]>
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> |
bf8f6e5b | 38 | |
bfd45be0 | 39 | #include <asm/sections.h> |
1da177e4 LT |
40 | #include <asm/cacheflush.h> |
41 | #include <asm/errno.h> | |
7c0f6ba6 | 42 | #include <linux/uaccess.h> |
1da177e4 LT |
43 | |
44 | #define KPROBE_HASH_BITS 6 | |
45 | #define KPROBE_TABLE_SIZE (1 << KPROBE_HASH_BITS) | |
46 | ||
3a872d89 | 47 | |
ef53d9c5 | 48 | static int kprobes_initialized; |
1da177e4 | 49 | static struct hlist_head kprobe_table[KPROBE_TABLE_SIZE]; |
b94cce92 | 50 | static struct hlist_head kretprobe_inst_table[KPROBE_TABLE_SIZE]; |
1da177e4 | 51 | |
bf8f6e5b | 52 | /* NOTE: change this value only with kprobe_mutex held */ |
e579abeb | 53 | static bool kprobes_all_disarmed; |
bf8f6e5b | 54 | |
43948f50 MH |
55 | /* This protects kprobe_table and optimizing_list */ |
56 | static DEFINE_MUTEX(kprobe_mutex); | |
e6584523 | 57 | static DEFINE_PER_CPU(struct kprobe *, kprobe_instance) = NULL; |
ef53d9c5 | 58 | static struct { |
ec484608 | 59 | raw_spinlock_t lock ____cacheline_aligned_in_smp; |
ef53d9c5 S |
60 | } kretprobe_table_locks[KPROBE_TABLE_SIZE]; |
61 | ||
290e3070 NR |
62 | kprobe_opcode_t * __weak kprobe_lookup_name(const char *name, |
63 | unsigned int __unused) | |
49e0b465 NR |
64 | { |
65 | return ((kprobe_opcode_t *)(kallsyms_lookup_name(name))); | |
66 | } | |
67 | ||
ec484608 | 68 | static raw_spinlock_t *kretprobe_table_lock_ptr(unsigned long hash) |
ef53d9c5 S |
69 | { |
70 | return &(kretprobe_table_locks[hash].lock); | |
71 | } | |
1da177e4 | 72 | |
376e2424 MH |
73 | /* Blacklist -- list of struct kprobe_blacklist_entry */ |
74 | static LIST_HEAD(kprobe_blacklist); | |
3d8d996e | 75 | |
2d14e39d | 76 | #ifdef __ARCH_WANT_KPROBES_INSN_SLOT |
9ec4b1f3 AM |
77 | /* |
78 | * kprobe->ainsn.insn points to the copy of the instruction to be | |
79 | * single-stepped. x86_64, POWER4 and above have no-exec support and | |
80 | * stepping on the instruction on a vmalloced/kmalloced/data page | |
81 | * is a recipe for disaster | |
82 | */ | |
9ec4b1f3 | 83 | struct kprobe_insn_page { |
c5cb5a2d | 84 | struct list_head list; |
9ec4b1f3 | 85 | kprobe_opcode_t *insns; /* Page of instruction slots */ |
af96397d | 86 | struct kprobe_insn_cache *cache; |
9ec4b1f3 | 87 | int nused; |
b4c6c34a | 88 | int ngarbage; |
4610ee1d | 89 | char slot_used[]; |
9ec4b1f3 AM |
90 | }; |
91 | ||
4610ee1d MH |
92 | #define KPROBE_INSN_PAGE_SIZE(slots) \ |
93 | (offsetof(struct kprobe_insn_page, slot_used) + \ | |
94 | (sizeof(char) * (slots))) | |
95 | ||
4610ee1d MH |
96 | static int slots_per_page(struct kprobe_insn_cache *c) |
97 | { | |
98 | return PAGE_SIZE/(c->insn_size * sizeof(kprobe_opcode_t)); | |
99 | } | |
100 | ||
ab40c5c6 MH |
101 | enum kprobe_slot_state { |
102 | SLOT_CLEAN = 0, | |
103 | SLOT_DIRTY = 1, | |
104 | SLOT_USED = 2, | |
105 | }; | |
106 | ||
63fef14f | 107 | void __weak *alloc_insn_page(void) |
af96397d HC |
108 | { |
109 | return module_alloc(PAGE_SIZE); | |
110 | } | |
111 | ||
c93f5cf5 | 112 | void __weak free_insn_page(void *page) |
af96397d | 113 | { |
be1f221c | 114 | module_memfree(page); |
af96397d HC |
115 | } |
116 | ||
c802d64a HC |
117 | struct kprobe_insn_cache kprobe_insn_slots = { |
118 | .mutex = __MUTEX_INITIALIZER(kprobe_insn_slots.mutex), | |
af96397d HC |
119 | .alloc = alloc_insn_page, |
120 | .free = free_insn_page, | |
4610ee1d MH |
121 | .pages = LIST_HEAD_INIT(kprobe_insn_slots.pages), |
122 | .insn_size = MAX_INSN_SIZE, | |
123 | .nr_garbage = 0, | |
124 | }; | |
55479f64 | 125 | static int collect_garbage_slots(struct kprobe_insn_cache *c); |
b4c6c34a | 126 | |
9ec4b1f3 | 127 | /** |
12941560 | 128 | * __get_insn_slot() - Find a slot on an executable page for an instruction. |
9ec4b1f3 AM |
129 | * We allocate an executable page if there's no room on existing ones. |
130 | */ | |
55479f64 | 131 | kprobe_opcode_t *__get_insn_slot(struct kprobe_insn_cache *c) |
9ec4b1f3 AM |
132 | { |
133 | struct kprobe_insn_page *kip; | |
c802d64a | 134 | kprobe_opcode_t *slot = NULL; |
9ec4b1f3 | 135 | |
5b485629 | 136 | /* Since the slot array is not protected by rcu, we need a mutex */ |
c802d64a | 137 | mutex_lock(&c->mutex); |
6f716acd | 138 | retry: |
5b485629 MH |
139 | rcu_read_lock(); |
140 | list_for_each_entry_rcu(kip, &c->pages, list) { | |
4610ee1d | 141 | if (kip->nused < slots_per_page(c)) { |
9ec4b1f3 | 142 | int i; |
4610ee1d | 143 | for (i = 0; i < slots_per_page(c); i++) { |
ab40c5c6 MH |
144 | if (kip->slot_used[i] == SLOT_CLEAN) { |
145 | kip->slot_used[i] = SLOT_USED; | |
9ec4b1f3 | 146 | kip->nused++; |
c802d64a | 147 | slot = kip->insns + (i * c->insn_size); |
5b485629 | 148 | rcu_read_unlock(); |
c802d64a | 149 | goto out; |
9ec4b1f3 AM |
150 | } |
151 | } | |
4610ee1d MH |
152 | /* kip->nused is broken. Fix it. */ |
153 | kip->nused = slots_per_page(c); | |
154 | WARN_ON(1); | |
9ec4b1f3 AM |
155 | } |
156 | } | |
5b485629 | 157 | rcu_read_unlock(); |
9ec4b1f3 | 158 | |
b4c6c34a | 159 | /* If there are any garbage slots, collect it and try again. */ |
4610ee1d | 160 | if (c->nr_garbage && collect_garbage_slots(c) == 0) |
b4c6c34a | 161 | goto retry; |
4610ee1d MH |
162 | |
163 | /* All out of space. Need to allocate a new page. */ | |
164 | kip = kmalloc(KPROBE_INSN_PAGE_SIZE(slots_per_page(c)), GFP_KERNEL); | |
6f716acd | 165 | if (!kip) |
c802d64a | 166 | goto out; |
9ec4b1f3 AM |
167 | |
168 | /* | |
169 | * Use module_alloc so this page is within +/- 2GB of where the | |
170 | * kernel image and loaded module images reside. This is required | |
171 | * so x86_64 can correctly handle the %rip-relative fixups. | |
172 | */ | |
af96397d | 173 | kip->insns = c->alloc(); |
9ec4b1f3 AM |
174 | if (!kip->insns) { |
175 | kfree(kip); | |
c802d64a | 176 | goto out; |
9ec4b1f3 | 177 | } |
c5cb5a2d | 178 | INIT_LIST_HEAD(&kip->list); |
4610ee1d | 179 | memset(kip->slot_used, SLOT_CLEAN, slots_per_page(c)); |
ab40c5c6 | 180 | kip->slot_used[0] = SLOT_USED; |
9ec4b1f3 | 181 | kip->nused = 1; |
b4c6c34a | 182 | kip->ngarbage = 0; |
af96397d | 183 | kip->cache = c; |
5b485629 | 184 | list_add_rcu(&kip->list, &c->pages); |
c802d64a HC |
185 | slot = kip->insns; |
186 | out: | |
187 | mutex_unlock(&c->mutex); | |
188 | return slot; | |
12941560 MH |
189 | } |
190 | ||
b4c6c34a | 191 | /* Return 1 if all garbages are collected, otherwise 0. */ |
55479f64 | 192 | static int collect_one_slot(struct kprobe_insn_page *kip, int idx) |
b4c6c34a | 193 | { |
ab40c5c6 | 194 | kip->slot_used[idx] = SLOT_CLEAN; |
b4c6c34a MH |
195 | kip->nused--; |
196 | if (kip->nused == 0) { | |
197 | /* | |
198 | * Page is no longer in use. Free it unless | |
199 | * it's the last one. We keep the last one | |
200 | * so as not to have to set it up again the | |
201 | * next time somebody inserts a probe. | |
202 | */ | |
4610ee1d | 203 | if (!list_is_singular(&kip->list)) { |
5b485629 MH |
204 | list_del_rcu(&kip->list); |
205 | synchronize_rcu(); | |
af96397d | 206 | kip->cache->free(kip->insns); |
b4c6c34a MH |
207 | kfree(kip); |
208 | } | |
209 | return 1; | |
210 | } | |
211 | return 0; | |
212 | } | |
213 | ||
55479f64 | 214 | static int collect_garbage_slots(struct kprobe_insn_cache *c) |
b4c6c34a | 215 | { |
c5cb5a2d | 216 | struct kprobe_insn_page *kip, *next; |
b4c6c34a | 217 | |
615d0ebb | 218 | /* Ensure no-one is interrupted on the garbages */ |
ae8b7ce7 | 219 | synchronize_rcu(); |
b4c6c34a | 220 | |
4610ee1d | 221 | list_for_each_entry_safe(kip, next, &c->pages, list) { |
b4c6c34a | 222 | int i; |
b4c6c34a MH |
223 | if (kip->ngarbage == 0) |
224 | continue; | |
225 | kip->ngarbage = 0; /* we will collect all garbages */ | |
4610ee1d | 226 | for (i = 0; i < slots_per_page(c); i++) { |
5b485629 | 227 | if (kip->slot_used[i] == SLOT_DIRTY && collect_one_slot(kip, i)) |
b4c6c34a MH |
228 | break; |
229 | } | |
230 | } | |
4610ee1d | 231 | c->nr_garbage = 0; |
b4c6c34a MH |
232 | return 0; |
233 | } | |
234 | ||
55479f64 MH |
235 | void __free_insn_slot(struct kprobe_insn_cache *c, |
236 | kprobe_opcode_t *slot, int dirty) | |
9ec4b1f3 AM |
237 | { |
238 | struct kprobe_insn_page *kip; | |
5b485629 | 239 | long idx; |
9ec4b1f3 | 240 | |
c802d64a | 241 | mutex_lock(&c->mutex); |
5b485629 MH |
242 | rcu_read_lock(); |
243 | list_for_each_entry_rcu(kip, &c->pages, list) { | |
244 | idx = ((long)slot - (long)kip->insns) / | |
245 | (c->insn_size * sizeof(kprobe_opcode_t)); | |
246 | if (idx >= 0 && idx < slots_per_page(c)) | |
c802d64a | 247 | goto out; |
9ec4b1f3 | 248 | } |
5b485629 | 249 | /* Could not find this slot. */ |
4610ee1d | 250 | WARN_ON(1); |
5b485629 | 251 | kip = NULL; |
c802d64a | 252 | out: |
5b485629 MH |
253 | rcu_read_unlock(); |
254 | /* Mark and sweep: this may sleep */ | |
255 | if (kip) { | |
256 | /* Check double free */ | |
257 | WARN_ON(kip->slot_used[idx] != SLOT_USED); | |
258 | if (dirty) { | |
259 | kip->slot_used[idx] = SLOT_DIRTY; | |
260 | kip->ngarbage++; | |
261 | if (++c->nr_garbage > slots_per_page(c)) | |
262 | collect_garbage_slots(c); | |
263 | } else { | |
264 | collect_one_slot(kip, idx); | |
265 | } | |
266 | } | |
c802d64a | 267 | mutex_unlock(&c->mutex); |
4610ee1d | 268 | } |
6f716acd | 269 | |
5b485629 MH |
270 | /* |
271 | * Check given address is on the page of kprobe instruction slots. | |
272 | * This will be used for checking whether the address on a stack | |
273 | * is on a text area or not. | |
274 | */ | |
275 | bool __is_insn_slot_addr(struct kprobe_insn_cache *c, unsigned long addr) | |
276 | { | |
277 | struct kprobe_insn_page *kip; | |
278 | bool ret = false; | |
279 | ||
280 | rcu_read_lock(); | |
281 | list_for_each_entry_rcu(kip, &c->pages, list) { | |
282 | if (addr >= (unsigned long)kip->insns && | |
283 | addr < (unsigned long)kip->insns + PAGE_SIZE) { | |
284 | ret = true; | |
285 | break; | |
286 | } | |
287 | } | |
288 | rcu_read_unlock(); | |
289 | ||
290 | return ret; | |
291 | } | |
292 | ||
afd66255 MH |
293 | #ifdef CONFIG_OPTPROBES |
294 | /* For optimized_kprobe buffer */ | |
c802d64a HC |
295 | struct kprobe_insn_cache kprobe_optinsn_slots = { |
296 | .mutex = __MUTEX_INITIALIZER(kprobe_optinsn_slots.mutex), | |
af96397d HC |
297 | .alloc = alloc_insn_page, |
298 | .free = free_insn_page, | |
afd66255 MH |
299 | .pages = LIST_HEAD_INIT(kprobe_optinsn_slots.pages), |
300 | /* .insn_size is initialized later */ | |
301 | .nr_garbage = 0, | |
302 | }; | |
afd66255 | 303 | #endif |
2d14e39d | 304 | #endif |
9ec4b1f3 | 305 | |
e6584523 AM |
306 | /* We have preemption disabled.. so it is safe to use __ versions */ |
307 | static inline void set_kprobe_instance(struct kprobe *kp) | |
308 | { | |
b76834bc | 309 | __this_cpu_write(kprobe_instance, kp); |
e6584523 AM |
310 | } |
311 | ||
312 | static inline void reset_kprobe_instance(void) | |
313 | { | |
b76834bc | 314 | __this_cpu_write(kprobe_instance, NULL); |
e6584523 AM |
315 | } |
316 | ||
3516a460 AM |
317 | /* |
318 | * This routine is called either: | |
49a2a1b8 | 319 | * - under the kprobe_mutex - during kprobe_[un]register() |
3516a460 | 320 | * OR |
d217d545 | 321 | * - with preemption disabled - from arch/xxx/kernel/kprobes.c |
3516a460 | 322 | */ |
820aede0 | 323 | struct kprobe *get_kprobe(void *addr) |
1da177e4 LT |
324 | { |
325 | struct hlist_head *head; | |
3516a460 | 326 | struct kprobe *p; |
1da177e4 LT |
327 | |
328 | head = &kprobe_table[hash_ptr(addr, KPROBE_HASH_BITS)]; | |
b67bfe0d | 329 | hlist_for_each_entry_rcu(p, head, hlist) { |
1da177e4 LT |
330 | if (p->addr == addr) |
331 | return p; | |
332 | } | |
afd66255 | 333 | |
1da177e4 LT |
334 | return NULL; |
335 | } | |
820aede0 | 336 | NOKPROBE_SYMBOL(get_kprobe); |
1da177e4 | 337 | |
820aede0 | 338 | static int aggr_pre_handler(struct kprobe *p, struct pt_regs *regs); |
afd66255 MH |
339 | |
340 | /* Return true if the kprobe is an aggregator */ | |
341 | static inline int kprobe_aggrprobe(struct kprobe *p) | |
342 | { | |
343 | return p->pre_handler == aggr_pre_handler; | |
344 | } | |
345 | ||
6274de49 MH |
346 | /* Return true(!0) if the kprobe is unused */ |
347 | static inline int kprobe_unused(struct kprobe *p) | |
348 | { | |
349 | return kprobe_aggrprobe(p) && kprobe_disabled(p) && | |
350 | list_empty(&p->list); | |
351 | } | |
352 | ||
afd66255 MH |
353 | /* |
354 | * Keep all fields in the kprobe consistent | |
355 | */ | |
6d8e40a8 | 356 | static inline void copy_kprobe(struct kprobe *ap, struct kprobe *p) |
afd66255 | 357 | { |
6d8e40a8 MH |
358 | memcpy(&p->opcode, &ap->opcode, sizeof(kprobe_opcode_t)); |
359 | memcpy(&p->ainsn, &ap->ainsn, sizeof(struct arch_specific_insn)); | |
afd66255 MH |
360 | } |
361 | ||
362 | #ifdef CONFIG_OPTPROBES | |
b2be84df MH |
363 | /* NOTE: change this value only with kprobe_mutex held */ |
364 | static bool kprobes_allow_optimization; | |
365 | ||
afd66255 MH |
366 | /* |
367 | * Call all pre_handler on the list, but ignores its return value. | |
368 | * This must be called from arch-dep optimized caller. | |
369 | */ | |
820aede0 | 370 | void opt_pre_handler(struct kprobe *p, struct pt_regs *regs) |
afd66255 MH |
371 | { |
372 | struct kprobe *kp; | |
373 | ||
374 | list_for_each_entry_rcu(kp, &p->list, list) { | |
375 | if (kp->pre_handler && likely(!kprobe_disabled(kp))) { | |
376 | set_kprobe_instance(kp); | |
4f3a8714 | 377 | kp->pre_handler(kp, regs); |
afd66255 MH |
378 | } |
379 | reset_kprobe_instance(); | |
380 | } | |
381 | } | |
820aede0 | 382 | NOKPROBE_SYMBOL(opt_pre_handler); |
afd66255 | 383 | |
6274de49 | 384 | /* Free optimized instructions and optimized_kprobe */ |
55479f64 | 385 | static void free_aggr_kprobe(struct kprobe *p) |
6274de49 MH |
386 | { |
387 | struct optimized_kprobe *op; | |
388 | ||
389 | op = container_of(p, struct optimized_kprobe, kp); | |
390 | arch_remove_optimized_kprobe(op); | |
391 | arch_remove_kprobe(p); | |
392 | kfree(op); | |
393 | } | |
394 | ||
afd66255 MH |
395 | /* Return true(!0) if the kprobe is ready for optimization. */ |
396 | static inline int kprobe_optready(struct kprobe *p) | |
397 | { | |
398 | struct optimized_kprobe *op; | |
399 | ||
400 | if (kprobe_aggrprobe(p)) { | |
401 | op = container_of(p, struct optimized_kprobe, kp); | |
402 | return arch_prepared_optinsn(&op->optinsn); | |
403 | } | |
404 | ||
405 | return 0; | |
406 | } | |
407 | ||
6274de49 MH |
408 | /* Return true(!0) if the kprobe is disarmed. Note: p must be on hash list */ |
409 | static inline int kprobe_disarmed(struct kprobe *p) | |
410 | { | |
411 | struct optimized_kprobe *op; | |
412 | ||
413 | /* If kprobe is not aggr/opt probe, just return kprobe is disabled */ | |
414 | if (!kprobe_aggrprobe(p)) | |
415 | return kprobe_disabled(p); | |
416 | ||
417 | op = container_of(p, struct optimized_kprobe, kp); | |
418 | ||
419 | return kprobe_disabled(p) && list_empty(&op->list); | |
420 | } | |
421 | ||
422 | /* Return true(!0) if the probe is queued on (un)optimizing lists */ | |
55479f64 | 423 | static int kprobe_queued(struct kprobe *p) |
6274de49 MH |
424 | { |
425 | struct optimized_kprobe *op; | |
426 | ||
427 | if (kprobe_aggrprobe(p)) { | |
428 | op = container_of(p, struct optimized_kprobe, kp); | |
429 | if (!list_empty(&op->list)) | |
430 | return 1; | |
431 | } | |
432 | return 0; | |
433 | } | |
434 | ||
afd66255 MH |
435 | /* |
436 | * Return an optimized kprobe whose optimizing code replaces | |
437 | * instructions including addr (exclude breakpoint). | |
438 | */ | |
55479f64 | 439 | static struct kprobe *get_optimized_kprobe(unsigned long addr) |
afd66255 MH |
440 | { |
441 | int i; | |
442 | struct kprobe *p = NULL; | |
443 | struct optimized_kprobe *op; | |
444 | ||
445 | /* Don't check i == 0, since that is a breakpoint case. */ | |
446 | for (i = 1; !p && i < MAX_OPTIMIZED_LENGTH; i++) | |
447 | p = get_kprobe((void *)(addr - i)); | |
448 | ||
449 | if (p && kprobe_optready(p)) { | |
450 | op = container_of(p, struct optimized_kprobe, kp); | |
451 | if (arch_within_optimized_kprobe(op, addr)) | |
452 | return p; | |
453 | } | |
454 | ||
455 | return NULL; | |
456 | } | |
457 | ||
458 | /* Optimization staging list, protected by kprobe_mutex */ | |
459 | static LIST_HEAD(optimizing_list); | |
6274de49 | 460 | static LIST_HEAD(unoptimizing_list); |
7b959fc5 | 461 | static LIST_HEAD(freeing_list); |
afd66255 MH |
462 | |
463 | static void kprobe_optimizer(struct work_struct *work); | |
464 | static DECLARE_DELAYED_WORK(optimizing_work, kprobe_optimizer); | |
465 | #define OPTIMIZE_DELAY 5 | |
466 | ||
61f4e13f MH |
467 | /* |
468 | * Optimize (replace a breakpoint with a jump) kprobes listed on | |
469 | * optimizing_list. | |
470 | */ | |
55479f64 | 471 | static void do_optimize_kprobes(void) |
afd66255 | 472 | { |
f1c6ece2 | 473 | lockdep_assert_held(&text_mutex); |
afd66255 MH |
474 | /* |
475 | * The optimization/unoptimization refers online_cpus via | |
476 | * stop_machine() and cpu-hotplug modifies online_cpus. | |
477 | * And same time, text_mutex will be held in cpu-hotplug and here. | |
478 | * This combination can cause a deadlock (cpu-hotplug try to lock | |
479 | * text_mutex but stop_machine can not be done because online_cpus | |
480 | * has been changed) | |
2d1e38f5 | 481 | * To avoid this deadlock, caller must have locked cpu hotplug |
afd66255 MH |
482 | * for preventing cpu-hotplug outside of text_mutex locking. |
483 | */ | |
2d1e38f5 TG |
484 | lockdep_assert_cpus_held(); |
485 | ||
486 | /* Optimization never be done when disarmed */ | |
487 | if (kprobes_all_disarmed || !kprobes_allow_optimization || | |
488 | list_empty(&optimizing_list)) | |
489 | return; | |
490 | ||
cd7ebe22 | 491 | arch_optimize_kprobes(&optimizing_list); |
61f4e13f MH |
492 | } |
493 | ||
6274de49 MH |
494 | /* |
495 | * Unoptimize (replace a jump with a breakpoint and remove the breakpoint | |
496 | * if need) kprobes listed on unoptimizing_list. | |
497 | */ | |
55479f64 | 498 | static void do_unoptimize_kprobes(void) |
6274de49 MH |
499 | { |
500 | struct optimized_kprobe *op, *tmp; | |
501 | ||
f1c6ece2 | 502 | lockdep_assert_held(&text_mutex); |
2d1e38f5 TG |
503 | /* See comment in do_optimize_kprobes() */ |
504 | lockdep_assert_cpus_held(); | |
505 | ||
6274de49 MH |
506 | /* Unoptimization must be done anytime */ |
507 | if (list_empty(&unoptimizing_list)) | |
508 | return; | |
509 | ||
7b959fc5 | 510 | arch_unoptimize_kprobes(&unoptimizing_list, &freeing_list); |
f984ba4e | 511 | /* Loop free_list for disarming */ |
7b959fc5 | 512 | list_for_each_entry_safe(op, tmp, &freeing_list, list) { |
f66c0447 MH |
513 | /* Switching from detour code to origin */ |
514 | op->kp.flags &= ~KPROBE_FLAG_OPTIMIZED; | |
6274de49 MH |
515 | /* Disarm probes if marked disabled */ |
516 | if (kprobe_disabled(&op->kp)) | |
517 | arch_disarm_kprobe(&op->kp); | |
518 | if (kprobe_unused(&op->kp)) { | |
519 | /* | |
520 | * Remove unused probes from hash list. After waiting | |
521 | * for synchronization, these probes are reclaimed. | |
522 | * (reclaiming is done by do_free_cleaned_kprobes.) | |
523 | */ | |
524 | hlist_del_rcu(&op->kp.hlist); | |
6274de49 MH |
525 | } else |
526 | list_del_init(&op->list); | |
527 | } | |
6274de49 MH |
528 | } |
529 | ||
530 | /* Reclaim all kprobes on the free_list */ | |
55479f64 | 531 | static void do_free_cleaned_kprobes(void) |
6274de49 MH |
532 | { |
533 | struct optimized_kprobe *op, *tmp; | |
534 | ||
7b959fc5 | 535 | list_for_each_entry_safe(op, tmp, &freeing_list, list) { |
6274de49 | 536 | list_del_init(&op->list); |
cbdd96f5 MH |
537 | if (WARN_ON_ONCE(!kprobe_unused(&op->kp))) { |
538 | /* | |
539 | * This must not happen, but if there is a kprobe | |
540 | * still in use, keep it on kprobes hash list. | |
541 | */ | |
542 | continue; | |
543 | } | |
6274de49 MH |
544 | free_aggr_kprobe(&op->kp); |
545 | } | |
546 | } | |
547 | ||
548 | /* Start optimizer after OPTIMIZE_DELAY passed */ | |
55479f64 | 549 | static void kick_kprobe_optimizer(void) |
6274de49 | 550 | { |
ad72b3be | 551 | schedule_delayed_work(&optimizing_work, OPTIMIZE_DELAY); |
6274de49 MH |
552 | } |
553 | ||
61f4e13f | 554 | /* Kprobe jump optimizer */ |
55479f64 | 555 | static void kprobe_optimizer(struct work_struct *work) |
61f4e13f | 556 | { |
72ef3794 | 557 | mutex_lock(&kprobe_mutex); |
2d1e38f5 | 558 | cpus_read_lock(); |
f1c6ece2 | 559 | mutex_lock(&text_mutex); |
61f4e13f MH |
560 | /* Lock modules while optimizing kprobes */ |
561 | mutex_lock(&module_mutex); | |
61f4e13f MH |
562 | |
563 | /* | |
6274de49 MH |
564 | * Step 1: Unoptimize kprobes and collect cleaned (unused and disarmed) |
565 | * kprobes before waiting for quiesence period. | |
566 | */ | |
7b959fc5 | 567 | do_unoptimize_kprobes(); |
6274de49 MH |
568 | |
569 | /* | |
a30b85df MH |
570 | * Step 2: Wait for quiesence period to ensure all potentially |
571 | * preempted tasks to have normally scheduled. Because optprobe | |
572 | * may modify multiple instructions, there is a chance that Nth | |
573 | * instruction is preempted. In that case, such tasks can return | |
574 | * to 2nd-Nth byte of jump instruction. This wait is for avoiding it. | |
575 | * Note that on non-preemptive kernel, this is transparently converted | |
576 | * to synchronoze_sched() to wait for all interrupts to have completed. | |
61f4e13f | 577 | */ |
a30b85df | 578 | synchronize_rcu_tasks(); |
61f4e13f | 579 | |
6274de49 | 580 | /* Step 3: Optimize kprobes after quiesence period */ |
61f4e13f | 581 | do_optimize_kprobes(); |
6274de49 MH |
582 | |
583 | /* Step 4: Free cleaned kprobes after quiesence period */ | |
7b959fc5 | 584 | do_free_cleaned_kprobes(); |
6274de49 | 585 | |
afd66255 | 586 | mutex_unlock(&module_mutex); |
f1c6ece2 | 587 | mutex_unlock(&text_mutex); |
2d1e38f5 | 588 | cpus_read_unlock(); |
72ef3794 | 589 | mutex_unlock(&kprobe_mutex); |
6274de49 | 590 | |
cd7ebe22 | 591 | /* Step 5: Kick optimizer again if needed */ |
f984ba4e | 592 | if (!list_empty(&optimizing_list) || !list_empty(&unoptimizing_list)) |
cd7ebe22 | 593 | kick_kprobe_optimizer(); |
6274de49 MH |
594 | } |
595 | ||
596 | /* Wait for completing optimization and unoptimization */ | |
30e7d894 | 597 | void wait_for_kprobe_optimizer(void) |
6274de49 | 598 | { |
ad72b3be TH |
599 | mutex_lock(&kprobe_mutex); |
600 | ||
601 | while (!list_empty(&optimizing_list) || !list_empty(&unoptimizing_list)) { | |
602 | mutex_unlock(&kprobe_mutex); | |
603 | ||
604 | /* this will also make optimizing_work execute immmediately */ | |
605 | flush_delayed_work(&optimizing_work); | |
606 | /* @optimizing_work might not have been queued yet, relax */ | |
607 | cpu_relax(); | |
608 | ||
609 | mutex_lock(&kprobe_mutex); | |
610 | } | |
611 | ||
612 | mutex_unlock(&kprobe_mutex); | |
afd66255 MH |
613 | } |
614 | ||
615 | /* Optimize kprobe if p is ready to be optimized */ | |
55479f64 | 616 | static void optimize_kprobe(struct kprobe *p) |
afd66255 MH |
617 | { |
618 | struct optimized_kprobe *op; | |
619 | ||
620 | /* Check if the kprobe is disabled or not ready for optimization. */ | |
b2be84df | 621 | if (!kprobe_optready(p) || !kprobes_allow_optimization || |
afd66255 MH |
622 | (kprobe_disabled(p) || kprobes_all_disarmed)) |
623 | return; | |
624 | ||
059053a2 MH |
625 | /* kprobes with post_handler can not be optimized */ |
626 | if (p->post_handler) | |
afd66255 MH |
627 | return; |
628 | ||
629 | op = container_of(p, struct optimized_kprobe, kp); | |
630 | ||
631 | /* Check there is no other kprobes at the optimized instructions */ | |
632 | if (arch_check_optimized_kprobe(op) < 0) | |
633 | return; | |
634 | ||
635 | /* Check if it is already optimized. */ | |
636 | if (op->kp.flags & KPROBE_FLAG_OPTIMIZED) | |
637 | return; | |
afd66255 | 638 | op->kp.flags |= KPROBE_FLAG_OPTIMIZED; |
6274de49 MH |
639 | |
640 | if (!list_empty(&op->list)) | |
641 | /* This is under unoptimizing. Just dequeue the probe */ | |
642 | list_del_init(&op->list); | |
643 | else { | |
644 | list_add(&op->list, &optimizing_list); | |
645 | kick_kprobe_optimizer(); | |
646 | } | |
647 | } | |
648 | ||
649 | /* Short cut to direct unoptimizing */ | |
55479f64 | 650 | static void force_unoptimize_kprobe(struct optimized_kprobe *op) |
6274de49 | 651 | { |
2d1e38f5 | 652 | lockdep_assert_cpus_held(); |
6274de49 | 653 | arch_unoptimize_kprobe(op); |
f66c0447 | 654 | op->kp.flags &= ~KPROBE_FLAG_OPTIMIZED; |
6274de49 MH |
655 | if (kprobe_disabled(&op->kp)) |
656 | arch_disarm_kprobe(&op->kp); | |
afd66255 MH |
657 | } |
658 | ||
659 | /* Unoptimize a kprobe if p is optimized */ | |
55479f64 | 660 | static void unoptimize_kprobe(struct kprobe *p, bool force) |
afd66255 MH |
661 | { |
662 | struct optimized_kprobe *op; | |
663 | ||
6274de49 MH |
664 | if (!kprobe_aggrprobe(p) || kprobe_disarmed(p)) |
665 | return; /* This is not an optprobe nor optimized */ | |
666 | ||
667 | op = container_of(p, struct optimized_kprobe, kp); | |
668 | if (!kprobe_optimized(p)) { | |
669 | /* Unoptimized or unoptimizing case */ | |
670 | if (force && !list_empty(&op->list)) { | |
671 | /* | |
672 | * Only if this is unoptimizing kprobe and forced, | |
673 | * forcibly unoptimize it. (No need to unoptimize | |
674 | * unoptimized kprobe again :) | |
675 | */ | |
afd66255 | 676 | list_del_init(&op->list); |
6274de49 MH |
677 | force_unoptimize_kprobe(op); |
678 | } | |
679 | return; | |
680 | } | |
681 | ||
6274de49 MH |
682 | if (!list_empty(&op->list)) { |
683 | /* Dequeue from the optimization queue */ | |
684 | list_del_init(&op->list); | |
685 | return; | |
686 | } | |
687 | /* Optimized kprobe case */ | |
688 | if (force) | |
689 | /* Forcibly update the code: this is a special case */ | |
690 | force_unoptimize_kprobe(op); | |
691 | else { | |
692 | list_add(&op->list, &unoptimizing_list); | |
693 | kick_kprobe_optimizer(); | |
afd66255 MH |
694 | } |
695 | } | |
696 | ||
0490cd1f | 697 | /* Cancel unoptimizing for reusing */ |
819319fc | 698 | static int reuse_unused_kprobe(struct kprobe *ap) |
0490cd1f MH |
699 | { |
700 | struct optimized_kprobe *op; | |
701 | ||
0490cd1f MH |
702 | /* |
703 | * Unused kprobe MUST be on the way of delayed unoptimizing (means | |
704 | * there is still a relative jump) and disabled. | |
705 | */ | |
706 | op = container_of(ap, struct optimized_kprobe, kp); | |
4458515b | 707 | WARN_ON_ONCE(list_empty(&op->list)); |
0490cd1f MH |
708 | /* Enable the probe again */ |
709 | ap->flags &= ~KPROBE_FLAG_DISABLED; | |
710 | /* Optimize it again (remove from op->list) */ | |
5f843ed4 MH |
711 | if (!kprobe_optready(ap)) |
712 | return -EINVAL; | |
819319fc | 713 | |
0490cd1f | 714 | optimize_kprobe(ap); |
819319fc | 715 | return 0; |
0490cd1f MH |
716 | } |
717 | ||
afd66255 | 718 | /* Remove optimized instructions */ |
55479f64 | 719 | static void kill_optimized_kprobe(struct kprobe *p) |
afd66255 MH |
720 | { |
721 | struct optimized_kprobe *op; | |
722 | ||
723 | op = container_of(p, struct optimized_kprobe, kp); | |
6274de49 MH |
724 | if (!list_empty(&op->list)) |
725 | /* Dequeue from the (un)optimization queue */ | |
afd66255 | 726 | list_del_init(&op->list); |
6274de49 | 727 | op->kp.flags &= ~KPROBE_FLAG_OPTIMIZED; |
7b959fc5 MH |
728 | |
729 | if (kprobe_unused(p)) { | |
730 | /* Enqueue if it is unused */ | |
731 | list_add(&op->list, &freeing_list); | |
732 | /* | |
733 | * Remove unused probes from the hash list. After waiting | |
734 | * for synchronization, this probe is reclaimed. | |
735 | * (reclaiming is done by do_free_cleaned_kprobes().) | |
736 | */ | |
737 | hlist_del_rcu(&op->kp.hlist); | |
738 | } | |
739 | ||
6274de49 | 740 | /* Don't touch the code, because it is already freed. */ |
afd66255 MH |
741 | arch_remove_optimized_kprobe(op); |
742 | } | |
743 | ||
a460246c MH |
744 | static inline |
745 | void __prepare_optimized_kprobe(struct optimized_kprobe *op, struct kprobe *p) | |
746 | { | |
747 | if (!kprobe_ftrace(p)) | |
748 | arch_prepare_optimized_kprobe(op, p); | |
749 | } | |
750 | ||
afd66255 | 751 | /* Try to prepare optimized instructions */ |
55479f64 | 752 | static void prepare_optimized_kprobe(struct kprobe *p) |
afd66255 MH |
753 | { |
754 | struct optimized_kprobe *op; | |
755 | ||
756 | op = container_of(p, struct optimized_kprobe, kp); | |
a460246c | 757 | __prepare_optimized_kprobe(op, p); |
afd66255 MH |
758 | } |
759 | ||
afd66255 | 760 | /* Allocate new optimized_kprobe and try to prepare optimized instructions */ |
55479f64 | 761 | static struct kprobe *alloc_aggr_kprobe(struct kprobe *p) |
afd66255 MH |
762 | { |
763 | struct optimized_kprobe *op; | |
764 | ||
765 | op = kzalloc(sizeof(struct optimized_kprobe), GFP_KERNEL); | |
766 | if (!op) | |
767 | return NULL; | |
768 | ||
769 | INIT_LIST_HEAD(&op->list); | |
770 | op->kp.addr = p->addr; | |
a460246c | 771 | __prepare_optimized_kprobe(op, p); |
afd66255 MH |
772 | |
773 | return &op->kp; | |
774 | } | |
775 | ||
55479f64 | 776 | static void init_aggr_kprobe(struct kprobe *ap, struct kprobe *p); |
afd66255 MH |
777 | |
778 | /* | |
779 | * Prepare an optimized_kprobe and optimize it | |
780 | * NOTE: p must be a normal registered kprobe | |
781 | */ | |
55479f64 | 782 | static void try_to_optimize_kprobe(struct kprobe *p) |
afd66255 MH |
783 | { |
784 | struct kprobe *ap; | |
785 | struct optimized_kprobe *op; | |
786 | ||
ae6aa16f MH |
787 | /* Impossible to optimize ftrace-based kprobe */ |
788 | if (kprobe_ftrace(p)) | |
789 | return; | |
790 | ||
25764288 | 791 | /* For preparing optimization, jump_label_text_reserved() is called */ |
2d1e38f5 | 792 | cpus_read_lock(); |
25764288 MH |
793 | jump_label_lock(); |
794 | mutex_lock(&text_mutex); | |
795 | ||
afd66255 MH |
796 | ap = alloc_aggr_kprobe(p); |
797 | if (!ap) | |
25764288 | 798 | goto out; |
afd66255 MH |
799 | |
800 | op = container_of(ap, struct optimized_kprobe, kp); | |
801 | if (!arch_prepared_optinsn(&op->optinsn)) { | |
802 | /* If failed to setup optimizing, fallback to kprobe */ | |
6274de49 MH |
803 | arch_remove_optimized_kprobe(op); |
804 | kfree(op); | |
25764288 | 805 | goto out; |
afd66255 MH |
806 | } |
807 | ||
808 | init_aggr_kprobe(ap, p); | |
25764288 MH |
809 | optimize_kprobe(ap); /* This just kicks optimizer thread */ |
810 | ||
811 | out: | |
812 | mutex_unlock(&text_mutex); | |
813 | jump_label_unlock(); | |
2d1e38f5 | 814 | cpus_read_unlock(); |
afd66255 MH |
815 | } |
816 | ||
b2be84df | 817 | #ifdef CONFIG_SYSCTL |
55479f64 | 818 | static void optimize_all_kprobes(void) |
b2be84df MH |
819 | { |
820 | struct hlist_head *head; | |
b2be84df MH |
821 | struct kprobe *p; |
822 | unsigned int i; | |
823 | ||
5c51543b | 824 | mutex_lock(&kprobe_mutex); |
b2be84df MH |
825 | /* If optimization is already allowed, just return */ |
826 | if (kprobes_allow_optimization) | |
5c51543b | 827 | goto out; |
b2be84df | 828 | |
2d1e38f5 | 829 | cpus_read_lock(); |
b2be84df | 830 | kprobes_allow_optimization = true; |
b2be84df MH |
831 | for (i = 0; i < KPROBE_TABLE_SIZE; i++) { |
832 | head = &kprobe_table[i]; | |
b67bfe0d | 833 | hlist_for_each_entry_rcu(p, head, hlist) |
b2be84df MH |
834 | if (!kprobe_disabled(p)) |
835 | optimize_kprobe(p); | |
836 | } | |
2d1e38f5 | 837 | cpus_read_unlock(); |
b2be84df | 838 | printk(KERN_INFO "Kprobes globally optimized\n"); |
5c51543b MH |
839 | out: |
840 | mutex_unlock(&kprobe_mutex); | |
b2be84df MH |
841 | } |
842 | ||
55479f64 | 843 | static void unoptimize_all_kprobes(void) |
b2be84df MH |
844 | { |
845 | struct hlist_head *head; | |
b2be84df MH |
846 | struct kprobe *p; |
847 | unsigned int i; | |
848 | ||
5c51543b | 849 | mutex_lock(&kprobe_mutex); |
b2be84df | 850 | /* If optimization is already prohibited, just return */ |
5c51543b MH |
851 | if (!kprobes_allow_optimization) { |
852 | mutex_unlock(&kprobe_mutex); | |
b2be84df | 853 | return; |
5c51543b | 854 | } |
b2be84df | 855 | |
2d1e38f5 | 856 | cpus_read_lock(); |
b2be84df | 857 | kprobes_allow_optimization = false; |
b2be84df MH |
858 | for (i = 0; i < KPROBE_TABLE_SIZE; i++) { |
859 | head = &kprobe_table[i]; | |
b67bfe0d | 860 | hlist_for_each_entry_rcu(p, head, hlist) { |
b2be84df | 861 | if (!kprobe_disabled(p)) |
6274de49 | 862 | unoptimize_kprobe(p, false); |
b2be84df MH |
863 | } |
864 | } | |
2d1e38f5 | 865 | cpus_read_unlock(); |
5c51543b MH |
866 | mutex_unlock(&kprobe_mutex); |
867 | ||
6274de49 MH |
868 | /* Wait for unoptimizing completion */ |
869 | wait_for_kprobe_optimizer(); | |
870 | printk(KERN_INFO "Kprobes globally unoptimized\n"); | |
b2be84df MH |
871 | } |
872 | ||
5c51543b | 873 | static DEFINE_MUTEX(kprobe_sysctl_mutex); |
b2be84df MH |
874 | int sysctl_kprobes_optimization; |
875 | int proc_kprobes_optimization_handler(struct ctl_table *table, int write, | |
876 | void __user *buffer, size_t *length, | |
877 | loff_t *ppos) | |
878 | { | |
879 | int ret; | |
880 | ||
5c51543b | 881 | mutex_lock(&kprobe_sysctl_mutex); |
b2be84df MH |
882 | sysctl_kprobes_optimization = kprobes_allow_optimization ? 1 : 0; |
883 | ret = proc_dointvec_minmax(table, write, buffer, length, ppos); | |
884 | ||
885 | if (sysctl_kprobes_optimization) | |
886 | optimize_all_kprobes(); | |
887 | else | |
888 | unoptimize_all_kprobes(); | |
5c51543b | 889 | mutex_unlock(&kprobe_sysctl_mutex); |
b2be84df MH |
890 | |
891 | return ret; | |
892 | } | |
893 | #endif /* CONFIG_SYSCTL */ | |
894 | ||
6274de49 | 895 | /* Put a breakpoint for a probe. Must be called with text_mutex locked */ |
55479f64 | 896 | static void __arm_kprobe(struct kprobe *p) |
afd66255 | 897 | { |
6d8e40a8 | 898 | struct kprobe *_p; |
afd66255 MH |
899 | |
900 | /* Check collision with other optimized kprobes */ | |
6d8e40a8 MH |
901 | _p = get_optimized_kprobe((unsigned long)p->addr); |
902 | if (unlikely(_p)) | |
6274de49 MH |
903 | /* Fallback to unoptimized kprobe */ |
904 | unoptimize_kprobe(_p, true); | |
afd66255 MH |
905 | |
906 | arch_arm_kprobe(p); | |
907 | optimize_kprobe(p); /* Try to optimize (add kprobe to a list) */ | |
908 | } | |
909 | ||
6274de49 | 910 | /* Remove the breakpoint of a probe. Must be called with text_mutex locked */ |
55479f64 | 911 | static void __disarm_kprobe(struct kprobe *p, bool reopt) |
afd66255 | 912 | { |
6d8e40a8 | 913 | struct kprobe *_p; |
afd66255 | 914 | |
69d54b91 WN |
915 | /* Try to unoptimize */ |
916 | unoptimize_kprobe(p, kprobes_all_disarmed); | |
afd66255 | 917 | |
6274de49 MH |
918 | if (!kprobe_queued(p)) { |
919 | arch_disarm_kprobe(p); | |
920 | /* If another kprobe was blocked, optimize it. */ | |
921 | _p = get_optimized_kprobe((unsigned long)p->addr); | |
922 | if (unlikely(_p) && reopt) | |
923 | optimize_kprobe(_p); | |
924 | } | |
925 | /* TODO: reoptimize others after unoptimized this probe */ | |
afd66255 MH |
926 | } |
927 | ||
928 | #else /* !CONFIG_OPTPROBES */ | |
929 | ||
930 | #define optimize_kprobe(p) do {} while (0) | |
6274de49 | 931 | #define unoptimize_kprobe(p, f) do {} while (0) |
afd66255 MH |
932 | #define kill_optimized_kprobe(p) do {} while (0) |
933 | #define prepare_optimized_kprobe(p) do {} while (0) | |
934 | #define try_to_optimize_kprobe(p) do {} while (0) | |
935 | #define __arm_kprobe(p) arch_arm_kprobe(p) | |
6274de49 MH |
936 | #define __disarm_kprobe(p, o) arch_disarm_kprobe(p) |
937 | #define kprobe_disarmed(p) kprobe_disabled(p) | |
938 | #define wait_for_kprobe_optimizer() do {} while (0) | |
afd66255 | 939 | |
819319fc | 940 | static int reuse_unused_kprobe(struct kprobe *ap) |
0490cd1f | 941 | { |
819319fc MH |
942 | /* |
943 | * If the optimized kprobe is NOT supported, the aggr kprobe is | |
944 | * released at the same time that the last aggregated kprobe is | |
945 | * unregistered. | |
946 | * Thus there should be no chance to reuse unused kprobe. | |
947 | */ | |
0490cd1f | 948 | printk(KERN_ERR "Error: There should be no unused kprobe here.\n"); |
819319fc | 949 | return -EINVAL; |
0490cd1f MH |
950 | } |
951 | ||
55479f64 | 952 | static void free_aggr_kprobe(struct kprobe *p) |
afd66255 | 953 | { |
6274de49 | 954 | arch_remove_kprobe(p); |
afd66255 MH |
955 | kfree(p); |
956 | } | |
957 | ||
55479f64 | 958 | static struct kprobe *alloc_aggr_kprobe(struct kprobe *p) |
afd66255 MH |
959 | { |
960 | return kzalloc(sizeof(struct kprobe), GFP_KERNEL); | |
961 | } | |
962 | #endif /* CONFIG_OPTPROBES */ | |
963 | ||
e7dbfe34 | 964 | #ifdef CONFIG_KPROBES_ON_FTRACE |
ae6aa16f | 965 | static struct ftrace_ops kprobe_ftrace_ops __read_mostly = { |
0bc11ed5 MH |
966 | .func = kprobe_ftrace_handler, |
967 | .flags = FTRACE_OPS_FL_SAVE_REGS, | |
968 | }; | |
969 | ||
970 | static struct ftrace_ops kprobe_ipmodify_ops __read_mostly = { | |
e5253896 | 971 | .func = kprobe_ftrace_handler, |
1d70be34 | 972 | .flags = FTRACE_OPS_FL_SAVE_REGS | FTRACE_OPS_FL_IPMODIFY, |
ae6aa16f | 973 | }; |
0bc11ed5 MH |
974 | |
975 | static int kprobe_ipmodify_enabled; | |
ae6aa16f MH |
976 | static int kprobe_ftrace_enabled; |
977 | ||
978 | /* Must ensure p->addr is really on ftrace */ | |
55479f64 | 979 | static int prepare_kprobe(struct kprobe *p) |
ae6aa16f MH |
980 | { |
981 | if (!kprobe_ftrace(p)) | |
982 | return arch_prepare_kprobe(p); | |
983 | ||
984 | return arch_prepare_kprobe_ftrace(p); | |
985 | } | |
986 | ||
987 | /* Caller must lock kprobe_mutex */ | |
0bc11ed5 MH |
988 | static int __arm_kprobe_ftrace(struct kprobe *p, struct ftrace_ops *ops, |
989 | int *cnt) | |
ae6aa16f | 990 | { |
12310e34 | 991 | int ret = 0; |
ae6aa16f | 992 | |
0bc11ed5 | 993 | ret = ftrace_set_filter_ip(ops, (unsigned long)p->addr, 0, 0); |
12310e34 | 994 | if (ret) { |
4458515b MH |
995 | pr_debug("Failed to arm kprobe-ftrace at %pS (%d)\n", |
996 | p->addr, ret); | |
12310e34 JY |
997 | return ret; |
998 | } | |
999 | ||
0bc11ed5 MH |
1000 | if (*cnt == 0) { |
1001 | ret = register_ftrace_function(ops); | |
12310e34 JY |
1002 | if (ret) { |
1003 | pr_debug("Failed to init kprobe-ftrace (%d)\n", ret); | |
1004 | goto err_ftrace; | |
1005 | } | |
ae6aa16f | 1006 | } |
12310e34 | 1007 | |
0bc11ed5 | 1008 | (*cnt)++; |
12310e34 JY |
1009 | return ret; |
1010 | ||
1011 | err_ftrace: | |
1012 | /* | |
0bc11ed5 MH |
1013 | * At this point, sinec ops is not registered, we should be sefe from |
1014 | * registering empty filter. | |
12310e34 | 1015 | */ |
0bc11ed5 | 1016 | ftrace_set_filter_ip(ops, (unsigned long)p->addr, 1, 0); |
12310e34 | 1017 | return ret; |
ae6aa16f MH |
1018 | } |
1019 | ||
0bc11ed5 MH |
1020 | static int arm_kprobe_ftrace(struct kprobe *p) |
1021 | { | |
1022 | bool ipmodify = (p->post_handler != NULL); | |
1023 | ||
1024 | return __arm_kprobe_ftrace(p, | |
1025 | ipmodify ? &kprobe_ipmodify_ops : &kprobe_ftrace_ops, | |
1026 | ipmodify ? &kprobe_ipmodify_enabled : &kprobe_ftrace_enabled); | |
1027 | } | |
1028 | ||
ae6aa16f | 1029 | /* Caller must lock kprobe_mutex */ |
0bc11ed5 MH |
1030 | static int __disarm_kprobe_ftrace(struct kprobe *p, struct ftrace_ops *ops, |
1031 | int *cnt) | |
ae6aa16f | 1032 | { |
297f9233 | 1033 | int ret = 0; |
ae6aa16f | 1034 | |
0bc11ed5 MH |
1035 | if (*cnt == 1) { |
1036 | ret = unregister_ftrace_function(ops); | |
297f9233 JY |
1037 | if (WARN(ret < 0, "Failed to unregister kprobe-ftrace (%d)\n", ret)) |
1038 | return ret; | |
ae6aa16f | 1039 | } |
297f9233 | 1040 | |
0bc11ed5 | 1041 | (*cnt)--; |
297f9233 | 1042 | |
0bc11ed5 | 1043 | ret = ftrace_set_filter_ip(ops, (unsigned long)p->addr, 1, 0); |
4458515b MH |
1044 | WARN_ONCE(ret < 0, "Failed to disarm kprobe-ftrace at %pS (%d)\n", |
1045 | p->addr, ret); | |
297f9233 | 1046 | return ret; |
ae6aa16f | 1047 | } |
0bc11ed5 MH |
1048 | |
1049 | static int disarm_kprobe_ftrace(struct kprobe *p) | |
1050 | { | |
1051 | bool ipmodify = (p->post_handler != NULL); | |
1052 | ||
1053 | return __disarm_kprobe_ftrace(p, | |
1054 | ipmodify ? &kprobe_ipmodify_ops : &kprobe_ftrace_ops, | |
1055 | ipmodify ? &kprobe_ipmodify_enabled : &kprobe_ftrace_enabled); | |
1056 | } | |
e7dbfe34 | 1057 | #else /* !CONFIG_KPROBES_ON_FTRACE */ |
ae6aa16f | 1058 | #define prepare_kprobe(p) arch_prepare_kprobe(p) |
12310e34 | 1059 | #define arm_kprobe_ftrace(p) (-ENODEV) |
297f9233 | 1060 | #define disarm_kprobe_ftrace(p) (-ENODEV) |
ae6aa16f MH |
1061 | #endif |
1062 | ||
201517a7 | 1063 | /* Arm a kprobe with text_mutex */ |
12310e34 | 1064 | static int arm_kprobe(struct kprobe *kp) |
201517a7 | 1065 | { |
12310e34 JY |
1066 | if (unlikely(kprobe_ftrace(kp))) |
1067 | return arm_kprobe_ftrace(kp); | |
1068 | ||
2d1e38f5 | 1069 | cpus_read_lock(); |
201517a7 | 1070 | mutex_lock(&text_mutex); |
afd66255 | 1071 | __arm_kprobe(kp); |
201517a7 | 1072 | mutex_unlock(&text_mutex); |
2d1e38f5 | 1073 | cpus_read_unlock(); |
12310e34 JY |
1074 | |
1075 | return 0; | |
201517a7 MH |
1076 | } |
1077 | ||
1078 | /* Disarm a kprobe with text_mutex */ | |
297f9233 | 1079 | static int disarm_kprobe(struct kprobe *kp, bool reopt) |
201517a7 | 1080 | { |
297f9233 JY |
1081 | if (unlikely(kprobe_ftrace(kp))) |
1082 | return disarm_kprobe_ftrace(kp); | |
2d1e38f5 TG |
1083 | |
1084 | cpus_read_lock(); | |
201517a7 | 1085 | mutex_lock(&text_mutex); |
ae6aa16f | 1086 | __disarm_kprobe(kp, reopt); |
201517a7 | 1087 | mutex_unlock(&text_mutex); |
2d1e38f5 | 1088 | cpus_read_unlock(); |
297f9233 JY |
1089 | |
1090 | return 0; | |
201517a7 MH |
1091 | } |
1092 | ||
64f562c6 AM |
1093 | /* |
1094 | * Aggregate handlers for multiple kprobes support - these handlers | |
1095 | * take care of invoking the individual kprobe handlers on p->list | |
1096 | */ | |
820aede0 | 1097 | static int aggr_pre_handler(struct kprobe *p, struct pt_regs *regs) |
64f562c6 AM |
1098 | { |
1099 | struct kprobe *kp; | |
1100 | ||
3516a460 | 1101 | list_for_each_entry_rcu(kp, &p->list, list) { |
de5bd88d | 1102 | if (kp->pre_handler && likely(!kprobe_disabled(kp))) { |
e6584523 | 1103 | set_kprobe_instance(kp); |
8b0914ea PP |
1104 | if (kp->pre_handler(kp, regs)) |
1105 | return 1; | |
64f562c6 | 1106 | } |
e6584523 | 1107 | reset_kprobe_instance(); |
64f562c6 AM |
1108 | } |
1109 | return 0; | |
1110 | } | |
820aede0 | 1111 | NOKPROBE_SYMBOL(aggr_pre_handler); |
64f562c6 | 1112 | |
820aede0 MH |
1113 | static void aggr_post_handler(struct kprobe *p, struct pt_regs *regs, |
1114 | unsigned long flags) | |
64f562c6 AM |
1115 | { |
1116 | struct kprobe *kp; | |
1117 | ||
3516a460 | 1118 | list_for_each_entry_rcu(kp, &p->list, list) { |
de5bd88d | 1119 | if (kp->post_handler && likely(!kprobe_disabled(kp))) { |
e6584523 | 1120 | set_kprobe_instance(kp); |
64f562c6 | 1121 | kp->post_handler(kp, regs, flags); |
e6584523 | 1122 | reset_kprobe_instance(); |
64f562c6 AM |
1123 | } |
1124 | } | |
64f562c6 | 1125 | } |
820aede0 | 1126 | NOKPROBE_SYMBOL(aggr_post_handler); |
64f562c6 | 1127 | |
820aede0 MH |
1128 | static int aggr_fault_handler(struct kprobe *p, struct pt_regs *regs, |
1129 | int trapnr) | |
64f562c6 | 1130 | { |
b76834bc | 1131 | struct kprobe *cur = __this_cpu_read(kprobe_instance); |
e6584523 | 1132 | |
64f562c6 AM |
1133 | /* |
1134 | * if we faulted "during" the execution of a user specified | |
1135 | * probe handler, invoke just that probe's fault handler | |
1136 | */ | |
e6584523 AM |
1137 | if (cur && cur->fault_handler) { |
1138 | if (cur->fault_handler(cur, regs, trapnr)) | |
64f562c6 AM |
1139 | return 1; |
1140 | } | |
1141 | return 0; | |
1142 | } | |
820aede0 | 1143 | NOKPROBE_SYMBOL(aggr_fault_handler); |
64f562c6 | 1144 | |
bf8d5c52 | 1145 | /* Walks the list and increments nmissed count for multiprobe case */ |
820aede0 | 1146 | void kprobes_inc_nmissed_count(struct kprobe *p) |
bf8d5c52 KA |
1147 | { |
1148 | struct kprobe *kp; | |
afd66255 | 1149 | if (!kprobe_aggrprobe(p)) { |
bf8d5c52 KA |
1150 | p->nmissed++; |
1151 | } else { | |
1152 | list_for_each_entry_rcu(kp, &p->list, list) | |
1153 | kp->nmissed++; | |
1154 | } | |
1155 | return; | |
1156 | } | |
820aede0 | 1157 | NOKPROBE_SYMBOL(kprobes_inc_nmissed_count); |
bf8d5c52 | 1158 | |
820aede0 MH |
1159 | void recycle_rp_inst(struct kretprobe_instance *ri, |
1160 | struct hlist_head *head) | |
b94cce92 | 1161 | { |
ef53d9c5 S |
1162 | struct kretprobe *rp = ri->rp; |
1163 | ||
b94cce92 HN |
1164 | /* remove rp inst off the rprobe_inst_table */ |
1165 | hlist_del(&ri->hlist); | |
ef53d9c5 S |
1166 | INIT_HLIST_NODE(&ri->hlist); |
1167 | if (likely(rp)) { | |
ec484608 | 1168 | raw_spin_lock(&rp->lock); |
ef53d9c5 | 1169 | hlist_add_head(&ri->hlist, &rp->free_instances); |
ec484608 | 1170 | raw_spin_unlock(&rp->lock); |
b94cce92 HN |
1171 | } else |
1172 | /* Unregistering */ | |
99219a3f | 1173 | hlist_add_head(&ri->hlist, head); |
b94cce92 | 1174 | } |
820aede0 | 1175 | NOKPROBE_SYMBOL(recycle_rp_inst); |
b94cce92 | 1176 | |
820aede0 | 1177 | void kretprobe_hash_lock(struct task_struct *tsk, |
ef53d9c5 | 1178 | struct hlist_head **head, unsigned long *flags) |
635c17c2 | 1179 | __acquires(hlist_lock) |
ef53d9c5 S |
1180 | { |
1181 | unsigned long hash = hash_ptr(tsk, KPROBE_HASH_BITS); | |
ec484608 | 1182 | raw_spinlock_t *hlist_lock; |
ef53d9c5 S |
1183 | |
1184 | *head = &kretprobe_inst_table[hash]; | |
1185 | hlist_lock = kretprobe_table_lock_ptr(hash); | |
ec484608 | 1186 | raw_spin_lock_irqsave(hlist_lock, *flags); |
ef53d9c5 | 1187 | } |
820aede0 | 1188 | NOKPROBE_SYMBOL(kretprobe_hash_lock); |
ef53d9c5 | 1189 | |
820aede0 MH |
1190 | static void kretprobe_table_lock(unsigned long hash, |
1191 | unsigned long *flags) | |
635c17c2 | 1192 | __acquires(hlist_lock) |
b94cce92 | 1193 | { |
ec484608 TG |
1194 | raw_spinlock_t *hlist_lock = kretprobe_table_lock_ptr(hash); |
1195 | raw_spin_lock_irqsave(hlist_lock, *flags); | |
ef53d9c5 | 1196 | } |
820aede0 | 1197 | NOKPROBE_SYMBOL(kretprobe_table_lock); |
ef53d9c5 | 1198 | |
820aede0 MH |
1199 | void kretprobe_hash_unlock(struct task_struct *tsk, |
1200 | unsigned long *flags) | |
635c17c2 | 1201 | __releases(hlist_lock) |
ef53d9c5 S |
1202 | { |
1203 | unsigned long hash = hash_ptr(tsk, KPROBE_HASH_BITS); | |
ec484608 | 1204 | raw_spinlock_t *hlist_lock; |
ef53d9c5 S |
1205 | |
1206 | hlist_lock = kretprobe_table_lock_ptr(hash); | |
ec484608 | 1207 | raw_spin_unlock_irqrestore(hlist_lock, *flags); |
ef53d9c5 | 1208 | } |
820aede0 | 1209 | NOKPROBE_SYMBOL(kretprobe_hash_unlock); |
ef53d9c5 | 1210 | |
820aede0 MH |
1211 | static void kretprobe_table_unlock(unsigned long hash, |
1212 | unsigned long *flags) | |
635c17c2 | 1213 | __releases(hlist_lock) |
ef53d9c5 | 1214 | { |
ec484608 TG |
1215 | raw_spinlock_t *hlist_lock = kretprobe_table_lock_ptr(hash); |
1216 | raw_spin_unlock_irqrestore(hlist_lock, *flags); | |
b94cce92 | 1217 | } |
820aede0 | 1218 | NOKPROBE_SYMBOL(kretprobe_table_unlock); |
b94cce92 | 1219 | |
b94cce92 | 1220 | /* |
c6fd91f0 | 1221 | * This function is called from finish_task_switch when task tk becomes dead, |
1222 | * so that we can recycle any function-return probe instances associated | |
1223 | * with this task. These left over instances represent probed functions | |
1224 | * that have been called but will never return. | |
b94cce92 | 1225 | */ |
820aede0 | 1226 | void kprobe_flush_task(struct task_struct *tk) |
b94cce92 | 1227 | { |
62c27be0 | 1228 | struct kretprobe_instance *ri; |
99219a3f | 1229 | struct hlist_head *head, empty_rp; |
b67bfe0d | 1230 | struct hlist_node *tmp; |
ef53d9c5 | 1231 | unsigned long hash, flags = 0; |
802eae7c | 1232 | |
ef53d9c5 S |
1233 | if (unlikely(!kprobes_initialized)) |
1234 | /* Early boot. kretprobe_table_locks not yet initialized. */ | |
1235 | return; | |
1236 | ||
d496aab5 | 1237 | INIT_HLIST_HEAD(&empty_rp); |
ef53d9c5 S |
1238 | hash = hash_ptr(tk, KPROBE_HASH_BITS); |
1239 | head = &kretprobe_inst_table[hash]; | |
1240 | kretprobe_table_lock(hash, &flags); | |
b67bfe0d | 1241 | hlist_for_each_entry_safe(ri, tmp, head, hlist) { |
62c27be0 | 1242 | if (ri->task == tk) |
99219a3f | 1243 | recycle_rp_inst(ri, &empty_rp); |
62c27be0 | 1244 | } |
ef53d9c5 | 1245 | kretprobe_table_unlock(hash, &flags); |
b67bfe0d | 1246 | hlist_for_each_entry_safe(ri, tmp, &empty_rp, hlist) { |
99219a3f | 1247 | hlist_del(&ri->hlist); |
1248 | kfree(ri); | |
1249 | } | |
b94cce92 | 1250 | } |
820aede0 | 1251 | NOKPROBE_SYMBOL(kprobe_flush_task); |
b94cce92 | 1252 | |
b94cce92 HN |
1253 | static inline void free_rp_inst(struct kretprobe *rp) |
1254 | { | |
1255 | struct kretprobe_instance *ri; | |
b67bfe0d | 1256 | struct hlist_node *next; |
4c4308cb | 1257 | |
b67bfe0d | 1258 | hlist_for_each_entry_safe(ri, next, &rp->free_instances, hlist) { |
ef53d9c5 | 1259 | hlist_del(&ri->hlist); |
b94cce92 HN |
1260 | kfree(ri); |
1261 | } | |
1262 | } | |
1263 | ||
820aede0 | 1264 | static void cleanup_rp_inst(struct kretprobe *rp) |
4a296e07 | 1265 | { |
ef53d9c5 | 1266 | unsigned long flags, hash; |
4a296e07 | 1267 | struct kretprobe_instance *ri; |
b67bfe0d | 1268 | struct hlist_node *next; |
ef53d9c5 S |
1269 | struct hlist_head *head; |
1270 | ||
4a296e07 | 1271 | /* No race here */ |
ef53d9c5 S |
1272 | for (hash = 0; hash < KPROBE_TABLE_SIZE; hash++) { |
1273 | kretprobe_table_lock(hash, &flags); | |
1274 | head = &kretprobe_inst_table[hash]; | |
b67bfe0d | 1275 | hlist_for_each_entry_safe(ri, next, head, hlist) { |
ef53d9c5 S |
1276 | if (ri->rp == rp) |
1277 | ri->rp = NULL; | |
1278 | } | |
1279 | kretprobe_table_unlock(hash, &flags); | |
4a296e07 | 1280 | } |
4a296e07 MH |
1281 | free_rp_inst(rp); |
1282 | } | |
820aede0 | 1283 | NOKPROBE_SYMBOL(cleanup_rp_inst); |
4a296e07 | 1284 | |
059053a2 | 1285 | /* Add the new probe to ap->list */ |
55479f64 | 1286 | static int add_new_kprobe(struct kprobe *ap, struct kprobe *p) |
8b0914ea | 1287 | { |
059053a2 | 1288 | if (p->post_handler) |
6274de49 | 1289 | unoptimize_kprobe(ap, true); /* Fall back to normal kprobe */ |
afd66255 | 1290 | |
059053a2 | 1291 | list_add_rcu(&p->list, &ap->list); |
b918e5e6 MH |
1292 | if (p->post_handler && !ap->post_handler) |
1293 | ap->post_handler = aggr_post_handler; | |
de5bd88d | 1294 | |
8b0914ea PP |
1295 | return 0; |
1296 | } | |
1297 | ||
64f562c6 AM |
1298 | /* |
1299 | * Fill in the required fields of the "manager kprobe". Replace the | |
1300 | * earlier kprobe in the hlist with the manager kprobe | |
1301 | */ | |
55479f64 | 1302 | static void init_aggr_kprobe(struct kprobe *ap, struct kprobe *p) |
64f562c6 | 1303 | { |
afd66255 | 1304 | /* Copy p's insn slot to ap */ |
8b0914ea | 1305 | copy_kprobe(p, ap); |
a9ad965e | 1306 | flush_insn_slot(ap); |
64f562c6 | 1307 | ap->addr = p->addr; |
afd66255 | 1308 | ap->flags = p->flags & ~KPROBE_FLAG_OPTIMIZED; |
64f562c6 | 1309 | ap->pre_handler = aggr_pre_handler; |
64f562c6 | 1310 | ap->fault_handler = aggr_fault_handler; |
e8386a0c MH |
1311 | /* We don't care the kprobe which has gone. */ |
1312 | if (p->post_handler && !kprobe_gone(p)) | |
36721656 | 1313 | ap->post_handler = aggr_post_handler; |
64f562c6 AM |
1314 | |
1315 | INIT_LIST_HEAD(&ap->list); | |
afd66255 | 1316 | INIT_HLIST_NODE(&ap->hlist); |
64f562c6 | 1317 | |
afd66255 | 1318 | list_add_rcu(&p->list, &ap->list); |
adad0f33 | 1319 | hlist_replace_rcu(&p->hlist, &ap->hlist); |
64f562c6 AM |
1320 | } |
1321 | ||
1322 | /* | |
1323 | * This is the second or subsequent kprobe at the address - handle | |
1324 | * the intricacies | |
64f562c6 | 1325 | */ |
55479f64 | 1326 | static int register_aggr_kprobe(struct kprobe *orig_p, struct kprobe *p) |
64f562c6 AM |
1327 | { |
1328 | int ret = 0; | |
6d8e40a8 | 1329 | struct kprobe *ap = orig_p; |
64f562c6 | 1330 | |
2d1e38f5 TG |
1331 | cpus_read_lock(); |
1332 | ||
25764288 MH |
1333 | /* For preparing optimization, jump_label_text_reserved() is called */ |
1334 | jump_label_lock(); | |
25764288 MH |
1335 | mutex_lock(&text_mutex); |
1336 | ||
6d8e40a8 MH |
1337 | if (!kprobe_aggrprobe(orig_p)) { |
1338 | /* If orig_p is not an aggr_kprobe, create new aggr_kprobe. */ | |
1339 | ap = alloc_aggr_kprobe(orig_p); | |
25764288 MH |
1340 | if (!ap) { |
1341 | ret = -ENOMEM; | |
1342 | goto out; | |
1343 | } | |
6d8e40a8 | 1344 | init_aggr_kprobe(ap, orig_p); |
819319fc | 1345 | } else if (kprobe_unused(ap)) { |
0490cd1f | 1346 | /* This probe is going to die. Rescue it */ |
819319fc MH |
1347 | ret = reuse_unused_kprobe(ap); |
1348 | if (ret) | |
1349 | goto out; | |
1350 | } | |
b918e5e6 MH |
1351 | |
1352 | if (kprobe_gone(ap)) { | |
e8386a0c MH |
1353 | /* |
1354 | * Attempting to insert new probe at the same location that | |
1355 | * had a probe in the module vaddr area which already | |
1356 | * freed. So, the instruction slot has already been | |
1357 | * released. We need a new slot for the new probe. | |
1358 | */ | |
b918e5e6 | 1359 | ret = arch_prepare_kprobe(ap); |
e8386a0c | 1360 | if (ret) |
b918e5e6 MH |
1361 | /* |
1362 | * Even if fail to allocate new slot, don't need to | |
1363 | * free aggr_probe. It will be used next time, or | |
1364 | * freed by unregister_kprobe. | |
1365 | */ | |
25764288 | 1366 | goto out; |
de5bd88d | 1367 | |
afd66255 MH |
1368 | /* Prepare optimized instructions if possible. */ |
1369 | prepare_optimized_kprobe(ap); | |
1370 | ||
e8386a0c | 1371 | /* |
de5bd88d MH |
1372 | * Clear gone flag to prevent allocating new slot again, and |
1373 | * set disabled flag because it is not armed yet. | |
e8386a0c | 1374 | */ |
de5bd88d MH |
1375 | ap->flags = (ap->flags & ~KPROBE_FLAG_GONE) |
1376 | | KPROBE_FLAG_DISABLED; | |
e8386a0c | 1377 | } |
b918e5e6 | 1378 | |
afd66255 | 1379 | /* Copy ap's insn slot to p */ |
b918e5e6 | 1380 | copy_kprobe(ap, p); |
25764288 MH |
1381 | ret = add_new_kprobe(ap, p); |
1382 | ||
1383 | out: | |
1384 | mutex_unlock(&text_mutex); | |
25764288 | 1385 | jump_label_unlock(); |
2d1e38f5 | 1386 | cpus_read_unlock(); |
25764288 MH |
1387 | |
1388 | if (ret == 0 && kprobe_disabled(ap) && !kprobe_disabled(p)) { | |
1389 | ap->flags &= ~KPROBE_FLAG_DISABLED; | |
12310e34 | 1390 | if (!kprobes_all_disarmed) { |
25764288 | 1391 | /* Arm the breakpoint again. */ |
12310e34 JY |
1392 | ret = arm_kprobe(ap); |
1393 | if (ret) { | |
1394 | ap->flags |= KPROBE_FLAG_DISABLED; | |
1395 | list_del_rcu(&p->list); | |
ae8b7ce7 | 1396 | synchronize_rcu(); |
12310e34 JY |
1397 | } |
1398 | } | |
25764288 MH |
1399 | } |
1400 | return ret; | |
64f562c6 AM |
1401 | } |
1402 | ||
be8f2743 MH |
1403 | bool __weak arch_within_kprobe_blacklist(unsigned long addr) |
1404 | { | |
1405 | /* The __kprobes marked functions and entry code must not be probed */ | |
1406 | return addr >= (unsigned long)__kprobes_text_start && | |
1407 | addr < (unsigned long)__kprobes_text_end; | |
1408 | } | |
1409 | ||
6143c6fb | 1410 | static bool __within_kprobe_blacklist(unsigned long addr) |
d0aaff97 | 1411 | { |
376e2424 | 1412 | struct kprobe_blacklist_entry *ent; |
3d8d996e | 1413 | |
be8f2743 | 1414 | if (arch_within_kprobe_blacklist(addr)) |
376e2424 | 1415 | return true; |
3d8d996e SD |
1416 | /* |
1417 | * If there exists a kprobe_blacklist, verify and | |
1418 | * fail any probe registration in the prohibited area | |
1419 | */ | |
376e2424 MH |
1420 | list_for_each_entry(ent, &kprobe_blacklist, list) { |
1421 | if (addr >= ent->start_addr && addr < ent->end_addr) | |
1422 | return true; | |
3d8d996e | 1423 | } |
6143c6fb MH |
1424 | return false; |
1425 | } | |
376e2424 | 1426 | |
6143c6fb MH |
1427 | bool within_kprobe_blacklist(unsigned long addr) |
1428 | { | |
1429 | char symname[KSYM_NAME_LEN], *p; | |
1430 | ||
1431 | if (__within_kprobe_blacklist(addr)) | |
1432 | return true; | |
1433 | ||
1434 | /* Check if the address is on a suffixed-symbol */ | |
1435 | if (!lookup_symbol_name(addr, symname)) { | |
1436 | p = strchr(symname, '.'); | |
1437 | if (!p) | |
1438 | return false; | |
1439 | *p = '\0'; | |
1440 | addr = (unsigned long)kprobe_lookup_name(symname, 0); | |
1441 | if (addr) | |
1442 | return __within_kprobe_blacklist(addr); | |
1443 | } | |
376e2424 | 1444 | return false; |
d0aaff97 PP |
1445 | } |
1446 | ||
b2a5cd69 MH |
1447 | /* |
1448 | * If we have a symbol_name argument, look it up and add the offset field | |
1449 | * to it. This way, we can specify a relative address to a symbol. | |
bc81d48d MH |
1450 | * This returns encoded errors if it fails to look up symbol or invalid |
1451 | * combination of parameters. | |
b2a5cd69 | 1452 | */ |
1d585e70 NR |
1453 | static kprobe_opcode_t *_kprobe_addr(kprobe_opcode_t *addr, |
1454 | const char *symbol_name, unsigned int offset) | |
b2a5cd69 | 1455 | { |
1d585e70 | 1456 | if ((symbol_name && addr) || (!symbol_name && !addr)) |
bc81d48d MH |
1457 | goto invalid; |
1458 | ||
1d585e70 | 1459 | if (symbol_name) { |
7246f600 | 1460 | addr = kprobe_lookup_name(symbol_name, offset); |
bc81d48d MH |
1461 | if (!addr) |
1462 | return ERR_PTR(-ENOENT); | |
b2a5cd69 MH |
1463 | } |
1464 | ||
1d585e70 | 1465 | addr = (kprobe_opcode_t *)(((char *)addr) + offset); |
bc81d48d MH |
1466 | if (addr) |
1467 | return addr; | |
1468 | ||
1469 | invalid: | |
1470 | return ERR_PTR(-EINVAL); | |
b2a5cd69 MH |
1471 | } |
1472 | ||
1d585e70 NR |
1473 | static kprobe_opcode_t *kprobe_addr(struct kprobe *p) |
1474 | { | |
1475 | return _kprobe_addr(p->addr, p->symbol_name, p->offset); | |
1476 | } | |
1477 | ||
1f0ab409 | 1478 | /* Check passed kprobe is valid and return kprobe in kprobe_table. */ |
55479f64 | 1479 | static struct kprobe *__get_valid_kprobe(struct kprobe *p) |
1f0ab409 | 1480 | { |
6d8e40a8 | 1481 | struct kprobe *ap, *list_p; |
1f0ab409 | 1482 | |
6d8e40a8 MH |
1483 | ap = get_kprobe(p->addr); |
1484 | if (unlikely(!ap)) | |
1f0ab409 AM |
1485 | return NULL; |
1486 | ||
6d8e40a8 MH |
1487 | if (p != ap) { |
1488 | list_for_each_entry_rcu(list_p, &ap->list, list) | |
1f0ab409 AM |
1489 | if (list_p == p) |
1490 | /* kprobe p is a valid probe */ | |
1491 | goto valid; | |
1492 | return NULL; | |
1493 | } | |
1494 | valid: | |
6d8e40a8 | 1495 | return ap; |
1f0ab409 AM |
1496 | } |
1497 | ||
1498 | /* Return error if the kprobe is being re-registered */ | |
1499 | static inline int check_kprobe_rereg(struct kprobe *p) | |
1500 | { | |
1501 | int ret = 0; | |
1f0ab409 AM |
1502 | |
1503 | mutex_lock(&kprobe_mutex); | |
6d8e40a8 | 1504 | if (__get_valid_kprobe(p)) |
1f0ab409 AM |
1505 | ret = -EINVAL; |
1506 | mutex_unlock(&kprobe_mutex); | |
6d8e40a8 | 1507 | |
1f0ab409 AM |
1508 | return ret; |
1509 | } | |
1510 | ||
f7f242ff | 1511 | int __weak arch_check_ftrace_location(struct kprobe *p) |
1da177e4 | 1512 | { |
ae6aa16f MH |
1513 | unsigned long ftrace_addr; |
1514 | ||
ae6aa16f MH |
1515 | ftrace_addr = ftrace_location((unsigned long)p->addr); |
1516 | if (ftrace_addr) { | |
e7dbfe34 | 1517 | #ifdef CONFIG_KPROBES_ON_FTRACE |
ae6aa16f MH |
1518 | /* Given address is not on the instruction boundary */ |
1519 | if ((unsigned long)p->addr != ftrace_addr) | |
1520 | return -EILSEQ; | |
ae6aa16f | 1521 | p->flags |= KPROBE_FLAG_FTRACE; |
e7dbfe34 | 1522 | #else /* !CONFIG_KPROBES_ON_FTRACE */ |
ae6aa16f MH |
1523 | return -EINVAL; |
1524 | #endif | |
1525 | } | |
f7f242ff HC |
1526 | return 0; |
1527 | } | |
1528 | ||
1529 | static int check_kprobe_address_safe(struct kprobe *p, | |
1530 | struct module **probed_mod) | |
1531 | { | |
1532 | int ret; | |
1f0ab409 | 1533 | |
f7f242ff HC |
1534 | ret = arch_check_ftrace_location(p); |
1535 | if (ret) | |
1536 | return ret; | |
91bad2f8 | 1537 | jump_label_lock(); |
de31c3ca | 1538 | preempt_disable(); |
f7fa6ef0 MH |
1539 | |
1540 | /* Ensure it is not in reserved area nor out of text */ | |
ec30c5f3 | 1541 | if (!kernel_text_address((unsigned long) p->addr) || |
376e2424 | 1542 | within_kprobe_blacklist((unsigned long) p->addr) || |
e336b402 MH |
1543 | jump_label_text_reserved(p->addr, p->addr) || |
1544 | find_bug((unsigned long)p->addr)) { | |
f986a499 | 1545 | ret = -EINVAL; |
f7fa6ef0 | 1546 | goto out; |
f986a499 | 1547 | } |
b3e55c72 | 1548 | |
f7fa6ef0 MH |
1549 | /* Check if are we probing a module */ |
1550 | *probed_mod = __module_text_address((unsigned long) p->addr); | |
1551 | if (*probed_mod) { | |
6f716acd | 1552 | /* |
e8386a0c MH |
1553 | * We must hold a refcount of the probed module while updating |
1554 | * its code to prohibit unexpected unloading. | |
df019b1d | 1555 | */ |
f7fa6ef0 MH |
1556 | if (unlikely(!try_module_get(*probed_mod))) { |
1557 | ret = -ENOENT; | |
1558 | goto out; | |
1559 | } | |
de31c3ca | 1560 | |
f24659d9 MH |
1561 | /* |
1562 | * If the module freed .init.text, we couldn't insert | |
1563 | * kprobes in there. | |
1564 | */ | |
f7fa6ef0 MH |
1565 | if (within_module_init((unsigned long)p->addr, *probed_mod) && |
1566 | (*probed_mod)->state != MODULE_STATE_COMING) { | |
1567 | module_put(*probed_mod); | |
1568 | *probed_mod = NULL; | |
1569 | ret = -ENOENT; | |
f24659d9 | 1570 | } |
df019b1d | 1571 | } |
f7fa6ef0 | 1572 | out: |
a189d035 | 1573 | preempt_enable(); |
de31c3ca | 1574 | jump_label_unlock(); |
1da177e4 | 1575 | |
f7fa6ef0 MH |
1576 | return ret; |
1577 | } | |
1578 | ||
55479f64 | 1579 | int register_kprobe(struct kprobe *p) |
f7fa6ef0 MH |
1580 | { |
1581 | int ret; | |
1582 | struct kprobe *old_p; | |
1583 | struct module *probed_mod; | |
1584 | kprobe_opcode_t *addr; | |
1585 | ||
1586 | /* Adjust probe address from symbol */ | |
1587 | addr = kprobe_addr(p); | |
1588 | if (IS_ERR(addr)) | |
1589 | return PTR_ERR(addr); | |
1590 | p->addr = addr; | |
1591 | ||
1592 | ret = check_kprobe_rereg(p); | |
1593 | if (ret) | |
1594 | return ret; | |
1595 | ||
1596 | /* User can pass only KPROBE_FLAG_DISABLED to register_kprobe */ | |
1597 | p->flags &= KPROBE_FLAG_DISABLED; | |
3516a460 | 1598 | p->nmissed = 0; |
9861668f | 1599 | INIT_LIST_HEAD(&p->list); |
afd66255 | 1600 | |
f7fa6ef0 MH |
1601 | ret = check_kprobe_address_safe(p, &probed_mod); |
1602 | if (ret) | |
1603 | return ret; | |
1604 | ||
1605 | mutex_lock(&kprobe_mutex); | |
afd66255 | 1606 | |
64f562c6 AM |
1607 | old_p = get_kprobe(p->addr); |
1608 | if (old_p) { | |
afd66255 | 1609 | /* Since this may unoptimize old_p, locking text_mutex. */ |
64f562c6 | 1610 | ret = register_aggr_kprobe(old_p, p); |
1da177e4 LT |
1611 | goto out; |
1612 | } | |
1da177e4 | 1613 | |
2d1e38f5 TG |
1614 | cpus_read_lock(); |
1615 | /* Prevent text modification */ | |
1616 | mutex_lock(&text_mutex); | |
ae6aa16f | 1617 | ret = prepare_kprobe(p); |
25764288 | 1618 | mutex_unlock(&text_mutex); |
2d1e38f5 | 1619 | cpus_read_unlock(); |
6f716acd | 1620 | if (ret) |
afd66255 | 1621 | goto out; |
49a2a1b8 | 1622 | |
64f562c6 | 1623 | INIT_HLIST_NODE(&p->hlist); |
3516a460 | 1624 | hlist_add_head_rcu(&p->hlist, |
1da177e4 LT |
1625 | &kprobe_table[hash_ptr(p->addr, KPROBE_HASH_BITS)]); |
1626 | ||
12310e34 JY |
1627 | if (!kprobes_all_disarmed && !kprobe_disabled(p)) { |
1628 | ret = arm_kprobe(p); | |
1629 | if (ret) { | |
1630 | hlist_del_rcu(&p->hlist); | |
ae8b7ce7 | 1631 | synchronize_rcu(); |
12310e34 JY |
1632 | goto out; |
1633 | } | |
1634 | } | |
afd66255 MH |
1635 | |
1636 | /* Try to optimize kprobe */ | |
1637 | try_to_optimize_kprobe(p); | |
1da177e4 | 1638 | out: |
7a7d1cf9 | 1639 | mutex_unlock(&kprobe_mutex); |
49a2a1b8 | 1640 | |
e8386a0c | 1641 | if (probed_mod) |
df019b1d | 1642 | module_put(probed_mod); |
e8386a0c | 1643 | |
1da177e4 LT |
1644 | return ret; |
1645 | } | |
99081ab5 | 1646 | EXPORT_SYMBOL_GPL(register_kprobe); |
1da177e4 | 1647 | |
6f0f1dd7 | 1648 | /* Check if all probes on the aggrprobe are disabled */ |
55479f64 | 1649 | static int aggr_kprobe_disabled(struct kprobe *ap) |
6f0f1dd7 MH |
1650 | { |
1651 | struct kprobe *kp; | |
1652 | ||
1653 | list_for_each_entry_rcu(kp, &ap->list, list) | |
1654 | if (!kprobe_disabled(kp)) | |
1655 | /* | |
1656 | * There is an active probe on the list. | |
1657 | * We can't disable this ap. | |
1658 | */ | |
1659 | return 0; | |
1660 | ||
1661 | return 1; | |
1662 | } | |
1663 | ||
1664 | /* Disable one kprobe: Make sure called under kprobe_mutex is locked */ | |
55479f64 | 1665 | static struct kprobe *__disable_kprobe(struct kprobe *p) |
6f0f1dd7 MH |
1666 | { |
1667 | struct kprobe *orig_p; | |
297f9233 | 1668 | int ret; |
6f0f1dd7 MH |
1669 | |
1670 | /* Get an original kprobe for return */ | |
1671 | orig_p = __get_valid_kprobe(p); | |
1672 | if (unlikely(orig_p == NULL)) | |
297f9233 | 1673 | return ERR_PTR(-EINVAL); |
6f0f1dd7 MH |
1674 | |
1675 | if (!kprobe_disabled(p)) { | |
1676 | /* Disable probe if it is a child probe */ | |
1677 | if (p != orig_p) | |
1678 | p->flags |= KPROBE_FLAG_DISABLED; | |
1679 | ||
1680 | /* Try to disarm and disable this/parent probe */ | |
1681 | if (p == orig_p || aggr_kprobe_disabled(orig_p)) { | |
69d54b91 WN |
1682 | /* |
1683 | * If kprobes_all_disarmed is set, orig_p | |
1684 | * should have already been disarmed, so | |
1685 | * skip unneed disarming process. | |
1686 | */ | |
297f9233 JY |
1687 | if (!kprobes_all_disarmed) { |
1688 | ret = disarm_kprobe(orig_p, true); | |
1689 | if (ret) { | |
1690 | p->flags &= ~KPROBE_FLAG_DISABLED; | |
1691 | return ERR_PTR(ret); | |
1692 | } | |
1693 | } | |
6f0f1dd7 MH |
1694 | orig_p->flags |= KPROBE_FLAG_DISABLED; |
1695 | } | |
1696 | } | |
1697 | ||
1698 | return orig_p; | |
1699 | } | |
1700 | ||
de5bd88d MH |
1701 | /* |
1702 | * Unregister a kprobe without a scheduler synchronization. | |
1703 | */ | |
55479f64 | 1704 | static int __unregister_kprobe_top(struct kprobe *p) |
de5bd88d | 1705 | { |
6d8e40a8 | 1706 | struct kprobe *ap, *list_p; |
de5bd88d | 1707 | |
6f0f1dd7 MH |
1708 | /* Disable kprobe. This will disarm it if needed. */ |
1709 | ap = __disable_kprobe(p); | |
297f9233 JY |
1710 | if (IS_ERR(ap)) |
1711 | return PTR_ERR(ap); | |
de5bd88d | 1712 | |
6f0f1dd7 | 1713 | if (ap == p) |
bf8f6e5b | 1714 | /* |
6f0f1dd7 MH |
1715 | * This probe is an independent(and non-optimized) kprobe |
1716 | * (not an aggrprobe). Remove from the hash list. | |
bf8f6e5b | 1717 | */ |
6f0f1dd7 MH |
1718 | goto disarmed; |
1719 | ||
1720 | /* Following process expects this probe is an aggrprobe */ | |
1721 | WARN_ON(!kprobe_aggrprobe(ap)); | |
1722 | ||
6274de49 MH |
1723 | if (list_is_singular(&ap->list) && kprobe_disarmed(ap)) |
1724 | /* | |
1725 | * !disarmed could be happen if the probe is under delayed | |
1726 | * unoptimizing. | |
1727 | */ | |
6f0f1dd7 MH |
1728 | goto disarmed; |
1729 | else { | |
1730 | /* If disabling probe has special handlers, update aggrprobe */ | |
e8386a0c | 1731 | if (p->post_handler && !kprobe_gone(p)) { |
6d8e40a8 | 1732 | list_for_each_entry_rcu(list_p, &ap->list, list) { |
9861668f MH |
1733 | if ((list_p != p) && (list_p->post_handler)) |
1734 | goto noclean; | |
1735 | } | |
6d8e40a8 | 1736 | ap->post_handler = NULL; |
9861668f MH |
1737 | } |
1738 | noclean: | |
6f0f1dd7 MH |
1739 | /* |
1740 | * Remove from the aggrprobe: this path will do nothing in | |
1741 | * __unregister_kprobe_bottom(). | |
1742 | */ | |
49a2a1b8 | 1743 | list_del_rcu(&p->list); |
6f0f1dd7 MH |
1744 | if (!kprobe_disabled(ap) && !kprobes_all_disarmed) |
1745 | /* | |
1746 | * Try to optimize this probe again, because post | |
1747 | * handler may have been changed. | |
1748 | */ | |
1749 | optimize_kprobe(ap); | |
49a2a1b8 | 1750 | } |
9861668f | 1751 | return 0; |
6f0f1dd7 MH |
1752 | |
1753 | disarmed: | |
1754 | hlist_del_rcu(&ap->hlist); | |
1755 | return 0; | |
9861668f | 1756 | } |
3516a460 | 1757 | |
55479f64 | 1758 | static void __unregister_kprobe_bottom(struct kprobe *p) |
9861668f | 1759 | { |
6d8e40a8 | 1760 | struct kprobe *ap; |
b3e55c72 | 1761 | |
e8386a0c | 1762 | if (list_empty(&p->list)) |
6274de49 | 1763 | /* This is an independent kprobe */ |
0498b635 | 1764 | arch_remove_kprobe(p); |
e8386a0c | 1765 | else if (list_is_singular(&p->list)) { |
6274de49 | 1766 | /* This is the last child of an aggrprobe */ |
6d8e40a8 | 1767 | ap = list_entry(p->list.next, struct kprobe, list); |
e8386a0c | 1768 | list_del(&p->list); |
6d8e40a8 | 1769 | free_aggr_kprobe(ap); |
9861668f | 1770 | } |
6274de49 | 1771 | /* Otherwise, do nothing. */ |
9861668f MH |
1772 | } |
1773 | ||
55479f64 | 1774 | int register_kprobes(struct kprobe **kps, int num) |
9861668f MH |
1775 | { |
1776 | int i, ret = 0; | |
1777 | ||
1778 | if (num <= 0) | |
1779 | return -EINVAL; | |
1780 | for (i = 0; i < num; i++) { | |
49ad2fd7 | 1781 | ret = register_kprobe(kps[i]); |
67dddaad MH |
1782 | if (ret < 0) { |
1783 | if (i > 0) | |
1784 | unregister_kprobes(kps, i); | |
9861668f | 1785 | break; |
36721656 | 1786 | } |
49a2a1b8 | 1787 | } |
9861668f MH |
1788 | return ret; |
1789 | } | |
99081ab5 | 1790 | EXPORT_SYMBOL_GPL(register_kprobes); |
9861668f | 1791 | |
55479f64 | 1792 | void unregister_kprobe(struct kprobe *p) |
9861668f MH |
1793 | { |
1794 | unregister_kprobes(&p, 1); | |
1795 | } | |
99081ab5 | 1796 | EXPORT_SYMBOL_GPL(unregister_kprobe); |
9861668f | 1797 | |
55479f64 | 1798 | void unregister_kprobes(struct kprobe **kps, int num) |
9861668f MH |
1799 | { |
1800 | int i; | |
1801 | ||
1802 | if (num <= 0) | |
1803 | return; | |
1804 | mutex_lock(&kprobe_mutex); | |
1805 | for (i = 0; i < num; i++) | |
1806 | if (__unregister_kprobe_top(kps[i]) < 0) | |
1807 | kps[i]->addr = NULL; | |
1808 | mutex_unlock(&kprobe_mutex); | |
1809 | ||
ae8b7ce7 | 1810 | synchronize_rcu(); |
9861668f MH |
1811 | for (i = 0; i < num; i++) |
1812 | if (kps[i]->addr) | |
1813 | __unregister_kprobe_bottom(kps[i]); | |
1da177e4 | 1814 | } |
99081ab5 | 1815 | EXPORT_SYMBOL_GPL(unregister_kprobes); |
1da177e4 | 1816 | |
5f6bee34 NR |
1817 | int __weak kprobe_exceptions_notify(struct notifier_block *self, |
1818 | unsigned long val, void *data) | |
fc62d020 NR |
1819 | { |
1820 | return NOTIFY_DONE; | |
1821 | } | |
5f6bee34 | 1822 | NOKPROBE_SYMBOL(kprobe_exceptions_notify); |
fc62d020 | 1823 | |
1da177e4 | 1824 | static struct notifier_block kprobe_exceptions_nb = { |
3d5631e0 AK |
1825 | .notifier_call = kprobe_exceptions_notify, |
1826 | .priority = 0x7fffffff /* we need to be notified first */ | |
1827 | }; | |
1828 | ||
3d7e3382 ME |
1829 | unsigned long __weak arch_deref_entry_point(void *entry) |
1830 | { | |
1831 | return (unsigned long)entry; | |
1832 | } | |
1da177e4 | 1833 | |
9edddaa2 | 1834 | #ifdef CONFIG_KRETPROBES |
e65cefe8 AB |
1835 | /* |
1836 | * This kprobe pre_handler is registered with every kretprobe. When probe | |
1837 | * hits it will set up the return probe. | |
1838 | */ | |
820aede0 | 1839 | static int pre_handler_kretprobe(struct kprobe *p, struct pt_regs *regs) |
e65cefe8 AB |
1840 | { |
1841 | struct kretprobe *rp = container_of(p, struct kretprobe, kp); | |
ef53d9c5 S |
1842 | unsigned long hash, flags = 0; |
1843 | struct kretprobe_instance *ri; | |
e65cefe8 | 1844 | |
f96f5678 MH |
1845 | /* |
1846 | * To avoid deadlocks, prohibit return probing in NMI contexts, | |
1847 | * just skip the probe and increase the (inexact) 'nmissed' | |
1848 | * statistical counter, so that the user is informed that | |
1849 | * something happened: | |
1850 | */ | |
1851 | if (unlikely(in_nmi())) { | |
1852 | rp->nmissed++; | |
1853 | return 0; | |
1854 | } | |
1855 | ||
1856 | /* TODO: consider to only swap the RA after the last pre_handler fired */ | |
ef53d9c5 | 1857 | hash = hash_ptr(current, KPROBE_HASH_BITS); |
ec484608 | 1858 | raw_spin_lock_irqsave(&rp->lock, flags); |
4c4308cb | 1859 | if (!hlist_empty(&rp->free_instances)) { |
4c4308cb | 1860 | ri = hlist_entry(rp->free_instances.first, |
ef53d9c5 S |
1861 | struct kretprobe_instance, hlist); |
1862 | hlist_del(&ri->hlist); | |
ec484608 | 1863 | raw_spin_unlock_irqrestore(&rp->lock, flags); |
ef53d9c5 | 1864 | |
4c4308cb CH |
1865 | ri->rp = rp; |
1866 | ri->task = current; | |
f47cd9b5 | 1867 | |
55ca6140 JL |
1868 | if (rp->entry_handler && rp->entry_handler(ri, regs)) { |
1869 | raw_spin_lock_irqsave(&rp->lock, flags); | |
1870 | hlist_add_head(&ri->hlist, &rp->free_instances); | |
1871 | raw_spin_unlock_irqrestore(&rp->lock, flags); | |
f47cd9b5 | 1872 | return 0; |
55ca6140 | 1873 | } |
f47cd9b5 | 1874 | |
4c4308cb CH |
1875 | arch_prepare_kretprobe(ri, regs); |
1876 | ||
1877 | /* XXX(hch): why is there no hlist_move_head? */ | |
ef53d9c5 S |
1878 | INIT_HLIST_NODE(&ri->hlist); |
1879 | kretprobe_table_lock(hash, &flags); | |
1880 | hlist_add_head(&ri->hlist, &kretprobe_inst_table[hash]); | |
1881 | kretprobe_table_unlock(hash, &flags); | |
1882 | } else { | |
4c4308cb | 1883 | rp->nmissed++; |
ec484608 | 1884 | raw_spin_unlock_irqrestore(&rp->lock, flags); |
ef53d9c5 | 1885 | } |
e65cefe8 AB |
1886 | return 0; |
1887 | } | |
820aede0 | 1888 | NOKPROBE_SYMBOL(pre_handler_kretprobe); |
e65cefe8 | 1889 | |
659b957f | 1890 | bool __weak arch_kprobe_on_func_entry(unsigned long offset) |
90ec5e89 NR |
1891 | { |
1892 | return !offset; | |
1893 | } | |
1894 | ||
659b957f | 1895 | bool kprobe_on_func_entry(kprobe_opcode_t *addr, const char *sym, unsigned long offset) |
1d585e70 NR |
1896 | { |
1897 | kprobe_opcode_t *kp_addr = _kprobe_addr(addr, sym, offset); | |
1898 | ||
1899 | if (IS_ERR(kp_addr)) | |
1900 | return false; | |
1901 | ||
1902 | if (!kallsyms_lookup_size_offset((unsigned long)kp_addr, NULL, &offset) || | |
659b957f | 1903 | !arch_kprobe_on_func_entry(offset)) |
1d585e70 NR |
1904 | return false; |
1905 | ||
1906 | return true; | |
1907 | } | |
1908 | ||
55479f64 | 1909 | int register_kretprobe(struct kretprobe *rp) |
b94cce92 HN |
1910 | { |
1911 | int ret = 0; | |
1912 | struct kretprobe_instance *inst; | |
1913 | int i; | |
b2a5cd69 | 1914 | void *addr; |
90ec5e89 | 1915 | |
659b957f | 1916 | if (!kprobe_on_func_entry(rp->kp.addr, rp->kp.symbol_name, rp->kp.offset)) |
90ec5e89 | 1917 | return -EINVAL; |
f438d914 MH |
1918 | |
1919 | if (kretprobe_blacklist_size) { | |
b2a5cd69 | 1920 | addr = kprobe_addr(&rp->kp); |
bc81d48d MH |
1921 | if (IS_ERR(addr)) |
1922 | return PTR_ERR(addr); | |
f438d914 MH |
1923 | |
1924 | for (i = 0; kretprobe_blacklist[i].name != NULL; i++) { | |
1925 | if (kretprobe_blacklist[i].addr == addr) | |
1926 | return -EINVAL; | |
1927 | } | |
1928 | } | |
b94cce92 HN |
1929 | |
1930 | rp->kp.pre_handler = pre_handler_kretprobe; | |
7522a842 AM |
1931 | rp->kp.post_handler = NULL; |
1932 | rp->kp.fault_handler = NULL; | |
b94cce92 HN |
1933 | |
1934 | /* Pre-allocate memory for max kretprobe instances */ | |
1935 | if (rp->maxactive <= 0) { | |
92616606 | 1936 | #ifdef CONFIG_PREEMPTION |
c2ef6661 | 1937 | rp->maxactive = max_t(unsigned int, 10, 2*num_possible_cpus()); |
b94cce92 | 1938 | #else |
4dae560f | 1939 | rp->maxactive = num_possible_cpus(); |
b94cce92 HN |
1940 | #endif |
1941 | } | |
ec484608 | 1942 | raw_spin_lock_init(&rp->lock); |
b94cce92 HN |
1943 | INIT_HLIST_HEAD(&rp->free_instances); |
1944 | for (i = 0; i < rp->maxactive; i++) { | |
f47cd9b5 AS |
1945 | inst = kmalloc(sizeof(struct kretprobe_instance) + |
1946 | rp->data_size, GFP_KERNEL); | |
b94cce92 HN |
1947 | if (inst == NULL) { |
1948 | free_rp_inst(rp); | |
1949 | return -ENOMEM; | |
1950 | } | |
ef53d9c5 S |
1951 | INIT_HLIST_NODE(&inst->hlist); |
1952 | hlist_add_head(&inst->hlist, &rp->free_instances); | |
b94cce92 HN |
1953 | } |
1954 | ||
1955 | rp->nmissed = 0; | |
1956 | /* Establish function entry probe point */ | |
49ad2fd7 | 1957 | ret = register_kprobe(&rp->kp); |
4a296e07 | 1958 | if (ret != 0) |
b94cce92 HN |
1959 | free_rp_inst(rp); |
1960 | return ret; | |
1961 | } | |
99081ab5 | 1962 | EXPORT_SYMBOL_GPL(register_kretprobe); |
b94cce92 | 1963 | |
55479f64 | 1964 | int register_kretprobes(struct kretprobe **rps, int num) |
4a296e07 MH |
1965 | { |
1966 | int ret = 0, i; | |
1967 | ||
1968 | if (num <= 0) | |
1969 | return -EINVAL; | |
1970 | for (i = 0; i < num; i++) { | |
49ad2fd7 | 1971 | ret = register_kretprobe(rps[i]); |
67dddaad MH |
1972 | if (ret < 0) { |
1973 | if (i > 0) | |
1974 | unregister_kretprobes(rps, i); | |
4a296e07 MH |
1975 | break; |
1976 | } | |
1977 | } | |
1978 | return ret; | |
1979 | } | |
99081ab5 | 1980 | EXPORT_SYMBOL_GPL(register_kretprobes); |
4a296e07 | 1981 | |
55479f64 | 1982 | void unregister_kretprobe(struct kretprobe *rp) |
4a296e07 MH |
1983 | { |
1984 | unregister_kretprobes(&rp, 1); | |
1985 | } | |
99081ab5 | 1986 | EXPORT_SYMBOL_GPL(unregister_kretprobe); |
4a296e07 | 1987 | |
55479f64 | 1988 | void unregister_kretprobes(struct kretprobe **rps, int num) |
4a296e07 MH |
1989 | { |
1990 | int i; | |
1991 | ||
1992 | if (num <= 0) | |
1993 | return; | |
1994 | mutex_lock(&kprobe_mutex); | |
1995 | for (i = 0; i < num; i++) | |
1996 | if (__unregister_kprobe_top(&rps[i]->kp) < 0) | |
1997 | rps[i]->kp.addr = NULL; | |
1998 | mutex_unlock(&kprobe_mutex); | |
1999 | ||
ae8b7ce7 | 2000 | synchronize_rcu(); |
4a296e07 MH |
2001 | for (i = 0; i < num; i++) { |
2002 | if (rps[i]->kp.addr) { | |
2003 | __unregister_kprobe_bottom(&rps[i]->kp); | |
2004 | cleanup_rp_inst(rps[i]); | |
2005 | } | |
2006 | } | |
2007 | } | |
99081ab5 | 2008 | EXPORT_SYMBOL_GPL(unregister_kretprobes); |
4a296e07 | 2009 | |
9edddaa2 | 2010 | #else /* CONFIG_KRETPROBES */ |
55479f64 | 2011 | int register_kretprobe(struct kretprobe *rp) |
b94cce92 HN |
2012 | { |
2013 | return -ENOSYS; | |
2014 | } | |
99081ab5 | 2015 | EXPORT_SYMBOL_GPL(register_kretprobe); |
b94cce92 | 2016 | |
55479f64 | 2017 | int register_kretprobes(struct kretprobe **rps, int num) |
346fd59b | 2018 | { |
4a296e07 | 2019 | return -ENOSYS; |
346fd59b | 2020 | } |
99081ab5 MH |
2021 | EXPORT_SYMBOL_GPL(register_kretprobes); |
2022 | ||
55479f64 | 2023 | void unregister_kretprobe(struct kretprobe *rp) |
b94cce92 | 2024 | { |
4a296e07 | 2025 | } |
99081ab5 | 2026 | EXPORT_SYMBOL_GPL(unregister_kretprobe); |
b94cce92 | 2027 | |
55479f64 | 2028 | void unregister_kretprobes(struct kretprobe **rps, int num) |
4a296e07 MH |
2029 | { |
2030 | } | |
99081ab5 | 2031 | EXPORT_SYMBOL_GPL(unregister_kretprobes); |
4c4308cb | 2032 | |
820aede0 | 2033 | static int pre_handler_kretprobe(struct kprobe *p, struct pt_regs *regs) |
4a296e07 MH |
2034 | { |
2035 | return 0; | |
b94cce92 | 2036 | } |
820aede0 | 2037 | NOKPROBE_SYMBOL(pre_handler_kretprobe); |
b94cce92 | 2038 | |
4a296e07 MH |
2039 | #endif /* CONFIG_KRETPROBES */ |
2040 | ||
e8386a0c | 2041 | /* Set the kprobe gone and remove its instruction buffer. */ |
55479f64 | 2042 | static void kill_kprobe(struct kprobe *p) |
e8386a0c MH |
2043 | { |
2044 | struct kprobe *kp; | |
de5bd88d | 2045 | |
e8386a0c | 2046 | p->flags |= KPROBE_FLAG_GONE; |
afd66255 | 2047 | if (kprobe_aggrprobe(p)) { |
e8386a0c MH |
2048 | /* |
2049 | * If this is an aggr_kprobe, we have to list all the | |
2050 | * chained probes and mark them GONE. | |
2051 | */ | |
2052 | list_for_each_entry_rcu(kp, &p->list, list) | |
2053 | kp->flags |= KPROBE_FLAG_GONE; | |
2054 | p->post_handler = NULL; | |
afd66255 | 2055 | kill_optimized_kprobe(p); |
e8386a0c MH |
2056 | } |
2057 | /* | |
2058 | * Here, we can remove insn_slot safely, because no thread calls | |
2059 | * the original probed function (which will be freed soon) any more. | |
2060 | */ | |
2061 | arch_remove_kprobe(p); | |
2062 | } | |
2063 | ||
c0614829 | 2064 | /* Disable one kprobe */ |
55479f64 | 2065 | int disable_kprobe(struct kprobe *kp) |
c0614829 MH |
2066 | { |
2067 | int ret = 0; | |
297f9233 | 2068 | struct kprobe *p; |
c0614829 MH |
2069 | |
2070 | mutex_lock(&kprobe_mutex); | |
2071 | ||
6f0f1dd7 | 2072 | /* Disable this kprobe */ |
297f9233 JY |
2073 | p = __disable_kprobe(kp); |
2074 | if (IS_ERR(p)) | |
2075 | ret = PTR_ERR(p); | |
c0614829 | 2076 | |
c0614829 MH |
2077 | mutex_unlock(&kprobe_mutex); |
2078 | return ret; | |
2079 | } | |
2080 | EXPORT_SYMBOL_GPL(disable_kprobe); | |
2081 | ||
2082 | /* Enable one kprobe */ | |
55479f64 | 2083 | int enable_kprobe(struct kprobe *kp) |
c0614829 MH |
2084 | { |
2085 | int ret = 0; | |
2086 | struct kprobe *p; | |
2087 | ||
2088 | mutex_lock(&kprobe_mutex); | |
2089 | ||
2090 | /* Check whether specified probe is valid. */ | |
2091 | p = __get_valid_kprobe(kp); | |
2092 | if (unlikely(p == NULL)) { | |
2093 | ret = -EINVAL; | |
2094 | goto out; | |
2095 | } | |
2096 | ||
2097 | if (kprobe_gone(kp)) { | |
2098 | /* This kprobe has gone, we couldn't enable it. */ | |
2099 | ret = -EINVAL; | |
2100 | goto out; | |
2101 | } | |
2102 | ||
2103 | if (p != kp) | |
2104 | kp->flags &= ~KPROBE_FLAG_DISABLED; | |
2105 | ||
2106 | if (!kprobes_all_disarmed && kprobe_disabled(p)) { | |
2107 | p->flags &= ~KPROBE_FLAG_DISABLED; | |
12310e34 JY |
2108 | ret = arm_kprobe(p); |
2109 | if (ret) | |
2110 | p->flags |= KPROBE_FLAG_DISABLED; | |
c0614829 MH |
2111 | } |
2112 | out: | |
2113 | mutex_unlock(&kprobe_mutex); | |
2114 | return ret; | |
2115 | } | |
2116 | EXPORT_SYMBOL_GPL(enable_kprobe); | |
2117 | ||
4458515b | 2118 | /* Caller must NOT call this in usual path. This is only for critical case */ |
820aede0 | 2119 | void dump_kprobe(struct kprobe *kp) |
24851d24 | 2120 | { |
4458515b MH |
2121 | pr_err("Dumping kprobe:\n"); |
2122 | pr_err("Name: %s\nOffset: %x\nAddress: %pS\n", | |
2123 | kp->symbol_name, kp->offset, kp->addr); | |
24851d24 | 2124 | } |
820aede0 | 2125 | NOKPROBE_SYMBOL(dump_kprobe); |
24851d24 | 2126 | |
fb1a59fa MH |
2127 | int kprobe_add_ksym_blacklist(unsigned long entry) |
2128 | { | |
2129 | struct kprobe_blacklist_entry *ent; | |
2130 | unsigned long offset = 0, size = 0; | |
2131 | ||
2132 | if (!kernel_text_address(entry) || | |
2133 | !kallsyms_lookup_size_offset(entry, &size, &offset)) | |
2134 | return -EINVAL; | |
2135 | ||
2136 | ent = kmalloc(sizeof(*ent), GFP_KERNEL); | |
2137 | if (!ent) | |
2138 | return -ENOMEM; | |
2139 | ent->start_addr = entry; | |
2140 | ent->end_addr = entry + size; | |
2141 | INIT_LIST_HEAD(&ent->list); | |
2142 | list_add_tail(&ent->list, &kprobe_blacklist); | |
2143 | ||
2144 | return (int)size; | |
2145 | } | |
2146 | ||
2147 | /* Add all symbols in given area into kprobe blacklist */ | |
2148 | int kprobe_add_area_blacklist(unsigned long start, unsigned long end) | |
2149 | { | |
2150 | unsigned long entry; | |
2151 | int ret = 0; | |
2152 | ||
2153 | for (entry = start; entry < end; entry += ret) { | |
2154 | ret = kprobe_add_ksym_blacklist(entry); | |
2155 | if (ret < 0) | |
2156 | return ret; | |
2157 | if (ret == 0) /* In case of alias symbol */ | |
2158 | ret = 1; | |
2159 | } | |
2160 | return 0; | |
2161 | } | |
2162 | ||
2163 | int __init __weak arch_populate_kprobe_blacklist(void) | |
2164 | { | |
2165 | return 0; | |
2166 | } | |
2167 | ||
376e2424 MH |
2168 | /* |
2169 | * Lookup and populate the kprobe_blacklist. | |
2170 | * | |
2171 | * Unlike the kretprobe blacklist, we'll need to determine | |
2172 | * the range of addresses that belong to the said functions, | |
2173 | * since a kprobe need not necessarily be at the beginning | |
2174 | * of a function. | |
2175 | */ | |
2176 | static int __init populate_kprobe_blacklist(unsigned long *start, | |
2177 | unsigned long *end) | |
2178 | { | |
fb1a59fa | 2179 | unsigned long entry; |
376e2424 | 2180 | unsigned long *iter; |
fb1a59fa | 2181 | int ret; |
376e2424 MH |
2182 | |
2183 | for (iter = start; iter < end; iter++) { | |
d81b4253 | 2184 | entry = arch_deref_entry_point((void *)*iter); |
fb1a59fa MH |
2185 | ret = kprobe_add_ksym_blacklist(entry); |
2186 | if (ret == -EINVAL) | |
376e2424 | 2187 | continue; |
fb1a59fa MH |
2188 | if (ret < 0) |
2189 | return ret; | |
376e2424 | 2190 | } |
fb1a59fa MH |
2191 | |
2192 | /* Symbols in __kprobes_text are blacklisted */ | |
2193 | ret = kprobe_add_area_blacklist((unsigned long)__kprobes_text_start, | |
2194 | (unsigned long)__kprobes_text_end); | |
2195 | ||
2196 | return ret ? : arch_populate_kprobe_blacklist(); | |
376e2424 MH |
2197 | } |
2198 | ||
e8386a0c | 2199 | /* Module notifier call back, checking kprobes on the module */ |
55479f64 MH |
2200 | static int kprobes_module_callback(struct notifier_block *nb, |
2201 | unsigned long val, void *data) | |
e8386a0c MH |
2202 | { |
2203 | struct module *mod = data; | |
2204 | struct hlist_head *head; | |
e8386a0c MH |
2205 | struct kprobe *p; |
2206 | unsigned int i; | |
f24659d9 | 2207 | int checkcore = (val == MODULE_STATE_GOING); |
e8386a0c | 2208 | |
f24659d9 | 2209 | if (val != MODULE_STATE_GOING && val != MODULE_STATE_LIVE) |
e8386a0c MH |
2210 | return NOTIFY_DONE; |
2211 | ||
2212 | /* | |
f24659d9 MH |
2213 | * When MODULE_STATE_GOING was notified, both of module .text and |
2214 | * .init.text sections would be freed. When MODULE_STATE_LIVE was | |
2215 | * notified, only .init.text section would be freed. We need to | |
2216 | * disable kprobes which have been inserted in the sections. | |
e8386a0c MH |
2217 | */ |
2218 | mutex_lock(&kprobe_mutex); | |
2219 | for (i = 0; i < KPROBE_TABLE_SIZE; i++) { | |
2220 | head = &kprobe_table[i]; | |
b67bfe0d | 2221 | hlist_for_each_entry_rcu(p, head, hlist) |
f24659d9 MH |
2222 | if (within_module_init((unsigned long)p->addr, mod) || |
2223 | (checkcore && | |
2224 | within_module_core((unsigned long)p->addr, mod))) { | |
e8386a0c MH |
2225 | /* |
2226 | * The vaddr this probe is installed will soon | |
2227 | * be vfreed buy not synced to disk. Hence, | |
2228 | * disarming the breakpoint isn't needed. | |
545a0281 SRV |
2229 | * |
2230 | * Note, this will also move any optimized probes | |
2231 | * that are pending to be removed from their | |
2232 | * corresponding lists to the freeing_list and | |
2233 | * will not be touched by the delayed | |
2234 | * kprobe_optimizer work handler. | |
e8386a0c MH |
2235 | */ |
2236 | kill_kprobe(p); | |
2237 | } | |
2238 | } | |
2239 | mutex_unlock(&kprobe_mutex); | |
2240 | return NOTIFY_DONE; | |
2241 | } | |
2242 | ||
2243 | static struct notifier_block kprobe_module_nb = { | |
2244 | .notifier_call = kprobes_module_callback, | |
2245 | .priority = 0 | |
2246 | }; | |
2247 | ||
376e2424 MH |
2248 | /* Markers of _kprobe_blacklist section */ |
2249 | extern unsigned long __start_kprobe_blacklist[]; | |
2250 | extern unsigned long __stop_kprobe_blacklist[]; | |
2251 | ||
1da177e4 LT |
2252 | static int __init init_kprobes(void) |
2253 | { | |
2254 | int i, err = 0; | |
2255 | ||
2256 | /* FIXME allocate the probe table, currently defined statically */ | |
2257 | /* initialize all list heads */ | |
b94cce92 | 2258 | for (i = 0; i < KPROBE_TABLE_SIZE; i++) { |
1da177e4 | 2259 | INIT_HLIST_HEAD(&kprobe_table[i]); |
b94cce92 | 2260 | INIT_HLIST_HEAD(&kretprobe_inst_table[i]); |
ec484608 | 2261 | raw_spin_lock_init(&(kretprobe_table_locks[i].lock)); |
b94cce92 | 2262 | } |
1da177e4 | 2263 | |
376e2424 MH |
2264 | err = populate_kprobe_blacklist(__start_kprobe_blacklist, |
2265 | __stop_kprobe_blacklist); | |
2266 | if (err) { | |
2267 | pr_err("kprobes: failed to populate blacklist: %d\n", err); | |
2268 | pr_err("Please take care of using kprobes.\n"); | |
3d8d996e SD |
2269 | } |
2270 | ||
f438d914 MH |
2271 | if (kretprobe_blacklist_size) { |
2272 | /* lookup the function address from its name */ | |
2273 | for (i = 0; kretprobe_blacklist[i].name != NULL; i++) { | |
49e0b465 | 2274 | kretprobe_blacklist[i].addr = |
290e3070 | 2275 | kprobe_lookup_name(kretprobe_blacklist[i].name, 0); |
f438d914 MH |
2276 | if (!kretprobe_blacklist[i].addr) |
2277 | printk("kretprobe: lookup failed: %s\n", | |
2278 | kretprobe_blacklist[i].name); | |
2279 | } | |
2280 | } | |
2281 | ||
b2be84df MH |
2282 | #if defined(CONFIG_OPTPROBES) |
2283 | #if defined(__ARCH_WANT_KPROBES_INSN_SLOT) | |
afd66255 MH |
2284 | /* Init kprobe_optinsn_slots */ |
2285 | kprobe_optinsn_slots.insn_size = MAX_OPTINSN_SIZE; | |
2286 | #endif | |
b2be84df MH |
2287 | /* By default, kprobes can be optimized */ |
2288 | kprobes_allow_optimization = true; | |
2289 | #endif | |
afd66255 | 2290 | |
e579abeb MH |
2291 | /* By default, kprobes are armed */ |
2292 | kprobes_all_disarmed = false; | |
bf8f6e5b | 2293 | |
6772926b | 2294 | err = arch_init_kprobes(); |
802eae7c RL |
2295 | if (!err) |
2296 | err = register_die_notifier(&kprobe_exceptions_nb); | |
e8386a0c MH |
2297 | if (!err) |
2298 | err = register_module_notifier(&kprobe_module_nb); | |
2299 | ||
ef53d9c5 | 2300 | kprobes_initialized = (err == 0); |
802eae7c | 2301 | |
8c1c9356 AM |
2302 | if (!err) |
2303 | init_test_probes(); | |
1da177e4 LT |
2304 | return err; |
2305 | } | |
65fc965c | 2306 | subsys_initcall(init_kprobes); |
1da177e4 | 2307 | |
346fd59b | 2308 | #ifdef CONFIG_DEBUG_FS |
55479f64 | 2309 | static void report_probe(struct seq_file *pi, struct kprobe *p, |
afd66255 | 2310 | const char *sym, int offset, char *modname, struct kprobe *pp) |
346fd59b SD |
2311 | { |
2312 | char *kprobe_type; | |
81365a94 | 2313 | void *addr = p->addr; |
346fd59b SD |
2314 | |
2315 | if (p->pre_handler == pre_handler_kretprobe) | |
2316 | kprobe_type = "r"; | |
346fd59b SD |
2317 | else |
2318 | kprobe_type = "k"; | |
afd66255 | 2319 | |
81365a94 MH |
2320 | if (!kallsyms_show_value()) |
2321 | addr = NULL; | |
2322 | ||
346fd59b | 2323 | if (sym) |
81365a94 MH |
2324 | seq_printf(pi, "%px %s %s+0x%x %s ", |
2325 | addr, kprobe_type, sym, offset, | |
afd66255 | 2326 | (modname ? modname : " ")); |
81365a94 MH |
2327 | else /* try to use %pS */ |
2328 | seq_printf(pi, "%px %s %pS ", | |
2329 | addr, kprobe_type, p->addr); | |
afd66255 MH |
2330 | |
2331 | if (!pp) | |
2332 | pp = p; | |
ae6aa16f | 2333 | seq_printf(pi, "%s%s%s%s\n", |
afd66255 MH |
2334 | (kprobe_gone(p) ? "[GONE]" : ""), |
2335 | ((kprobe_disabled(p) && !kprobe_gone(p)) ? "[DISABLED]" : ""), | |
ae6aa16f MH |
2336 | (kprobe_optimized(pp) ? "[OPTIMIZED]" : ""), |
2337 | (kprobe_ftrace(pp) ? "[FTRACE]" : "")); | |
346fd59b SD |
2338 | } |
2339 | ||
55479f64 | 2340 | static void *kprobe_seq_start(struct seq_file *f, loff_t *pos) |
346fd59b SD |
2341 | { |
2342 | return (*pos < KPROBE_TABLE_SIZE) ? pos : NULL; | |
2343 | } | |
2344 | ||
55479f64 | 2345 | static void *kprobe_seq_next(struct seq_file *f, void *v, loff_t *pos) |
346fd59b SD |
2346 | { |
2347 | (*pos)++; | |
2348 | if (*pos >= KPROBE_TABLE_SIZE) | |
2349 | return NULL; | |
2350 | return pos; | |
2351 | } | |
2352 | ||
55479f64 | 2353 | static void kprobe_seq_stop(struct seq_file *f, void *v) |
346fd59b SD |
2354 | { |
2355 | /* Nothing to do */ | |
2356 | } | |
2357 | ||
55479f64 | 2358 | static int show_kprobe_addr(struct seq_file *pi, void *v) |
346fd59b SD |
2359 | { |
2360 | struct hlist_head *head; | |
346fd59b SD |
2361 | struct kprobe *p, *kp; |
2362 | const char *sym = NULL; | |
2363 | unsigned int i = *(loff_t *) v; | |
ffb45122 | 2364 | unsigned long offset = 0; |
ab767865 | 2365 | char *modname, namebuf[KSYM_NAME_LEN]; |
346fd59b SD |
2366 | |
2367 | head = &kprobe_table[i]; | |
2368 | preempt_disable(); | |
b67bfe0d | 2369 | hlist_for_each_entry_rcu(p, head, hlist) { |
ffb45122 | 2370 | sym = kallsyms_lookup((unsigned long)p->addr, NULL, |
346fd59b | 2371 | &offset, &modname, namebuf); |
afd66255 | 2372 | if (kprobe_aggrprobe(p)) { |
346fd59b | 2373 | list_for_each_entry_rcu(kp, &p->list, list) |
afd66255 | 2374 | report_probe(pi, kp, sym, offset, modname, p); |
346fd59b | 2375 | } else |
afd66255 | 2376 | report_probe(pi, p, sym, offset, modname, NULL); |
346fd59b SD |
2377 | } |
2378 | preempt_enable(); | |
2379 | return 0; | |
2380 | } | |
2381 | ||
88e9d34c | 2382 | static const struct seq_operations kprobes_seq_ops = { |
346fd59b SD |
2383 | .start = kprobe_seq_start, |
2384 | .next = kprobe_seq_next, | |
2385 | .stop = kprobe_seq_stop, | |
2386 | .show = show_kprobe_addr | |
2387 | }; | |
2388 | ||
55479f64 | 2389 | static int kprobes_open(struct inode *inode, struct file *filp) |
346fd59b SD |
2390 | { |
2391 | return seq_open(filp, &kprobes_seq_ops); | |
2392 | } | |
2393 | ||
828c0950 | 2394 | static const struct file_operations debugfs_kprobes_operations = { |
346fd59b SD |
2395 | .open = kprobes_open, |
2396 | .read = seq_read, | |
2397 | .llseek = seq_lseek, | |
2398 | .release = seq_release, | |
2399 | }; | |
2400 | ||
63724740 MH |
2401 | /* kprobes/blacklist -- shows which functions can not be probed */ |
2402 | static void *kprobe_blacklist_seq_start(struct seq_file *m, loff_t *pos) | |
2403 | { | |
2404 | return seq_list_start(&kprobe_blacklist, *pos); | |
2405 | } | |
2406 | ||
2407 | static void *kprobe_blacklist_seq_next(struct seq_file *m, void *v, loff_t *pos) | |
2408 | { | |
2409 | return seq_list_next(v, &kprobe_blacklist, pos); | |
2410 | } | |
2411 | ||
2412 | static int kprobe_blacklist_seq_show(struct seq_file *m, void *v) | |
2413 | { | |
2414 | struct kprobe_blacklist_entry *ent = | |
2415 | list_entry(v, struct kprobe_blacklist_entry, list); | |
2416 | ||
ffb9bd68 MH |
2417 | /* |
2418 | * If /proc/kallsyms is not showing kernel address, we won't | |
2419 | * show them here either. | |
2420 | */ | |
2421 | if (!kallsyms_show_value()) | |
2422 | seq_printf(m, "0x%px-0x%px\t%ps\n", NULL, NULL, | |
2423 | (void *)ent->start_addr); | |
2424 | else | |
2425 | seq_printf(m, "0x%px-0x%px\t%ps\n", (void *)ent->start_addr, | |
2426 | (void *)ent->end_addr, (void *)ent->start_addr); | |
63724740 MH |
2427 | return 0; |
2428 | } | |
2429 | ||
2430 | static const struct seq_operations kprobe_blacklist_seq_ops = { | |
2431 | .start = kprobe_blacklist_seq_start, | |
2432 | .next = kprobe_blacklist_seq_next, | |
2433 | .stop = kprobe_seq_stop, /* Reuse void function */ | |
2434 | .show = kprobe_blacklist_seq_show, | |
2435 | }; | |
2436 | ||
2437 | static int kprobe_blacklist_open(struct inode *inode, struct file *filp) | |
2438 | { | |
2439 | return seq_open(filp, &kprobe_blacklist_seq_ops); | |
2440 | } | |
2441 | ||
2442 | static const struct file_operations debugfs_kprobe_blacklist_ops = { | |
2443 | .open = kprobe_blacklist_open, | |
2444 | .read = seq_read, | |
2445 | .llseek = seq_lseek, | |
2446 | .release = seq_release, | |
2447 | }; | |
2448 | ||
12310e34 | 2449 | static int arm_all_kprobes(void) |
bf8f6e5b AM |
2450 | { |
2451 | struct hlist_head *head; | |
bf8f6e5b | 2452 | struct kprobe *p; |
12310e34 JY |
2453 | unsigned int i, total = 0, errors = 0; |
2454 | int err, ret = 0; | |
bf8f6e5b AM |
2455 | |
2456 | mutex_lock(&kprobe_mutex); | |
2457 | ||
e579abeb MH |
2458 | /* If kprobes are armed, just return */ |
2459 | if (!kprobes_all_disarmed) | |
bf8f6e5b AM |
2460 | goto already_enabled; |
2461 | ||
977ad481 WN |
2462 | /* |
2463 | * optimize_kprobe() called by arm_kprobe() checks | |
2464 | * kprobes_all_disarmed, so set kprobes_all_disarmed before | |
2465 | * arm_kprobe. | |
2466 | */ | |
2467 | kprobes_all_disarmed = false; | |
afd66255 | 2468 | /* Arming kprobes doesn't optimize kprobe itself */ |
bf8f6e5b AM |
2469 | for (i = 0; i < KPROBE_TABLE_SIZE; i++) { |
2470 | head = &kprobe_table[i]; | |
12310e34 JY |
2471 | /* Arm all kprobes on a best-effort basis */ |
2472 | hlist_for_each_entry_rcu(p, head, hlist) { | |
2473 | if (!kprobe_disabled(p)) { | |
2474 | err = arm_kprobe(p); | |
2475 | if (err) { | |
2476 | errors++; | |
2477 | ret = err; | |
2478 | } | |
2479 | total++; | |
2480 | } | |
2481 | } | |
bf8f6e5b AM |
2482 | } |
2483 | ||
12310e34 JY |
2484 | if (errors) |
2485 | pr_warn("Kprobes globally enabled, but failed to arm %d out of %d probes\n", | |
2486 | errors, total); | |
2487 | else | |
2488 | pr_info("Kprobes globally enabled\n"); | |
bf8f6e5b AM |
2489 | |
2490 | already_enabled: | |
2491 | mutex_unlock(&kprobe_mutex); | |
12310e34 | 2492 | return ret; |
bf8f6e5b AM |
2493 | } |
2494 | ||
297f9233 | 2495 | static int disarm_all_kprobes(void) |
bf8f6e5b AM |
2496 | { |
2497 | struct hlist_head *head; | |
bf8f6e5b | 2498 | struct kprobe *p; |
297f9233 JY |
2499 | unsigned int i, total = 0, errors = 0; |
2500 | int err, ret = 0; | |
bf8f6e5b AM |
2501 | |
2502 | mutex_lock(&kprobe_mutex); | |
2503 | ||
e579abeb | 2504 | /* If kprobes are already disarmed, just return */ |
6274de49 MH |
2505 | if (kprobes_all_disarmed) { |
2506 | mutex_unlock(&kprobe_mutex); | |
297f9233 | 2507 | return 0; |
6274de49 | 2508 | } |
bf8f6e5b | 2509 | |
e579abeb | 2510 | kprobes_all_disarmed = true; |
afd66255 | 2511 | |
bf8f6e5b AM |
2512 | for (i = 0; i < KPROBE_TABLE_SIZE; i++) { |
2513 | head = &kprobe_table[i]; | |
297f9233 | 2514 | /* Disarm all kprobes on a best-effort basis */ |
b67bfe0d | 2515 | hlist_for_each_entry_rcu(p, head, hlist) { |
297f9233 JY |
2516 | if (!arch_trampoline_kprobe(p) && !kprobe_disabled(p)) { |
2517 | err = disarm_kprobe(p, false); | |
2518 | if (err) { | |
2519 | errors++; | |
2520 | ret = err; | |
2521 | } | |
2522 | total++; | |
2523 | } | |
bf8f6e5b AM |
2524 | } |
2525 | } | |
297f9233 JY |
2526 | |
2527 | if (errors) | |
2528 | pr_warn("Kprobes globally disabled, but failed to disarm %d out of %d probes\n", | |
2529 | errors, total); | |
2530 | else | |
2531 | pr_info("Kprobes globally disabled\n"); | |
2532 | ||
bf8f6e5b | 2533 | mutex_unlock(&kprobe_mutex); |
bf8f6e5b | 2534 | |
6274de49 MH |
2535 | /* Wait for disarming all kprobes by optimizer */ |
2536 | wait_for_kprobe_optimizer(); | |
297f9233 JY |
2537 | |
2538 | return ret; | |
bf8f6e5b AM |
2539 | } |
2540 | ||
2541 | /* | |
2542 | * XXX: The debugfs bool file interface doesn't allow for callbacks | |
2543 | * when the bool state is switched. We can reuse that facility when | |
2544 | * available | |
2545 | */ | |
2546 | static ssize_t read_enabled_file_bool(struct file *file, | |
2547 | char __user *user_buf, size_t count, loff_t *ppos) | |
2548 | { | |
2549 | char buf[3]; | |
2550 | ||
e579abeb | 2551 | if (!kprobes_all_disarmed) |
bf8f6e5b AM |
2552 | buf[0] = '1'; |
2553 | else | |
2554 | buf[0] = '0'; | |
2555 | buf[1] = '\n'; | |
2556 | buf[2] = 0x00; | |
2557 | return simple_read_from_buffer(user_buf, count, ppos, buf, 2); | |
2558 | } | |
2559 | ||
2560 | static ssize_t write_enabled_file_bool(struct file *file, | |
2561 | const char __user *user_buf, size_t count, loff_t *ppos) | |
2562 | { | |
2563 | char buf[32]; | |
efeb156e | 2564 | size_t buf_size; |
12310e34 | 2565 | int ret = 0; |
bf8f6e5b AM |
2566 | |
2567 | buf_size = min(count, (sizeof(buf)-1)); | |
2568 | if (copy_from_user(buf, user_buf, buf_size)) | |
2569 | return -EFAULT; | |
2570 | ||
10fb46d5 | 2571 | buf[buf_size] = '\0'; |
bf8f6e5b AM |
2572 | switch (buf[0]) { |
2573 | case 'y': | |
2574 | case 'Y': | |
2575 | case '1': | |
12310e34 | 2576 | ret = arm_all_kprobes(); |
bf8f6e5b AM |
2577 | break; |
2578 | case 'n': | |
2579 | case 'N': | |
2580 | case '0': | |
297f9233 | 2581 | ret = disarm_all_kprobes(); |
bf8f6e5b | 2582 | break; |
10fb46d5 MK |
2583 | default: |
2584 | return -EINVAL; | |
bf8f6e5b AM |
2585 | } |
2586 | ||
12310e34 JY |
2587 | if (ret) |
2588 | return ret; | |
2589 | ||
bf8f6e5b AM |
2590 | return count; |
2591 | } | |
2592 | ||
828c0950 | 2593 | static const struct file_operations fops_kp = { |
bf8f6e5b AM |
2594 | .read = read_enabled_file_bool, |
2595 | .write = write_enabled_file_bool, | |
6038f373 | 2596 | .llseek = default_llseek, |
bf8f6e5b AM |
2597 | }; |
2598 | ||
55479f64 | 2599 | static int __init debugfs_kprobe_init(void) |
346fd59b | 2600 | { |
8c0fd1fa | 2601 | struct dentry *dir; |
bf8f6e5b | 2602 | unsigned int value = 1; |
346fd59b SD |
2603 | |
2604 | dir = debugfs_create_dir("kprobes", NULL); | |
346fd59b | 2605 | |
8c0fd1fa GKH |
2606 | debugfs_create_file("list", 0400, dir, NULL, |
2607 | &debugfs_kprobes_operations); | |
346fd59b | 2608 | |
8c0fd1fa | 2609 | debugfs_create_file("enabled", 0600, dir, &value, &fops_kp); |
63724740 | 2610 | |
8c0fd1fa GKH |
2611 | debugfs_create_file("blacklist", 0400, dir, NULL, |
2612 | &debugfs_kprobe_blacklist_ops); | |
bf8f6e5b | 2613 | |
346fd59b SD |
2614 | return 0; |
2615 | } | |
2616 | ||
2617 | late_initcall(debugfs_kprobe_init); | |
2618 | #endif /* CONFIG_DEBUG_FS */ |