]>
Commit | Line | Data |
---|---|---|
1da177e4 LT |
1 | /* |
2 | * Kernel Probes (KProbes) | |
3 | * kernel/kprobes.c | |
4 | * | |
5 | * This program is free software; you can redistribute it and/or modify | |
6 | * it under the terms of the GNU General Public License as published by | |
7 | * the Free Software Foundation; either version 2 of the License, or | |
8 | * (at your option) any later version. | |
9 | * | |
10 | * This program is distributed in the hope that it will be useful, | |
11 | * but WITHOUT ANY WARRANTY; without even the implied warranty of | |
12 | * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the | |
13 | * GNU General Public License for more details. | |
14 | * | |
15 | * You should have received a copy of the GNU General Public License | |
16 | * along with this program; if not, write to the Free Software | |
17 | * Foundation, Inc., 59 Temple Place - Suite 330, Boston, MA 02111-1307, USA. | |
18 | * | |
19 | * Copyright (C) IBM Corporation, 2002, 2004 | |
20 | * | |
21 | * 2002-Oct Created by Vamsi Krishna S <vamsi_krishna@in.ibm.com> Kernel | |
22 | * Probes initial implementation (includes suggestions from | |
23 | * Rusty Russell). | |
24 | * 2004-Aug Updated by Prasanna S Panchamukhi <prasanna@in.ibm.com> with | |
25 | * hlists and exceptions notifier as suggested by Andi Kleen. | |
26 | * 2004-July Suparna Bhattacharya <suparna@in.ibm.com> added jumper probes | |
27 | * interface to access function arguments. | |
28 | * 2004-Sep Prasanna S Panchamukhi <prasanna@in.ibm.com> Changed Kprobes | |
29 | * exceptions notifier to be first on the priority list. | |
b94cce92 HN |
30 | * 2005-May Hien Nguyen <hien@us.ibm.com>, Jim Keniston |
31 | * <jkenisto@us.ibm.com> and Prasanna S Panchamukhi | |
32 | * <prasanna@in.ibm.com> added function-return probes. | |
1da177e4 LT |
33 | */ |
34 | #include <linux/kprobes.h> | |
1da177e4 LT |
35 | #include <linux/hash.h> |
36 | #include <linux/init.h> | |
4e57b681 | 37 | #include <linux/slab.h> |
e3869792 | 38 | #include <linux/stddef.h> |
9984de1a | 39 | #include <linux/export.h> |
9ec4b1f3 | 40 | #include <linux/moduleloader.h> |
3a872d89 | 41 | #include <linux/kallsyms.h> |
b4c6c34a | 42 | #include <linux/freezer.h> |
346fd59b SD |
43 | #include <linux/seq_file.h> |
44 | #include <linux/debugfs.h> | |
b2be84df | 45 | #include <linux/sysctl.h> |
1eeb66a1 | 46 | #include <linux/kdebug.h> |
4460fdad | 47 | #include <linux/memory.h> |
4554dbcb | 48 | #include <linux/ftrace.h> |
afd66255 | 49 | #include <linux/cpu.h> |
bf5438fc | 50 | #include <linux/jump_label.h> |
bf8f6e5b | 51 | |
d0aaff97 | 52 | #include <asm-generic/sections.h> |
1da177e4 LT |
53 | #include <asm/cacheflush.h> |
54 | #include <asm/errno.h> | |
bf8f6e5b | 55 | #include <asm/uaccess.h> |
1da177e4 LT |
56 | |
57 | #define KPROBE_HASH_BITS 6 | |
58 | #define KPROBE_TABLE_SIZE (1 << KPROBE_HASH_BITS) | |
59 | ||
3a872d89 AM |
60 | |
61 | /* | |
62 | * Some oddball architectures like 64bit powerpc have function descriptors | |
63 | * so this must be overridable. | |
64 | */ | |
65 | #ifndef kprobe_lookup_name | |
66 | #define kprobe_lookup_name(name, addr) \ | |
67 | addr = ((kprobe_opcode_t *)(kallsyms_lookup_name(name))) | |
68 | #endif | |
69 | ||
ef53d9c5 | 70 | static int kprobes_initialized; |
1da177e4 | 71 | static struct hlist_head kprobe_table[KPROBE_TABLE_SIZE]; |
b94cce92 | 72 | static struct hlist_head kretprobe_inst_table[KPROBE_TABLE_SIZE]; |
1da177e4 | 73 | |
bf8f6e5b | 74 | /* NOTE: change this value only with kprobe_mutex held */ |
e579abeb | 75 | static bool kprobes_all_disarmed; |
bf8f6e5b | 76 | |
43948f50 MH |
77 | /* This protects kprobe_table and optimizing_list */ |
78 | static DEFINE_MUTEX(kprobe_mutex); | |
e6584523 | 79 | static DEFINE_PER_CPU(struct kprobe *, kprobe_instance) = NULL; |
ef53d9c5 | 80 | static struct { |
ec484608 | 81 | raw_spinlock_t lock ____cacheline_aligned_in_smp; |
ef53d9c5 S |
82 | } kretprobe_table_locks[KPROBE_TABLE_SIZE]; |
83 | ||
ec484608 | 84 | static raw_spinlock_t *kretprobe_table_lock_ptr(unsigned long hash) |
ef53d9c5 S |
85 | { |
86 | return &(kretprobe_table_locks[hash].lock); | |
87 | } | |
1da177e4 | 88 | |
376e2424 MH |
89 | /* Blacklist -- list of struct kprobe_blacklist_entry */ |
90 | static LIST_HEAD(kprobe_blacklist); | |
3d8d996e | 91 | |
2d14e39d | 92 | #ifdef __ARCH_WANT_KPROBES_INSN_SLOT |
9ec4b1f3 AM |
93 | /* |
94 | * kprobe->ainsn.insn points to the copy of the instruction to be | |
95 | * single-stepped. x86_64, POWER4 and above have no-exec support and | |
96 | * stepping on the instruction on a vmalloced/kmalloced/data page | |
97 | * is a recipe for disaster | |
98 | */ | |
9ec4b1f3 | 99 | struct kprobe_insn_page { |
c5cb5a2d | 100 | struct list_head list; |
9ec4b1f3 | 101 | kprobe_opcode_t *insns; /* Page of instruction slots */ |
af96397d | 102 | struct kprobe_insn_cache *cache; |
9ec4b1f3 | 103 | int nused; |
b4c6c34a | 104 | int ngarbage; |
4610ee1d | 105 | char slot_used[]; |
9ec4b1f3 AM |
106 | }; |
107 | ||
4610ee1d MH |
108 | #define KPROBE_INSN_PAGE_SIZE(slots) \ |
109 | (offsetof(struct kprobe_insn_page, slot_used) + \ | |
110 | (sizeof(char) * (slots))) | |
111 | ||
4610ee1d MH |
112 | static int slots_per_page(struct kprobe_insn_cache *c) |
113 | { | |
114 | return PAGE_SIZE/(c->insn_size * sizeof(kprobe_opcode_t)); | |
115 | } | |
116 | ||
ab40c5c6 MH |
117 | enum kprobe_slot_state { |
118 | SLOT_CLEAN = 0, | |
119 | SLOT_DIRTY = 1, | |
120 | SLOT_USED = 2, | |
121 | }; | |
122 | ||
af96397d HC |
123 | static void *alloc_insn_page(void) |
124 | { | |
125 | return module_alloc(PAGE_SIZE); | |
126 | } | |
127 | ||
128 | static void free_insn_page(void *page) | |
129 | { | |
130 | module_free(NULL, page); | |
131 | } | |
132 | ||
c802d64a HC |
133 | struct kprobe_insn_cache kprobe_insn_slots = { |
134 | .mutex = __MUTEX_INITIALIZER(kprobe_insn_slots.mutex), | |
af96397d HC |
135 | .alloc = alloc_insn_page, |
136 | .free = free_insn_page, | |
4610ee1d MH |
137 | .pages = LIST_HEAD_INIT(kprobe_insn_slots.pages), |
138 | .insn_size = MAX_INSN_SIZE, | |
139 | .nr_garbage = 0, | |
140 | }; | |
55479f64 | 141 | static int collect_garbage_slots(struct kprobe_insn_cache *c); |
b4c6c34a | 142 | |
9ec4b1f3 | 143 | /** |
12941560 | 144 | * __get_insn_slot() - Find a slot on an executable page for an instruction. |
9ec4b1f3 AM |
145 | * We allocate an executable page if there's no room on existing ones. |
146 | */ | |
55479f64 | 147 | kprobe_opcode_t *__get_insn_slot(struct kprobe_insn_cache *c) |
9ec4b1f3 AM |
148 | { |
149 | struct kprobe_insn_page *kip; | |
c802d64a | 150 | kprobe_opcode_t *slot = NULL; |
9ec4b1f3 | 151 | |
c802d64a | 152 | mutex_lock(&c->mutex); |
6f716acd | 153 | retry: |
4610ee1d MH |
154 | list_for_each_entry(kip, &c->pages, list) { |
155 | if (kip->nused < slots_per_page(c)) { | |
9ec4b1f3 | 156 | int i; |
4610ee1d | 157 | for (i = 0; i < slots_per_page(c); i++) { |
ab40c5c6 MH |
158 | if (kip->slot_used[i] == SLOT_CLEAN) { |
159 | kip->slot_used[i] = SLOT_USED; | |
9ec4b1f3 | 160 | kip->nused++; |
c802d64a HC |
161 | slot = kip->insns + (i * c->insn_size); |
162 | goto out; | |
9ec4b1f3 AM |
163 | } |
164 | } | |
4610ee1d MH |
165 | /* kip->nused is broken. Fix it. */ |
166 | kip->nused = slots_per_page(c); | |
167 | WARN_ON(1); | |
9ec4b1f3 AM |
168 | } |
169 | } | |
170 | ||
b4c6c34a | 171 | /* If there are any garbage slots, collect it and try again. */ |
4610ee1d | 172 | if (c->nr_garbage && collect_garbage_slots(c) == 0) |
b4c6c34a | 173 | goto retry; |
4610ee1d MH |
174 | |
175 | /* All out of space. Need to allocate a new page. */ | |
176 | kip = kmalloc(KPROBE_INSN_PAGE_SIZE(slots_per_page(c)), GFP_KERNEL); | |
6f716acd | 177 | if (!kip) |
c802d64a | 178 | goto out; |
9ec4b1f3 AM |
179 | |
180 | /* | |
181 | * Use module_alloc so this page is within +/- 2GB of where the | |
182 | * kernel image and loaded module images reside. This is required | |
183 | * so x86_64 can correctly handle the %rip-relative fixups. | |
184 | */ | |
af96397d | 185 | kip->insns = c->alloc(); |
9ec4b1f3 AM |
186 | if (!kip->insns) { |
187 | kfree(kip); | |
c802d64a | 188 | goto out; |
9ec4b1f3 | 189 | } |
c5cb5a2d | 190 | INIT_LIST_HEAD(&kip->list); |
4610ee1d | 191 | memset(kip->slot_used, SLOT_CLEAN, slots_per_page(c)); |
ab40c5c6 | 192 | kip->slot_used[0] = SLOT_USED; |
9ec4b1f3 | 193 | kip->nused = 1; |
b4c6c34a | 194 | kip->ngarbage = 0; |
af96397d | 195 | kip->cache = c; |
4610ee1d | 196 | list_add(&kip->list, &c->pages); |
c802d64a HC |
197 | slot = kip->insns; |
198 | out: | |
199 | mutex_unlock(&c->mutex); | |
200 | return slot; | |
12941560 MH |
201 | } |
202 | ||
b4c6c34a | 203 | /* Return 1 if all garbages are collected, otherwise 0. */ |
55479f64 | 204 | static int collect_one_slot(struct kprobe_insn_page *kip, int idx) |
b4c6c34a | 205 | { |
ab40c5c6 | 206 | kip->slot_used[idx] = SLOT_CLEAN; |
b4c6c34a MH |
207 | kip->nused--; |
208 | if (kip->nused == 0) { | |
209 | /* | |
210 | * Page is no longer in use. Free it unless | |
211 | * it's the last one. We keep the last one | |
212 | * so as not to have to set it up again the | |
213 | * next time somebody inserts a probe. | |
214 | */ | |
4610ee1d | 215 | if (!list_is_singular(&kip->list)) { |
c5cb5a2d | 216 | list_del(&kip->list); |
af96397d | 217 | kip->cache->free(kip->insns); |
b4c6c34a MH |
218 | kfree(kip); |
219 | } | |
220 | return 1; | |
221 | } | |
222 | return 0; | |
223 | } | |
224 | ||
55479f64 | 225 | static int collect_garbage_slots(struct kprobe_insn_cache *c) |
b4c6c34a | 226 | { |
c5cb5a2d | 227 | struct kprobe_insn_page *kip, *next; |
b4c6c34a | 228 | |
615d0ebb MH |
229 | /* Ensure no-one is interrupted on the garbages */ |
230 | synchronize_sched(); | |
b4c6c34a | 231 | |
4610ee1d | 232 | list_for_each_entry_safe(kip, next, &c->pages, list) { |
b4c6c34a | 233 | int i; |
b4c6c34a MH |
234 | if (kip->ngarbage == 0) |
235 | continue; | |
236 | kip->ngarbage = 0; /* we will collect all garbages */ | |
4610ee1d | 237 | for (i = 0; i < slots_per_page(c); i++) { |
ab40c5c6 | 238 | if (kip->slot_used[i] == SLOT_DIRTY && |
b4c6c34a MH |
239 | collect_one_slot(kip, i)) |
240 | break; | |
241 | } | |
242 | } | |
4610ee1d | 243 | c->nr_garbage = 0; |
b4c6c34a MH |
244 | return 0; |
245 | } | |
246 | ||
55479f64 MH |
247 | void __free_insn_slot(struct kprobe_insn_cache *c, |
248 | kprobe_opcode_t *slot, int dirty) | |
9ec4b1f3 AM |
249 | { |
250 | struct kprobe_insn_page *kip; | |
9ec4b1f3 | 251 | |
c802d64a | 252 | mutex_lock(&c->mutex); |
4610ee1d | 253 | list_for_each_entry(kip, &c->pages, list) { |
83ff56f4 MH |
254 | long idx = ((long)slot - (long)kip->insns) / |
255 | (c->insn_size * sizeof(kprobe_opcode_t)); | |
4610ee1d MH |
256 | if (idx >= 0 && idx < slots_per_page(c)) { |
257 | WARN_ON(kip->slot_used[idx] != SLOT_USED); | |
b4c6c34a | 258 | if (dirty) { |
4610ee1d | 259 | kip->slot_used[idx] = SLOT_DIRTY; |
b4c6c34a | 260 | kip->ngarbage++; |
4610ee1d MH |
261 | if (++c->nr_garbage > slots_per_page(c)) |
262 | collect_garbage_slots(c); | |
c5cb5a2d | 263 | } else |
4610ee1d | 264 | collect_one_slot(kip, idx); |
c802d64a | 265 | goto out; |
9ec4b1f3 AM |
266 | } |
267 | } | |
4610ee1d MH |
268 | /* Could not free this slot. */ |
269 | WARN_ON(1); | |
c802d64a HC |
270 | out: |
271 | mutex_unlock(&c->mutex); | |
4610ee1d | 272 | } |
6f716acd | 273 | |
afd66255 MH |
274 | #ifdef CONFIG_OPTPROBES |
275 | /* For optimized_kprobe buffer */ | |
c802d64a HC |
276 | struct kprobe_insn_cache kprobe_optinsn_slots = { |
277 | .mutex = __MUTEX_INITIALIZER(kprobe_optinsn_slots.mutex), | |
af96397d HC |
278 | .alloc = alloc_insn_page, |
279 | .free = free_insn_page, | |
afd66255 MH |
280 | .pages = LIST_HEAD_INIT(kprobe_optinsn_slots.pages), |
281 | /* .insn_size is initialized later */ | |
282 | .nr_garbage = 0, | |
283 | }; | |
afd66255 | 284 | #endif |
2d14e39d | 285 | #endif |
9ec4b1f3 | 286 | |
e6584523 AM |
287 | /* We have preemption disabled.. so it is safe to use __ versions */ |
288 | static inline void set_kprobe_instance(struct kprobe *kp) | |
289 | { | |
b76834bc | 290 | __this_cpu_write(kprobe_instance, kp); |
e6584523 AM |
291 | } |
292 | ||
293 | static inline void reset_kprobe_instance(void) | |
294 | { | |
b76834bc | 295 | __this_cpu_write(kprobe_instance, NULL); |
e6584523 AM |
296 | } |
297 | ||
3516a460 AM |
298 | /* |
299 | * This routine is called either: | |
49a2a1b8 | 300 | * - under the kprobe_mutex - during kprobe_[un]register() |
3516a460 | 301 | * OR |
d217d545 | 302 | * - with preemption disabled - from arch/xxx/kernel/kprobes.c |
3516a460 | 303 | */ |
d0aaff97 | 304 | struct kprobe __kprobes *get_kprobe(void *addr) |
1da177e4 LT |
305 | { |
306 | struct hlist_head *head; | |
3516a460 | 307 | struct kprobe *p; |
1da177e4 LT |
308 | |
309 | head = &kprobe_table[hash_ptr(addr, KPROBE_HASH_BITS)]; | |
b67bfe0d | 310 | hlist_for_each_entry_rcu(p, head, hlist) { |
1da177e4 LT |
311 | if (p->addr == addr) |
312 | return p; | |
313 | } | |
afd66255 | 314 | |
1da177e4 LT |
315 | return NULL; |
316 | } | |
317 | ||
afd66255 MH |
318 | static int __kprobes aggr_pre_handler(struct kprobe *p, struct pt_regs *regs); |
319 | ||
320 | /* Return true if the kprobe is an aggregator */ | |
321 | static inline int kprobe_aggrprobe(struct kprobe *p) | |
322 | { | |
323 | return p->pre_handler == aggr_pre_handler; | |
324 | } | |
325 | ||
6274de49 MH |
326 | /* Return true(!0) if the kprobe is unused */ |
327 | static inline int kprobe_unused(struct kprobe *p) | |
328 | { | |
329 | return kprobe_aggrprobe(p) && kprobe_disabled(p) && | |
330 | list_empty(&p->list); | |
331 | } | |
332 | ||
afd66255 MH |
333 | /* |
334 | * Keep all fields in the kprobe consistent | |
335 | */ | |
6d8e40a8 | 336 | static inline void copy_kprobe(struct kprobe *ap, struct kprobe *p) |
afd66255 | 337 | { |
6d8e40a8 MH |
338 | memcpy(&p->opcode, &ap->opcode, sizeof(kprobe_opcode_t)); |
339 | memcpy(&p->ainsn, &ap->ainsn, sizeof(struct arch_specific_insn)); | |
afd66255 MH |
340 | } |
341 | ||
342 | #ifdef CONFIG_OPTPROBES | |
b2be84df MH |
343 | /* NOTE: change this value only with kprobe_mutex held */ |
344 | static bool kprobes_allow_optimization; | |
345 | ||
afd66255 MH |
346 | /* |
347 | * Call all pre_handler on the list, but ignores its return value. | |
348 | * This must be called from arch-dep optimized caller. | |
349 | */ | |
350 | void __kprobes opt_pre_handler(struct kprobe *p, struct pt_regs *regs) | |
351 | { | |
352 | struct kprobe *kp; | |
353 | ||
354 | list_for_each_entry_rcu(kp, &p->list, list) { | |
355 | if (kp->pre_handler && likely(!kprobe_disabled(kp))) { | |
356 | set_kprobe_instance(kp); | |
357 | kp->pre_handler(kp, regs); | |
358 | } | |
359 | reset_kprobe_instance(); | |
360 | } | |
361 | } | |
362 | ||
6274de49 | 363 | /* Free optimized instructions and optimized_kprobe */ |
55479f64 | 364 | static void free_aggr_kprobe(struct kprobe *p) |
6274de49 MH |
365 | { |
366 | struct optimized_kprobe *op; | |
367 | ||
368 | op = container_of(p, struct optimized_kprobe, kp); | |
369 | arch_remove_optimized_kprobe(op); | |
370 | arch_remove_kprobe(p); | |
371 | kfree(op); | |
372 | } | |
373 | ||
afd66255 MH |
374 | /* Return true(!0) if the kprobe is ready for optimization. */ |
375 | static inline int kprobe_optready(struct kprobe *p) | |
376 | { | |
377 | struct optimized_kprobe *op; | |
378 | ||
379 | if (kprobe_aggrprobe(p)) { | |
380 | op = container_of(p, struct optimized_kprobe, kp); | |
381 | return arch_prepared_optinsn(&op->optinsn); | |
382 | } | |
383 | ||
384 | return 0; | |
385 | } | |
386 | ||
6274de49 MH |
387 | /* Return true(!0) if the kprobe is disarmed. Note: p must be on hash list */ |
388 | static inline int kprobe_disarmed(struct kprobe *p) | |
389 | { | |
390 | struct optimized_kprobe *op; | |
391 | ||
392 | /* If kprobe is not aggr/opt probe, just return kprobe is disabled */ | |
393 | if (!kprobe_aggrprobe(p)) | |
394 | return kprobe_disabled(p); | |
395 | ||
396 | op = container_of(p, struct optimized_kprobe, kp); | |
397 | ||
398 | return kprobe_disabled(p) && list_empty(&op->list); | |
399 | } | |
400 | ||
401 | /* Return true(!0) if the probe is queued on (un)optimizing lists */ | |
55479f64 | 402 | static int kprobe_queued(struct kprobe *p) |
6274de49 MH |
403 | { |
404 | struct optimized_kprobe *op; | |
405 | ||
406 | if (kprobe_aggrprobe(p)) { | |
407 | op = container_of(p, struct optimized_kprobe, kp); | |
408 | if (!list_empty(&op->list)) | |
409 | return 1; | |
410 | } | |
411 | return 0; | |
412 | } | |
413 | ||
afd66255 MH |
414 | /* |
415 | * Return an optimized kprobe whose optimizing code replaces | |
416 | * instructions including addr (exclude breakpoint). | |
417 | */ | |
55479f64 | 418 | static struct kprobe *get_optimized_kprobe(unsigned long addr) |
afd66255 MH |
419 | { |
420 | int i; | |
421 | struct kprobe *p = NULL; | |
422 | struct optimized_kprobe *op; | |
423 | ||
424 | /* Don't check i == 0, since that is a breakpoint case. */ | |
425 | for (i = 1; !p && i < MAX_OPTIMIZED_LENGTH; i++) | |
426 | p = get_kprobe((void *)(addr - i)); | |
427 | ||
428 | if (p && kprobe_optready(p)) { | |
429 | op = container_of(p, struct optimized_kprobe, kp); | |
430 | if (arch_within_optimized_kprobe(op, addr)) | |
431 | return p; | |
432 | } | |
433 | ||
434 | return NULL; | |
435 | } | |
436 | ||
437 | /* Optimization staging list, protected by kprobe_mutex */ | |
438 | static LIST_HEAD(optimizing_list); | |
6274de49 | 439 | static LIST_HEAD(unoptimizing_list); |
7b959fc5 | 440 | static LIST_HEAD(freeing_list); |
afd66255 MH |
441 | |
442 | static void kprobe_optimizer(struct work_struct *work); | |
443 | static DECLARE_DELAYED_WORK(optimizing_work, kprobe_optimizer); | |
444 | #define OPTIMIZE_DELAY 5 | |
445 | ||
61f4e13f MH |
446 | /* |
447 | * Optimize (replace a breakpoint with a jump) kprobes listed on | |
448 | * optimizing_list. | |
449 | */ | |
55479f64 | 450 | static void do_optimize_kprobes(void) |
afd66255 | 451 | { |
6274de49 MH |
452 | /* Optimization never be done when disarmed */ |
453 | if (kprobes_all_disarmed || !kprobes_allow_optimization || | |
454 | list_empty(&optimizing_list)) | |
455 | return; | |
456 | ||
afd66255 MH |
457 | /* |
458 | * The optimization/unoptimization refers online_cpus via | |
459 | * stop_machine() and cpu-hotplug modifies online_cpus. | |
460 | * And same time, text_mutex will be held in cpu-hotplug and here. | |
461 | * This combination can cause a deadlock (cpu-hotplug try to lock | |
462 | * text_mutex but stop_machine can not be done because online_cpus | |
463 | * has been changed) | |
464 | * To avoid this deadlock, we need to call get_online_cpus() | |
465 | * for preventing cpu-hotplug outside of text_mutex locking. | |
466 | */ | |
467 | get_online_cpus(); | |
468 | mutex_lock(&text_mutex); | |
cd7ebe22 | 469 | arch_optimize_kprobes(&optimizing_list); |
afd66255 MH |
470 | mutex_unlock(&text_mutex); |
471 | put_online_cpus(); | |
61f4e13f MH |
472 | } |
473 | ||
6274de49 MH |
474 | /* |
475 | * Unoptimize (replace a jump with a breakpoint and remove the breakpoint | |
476 | * if need) kprobes listed on unoptimizing_list. | |
477 | */ | |
55479f64 | 478 | static void do_unoptimize_kprobes(void) |
6274de49 MH |
479 | { |
480 | struct optimized_kprobe *op, *tmp; | |
481 | ||
482 | /* Unoptimization must be done anytime */ | |
483 | if (list_empty(&unoptimizing_list)) | |
484 | return; | |
485 | ||
486 | /* Ditto to do_optimize_kprobes */ | |
487 | get_online_cpus(); | |
488 | mutex_lock(&text_mutex); | |
7b959fc5 | 489 | arch_unoptimize_kprobes(&unoptimizing_list, &freeing_list); |
f984ba4e | 490 | /* Loop free_list for disarming */ |
7b959fc5 | 491 | list_for_each_entry_safe(op, tmp, &freeing_list, list) { |
6274de49 MH |
492 | /* Disarm probes if marked disabled */ |
493 | if (kprobe_disabled(&op->kp)) | |
494 | arch_disarm_kprobe(&op->kp); | |
495 | if (kprobe_unused(&op->kp)) { | |
496 | /* | |
497 | * Remove unused probes from hash list. After waiting | |
498 | * for synchronization, these probes are reclaimed. | |
499 | * (reclaiming is done by do_free_cleaned_kprobes.) | |
500 | */ | |
501 | hlist_del_rcu(&op->kp.hlist); | |
6274de49 MH |
502 | } else |
503 | list_del_init(&op->list); | |
504 | } | |
505 | mutex_unlock(&text_mutex); | |
506 | put_online_cpus(); | |
507 | } | |
508 | ||
509 | /* Reclaim all kprobes on the free_list */ | |
55479f64 | 510 | static void do_free_cleaned_kprobes(void) |
6274de49 MH |
511 | { |
512 | struct optimized_kprobe *op, *tmp; | |
513 | ||
7b959fc5 | 514 | list_for_each_entry_safe(op, tmp, &freeing_list, list) { |
6274de49 MH |
515 | BUG_ON(!kprobe_unused(&op->kp)); |
516 | list_del_init(&op->list); | |
517 | free_aggr_kprobe(&op->kp); | |
518 | } | |
519 | } | |
520 | ||
521 | /* Start optimizer after OPTIMIZE_DELAY passed */ | |
55479f64 | 522 | static void kick_kprobe_optimizer(void) |
6274de49 | 523 | { |
ad72b3be | 524 | schedule_delayed_work(&optimizing_work, OPTIMIZE_DELAY); |
6274de49 MH |
525 | } |
526 | ||
61f4e13f | 527 | /* Kprobe jump optimizer */ |
55479f64 | 528 | static void kprobe_optimizer(struct work_struct *work) |
61f4e13f | 529 | { |
72ef3794 | 530 | mutex_lock(&kprobe_mutex); |
61f4e13f MH |
531 | /* Lock modules while optimizing kprobes */ |
532 | mutex_lock(&module_mutex); | |
61f4e13f MH |
533 | |
534 | /* | |
6274de49 MH |
535 | * Step 1: Unoptimize kprobes and collect cleaned (unused and disarmed) |
536 | * kprobes before waiting for quiesence period. | |
537 | */ | |
7b959fc5 | 538 | do_unoptimize_kprobes(); |
6274de49 MH |
539 | |
540 | /* | |
541 | * Step 2: Wait for quiesence period to ensure all running interrupts | |
61f4e13f MH |
542 | * are done. Because optprobe may modify multiple instructions |
543 | * there is a chance that Nth instruction is interrupted. In that | |
544 | * case, running interrupt can return to 2nd-Nth byte of jump | |
545 | * instruction. This wait is for avoiding it. | |
546 | */ | |
547 | synchronize_sched(); | |
548 | ||
6274de49 | 549 | /* Step 3: Optimize kprobes after quiesence period */ |
61f4e13f | 550 | do_optimize_kprobes(); |
6274de49 MH |
551 | |
552 | /* Step 4: Free cleaned kprobes after quiesence period */ | |
7b959fc5 | 553 | do_free_cleaned_kprobes(); |
6274de49 | 554 | |
afd66255 | 555 | mutex_unlock(&module_mutex); |
72ef3794 | 556 | mutex_unlock(&kprobe_mutex); |
6274de49 | 557 | |
cd7ebe22 | 558 | /* Step 5: Kick optimizer again if needed */ |
f984ba4e | 559 | if (!list_empty(&optimizing_list) || !list_empty(&unoptimizing_list)) |
cd7ebe22 | 560 | kick_kprobe_optimizer(); |
6274de49 MH |
561 | } |
562 | ||
563 | /* Wait for completing optimization and unoptimization */ | |
55479f64 | 564 | static void wait_for_kprobe_optimizer(void) |
6274de49 | 565 | { |
ad72b3be TH |
566 | mutex_lock(&kprobe_mutex); |
567 | ||
568 | while (!list_empty(&optimizing_list) || !list_empty(&unoptimizing_list)) { | |
569 | mutex_unlock(&kprobe_mutex); | |
570 | ||
571 | /* this will also make optimizing_work execute immmediately */ | |
572 | flush_delayed_work(&optimizing_work); | |
573 | /* @optimizing_work might not have been queued yet, relax */ | |
574 | cpu_relax(); | |
575 | ||
576 | mutex_lock(&kprobe_mutex); | |
577 | } | |
578 | ||
579 | mutex_unlock(&kprobe_mutex); | |
afd66255 MH |
580 | } |
581 | ||
582 | /* Optimize kprobe if p is ready to be optimized */ | |
55479f64 | 583 | static void optimize_kprobe(struct kprobe *p) |
afd66255 MH |
584 | { |
585 | struct optimized_kprobe *op; | |
586 | ||
587 | /* Check if the kprobe is disabled or not ready for optimization. */ | |
b2be84df | 588 | if (!kprobe_optready(p) || !kprobes_allow_optimization || |
afd66255 MH |
589 | (kprobe_disabled(p) || kprobes_all_disarmed)) |
590 | return; | |
591 | ||
592 | /* Both of break_handler and post_handler are not supported. */ | |
593 | if (p->break_handler || p->post_handler) | |
594 | return; | |
595 | ||
596 | op = container_of(p, struct optimized_kprobe, kp); | |
597 | ||
598 | /* Check there is no other kprobes at the optimized instructions */ | |
599 | if (arch_check_optimized_kprobe(op) < 0) | |
600 | return; | |
601 | ||
602 | /* Check if it is already optimized. */ | |
603 | if (op->kp.flags & KPROBE_FLAG_OPTIMIZED) | |
604 | return; | |
afd66255 | 605 | op->kp.flags |= KPROBE_FLAG_OPTIMIZED; |
6274de49 MH |
606 | |
607 | if (!list_empty(&op->list)) | |
608 | /* This is under unoptimizing. Just dequeue the probe */ | |
609 | list_del_init(&op->list); | |
610 | else { | |
611 | list_add(&op->list, &optimizing_list); | |
612 | kick_kprobe_optimizer(); | |
613 | } | |
614 | } | |
615 | ||
616 | /* Short cut to direct unoptimizing */ | |
55479f64 | 617 | static void force_unoptimize_kprobe(struct optimized_kprobe *op) |
6274de49 MH |
618 | { |
619 | get_online_cpus(); | |
620 | arch_unoptimize_kprobe(op); | |
621 | put_online_cpus(); | |
622 | if (kprobe_disabled(&op->kp)) | |
623 | arch_disarm_kprobe(&op->kp); | |
afd66255 MH |
624 | } |
625 | ||
626 | /* Unoptimize a kprobe if p is optimized */ | |
55479f64 | 627 | static void unoptimize_kprobe(struct kprobe *p, bool force) |
afd66255 MH |
628 | { |
629 | struct optimized_kprobe *op; | |
630 | ||
6274de49 MH |
631 | if (!kprobe_aggrprobe(p) || kprobe_disarmed(p)) |
632 | return; /* This is not an optprobe nor optimized */ | |
633 | ||
634 | op = container_of(p, struct optimized_kprobe, kp); | |
635 | if (!kprobe_optimized(p)) { | |
636 | /* Unoptimized or unoptimizing case */ | |
637 | if (force && !list_empty(&op->list)) { | |
638 | /* | |
639 | * Only if this is unoptimizing kprobe and forced, | |
640 | * forcibly unoptimize it. (No need to unoptimize | |
641 | * unoptimized kprobe again :) | |
642 | */ | |
afd66255 | 643 | list_del_init(&op->list); |
6274de49 MH |
644 | force_unoptimize_kprobe(op); |
645 | } | |
646 | return; | |
647 | } | |
648 | ||
649 | op->kp.flags &= ~KPROBE_FLAG_OPTIMIZED; | |
650 | if (!list_empty(&op->list)) { | |
651 | /* Dequeue from the optimization queue */ | |
652 | list_del_init(&op->list); | |
653 | return; | |
654 | } | |
655 | /* Optimized kprobe case */ | |
656 | if (force) | |
657 | /* Forcibly update the code: this is a special case */ | |
658 | force_unoptimize_kprobe(op); | |
659 | else { | |
660 | list_add(&op->list, &unoptimizing_list); | |
661 | kick_kprobe_optimizer(); | |
afd66255 MH |
662 | } |
663 | } | |
664 | ||
0490cd1f MH |
665 | /* Cancel unoptimizing for reusing */ |
666 | static void reuse_unused_kprobe(struct kprobe *ap) | |
667 | { | |
668 | struct optimized_kprobe *op; | |
669 | ||
670 | BUG_ON(!kprobe_unused(ap)); | |
671 | /* | |
672 | * Unused kprobe MUST be on the way of delayed unoptimizing (means | |
673 | * there is still a relative jump) and disabled. | |
674 | */ | |
675 | op = container_of(ap, struct optimized_kprobe, kp); | |
676 | if (unlikely(list_empty(&op->list))) | |
677 | printk(KERN_WARNING "Warning: found a stray unused " | |
678 | "aggrprobe@%p\n", ap->addr); | |
679 | /* Enable the probe again */ | |
680 | ap->flags &= ~KPROBE_FLAG_DISABLED; | |
681 | /* Optimize it again (remove from op->list) */ | |
682 | BUG_ON(!kprobe_optready(ap)); | |
683 | optimize_kprobe(ap); | |
684 | } | |
685 | ||
afd66255 | 686 | /* Remove optimized instructions */ |
55479f64 | 687 | static void kill_optimized_kprobe(struct kprobe *p) |
afd66255 MH |
688 | { |
689 | struct optimized_kprobe *op; | |
690 | ||
691 | op = container_of(p, struct optimized_kprobe, kp); | |
6274de49 MH |
692 | if (!list_empty(&op->list)) |
693 | /* Dequeue from the (un)optimization queue */ | |
afd66255 | 694 | list_del_init(&op->list); |
6274de49 | 695 | op->kp.flags &= ~KPROBE_FLAG_OPTIMIZED; |
7b959fc5 MH |
696 | |
697 | if (kprobe_unused(p)) { | |
698 | /* Enqueue if it is unused */ | |
699 | list_add(&op->list, &freeing_list); | |
700 | /* | |
701 | * Remove unused probes from the hash list. After waiting | |
702 | * for synchronization, this probe is reclaimed. | |
703 | * (reclaiming is done by do_free_cleaned_kprobes().) | |
704 | */ | |
705 | hlist_del_rcu(&op->kp.hlist); | |
706 | } | |
707 | ||
6274de49 | 708 | /* Don't touch the code, because it is already freed. */ |
afd66255 MH |
709 | arch_remove_optimized_kprobe(op); |
710 | } | |
711 | ||
712 | /* Try to prepare optimized instructions */ | |
55479f64 | 713 | static void prepare_optimized_kprobe(struct kprobe *p) |
afd66255 MH |
714 | { |
715 | struct optimized_kprobe *op; | |
716 | ||
717 | op = container_of(p, struct optimized_kprobe, kp); | |
718 | arch_prepare_optimized_kprobe(op); | |
719 | } | |
720 | ||
afd66255 | 721 | /* Allocate new optimized_kprobe and try to prepare optimized instructions */ |
55479f64 | 722 | static struct kprobe *alloc_aggr_kprobe(struct kprobe *p) |
afd66255 MH |
723 | { |
724 | struct optimized_kprobe *op; | |
725 | ||
726 | op = kzalloc(sizeof(struct optimized_kprobe), GFP_KERNEL); | |
727 | if (!op) | |
728 | return NULL; | |
729 | ||
730 | INIT_LIST_HEAD(&op->list); | |
731 | op->kp.addr = p->addr; | |
732 | arch_prepare_optimized_kprobe(op); | |
733 | ||
734 | return &op->kp; | |
735 | } | |
736 | ||
55479f64 | 737 | static void init_aggr_kprobe(struct kprobe *ap, struct kprobe *p); |
afd66255 MH |
738 | |
739 | /* | |
740 | * Prepare an optimized_kprobe and optimize it | |
741 | * NOTE: p must be a normal registered kprobe | |
742 | */ | |
55479f64 | 743 | static void try_to_optimize_kprobe(struct kprobe *p) |
afd66255 MH |
744 | { |
745 | struct kprobe *ap; | |
746 | struct optimized_kprobe *op; | |
747 | ||
ae6aa16f MH |
748 | /* Impossible to optimize ftrace-based kprobe */ |
749 | if (kprobe_ftrace(p)) | |
750 | return; | |
751 | ||
25764288 MH |
752 | /* For preparing optimization, jump_label_text_reserved() is called */ |
753 | jump_label_lock(); | |
754 | mutex_lock(&text_mutex); | |
755 | ||
afd66255 MH |
756 | ap = alloc_aggr_kprobe(p); |
757 | if (!ap) | |
25764288 | 758 | goto out; |
afd66255 MH |
759 | |
760 | op = container_of(ap, struct optimized_kprobe, kp); | |
761 | if (!arch_prepared_optinsn(&op->optinsn)) { | |
762 | /* If failed to setup optimizing, fallback to kprobe */ | |
6274de49 MH |
763 | arch_remove_optimized_kprobe(op); |
764 | kfree(op); | |
25764288 | 765 | goto out; |
afd66255 MH |
766 | } |
767 | ||
768 | init_aggr_kprobe(ap, p); | |
25764288 MH |
769 | optimize_kprobe(ap); /* This just kicks optimizer thread */ |
770 | ||
771 | out: | |
772 | mutex_unlock(&text_mutex); | |
773 | jump_label_unlock(); | |
afd66255 MH |
774 | } |
775 | ||
b2be84df | 776 | #ifdef CONFIG_SYSCTL |
55479f64 | 777 | static void optimize_all_kprobes(void) |
b2be84df MH |
778 | { |
779 | struct hlist_head *head; | |
b2be84df MH |
780 | struct kprobe *p; |
781 | unsigned int i; | |
782 | ||
5c51543b | 783 | mutex_lock(&kprobe_mutex); |
b2be84df MH |
784 | /* If optimization is already allowed, just return */ |
785 | if (kprobes_allow_optimization) | |
5c51543b | 786 | goto out; |
b2be84df MH |
787 | |
788 | kprobes_allow_optimization = true; | |
b2be84df MH |
789 | for (i = 0; i < KPROBE_TABLE_SIZE; i++) { |
790 | head = &kprobe_table[i]; | |
b67bfe0d | 791 | hlist_for_each_entry_rcu(p, head, hlist) |
b2be84df MH |
792 | if (!kprobe_disabled(p)) |
793 | optimize_kprobe(p); | |
794 | } | |
b2be84df | 795 | printk(KERN_INFO "Kprobes globally optimized\n"); |
5c51543b MH |
796 | out: |
797 | mutex_unlock(&kprobe_mutex); | |
b2be84df MH |
798 | } |
799 | ||
55479f64 | 800 | static void unoptimize_all_kprobes(void) |
b2be84df MH |
801 | { |
802 | struct hlist_head *head; | |
b2be84df MH |
803 | struct kprobe *p; |
804 | unsigned int i; | |
805 | ||
5c51543b | 806 | mutex_lock(&kprobe_mutex); |
b2be84df | 807 | /* If optimization is already prohibited, just return */ |
5c51543b MH |
808 | if (!kprobes_allow_optimization) { |
809 | mutex_unlock(&kprobe_mutex); | |
b2be84df | 810 | return; |
5c51543b | 811 | } |
b2be84df MH |
812 | |
813 | kprobes_allow_optimization = false; | |
b2be84df MH |
814 | for (i = 0; i < KPROBE_TABLE_SIZE; i++) { |
815 | head = &kprobe_table[i]; | |
b67bfe0d | 816 | hlist_for_each_entry_rcu(p, head, hlist) { |
b2be84df | 817 | if (!kprobe_disabled(p)) |
6274de49 | 818 | unoptimize_kprobe(p, false); |
b2be84df MH |
819 | } |
820 | } | |
5c51543b MH |
821 | mutex_unlock(&kprobe_mutex); |
822 | ||
6274de49 MH |
823 | /* Wait for unoptimizing completion */ |
824 | wait_for_kprobe_optimizer(); | |
825 | printk(KERN_INFO "Kprobes globally unoptimized\n"); | |
b2be84df MH |
826 | } |
827 | ||
5c51543b | 828 | static DEFINE_MUTEX(kprobe_sysctl_mutex); |
b2be84df MH |
829 | int sysctl_kprobes_optimization; |
830 | int proc_kprobes_optimization_handler(struct ctl_table *table, int write, | |
831 | void __user *buffer, size_t *length, | |
832 | loff_t *ppos) | |
833 | { | |
834 | int ret; | |
835 | ||
5c51543b | 836 | mutex_lock(&kprobe_sysctl_mutex); |
b2be84df MH |
837 | sysctl_kprobes_optimization = kprobes_allow_optimization ? 1 : 0; |
838 | ret = proc_dointvec_minmax(table, write, buffer, length, ppos); | |
839 | ||
840 | if (sysctl_kprobes_optimization) | |
841 | optimize_all_kprobes(); | |
842 | else | |
843 | unoptimize_all_kprobes(); | |
5c51543b | 844 | mutex_unlock(&kprobe_sysctl_mutex); |
b2be84df MH |
845 | |
846 | return ret; | |
847 | } | |
848 | #endif /* CONFIG_SYSCTL */ | |
849 | ||
6274de49 | 850 | /* Put a breakpoint for a probe. Must be called with text_mutex locked */ |
55479f64 | 851 | static void __arm_kprobe(struct kprobe *p) |
afd66255 | 852 | { |
6d8e40a8 | 853 | struct kprobe *_p; |
afd66255 MH |
854 | |
855 | /* Check collision with other optimized kprobes */ | |
6d8e40a8 MH |
856 | _p = get_optimized_kprobe((unsigned long)p->addr); |
857 | if (unlikely(_p)) | |
6274de49 MH |
858 | /* Fallback to unoptimized kprobe */ |
859 | unoptimize_kprobe(_p, true); | |
afd66255 MH |
860 | |
861 | arch_arm_kprobe(p); | |
862 | optimize_kprobe(p); /* Try to optimize (add kprobe to a list) */ | |
863 | } | |
864 | ||
6274de49 | 865 | /* Remove the breakpoint of a probe. Must be called with text_mutex locked */ |
55479f64 | 866 | static void __disarm_kprobe(struct kprobe *p, bool reopt) |
afd66255 | 867 | { |
6d8e40a8 | 868 | struct kprobe *_p; |
afd66255 | 869 | |
6274de49 | 870 | unoptimize_kprobe(p, false); /* Try to unoptimize */ |
afd66255 | 871 | |
6274de49 MH |
872 | if (!kprobe_queued(p)) { |
873 | arch_disarm_kprobe(p); | |
874 | /* If another kprobe was blocked, optimize it. */ | |
875 | _p = get_optimized_kprobe((unsigned long)p->addr); | |
876 | if (unlikely(_p) && reopt) | |
877 | optimize_kprobe(_p); | |
878 | } | |
879 | /* TODO: reoptimize others after unoptimized this probe */ | |
afd66255 MH |
880 | } |
881 | ||
882 | #else /* !CONFIG_OPTPROBES */ | |
883 | ||
884 | #define optimize_kprobe(p) do {} while (0) | |
6274de49 | 885 | #define unoptimize_kprobe(p, f) do {} while (0) |
afd66255 MH |
886 | #define kill_optimized_kprobe(p) do {} while (0) |
887 | #define prepare_optimized_kprobe(p) do {} while (0) | |
888 | #define try_to_optimize_kprobe(p) do {} while (0) | |
889 | #define __arm_kprobe(p) arch_arm_kprobe(p) | |
6274de49 MH |
890 | #define __disarm_kprobe(p, o) arch_disarm_kprobe(p) |
891 | #define kprobe_disarmed(p) kprobe_disabled(p) | |
892 | #define wait_for_kprobe_optimizer() do {} while (0) | |
afd66255 | 893 | |
0490cd1f MH |
894 | /* There should be no unused kprobes can be reused without optimization */ |
895 | static void reuse_unused_kprobe(struct kprobe *ap) | |
896 | { | |
897 | printk(KERN_ERR "Error: There should be no unused kprobe here.\n"); | |
898 | BUG_ON(kprobe_unused(ap)); | |
899 | } | |
900 | ||
55479f64 | 901 | static void free_aggr_kprobe(struct kprobe *p) |
afd66255 | 902 | { |
6274de49 | 903 | arch_remove_kprobe(p); |
afd66255 MH |
904 | kfree(p); |
905 | } | |
906 | ||
55479f64 | 907 | static struct kprobe *alloc_aggr_kprobe(struct kprobe *p) |
afd66255 MH |
908 | { |
909 | return kzalloc(sizeof(struct kprobe), GFP_KERNEL); | |
910 | } | |
911 | #endif /* CONFIG_OPTPROBES */ | |
912 | ||
e7dbfe34 | 913 | #ifdef CONFIG_KPROBES_ON_FTRACE |
ae6aa16f | 914 | static struct ftrace_ops kprobe_ftrace_ops __read_mostly = { |
e5253896 | 915 | .func = kprobe_ftrace_handler, |
ae6aa16f MH |
916 | .flags = FTRACE_OPS_FL_SAVE_REGS, |
917 | }; | |
918 | static int kprobe_ftrace_enabled; | |
919 | ||
920 | /* Must ensure p->addr is really on ftrace */ | |
55479f64 | 921 | static int prepare_kprobe(struct kprobe *p) |
ae6aa16f MH |
922 | { |
923 | if (!kprobe_ftrace(p)) | |
924 | return arch_prepare_kprobe(p); | |
925 | ||
926 | return arch_prepare_kprobe_ftrace(p); | |
927 | } | |
928 | ||
929 | /* Caller must lock kprobe_mutex */ | |
55479f64 | 930 | static void arm_kprobe_ftrace(struct kprobe *p) |
ae6aa16f MH |
931 | { |
932 | int ret; | |
933 | ||
934 | ret = ftrace_set_filter_ip(&kprobe_ftrace_ops, | |
935 | (unsigned long)p->addr, 0, 0); | |
936 | WARN(ret < 0, "Failed to arm kprobe-ftrace at %p (%d)\n", p->addr, ret); | |
937 | kprobe_ftrace_enabled++; | |
938 | if (kprobe_ftrace_enabled == 1) { | |
939 | ret = register_ftrace_function(&kprobe_ftrace_ops); | |
940 | WARN(ret < 0, "Failed to init kprobe-ftrace (%d)\n", ret); | |
941 | } | |
942 | } | |
943 | ||
944 | /* Caller must lock kprobe_mutex */ | |
55479f64 | 945 | static void disarm_kprobe_ftrace(struct kprobe *p) |
ae6aa16f MH |
946 | { |
947 | int ret; | |
948 | ||
949 | kprobe_ftrace_enabled--; | |
950 | if (kprobe_ftrace_enabled == 0) { | |
951 | ret = unregister_ftrace_function(&kprobe_ftrace_ops); | |
952 | WARN(ret < 0, "Failed to init kprobe-ftrace (%d)\n", ret); | |
953 | } | |
954 | ret = ftrace_set_filter_ip(&kprobe_ftrace_ops, | |
955 | (unsigned long)p->addr, 1, 0); | |
956 | WARN(ret < 0, "Failed to disarm kprobe-ftrace at %p (%d)\n", p->addr, ret); | |
957 | } | |
e7dbfe34 | 958 | #else /* !CONFIG_KPROBES_ON_FTRACE */ |
ae6aa16f MH |
959 | #define prepare_kprobe(p) arch_prepare_kprobe(p) |
960 | #define arm_kprobe_ftrace(p) do {} while (0) | |
961 | #define disarm_kprobe_ftrace(p) do {} while (0) | |
962 | #endif | |
963 | ||
201517a7 | 964 | /* Arm a kprobe with text_mutex */ |
55479f64 | 965 | static void arm_kprobe(struct kprobe *kp) |
201517a7 | 966 | { |
ae6aa16f MH |
967 | if (unlikely(kprobe_ftrace(kp))) { |
968 | arm_kprobe_ftrace(kp); | |
969 | return; | |
970 | } | |
afd66255 MH |
971 | /* |
972 | * Here, since __arm_kprobe() doesn't use stop_machine(), | |
973 | * this doesn't cause deadlock on text_mutex. So, we don't | |
974 | * need get_online_cpus(). | |
975 | */ | |
201517a7 | 976 | mutex_lock(&text_mutex); |
afd66255 | 977 | __arm_kprobe(kp); |
201517a7 MH |
978 | mutex_unlock(&text_mutex); |
979 | } | |
980 | ||
981 | /* Disarm a kprobe with text_mutex */ | |
55479f64 | 982 | static void disarm_kprobe(struct kprobe *kp, bool reopt) |
201517a7 | 983 | { |
ae6aa16f MH |
984 | if (unlikely(kprobe_ftrace(kp))) { |
985 | disarm_kprobe_ftrace(kp); | |
986 | return; | |
987 | } | |
6274de49 | 988 | /* Ditto */ |
201517a7 | 989 | mutex_lock(&text_mutex); |
ae6aa16f | 990 | __disarm_kprobe(kp, reopt); |
201517a7 MH |
991 | mutex_unlock(&text_mutex); |
992 | } | |
993 | ||
64f562c6 AM |
994 | /* |
995 | * Aggregate handlers for multiple kprobes support - these handlers | |
996 | * take care of invoking the individual kprobe handlers on p->list | |
997 | */ | |
d0aaff97 | 998 | static int __kprobes aggr_pre_handler(struct kprobe *p, struct pt_regs *regs) |
64f562c6 AM |
999 | { |
1000 | struct kprobe *kp; | |
1001 | ||
3516a460 | 1002 | list_for_each_entry_rcu(kp, &p->list, list) { |
de5bd88d | 1003 | if (kp->pre_handler && likely(!kprobe_disabled(kp))) { |
e6584523 | 1004 | set_kprobe_instance(kp); |
8b0914ea PP |
1005 | if (kp->pre_handler(kp, regs)) |
1006 | return 1; | |
64f562c6 | 1007 | } |
e6584523 | 1008 | reset_kprobe_instance(); |
64f562c6 AM |
1009 | } |
1010 | return 0; | |
1011 | } | |
1012 | ||
d0aaff97 PP |
1013 | static void __kprobes aggr_post_handler(struct kprobe *p, struct pt_regs *regs, |
1014 | unsigned long flags) | |
64f562c6 AM |
1015 | { |
1016 | struct kprobe *kp; | |
1017 | ||
3516a460 | 1018 | list_for_each_entry_rcu(kp, &p->list, list) { |
de5bd88d | 1019 | if (kp->post_handler && likely(!kprobe_disabled(kp))) { |
e6584523 | 1020 | set_kprobe_instance(kp); |
64f562c6 | 1021 | kp->post_handler(kp, regs, flags); |
e6584523 | 1022 | reset_kprobe_instance(); |
64f562c6 AM |
1023 | } |
1024 | } | |
64f562c6 AM |
1025 | } |
1026 | ||
d0aaff97 PP |
1027 | static int __kprobes aggr_fault_handler(struct kprobe *p, struct pt_regs *regs, |
1028 | int trapnr) | |
64f562c6 | 1029 | { |
b76834bc | 1030 | struct kprobe *cur = __this_cpu_read(kprobe_instance); |
e6584523 | 1031 | |
64f562c6 AM |
1032 | /* |
1033 | * if we faulted "during" the execution of a user specified | |
1034 | * probe handler, invoke just that probe's fault handler | |
1035 | */ | |
e6584523 AM |
1036 | if (cur && cur->fault_handler) { |
1037 | if (cur->fault_handler(cur, regs, trapnr)) | |
64f562c6 AM |
1038 | return 1; |
1039 | } | |
1040 | return 0; | |
1041 | } | |
1042 | ||
d0aaff97 | 1043 | static int __kprobes aggr_break_handler(struct kprobe *p, struct pt_regs *regs) |
8b0914ea | 1044 | { |
b76834bc | 1045 | struct kprobe *cur = __this_cpu_read(kprobe_instance); |
e6584523 AM |
1046 | int ret = 0; |
1047 | ||
1048 | if (cur && cur->break_handler) { | |
1049 | if (cur->break_handler(cur, regs)) | |
1050 | ret = 1; | |
8b0914ea | 1051 | } |
e6584523 AM |
1052 | reset_kprobe_instance(); |
1053 | return ret; | |
8b0914ea PP |
1054 | } |
1055 | ||
bf8d5c52 KA |
1056 | /* Walks the list and increments nmissed count for multiprobe case */ |
1057 | void __kprobes kprobes_inc_nmissed_count(struct kprobe *p) | |
1058 | { | |
1059 | struct kprobe *kp; | |
afd66255 | 1060 | if (!kprobe_aggrprobe(p)) { |
bf8d5c52 KA |
1061 | p->nmissed++; |
1062 | } else { | |
1063 | list_for_each_entry_rcu(kp, &p->list, list) | |
1064 | kp->nmissed++; | |
1065 | } | |
1066 | return; | |
1067 | } | |
1068 | ||
99219a3f | 1069 | void __kprobes recycle_rp_inst(struct kretprobe_instance *ri, |
1070 | struct hlist_head *head) | |
b94cce92 | 1071 | { |
ef53d9c5 S |
1072 | struct kretprobe *rp = ri->rp; |
1073 | ||
b94cce92 HN |
1074 | /* remove rp inst off the rprobe_inst_table */ |
1075 | hlist_del(&ri->hlist); | |
ef53d9c5 S |
1076 | INIT_HLIST_NODE(&ri->hlist); |
1077 | if (likely(rp)) { | |
ec484608 | 1078 | raw_spin_lock(&rp->lock); |
ef53d9c5 | 1079 | hlist_add_head(&ri->hlist, &rp->free_instances); |
ec484608 | 1080 | raw_spin_unlock(&rp->lock); |
b94cce92 HN |
1081 | } else |
1082 | /* Unregistering */ | |
99219a3f | 1083 | hlist_add_head(&ri->hlist, head); |
b94cce92 HN |
1084 | } |
1085 | ||
017c39bd | 1086 | void __kprobes kretprobe_hash_lock(struct task_struct *tsk, |
ef53d9c5 | 1087 | struct hlist_head **head, unsigned long *flags) |
635c17c2 | 1088 | __acquires(hlist_lock) |
ef53d9c5 S |
1089 | { |
1090 | unsigned long hash = hash_ptr(tsk, KPROBE_HASH_BITS); | |
ec484608 | 1091 | raw_spinlock_t *hlist_lock; |
ef53d9c5 S |
1092 | |
1093 | *head = &kretprobe_inst_table[hash]; | |
1094 | hlist_lock = kretprobe_table_lock_ptr(hash); | |
ec484608 | 1095 | raw_spin_lock_irqsave(hlist_lock, *flags); |
ef53d9c5 S |
1096 | } |
1097 | ||
017c39bd MH |
1098 | static void __kprobes kretprobe_table_lock(unsigned long hash, |
1099 | unsigned long *flags) | |
635c17c2 | 1100 | __acquires(hlist_lock) |
b94cce92 | 1101 | { |
ec484608 TG |
1102 | raw_spinlock_t *hlist_lock = kretprobe_table_lock_ptr(hash); |
1103 | raw_spin_lock_irqsave(hlist_lock, *flags); | |
ef53d9c5 S |
1104 | } |
1105 | ||
017c39bd MH |
1106 | void __kprobes kretprobe_hash_unlock(struct task_struct *tsk, |
1107 | unsigned long *flags) | |
635c17c2 | 1108 | __releases(hlist_lock) |
ef53d9c5 S |
1109 | { |
1110 | unsigned long hash = hash_ptr(tsk, KPROBE_HASH_BITS); | |
ec484608 | 1111 | raw_spinlock_t *hlist_lock; |
ef53d9c5 S |
1112 | |
1113 | hlist_lock = kretprobe_table_lock_ptr(hash); | |
ec484608 | 1114 | raw_spin_unlock_irqrestore(hlist_lock, *flags); |
ef53d9c5 S |
1115 | } |
1116 | ||
6376b229 NK |
1117 | static void __kprobes kretprobe_table_unlock(unsigned long hash, |
1118 | unsigned long *flags) | |
635c17c2 | 1119 | __releases(hlist_lock) |
ef53d9c5 | 1120 | { |
ec484608 TG |
1121 | raw_spinlock_t *hlist_lock = kretprobe_table_lock_ptr(hash); |
1122 | raw_spin_unlock_irqrestore(hlist_lock, *flags); | |
b94cce92 HN |
1123 | } |
1124 | ||
b94cce92 | 1125 | /* |
c6fd91f0 | 1126 | * This function is called from finish_task_switch when task tk becomes dead, |
1127 | * so that we can recycle any function-return probe instances associated | |
1128 | * with this task. These left over instances represent probed functions | |
1129 | * that have been called but will never return. | |
b94cce92 | 1130 | */ |
d0aaff97 | 1131 | void __kprobes kprobe_flush_task(struct task_struct *tk) |
b94cce92 | 1132 | { |
62c27be0 | 1133 | struct kretprobe_instance *ri; |
99219a3f | 1134 | struct hlist_head *head, empty_rp; |
b67bfe0d | 1135 | struct hlist_node *tmp; |
ef53d9c5 | 1136 | unsigned long hash, flags = 0; |
802eae7c | 1137 | |
ef53d9c5 S |
1138 | if (unlikely(!kprobes_initialized)) |
1139 | /* Early boot. kretprobe_table_locks not yet initialized. */ | |
1140 | return; | |
1141 | ||
d496aab5 | 1142 | INIT_HLIST_HEAD(&empty_rp); |
ef53d9c5 S |
1143 | hash = hash_ptr(tk, KPROBE_HASH_BITS); |
1144 | head = &kretprobe_inst_table[hash]; | |
1145 | kretprobe_table_lock(hash, &flags); | |
b67bfe0d | 1146 | hlist_for_each_entry_safe(ri, tmp, head, hlist) { |
62c27be0 | 1147 | if (ri->task == tk) |
99219a3f | 1148 | recycle_rp_inst(ri, &empty_rp); |
62c27be0 | 1149 | } |
ef53d9c5 | 1150 | kretprobe_table_unlock(hash, &flags); |
b67bfe0d | 1151 | hlist_for_each_entry_safe(ri, tmp, &empty_rp, hlist) { |
99219a3f | 1152 | hlist_del(&ri->hlist); |
1153 | kfree(ri); | |
1154 | } | |
b94cce92 HN |
1155 | } |
1156 | ||
b94cce92 HN |
1157 | static inline void free_rp_inst(struct kretprobe *rp) |
1158 | { | |
1159 | struct kretprobe_instance *ri; | |
b67bfe0d | 1160 | struct hlist_node *next; |
4c4308cb | 1161 | |
b67bfe0d | 1162 | hlist_for_each_entry_safe(ri, next, &rp->free_instances, hlist) { |
ef53d9c5 | 1163 | hlist_del(&ri->hlist); |
b94cce92 HN |
1164 | kfree(ri); |
1165 | } | |
1166 | } | |
1167 | ||
4a296e07 MH |
1168 | static void __kprobes cleanup_rp_inst(struct kretprobe *rp) |
1169 | { | |
ef53d9c5 | 1170 | unsigned long flags, hash; |
4a296e07 | 1171 | struct kretprobe_instance *ri; |
b67bfe0d | 1172 | struct hlist_node *next; |
ef53d9c5 S |
1173 | struct hlist_head *head; |
1174 | ||
4a296e07 | 1175 | /* No race here */ |
ef53d9c5 S |
1176 | for (hash = 0; hash < KPROBE_TABLE_SIZE; hash++) { |
1177 | kretprobe_table_lock(hash, &flags); | |
1178 | head = &kretprobe_inst_table[hash]; | |
b67bfe0d | 1179 | hlist_for_each_entry_safe(ri, next, head, hlist) { |
ef53d9c5 S |
1180 | if (ri->rp == rp) |
1181 | ri->rp = NULL; | |
1182 | } | |
1183 | kretprobe_table_unlock(hash, &flags); | |
4a296e07 | 1184 | } |
4a296e07 MH |
1185 | free_rp_inst(rp); |
1186 | } | |
1187 | ||
8b0914ea | 1188 | /* |
b918e5e6 | 1189 | * Add the new probe to ap->list. Fail if this is the |
8b0914ea PP |
1190 | * second jprobe at the address - two jprobes can't coexist |
1191 | */ | |
55479f64 | 1192 | static int add_new_kprobe(struct kprobe *ap, struct kprobe *p) |
8b0914ea | 1193 | { |
de5bd88d | 1194 | BUG_ON(kprobe_gone(ap) || kprobe_gone(p)); |
afd66255 MH |
1195 | |
1196 | if (p->break_handler || p->post_handler) | |
6274de49 | 1197 | unoptimize_kprobe(ap, true); /* Fall back to normal kprobe */ |
afd66255 | 1198 | |
8b0914ea | 1199 | if (p->break_handler) { |
b918e5e6 | 1200 | if (ap->break_handler) |
36721656 | 1201 | return -EEXIST; |
b918e5e6 MH |
1202 | list_add_tail_rcu(&p->list, &ap->list); |
1203 | ap->break_handler = aggr_break_handler; | |
8b0914ea | 1204 | } else |
b918e5e6 MH |
1205 | list_add_rcu(&p->list, &ap->list); |
1206 | if (p->post_handler && !ap->post_handler) | |
1207 | ap->post_handler = aggr_post_handler; | |
de5bd88d | 1208 | |
8b0914ea PP |
1209 | return 0; |
1210 | } | |
1211 | ||
64f562c6 AM |
1212 | /* |
1213 | * Fill in the required fields of the "manager kprobe". Replace the | |
1214 | * earlier kprobe in the hlist with the manager kprobe | |
1215 | */ | |
55479f64 | 1216 | static void init_aggr_kprobe(struct kprobe *ap, struct kprobe *p) |
64f562c6 | 1217 | { |
afd66255 | 1218 | /* Copy p's insn slot to ap */ |
8b0914ea | 1219 | copy_kprobe(p, ap); |
a9ad965e | 1220 | flush_insn_slot(ap); |
64f562c6 | 1221 | ap->addr = p->addr; |
afd66255 | 1222 | ap->flags = p->flags & ~KPROBE_FLAG_OPTIMIZED; |
64f562c6 | 1223 | ap->pre_handler = aggr_pre_handler; |
64f562c6 | 1224 | ap->fault_handler = aggr_fault_handler; |
e8386a0c MH |
1225 | /* We don't care the kprobe which has gone. */ |
1226 | if (p->post_handler && !kprobe_gone(p)) | |
36721656 | 1227 | ap->post_handler = aggr_post_handler; |
e8386a0c | 1228 | if (p->break_handler && !kprobe_gone(p)) |
36721656 | 1229 | ap->break_handler = aggr_break_handler; |
64f562c6 AM |
1230 | |
1231 | INIT_LIST_HEAD(&ap->list); | |
afd66255 | 1232 | INIT_HLIST_NODE(&ap->hlist); |
64f562c6 | 1233 | |
afd66255 | 1234 | list_add_rcu(&p->list, &ap->list); |
adad0f33 | 1235 | hlist_replace_rcu(&p->hlist, &ap->hlist); |
64f562c6 AM |
1236 | } |
1237 | ||
1238 | /* | |
1239 | * This is the second or subsequent kprobe at the address - handle | |
1240 | * the intricacies | |
64f562c6 | 1241 | */ |
55479f64 | 1242 | static int register_aggr_kprobe(struct kprobe *orig_p, struct kprobe *p) |
64f562c6 AM |
1243 | { |
1244 | int ret = 0; | |
6d8e40a8 | 1245 | struct kprobe *ap = orig_p; |
64f562c6 | 1246 | |
25764288 MH |
1247 | /* For preparing optimization, jump_label_text_reserved() is called */ |
1248 | jump_label_lock(); | |
1249 | /* | |
1250 | * Get online CPUs to avoid text_mutex deadlock.with stop machine, | |
1251 | * which is invoked by unoptimize_kprobe() in add_new_kprobe() | |
1252 | */ | |
1253 | get_online_cpus(); | |
1254 | mutex_lock(&text_mutex); | |
1255 | ||
6d8e40a8 MH |
1256 | if (!kprobe_aggrprobe(orig_p)) { |
1257 | /* If orig_p is not an aggr_kprobe, create new aggr_kprobe. */ | |
1258 | ap = alloc_aggr_kprobe(orig_p); | |
25764288 MH |
1259 | if (!ap) { |
1260 | ret = -ENOMEM; | |
1261 | goto out; | |
1262 | } | |
6d8e40a8 | 1263 | init_aggr_kprobe(ap, orig_p); |
6274de49 | 1264 | } else if (kprobe_unused(ap)) |
0490cd1f MH |
1265 | /* This probe is going to die. Rescue it */ |
1266 | reuse_unused_kprobe(ap); | |
b918e5e6 MH |
1267 | |
1268 | if (kprobe_gone(ap)) { | |
e8386a0c MH |
1269 | /* |
1270 | * Attempting to insert new probe at the same location that | |
1271 | * had a probe in the module vaddr area which already | |
1272 | * freed. So, the instruction slot has already been | |
1273 | * released. We need a new slot for the new probe. | |
1274 | */ | |
b918e5e6 | 1275 | ret = arch_prepare_kprobe(ap); |
e8386a0c | 1276 | if (ret) |
b918e5e6 MH |
1277 | /* |
1278 | * Even if fail to allocate new slot, don't need to | |
1279 | * free aggr_probe. It will be used next time, or | |
1280 | * freed by unregister_kprobe. | |
1281 | */ | |
25764288 | 1282 | goto out; |
de5bd88d | 1283 | |
afd66255 MH |
1284 | /* Prepare optimized instructions if possible. */ |
1285 | prepare_optimized_kprobe(ap); | |
1286 | ||
e8386a0c | 1287 | /* |
de5bd88d MH |
1288 | * Clear gone flag to prevent allocating new slot again, and |
1289 | * set disabled flag because it is not armed yet. | |
e8386a0c | 1290 | */ |
de5bd88d MH |
1291 | ap->flags = (ap->flags & ~KPROBE_FLAG_GONE) |
1292 | | KPROBE_FLAG_DISABLED; | |
e8386a0c | 1293 | } |
b918e5e6 | 1294 | |
afd66255 | 1295 | /* Copy ap's insn slot to p */ |
b918e5e6 | 1296 | copy_kprobe(ap, p); |
25764288 MH |
1297 | ret = add_new_kprobe(ap, p); |
1298 | ||
1299 | out: | |
1300 | mutex_unlock(&text_mutex); | |
1301 | put_online_cpus(); | |
1302 | jump_label_unlock(); | |
1303 | ||
1304 | if (ret == 0 && kprobe_disabled(ap) && !kprobe_disabled(p)) { | |
1305 | ap->flags &= ~KPROBE_FLAG_DISABLED; | |
1306 | if (!kprobes_all_disarmed) | |
1307 | /* Arm the breakpoint again. */ | |
1308 | arm_kprobe(ap); | |
1309 | } | |
1310 | return ret; | |
64f562c6 AM |
1311 | } |
1312 | ||
be8f2743 MH |
1313 | bool __weak arch_within_kprobe_blacklist(unsigned long addr) |
1314 | { | |
1315 | /* The __kprobes marked functions and entry code must not be probed */ | |
1316 | return addr >= (unsigned long)__kprobes_text_start && | |
1317 | addr < (unsigned long)__kprobes_text_end; | |
1318 | } | |
1319 | ||
55479f64 | 1320 | static bool within_kprobe_blacklist(unsigned long addr) |
d0aaff97 | 1321 | { |
376e2424 | 1322 | struct kprobe_blacklist_entry *ent; |
3d8d996e | 1323 | |
be8f2743 | 1324 | if (arch_within_kprobe_blacklist(addr)) |
376e2424 | 1325 | return true; |
3d8d996e SD |
1326 | /* |
1327 | * If there exists a kprobe_blacklist, verify and | |
1328 | * fail any probe registration in the prohibited area | |
1329 | */ | |
376e2424 MH |
1330 | list_for_each_entry(ent, &kprobe_blacklist, list) { |
1331 | if (addr >= ent->start_addr && addr < ent->end_addr) | |
1332 | return true; | |
3d8d996e | 1333 | } |
376e2424 MH |
1334 | |
1335 | return false; | |
d0aaff97 PP |
1336 | } |
1337 | ||
b2a5cd69 MH |
1338 | /* |
1339 | * If we have a symbol_name argument, look it up and add the offset field | |
1340 | * to it. This way, we can specify a relative address to a symbol. | |
bc81d48d MH |
1341 | * This returns encoded errors if it fails to look up symbol or invalid |
1342 | * combination of parameters. | |
b2a5cd69 | 1343 | */ |
55479f64 | 1344 | static kprobe_opcode_t *kprobe_addr(struct kprobe *p) |
b2a5cd69 MH |
1345 | { |
1346 | kprobe_opcode_t *addr = p->addr; | |
bc81d48d MH |
1347 | |
1348 | if ((p->symbol_name && p->addr) || | |
1349 | (!p->symbol_name && !p->addr)) | |
1350 | goto invalid; | |
1351 | ||
b2a5cd69 | 1352 | if (p->symbol_name) { |
b2a5cd69 | 1353 | kprobe_lookup_name(p->symbol_name, addr); |
bc81d48d MH |
1354 | if (!addr) |
1355 | return ERR_PTR(-ENOENT); | |
b2a5cd69 MH |
1356 | } |
1357 | ||
bc81d48d MH |
1358 | addr = (kprobe_opcode_t *)(((char *)addr) + p->offset); |
1359 | if (addr) | |
1360 | return addr; | |
1361 | ||
1362 | invalid: | |
1363 | return ERR_PTR(-EINVAL); | |
b2a5cd69 MH |
1364 | } |
1365 | ||
1f0ab409 | 1366 | /* Check passed kprobe is valid and return kprobe in kprobe_table. */ |
55479f64 | 1367 | static struct kprobe *__get_valid_kprobe(struct kprobe *p) |
1f0ab409 | 1368 | { |
6d8e40a8 | 1369 | struct kprobe *ap, *list_p; |
1f0ab409 | 1370 | |
6d8e40a8 MH |
1371 | ap = get_kprobe(p->addr); |
1372 | if (unlikely(!ap)) | |
1f0ab409 AM |
1373 | return NULL; |
1374 | ||
6d8e40a8 MH |
1375 | if (p != ap) { |
1376 | list_for_each_entry_rcu(list_p, &ap->list, list) | |
1f0ab409 AM |
1377 | if (list_p == p) |
1378 | /* kprobe p is a valid probe */ | |
1379 | goto valid; | |
1380 | return NULL; | |
1381 | } | |
1382 | valid: | |
6d8e40a8 | 1383 | return ap; |
1f0ab409 AM |
1384 | } |
1385 | ||
1386 | /* Return error if the kprobe is being re-registered */ | |
1387 | static inline int check_kprobe_rereg(struct kprobe *p) | |
1388 | { | |
1389 | int ret = 0; | |
1f0ab409 AM |
1390 | |
1391 | mutex_lock(&kprobe_mutex); | |
6d8e40a8 | 1392 | if (__get_valid_kprobe(p)) |
1f0ab409 AM |
1393 | ret = -EINVAL; |
1394 | mutex_unlock(&kprobe_mutex); | |
6d8e40a8 | 1395 | |
1f0ab409 AM |
1396 | return ret; |
1397 | } | |
1398 | ||
55479f64 MH |
1399 | static int check_kprobe_address_safe(struct kprobe *p, |
1400 | struct module **probed_mod) | |
1da177e4 LT |
1401 | { |
1402 | int ret = 0; | |
ae6aa16f MH |
1403 | unsigned long ftrace_addr; |
1404 | ||
1405 | /* | |
1406 | * If the address is located on a ftrace nop, set the | |
1407 | * breakpoint to the following instruction. | |
1408 | */ | |
1409 | ftrace_addr = ftrace_location((unsigned long)p->addr); | |
1410 | if (ftrace_addr) { | |
e7dbfe34 | 1411 | #ifdef CONFIG_KPROBES_ON_FTRACE |
ae6aa16f MH |
1412 | /* Given address is not on the instruction boundary */ |
1413 | if ((unsigned long)p->addr != ftrace_addr) | |
1414 | return -EILSEQ; | |
ae6aa16f | 1415 | p->flags |= KPROBE_FLAG_FTRACE; |
e7dbfe34 | 1416 | #else /* !CONFIG_KPROBES_ON_FTRACE */ |
ae6aa16f MH |
1417 | return -EINVAL; |
1418 | #endif | |
1419 | } | |
1f0ab409 | 1420 | |
91bad2f8 | 1421 | jump_label_lock(); |
de31c3ca | 1422 | preempt_disable(); |
f7fa6ef0 MH |
1423 | |
1424 | /* Ensure it is not in reserved area nor out of text */ | |
ec30c5f3 | 1425 | if (!kernel_text_address((unsigned long) p->addr) || |
376e2424 | 1426 | within_kprobe_blacklist((unsigned long) p->addr) || |
f986a499 PN |
1427 | jump_label_text_reserved(p->addr, p->addr)) { |
1428 | ret = -EINVAL; | |
f7fa6ef0 | 1429 | goto out; |
f986a499 | 1430 | } |
b3e55c72 | 1431 | |
f7fa6ef0 MH |
1432 | /* Check if are we probing a module */ |
1433 | *probed_mod = __module_text_address((unsigned long) p->addr); | |
1434 | if (*probed_mod) { | |
6f716acd | 1435 | /* |
e8386a0c MH |
1436 | * We must hold a refcount of the probed module while updating |
1437 | * its code to prohibit unexpected unloading. | |
df019b1d | 1438 | */ |
f7fa6ef0 MH |
1439 | if (unlikely(!try_module_get(*probed_mod))) { |
1440 | ret = -ENOENT; | |
1441 | goto out; | |
1442 | } | |
de31c3ca | 1443 | |
f24659d9 MH |
1444 | /* |
1445 | * If the module freed .init.text, we couldn't insert | |
1446 | * kprobes in there. | |
1447 | */ | |
f7fa6ef0 MH |
1448 | if (within_module_init((unsigned long)p->addr, *probed_mod) && |
1449 | (*probed_mod)->state != MODULE_STATE_COMING) { | |
1450 | module_put(*probed_mod); | |
1451 | *probed_mod = NULL; | |
1452 | ret = -ENOENT; | |
f24659d9 | 1453 | } |
df019b1d | 1454 | } |
f7fa6ef0 | 1455 | out: |
a189d035 | 1456 | preempt_enable(); |
de31c3ca | 1457 | jump_label_unlock(); |
1da177e4 | 1458 | |
f7fa6ef0 MH |
1459 | return ret; |
1460 | } | |
1461 | ||
55479f64 | 1462 | int register_kprobe(struct kprobe *p) |
f7fa6ef0 MH |
1463 | { |
1464 | int ret; | |
1465 | struct kprobe *old_p; | |
1466 | struct module *probed_mod; | |
1467 | kprobe_opcode_t *addr; | |
1468 | ||
1469 | /* Adjust probe address from symbol */ | |
1470 | addr = kprobe_addr(p); | |
1471 | if (IS_ERR(addr)) | |
1472 | return PTR_ERR(addr); | |
1473 | p->addr = addr; | |
1474 | ||
1475 | ret = check_kprobe_rereg(p); | |
1476 | if (ret) | |
1477 | return ret; | |
1478 | ||
1479 | /* User can pass only KPROBE_FLAG_DISABLED to register_kprobe */ | |
1480 | p->flags &= KPROBE_FLAG_DISABLED; | |
3516a460 | 1481 | p->nmissed = 0; |
9861668f | 1482 | INIT_LIST_HEAD(&p->list); |
afd66255 | 1483 | |
f7fa6ef0 MH |
1484 | ret = check_kprobe_address_safe(p, &probed_mod); |
1485 | if (ret) | |
1486 | return ret; | |
1487 | ||
1488 | mutex_lock(&kprobe_mutex); | |
afd66255 | 1489 | |
64f562c6 AM |
1490 | old_p = get_kprobe(p->addr); |
1491 | if (old_p) { | |
afd66255 | 1492 | /* Since this may unoptimize old_p, locking text_mutex. */ |
64f562c6 | 1493 | ret = register_aggr_kprobe(old_p, p); |
1da177e4 LT |
1494 | goto out; |
1495 | } | |
1da177e4 | 1496 | |
25764288 | 1497 | mutex_lock(&text_mutex); /* Avoiding text modification */ |
ae6aa16f | 1498 | ret = prepare_kprobe(p); |
25764288 | 1499 | mutex_unlock(&text_mutex); |
6f716acd | 1500 | if (ret) |
afd66255 | 1501 | goto out; |
49a2a1b8 | 1502 | |
64f562c6 | 1503 | INIT_HLIST_NODE(&p->hlist); |
3516a460 | 1504 | hlist_add_head_rcu(&p->hlist, |
1da177e4 LT |
1505 | &kprobe_table[hash_ptr(p->addr, KPROBE_HASH_BITS)]); |
1506 | ||
de5bd88d | 1507 | if (!kprobes_all_disarmed && !kprobe_disabled(p)) |
25764288 | 1508 | arm_kprobe(p); |
afd66255 MH |
1509 | |
1510 | /* Try to optimize kprobe */ | |
1511 | try_to_optimize_kprobe(p); | |
74a0b576 | 1512 | |
1da177e4 | 1513 | out: |
7a7d1cf9 | 1514 | mutex_unlock(&kprobe_mutex); |
49a2a1b8 | 1515 | |
e8386a0c | 1516 | if (probed_mod) |
df019b1d | 1517 | module_put(probed_mod); |
e8386a0c | 1518 | |
1da177e4 LT |
1519 | return ret; |
1520 | } | |
99081ab5 | 1521 | EXPORT_SYMBOL_GPL(register_kprobe); |
1da177e4 | 1522 | |
6f0f1dd7 | 1523 | /* Check if all probes on the aggrprobe are disabled */ |
55479f64 | 1524 | static int aggr_kprobe_disabled(struct kprobe *ap) |
6f0f1dd7 MH |
1525 | { |
1526 | struct kprobe *kp; | |
1527 | ||
1528 | list_for_each_entry_rcu(kp, &ap->list, list) | |
1529 | if (!kprobe_disabled(kp)) | |
1530 | /* | |
1531 | * There is an active probe on the list. | |
1532 | * We can't disable this ap. | |
1533 | */ | |
1534 | return 0; | |
1535 | ||
1536 | return 1; | |
1537 | } | |
1538 | ||
1539 | /* Disable one kprobe: Make sure called under kprobe_mutex is locked */ | |
55479f64 | 1540 | static struct kprobe *__disable_kprobe(struct kprobe *p) |
6f0f1dd7 MH |
1541 | { |
1542 | struct kprobe *orig_p; | |
1543 | ||
1544 | /* Get an original kprobe for return */ | |
1545 | orig_p = __get_valid_kprobe(p); | |
1546 | if (unlikely(orig_p == NULL)) | |
1547 | return NULL; | |
1548 | ||
1549 | if (!kprobe_disabled(p)) { | |
1550 | /* Disable probe if it is a child probe */ | |
1551 | if (p != orig_p) | |
1552 | p->flags |= KPROBE_FLAG_DISABLED; | |
1553 | ||
1554 | /* Try to disarm and disable this/parent probe */ | |
1555 | if (p == orig_p || aggr_kprobe_disabled(orig_p)) { | |
ae6aa16f | 1556 | disarm_kprobe(orig_p, true); |
6f0f1dd7 MH |
1557 | orig_p->flags |= KPROBE_FLAG_DISABLED; |
1558 | } | |
1559 | } | |
1560 | ||
1561 | return orig_p; | |
1562 | } | |
1563 | ||
de5bd88d MH |
1564 | /* |
1565 | * Unregister a kprobe without a scheduler synchronization. | |
1566 | */ | |
55479f64 | 1567 | static int __unregister_kprobe_top(struct kprobe *p) |
de5bd88d | 1568 | { |
6d8e40a8 | 1569 | struct kprobe *ap, *list_p; |
de5bd88d | 1570 | |
6f0f1dd7 MH |
1571 | /* Disable kprobe. This will disarm it if needed. */ |
1572 | ap = __disable_kprobe(p); | |
6d8e40a8 | 1573 | if (ap == NULL) |
de5bd88d MH |
1574 | return -EINVAL; |
1575 | ||
6f0f1dd7 | 1576 | if (ap == p) |
bf8f6e5b | 1577 | /* |
6f0f1dd7 MH |
1578 | * This probe is an independent(and non-optimized) kprobe |
1579 | * (not an aggrprobe). Remove from the hash list. | |
bf8f6e5b | 1580 | */ |
6f0f1dd7 MH |
1581 | goto disarmed; |
1582 | ||
1583 | /* Following process expects this probe is an aggrprobe */ | |
1584 | WARN_ON(!kprobe_aggrprobe(ap)); | |
1585 | ||
6274de49 MH |
1586 | if (list_is_singular(&ap->list) && kprobe_disarmed(ap)) |
1587 | /* | |
1588 | * !disarmed could be happen if the probe is under delayed | |
1589 | * unoptimizing. | |
1590 | */ | |
6f0f1dd7 MH |
1591 | goto disarmed; |
1592 | else { | |
1593 | /* If disabling probe has special handlers, update aggrprobe */ | |
e8386a0c | 1594 | if (p->break_handler && !kprobe_gone(p)) |
6d8e40a8 | 1595 | ap->break_handler = NULL; |
e8386a0c | 1596 | if (p->post_handler && !kprobe_gone(p)) { |
6d8e40a8 | 1597 | list_for_each_entry_rcu(list_p, &ap->list, list) { |
9861668f MH |
1598 | if ((list_p != p) && (list_p->post_handler)) |
1599 | goto noclean; | |
1600 | } | |
6d8e40a8 | 1601 | ap->post_handler = NULL; |
9861668f MH |
1602 | } |
1603 | noclean: | |
6f0f1dd7 MH |
1604 | /* |
1605 | * Remove from the aggrprobe: this path will do nothing in | |
1606 | * __unregister_kprobe_bottom(). | |
1607 | */ | |
49a2a1b8 | 1608 | list_del_rcu(&p->list); |
6f0f1dd7 MH |
1609 | if (!kprobe_disabled(ap) && !kprobes_all_disarmed) |
1610 | /* | |
1611 | * Try to optimize this probe again, because post | |
1612 | * handler may have been changed. | |
1613 | */ | |
1614 | optimize_kprobe(ap); | |
49a2a1b8 | 1615 | } |
9861668f | 1616 | return 0; |
6f0f1dd7 MH |
1617 | |
1618 | disarmed: | |
6274de49 | 1619 | BUG_ON(!kprobe_disarmed(ap)); |
6f0f1dd7 MH |
1620 | hlist_del_rcu(&ap->hlist); |
1621 | return 0; | |
9861668f | 1622 | } |
3516a460 | 1623 | |
55479f64 | 1624 | static void __unregister_kprobe_bottom(struct kprobe *p) |
9861668f | 1625 | { |
6d8e40a8 | 1626 | struct kprobe *ap; |
b3e55c72 | 1627 | |
e8386a0c | 1628 | if (list_empty(&p->list)) |
6274de49 | 1629 | /* This is an independent kprobe */ |
0498b635 | 1630 | arch_remove_kprobe(p); |
e8386a0c | 1631 | else if (list_is_singular(&p->list)) { |
6274de49 | 1632 | /* This is the last child of an aggrprobe */ |
6d8e40a8 | 1633 | ap = list_entry(p->list.next, struct kprobe, list); |
e8386a0c | 1634 | list_del(&p->list); |
6d8e40a8 | 1635 | free_aggr_kprobe(ap); |
9861668f | 1636 | } |
6274de49 | 1637 | /* Otherwise, do nothing. */ |
9861668f MH |
1638 | } |
1639 | ||
55479f64 | 1640 | int register_kprobes(struct kprobe **kps, int num) |
9861668f MH |
1641 | { |
1642 | int i, ret = 0; | |
1643 | ||
1644 | if (num <= 0) | |
1645 | return -EINVAL; | |
1646 | for (i = 0; i < num; i++) { | |
49ad2fd7 | 1647 | ret = register_kprobe(kps[i]); |
67dddaad MH |
1648 | if (ret < 0) { |
1649 | if (i > 0) | |
1650 | unregister_kprobes(kps, i); | |
9861668f | 1651 | break; |
36721656 | 1652 | } |
49a2a1b8 | 1653 | } |
9861668f MH |
1654 | return ret; |
1655 | } | |
99081ab5 | 1656 | EXPORT_SYMBOL_GPL(register_kprobes); |
9861668f | 1657 | |
55479f64 | 1658 | void unregister_kprobe(struct kprobe *p) |
9861668f MH |
1659 | { |
1660 | unregister_kprobes(&p, 1); | |
1661 | } | |
99081ab5 | 1662 | EXPORT_SYMBOL_GPL(unregister_kprobe); |
9861668f | 1663 | |
55479f64 | 1664 | void unregister_kprobes(struct kprobe **kps, int num) |
9861668f MH |
1665 | { |
1666 | int i; | |
1667 | ||
1668 | if (num <= 0) | |
1669 | return; | |
1670 | mutex_lock(&kprobe_mutex); | |
1671 | for (i = 0; i < num; i++) | |
1672 | if (__unregister_kprobe_top(kps[i]) < 0) | |
1673 | kps[i]->addr = NULL; | |
1674 | mutex_unlock(&kprobe_mutex); | |
1675 | ||
1676 | synchronize_sched(); | |
1677 | for (i = 0; i < num; i++) | |
1678 | if (kps[i]->addr) | |
1679 | __unregister_kprobe_bottom(kps[i]); | |
1da177e4 | 1680 | } |
99081ab5 | 1681 | EXPORT_SYMBOL_GPL(unregister_kprobes); |
1da177e4 LT |
1682 | |
1683 | static struct notifier_block kprobe_exceptions_nb = { | |
3d5631e0 AK |
1684 | .notifier_call = kprobe_exceptions_notify, |
1685 | .priority = 0x7fffffff /* we need to be notified first */ | |
1686 | }; | |
1687 | ||
3d7e3382 ME |
1688 | unsigned long __weak arch_deref_entry_point(void *entry) |
1689 | { | |
1690 | return (unsigned long)entry; | |
1691 | } | |
1da177e4 | 1692 | |
55479f64 | 1693 | int register_jprobes(struct jprobe **jps, int num) |
1da177e4 | 1694 | { |
26b31c19 MH |
1695 | struct jprobe *jp; |
1696 | int ret = 0, i; | |
3d7e3382 | 1697 | |
26b31c19 | 1698 | if (num <= 0) |
3d7e3382 | 1699 | return -EINVAL; |
26b31c19 | 1700 | for (i = 0; i < num; i++) { |
05662bdb | 1701 | unsigned long addr, offset; |
26b31c19 MH |
1702 | jp = jps[i]; |
1703 | addr = arch_deref_entry_point(jp->entry); | |
1704 | ||
05662bdb NK |
1705 | /* Verify probepoint is a function entry point */ |
1706 | if (kallsyms_lookup_size_offset(addr, NULL, &offset) && | |
1707 | offset == 0) { | |
1708 | jp->kp.pre_handler = setjmp_pre_handler; | |
1709 | jp->kp.break_handler = longjmp_break_handler; | |
1710 | ret = register_kprobe(&jp->kp); | |
1711 | } else | |
1712 | ret = -EINVAL; | |
edbaadbe | 1713 | |
67dddaad MH |
1714 | if (ret < 0) { |
1715 | if (i > 0) | |
1716 | unregister_jprobes(jps, i); | |
26b31c19 MH |
1717 | break; |
1718 | } | |
1719 | } | |
1720 | return ret; | |
1721 | } | |
99081ab5 | 1722 | EXPORT_SYMBOL_GPL(register_jprobes); |
3d7e3382 | 1723 | |
55479f64 | 1724 | int register_jprobe(struct jprobe *jp) |
26b31c19 | 1725 | { |
49ad2fd7 | 1726 | return register_jprobes(&jp, 1); |
1da177e4 | 1727 | } |
99081ab5 | 1728 | EXPORT_SYMBOL_GPL(register_jprobe); |
1da177e4 | 1729 | |
55479f64 | 1730 | void unregister_jprobe(struct jprobe *jp) |
1da177e4 | 1731 | { |
26b31c19 MH |
1732 | unregister_jprobes(&jp, 1); |
1733 | } | |
99081ab5 | 1734 | EXPORT_SYMBOL_GPL(unregister_jprobe); |
26b31c19 | 1735 | |
55479f64 | 1736 | void unregister_jprobes(struct jprobe **jps, int num) |
26b31c19 MH |
1737 | { |
1738 | int i; | |
1739 | ||
1740 | if (num <= 0) | |
1741 | return; | |
1742 | mutex_lock(&kprobe_mutex); | |
1743 | for (i = 0; i < num; i++) | |
1744 | if (__unregister_kprobe_top(&jps[i]->kp) < 0) | |
1745 | jps[i]->kp.addr = NULL; | |
1746 | mutex_unlock(&kprobe_mutex); | |
1747 | ||
1748 | synchronize_sched(); | |
1749 | for (i = 0; i < num; i++) { | |
1750 | if (jps[i]->kp.addr) | |
1751 | __unregister_kprobe_bottom(&jps[i]->kp); | |
1752 | } | |
1da177e4 | 1753 | } |
99081ab5 | 1754 | EXPORT_SYMBOL_GPL(unregister_jprobes); |
1da177e4 | 1755 | |
9edddaa2 | 1756 | #ifdef CONFIG_KRETPROBES |
e65cefe8 AB |
1757 | /* |
1758 | * This kprobe pre_handler is registered with every kretprobe. When probe | |
1759 | * hits it will set up the return probe. | |
1760 | */ | |
1761 | static int __kprobes pre_handler_kretprobe(struct kprobe *p, | |
1762 | struct pt_regs *regs) | |
1763 | { | |
1764 | struct kretprobe *rp = container_of(p, struct kretprobe, kp); | |
ef53d9c5 S |
1765 | unsigned long hash, flags = 0; |
1766 | struct kretprobe_instance *ri; | |
e65cefe8 AB |
1767 | |
1768 | /*TODO: consider to only swap the RA after the last pre_handler fired */ | |
ef53d9c5 | 1769 | hash = hash_ptr(current, KPROBE_HASH_BITS); |
ec484608 | 1770 | raw_spin_lock_irqsave(&rp->lock, flags); |
4c4308cb | 1771 | if (!hlist_empty(&rp->free_instances)) { |
4c4308cb | 1772 | ri = hlist_entry(rp->free_instances.first, |
ef53d9c5 S |
1773 | struct kretprobe_instance, hlist); |
1774 | hlist_del(&ri->hlist); | |
ec484608 | 1775 | raw_spin_unlock_irqrestore(&rp->lock, flags); |
ef53d9c5 | 1776 | |
4c4308cb CH |
1777 | ri->rp = rp; |
1778 | ri->task = current; | |
f47cd9b5 | 1779 | |
55ca6140 JL |
1780 | if (rp->entry_handler && rp->entry_handler(ri, regs)) { |
1781 | raw_spin_lock_irqsave(&rp->lock, flags); | |
1782 | hlist_add_head(&ri->hlist, &rp->free_instances); | |
1783 | raw_spin_unlock_irqrestore(&rp->lock, flags); | |
f47cd9b5 | 1784 | return 0; |
55ca6140 | 1785 | } |
f47cd9b5 | 1786 | |
4c4308cb CH |
1787 | arch_prepare_kretprobe(ri, regs); |
1788 | ||
1789 | /* XXX(hch): why is there no hlist_move_head? */ | |
ef53d9c5 S |
1790 | INIT_HLIST_NODE(&ri->hlist); |
1791 | kretprobe_table_lock(hash, &flags); | |
1792 | hlist_add_head(&ri->hlist, &kretprobe_inst_table[hash]); | |
1793 | kretprobe_table_unlock(hash, &flags); | |
1794 | } else { | |
4c4308cb | 1795 | rp->nmissed++; |
ec484608 | 1796 | raw_spin_unlock_irqrestore(&rp->lock, flags); |
ef53d9c5 | 1797 | } |
e65cefe8 AB |
1798 | return 0; |
1799 | } | |
1800 | ||
55479f64 | 1801 | int register_kretprobe(struct kretprobe *rp) |
b94cce92 HN |
1802 | { |
1803 | int ret = 0; | |
1804 | struct kretprobe_instance *inst; | |
1805 | int i; | |
b2a5cd69 | 1806 | void *addr; |
f438d914 MH |
1807 | |
1808 | if (kretprobe_blacklist_size) { | |
b2a5cd69 | 1809 | addr = kprobe_addr(&rp->kp); |
bc81d48d MH |
1810 | if (IS_ERR(addr)) |
1811 | return PTR_ERR(addr); | |
f438d914 MH |
1812 | |
1813 | for (i = 0; kretprobe_blacklist[i].name != NULL; i++) { | |
1814 | if (kretprobe_blacklist[i].addr == addr) | |
1815 | return -EINVAL; | |
1816 | } | |
1817 | } | |
b94cce92 HN |
1818 | |
1819 | rp->kp.pre_handler = pre_handler_kretprobe; | |
7522a842 AM |
1820 | rp->kp.post_handler = NULL; |
1821 | rp->kp.fault_handler = NULL; | |
1822 | rp->kp.break_handler = NULL; | |
b94cce92 HN |
1823 | |
1824 | /* Pre-allocate memory for max kretprobe instances */ | |
1825 | if (rp->maxactive <= 0) { | |
1826 | #ifdef CONFIG_PREEMPT | |
c2ef6661 | 1827 | rp->maxactive = max_t(unsigned int, 10, 2*num_possible_cpus()); |
b94cce92 | 1828 | #else |
4dae560f | 1829 | rp->maxactive = num_possible_cpus(); |
b94cce92 HN |
1830 | #endif |
1831 | } | |
ec484608 | 1832 | raw_spin_lock_init(&rp->lock); |
b94cce92 HN |
1833 | INIT_HLIST_HEAD(&rp->free_instances); |
1834 | for (i = 0; i < rp->maxactive; i++) { | |
f47cd9b5 AS |
1835 | inst = kmalloc(sizeof(struct kretprobe_instance) + |
1836 | rp->data_size, GFP_KERNEL); | |
b94cce92 HN |
1837 | if (inst == NULL) { |
1838 | free_rp_inst(rp); | |
1839 | return -ENOMEM; | |
1840 | } | |
ef53d9c5 S |
1841 | INIT_HLIST_NODE(&inst->hlist); |
1842 | hlist_add_head(&inst->hlist, &rp->free_instances); | |
b94cce92 HN |
1843 | } |
1844 | ||
1845 | rp->nmissed = 0; | |
1846 | /* Establish function entry probe point */ | |
49ad2fd7 | 1847 | ret = register_kprobe(&rp->kp); |
4a296e07 | 1848 | if (ret != 0) |
b94cce92 HN |
1849 | free_rp_inst(rp); |
1850 | return ret; | |
1851 | } | |
99081ab5 | 1852 | EXPORT_SYMBOL_GPL(register_kretprobe); |
b94cce92 | 1853 | |
55479f64 | 1854 | int register_kretprobes(struct kretprobe **rps, int num) |
4a296e07 MH |
1855 | { |
1856 | int ret = 0, i; | |
1857 | ||
1858 | if (num <= 0) | |
1859 | return -EINVAL; | |
1860 | for (i = 0; i < num; i++) { | |
49ad2fd7 | 1861 | ret = register_kretprobe(rps[i]); |
67dddaad MH |
1862 | if (ret < 0) { |
1863 | if (i > 0) | |
1864 | unregister_kretprobes(rps, i); | |
4a296e07 MH |
1865 | break; |
1866 | } | |
1867 | } | |
1868 | return ret; | |
1869 | } | |
99081ab5 | 1870 | EXPORT_SYMBOL_GPL(register_kretprobes); |
4a296e07 | 1871 | |
55479f64 | 1872 | void unregister_kretprobe(struct kretprobe *rp) |
4a296e07 MH |
1873 | { |
1874 | unregister_kretprobes(&rp, 1); | |
1875 | } | |
99081ab5 | 1876 | EXPORT_SYMBOL_GPL(unregister_kretprobe); |
4a296e07 | 1877 | |
55479f64 | 1878 | void unregister_kretprobes(struct kretprobe **rps, int num) |
4a296e07 MH |
1879 | { |
1880 | int i; | |
1881 | ||
1882 | if (num <= 0) | |
1883 | return; | |
1884 | mutex_lock(&kprobe_mutex); | |
1885 | for (i = 0; i < num; i++) | |
1886 | if (__unregister_kprobe_top(&rps[i]->kp) < 0) | |
1887 | rps[i]->kp.addr = NULL; | |
1888 | mutex_unlock(&kprobe_mutex); | |
1889 | ||
1890 | synchronize_sched(); | |
1891 | for (i = 0; i < num; i++) { | |
1892 | if (rps[i]->kp.addr) { | |
1893 | __unregister_kprobe_bottom(&rps[i]->kp); | |
1894 | cleanup_rp_inst(rps[i]); | |
1895 | } | |
1896 | } | |
1897 | } | |
99081ab5 | 1898 | EXPORT_SYMBOL_GPL(unregister_kretprobes); |
4a296e07 | 1899 | |
9edddaa2 | 1900 | #else /* CONFIG_KRETPROBES */ |
55479f64 | 1901 | int register_kretprobe(struct kretprobe *rp) |
b94cce92 HN |
1902 | { |
1903 | return -ENOSYS; | |
1904 | } | |
99081ab5 | 1905 | EXPORT_SYMBOL_GPL(register_kretprobe); |
b94cce92 | 1906 | |
55479f64 | 1907 | int register_kretprobes(struct kretprobe **rps, int num) |
346fd59b | 1908 | { |
4a296e07 | 1909 | return -ENOSYS; |
346fd59b | 1910 | } |
99081ab5 MH |
1911 | EXPORT_SYMBOL_GPL(register_kretprobes); |
1912 | ||
55479f64 | 1913 | void unregister_kretprobe(struct kretprobe *rp) |
b94cce92 | 1914 | { |
4a296e07 | 1915 | } |
99081ab5 | 1916 | EXPORT_SYMBOL_GPL(unregister_kretprobe); |
b94cce92 | 1917 | |
55479f64 | 1918 | void unregister_kretprobes(struct kretprobe **rps, int num) |
4a296e07 MH |
1919 | { |
1920 | } | |
99081ab5 | 1921 | EXPORT_SYMBOL_GPL(unregister_kretprobes); |
4c4308cb | 1922 | |
4a296e07 MH |
1923 | static int __kprobes pre_handler_kretprobe(struct kprobe *p, |
1924 | struct pt_regs *regs) | |
1925 | { | |
1926 | return 0; | |
b94cce92 HN |
1927 | } |
1928 | ||
4a296e07 MH |
1929 | #endif /* CONFIG_KRETPROBES */ |
1930 | ||
e8386a0c | 1931 | /* Set the kprobe gone and remove its instruction buffer. */ |
55479f64 | 1932 | static void kill_kprobe(struct kprobe *p) |
e8386a0c MH |
1933 | { |
1934 | struct kprobe *kp; | |
de5bd88d | 1935 | |
e8386a0c | 1936 | p->flags |= KPROBE_FLAG_GONE; |
afd66255 | 1937 | if (kprobe_aggrprobe(p)) { |
e8386a0c MH |
1938 | /* |
1939 | * If this is an aggr_kprobe, we have to list all the | |
1940 | * chained probes and mark them GONE. | |
1941 | */ | |
1942 | list_for_each_entry_rcu(kp, &p->list, list) | |
1943 | kp->flags |= KPROBE_FLAG_GONE; | |
1944 | p->post_handler = NULL; | |
1945 | p->break_handler = NULL; | |
afd66255 | 1946 | kill_optimized_kprobe(p); |
e8386a0c MH |
1947 | } |
1948 | /* | |
1949 | * Here, we can remove insn_slot safely, because no thread calls | |
1950 | * the original probed function (which will be freed soon) any more. | |
1951 | */ | |
1952 | arch_remove_kprobe(p); | |
1953 | } | |
1954 | ||
c0614829 | 1955 | /* Disable one kprobe */ |
55479f64 | 1956 | int disable_kprobe(struct kprobe *kp) |
c0614829 MH |
1957 | { |
1958 | int ret = 0; | |
c0614829 MH |
1959 | |
1960 | mutex_lock(&kprobe_mutex); | |
1961 | ||
6f0f1dd7 MH |
1962 | /* Disable this kprobe */ |
1963 | if (__disable_kprobe(kp) == NULL) | |
c0614829 | 1964 | ret = -EINVAL; |
c0614829 | 1965 | |
c0614829 MH |
1966 | mutex_unlock(&kprobe_mutex); |
1967 | return ret; | |
1968 | } | |
1969 | EXPORT_SYMBOL_GPL(disable_kprobe); | |
1970 | ||
1971 | /* Enable one kprobe */ | |
55479f64 | 1972 | int enable_kprobe(struct kprobe *kp) |
c0614829 MH |
1973 | { |
1974 | int ret = 0; | |
1975 | struct kprobe *p; | |
1976 | ||
1977 | mutex_lock(&kprobe_mutex); | |
1978 | ||
1979 | /* Check whether specified probe is valid. */ | |
1980 | p = __get_valid_kprobe(kp); | |
1981 | if (unlikely(p == NULL)) { | |
1982 | ret = -EINVAL; | |
1983 | goto out; | |
1984 | } | |
1985 | ||
1986 | if (kprobe_gone(kp)) { | |
1987 | /* This kprobe has gone, we couldn't enable it. */ | |
1988 | ret = -EINVAL; | |
1989 | goto out; | |
1990 | } | |
1991 | ||
1992 | if (p != kp) | |
1993 | kp->flags &= ~KPROBE_FLAG_DISABLED; | |
1994 | ||
1995 | if (!kprobes_all_disarmed && kprobe_disabled(p)) { | |
1996 | p->flags &= ~KPROBE_FLAG_DISABLED; | |
1997 | arm_kprobe(p); | |
1998 | } | |
1999 | out: | |
2000 | mutex_unlock(&kprobe_mutex); | |
2001 | return ret; | |
2002 | } | |
2003 | EXPORT_SYMBOL_GPL(enable_kprobe); | |
2004 | ||
24851d24 FW |
2005 | void __kprobes dump_kprobe(struct kprobe *kp) |
2006 | { | |
2007 | printk(KERN_WARNING "Dumping kprobe:\n"); | |
2008 | printk(KERN_WARNING "Name: %s\nAddress: %p\nOffset: %x\n", | |
2009 | kp->symbol_name, kp->addr, kp->offset); | |
2010 | } | |
2011 | ||
376e2424 MH |
2012 | /* |
2013 | * Lookup and populate the kprobe_blacklist. | |
2014 | * | |
2015 | * Unlike the kretprobe blacklist, we'll need to determine | |
2016 | * the range of addresses that belong to the said functions, | |
2017 | * since a kprobe need not necessarily be at the beginning | |
2018 | * of a function. | |
2019 | */ | |
2020 | static int __init populate_kprobe_blacklist(unsigned long *start, | |
2021 | unsigned long *end) | |
2022 | { | |
2023 | unsigned long *iter; | |
2024 | struct kprobe_blacklist_entry *ent; | |
2025 | unsigned long offset = 0, size = 0; | |
2026 | ||
2027 | for (iter = start; iter < end; iter++) { | |
2028 | if (!kallsyms_lookup_size_offset(*iter, &size, &offset)) { | |
2029 | pr_err("Failed to find blacklist %p\n", (void *)*iter); | |
2030 | continue; | |
2031 | } | |
2032 | ||
2033 | ent = kmalloc(sizeof(*ent), GFP_KERNEL); | |
2034 | if (!ent) | |
2035 | return -ENOMEM; | |
2036 | ent->start_addr = *iter; | |
2037 | ent->end_addr = *iter + size; | |
2038 | INIT_LIST_HEAD(&ent->list); | |
2039 | list_add_tail(&ent->list, &kprobe_blacklist); | |
2040 | } | |
2041 | return 0; | |
2042 | } | |
2043 | ||
e8386a0c | 2044 | /* Module notifier call back, checking kprobes on the module */ |
55479f64 MH |
2045 | static int kprobes_module_callback(struct notifier_block *nb, |
2046 | unsigned long val, void *data) | |
e8386a0c MH |
2047 | { |
2048 | struct module *mod = data; | |
2049 | struct hlist_head *head; | |
e8386a0c MH |
2050 | struct kprobe *p; |
2051 | unsigned int i; | |
f24659d9 | 2052 | int checkcore = (val == MODULE_STATE_GOING); |
e8386a0c | 2053 | |
f24659d9 | 2054 | if (val != MODULE_STATE_GOING && val != MODULE_STATE_LIVE) |
e8386a0c MH |
2055 | return NOTIFY_DONE; |
2056 | ||
2057 | /* | |
f24659d9 MH |
2058 | * When MODULE_STATE_GOING was notified, both of module .text and |
2059 | * .init.text sections would be freed. When MODULE_STATE_LIVE was | |
2060 | * notified, only .init.text section would be freed. We need to | |
2061 | * disable kprobes which have been inserted in the sections. | |
e8386a0c MH |
2062 | */ |
2063 | mutex_lock(&kprobe_mutex); | |
2064 | for (i = 0; i < KPROBE_TABLE_SIZE; i++) { | |
2065 | head = &kprobe_table[i]; | |
b67bfe0d | 2066 | hlist_for_each_entry_rcu(p, head, hlist) |
f24659d9 MH |
2067 | if (within_module_init((unsigned long)p->addr, mod) || |
2068 | (checkcore && | |
2069 | within_module_core((unsigned long)p->addr, mod))) { | |
e8386a0c MH |
2070 | /* |
2071 | * The vaddr this probe is installed will soon | |
2072 | * be vfreed buy not synced to disk. Hence, | |
2073 | * disarming the breakpoint isn't needed. | |
2074 | */ | |
2075 | kill_kprobe(p); | |
2076 | } | |
2077 | } | |
2078 | mutex_unlock(&kprobe_mutex); | |
2079 | return NOTIFY_DONE; | |
2080 | } | |
2081 | ||
2082 | static struct notifier_block kprobe_module_nb = { | |
2083 | .notifier_call = kprobes_module_callback, | |
2084 | .priority = 0 | |
2085 | }; | |
2086 | ||
376e2424 MH |
2087 | /* Markers of _kprobe_blacklist section */ |
2088 | extern unsigned long __start_kprobe_blacklist[]; | |
2089 | extern unsigned long __stop_kprobe_blacklist[]; | |
2090 | ||
1da177e4 LT |
2091 | static int __init init_kprobes(void) |
2092 | { | |
2093 | int i, err = 0; | |
2094 | ||
2095 | /* FIXME allocate the probe table, currently defined statically */ | |
2096 | /* initialize all list heads */ | |
b94cce92 | 2097 | for (i = 0; i < KPROBE_TABLE_SIZE; i++) { |
1da177e4 | 2098 | INIT_HLIST_HEAD(&kprobe_table[i]); |
b94cce92 | 2099 | INIT_HLIST_HEAD(&kretprobe_inst_table[i]); |
ec484608 | 2100 | raw_spin_lock_init(&(kretprobe_table_locks[i].lock)); |
b94cce92 | 2101 | } |
1da177e4 | 2102 | |
376e2424 MH |
2103 | err = populate_kprobe_blacklist(__start_kprobe_blacklist, |
2104 | __stop_kprobe_blacklist); | |
2105 | if (err) { | |
2106 | pr_err("kprobes: failed to populate blacklist: %d\n", err); | |
2107 | pr_err("Please take care of using kprobes.\n"); | |
3d8d996e SD |
2108 | } |
2109 | ||
f438d914 MH |
2110 | if (kretprobe_blacklist_size) { |
2111 | /* lookup the function address from its name */ | |
2112 | for (i = 0; kretprobe_blacklist[i].name != NULL; i++) { | |
2113 | kprobe_lookup_name(kretprobe_blacklist[i].name, | |
2114 | kretprobe_blacklist[i].addr); | |
2115 | if (!kretprobe_blacklist[i].addr) | |
2116 | printk("kretprobe: lookup failed: %s\n", | |
2117 | kretprobe_blacklist[i].name); | |
2118 | } | |
2119 | } | |
2120 | ||
b2be84df MH |
2121 | #if defined(CONFIG_OPTPROBES) |
2122 | #if defined(__ARCH_WANT_KPROBES_INSN_SLOT) | |
afd66255 MH |
2123 | /* Init kprobe_optinsn_slots */ |
2124 | kprobe_optinsn_slots.insn_size = MAX_OPTINSN_SIZE; | |
2125 | #endif | |
b2be84df MH |
2126 | /* By default, kprobes can be optimized */ |
2127 | kprobes_allow_optimization = true; | |
2128 | #endif | |
afd66255 | 2129 | |
e579abeb MH |
2130 | /* By default, kprobes are armed */ |
2131 | kprobes_all_disarmed = false; | |
bf8f6e5b | 2132 | |
6772926b | 2133 | err = arch_init_kprobes(); |
802eae7c RL |
2134 | if (!err) |
2135 | err = register_die_notifier(&kprobe_exceptions_nb); | |
e8386a0c MH |
2136 | if (!err) |
2137 | err = register_module_notifier(&kprobe_module_nb); | |
2138 | ||
ef53d9c5 | 2139 | kprobes_initialized = (err == 0); |
802eae7c | 2140 | |
8c1c9356 AM |
2141 | if (!err) |
2142 | init_test_probes(); | |
1da177e4 LT |
2143 | return err; |
2144 | } | |
2145 | ||
346fd59b | 2146 | #ifdef CONFIG_DEBUG_FS |
55479f64 | 2147 | static void report_probe(struct seq_file *pi, struct kprobe *p, |
afd66255 | 2148 | const char *sym, int offset, char *modname, struct kprobe *pp) |
346fd59b SD |
2149 | { |
2150 | char *kprobe_type; | |
2151 | ||
2152 | if (p->pre_handler == pre_handler_kretprobe) | |
2153 | kprobe_type = "r"; | |
2154 | else if (p->pre_handler == setjmp_pre_handler) | |
2155 | kprobe_type = "j"; | |
2156 | else | |
2157 | kprobe_type = "k"; | |
afd66255 | 2158 | |
346fd59b | 2159 | if (sym) |
afd66255 | 2160 | seq_printf(pi, "%p %s %s+0x%x %s ", |
de5bd88d | 2161 | p->addr, kprobe_type, sym, offset, |
afd66255 | 2162 | (modname ? modname : " ")); |
346fd59b | 2163 | else |
afd66255 MH |
2164 | seq_printf(pi, "%p %s %p ", |
2165 | p->addr, kprobe_type, p->addr); | |
2166 | ||
2167 | if (!pp) | |
2168 | pp = p; | |
ae6aa16f | 2169 | seq_printf(pi, "%s%s%s%s\n", |
afd66255 MH |
2170 | (kprobe_gone(p) ? "[GONE]" : ""), |
2171 | ((kprobe_disabled(p) && !kprobe_gone(p)) ? "[DISABLED]" : ""), | |
ae6aa16f MH |
2172 | (kprobe_optimized(pp) ? "[OPTIMIZED]" : ""), |
2173 | (kprobe_ftrace(pp) ? "[FTRACE]" : "")); | |
346fd59b SD |
2174 | } |
2175 | ||
55479f64 | 2176 | static void *kprobe_seq_start(struct seq_file *f, loff_t *pos) |
346fd59b SD |
2177 | { |
2178 | return (*pos < KPROBE_TABLE_SIZE) ? pos : NULL; | |
2179 | } | |
2180 | ||
55479f64 | 2181 | static void *kprobe_seq_next(struct seq_file *f, void *v, loff_t *pos) |
346fd59b SD |
2182 | { |
2183 | (*pos)++; | |
2184 | if (*pos >= KPROBE_TABLE_SIZE) | |
2185 | return NULL; | |
2186 | return pos; | |
2187 | } | |
2188 | ||
55479f64 | 2189 | static void kprobe_seq_stop(struct seq_file *f, void *v) |
346fd59b SD |
2190 | { |
2191 | /* Nothing to do */ | |
2192 | } | |
2193 | ||
55479f64 | 2194 | static int show_kprobe_addr(struct seq_file *pi, void *v) |
346fd59b SD |
2195 | { |
2196 | struct hlist_head *head; | |
346fd59b SD |
2197 | struct kprobe *p, *kp; |
2198 | const char *sym = NULL; | |
2199 | unsigned int i = *(loff_t *) v; | |
ffb45122 | 2200 | unsigned long offset = 0; |
ab767865 | 2201 | char *modname, namebuf[KSYM_NAME_LEN]; |
346fd59b SD |
2202 | |
2203 | head = &kprobe_table[i]; | |
2204 | preempt_disable(); | |
b67bfe0d | 2205 | hlist_for_each_entry_rcu(p, head, hlist) { |
ffb45122 | 2206 | sym = kallsyms_lookup((unsigned long)p->addr, NULL, |
346fd59b | 2207 | &offset, &modname, namebuf); |
afd66255 | 2208 | if (kprobe_aggrprobe(p)) { |
346fd59b | 2209 | list_for_each_entry_rcu(kp, &p->list, list) |
afd66255 | 2210 | report_probe(pi, kp, sym, offset, modname, p); |
346fd59b | 2211 | } else |
afd66255 | 2212 | report_probe(pi, p, sym, offset, modname, NULL); |
346fd59b SD |
2213 | } |
2214 | preempt_enable(); | |
2215 | return 0; | |
2216 | } | |
2217 | ||
88e9d34c | 2218 | static const struct seq_operations kprobes_seq_ops = { |
346fd59b SD |
2219 | .start = kprobe_seq_start, |
2220 | .next = kprobe_seq_next, | |
2221 | .stop = kprobe_seq_stop, | |
2222 | .show = show_kprobe_addr | |
2223 | }; | |
2224 | ||
55479f64 | 2225 | static int kprobes_open(struct inode *inode, struct file *filp) |
346fd59b SD |
2226 | { |
2227 | return seq_open(filp, &kprobes_seq_ops); | |
2228 | } | |
2229 | ||
828c0950 | 2230 | static const struct file_operations debugfs_kprobes_operations = { |
346fd59b SD |
2231 | .open = kprobes_open, |
2232 | .read = seq_read, | |
2233 | .llseek = seq_lseek, | |
2234 | .release = seq_release, | |
2235 | }; | |
2236 | ||
55479f64 | 2237 | static void arm_all_kprobes(void) |
bf8f6e5b AM |
2238 | { |
2239 | struct hlist_head *head; | |
bf8f6e5b AM |
2240 | struct kprobe *p; |
2241 | unsigned int i; | |
2242 | ||
2243 | mutex_lock(&kprobe_mutex); | |
2244 | ||
e579abeb MH |
2245 | /* If kprobes are armed, just return */ |
2246 | if (!kprobes_all_disarmed) | |
bf8f6e5b AM |
2247 | goto already_enabled; |
2248 | ||
afd66255 | 2249 | /* Arming kprobes doesn't optimize kprobe itself */ |
bf8f6e5b AM |
2250 | for (i = 0; i < KPROBE_TABLE_SIZE; i++) { |
2251 | head = &kprobe_table[i]; | |
b67bfe0d | 2252 | hlist_for_each_entry_rcu(p, head, hlist) |
de5bd88d | 2253 | if (!kprobe_disabled(p)) |
ae6aa16f | 2254 | arm_kprobe(p); |
bf8f6e5b AM |
2255 | } |
2256 | ||
e579abeb | 2257 | kprobes_all_disarmed = false; |
bf8f6e5b AM |
2258 | printk(KERN_INFO "Kprobes globally enabled\n"); |
2259 | ||
2260 | already_enabled: | |
2261 | mutex_unlock(&kprobe_mutex); | |
2262 | return; | |
2263 | } | |
2264 | ||
55479f64 | 2265 | static void disarm_all_kprobes(void) |
bf8f6e5b AM |
2266 | { |
2267 | struct hlist_head *head; | |
bf8f6e5b AM |
2268 | struct kprobe *p; |
2269 | unsigned int i; | |
2270 | ||
2271 | mutex_lock(&kprobe_mutex); | |
2272 | ||
e579abeb | 2273 | /* If kprobes are already disarmed, just return */ |
6274de49 MH |
2274 | if (kprobes_all_disarmed) { |
2275 | mutex_unlock(&kprobe_mutex); | |
2276 | return; | |
2277 | } | |
bf8f6e5b | 2278 | |
e579abeb | 2279 | kprobes_all_disarmed = true; |
bf8f6e5b | 2280 | printk(KERN_INFO "Kprobes globally disabled\n"); |
afd66255 | 2281 | |
bf8f6e5b AM |
2282 | for (i = 0; i < KPROBE_TABLE_SIZE; i++) { |
2283 | head = &kprobe_table[i]; | |
b67bfe0d | 2284 | hlist_for_each_entry_rcu(p, head, hlist) { |
de5bd88d | 2285 | if (!arch_trampoline_kprobe(p) && !kprobe_disabled(p)) |
ae6aa16f | 2286 | disarm_kprobe(p, false); |
bf8f6e5b AM |
2287 | } |
2288 | } | |
bf8f6e5b | 2289 | mutex_unlock(&kprobe_mutex); |
bf8f6e5b | 2290 | |
6274de49 MH |
2291 | /* Wait for disarming all kprobes by optimizer */ |
2292 | wait_for_kprobe_optimizer(); | |
bf8f6e5b AM |
2293 | } |
2294 | ||
2295 | /* | |
2296 | * XXX: The debugfs bool file interface doesn't allow for callbacks | |
2297 | * when the bool state is switched. We can reuse that facility when | |
2298 | * available | |
2299 | */ | |
2300 | static ssize_t read_enabled_file_bool(struct file *file, | |
2301 | char __user *user_buf, size_t count, loff_t *ppos) | |
2302 | { | |
2303 | char buf[3]; | |
2304 | ||
e579abeb | 2305 | if (!kprobes_all_disarmed) |
bf8f6e5b AM |
2306 | buf[0] = '1'; |
2307 | else | |
2308 | buf[0] = '0'; | |
2309 | buf[1] = '\n'; | |
2310 | buf[2] = 0x00; | |
2311 | return simple_read_from_buffer(user_buf, count, ppos, buf, 2); | |
2312 | } | |
2313 | ||
2314 | static ssize_t write_enabled_file_bool(struct file *file, | |
2315 | const char __user *user_buf, size_t count, loff_t *ppos) | |
2316 | { | |
2317 | char buf[32]; | |
efeb156e | 2318 | size_t buf_size; |
bf8f6e5b AM |
2319 | |
2320 | buf_size = min(count, (sizeof(buf)-1)); | |
2321 | if (copy_from_user(buf, user_buf, buf_size)) | |
2322 | return -EFAULT; | |
2323 | ||
10fb46d5 | 2324 | buf[buf_size] = '\0'; |
bf8f6e5b AM |
2325 | switch (buf[0]) { |
2326 | case 'y': | |
2327 | case 'Y': | |
2328 | case '1': | |
e579abeb | 2329 | arm_all_kprobes(); |
bf8f6e5b AM |
2330 | break; |
2331 | case 'n': | |
2332 | case 'N': | |
2333 | case '0': | |
e579abeb | 2334 | disarm_all_kprobes(); |
bf8f6e5b | 2335 | break; |
10fb46d5 MK |
2336 | default: |
2337 | return -EINVAL; | |
bf8f6e5b AM |
2338 | } |
2339 | ||
2340 | return count; | |
2341 | } | |
2342 | ||
828c0950 | 2343 | static const struct file_operations fops_kp = { |
bf8f6e5b AM |
2344 | .read = read_enabled_file_bool, |
2345 | .write = write_enabled_file_bool, | |
6038f373 | 2346 | .llseek = default_llseek, |
bf8f6e5b AM |
2347 | }; |
2348 | ||
55479f64 | 2349 | static int __init debugfs_kprobe_init(void) |
346fd59b SD |
2350 | { |
2351 | struct dentry *dir, *file; | |
bf8f6e5b | 2352 | unsigned int value = 1; |
346fd59b SD |
2353 | |
2354 | dir = debugfs_create_dir("kprobes", NULL); | |
2355 | if (!dir) | |
2356 | return -ENOMEM; | |
2357 | ||
e3869792 | 2358 | file = debugfs_create_file("list", 0444, dir, NULL, |
346fd59b SD |
2359 | &debugfs_kprobes_operations); |
2360 | if (!file) { | |
2361 | debugfs_remove(dir); | |
2362 | return -ENOMEM; | |
2363 | } | |
2364 | ||
bf8f6e5b AM |
2365 | file = debugfs_create_file("enabled", 0600, dir, |
2366 | &value, &fops_kp); | |
2367 | if (!file) { | |
2368 | debugfs_remove(dir); | |
2369 | return -ENOMEM; | |
2370 | } | |
2371 | ||
346fd59b SD |
2372 | return 0; |
2373 | } | |
2374 | ||
2375 | late_initcall(debugfs_kprobe_init); | |
2376 | #endif /* CONFIG_DEBUG_FS */ | |
2377 | ||
2378 | module_init(init_kprobes); | |
1da177e4 | 2379 | |
99081ab5 | 2380 | /* defined in arch/.../kernel/kprobes.c */ |
1da177e4 | 2381 | EXPORT_SYMBOL_GPL(jprobe_return); |