | |
291 | list_for_each_entry_rcu(kip, &c->pages, list) { | |
292 | if (addr >= (unsigned long)kip->insns && | |
293 | addr < (unsigned long)kip->insns + PAGE_SIZE) { | |
294 | ret = true; | |
295 | break; | |
296 | } | |
297 | } | |
298 | rcu_read_unlock(); | |
299 | ||
300 | return ret; | |
301 | } | |
302 | ||
d002b8bc AH |
303 | int kprobe_cache_get_kallsym(struct kprobe_insn_cache *c, unsigned int *symnum, |
304 | unsigned long *value, char *type, char *sym) | |
305 | { | |
306 | struct kprobe_insn_page *kip; | |
307 | int ret = -ERANGE; | |
308 | ||
309 | rcu_read_lock(); | |
310 | list_for_each_entry_rcu(kip, &c->pages, list) { | |
311 | if ((*symnum)--) | |
312 | continue; | |
313 | strlcpy(sym, c->sym, KSYM_NAME_LEN); | |
314 | *type = 't'; | |
315 | *value = (unsigned long)kip->insns; | |
316 | ret = 0; | |
317 | break; | |
318 | } | |
319 | rcu_read_unlock(); | |
320 | ||
321 | return ret; | |
322 | } | |
323 | ||
afd66255 | 324 | #ifdef CONFIG_OPTPROBES |
7ee3e97e CL |
325 | void __weak *alloc_optinsn_page(void) |
326 | { | |
327 | return alloc_insn_page(); | |
328 | } | |
329 | ||
330 | void __weak free_optinsn_page(void *page) | |
331 | { | |
332 | free_insn_page(page); | |
333 | } | |
334 | ||
afd66255 | 335 | /* For optimized_kprobe buffer */ |
c802d64a HC |
336 | struct kprobe_insn_cache kprobe_optinsn_slots = { |
337 | .mutex = __MUTEX_INITIALIZER(kprobe_optinsn_slots.mutex), | |
7ee3e97e CL |
338 | .alloc = alloc_optinsn_page, |
339 | .free = free_optinsn_page, | |
d002b8bc | 340 | .sym = KPROBE_OPTINSN_PAGE_SYM, |
afd66255 MH |
341 | .pages = LIST_HEAD_INIT(kprobe_optinsn_slots.pages), |
342 | /* .insn_size is initialized later */ | |
343 | .nr_garbage = 0, | |
344 | }; | |
afd66255 | 345 | #endif |
2d14e39d | 346 | #endif |
9ec4b1f3 | 347 | |
e6584523 AM |
348 | /* We have preemption disabled.. so it is safe to use __ versions */ |
349 | static inline void set_kprobe_instance(struct kprobe *kp) | |
350 | { | |
b76834bc | 351 | __this_cpu_write(kprobe_instance, kp); |
e6584523 AM |
352 | } |
353 | ||
354 | static inline void reset_kprobe_instance(void) | |
355 | { | |
b76834bc | 356 | __this_cpu_write(kprobe_instance, NULL); |
e6584523 AM |
357 | } |
358 | ||
3516a460 AM |
359 | /* |
360 | * This routine is called either: | |
49a2a1b8 | 361 | * - under the kprobe_mutex - during kprobe_[un]register() |
3516a460 | 362 | * OR |
d217d545 | 363 | * - with preemption disabled - from arch/xxx/kernel/kprobes.c |
3516a460 | 364 | */ |
820aede0 | 365 | struct kprobe *get_kprobe(void *addr) |
1da177e4 LT |
366 | { |
367 | struct hlist_head *head; | |
3516a460 | 368 | struct kprobe *p; |
1da177e4 LT |
369 | |
370 | head = &kprobe_table[hash_ptr(addr, KPROBE_HASH_BITS)]; | |
6743ad43 MH |
371 | hlist_for_each_entry_rcu(p, head, hlist, |
372 | lockdep_is_held(&kprobe_mutex)) { | |
1da177e4 LT |
373 | if (p->addr == addr) |
374 | return p; | |
375 | } | |
afd66255 | 376 | |
1da177e4 LT |
377 | return NULL; |
378 | } | |
820aede0 | 379 | NOKPROBE_SYMBOL(get_kprobe); |
1da177e4 | 380 | |
820aede0 | 381 | static int aggr_pre_handler(struct kprobe *p, struct pt_regs *regs); |
afd66255 MH |
382 | |
383 | /* Return true if the kprobe is an aggregator */ | |
384 | static inline int kprobe_aggrprobe(struct kprobe *p) | |
385 | { | |
386 | return p->pre_handler == aggr_pre_handler; | |
387 | } | |
388 | ||
6274de49 MH |
389 | /* Return true(!0) if the kprobe is unused */ |
390 | static inline int kprobe_unused(struct kprobe *p) | |
391 | { | |
392 | return kprobe_aggrprobe(p) && kprobe_disabled(p) && | |
393 | list_empty(&p->list); | |
394 | } | |
395 | ||
afd66255 MH |
396 | /* |
397 | * Keep all fields in the kprobe consistent | |
398 | */ | |
6d8e40a8 | 399 | static inline void copy_kprobe(struct kprobe *ap, struct kprobe *p) |
afd66255 | 400 | { |
6d8e40a8 MH |
401 | memcpy(&p->opcode, &ap->opcode, sizeof(kprobe_opcode_t)); |
402 | memcpy(&p->ainsn, &ap->ainsn, sizeof(struct arch_specific_insn)); | |
afd66255 MH |
403 | } |
404 | ||
405 | #ifdef CONFIG_OPTPROBES | |
b2be84df MH |
406 | /* NOTE: change this value only with kprobe_mutex held */ |
407 | static bool kprobes_allow_optimization; | |
408 | ||
afd66255 MH |
409 | /* |
410 | * Call all pre_handler on the list, but ignores its return value. | |
411 | * This must be called from arch-dep optimized caller. | |
412 | */ | |
820aede0 | 413 | void opt_pre_handler(struct kprobe *p, struct pt_regs *regs) |
afd66255 MH |
414 | { |
415 | struct kprobe *kp; | |
416 | ||
417 | list_for_each_entry_rcu(kp, &p->list, list) { | |
418 | if (kp->pre_handler && likely(!kprobe_disabled(kp))) { | |
419 | set_kprobe_instance(kp); | |
4f3a8714 | 420 | kp->pre_handler(kp, regs); |
afd66255 MH |
421 | } |
422 | reset_kprobe_instance(); | |
423 | } | |
424 | } | |
820aede0 | 425 | NOKPROBE_SYMBOL(opt_pre_handler); |
afd66255 | 426 | |
6274de49 | 427 | /* Free optimized instructions and optimized_kprobe */ |
55479f64 | 428 | static void free_aggr_kprobe(struct kprobe *p) |
6274de49 MH |
429 | { |
430 | struct optimized_kprobe *op; | |
431 | ||
432 | op = container_of(p, struct optimized_kprobe, kp); | |
433 | arch_remove_optimized_kprobe(op); | |
434 | arch_remove_kprobe(p); | |
435 | kfree(op); | |
436 | } | |
437 | ||
afd66255 MH |
438 | /* Return true(!0) if the kprobe is ready for optimization. */ |
439 | static inline int kprobe_optready(struct kprobe *p) | |
440 | { | |
441 | struct optimized_kprobe *op; | |
442 | ||
443 | if (kprobe_aggrprobe(p)) { | |
444 | op = container_of(p, struct optimized_kprobe, kp); | |
445 | return arch_prepared_optinsn(&op->optinsn); | |
446 | } | |
447 | ||
448 | return 0; | |
449 | } | |
450 | ||
6274de49 MH |
451 | /* Return true(!0) if the kprobe is disarmed. Note: p must be on hash list */ |
452 | static inline int kprobe_disarmed(struct kprobe *p) | |
453 | { | |
454 | struct optimized_kprobe *op; | |
455 | ||
456 | /* If kprobe is not aggr/opt probe, just return kprobe is disabled */ | |
457 | if (!kprobe_aggrprobe(p)) | |
458 | return kprobe_disabled(p); | |
459 | ||
460 | op = container_of(p, struct optimized_kprobe, kp); | |
461 | ||
462 | return kprobe_disabled(p) && list_empty(&op->list); | |
463 | } | |
464 | ||
465 | /* Return true(!0) if the probe is queued on (un)optimizing lists */ | |
55479f64 | 466 | static int kprobe_queued(struct kprobe *p) |
6274de49 MH |
467 | { |
468 | struct optimized_kprobe *op; | |
469 | ||
470 | if (kprobe_aggrprobe(p)) { | |
471 | op = container_of(p, struct optimized_kprobe, kp); | |
472 | if (!list_empty(&op->list)) | |
473 | return 1; | |
474 | } | |
475 | return 0; | |
476 | } | |
477 | ||
afd66255 MH |
478 | /* |
479 | * Return an optimized kprobe whose optimizing code replaces | |
480 | * instructions including addr (exclude breakpoint). | |
481 | */ | |
55479f64 | 482 | static struct kprobe *get_optimized_kprobe(unsigned long addr) |
afd66255 MH |
483 | { |
484 | int i; | |
485 | struct kprobe *p = NULL; | |
486 | struct optimized_kprobe *op; | |
487 | ||
488 | /* Don't check i == 0, since that is a breakpoint case. */ | |
489 | for (i = 1; !p && i < MAX_OPTIMIZED_LENGTH; i++) | |
490 | p = get_kprobe((void *)(addr - i)); | |
491 | ||
492 | if (p && kprobe_optready(p)) { | |
493 | op = container_of(p, struct optimized_kprobe, kp); | |
494 | if (arch_within_optimized_kprobe(op, addr)) | |
495 | return p; | |
496 | } | |
497 | ||
498 | return NULL; | |
499 | } | |
500 | ||
501 | /* Optimization staging list, protected by kprobe_mutex */ | |
502 | static LIST_HEAD(optimizing_list); | |
6274de49 | 503 | static LIST_HEAD(unoptimizing_list); |
7b959fc5 | 504 | static LIST_HEAD(freeing_list); |
afd66255 MH |
505 | |
506 | static void kprobe_optimizer(struct work_struct *work); | |
507 | static DECLARE_DELAYED_WORK(optimizing_work, kprobe_optimizer); | |
508 | #define OPTIMIZE_DELAY 5 | |
509 | ||
61f4e13f MH |
510 | /* |
511 | * Optimize (replace a breakpoint with a jump) kprobes listed on | |
512 | * optimizing_list. | |
513 | */ | |
55479f64 | 514 | static void do_optimize_kprobes(void) |
afd66255 | 515 | { |
f1c6ece2 | 516 | lockdep_assert_held(&text_mutex); |
afd66255 MH |
517 | /* |
518 | * The optimization/unoptimization refers online_cpus via | |
519 | * stop_machine() and cpu-hotplug modifies online_cpus. | |
520 | * And same time, text_mutex will be held in cpu-hotplug and here. | |
521 | * This combination can cause a deadlock (cpu-hotplug try to lock | |
522 | * text_mutex but stop_machine can not be done because online_cpus | |
523 | * has been changed) | |
2d1e38f5 | 524 | * To avoid this deadlock, caller must have locked cpu hotplug |
afd66255 MH |
525 | * for preventing cpu-hotplug outside of text_mutex locking. |
526 | */ | |
2d1e38f5 TG |
527 | lockdep_assert_cpus_held(); |
528 | ||
529 | /* Optimization never be done when disarmed */ | |
530 | if (kprobes_all_disarmed || !kprobes_allow_optimization || | |
531 | list_empty(&optimizing_list)) | |
532 | return; | |
533 | ||
cd7ebe22 | 534 | arch_optimize_kprobes(&optimizing_list); |
61f4e13f MH |
535 | } |
536 | ||
6274de49 MH |
537 | /* |
538 | * Unoptimize (replace a jump with a breakpoint and remove the breakpoint | |
539 | * if need) kprobes listed on unoptimizing_list. | |
540 | */ | |
55479f64 | 541 | static void do_unoptimize_kprobes(void) |
6274de49 MH |
542 | { |
543 | struct optimized_kprobe *op, *tmp; | |
544 | ||
f1c6ece2 | 545 | lockdep_assert_held(&text_mutex); |
2d1e38f5 TG |
546 | /* See comment in do_optimize_kprobes() */ |
547 | lockdep_assert_cpus_held(); | |
548 | ||
6274de49 MH |
549 | /* Unoptimization must be done anytime */ |
550 | if (list_empty(&unoptimizing_list)) | |
551 | return; | |
552 | ||
7b959fc5 | 553 | arch_unoptimize_kprobes(&unoptimizing_list, &freeing_list); |
f984ba4e | 554 | /* Loop free_list for disarming */ |
7b959fc5 | 555 | list_for_each_entry_safe(op, tmp, &freeing_list, list) { |
f66c0447 MH |
556 | /* Switching from detour code to origin */ |
557 | op->kp.flags &= ~KPROBE_FLAG_OPTIMIZED; | |
6274de49 MH |
558 | /* Disarm probes if marked disabled */ |
559 | if (kprobe_disabled(&op->kp)) | |
560 | arch_disarm_kprobe(&op->kp); | |
561 | if (kprobe_unused(&op->kp)) { | |
562 | /* | |
563 | * Remove unused probes from hash list. After waiting | |
564 | * for synchronization, these probes are reclaimed. | |
565 | * (reclaiming is done by do_free_cleaned_kprobes.) | |
566 | */ | |
567 | hlist_del_rcu(&op->kp.hlist); | |
6274de49 MH |
568 | } else |
569 | list_del_init(&op->list); | |
570 | } | |
6274de49 MH |
571 | } |
572 | ||
573 | /* Reclaim all kprobes on the free_list */ | |
55479f64 | 574 | static void do_free_cleaned_kprobes(void) |
6274de49 MH |
575 | { |
576 | struct optimized_kprobe *op, *tmp; | |
577 | ||
7b959fc5 | 578 | list_for_each_entry_safe(op, tmp, &freeing_list, list) { |
6274de49 | 579 | list_del_init(&op->list); |
cbdd96f5 MH |
580 | if (WARN_ON_ONCE(!kprobe_unused(&op->kp))) { |
581 | /* | |
582 | * This must not happen, but if there is a kprobe | |
583 | * still in use, keep it on kprobes hash list. | |
584 | */ | |
585 | continue; | |
586 | } | |
6274de49 MH |
587 | free_aggr_kprobe(&op->kp); |
588 | } | |
589 | } | |
590 | ||
591 | /* Start optimizer after OPTIMIZE_DELAY passed */ | |
55479f64 | 592 | static void kick_kprobe_optimizer(void) |
6274de49 | 593 | { |
ad72b3be | 594 | schedule_delayed_work(&optimizing_work, OPTIMIZE_DELAY); |
6274de49 MH |
595 | } |
596 | ||
61f4e13f | 597 | /* Kprobe jump optimizer */ |
55479f64 | 598 | static void kprobe_optimizer(struct work_struct *work) |
61f4e13f | 599 | { |
72ef3794 | 600 | mutex_lock(&kprobe_mutex); |
2d1e38f5 | 601 | cpus_read_lock(); |
f1c6ece2 | 602 | mutex_lock(&text_mutex); |
61f4e13f MH |
603 | |
604 | /* | |
6274de49 MH |
605 | * Step 1: Unoptimize kprobes and collect cleaned (unused and disarmed) |
606 | * kprobes before waiting for quiesence period. | |
607 | */ | |
7b959fc5 | 608 | do_unoptimize_kprobes(); |
6274de49 MH |
609 | |
610 | /* | |
a30b85df MH |
611 | * Step 2: Wait for quiesence period to ensure all potentially |
612 | * preempted tasks to have normally scheduled. Because optprobe | |
613 | * may modify multiple instructions, there is a chance that Nth | |
614 | * instruction is preempted. In that case, such tasks can return | |
615 | * to 2nd-Nth byte of jump instruction. This wait is for avoiding it. | |
616 | * Note that on non-preemptive kernel, this is transparently converted | |
617 | * to synchronoze_sched() to wait for all interrupts to have completed. | |
61f4e13f | 618 | */ |
a30b85df | 619 | synchronize_rcu_tasks(); |
61f4e13f | 620 | |
6274de49 | 621 | /* Step 3: Optimize kprobes after quiesence period */ |
61f4e13f | 622 | do_optimize_kprobes(); |
6274de49 MH |
623 | |
624 | /* Step 4: Free cleaned kprobes after quiesence period */ | |
7b959fc5 | 625 | do_free_cleaned_kprobes(); |
6274de49 | 626 | |
f1c6ece2 | 627 | mutex_unlock(&text_mutex); |
2d1e38f5 | 628 | cpus_read_unlock(); |
6274de49 | 629 | |
cd7ebe22 | 630 | /* Step 5: Kick optimizer again if needed */ |
f984ba4e | 631 | if (!list_empty(&optimizing_list) || !list_empty(&unoptimizing_list)) |
cd7ebe22 | 632 | kick_kprobe_optimizer(); |
1a0aa991 MH |
633 | |
634 | mutex_unlock(&kprobe_mutex); | |
6274de49 MH |
635 | } |
636 | ||
637 | /* Wait for completing optimization and unoptimization */ | |
30e7d894 | 638 | void wait_for_kprobe_optimizer(void) |
6274de49 | 639 | { |
ad72b3be TH |
640 | mutex_lock(&kprobe_mutex); |
641 | ||
642 | while (!list_empty(&optimizing_list) || !list_empty(&unoptimizing_list)) { | |
643 | mutex_unlock(&kprobe_mutex); | |
644 | ||
645 | /* this will also make optimizing_work execute immmediately */ | |
646 | flush_delayed_work(&optimizing_work); | |
647 | /* @optimizing_work might not have been queued yet, relax */ | |
648 | cpu_relax(); | |
649 | ||
650 | mutex_lock(&kprobe_mutex); | |
651 | } | |
652 | ||
653 | mutex_unlock(&kprobe_mutex); | |
afd66255 MH |
654 | } |
655 | ||
e4add247 MH |
656 | static bool optprobe_queued_unopt(struct optimized_kprobe *op) |
657 | { | |
658 | struct optimized_kprobe *_op; | |
659 | ||
660 | list_for_each_entry(_op, &unoptimizing_list, list) { | |
661 | if (op == _op) | |
662 | return true; | |
663 | } | |
664 | ||
665 | return false; | |
666 | } | |
667 | ||
afd66255 | 668 | /* Optimize kprobe if p is ready to be optimized */ |
55479f64 | 669 | static void optimize_kprobe(struct kprobe *p) |
afd66255 MH |
670 | { |
671 | struct optimized_kprobe *op; | |
672 | ||
673 | /* Check if the kprobe is disabled or not ready for optimization. */ | |
b2be84df | 674 | if (!kprobe_optready(p) || !kprobes_allow_optimization || |
afd66255 MH |
675 | (kprobe_disabled(p) || kprobes_all_disarmed)) |
676 | return; | |
677 | ||
059053a2 MH |
678 | /* kprobes with post_handler can not be optimized */ |
679 | if (p->post_handler) | |
afd66255 MH |
680 | return; |
681 | ||
682 | op = container_of(p, struct optimized_kprobe, kp); | |
683 | ||
684 | /* Check there is no other kprobes at the optimized instructions */ | |
685 | if (arch_check_optimized_kprobe(op) < 0) | |
686 | return; | |
687 | ||
688 | /* Check if it is already optimized. */ | |
e4add247 MH |
689 | if (op->kp.flags & KPROBE_FLAG_OPTIMIZED) { |
690 | if (optprobe_queued_unopt(op)) { | |
691 | /* This is under unoptimizing. Just dequeue the probe */ | |
692 | list_del_init(&op->list); | |
693 | } | |
afd66255 | 694 | return; |
e4add247 | 695 | } |
afd66255 | 696 | op->kp.flags |= KPROBE_FLAG_OPTIMIZED; |
6274de49 | 697 | |
e4add247 MH |
698 | /* On unoptimizing/optimizing_list, op must have OPTIMIZED flag */ |
699 | if (WARN_ON_ONCE(!list_empty(&op->list))) | |
700 | return; | |
701 | ||
702 | list_add(&op->list, &optimizing_list); | |
703 | kick_kprobe_optimizer(); | |
6274de49 MH |
704 | } |
705 | ||
706 | /* Short cut to direct unoptimizing */ | |
55479f64 | 707 | static void force_unoptimize_kprobe(struct optimized_kprobe *op) |
6274de49 | 708 | { |
2d1e38f5 | 709 | lockdep_assert_cpus_held(); |
6274de49 | 710 | arch_unoptimize_kprobe(op); |
f66c0447 | 711 | op->kp.flags &= ~KPROBE_FLAG_OPTIMIZED; |
afd66255 MH |
712 | } |
713 | ||
714 | /* Unoptimize a kprobe if p is optimized */ | |
55479f64 | 715 | static void unoptimize_kprobe(struct kprobe *p, bool force) |
afd66255 MH |
716 | { |
717 | struct optimized_kprobe *op; | |
718 | ||
6274de49 MH |
719 | if (!kprobe_aggrprobe(p) || kprobe_disarmed(p)) |
720 | return; /* This is not an optprobe nor optimized */ | |
721 | ||
722 | op = container_of(p, struct optimized_kprobe, kp); | |
e4add247 | 723 | if (!kprobe_optimized(p)) |
6274de49 | 724 | return; |
6274de49 | 725 | |
6274de49 | 726 | if (!list_empty(&op->list)) { |
e4add247 MH |
727 | if (optprobe_queued_unopt(op)) { |
728 | /* Queued in unoptimizing queue */ | |
729 | if (force) { | |
730 | /* | |
731 | * Forcibly unoptimize the kprobe here, and queue it | |
732 | * in the freeing list for release afterwards. | |
733 | */ | |
734 | force_unoptimize_kprobe(op); | |
735 | list_move(&op->list, &freeing_list); | |
736 | } | |
737 | } else { | |
738 | /* Dequeue from the optimizing queue */ | |
739 | list_del_init(&op->list); | |
740 | op->kp.flags &= ~KPROBE_FLAG_OPTIMIZED; | |
741 | } | |
6274de49 MH |
742 | return; |
743 | } | |
e4add247 | 744 | |
6274de49 | 745 | /* Optimized kprobe case */ |
e4add247 | 746 | if (force) { |
6274de49 MH |
747 | /* Forcibly update the code: this is a special case */ |
748 | force_unoptimize_kprobe(op); | |
e4add247 | 749 | } else { |
6274de49 MH |
750 | list_add(&op->list, &unoptimizing_list); |
751 | kick_kprobe_optimizer(); | |
afd66255 MH |
752 | } |
753 | } | |
754 | ||
0490cd1f | 755 | /* Cancel unoptimizing for reusing */ |
819319fc | 756 | static int reuse_unused_kprobe(struct kprobe *ap) |
0490cd1f MH |
757 | { |
758 | struct optimized_kprobe *op; | |
759 | ||
0490cd1f MH |
760 | /* |
761 | * Unused kprobe MUST be on the way of delayed unoptimizing (means | |
762 | * there is still a relative jump) and disabled. | |
763 | */ | |
764 | op = container_of(ap, struct optimized_kprobe, kp); | |
4458515b | 765 | WARN_ON_ONCE(list_empty(&op->list)); |
0490cd1f MH |
766 | /* Enable the probe again */ |
767 | ap->flags &= ~KPROBE_FLAG_DISABLED; | |
768 | /* Optimize it again (remove from op->list) */ | |
5f843ed4 MH |
769 | if (!kprobe_optready(ap)) |
770 | return -EINVAL; | |
819319fc | 771 | |
0490cd1f | 772 | optimize_kprobe(ap); |
819319fc | 773 | return 0; |
0490cd1f MH |
774 | } |
775 | ||
afd66255 | 776 | /* Remove optimized instructions */ |
55479f64 | 777 | static void kill_optimized_kprobe(struct kprobe *p) |
afd66255 MH |
778 | { |
779 | struct optimized_kprobe *op; | |
780 | ||
781 | op = container_of(p, struct optimized_kprobe, kp); | |
6274de49 MH |
782 | if (!list_empty(&op->list)) |
783 | /* Dequeue from the (un)optimization queue */ | |
afd66255 | 784 | list_del_init(&op->list); |
6274de49 | 785 | op->kp.flags &= ~KPROBE_FLAG_OPTIMIZED; |
7b959fc5 MH |
786 | |
787 | if (kprobe_unused(p)) { | |
788 | /* Enqueue if it is unused */ | |
789 | list_add(&op->list, &freeing_list); | |
790 | /* | |
791 | * Remove unused probes from the hash list. After waiting | |
792 | * for synchronization, this probe is reclaimed. | |
793 | * (reclaiming is done by do_free_cleaned_kprobes().) | |
794 | */ | |
795 | hlist_del_rcu(&op->kp.hlist); | |
796 | } | |
797 | ||
6274de49 | 798 | /* Don't touch the code, because it is already freed. */ |
afd66255 MH |
799 | arch_remove_optimized_kprobe(op); |
800 | } | |
801 | ||
a460246c MH |
802 | static inline |
803 | void __prepare_optimized_kprobe(struct optimized_kprobe *op, struct kprobe *p) | |
804 | { | |
805 | if (!kprobe_ftrace(p)) | |
806 | arch_prepare_optimized_kprobe(op, p); | |
807 | } | |
808 | ||
afd66255 | 809 | /* Try to prepare optimized instructions */ |
55479f64 | 810 | static void prepare_optimized_kprobe(struct kprobe *p) |
afd66255 MH |
811 | { |
812 | struct optimized_kprobe *op; | |
813 | ||
814 | op = container_of(p, struct optimized_kprobe, kp); | |
a460246c | 815 | __prepare_optimized_kprobe(op, p); |
afd66255 MH |
816 | } |
817 | ||
afd66255 | 818 | /* Allocate new optimized_kprobe and try to prepare optimized instructions */ |
55479f64 | 819 | static struct kprobe *alloc_aggr_kprobe(struct kprobe *p) |
afd66255 MH |
820 | { |
821 | struct optimized_kprobe *op; | |
822 | ||
823 | op = kzalloc(sizeof(struct optimized_kprobe), GFP_KERNEL); | |
824 | if (!op) | |
825 | return NULL; | |
826 | ||
827 | INIT_LIST_HEAD(&op->list); | |
828 | op->kp.addr = p->addr; | |
a460246c | 829 | __prepare_optimized_kprobe(op, p); |
afd66255 MH |
830 | |
831 | return &op->kp; | |
832 | } | |
833 | ||
55479f64 | 834 | static void init_aggr_kprobe(struct kprobe *ap, struct kprobe *p); |
afd66255 MH |
835 | |
836 | /* | |
837 | * Prepare an optimized_kprobe and optimize it | |
838 | * NOTE: p must be a normal registered kprobe | |
839 | */ | |
55479f64 | 840 | static void try_to_optimize_kprobe(struct kprobe *p) |
afd66255 MH |
841 | { |
842 | struct kprobe *ap; | |
843 | struct optimized_kprobe *op; | |
844 | ||
ae6aa16f MH |
845 | /* Impossible to optimize ftrace-based kprobe */ |
846 | if (kprobe_ftrace(p)) | |
847 | return; | |
848 | ||
25764288 | 849 | /* For preparing optimization, jump_label_text_reserved() is called */ |
2d1e38f5 | 850 | cpus_read_lock(); |
25764288 MH |
851 | jump_label_lock(); |
852 | mutex_lock(&text_mutex); | |
853 | ||
afd66255 MH |
854 | ap = alloc_aggr_kprobe(p); |
855 | if (!ap) | |
25764288 | 856 | goto out; |
afd66255 MH |
857 | |
858 | op = container_of(ap, struct optimized_kprobe, kp); | |
859 | if (!arch_prepared_optinsn(&op->optinsn)) { | |
860 | /* If failed to setup optimizing, fallback to kprobe */ | |
6274de49 MH |
861 | arch_remove_optimized_kprobe(op); |
862 | kfree(op); | |
25764288 | 863 | goto out; |
afd66255 MH |
864 | } |
865 | ||
866 | init_aggr_kprobe(ap, p); | |
25764288 MH |
867 | optimize_kprobe(ap); /* This just kicks optimizer thread */ |
868 | ||
869 | out: | |
870 | mutex_unlock(&text_mutex); | |
871 | jump_label_unlock(); | |
2d1e38f5 | 872 | cpus_read_unlock(); |
afd66255 MH |
873 | } |
874 | ||
55479f64 | 875 | static void optimize_all_kprobes(void) |
b2be84df MH |
876 | { |
877 | struct hlist_head *head; | |
b2be84df MH |
878 | struct kprobe *p; |
879 | unsigned int i; | |
880 | ||
5c51543b | 881 | mutex_lock(&kprobe_mutex); |
b2be84df MH |
882 | /* If optimization is already allowed, just return */ |
883 | if (kprobes_allow_optimization) | |
5c51543b | 884 | goto out; |
b2be84df | 885 | |
2d1e38f5 | 886 | cpus_read_lock(); |
b2be84df | 887 | kprobes_allow_optimization = true; |
b2be84df MH |
888 | for (i = 0; i < KPROBE_TABLE_SIZE; i++) { |
889 | head = &kprobe_table[i]; | |
7e6a71d8 | 890 | hlist_for_each_entry(p, head, hlist) |
b2be84df MH |
891 | if (!kprobe_disabled(p)) |
892 | optimize_kprobe(p); | |
893 | } | |
2d1e38f5 | 894 | cpus_read_unlock(); |
b2be84df | 895 | printk(KERN_INFO "Kprobes globally optimized\n"); |
5c51543b MH |
896 | out: |
897 | mutex_unlock(&kprobe_mutex); | |
b2be84df MH |
898 | } |
899 | ||
c85c9a2c | 900 | #ifdef CONFIG_SYSCTL |
55479f64 | 901 | static void unoptimize_all_kprobes(void) |
b2be84df MH |
902 | { |
903 | struct hlist_head *head; | |
b2be84df MH |
904 | struct kprobe *p; |
905 | unsigned int i; | |
906 | ||
5c51543b | 907 | mutex_lock(&kprobe_mutex); |
b2be84df | 908 | /* If optimization is already prohibited, just return */ |
5c51543b MH |
909 | if (!kprobes_allow_optimization) { |
910 | mutex_unlock(&kprobe_mutex); | |
b2be84df | 911 | return; |
5c51543b | 912 | } |
b2be84df | 913 | |
2d1e38f5 | 914 | cpus_read_lock(); |
b2be84df | 915 | kprobes_allow_optimization = false; |
b2be84df MH |
916 | for (i = 0; i < KPROBE_TABLE_SIZE; i++) { |
917 | head = &kprobe_table[i]; | |
7e6a71d8 | 918 | hlist_for_each_entry(p, head, hlist) { |
b2be84df | 919 | if (!kprobe_disabled(p)) |
6274de49 | 920 | unoptimize_kprobe(p, false); |
b2be84df MH |
921 | } |
922 | } | |
2d1e38f5 | 923 | cpus_read_unlock(); |
5c51543b MH |
924 | mutex_unlock(&kprobe_mutex); |
925 | ||
6274de49 MH |
926 | /* Wait for unoptimizing completion */ |
927 | wait_for_kprobe_optimizer(); | |
928 | printk(KERN_INFO "Kprobes globally unoptimized\n"); | |
b2be84df MH |
929 | } |
930 | ||
5c51543b | 931 | static DEFINE_MUTEX(kprobe_sysctl_mutex); |
b2be84df MH |
932 | int sysctl_kprobes_optimization; |
933 | int proc_kprobes_optimization_handler(struct ctl_table *table, int write, | |
32927393 | 934 | void *buffer, size_t *length, |
b2be84df MH |
935 | loff_t *ppos) |
936 | { | |
937 | int ret; | |
938 | ||
5c51543b | 939 | mutex_lock(&kprobe_sysctl_mutex); |
b2be84df MH |
940 | sysctl_kprobes_optimization = kprobes_allow_optimization ? 1 : 0; |
941 | ret = proc_dointvec_minmax(table, write, buffer, length, ppos); | |
942 | ||
943 | if (sysctl_kprobes_optimization) | |
944 | optimize_all_kprobes(); | |
945 | else | |
946 | unoptimize_all_kprobes(); | |
5c51543b | 947 | mutex_unlock(&kprobe_sysctl_mutex); |
b2be84df MH |
948 | |
949 | return ret; | |
950 | } | |
951 | #endif /* CONFIG_SYSCTL */ | |
952 | ||
6274de49 | 953 | /* Put a breakpoint for a probe. Must be called with text_mutex locked */ |
55479f64 | 954 | static void __arm_kprobe(struct kprobe *p) |
afd66255 | 955 | { |
6d8e40a8 | 956 | struct kprobe *_p; |
afd66255 MH |
957 | |
958 | /* Check collision with other optimized kprobes */ | |
6d8e40a8 MH |
959 | _p = get_optimized_kprobe((unsigned long)p->addr); |
960 | if (unlikely(_p)) | |
6274de49 MH |
961 | /* Fallback to unoptimized kprobe */ |
962 | unoptimize_kprobe(_p, true); | |
afd66255 MH |
963 | |
964 | arch_arm_kprobe(p); | |
965 | optimize_kprobe(p); /* Try to optimize (add kprobe to a list) */ | |
966 | } | |
967 | ||
6274de49 | 968 | /* Remove the breakpoint of a probe. Must be called with text_mutex locked */ |
55479f64 | 969 | static void __disarm_kprobe(struct kprobe *p, bool reopt) |
afd66255 | 970 | { |
6d8e40a8 | 971 | struct kprobe *_p; |
afd66255 | 972 | |
69d54b91 WN |
973 | /* Try to unoptimize */ |
974 | unoptimize_kprobe(p, kprobes_all_disarmed); | |
afd66255 | 975 | |
6274de49 MH |
976 | if (!kprobe_queued(p)) { |
977 | arch_disarm_kprobe(p); | |
978 | /* If another kprobe was blocked, optimize it. */ | |
979 | _p = get_optimized_kprobe((unsigned long)p->addr); | |
980 | if (unlikely(_p) && reopt) | |
981 | optimize_kprobe(_p); | |
982 | } | |
983 | /* TODO: reoptimize others after unoptimized this probe */ | |
afd66255 MH |
984 | } |
985 | ||
986 | #else /* !CONFIG_OPTPROBES */ | |
987 | ||
988 | #define optimize_kprobe(p) do {} while (0) | |
6274de49 | 989 | #define unoptimize_kprobe(p, f) do {} while (0) |
afd66255 MH |
990 | #define kill_optimized_kprobe(p) do {} while (0) |
991 | #define prepare_optimized_kprobe(p) do {} while (0) | |
992 | #define try_to_optimize_kprobe(p) do {} while (0) | |
993 | #define __arm_kprobe(p) arch_arm_kprobe(p) | |
6274de49 MH |
994 | #define __disarm_kprobe(p, o) arch_disarm_kprobe(p) |
995 | #define kprobe_disarmed(p) kprobe_disabled(p) | |
996 | #define wait_for_kprobe_optimizer() do {} while (0) | |
afd66255 | 997 | |
819319fc | 998 | static int reuse_unused_kprobe(struct kprobe *ap) |
0490cd1f | 999 | { |
819319fc MH |
1000 | /* |
1001 | * If the optimized kprobe is NOT supported, the aggr kprobe is | |
1002 | * released at the same time that the last aggregated kprobe is | |
1003 | * unregistered. | |
1004 | * Thus there should be no chance to reuse unused kprobe. | |
1005 | */ | |
0490cd1f | 1006 | printk(KERN_ERR "Error: There should be no unused kprobe here.\n"); |
819319fc | 1007 | return -EINVAL; |
0490cd1f MH |
1008 | } |
1009 | ||
55479f64 | 1010 | static void free_aggr_kprobe(struct kprobe *p) |
afd66255 | 1011 | { |
6274de49 | 1012 | arch_remove_kprobe(p); |
afd66255 MH |
1013 | kfree(p); |
1014 | } | |
1015 | ||
55479f64 | 1016 | static struct kprobe *alloc_aggr_kprobe(struct kprobe *p) |
afd66255 MH |
1017 | { |
1018 | return kzalloc(sizeof(struct kprobe), GFP_KERNEL); | |
1019 | } | |
1020 | #endif /* CONFIG_OPTPROBES */ | |
1021 | ||
e7dbfe34 | 1022 | #ifdef CONFIG_KPROBES_ON_FTRACE |
ae6aa16f | 1023 | static struct ftrace_ops kprobe_ftrace_ops __read_mostly = { |
0bc11ed5 MH |
1024 | .func = kprobe_ftrace_handler, |
1025 | .flags = FTRACE_OPS_FL_SAVE_REGS, | |
1026 | }; | |
1027 | ||
1028 | static struct ftrace_ops kprobe_ipmodify_ops __read_mostly = { | |
e5253896 | 1029 | .func = kprobe_ftrace_handler, |
1d70be34 | 1030 | .flags = FTRACE_OPS_FL_SAVE_REGS | FTRACE_OPS_FL_IPMODIFY, |
ae6aa16f | 1031 | }; |
0bc11ed5 MH |
1032 | |
1033 | static int kprobe_ipmodify_enabled; | |
ae6aa16f MH |
1034 | static int kprobe_ftrace_enabled; |
1035 | ||
ae6aa16f | 1036 | /* Caller must lock kprobe_mutex */ |
0bc11ed5 MH |
1037 | static int __arm_kprobe_ftrace(struct kprobe *p, struct ftrace_ops *ops, |
1038 | int *cnt) | |
ae6aa16f | 1039 | { |
12310e34 | 1040 | int ret = 0; |
ae6aa16f | 1041 | |
0bc11ed5 | 1042 | ret = ftrace_set_filter_ip(ops, (unsigned long)p->addr, 0, 0); |
12310e34 | 1043 | if (ret) { |
4458515b MH |
1044 | pr_debug("Failed to arm kprobe-ftrace at %pS (%d)\n", |
1045 | p->addr, ret); | |
12310e34 JY |
1046 | return ret; |
1047 | } | |
1048 | ||
0bc11ed5 MH |
1049 | if (*cnt == 0) { |
1050 | ret = register_ftrace_function(ops); | |
12310e34 JY |
1051 | if (ret) { |
1052 | pr_debug("Failed to init kprobe-ftrace (%d)\n", ret); | |
1053 | goto err_ftrace; | |
1054 | } | |
ae6aa16f | 1055 | } |
12310e34 | 1056 | |
0bc11ed5 | 1057 | (*cnt)++; |
12310e34 JY |
1058 | return ret; |
1059 | ||
1060 | err_ftrace: | |
1061 | /* | |
0bc11ed5 MH |
1062 | * At this point, sinec ops is not registered, we should be sefe from |
1063 | * registering empty filter. | |
12310e34 | 1064 | */ |
0bc11ed5 | 1065 | ftrace_set_filter_ip(ops, (unsigned long)p->addr, 1, 0); |
12310e34 | 1066 | return ret; |
ae6aa16f MH |
1067 | } |
1068 | ||
0bc11ed5 MH |
1069 | static int arm_kprobe_ftrace(struct kprobe *p) |
1070 | { | |
1071 | bool ipmodify = (p->post_handler != NULL); | |
1072 | ||
1073 | return __arm_kprobe_ftrace(p, | |
1074 | ipmodify ? &kprobe_ipmodify_ops : &kprobe_ftrace_ops, | |
1075 | ipmodify ? &kprobe_ipmodify_enabled : &kprobe_ftrace_enabled); | |
1076 | } | |
1077 | ||
ae6aa16f | 1078 | /* Caller must lock kprobe_mutex */ |
0bc11ed5 MH |
1079 | static int __disarm_kprobe_ftrace(struct kprobe *p, struct ftrace_ops *ops, |
1080 | int *cnt) | |
ae6aa16f | 1081 | { |
297f9233 | 1082 | int ret = 0; |
ae6aa16f | 1083 | |
0bc11ed5 MH |
1084 | if (*cnt == 1) { |
1085 | ret = unregister_ftrace_function(ops); | |
297f9233 JY |
1086 | if (WARN(ret < 0, "Failed to unregister kprobe-ftrace (%d)\n", ret)) |
1087 | return ret; | |
ae6aa16f | 1088 | } |
297f9233 | 1089 | |
0bc11ed5 | 1090 | (*cnt)--; |
297f9233 | 1091 | |
0bc11ed5 | 1092 | ret = ftrace_set_filter_ip(ops, (unsigned long)p->addr, 1, 0); |
4458515b MH |
1093 | WARN_ONCE(ret < 0, "Failed to disarm kprobe-ftrace at %pS (%d)\n", |
1094 | p->addr, ret); | |
297f9233 | 1095 | return ret; |
ae6aa16f | 1096 | } |
0bc11ed5 MH |
1097 | |
1098 | static int disarm_kprobe_ftrace(struct kprobe *p) | |
1099 | { | |
1100 | bool ipmodify = (p->post_handler != NULL); | |
1101 | ||
1102 | return __disarm_kprobe_ftrace(p, | |
1103 | ipmodify ? &kprobe_ipmodify_ops : &kprobe_ftrace_ops, | |
1104 | ipmodify ? &kprobe_ipmodify_enabled : &kprobe_ftrace_enabled); | |
1105 | } | |
e7dbfe34 | 1106 | #else /* !CONFIG_KPROBES_ON_FTRACE */ |
10de795a MS |
1107 | static inline int arm_kprobe_ftrace(struct kprobe *p) |
1108 | { | |
1109 | return -ENODEV; | |
1110 | } | |
1111 | ||
1112 | static inline int disarm_kprobe_ftrace(struct kprobe *p) | |
1113 | { | |
1114 | return -ENODEV; | |
1115 | } | |
ae6aa16f MH |
1116 | #endif |
1117 | ||
02afb8d6 PA |
1118 | static int prepare_kprobe(struct kprobe *p) |
1119 | { | |
1120 | /* Must ensure p->addr is really on ftrace */ | |
1121 | if (kprobe_ftrace(p)) | |
1122 | return arch_prepare_kprobe_ftrace(p); | |
1123 | ||
1124 | return arch_prepare_kprobe(p); | |
1125 | } | |
1126 | ||
201517a7 | 1127 | /* Arm a kprobe with text_mutex */ |
12310e34 | 1128 | static int arm_kprobe(struct kprobe *kp) |
201517a7 | 1129 | { |
12310e34 JY |
1130 | if (unlikely(kprobe_ftrace(kp))) |
1131 | return arm_kprobe_ftrace(kp); | |
1132 | ||
2d1e38f5 | 1133 | cpus_read_lock(); |
201517a7 | 1134 | mutex_lock(&text_mutex); |
afd66255 | 1135 | __arm_kprobe(kp); |
201517a7 | 1136 | mutex_unlock(&text_mutex); |
2d1e38f5 | 1137 | cpus_read_unlock(); |
12310e34 JY |
1138 | |
1139 | return 0; | |
201517a7 MH |
1140 | } |
1141 | ||
1142 | /* Disarm a kprobe with text_mutex */ | |
297f9233 | 1143 | static int disarm_kprobe(struct kprobe *kp, bool reopt) |
201517a7 | 1144 | { |
297f9233 JY |
1145 | if (unlikely(kprobe_ftrace(kp))) |
1146 | return disarm_kprobe_ftrace(kp); | |
2d1e38f5 TG |
1147 | |
1148 | cpus_read_lock(); | |
201517a7 | 1149 | mutex_lock(&text_mutex); |
ae6aa16f | 1150 | __disarm_kprobe(kp, reopt); |
201517a7 | 1151 | mutex_unlock(&text_mutex); |
2d1e38f5 | 1152 | cpus_read_unlock(); |
297f9233 JY |
1153 | |
1154 | return 0; | |
201517a7 MH |
1155 | } |
1156 | ||
64f562c6 AM |
1157 | /* |
1158 | * Aggregate handlers for multiple kprobes support - these handlers | |
1159 | * take care of invoking the individual kprobe handlers on p->list | |
1160 | */ | |
820aede0 | 1161 | static int aggr_pre_handler(struct kprobe *p, struct pt_regs *regs) |
64f562c6 AM |
1162 | { |
1163 | struct kprobe *kp; | |
1164 | ||
3516a460 | 1165 | list_for_each_entry_rcu(kp, &p->list, list) { |
de5bd88d | 1166 | if (kp->pre_handler && likely(!kprobe_disabled(kp))) { |
e6584523 | 1167 | set_kprobe_instance(kp); |
8b0914ea PP |
1168 | if (kp->pre_handler(kp, regs)) |
1169 | return 1; | |
64f562c6 | 1170 | } |
e6584523 | 1171 | reset_kprobe_instance(); |
64f562c6 AM |
1172 | } |
1173 | return 0; | |
1174 | } | |
820aede0 | 1175 | NOKPROBE_SYMBOL(aggr_pre_handler); |
64f562c6 | 1176 | |
820aede0 MH |
1177 | static void aggr_post_handler(struct kprobe *p, struct pt_regs *regs, |
1178 | unsigned long flags) | |
64f562c6 AM |
1179 | { |
1180 | struct kprobe *kp; | |
1181 | ||
3516a460 | 1182 | list_for_each_entry_rcu(kp, &p->list, list) { |
de5bd88d | 1183 | if (kp->post_handler && likely(!kprobe_disabled(kp))) { |
e6584523 | 1184 | set_kprobe_instance(kp); |
64f562c6 | 1185 | kp->post_handler(kp, regs, flags); |
e6584523 | 1186 | reset_kprobe_instance(); |
64f562c6 AM |
1187 | } |
1188 | } | |
64f562c6 | 1189 | } |
820aede0 | 1190 | NOKPROBE_SYMBOL(aggr_post_handler); |
64f562c6 | 1191 | |
bf8d5c52 | 1192 | /* Walks the list and increments nmissed count for multiprobe case */ |
820aede0 | 1193 | void kprobes_inc_nmissed_count(struct kprobe *p) |
bf8d5c52 KA |
1194 | { |
1195 | struct kprobe *kp; | |
afd66255 | 1196 | if (!kprobe_aggrprobe(p)) { |
bf8d5c52 KA |
1197 | p->nmissed++; |
1198 | } else { | |
1199 | list_for_each_entry_rcu(kp, &p->list, list) | |
1200 | kp->nmissed++; | |
1201 | } | |
1202 | return; | |
1203 | } | |
820aede0 | 1204 | NOKPROBE_SYMBOL(kprobes_inc_nmissed_count); |
bf8d5c52 | 1205 | |
d741bf41 PZ |
1206 | static void free_rp_inst_rcu(struct rcu_head *head) |
1207 | { | |
1208 | struct kretprobe_instance *ri = container_of(head, struct kretprobe_instance, rcu); | |
1209 | ||
1210 | if (refcount_dec_and_test(&ri->rph->ref)) | |
1211 | kfree(ri->rph); | |
1212 | kfree(ri); | |
1213 | } | |
1214 | NOKPROBE_SYMBOL(free_rp_inst_rcu); | |
1215 | ||
b3388178 | 1216 | static void recycle_rp_inst(struct kretprobe_instance *ri) |
b94cce92 | 1217 | { |
d741bf41 | 1218 | struct kretprobe *rp = get_kretprobe(ri); |
ef53d9c5 | 1219 | |
ef53d9c5 | 1220 | if (likely(rp)) { |
6e426e0f | 1221 | freelist_add(&ri->freelist, &rp->freelist); |
b94cce92 | 1222 | } else |
d741bf41 | 1223 | call_rcu(&ri->rcu, free_rp_inst_rcu); |
b94cce92 | 1224 | } |
820aede0 | 1225 | NOKPROBE_SYMBOL(recycle_rp_inst); |
b94cce92 | 1226 | |
319f0ce2 | 1227 | static struct kprobe kprobe_busy = { |
9b38cc70 JO |
1228 | .addr = (void *) get_kprobe, |
1229 | }; | |
1230 | ||
1231 | void kprobe_busy_begin(void) | |
1232 | { | |
1233 | struct kprobe_ctlblk *kcb; | |
1234 | ||
1235 | preempt_disable(); | |
1236 | __this_cpu_write(current_kprobe, &kprobe_busy); | |
1237 | kcb = get_kprobe_ctlblk(); | |
1238 | kcb->kprobe_status = KPROBE_HIT_ACTIVE; | |
1239 | } | |
1240 | ||
1241 | void kprobe_busy_end(void) | |
1242 | { | |
1243 | __this_cpu_write(current_kprobe, NULL); | |
1244 | preempt_enable(); | |
1245 | } | |
1246 | ||
b94cce92 | 1247 | /* |
c6fd91f0 | 1248 | * This function is called from finish_task_switch when task tk becomes dead, |
1249 | * so that we can recycle any function-return probe instances associated | |
1250 | * with this task. These left over instances represent probed functions | |
1251 | * that have been called but will never return. | |
b94cce92 | 1252 | */ |
820aede0 | 1253 | void kprobe_flush_task(struct task_struct *tk) |
b94cce92 | 1254 | { |
62c27be0 | 1255 | struct kretprobe_instance *ri; |
d741bf41 | 1256 | struct llist_node *node; |
802eae7c | 1257 | |
d741bf41 | 1258 | /* Early boot, not yet initialized. */ |
ef53d9c5 | 1259 | if (unlikely(!kprobes_initialized)) |
ef53d9c5 S |
1260 | return; |
1261 | ||
9b38cc70 JO |
1262 | kprobe_busy_begin(); |
1263 | ||
d741bf41 PZ |
1264 | node = __llist_del_all(&tk->kretprobe_instances); |
1265 | while (node) { | |
1266 | ri = container_of(node, struct kretprobe_instance, llist); | |
1267 | node = node->next; | |
1268 | ||
1269 | recycle_rp_inst(ri); | |
62c27be0 | 1270 | } |
9b38cc70 JO |
1271 | |
1272 | kprobe_busy_end(); | |
b94cce92 | 1273 | } |
820aede0 | 1274 | NOKPROBE_SYMBOL(kprobe_flush_task); |
b94cce92 | 1275 | |
b94cce92 HN |
1276 | static inline void free_rp_inst(struct kretprobe *rp) |
1277 | { | |
1278 | struct kretprobe_instance *ri; | |
6e426e0f | 1279 | struct freelist_node *node; |
d741bf41 | 1280 | int count = 0; |
4c4308cb | 1281 | |
6e426e0f PZ |
1282 | node = rp->freelist.head; |
1283 | while (node) { | |
1284 | ri = container_of(node, struct kretprobe_instance, freelist); | |
1285 | node = node->next; | |
1286 | ||
b94cce92 | 1287 | kfree(ri); |
d741bf41 | 1288 | count++; |
b94cce92 | 1289 | } |
ef53d9c5 | 1290 | |
d741bf41 PZ |
1291 | if (refcount_sub_and_test(count, &rp->rph->ref)) { |
1292 | kfree(rp->rph); | |
1293 | rp->rph = NULL; | |
4a296e07 | 1294 | } |
4a296e07 MH |
1295 | } |
1296 | ||
059053a2 | 1297 | /* Add the new probe to ap->list */ |
55479f64 | 1298 | static int add_new_kprobe(struct kprobe *ap, struct kprobe *p) |
8b0914ea | 1299 | { |
059053a2 | 1300 | if (p->post_handler) |
6274de49 | 1301 | unoptimize_kprobe(ap, true); /* Fall back to normal kprobe */ |
afd66255 | 1302 | |
059053a2 | 1303 | list_add_rcu(&p->list, &ap->list); |
b918e5e6 MH |
1304 | if (p->post_handler && !ap->post_handler) |
1305 | ap->post_handler = aggr_post_handler; | |
de5bd88d | 1306 | |
8b0914ea PP |
1307 | return 0; |
1308 | } | |
1309 | ||
64f562c6 AM |
1310 | /* |
1311 | * Fill in the required fields of the "manager kprobe". Replace the | |
1312 | * earlier kprobe in the hlist with the manager kprobe | |
1313 | */ | |
55479f64 | 1314 | static void init_aggr_kprobe(struct kprobe *ap, struct kprobe *p) |
64f562c6 | 1315 | { |
afd66255 | 1316 | /* Copy p's insn slot to ap */ |
8b0914ea | 1317 | copy_kprobe(p, ap); |
a9ad965e | 1318 | flush_insn_slot(ap); |
64f562c6 | 1319 | ap->addr = p->addr; |
afd66255 | 1320 | ap->flags = p->flags & ~KPROBE_FLAG_OPTIMIZED; |
64f562c6 | 1321 | ap->pre_handler = aggr_pre_handler; |
e8386a0c MH |
1322 | /* We don't care the kprobe which has gone. */ |
1323 | if (p->post_handler && !kprobe_gone(p)) | |
36721656 | 1324 | ap->post_handler = aggr_post_handler; |
64f562c6 AM |
1325 | |
1326 | INIT_LIST_HEAD(&ap->list); | |
afd66255 | 1327 | INIT_HLIST_NODE(&ap->hlist); |
64f562c6 | 1328 | |
afd66255 | 1329 | list_add_rcu(&p->list, &ap->list); |
adad0f33 | 1330 | hlist_replace_rcu(&p->hlist, &ap->hlist); |
64f562c6 AM |
1331 | } |
1332 | ||
1333 | /* | |
1334 | * This is the second or subsequent kprobe at the address - handle | |
1335 | * the intricacies | |
64f562c6 | 1336 | */ |
55479f64 | 1337 | static int register_aggr_kprobe(struct kprobe *orig_p, struct kprobe *p) |
64f562c6 AM |
1338 | { |
1339 | int ret = 0; | |
6d8e40a8 | 1340 | struct kprobe *ap = orig_p; |
64f562c6 | 1341 | |
2d1e38f5 TG |
1342 | cpus_read_lock(); |
1343 | ||
25764288 MH |
1344 | /* For preparing optimization, jump_label_text_reserved() is called */ |
1345 | jump_label_lock(); | |
25764288 MH |
1346 | mutex_lock(&text_mutex); |
1347 | ||
6d8e40a8 MH |
1348 | if (!kprobe_aggrprobe(orig_p)) { |
1349 | /* If orig_p is not an aggr_kprobe, create new aggr_kprobe. */ | |
1350 | ap = alloc_aggr_kprobe(orig_p); | |
25764288 MH |
1351 | if (!ap) { |
1352 | ret = -ENOMEM; | |
1353 | goto out; | |
1354 | } | |
6d8e40a8 | 1355 | init_aggr_kprobe(ap, orig_p); |
819319fc | 1356 | } else if (kprobe_unused(ap)) { |
0490cd1f | 1357 | /* This probe is going to die. Rescue it */ |
819319fc MH |
1358 | ret = reuse_unused_kprobe(ap); |
1359 | if (ret) | |
1360 | goto out; | |
1361 | } | |
b918e5e6 MH |
1362 | |
1363 | if (kprobe_gone(ap)) { | |
e8386a0c MH |
1364 | /* |
1365 | * Attempting to insert new probe at the same location that | |
1366 | * had a probe in the module vaddr area which already | |
1367 | * freed. So, the instruction slot has already been | |
1368 | * released. We need a new slot for the new probe. | |
1369 | */ | |
b918e5e6 | 1370 | ret = arch_prepare_kprobe(ap); |
e8386a0c | 1371 | if (ret) |
b918e5e6 MH |
1372 | /* |
1373 | * Even if fail to allocate new slot, don't need to | |
1374 | * free aggr_probe. It will be used next time, or | |
1375 | * freed by unregister_kprobe. | |
1376 | */ | |
25764288 | 1377 | goto out; |
de5bd88d | 1378 | |
afd66255 MH |
1379 | /* Prepare optimized instructions if possible. */ |
1380 | prepare_optimized_kprobe(ap); | |
1381 | ||
e8386a0c | 1382 | /* |
de5bd88d MH |
1383 | * Clear gone flag to prevent allocating new slot again, and |
1384 | * set disabled flag because it is not armed yet. | |
e8386a0c | 1385 | */ |
de5bd88d MH |
1386 | ap->flags = (ap->flags & ~KPROBE_FLAG_GONE) |
1387 | | KPROBE_FLAG_DISABLED; | |
e8386a0c | 1388 | } |
b918e5e6 | 1389 | |
afd66255 | 1390 | /* Copy ap's insn slot to p */ |
b918e5e6 | 1391 | copy_kprobe(ap, p); |
25764288 MH |
1392 | ret = add_new_kprobe(ap, p); |
1393 | ||
1394 | out: | |
1395 | mutex_unlock(&text_mutex); | |
25764288 | 1396 | jump_label_unlock(); |
2d1e38f5 | 1397 | cpus_read_unlock(); |
25764288 MH |
1398 | |
1399 | if (ret == 0 && kprobe_disabled(ap) && !kprobe_disabled(p)) { | |
1400 | ap->flags &= ~KPROBE_FLAG_DISABLED; | |
12310e34 | 1401 | if (!kprobes_all_disarmed) { |
25764288 | 1402 | /* Arm the breakpoint again. */ |
12310e34 JY |
1403 | ret = arm_kprobe(ap); |
1404 | if (ret) { | |
1405 | ap->flags |= KPROBE_FLAG_DISABLED; | |
1406 | list_del_rcu(&p->list); | |
ae8b7ce7 | 1407 | synchronize_rcu(); |
12310e34 JY |
1408 | } |
1409 | } | |
25764288 MH |
1410 | } |
1411 | return ret; | |
64f562c6 AM |
1412 | } |
1413 | ||
be8f2743 MH |
1414 | bool __weak arch_within_kprobe_blacklist(unsigned long addr) |
1415 | { | |
1416 | /* The __kprobes marked functions and entry code must not be probed */ | |
1417 | return addr >= (unsigned long)__kprobes_text_start && | |
1418 | addr < (unsigned long)__kprobes_text_end; | |
1419 | } | |
1420 | ||
6143c6fb | 1421 | static bool __within_kprobe_blacklist(unsigned long addr) |
d0aaff97 | 1422 | { |
376e2424 | 1423 | struct kprobe_blacklist_entry *ent; |
3d8d996e | 1424 | |
be8f2743 | 1425 | if (arch_within_kprobe_blacklist(addr)) |
376e2424 | 1426 | return true; |
3d8d996e SD |
1427 | /* |
1428 | * If there exists a kprobe_blacklist, verify and | |
1429 | * fail any probe registration in the prohibited area | |
1430 | */ | |
376e2424 MH |
1431 | list_for_each_entry(ent, &kprobe_blacklist, list) { |
1432 | if (addr >= ent->start_addr && addr < ent->end_addr) | |
1433 | return true; | |
3d8d996e | 1434 | } |
6143c6fb MH |
1435 | return false; |
1436 | } | |
376e2424 | 1437 | |
6143c6fb MH |
1438 | bool within_kprobe_blacklist(unsigned long addr) |
1439 | { | |
1440 | char symname[KSYM_NAME_LEN], *p; | |
1441 | ||
1442 | if (__within_kprobe_blacklist(addr)) | |
1443 | return true; | |
1444 | ||
1445 | /* Check if the address is on a suffixed-symbol */ | |
1446 | if (!lookup_symbol_name(addr, symname)) { | |
1447 | p = strchr(symname, '.'); | |
1448 | if (!p) | |
1449 | return false; | |
1450 | *p = '\0'; | |
1451 | addr = (unsigned long)kprobe_lookup_name(symname, 0); | |
1452 | if (addr) | |
1453 | return __within_kprobe_blacklist(addr); | |
1454 | } | |
376e2424 | 1455 | return false; |
d0aaff97 PP |
1456 | } |
1457 | ||
b2a5cd69 MH |
1458 | /* |
1459 | * If we have a symbol_name argument, look it up and add the offset field | |
1460 | * to it. This way, we can specify a relative address to a symbol. | |
bc81d48d MH |
1461 | * This returns encoded errors if it fails to look up symbol or invalid |
1462 | * combination of parameters. | |
b2a5cd69 | 1463 | */ |
1d585e70 NR |
1464 | static kprobe_opcode_t *_kprobe_addr(kprobe_opcode_t *addr, |
1465 | const char *symbol_name, unsigned int offset) | |
b2a5cd69 | 1466 | { |
1d585e70 | 1467 | if ((symbol_name && addr) || (!symbol_name && !addr)) |
bc81d48d MH |
1468 | goto invalid; |
1469 | ||
1d585e70 | 1470 | if (symbol_name) { |
7246f600 | 1471 | addr = kprobe_lookup_name(symbol_name, offset); |
bc81d48d MH |
1472 | if (!addr) |
1473 | return ERR_PTR(-ENOENT); | |
b2a5cd69 MH |
1474 | } |
1475 | ||
1d585e70 | 1476 | addr = (kprobe_opcode_t *)(((char *)addr) + offset); |
bc81d48d MH |
1477 | if (addr) |
1478 | return addr; | |
1479 | ||
1480 | invalid: | |
1481 | return ERR_PTR(-EINVAL); | |
b2a5cd69 MH |
1482 | } |
1483 | ||
1d585e70 NR |
1484 | static kprobe_opcode_t *kprobe_addr(struct kprobe *p) |
1485 | { | |
1486 | return _kprobe_addr(p->addr, p->symbol_name, p->offset); | |
1487 | } | |
1488 | ||
1f0ab409 | 1489 | /* Check passed kprobe is valid and return kprobe in kprobe_table. */ |
55479f64 | 1490 | static struct kprobe *__get_valid_kprobe(struct kprobe *p) |
1f0ab409 | 1491 | { |
6d8e40a8 | 1492 | struct kprobe *ap, *list_p; |
1f0ab409 | 1493 | |
7e6a71d8 MH |
1494 | lockdep_assert_held(&kprobe_mutex); |
1495 | ||
6d8e40a8 MH |
1496 | ap = get_kprobe(p->addr); |
1497 | if (unlikely(!ap)) | |
1f0ab409 AM |
1498 | return NULL; |
1499 | ||
6d8e40a8 | 1500 | if (p != ap) { |
7e6a71d8 | 1501 | list_for_each_entry(list_p, &ap->list, list) |
1f0ab409 AM |
1502 | if (list_p == p) |
1503 | /* kprobe p is a valid probe */ | |
1504 | goto valid; | |
1505 | return NULL; | |
1506 | } | |
1507 | valid: | |
6d8e40a8 | 1508 | return ap; |
1f0ab409 AM |
1509 | } |
1510 | ||
33b1d146 MH |
1511 | /* |
1512 | * Warn and return error if the kprobe is being re-registered since | |
1513 | * there must be a software bug. | |
1514 | */ | |
1515 | static inline int warn_kprobe_rereg(struct kprobe *p) | |
1f0ab409 AM |
1516 | { |
1517 | int ret = 0; | |
1f0ab409 AM |
1518 | |
1519 | mutex_lock(&kprobe_mutex); | |
33b1d146 | 1520 | if (WARN_ON_ONCE(__get_valid_kprobe(p))) |
1f0ab409 AM |
1521 | ret = -EINVAL; |
1522 | mutex_unlock(&kprobe_mutex); | |
6d8e40a8 | 1523 | |
1f0ab409 AM |
1524 | return ret; |
1525 | } | |
1526 | ||
4402deae | 1527 | static int check_ftrace_location(struct kprobe *p) |
1da177e4 | 1528 | { |
ae6aa16f MH |
1529 | unsigned long ftrace_addr; |
1530 | ||
ae6aa16f MH |
1531 | ftrace_addr = ftrace_location((unsigned long)p->addr); |
1532 | if (ftrace_addr) { | |
e7dbfe34 | 1533 | #ifdef CONFIG_KPROBES_ON_FTRACE |
ae6aa16f MH |
1534 | /* Given address is not on the instruction boundary */ |
1535 | if ((unsigned long)p->addr != ftrace_addr) | |
1536 | return -EILSEQ; | |
ae6aa16f | 1537 | p->flags |= KPROBE_FLAG_FTRACE; |
e7dbfe34 | 1538 | #else /* !CONFIG_KPROBES_ON_FTRACE */ |
ae6aa16f MH |
1539 | return -EINVAL; |
1540 | #endif | |
1541 | } | |
f7f242ff HC |
1542 | return 0; |
1543 | } | |
1544 | ||
1545 | static int check_kprobe_address_safe(struct kprobe *p, | |
1546 | struct module **probed_mod) | |
1547 | { | |
1548 | int ret; | |
1f0ab409 | 1549 | |
4402deae | 1550 | ret = check_ftrace_location(p); |
f7f242ff HC |
1551 | if (ret) |
1552 | return ret; | |
91bad2f8 | 1553 | jump_label_lock(); |
de31c3ca | 1554 | preempt_disable(); |
f7fa6ef0 MH |
1555 | |
1556 | /* Ensure it is not in reserved area nor out of text */ | |
ec30c5f3 | 1557 | if (!kernel_text_address((unsigned long) p->addr) || |
376e2424 | 1558 | within_kprobe_blacklist((unsigned long) p->addr) || |
e336b402 | 1559 | jump_label_text_reserved(p->addr, p->addr) || |
fa68bd09 | 1560 | static_call_text_reserved(p->addr, p->addr) || |
e336b402 | 1561 | find_bug((unsigned long)p->addr)) { |
f986a499 | 1562 | ret = -EINVAL; |
f7fa6ef0 | 1563 | goto out; |
f986a499 | 1564 | } |
b3e55c72 | 1565 | |
f7fa6ef0 MH |
1566 | /* Check if are we probing a module */ |
1567 | *probed_mod = __module_text_address((unsigned long) p->addr); | |
1568 | if (*probed_mod) { | |
6f716acd | 1569 | /* |
e8386a0c MH |
1570 | * We must hold a refcount of the probed module while updating |
1571 | * its code to prohibit unexpected unloading. | |
df019b1d | 1572 | */ |
f7fa6ef0 MH |
1573 | if (unlikely(!try_module_get(*probed_mod))) { |
1574 | ret = -ENOENT; | |
1575 | goto out; | |
1576 | } | |
de31c3ca | 1577 | |
f24659d9 MH |
1578 | /* |
1579 | * If the module freed .init.text, we couldn't insert | |
1580 | * kprobes in there. | |
1581 | */ | |
f7fa6ef0 MH |
1582 | if (within_module_init((unsigned long)p->addr, *probed_mod) && |
1583 | (*probed_mod)->state != MODULE_STATE_COMING) { | |
1584 | module_put(*probed_mod); | |
1585 | *probed_mod = NULL; | |
1586 | ret = -ENOENT; | |
f24659d9 | 1587 | } |
df019b1d | 1588 | } |
f7fa6ef0 | 1589 | out: |
a189d035 | 1590 | preempt_enable(); |
de31c3ca | 1591 | jump_label_unlock(); |
1da177e4 | 1592 | |
f7fa6ef0 MH |
1593 | return ret; |
1594 | } | |
1595 | ||
55479f64 | 1596 | int register_kprobe(struct kprobe *p) |
f7fa6ef0 MH |
1597 | { |
1598 | int ret; | |
1599 | struct kprobe *old_p; | |
1600 | struct module *probed_mod; | |
1601 | kprobe_opcode_t *addr; | |
1602 | ||
1603 | /* Adjust probe address from symbol */ | |
1604 | addr = kprobe_addr(p); | |
1605 | if (IS_ERR(addr)) | |
1606 | return PTR_ERR(addr); | |
1607 | p->addr = addr; | |
1608 | ||
33b1d146 | 1609 | ret = warn_kprobe_rereg(p); |
f7fa6ef0 MH |
1610 | if (ret) |
1611 | return ret; | |
1612 | ||
1613 | /* User can pass only KPROBE_FLAG_DISABLED to register_kprobe */ | |
1614 | p->flags &= KPROBE_FLAG_DISABLED; | |
3516a460 | 1615 | p->nmissed = 0; |
9861668f | 1616 | INIT_LIST_HEAD(&p->list); |
afd66255 | 1617 | |
f7fa6ef0 MH |
1618 | ret = check_kprobe_address_safe(p, &probed_mod); |
1619 | if (ret) | |
1620 | return ret; | |
1621 | ||
1622 | mutex_lock(&kprobe_mutex); | |
afd66255 | 1623 | |
64f562c6 AM |
1624 | old_p = get_kprobe(p->addr); |
1625 | if (old_p) { | |
afd66255 | 1626 | /* Since this may unoptimize old_p, locking text_mutex. */ |
64f562c6 | 1627 | ret = register_aggr_kprobe(old_p, p); |
1da177e4 LT |
1628 | goto out; |
1629 | } | |
1da177e4 | 1630 | |
2d1e38f5 TG |
1631 | cpus_read_lock(); |
1632 | /* Prevent text modification */ | |
1633 | mutex_lock(&text_mutex); | |
ae6aa16f | 1634 | ret = prepare_kprobe(p); |
25764288 | 1635 | mutex_unlock(&text_mutex); |
2d1e38f5 | 1636 | cpus_read_unlock(); |
6f716acd | 1637 | if (ret) |
afd66255 | 1638 | goto out; |
49a2a1b8 | 1639 | |
64f562c6 | 1640 | INIT_HLIST_NODE(&p->hlist); |
3516a460 | 1641 | hlist_add_head_rcu(&p->hlist, |
1da177e4 LT |
1642 | &kprobe_table[hash_ptr(p->addr, KPROBE_HASH_BITS)]); |
1643 | ||
12310e34 JY |
1644 | if (!kprobes_all_disarmed && !kprobe_disabled(p)) { |
1645 | ret = arm_kprobe(p); | |
1646 | if (ret) { | |
1647 | hlist_del_rcu(&p->hlist); | |
ae8b7ce7 | 1648 | synchronize_rcu(); |
12310e34 JY |
1649 | goto out; |
1650 | } | |
1651 | } | |
afd66255 MH |
1652 | |
1653 | /* Try to optimize kprobe */ | |
1654 | try_to_optimize_kprobe(p); | |
1da177e4 | 1655 | out: |
7a7d1cf9 | 1656 | mutex_unlock(&kprobe_mutex); |
49a2a1b8 | 1657 | |
e8386a0c | 1658 | if (probed_mod) |
df019b1d | 1659 | module_put(probed_mod); |
e8386a0c | 1660 | |
1da177e4 LT |
1661 | return ret; |
1662 | } | |
99081ab5 | 1663 | EXPORT_SYMBOL_GPL(register_kprobe); |
1da177e4 | 1664 | |
6f0f1dd7 | 1665 | /* Check if all probes on the aggrprobe are disabled */ |
55479f64 | 1666 | static int aggr_kprobe_disabled(struct kprobe *ap) |
6f0f1dd7 MH |
1667 | { |
1668 | struct kprobe *kp; | |
1669 | ||
7e6a71d8 MH |
1670 | lockdep_assert_held(&kprobe_mutex); |
1671 | ||
1672 | list_for_each_entry(kp, &ap->list, list) | |
6f0f1dd7 MH |
1673 | if (!kprobe_disabled(kp)) |
1674 | /* | |
1675 | * There is an active probe on the list. | |
1676 | * We can't disable this ap. | |
1677 | */ | |
1678 | return 0; | |
1679 | ||
1680 | return 1; | |
1681 | } | |
1682 | ||
1683 | /* Disable one kprobe: Make sure called under kprobe_mutex is locked */ | |
55479f64 | 1684 | static struct kprobe *__disable_kprobe(struct kprobe *p) |
6f0f1dd7 MH |
1685 | { |
1686 | struct kprobe *orig_p; | |
297f9233 | 1687 | int ret; |
6f0f1dd7 MH |
1688 | |
1689 | /* Get an original kprobe for return */ | |
1690 | orig_p = __get_valid_kprobe(p); | |
1691 | if (unlikely(orig_p == NULL)) | |
297f9233 | 1692 | return ERR_PTR(-EINVAL); |
6f0f1dd7 MH |
1693 | |
1694 | if (!kprobe_disabled(p)) { | |
1695 | /* Disable probe if it is a child probe */ | |
1696 | if (p != orig_p) | |
1697 | p->flags |= KPROBE_FLAG_DISABLED; | |
1698 | ||
1699 | /* Try to disarm and disable this/parent probe */ | |
1700 | if (p == orig_p || aggr_kprobe_disabled(orig_p)) { | |
69d54b91 WN |
1701 | /* |
1702 | * If kprobes_all_disarmed is set, orig_p | |
1703 | * should have already been disarmed, so | |
1704 | * skip unneed disarming process. | |
1705 | */ | |
297f9233 JY |
1706 | if (!kprobes_all_disarmed) { |
1707 | ret = disarm_kprobe(orig_p, true); | |
1708 | if (ret) { | |
1709 | p->flags &= ~KPROBE_FLAG_DISABLED; | |
1710 | return ERR_PTR(ret); | |
1711 | } | |
1712 | } | |
6f0f1dd7 MH |
1713 | orig_p->flags |= KPROBE_FLAG_DISABLED; |
1714 | } | |
1715 | } | |
1716 | ||
1717 | return orig_p; | |
1718 | } | |
1719 | ||
de5bd88d MH |
1720 | /* |
1721 | * Unregister a kprobe without a scheduler synchronization. | |
1722 | */ | |
55479f64 | 1723 | static int __unregister_kprobe_top(struct kprobe *p) |
de5bd88d | 1724 | { |
6d8e40a8 | 1725 | struct kprobe *ap, *list_p; |
de5bd88d | 1726 | |
6f0f1dd7 MH |
1727 | /* Disable kprobe. This will disarm it if needed. */ |
1728 | ap = __disable_kprobe(p); | |
297f9233 JY |
1729 | if (IS_ERR(ap)) |
1730 | return PTR_ERR(ap); | |
de5bd88d | 1731 | |
6f0f1dd7 | 1732 | if (ap == p) |
bf8f6e5b | 1733 | /* |
6f0f1dd7 MH |
1734 | * This probe is an independent(and non-optimized) kprobe |
1735 | * (not an aggrprobe). Remove from the hash list. | |
bf8f6e5b | 1736 | */ |
6f0f1dd7 MH |
1737 | goto disarmed; |
1738 | ||
1739 | /* Following process expects this probe is an aggrprobe */ | |
1740 | WARN_ON(!kprobe_aggrprobe(ap)); | |
1741 | ||
6274de49 MH |
1742 | if (list_is_singular(&ap->list) && kprobe_disarmed(ap)) |
1743 | /* | |
1744 | * !disarmed could be happen if the probe is under delayed | |
1745 | * unoptimizing. | |
1746 | */ | |
6f0f1dd7 MH |
1747 | goto disarmed; |
1748 | else { | |
1749 | /* If disabling probe has special handlers, update aggrprobe */ | |
e8386a0c | 1750 | if (p->post_handler && !kprobe_gone(p)) { |
7e6a71d8 | 1751 | list_for_each_entry(list_p, &ap->list, list) { |
9861668f MH |
1752 | if ((list_p != p) && (list_p->post_handler)) |
1753 | goto noclean; | |
1754 | } | |
6d8e40a8 | 1755 | ap->post_handler = NULL; |
9861668f MH |
1756 | } |
1757 | noclean: | |
6f0f1dd7 MH |
1758 | /* |
1759 | * Remove from the aggrprobe: this path will do nothing in | |
1760 | * __unregister_kprobe_bottom(). | |
1761 | */ | |
49a2a1b8 | 1762 | list_del_rcu(&p->list); |
6f0f1dd7 MH |
1763 | if (!kprobe_disabled(ap) && !kprobes_all_disarmed) |
1764 | /* | |
1765 | * Try to optimize this probe again, because post | |
1766 | * handler may have been changed. | |
1767 | */ | |
1768 | optimize_kprobe(ap); | |
49a2a1b8 | 1769 | } |
9861668f | 1770 | return 0; |
6f0f1dd7 MH |
1771 | |
1772 | disarmed: | |
1773 | hlist_del_rcu(&ap->hlist); | |
1774 | return 0; | |
9861668f | 1775 | } |
3516a460 | 1776 | |
55479f64 | 1777 | static void __unregister_kprobe_bottom(struct kprobe *p) |
9861668f | 1778 | { |
6d8e40a8 | 1779 | struct kprobe *ap; |
b3e55c72 | 1780 | |
e8386a0c | 1781 | if (list_empty(&p->list)) |
6274de49 | 1782 | /* This is an independent kprobe */ |
0498b635 | 1783 | arch_remove_kprobe(p); |
e8386a0c | 1784 | else if (list_is_singular(&p->list)) { |
6274de49 | 1785 | /* This is the last child of an aggrprobe */ |
6d8e40a8 | 1786 | ap = list_entry(p->list.next, struct kprobe, list); |
e8386a0c | 1787 | list_del(&p->list); |
6d8e40a8 | 1788 | free_aggr_kprobe(ap); |
9861668f | 1789 | } |
6274de49 | 1790 | /* Otherwise, do nothing. */ |
9861668f MH |
1791 | } |
1792 | ||
55479f64 | 1793 | int register_kprobes(struct kprobe **kps, int num) |
9861668f MH |
1794 | { |
1795 | int i, ret = 0; | |
1796 | ||
1797 | if (num <= 0) | |
1798 | return -EINVAL; | |
1799 | for (i = 0; i < num; i++) { | |
49ad2fd7 | 1800 | ret = register_kprobe(kps[i]); |
67dddaad MH |
1801 | if (ret < 0) { |
1802 | if (i > 0) | |
1803 | unregister_kprobes(kps, i); | |
9861668f | 1804 | break; |
36721656 | 1805 | } |
49a2a1b8 | 1806 | } |
9861668f MH |
1807 | return ret; |
1808 | } | |
99081ab5 | 1809 | EXPORT_SYMBOL_GPL(register_kprobes); |
9861668f | 1810 | |
55479f64 | 1811 | void unregister_kprobe(struct kprobe *p) |
9861668f MH |
1812 | { |
1813 | unregister_kprobes(&p, 1); | |
1814 | } | |
99081ab5 | 1815 | EXPORT_SYMBOL_GPL(unregister_kprobe); |
9861668f | 1816 | |
55479f64 | 1817 | void unregister_kprobes(struct kprobe **kps, int num) |
9861668f MH |
1818 | { |
1819 | int i; | |
1820 | ||
1821 | if (num <= 0) | |
1822 | return; | |
1823 | mutex_lock(&kprobe_mutex); | |
1824 | for (i = 0; i < num; i++) | |
1825 | if (__unregister_kprobe_top(kps[i]) < 0) | |
1826 | kps[i]->addr = NULL; | |
1827 | mutex_unlock(&kprobe_mutex); | |
1828 | ||
ae8b7ce7 | 1829 | synchronize_rcu(); |
9861668f MH |
1830 | for (i = 0; i < num; i++) |
1831 | if (kps[i]->addr) | |
1832 | __unregister_kprobe_bottom(kps[i]); | |
1da177e4 | 1833 | } |
99081ab5 | 1834 | EXPORT_SYMBOL_GPL(unregister_kprobes); |
1da177e4 | 1835 | |
5f6bee34 NR |
1836 | int __weak kprobe_exceptions_notify(struct notifier_block *self, |
1837 | unsigned long val, void *data) | |
fc62d020 NR |
1838 | { |
1839 | return NOTIFY_DONE; | |
1840 | } | |
5f6bee34 | 1841 | NOKPROBE_SYMBOL(kprobe_exceptions_notify); |
fc62d020 | 1842 | |
1da177e4 | 1843 | static struct notifier_block kprobe_exceptions_nb = { |
3d5631e0 AK |
1844 | .notifier_call = kprobe_exceptions_notify, |
1845 | .priority = 0x7fffffff /* we need to be notified first */ | |
1846 | }; | |
1847 | ||
3d7e3382 ME |
1848 | unsigned long __weak arch_deref_entry_point(void *entry) |
1849 | { | |
1850 | return (unsigned long)entry; | |
1851 | } | |
1da177e4 | 1852 | |
9edddaa2 | 1853 | #ifdef CONFIG_KRETPROBES |
66ada2cc MH |
1854 | |
1855 | unsigned long __kretprobe_trampoline_handler(struct pt_regs *regs, | |
1856 | void *trampoline_address, | |
1857 | void *frame_pointer) | |
1858 | { | |
66ada2cc | 1859 | kprobe_opcode_t *correct_ret_addr = NULL; |
d741bf41 PZ |
1860 | struct kretprobe_instance *ri = NULL; |
1861 | struct llist_node *first, *node; | |
1862 | struct kretprobe *rp; | |
66ada2cc | 1863 | |
d741bf41 PZ |
1864 | /* Find all nodes for this frame. */ |
1865 | first = node = current->kretprobe_instances.first; | |
1866 | while (node) { | |
1867 | ri = container_of(node, struct kretprobe_instance, llist); | |
66ada2cc | 1868 | |
d741bf41 | 1869 | BUG_ON(ri->fp != frame_pointer); |
66ada2cc | 1870 | |
d741bf41 PZ |
1871 | if (ri->ret_addr != trampoline_address) { |
1872 | correct_ret_addr = ri->ret_addr; | |
66ada2cc MH |
1873 | /* |
1874 | * This is the real return address. Any other | |
1875 | * instances associated with this task are for | |
1876 | * other calls deeper on the call stack | |
1877 | */ | |
d741bf41 PZ |
1878 | goto found; |
1879 | } | |
1880 | ||
1881 | node = node->next; | |
66ada2cc | 1882 | } |
d741bf41 PZ |
1883 | pr_err("Oops! Kretprobe fails to find correct return address.\n"); |
1884 | BUG_ON(1); | |
66ada2cc | 1885 | |
d741bf41 PZ |
1886 | found: |
1887 | /* Unlink all nodes for this frame. */ | |
1888 | current->kretprobe_instances.first = node->next; | |
1889 | node->next = NULL; | |
66ada2cc | 1890 | |
d741bf41 PZ |
1891 | /* Run them.. */ |
1892 | while (first) { | |
1893 | ri = container_of(first, struct kretprobe_instance, llist); | |
1894 | first = first->next; | |
66ada2cc | 1895 | |
d741bf41 PZ |
1896 | rp = get_kretprobe(ri); |
1897 | if (rp && rp->handler) { | |
66ada2cc MH |
1898 | struct kprobe *prev = kprobe_running(); |
1899 | ||
d741bf41 | 1900 | __this_cpu_write(current_kprobe, &rp->kp); |
66ada2cc | 1901 | ri->ret_addr = correct_ret_addr; |
d741bf41 | 1902 | rp->handler(ri, regs); |
66ada2cc MH |
1903 | __this_cpu_write(current_kprobe, prev); |
1904 | } | |
1905 | ||
b3388178 | 1906 | recycle_rp_inst(ri); |
66ada2cc MH |
1907 | } |
1908 | ||
66ada2cc MH |
1909 | return (unsigned long)correct_ret_addr; |
1910 | } | |
1911 | NOKPROBE_SYMBOL(__kretprobe_trampoline_handler) | |
1912 | ||
e65cefe8 AB |
1913 | /* |
1914 | * This kprobe pre_handler is registered with every kretprobe. When probe | |
1915 | * hits it will set up the return probe. | |
1916 | */ | |
820aede0 | 1917 | static int pre_handler_kretprobe(struct kprobe *p, struct pt_regs *regs) |
e65cefe8 AB |
1918 | { |
1919 | struct kretprobe *rp = container_of(p, struct kretprobe, kp); | |
ef53d9c5 | 1920 | struct kretprobe_instance *ri; |
6e426e0f | 1921 | struct freelist_node *fn; |
e65cefe8 | 1922 | |
6e426e0f PZ |
1923 | fn = freelist_try_get(&rp->freelist); |
1924 | if (!fn) { | |
1925 | rp->nmissed++; | |
1926 | return 0; | |
1927 | } | |
4c4308cb | 1928 | |
6e426e0f | 1929 | ri = container_of(fn, struct kretprobe_instance, freelist); |
d741bf41 | 1930 | |
6e426e0f PZ |
1931 | if (rp->entry_handler && rp->entry_handler(ri, regs)) { |
1932 | freelist_add(&ri->freelist, &rp->freelist); | |
1933 | return 0; | |
ef53d9c5 | 1934 | } |
6e426e0f PZ |
1935 | |
1936 | arch_prepare_kretprobe(ri, regs); | |
1937 | ||
1938 | __llist_add(&ri->llist, ¤t->kretprobe_instances); | |
1939 | ||
e65cefe8 AB |
1940 | return 0; |
1941 | } | |
820aede0 | 1942 | NOKPROBE_SYMBOL(pre_handler_kretprobe); |
e65cefe8 | 1943 | |
659b957f | 1944 | bool __weak arch_kprobe_on_func_entry(unsigned long offset) |
90ec5e89 NR |
1945 | { |
1946 | return !offset; | |
1947 | } | |
1948 | ||
97c753e6 MH |
1949 | /** |
1950 | * kprobe_on_func_entry() -- check whether given address is function entry | |
1951 | * @addr: Target address | |
1952 | * @sym: Target symbol name | |
1953 | * @offset: The offset from the symbol or the address | |
1954 | * | |
1955 | * This checks whether the given @addr+@offset or @sym+@offset is on the | |
1956 | * function entry address or not. | |
1957 | * This returns 0 if it is the function entry, or -EINVAL if it is not. | |
1958 | * And also it returns -ENOENT if it fails the symbol or address lookup. | |
1959 | * Caller must pass @addr or @sym (either one must be NULL), or this | |
1960 | * returns -EINVAL. | |
1961 | */ | |
1962 | int kprobe_on_func_entry(kprobe_opcode_t *addr, const char *sym, unsigned long offset) | |
1d585e70 NR |
1963 | { |
1964 | kprobe_opcode_t *kp_addr = _kprobe_addr(addr, sym, offset); | |
1965 | ||
1966 | if (IS_ERR(kp_addr)) | |
97c753e6 | 1967 | return PTR_ERR(kp_addr); |
1d585e70 | 1968 | |
97c753e6 MH |
1969 | if (!kallsyms_lookup_size_offset((unsigned long)kp_addr, NULL, &offset)) |
1970 | return -ENOENT; | |
1d585e70 | 1971 | |
97c753e6 MH |
1972 | if (!arch_kprobe_on_func_entry(offset)) |
1973 | return -EINVAL; | |
1974 | ||
1975 | return 0; | |
1d585e70 NR |
1976 | } |
1977 | ||
55479f64 | 1978 | int register_kretprobe(struct kretprobe *rp) |
b94cce92 | 1979 | { |
97c753e6 | 1980 | int ret; |
b94cce92 HN |
1981 | struct kretprobe_instance *inst; |
1982 | int i; | |
b2a5cd69 | 1983 | void *addr; |
90ec5e89 | 1984 | |
97c753e6 MH |
1985 | ret = kprobe_on_func_entry(rp->kp.addr, rp->kp.symbol_name, rp->kp.offset); |
1986 | if (ret) | |
1987 | return ret; | |
f438d914 | 1988 | |
0188b878 | 1989 | /* If only rp->kp.addr is specified, check reregistering kprobes */ |
33b1d146 | 1990 | if (rp->kp.addr && warn_kprobe_rereg(&rp->kp)) |
0188b878 WS |
1991 | return -EINVAL; |
1992 | ||
f438d914 | 1993 | if (kretprobe_blacklist_size) { |
b2a5cd69 | 1994 | addr = kprobe_addr(&rp->kp); |
bc81d48d MH |
1995 | if (IS_ERR(addr)) |
1996 | return PTR_ERR(addr); | |
f438d914 MH |
1997 | |
1998 | for (i = 0; kretprobe_blacklist[i].name != NULL; i++) { | |
1999 | if (kretprobe_blacklist[i].addr == addr) | |
2000 | return -EINVAL; | |
2001 | } | |
2002 | } | |
b94cce92 HN |
2003 | |
2004 | rp->kp.pre_handler = pre_handler_kretprobe; | |
7522a842 | 2005 | rp->kp.post_handler = NULL; |
b94cce92 HN |
2006 | |
2007 | /* Pre-allocate memory for max kretprobe instances */ | |
2008 | if (rp->maxactive <= 0) { | |
92616606 | 2009 | #ifdef CONFIG_PREEMPTION |
c2ef6661 | 2010 | rp->maxactive = max_t(unsigned int, 10, 2*num_possible_cpus()); |
b94cce92 | 2011 | #else |
4dae560f | 2012 | rp->maxactive = num_possible_cpus(); |
b94cce92 HN |
2013 | #endif |
2014 | } | |
6e426e0f | 2015 | rp->freelist.head = NULL; |
d741bf41 PZ |
2016 | rp->rph = kzalloc(sizeof(struct kretprobe_holder), GFP_KERNEL); |
2017 | if (!rp->rph) | |
2018 | return -ENOMEM; | |
2019 | ||
2020 | rp->rph->rp = rp; | |
b94cce92 | 2021 | for (i = 0; i < rp->maxactive; i++) { |
d741bf41 | 2022 | inst = kzalloc(sizeof(struct kretprobe_instance) + |
f47cd9b5 | 2023 | rp->data_size, GFP_KERNEL); |
b94cce92 | 2024 | if (inst == NULL) { |
d741bf41 | 2025 | refcount_set(&rp->rph->ref, i); |
b94cce92 HN |
2026 | free_rp_inst(rp); |
2027 | return -ENOMEM; | |
2028 | } | |
d741bf41 | 2029 | inst->rph = rp->rph; |
6e426e0f | 2030 | freelist_add(&inst->freelist, &rp->freelist); |
b94cce92 | 2031 | } |
d741bf41 | 2032 | refcount_set(&rp->rph->ref, i); |
b94cce92 HN |
2033 | |
2034 | rp->nmissed = 0; | |
2035 | /* Establish function entry probe point */ | |
49ad2fd7 | 2036 | ret = register_kprobe(&rp->kp); |
4a296e07 | 2037 | if (ret != 0) |
b94cce92 HN |
2038 | free_rp_inst(rp); |
2039 | return ret; | |
2040 | } | |
99081ab5 | 2041 | EXPORT_SYMBOL_GPL(register_kretprobe); |
b94cce92 | 2042 | |
55479f64 | 2043 | int register_kretprobes(struct kretprobe **rps, int num) |
4a296e07 MH |
2044 | { |
2045 | int ret = 0, i; | |
2046 | ||
2047 | if (num <= 0) | |
2048 | return -EINVAL; | |
2049 | for (i = 0; i < num; i++) { | |
49ad2fd7 | 2050 | ret = register_kretprobe(rps[i]); |
67dddaad MH |
2051 | if (ret < 0) { |
2052 | if (i > 0) | |
2053 | unregister_kretprobes(rps, i); | |
4a296e07 MH |
2054 | break; |
2055 | } | |
2056 | } | |
2057 | return ret; | |
2058 | } | |
99081ab5 | 2059 | EXPORT_SYMBOL_GPL(register_kretprobes); |
4a296e07 | 2060 | |
55479f64 | 2061 | void unregister_kretprobe(struct kretprobe *rp) |
4a296e07 MH |
2062 | { |
2063 | unregister_kretprobes(&rp, 1); | |
2064 | } | |
99081ab5 | 2065 | EXPORT_SYMBOL_GPL(unregister_kretprobe); |
4a296e07 | 2066 | |
55479f64 | 2067 | void unregister_kretprobes(struct kretprobe **rps, int num) |
4a296e07 MH |
2068 | { |
2069 | int i; | |
2070 | ||
2071 | if (num <= 0) | |
2072 | return; | |
2073 | mutex_lock(&kprobe_mutex); | |
d741bf41 | 2074 | for (i = 0; i < num; i++) { |
4a296e07 MH |
2075 | if (__unregister_kprobe_top(&rps[i]->kp) < 0) |
2076 | rps[i]->kp.addr = NULL; | |
d741bf41 PZ |
2077 | rps[i]->rph->rp = NULL; |
2078 | } | |
4a296e07 MH |
2079 | mutex_unlock(&kprobe_mutex); |
2080 | ||
ae8b7ce7 | 2081 | synchronize_rcu(); |
4a296e07 MH |
2082 | for (i = 0; i < num; i++) { |
2083 | if (rps[i]->kp.addr) { | |
2084 | __unregister_kprobe_bottom(&rps[i]->kp); | |
d741bf41 | 2085 | free_rp_inst(rps[i]); |
4a296e07 MH |
2086 | } |
2087 | } | |
2088 | } | |
99081ab5 | 2089 | EXPORT_SYMBOL_GPL(unregister_kretprobes); |
4a296e07 | 2090 | |
9edddaa2 | 2091 | #else /* CONFIG_KRETPROBES */ |
55479f64 | 2092 | int register_kretprobe(struct kretprobe *rp) |
b94cce92 HN |
2093 | { |
2094 | return -ENOSYS; | |
2095 | } | |
99081ab5 | 2096 | EXPORT_SYMBOL_GPL(register_kretprobe); |
b94cce92 | 2097 | |
55479f64 | 2098 | int register_kretprobes(struct kretprobe **rps, int num) |
346fd59b | 2099 | { |
4a296e07 | 2100 | return -ENOSYS; |
346fd59b | 2101 | } |
99081ab5 MH |
2102 | EXPORT_SYMBOL_GPL(register_kretprobes); |
2103 | ||
55479f64 | 2104 | void unregister_kretprobe(struct kretprobe *rp) |
b94cce92 | 2105 | { |
4a296e07 | 2106 | } |
99081ab5 | 2107 | EXPORT_SYMBOL_GPL(unregister_kretprobe); |
b94cce92 | 2108 | |
55479f64 | 2109 | void unregister_kretprobes(struct kretprobe **rps, int num) |
4a296e07 MH |
2110 | { |
2111 | } | |
99081ab5 | 2112 | EXPORT_SYMBOL_GPL(unregister_kretprobes); |
4c4308cb | 2113 | |
820aede0 | 2114 | static int pre_handler_kretprobe(struct kprobe *p, struct pt_regs *regs) |
4a296e07 MH |
2115 | { |
2116 | return 0; | |
b94cce92 | 2117 | } |
820aede0 | 2118 | NOKPROBE_SYMBOL(pre_handler_kretprobe); |
b94cce92 | 2119 | |
4a296e07 MH |
2120 | #endif /* CONFIG_KRETPROBES */ |
2121 | ||
e8386a0c | 2122 | /* Set the kprobe gone and remove its instruction buffer. */ |
55479f64 | 2123 | static void kill_kprobe(struct kprobe *p) |
e8386a0c MH |
2124 | { |
2125 | struct kprobe *kp; | |
de5bd88d | 2126 | |
7e6a71d8 MH |
2127 | lockdep_assert_held(&kprobe_mutex); |
2128 | ||
e8386a0c | 2129 | p->flags |= KPROBE_FLAG_GONE; |
afd66255 | 2130 | if (kprobe_aggrprobe(p)) { |
e8386a0c MH |
2131 | /* |
2132 | * If this is an aggr_kprobe, we have to list all the | |
2133 | * chained probes and mark them GONE. | |
2134 | */ | |
7e6a71d8 | 2135 | list_for_each_entry(kp, &p->list, list) |
e8386a0c MH |
2136 | kp->flags |= KPROBE_FLAG_GONE; |
2137 | p->post_handler = NULL; | |
afd66255 | 2138 | kill_optimized_kprobe(p); |
e8386a0c MH |
2139 | } |
2140 | /* | |
2141 | * Here, we can remove insn_slot safely, because no thread calls | |
2142 | * the original probed function (which will be freed soon) any more. | |
2143 | */ | |
2144 | arch_remove_kprobe(p); | |
0cb2f137 MS |
2145 | |
2146 | /* | |
2147 | * The module is going away. We should disarm the kprobe which | |
bcb53209 MH |
2148 | * is using ftrace, because ftrace framework is still available at |
2149 | * MODULE_STATE_GOING notification. | |
0cb2f137 | 2150 | */ |
bcb53209 | 2151 | if (kprobe_ftrace(p) && !kprobe_disabled(p) && !kprobes_all_disarmed) |
0cb2f137 | 2152 | disarm_kprobe_ftrace(p); |
e8386a0c MH |
2153 | } |
2154 | ||
c0614829 | 2155 | /* Disable one kprobe */ |
55479f64 | 2156 | int disable_kprobe(struct kprobe *kp) |
c0614829 MH |
2157 | { |
2158 | int ret = 0; | |
297f9233 | 2159 | struct kprobe *p; |
c0614829 MH |
2160 | |
2161 | mutex_lock(&kprobe_mutex); | |
2162 | ||
6f0f1dd7 | 2163 | /* Disable this kprobe */ |
297f9233 JY |
2164 | p = __disable_kprobe(kp); |
2165 | if (IS_ERR(p)) | |
2166 | ret = PTR_ERR(p); | |
c0614829 | 2167 | |
c0614829 MH |
2168 | mutex_unlock(&kprobe_mutex); |
2169 | return ret; | |
2170 | } | |
2171 | EXPORT_SYMBOL_GPL(disable_kprobe); | |
2172 | ||
2173 | /* Enable one kprobe */ | |
55479f64 | 2174 | int enable_kprobe(struct kprobe *kp) |
c0614829 MH |
2175 | { |
2176 | int ret = 0; | |
2177 | struct kprobe *p; | |
2178 | ||
2179 | mutex_lock(&kprobe_mutex); | |
2180 | ||
2181 | /* Check whether specified probe is valid. */ | |
2182 | p = __get_valid_kprobe(kp); | |
2183 | if (unlikely(p == NULL)) { | |
2184 | ret = -EINVAL; | |
2185 | goto out; | |
2186 | } | |
2187 | ||
2188 | if (kprobe_gone(kp)) { | |
2189 | /* This kprobe has gone, we couldn't enable it. */ | |
2190 | ret = -EINVAL; | |
2191 | goto out; | |
2192 | } | |
2193 | ||
2194 | if (p != kp) | |
2195 | kp->flags &= ~KPROBE_FLAG_DISABLED; | |
2196 | ||
2197 | if (!kprobes_all_disarmed && kprobe_disabled(p)) { | |
2198 | p->flags &= ~KPROBE_FLAG_DISABLED; | |
12310e34 JY |
2199 | ret = arm_kprobe(p); |
2200 | if (ret) | |
2201 | p->flags |= KPROBE_FLAG_DISABLED; | |
c0614829 MH |
2202 | } |
2203 | out: | |
2204 | mutex_unlock(&kprobe_mutex); | |
2205 | return ret; | |
2206 | } | |
2207 | EXPORT_SYMBOL_GPL(enable_kprobe); | |
2208 | ||
4458515b | 2209 | /* Caller must NOT call this in usual path. This is only for critical case */ |
820aede0 | 2210 | void dump_kprobe(struct kprobe *kp) |
24851d24 | 2211 | { |
4458515b MH |
2212 | pr_err("Dumping kprobe:\n"); |
2213 | pr_err("Name: %s\nOffset: %x\nAddress: %pS\n", | |
2214 | kp->symbol_name, kp->offset, kp->addr); | |
24851d24 | 2215 | } |
820aede0 | 2216 | NOKPROBE_SYMBOL(dump_kprobe); |
24851d24 | 2217 | |
fb1a59fa MH |
2218 | int kprobe_add_ksym_blacklist(unsigned long entry) |
2219 | { | |
2220 | struct kprobe_blacklist_entry *ent; | |
2221 | unsigned long offset = 0, size = 0; | |
2222 | ||
2223 | if (!kernel_text_address(entry) || | |
2224 | !kallsyms_lookup_size_offset(entry, &size, &offset)) | |
2225 | return -EINVAL; | |
2226 | ||
2227 | ent = kmalloc(sizeof(*ent), GFP_KERNEL); | |
2228 | if (!ent) | |
2229 | return -ENOMEM; | |
2230 | ent->start_addr = entry; | |
2231 | ent->end_addr = entry + size; | |
2232 | INIT_LIST_HEAD(&ent->list); | |
2233 | list_add_tail(&ent->list, &kprobe_blacklist); | |
2234 | ||
2235 | return (int)size; | |
2236 | } | |
2237 | ||
2238 | /* Add all symbols in given area into kprobe blacklist */ | |
2239 | int kprobe_add_area_blacklist(unsigned long start, unsigned long end) | |
2240 | { | |
2241 | unsigned long entry; | |
2242 | int ret = 0; | |
2243 | ||
2244 | for (entry = start; entry < end; entry += ret) { | |
2245 | ret = kprobe_add_ksym_blacklist(entry); | |
2246 | if (ret < 0) | |
2247 | return ret; | |
2248 | if (ret == 0) /* In case of alias symbol */ | |
2249 | ret = 1; | |
2250 | } | |
2251 | return 0; | |
2252 | } | |
2253 | ||
1e6769b0 MH |
2254 | /* Remove all symbols in given area from kprobe blacklist */ |
2255 | static void kprobe_remove_area_blacklist(unsigned long start, unsigned long end) | |
2256 | { | |
2257 | struct kprobe_blacklist_entry *ent, *n; | |
2258 | ||
2259 | list_for_each_entry_safe(ent, n, &kprobe_blacklist, list) { | |
2260 | if (ent->start_addr < start || ent->start_addr >= end) | |
2261 | continue; | |
2262 | list_del(&ent->list); | |
2263 | kfree(ent); | |
2264 | } | |
2265 | } | |
2266 | ||
16db6264 MH |
2267 | static void kprobe_remove_ksym_blacklist(unsigned long entry) |
2268 | { | |
2269 | kprobe_remove_area_blacklist(entry, entry + 1); | |
2270 | } | |
2271 | ||
d002b8bc AH |
2272 | int __weak arch_kprobe_get_kallsym(unsigned int *symnum, unsigned long *value, |
2273 | char *type, char *sym) | |
2274 | { | |
2275 | return -ERANGE; | |
2276 | } | |
2277 | ||
2278 | int kprobe_get_kallsym(unsigned int symnum, unsigned long *value, char *type, | |
2279 | char *sym) | |
2280 | { | |
2281 | #ifdef __ARCH_WANT_KPROBES_INSN_SLOT | |
2282 | if (!kprobe_cache_get_kallsym(&kprobe_insn_slots, &symnum, value, type, sym)) | |
2283 | return 0; | |
2284 | #ifdef CONFIG_OPTPROBES | |
2285 | if (!kprobe_cache_get_kallsym(&kprobe_optinsn_slots, &symnum, value, type, sym)) | |
2286 | return 0; | |
2287 | #endif | |
2288 | #endif | |
2289 | if (!arch_kprobe_get_kallsym(&symnum, value, type, sym)) | |
2290 | return 0; | |
2291 | return -ERANGE; | |
2292 | } | |
2293 | ||
fb1a59fa MH |
2294 | int __init __weak arch_populate_kprobe_blacklist(void) |
2295 | { | |
2296 | return 0; | |
2297 | } | |
2298 | ||
376e2424 MH |
2299 | /* |
2300 | * Lookup and populate the kprobe_blacklist. | |
2301 | * | |
2302 | * Unlike the kretprobe blacklist, we'll need to determine | |
2303 | * the range of addresses that belong to the said functions, | |
2304 | * since a kprobe need not necessarily be at the beginning | |
2305 | * of a function. | |
2306 | */ | |
2307 | static int __init populate_kprobe_blacklist(unsigned long *start, | |
2308 | unsigned long *end) | |
2309 | { | |
fb1a59fa | 2310 | unsigned long entry; |
376e2424 | 2311 | unsigned long *iter; |
fb1a59fa | 2312 | int ret; |
376e2424 MH |
2313 | |
2314 | for (iter = start; iter < end; iter++) { | |
d81b4253 | 2315 | entry = arch_deref_entry_point((void *)*iter); |
fb1a59fa MH |
2316 | ret = kprobe_add_ksym_blacklist(entry); |
2317 | if (ret == -EINVAL) | |
376e2424 | 2318 | continue; |
fb1a59fa MH |
2319 | if (ret < 0) |
2320 | return ret; | |
376e2424 | 2321 | } |
fb1a59fa MH |
2322 | |
2323 | /* Symbols in __kprobes_text are blacklisted */ | |
2324 | ret = kprobe_add_area_blacklist((unsigned long)__kprobes_text_start, | |
2325 | (unsigned long)__kprobes_text_end); | |
66e9b071 TG |
2326 | if (ret) |
2327 | return ret; | |
2328 | ||
2329 | /* Symbols in noinstr section are blacklisted */ | |
2330 | ret = kprobe_add_area_blacklist((unsigned long)__noinstr_text_start, | |
2331 | (unsigned long)__noinstr_text_end); | |
fb1a59fa MH |
2332 | |
2333 | return ret ? : arch_populate_kprobe_blacklist(); | |
376e2424 MH |
2334 | } |
2335 | ||
1e6769b0 MH |
2336 | static void add_module_kprobe_blacklist(struct module *mod) |
2337 | { | |
2338 | unsigned long start, end; | |
16db6264 MH |
2339 | int i; |
2340 | ||
2341 | if (mod->kprobe_blacklist) { | |
2342 | for (i = 0; i < mod->num_kprobe_blacklist; i++) | |
2343 | kprobe_add_ksym_blacklist(mod->kprobe_blacklist[i]); | |
2344 | } | |
1e6769b0 MH |
2345 | |
2346 | start = (unsigned long)mod->kprobes_text_start; | |
2347 | if (start) { | |
2348 | end = start + mod->kprobes_text_size; | |
2349 | kprobe_add_area_blacklist(start, end); | |
2350 | } | |
66e9b071 TG |
2351 | |
2352 | start = (unsigned long)mod->noinstr_text_start; | |
2353 | if (start) { | |
2354 | end = start + mod->noinstr_text_size; | |
2355 | kprobe_add_area_blacklist(start, end); | |
2356 | } | |
1e6769b0 MH |
2357 | } |
2358 | ||
2359 | static void remove_module_kprobe_blacklist(struct module *mod) | |
2360 | { | |
2361 | unsigned long start, end; | |
16db6264 MH |
2362 | int i; |
2363 | ||
2364 | if (mod->kprobe_blacklist) { | |
2365 | for (i = 0; i < mod->num_kprobe_blacklist; i++) | |
2366 | kprobe_remove_ksym_blacklist(mod->kprobe_blacklist[i]); | |
2367 | } | |
1e6769b0 MH |
2368 | |
2369 | start = (unsigned long)mod->kprobes_text_start; | |
2370 | if (start) { | |
2371 | end = start + mod->kprobes_text_size; | |
2372 | kprobe_remove_area_blacklist(start, end); | |
2373 | } | |
66e9b071 TG |
2374 | |
2375 | start = (unsigned long)mod->noinstr_text_start; | |
2376 | if (start) { | |
2377 | end = start + mod->noinstr_text_size; | |
2378 | kprobe_remove_area_blacklist(start, end); | |
2379 | } | |
1e6769b0 MH |
2380 | } |
2381 | ||
e8386a0c | 2382 | /* Module notifier call back, checking kprobes on the module */ |
55479f64 MH |
2383 | static int kprobes_module_callback(struct notifier_block *nb, |
2384 | unsigned long val, void *data) | |
e8386a0c MH |
2385 | { |
2386 | struct module *mod = data; | |
2387 | struct hlist_head *head; | |
e8386a0c MH |
2388 | struct kprobe *p; |
2389 | unsigned int i; | |
f24659d9 | 2390 | int checkcore = (val == MODULE_STATE_GOING); |
e8386a0c | 2391 | |
1e6769b0 MH |
2392 | if (val == MODULE_STATE_COMING) { |
2393 | mutex_lock(&kprobe_mutex); | |
2394 | add_module_kprobe_blacklist(mod); | |
2395 | mutex_unlock(&kprobe_mutex); | |
2396 | } | |
f24659d9 | 2397 | if (val != MODULE_STATE_GOING && val != MODULE_STATE_LIVE) |
e8386a0c MH |
2398 | return NOTIFY_DONE; |
2399 | ||
2400 | /* | |
f24659d9 MH |
2401 | * When MODULE_STATE_GOING was notified, both of module .text and |
2402 | * .init.text sections would be freed. When MODULE_STATE_LIVE was | |
2403 | * notified, only .init.text section would be freed. We need to | |
2404 | * disable kprobes which have been inserted in the sections. | |
e8386a0c MH |
2405 | */ |
2406 | mutex_lock(&kprobe_mutex); | |
2407 | for (i = 0; i < KPROBE_TABLE_SIZE; i++) { | |
2408 | head = &kprobe_table[i]; | |
7e6a71d8 | 2409 | hlist_for_each_entry(p, head, hlist) |
f24659d9 MH |
2410 | if (within_module_init((unsigned long)p->addr, mod) || |
2411 | (checkcore && | |
2412 | within_module_core((unsigned long)p->addr, mod))) { | |
e8386a0c MH |
2413 | /* |
2414 | * The vaddr this probe is installed will soon | |
2415 | * be vfreed buy not synced to disk. Hence, | |
2416 | * disarming the breakpoint isn't needed. | |
545a0281 SRV |
2417 | * |
2418 | * Note, this will also move any optimized probes | |
2419 | * that are pending to be removed from their | |
2420 | * corresponding lists to the freeing_list and | |
2421 | * will not be touched by the delayed | |
2422 | * kprobe_optimizer work handler. | |
e8386a0c MH |
2423 | */ |
2424 | kill_kprobe(p); | |
2425 | } | |
2426 | } | |
1e6769b0 MH |
2427 | if (val == MODULE_STATE_GOING) |
2428 | remove_module_kprobe_blacklist(mod); | |
e8386a0c MH |
2429 | mutex_unlock(&kprobe_mutex); |
2430 | return NOTIFY_DONE; | |
2431 | } | |
2432 | ||
2433 | static struct notifier_block kprobe_module_nb = { | |
2434 | .notifier_call = kprobes_module_callback, | |
2435 | .priority = 0 | |
2436 | }; | |
2437 | ||
376e2424 MH |
2438 | /* Markers of _kprobe_blacklist section */ |
2439 | extern unsigned long __start_kprobe_blacklist[]; | |
2440 | extern unsigned long __stop_kprobe_blacklist[]; | |
2441 | ||
82d083ab MH |
2442 | void kprobe_free_init_mem(void) |
2443 | { | |
2444 | void *start = (void *)(&__init_begin); | |
2445 | void *end = (void *)(&__init_end); | |
2446 | struct hlist_head *head; | |
2447 | struct kprobe *p; | |
2448 | int i; | |
2449 | ||
2450 | mutex_lock(&kprobe_mutex); | |
2451 | ||
2452 | /* Kill all kprobes on initmem */ | |
2453 | for (i = 0; i < KPROBE_TABLE_SIZE; i++) { | |
2454 | head = &kprobe_table[i]; | |
2455 | hlist_for_each_entry(p, head, hlist) { | |
2456 | if (start <= (void *)p->addr && (void *)p->addr < end) | |
2457 | kill_kprobe(p); | |
2458 | } | |
2459 | } | |
2460 | ||
2461 | mutex_unlock(&kprobe_mutex); | |
2462 | } | |
2463 | ||
1da177e4 LT |
2464 | static int __init init_kprobes(void) |
2465 | { | |
2466 | int i, err = 0; | |
2467 | ||
2468 | /* FIXME allocate the probe table, currently defined statically */ | |
2469 | /* initialize all list heads */ | |
d741bf41 | 2470 | for (i = 0; i < KPROBE_TABLE_SIZE; i++) |
1da177e4 LT |
2471 | INIT_HLIST_HEAD(&kprobe_table[i]); |
2472 | ||
376e2424 MH |
2473 | err = populate_kprobe_blacklist(__start_kprobe_blacklist, |
2474 | __stop_kprobe_blacklist); | |
2475 | if (err) { | |
2476 | pr_err("kprobes: failed to populate blacklist: %d\n", err); | |
2477 | pr_err("Please take care of using kprobes.\n"); | |
3d8d996e SD |
2478 | } |
2479 | ||
f438d914 MH |
2480 | if (kretprobe_blacklist_size) { |
2481 | /* lookup the function address from its name */ | |
2482 | for (i = 0; kretprobe_blacklist[i].name != NULL; i++) { | |
49e0b465 | 2483 | kretprobe_blacklist[i].addr = |
290e3070 | 2484 | kprobe_lookup_name(kretprobe_blacklist[i].name, 0); |
f438d914 MH |
2485 | if (!kretprobe_blacklist[i].addr) |
2486 | printk("kretprobe: lookup failed: %s\n", | |
2487 | kretprobe_blacklist[i].name); | |
2488 | } | |
2489 | } | |
2490 | ||
e579abeb MH |
2491 | /* By default, kprobes are armed */ |
2492 | kprobes_all_disarmed = false; | |
bf8f6e5b | 2493 | |
c85c9a2c MH |
2494 | #if defined(CONFIG_OPTPROBES) && defined(__ARCH_WANT_KPROBES_INSN_SLOT) |
2495 | /* Init kprobe_optinsn_slots for allocation */ | |
2496 | kprobe_optinsn_slots.insn_size = MAX_OPTINSN_SIZE; | |
2497 | #endif | |
2498 | ||
6772926b | 2499 | err = arch_init_kprobes(); |
802eae7c RL |
2500 | if (!err) |
2501 | err = register_die_notifier(&kprobe_exceptions_nb); | |
e8386a0c MH |
2502 | if (!err) |
2503 | err = register_module_notifier(&kprobe_module_nb); | |
2504 | ||
ef53d9c5 | 2505 | kprobes_initialized = (err == 0); |
802eae7c | 2506 | |
8c1c9356 AM |
2507 | if (!err) |
2508 | init_test_probes(); | |
1da177e4 LT |
2509 | return err; |
2510 | } | |
36dadef2 | 2511 | early_initcall(init_kprobes); |
1da177e4 | 2512 | |
c85c9a2c MH |
2513 | #if defined(CONFIG_OPTPROBES) |
2514 | static int __init init_optprobes(void) | |
2515 | { | |
2516 | /* | |
2517 | * Enable kprobe optimization - this kicks the optimizer which | |
2518 | * depends on synchronize_rcu_tasks() and ksoftirqd, that is | |
2519 | * not spawned in early initcall. So delay the optimization. | |
2520 | */ | |
2521 | optimize_all_kprobes(); | |
2522 | ||
2523 | return 0; | |
2524 | } | |
2525 | subsys_initcall(init_optprobes); | |
2526 | #endif | |
2527 | ||
346fd59b | 2528 | #ifdef CONFIG_DEBUG_FS |
55479f64 | 2529 | static void report_probe(struct seq_file *pi, struct kprobe *p, |
afd66255 | 2530 | const char *sym, int offset, char *modname, struct kprobe *pp) |
346fd59b SD |
2531 | { |
2532 | char *kprobe_type; | |
81365a94 | 2533 | void *addr = p->addr; |
346fd59b SD |
2534 | |
2535 | if (p->pre_handler == pre_handler_kretprobe) | |
2536 | kprobe_type = "r"; | |
346fd59b SD |
2537 | else |
2538 | kprobe_type = "k"; | |
afd66255 | 2539 | |
60f7bb66 | 2540 | if (!kallsyms_show_value(pi->file->f_cred)) |
81365a94 MH |
2541 | addr = NULL; |
2542 | ||
346fd59b | 2543 | if (sym) |
81365a94 MH |
2544 | seq_printf(pi, "%px %s %s+0x%x %s ", |
2545 | addr, kprobe_type, sym, offset, | |
afd66255 | 2546 | (modname ? modname : " ")); |
81365a94 MH |
2547 | else /* try to use %pS */ |
2548 | seq_printf(pi, "%px %s %pS ", | |
2549 | addr, kprobe_type, p->addr); | |
afd66255 MH |
2550 | |
2551 | if (!pp) | |
2552 | pp = p; | |
ae6aa16f | 2553 | seq_printf(pi, "%s%s%s%s\n", |
afd66255 MH |
2554 | (kprobe_gone(p) ? "[GONE]" : ""), |
2555 | ((kprobe_disabled(p) && !kprobe_gone(p)) ? "[DISABLED]" : ""), | |
ae6aa16f MH |
2556 | (kprobe_optimized(pp) ? "[OPTIMIZED]" : ""), |
2557 | (kprobe_ftrace(pp) ? "[FTRACE]" : "")); | |
346fd59b SD |
2558 | } |
2559 | ||
55479f64 | 2560 | static void *kprobe_seq_start(struct seq_file *f, loff_t *pos) |
346fd59b SD |
2561 | { |
2562 | return (*pos < KPROBE_TABLE_SIZE) ? pos : NULL; | |
2563 | } | |
2564 | ||
55479f64 | 2565 | static void *kprobe_seq_next(struct seq_file *f, void *v, loff_t *pos) |
346fd59b SD |
2566 | { |
2567 | (*pos)++; | |
2568 | if (*pos >= KPROBE_TABLE_SIZE) | |
2569 | return NULL; | |
2570 | return pos; | |
2571 | } | |
2572 | ||
55479f64 | 2573 | static void kprobe_seq_stop(struct seq_file *f, void *v) |
346fd59b SD |
2574 | { |
2575 | /* Nothing to do */ | |
2576 | } | |
2577 | ||
55479f64 | 2578 | static int show_kprobe_addr(struct seq_file *pi, void *v) |
346fd59b SD |
2579 | { |
2580 | struct hlist_head *head; | |
346fd59b SD |
2581 | struct kprobe *p, *kp; |
2582 | const char *sym = NULL; | |
2583 | unsigned int i = *(loff_t *) v; | |
ffb45122 | 2584 | unsigned long offset = 0; |
ab767865 | 2585 | char *modname, namebuf[KSYM_NAME_LEN]; |
346fd59b SD |
2586 | |
2587 | head = &kprobe_table[i]; | |
2588 | preempt_disable(); | |
b67bfe0d | 2589 | hlist_for_each_entry_rcu(p, head, hlist) { |
ffb45122 | 2590 | sym = kallsyms_lookup((unsigned long)p->addr, NULL, |
346fd59b | 2591 | &offset, &modname, namebuf); |
afd66255 | 2592 | if (kprobe_aggrprobe(p)) { |
346fd59b | 2593 | list_for_each_entry_rcu(kp, &p->list, list) |
afd66255 | 2594 | report_probe(pi, kp, sym, offset, modname, p); |
346fd59b | 2595 | } else |
afd66255 | 2596 | report_probe(pi, p, sym, offset, modname, NULL); |
346fd59b SD |
2597 | } |
2598 | preempt_enable(); | |
2599 | return 0; | |
2600 | } | |
2601 | ||
eac2cece | 2602 | static const struct seq_operations kprobes_sops = { |
346fd59b SD |
2603 | .start = kprobe_seq_start, |
2604 | .next = kprobe_seq_next, | |
2605 | .stop = kprobe_seq_stop, | |
2606 | .show = show_kprobe_addr | |
2607 | }; | |
2608 | ||
eac2cece | 2609 | DEFINE_SEQ_ATTRIBUTE(kprobes); |
346fd59b | 2610 | |
63724740 MH |
2611 | /* kprobes/blacklist -- shows which functions can not be probed */ |
2612 | static void *kprobe_blacklist_seq_start(struct seq_file *m, loff_t *pos) | |
2613 | { | |
4fdd8887 | 2614 | mutex_lock(&kprobe_mutex); |
63724740 MH |
2615 | return seq_list_start(&kprobe_blacklist, *pos); |
2616 | } | |
2617 | ||
2618 | static void *kprobe_blacklist_seq_next(struct seq_file *m, void *v, loff_t *pos) | |
2619 | { | |
2620 | return seq_list_next(v, &kprobe_blacklist, pos); | |
2621 | } | |
2622 | ||
2623 | static int kprobe_blacklist_seq_show(struct seq_file *m, void *v) | |
2624 | { | |
2625 | struct kprobe_blacklist_entry *ent = | |
2626 | list_entry(v, struct kprobe_blacklist_entry, list); | |
2627 | ||
ffb9bd68 MH |
2628 | /* |
2629 | * If /proc/kallsyms is not showing kernel address, we won't | |
2630 | * show them here either. | |
2631 | */ | |
60f7bb66 | 2632 | if (!kallsyms_show_value(m->file->f_cred)) |
ffb9bd68 MH |
2633 | seq_printf(m, "0x%px-0x%px\t%ps\n", NULL, NULL, |
2634 | (void *)ent->start_addr); | |
2635 | else | |
2636 | seq_printf(m, "0x%px-0x%px\t%ps\n", (void *)ent->start_addr, | |
2637 | (void *)ent->end_addr, (void *)ent->start_addr); | |
63724740 MH |
2638 | return 0; |
2639 | } | |
2640 | ||
4fdd8887 MH |
2641 | static void kprobe_blacklist_seq_stop(struct seq_file *f, void *v) |
2642 | { | |
2643 | mutex_unlock(&kprobe_mutex); | |
2644 | } | |
2645 | ||
eac2cece | 2646 | static const struct seq_operations kprobe_blacklist_sops = { |
63724740 MH |
2647 | .start = kprobe_blacklist_seq_start, |
2648 | .next = kprobe_blacklist_seq_next, | |
4fdd8887 | 2649 | .stop = kprobe_blacklist_seq_stop, |
63724740 MH |
2650 | .show = kprobe_blacklist_seq_show, |
2651 | }; | |
eac2cece | 2652 | DEFINE_SEQ_ATTRIBUTE(kprobe_blacklist); |
63724740 | 2653 | |
12310e34 | 2654 | static int arm_all_kprobes(void) |
bf8f6e5b AM |
2655 | { |
2656 | struct hlist_head *head; | |
bf8f6e5b | 2657 | struct kprobe *p; |
12310e34 JY |
2658 | unsigned int i, total = 0, errors = 0; |
2659 | int err, ret = 0; | |
bf8f6e5b AM |
2660 | |
2661 | mutex_lock(&kprobe_mutex); | |
2662 | ||
e579abeb MH |
2663 | /* If kprobes are armed, just return */ |
2664 | if (!kprobes_all_disarmed) | |
bf8f6e5b AM |
2665 | goto already_enabled; |
2666 | ||
977ad481 WN |
2667 | /* |
2668 | * optimize_kprobe() called by arm_kprobe() checks | |
2669 | * kprobes_all_disarmed, so set kprobes_all_disarmed before | |
2670 | * arm_kprobe. | |
2671 | */ | |
2672 | kprobes_all_disarmed = false; | |
afd66255 | 2673 | /* Arming kprobes doesn't optimize kprobe itself */ |
bf8f6e5b AM |
2674 | for (i = 0; i < KPROBE_TABLE_SIZE; i++) { |
2675 | head = &kprobe_table[i]; | |
12310e34 | 2676 | /* Arm all kprobes on a best-effort basis */ |
7e6a71d8 | 2677 | hlist_for_each_entry(p, head, hlist) { |
12310e34 JY |
2678 | if (!kprobe_disabled(p)) { |
2679 | err = arm_kprobe(p); | |
2680 | if (err) { | |
2681 | errors++; | |
2682 | ret = err; | |
2683 | } | |
2684 | total++; | |
2685 | } | |
2686 | } | |
bf8f6e5b AM |
2687 | } |
2688 | ||
12310e34 JY |
2689 | if (errors) |
2690 | pr_warn("Kprobes globally enabled, but failed to arm %d out of %d probes\n", | |
2691 | errors, total); | |
2692 | else | |
2693 | pr_info("Kprobes globally enabled\n"); | |
bf8f6e5b AM |
2694 | |
2695 | already_enabled: | |
2696 | mutex_unlock(&kprobe_mutex); | |
12310e34 | 2697 | return ret; |
bf8f6e5b AM |
2698 | } |
2699 | ||
297f9233 | 2700 | static int disarm_all_kprobes(void) |
bf8f6e5b AM |
2701 | { |
2702 | struct hlist_head *head; | |
bf8f6e5b | 2703 | struct kprobe *p; |
297f9233 JY |
2704 | unsigned int i, total = 0, errors = 0; |
2705 | int err, ret = 0; | |
bf8f6e5b AM |
2706 | |
2707 | mutex_lock(&kprobe_mutex); | |
2708 | ||
e579abeb | 2709 | /* If kprobes are already disarmed, just return */ |
6274de49 MH |
2710 | if (kprobes_all_disarmed) { |
2711 | mutex_unlock(&kprobe_mutex); | |
297f9233 | 2712 | return 0; |
6274de49 | 2713 | } |
bf8f6e5b | 2714 | |
e579abeb | 2715 | kprobes_all_disarmed = true; |
afd66255 | 2716 | |
bf8f6e5b AM |
2717 | for (i = 0; i < KPROBE_TABLE_SIZE; i++) { |
2718 | head = &kprobe_table[i]; | |
297f9233 | 2719 | /* Disarm all kprobes on a best-effort basis */ |
7e6a71d8 | 2720 | hlist_for_each_entry(p, head, hlist) { |
297f9233 JY |
2721 | if (!arch_trampoline_kprobe(p) && !kprobe_disabled(p)) { |
2722 | err = disarm_kprobe(p, false); | |
2723 | if (err) { | |
2724 | errors++; | |
2725 | ret = err; | |
2726 | } | |
2727 | total++; | |
2728 | } | |
bf8f6e5b AM |
2729 | } |
2730 | } | |
297f9233 JY |
2731 | |
2732 | if (errors) | |
2733 | pr_warn("Kprobes globally disabled, but failed to disarm %d out of %d probes\n", | |
2734 | errors, total); | |
2735 | else | |
2736 | pr_info("Kprobes globally disabled\n"); | |
2737 | ||
bf8f6e5b | 2738 | mutex_unlock(&kprobe_mutex); |
bf8f6e5b | 2739 | |
6274de49 MH |
2740 | /* Wait for disarming all kprobes by optimizer */ |
2741 | wait_for_kprobe_optimizer(); | |
297f9233 JY |
2742 | |
2743 | return ret; | |
bf8f6e5b AM |
2744 | } |
2745 | ||
2746 | /* | |
2747 | * XXX: The debugfs bool file interface doesn't allow for callbacks | |
2748 | * when the bool state is switched. We can reuse that facility when | |
2749 | * available | |
2750 | */ | |
2751 | static ssize_t read_enabled_file_bool(struct file *file, | |
2752 | char __user *user_buf, size_t count, loff_t *ppos) | |
2753 | { | |
2754 | char buf[3]; | |
2755 | ||
e579abeb | 2756 | if (!kprobes_all_disarmed) |
bf8f6e5b AM |
2757 | buf[0] = '1'; |
2758 | else | |
2759 | buf[0] = '0'; | |
2760 | buf[1] = '\n'; | |
2761 | buf[2] = 0x00; | |
2762 | return simple_read_from_buffer(user_buf, count, ppos, buf, 2); | |
2763 | } | |
2764 | ||
2765 | static ssize_t write_enabled_file_bool(struct file *file, | |
2766 | const char __user *user_buf, size_t count, loff_t *ppos) | |
2767 | { | |
5d6de7d7 PA |
2768 | bool enable; |
2769 | int ret; | |
bf8f6e5b | 2770 | |
5d6de7d7 PA |
2771 | ret = kstrtobool_from_user(user_buf, count, &enable); |
2772 | if (ret) | |
2773 | return ret; | |
bf8f6e5b | 2774 | |
5d6de7d7 | 2775 | ret = enable ? arm_all_kprobes() : disarm_all_kprobes(); |
12310e34 JY |
2776 | if (ret) |
2777 | return ret; | |
2778 | ||
bf8f6e5b AM |
2779 | return count; |
2780 | } | |
2781 | ||
828c0950 | 2782 | static const struct file_operations fops_kp = { |
bf8f6e5b AM |
2783 | .read = read_enabled_file_bool, |
2784 | .write = write_enabled_file_bool, | |
6038f373 | 2785 | .llseek = default_llseek, |
bf8f6e5b AM |
2786 | }; |
2787 | ||
55479f64 | 2788 | static int __init debugfs_kprobe_init(void) |
346fd59b | 2789 | { |
8c0fd1fa | 2790 | struct dentry *dir; |
346fd59b SD |
2791 | |
2792 | dir = debugfs_create_dir("kprobes", NULL); | |
346fd59b | 2793 | |
eac2cece | 2794 | debugfs_create_file("list", 0400, dir, NULL, &kprobes_fops); |
346fd59b | 2795 | |
8f7262cd | 2796 | debugfs_create_file("enabled", 0600, dir, NULL, &fops_kp); |
63724740 | 2797 | |
8c0fd1fa | 2798 | debugfs_create_file("blacklist", 0400, dir, NULL, |
eac2cece | 2799 | &kprobe_blacklist_fops); |
bf8f6e5b | 2800 | |
346fd59b SD |
2801 | return 0; |
2802 | } | |
2803 | ||
2804 | late_initcall(debugfs_kprobe_init); | |
2805 | #endif /* CONFIG_DEBUG_FS */ |