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> |
1da177e4 | 39 | #include <linux/module.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 { |
7e036d04 | 81 | spinlock_t lock ____cacheline_aligned_in_smp; |
ef53d9c5 S |
82 | } kretprobe_table_locks[KPROBE_TABLE_SIZE]; |
83 | ||
84 | static spinlock_t *kretprobe_table_lock_ptr(unsigned long hash) | |
85 | { | |
86 | return &(kretprobe_table_locks[hash].lock); | |
87 | } | |
1da177e4 | 88 | |
3d8d996e SD |
89 | /* |
90 | * Normally, functions that we'd want to prohibit kprobes in, are marked | |
91 | * __kprobes. But, there are cases where such functions already belong to | |
92 | * a different section (__sched for preempt_schedule) | |
93 | * | |
94 | * For such cases, we now have a blacklist | |
95 | */ | |
544304b2 | 96 | static struct kprobe_blackpoint kprobe_blacklist[] = { |
3d8d996e | 97 | {"preempt_schedule",}, |
65e234ec | 98 | {"native_get_debugreg",}, |
a00e817f MH |
99 | {"irq_entries_start",}, |
100 | {"common_interrupt",}, | |
5ecaafdb | 101 | {"mcount",}, /* mcount can be called from everywhere */ |
3d8d996e SD |
102 | {NULL} /* Terminator */ |
103 | }; | |
104 | ||
2d14e39d | 105 | #ifdef __ARCH_WANT_KPROBES_INSN_SLOT |
9ec4b1f3 AM |
106 | /* |
107 | * kprobe->ainsn.insn points to the copy of the instruction to be | |
108 | * single-stepped. x86_64, POWER4 and above have no-exec support and | |
109 | * stepping on the instruction on a vmalloced/kmalloced/data page | |
110 | * is a recipe for disaster | |
111 | */ | |
9ec4b1f3 | 112 | struct kprobe_insn_page { |
c5cb5a2d | 113 | struct list_head list; |
9ec4b1f3 | 114 | kprobe_opcode_t *insns; /* Page of instruction slots */ |
9ec4b1f3 | 115 | int nused; |
b4c6c34a | 116 | int ngarbage; |
4610ee1d | 117 | char slot_used[]; |
9ec4b1f3 AM |
118 | }; |
119 | ||
4610ee1d MH |
120 | #define KPROBE_INSN_PAGE_SIZE(slots) \ |
121 | (offsetof(struct kprobe_insn_page, slot_used) + \ | |
122 | (sizeof(char) * (slots))) | |
123 | ||
124 | struct kprobe_insn_cache { | |
125 | struct list_head pages; /* list of kprobe_insn_page */ | |
126 | size_t insn_size; /* size of instruction slot */ | |
127 | int nr_garbage; | |
128 | }; | |
129 | ||
130 | static int slots_per_page(struct kprobe_insn_cache *c) | |
131 | { | |
132 | return PAGE_SIZE/(c->insn_size * sizeof(kprobe_opcode_t)); | |
133 | } | |
134 | ||
ab40c5c6 MH |
135 | enum kprobe_slot_state { |
136 | SLOT_CLEAN = 0, | |
137 | SLOT_DIRTY = 1, | |
138 | SLOT_USED = 2, | |
139 | }; | |
140 | ||
4610ee1d MH |
141 | static DEFINE_MUTEX(kprobe_insn_mutex); /* Protects kprobe_insn_slots */ |
142 | static struct kprobe_insn_cache kprobe_insn_slots = { | |
143 | .pages = LIST_HEAD_INIT(kprobe_insn_slots.pages), | |
144 | .insn_size = MAX_INSN_SIZE, | |
145 | .nr_garbage = 0, | |
146 | }; | |
147 | static int __kprobes collect_garbage_slots(struct kprobe_insn_cache *c); | |
b4c6c34a | 148 | |
9ec4b1f3 | 149 | /** |
12941560 | 150 | * __get_insn_slot() - Find a slot on an executable page for an instruction. |
9ec4b1f3 AM |
151 | * We allocate an executable page if there's no room on existing ones. |
152 | */ | |
4610ee1d | 153 | static kprobe_opcode_t __kprobes *__get_insn_slot(struct kprobe_insn_cache *c) |
9ec4b1f3 AM |
154 | { |
155 | struct kprobe_insn_page *kip; | |
9ec4b1f3 | 156 | |
6f716acd | 157 | retry: |
4610ee1d MH |
158 | list_for_each_entry(kip, &c->pages, list) { |
159 | if (kip->nused < slots_per_page(c)) { | |
9ec4b1f3 | 160 | int i; |
4610ee1d | 161 | for (i = 0; i < slots_per_page(c); i++) { |
ab40c5c6 MH |
162 | if (kip->slot_used[i] == SLOT_CLEAN) { |
163 | kip->slot_used[i] = SLOT_USED; | |
9ec4b1f3 | 164 | kip->nused++; |
4610ee1d | 165 | return kip->insns + (i * c->insn_size); |
9ec4b1f3 AM |
166 | } |
167 | } | |
4610ee1d MH |
168 | /* kip->nused is broken. Fix it. */ |
169 | kip->nused = slots_per_page(c); | |
170 | WARN_ON(1); | |
9ec4b1f3 AM |
171 | } |
172 | } | |
173 | ||
b4c6c34a | 174 | /* If there are any garbage slots, collect it and try again. */ |
4610ee1d | 175 | if (c->nr_garbage && collect_garbage_slots(c) == 0) |
b4c6c34a | 176 | goto retry; |
4610ee1d MH |
177 | |
178 | /* All out of space. Need to allocate a new page. */ | |
179 | kip = kmalloc(KPROBE_INSN_PAGE_SIZE(slots_per_page(c)), GFP_KERNEL); | |
6f716acd | 180 | if (!kip) |
9ec4b1f3 | 181 | return NULL; |
9ec4b1f3 AM |
182 | |
183 | /* | |
184 | * Use module_alloc so this page is within +/- 2GB of where the | |
185 | * kernel image and loaded module images reside. This is required | |
186 | * so x86_64 can correctly handle the %rip-relative fixups. | |
187 | */ | |
188 | kip->insns = module_alloc(PAGE_SIZE); | |
189 | if (!kip->insns) { | |
190 | kfree(kip); | |
191 | return NULL; | |
192 | } | |
c5cb5a2d | 193 | INIT_LIST_HEAD(&kip->list); |
4610ee1d | 194 | memset(kip->slot_used, SLOT_CLEAN, slots_per_page(c)); |
ab40c5c6 | 195 | kip->slot_used[0] = SLOT_USED; |
9ec4b1f3 | 196 | kip->nused = 1; |
b4c6c34a | 197 | kip->ngarbage = 0; |
4610ee1d | 198 | list_add(&kip->list, &c->pages); |
9ec4b1f3 AM |
199 | return kip->insns; |
200 | } | |
201 | ||
4610ee1d | 202 | |
12941560 MH |
203 | kprobe_opcode_t __kprobes *get_insn_slot(void) |
204 | { | |
4610ee1d MH |
205 | kprobe_opcode_t *ret = NULL; |
206 | ||
12941560 | 207 | mutex_lock(&kprobe_insn_mutex); |
4610ee1d | 208 | ret = __get_insn_slot(&kprobe_insn_slots); |
12941560 | 209 | mutex_unlock(&kprobe_insn_mutex); |
4610ee1d | 210 | |
12941560 MH |
211 | return ret; |
212 | } | |
213 | ||
b4c6c34a MH |
214 | /* Return 1 if all garbages are collected, otherwise 0. */ |
215 | static int __kprobes collect_one_slot(struct kprobe_insn_page *kip, int idx) | |
216 | { | |
ab40c5c6 | 217 | kip->slot_used[idx] = SLOT_CLEAN; |
b4c6c34a MH |
218 | kip->nused--; |
219 | if (kip->nused == 0) { | |
220 | /* | |
221 | * Page is no longer in use. Free it unless | |
222 | * it's the last one. We keep the last one | |
223 | * so as not to have to set it up again the | |
224 | * next time somebody inserts a probe. | |
225 | */ | |
4610ee1d | 226 | if (!list_is_singular(&kip->list)) { |
c5cb5a2d | 227 | list_del(&kip->list); |
b4c6c34a MH |
228 | module_free(NULL, kip->insns); |
229 | kfree(kip); | |
230 | } | |
231 | return 1; | |
232 | } | |
233 | return 0; | |
234 | } | |
235 | ||
4610ee1d | 236 | static int __kprobes collect_garbage_slots(struct kprobe_insn_cache *c) |
b4c6c34a | 237 | { |
c5cb5a2d | 238 | struct kprobe_insn_page *kip, *next; |
b4c6c34a | 239 | |
615d0ebb MH |
240 | /* Ensure no-one is interrupted on the garbages */ |
241 | synchronize_sched(); | |
b4c6c34a | 242 | |
4610ee1d | 243 | list_for_each_entry_safe(kip, next, &c->pages, list) { |
b4c6c34a | 244 | int i; |
b4c6c34a MH |
245 | if (kip->ngarbage == 0) |
246 | continue; | |
247 | kip->ngarbage = 0; /* we will collect all garbages */ | |
4610ee1d | 248 | for (i = 0; i < slots_per_page(c); i++) { |
ab40c5c6 | 249 | if (kip->slot_used[i] == SLOT_DIRTY && |
b4c6c34a MH |
250 | collect_one_slot(kip, i)) |
251 | break; | |
252 | } | |
253 | } | |
4610ee1d | 254 | c->nr_garbage = 0; |
b4c6c34a MH |
255 | return 0; |
256 | } | |
257 | ||
4610ee1d MH |
258 | static void __kprobes __free_insn_slot(struct kprobe_insn_cache *c, |
259 | kprobe_opcode_t *slot, int dirty) | |
9ec4b1f3 AM |
260 | { |
261 | struct kprobe_insn_page *kip; | |
9ec4b1f3 | 262 | |
4610ee1d | 263 | list_for_each_entry(kip, &c->pages, list) { |
83ff56f4 MH |
264 | long idx = ((long)slot - (long)kip->insns) / |
265 | (c->insn_size * sizeof(kprobe_opcode_t)); | |
4610ee1d MH |
266 | if (idx >= 0 && idx < slots_per_page(c)) { |
267 | WARN_ON(kip->slot_used[idx] != SLOT_USED); | |
b4c6c34a | 268 | if (dirty) { |
4610ee1d | 269 | kip->slot_used[idx] = SLOT_DIRTY; |
b4c6c34a | 270 | kip->ngarbage++; |
4610ee1d MH |
271 | if (++c->nr_garbage > slots_per_page(c)) |
272 | collect_garbage_slots(c); | |
c5cb5a2d | 273 | } else |
4610ee1d MH |
274 | collect_one_slot(kip, idx); |
275 | return; | |
9ec4b1f3 AM |
276 | } |
277 | } | |
4610ee1d MH |
278 | /* Could not free this slot. */ |
279 | WARN_ON(1); | |
280 | } | |
6f716acd | 281 | |
4610ee1d MH |
282 | void __kprobes free_insn_slot(kprobe_opcode_t * slot, int dirty) |
283 | { | |
284 | mutex_lock(&kprobe_insn_mutex); | |
285 | __free_insn_slot(&kprobe_insn_slots, slot, dirty); | |
12941560 | 286 | mutex_unlock(&kprobe_insn_mutex); |
9ec4b1f3 | 287 | } |
afd66255 MH |
288 | #ifdef CONFIG_OPTPROBES |
289 | /* For optimized_kprobe buffer */ | |
290 | static DEFINE_MUTEX(kprobe_optinsn_mutex); /* Protects kprobe_optinsn_slots */ | |
291 | static struct kprobe_insn_cache kprobe_optinsn_slots = { | |
292 | .pages = LIST_HEAD_INIT(kprobe_optinsn_slots.pages), | |
293 | /* .insn_size is initialized later */ | |
294 | .nr_garbage = 0, | |
295 | }; | |
296 | /* Get a slot for optimized_kprobe buffer */ | |
297 | kprobe_opcode_t __kprobes *get_optinsn_slot(void) | |
298 | { | |
299 | kprobe_opcode_t *ret = NULL; | |
300 | ||
301 | mutex_lock(&kprobe_optinsn_mutex); | |
302 | ret = __get_insn_slot(&kprobe_optinsn_slots); | |
303 | mutex_unlock(&kprobe_optinsn_mutex); | |
304 | ||
305 | return ret; | |
306 | } | |
307 | ||
308 | void __kprobes free_optinsn_slot(kprobe_opcode_t * slot, int dirty) | |
309 | { | |
310 | mutex_lock(&kprobe_optinsn_mutex); | |
311 | __free_insn_slot(&kprobe_optinsn_slots, slot, dirty); | |
312 | mutex_unlock(&kprobe_optinsn_mutex); | |
313 | } | |
314 | #endif | |
2d14e39d | 315 | #endif |
9ec4b1f3 | 316 | |
e6584523 AM |
317 | /* We have preemption disabled.. so it is safe to use __ versions */ |
318 | static inline void set_kprobe_instance(struct kprobe *kp) | |
319 | { | |
320 | __get_cpu_var(kprobe_instance) = kp; | |
321 | } | |
322 | ||
323 | static inline void reset_kprobe_instance(void) | |
324 | { | |
325 | __get_cpu_var(kprobe_instance) = NULL; | |
326 | } | |
327 | ||
3516a460 AM |
328 | /* |
329 | * This routine is called either: | |
49a2a1b8 | 330 | * - under the kprobe_mutex - during kprobe_[un]register() |
3516a460 | 331 | * OR |
d217d545 | 332 | * - with preemption disabled - from arch/xxx/kernel/kprobes.c |
3516a460 | 333 | */ |
d0aaff97 | 334 | struct kprobe __kprobes *get_kprobe(void *addr) |
1da177e4 LT |
335 | { |
336 | struct hlist_head *head; | |
337 | struct hlist_node *node; | |
3516a460 | 338 | struct kprobe *p; |
1da177e4 LT |
339 | |
340 | head = &kprobe_table[hash_ptr(addr, KPROBE_HASH_BITS)]; | |
3516a460 | 341 | hlist_for_each_entry_rcu(p, node, head, hlist) { |
1da177e4 LT |
342 | if (p->addr == addr) |
343 | return p; | |
344 | } | |
afd66255 | 345 | |
1da177e4 LT |
346 | return NULL; |
347 | } | |
348 | ||
afd66255 MH |
349 | static int __kprobes aggr_pre_handler(struct kprobe *p, struct pt_regs *regs); |
350 | ||
351 | /* Return true if the kprobe is an aggregator */ | |
352 | static inline int kprobe_aggrprobe(struct kprobe *p) | |
353 | { | |
354 | return p->pre_handler == aggr_pre_handler; | |
355 | } | |
356 | ||
6274de49 MH |
357 | /* Return true(!0) if the kprobe is unused */ |
358 | static inline int kprobe_unused(struct kprobe *p) | |
359 | { | |
360 | return kprobe_aggrprobe(p) && kprobe_disabled(p) && | |
361 | list_empty(&p->list); | |
362 | } | |
363 | ||
afd66255 MH |
364 | /* |
365 | * Keep all fields in the kprobe consistent | |
366 | */ | |
6d8e40a8 | 367 | static inline void copy_kprobe(struct kprobe *ap, struct kprobe *p) |
afd66255 | 368 | { |
6d8e40a8 MH |
369 | memcpy(&p->opcode, &ap->opcode, sizeof(kprobe_opcode_t)); |
370 | memcpy(&p->ainsn, &ap->ainsn, sizeof(struct arch_specific_insn)); | |
afd66255 MH |
371 | } |
372 | ||
373 | #ifdef CONFIG_OPTPROBES | |
b2be84df MH |
374 | /* NOTE: change this value only with kprobe_mutex held */ |
375 | static bool kprobes_allow_optimization; | |
376 | ||
afd66255 MH |
377 | /* |
378 | * Call all pre_handler on the list, but ignores its return value. | |
379 | * This must be called from arch-dep optimized caller. | |
380 | */ | |
381 | void __kprobes opt_pre_handler(struct kprobe *p, struct pt_regs *regs) | |
382 | { | |
383 | struct kprobe *kp; | |
384 | ||
385 | list_for_each_entry_rcu(kp, &p->list, list) { | |
386 | if (kp->pre_handler && likely(!kprobe_disabled(kp))) { | |
387 | set_kprobe_instance(kp); | |
388 | kp->pre_handler(kp, regs); | |
389 | } | |
390 | reset_kprobe_instance(); | |
391 | } | |
392 | } | |
393 | ||
6274de49 MH |
394 | /* Free optimized instructions and optimized_kprobe */ |
395 | static __kprobes void free_aggr_kprobe(struct kprobe *p) | |
396 | { | |
397 | struct optimized_kprobe *op; | |
398 | ||
399 | op = container_of(p, struct optimized_kprobe, kp); | |
400 | arch_remove_optimized_kprobe(op); | |
401 | arch_remove_kprobe(p); | |
402 | kfree(op); | |
403 | } | |
404 | ||
afd66255 MH |
405 | /* Return true(!0) if the kprobe is ready for optimization. */ |
406 | static inline int kprobe_optready(struct kprobe *p) | |
407 | { | |
408 | struct optimized_kprobe *op; | |
409 | ||
410 | if (kprobe_aggrprobe(p)) { | |
411 | op = container_of(p, struct optimized_kprobe, kp); | |
412 | return arch_prepared_optinsn(&op->optinsn); | |
413 | } | |
414 | ||
415 | return 0; | |
416 | } | |
417 | ||
6274de49 MH |
418 | /* Return true(!0) if the kprobe is disarmed. Note: p must be on hash list */ |
419 | static inline int kprobe_disarmed(struct kprobe *p) | |
420 | { | |
421 | struct optimized_kprobe *op; | |
422 | ||
423 | /* If kprobe is not aggr/opt probe, just return kprobe is disabled */ | |
424 | if (!kprobe_aggrprobe(p)) | |
425 | return kprobe_disabled(p); | |
426 | ||
427 | op = container_of(p, struct optimized_kprobe, kp); | |
428 | ||
429 | return kprobe_disabled(p) && list_empty(&op->list); | |
430 | } | |
431 | ||
432 | /* Return true(!0) if the probe is queued on (un)optimizing lists */ | |
433 | static int __kprobes kprobe_queued(struct kprobe *p) | |
434 | { | |
435 | struct optimized_kprobe *op; | |
436 | ||
437 | if (kprobe_aggrprobe(p)) { | |
438 | op = container_of(p, struct optimized_kprobe, kp); | |
439 | if (!list_empty(&op->list)) | |
440 | return 1; | |
441 | } | |
442 | return 0; | |
443 | } | |
444 | ||
afd66255 MH |
445 | /* |
446 | * Return an optimized kprobe whose optimizing code replaces | |
447 | * instructions including addr (exclude breakpoint). | |
448 | */ | |
6376b229 | 449 | static struct kprobe *__kprobes get_optimized_kprobe(unsigned long addr) |
afd66255 MH |
450 | { |
451 | int i; | |
452 | struct kprobe *p = NULL; | |
453 | struct optimized_kprobe *op; | |
454 | ||
455 | /* Don't check i == 0, since that is a breakpoint case. */ | |
456 | for (i = 1; !p && i < MAX_OPTIMIZED_LENGTH; i++) | |
457 | p = get_kprobe((void *)(addr - i)); | |
458 | ||
459 | if (p && kprobe_optready(p)) { | |
460 | op = container_of(p, struct optimized_kprobe, kp); | |
461 | if (arch_within_optimized_kprobe(op, addr)) | |
462 | return p; | |
463 | } | |
464 | ||
465 | return NULL; | |
466 | } | |
467 | ||
468 | /* Optimization staging list, protected by kprobe_mutex */ | |
469 | static LIST_HEAD(optimizing_list); | |
6274de49 | 470 | static LIST_HEAD(unoptimizing_list); |
afd66255 MH |
471 | |
472 | static void kprobe_optimizer(struct work_struct *work); | |
473 | static DECLARE_DELAYED_WORK(optimizing_work, kprobe_optimizer); | |
6274de49 | 474 | static DECLARE_COMPLETION(optimizer_comp); |
afd66255 MH |
475 | #define OPTIMIZE_DELAY 5 |
476 | ||
61f4e13f MH |
477 | /* |
478 | * Optimize (replace a breakpoint with a jump) kprobes listed on | |
479 | * optimizing_list. | |
480 | */ | |
481 | static __kprobes void do_optimize_kprobes(void) | |
afd66255 | 482 | { |
6274de49 MH |
483 | /* Optimization never be done when disarmed */ |
484 | if (kprobes_all_disarmed || !kprobes_allow_optimization || | |
485 | list_empty(&optimizing_list)) | |
486 | return; | |
487 | ||
afd66255 MH |
488 | /* |
489 | * The optimization/unoptimization refers online_cpus via | |
490 | * stop_machine() and cpu-hotplug modifies online_cpus. | |
491 | * And same time, text_mutex will be held in cpu-hotplug and here. | |
492 | * This combination can cause a deadlock (cpu-hotplug try to lock | |
493 | * text_mutex but stop_machine can not be done because online_cpus | |
494 | * has been changed) | |
495 | * To avoid this deadlock, we need to call get_online_cpus() | |
496 | * for preventing cpu-hotplug outside of text_mutex locking. | |
497 | */ | |
498 | get_online_cpus(); | |
499 | mutex_lock(&text_mutex); | |
cd7ebe22 | 500 | arch_optimize_kprobes(&optimizing_list); |
afd66255 MH |
501 | mutex_unlock(&text_mutex); |
502 | put_online_cpus(); | |
61f4e13f MH |
503 | } |
504 | ||
6274de49 MH |
505 | /* |
506 | * Unoptimize (replace a jump with a breakpoint and remove the breakpoint | |
507 | * if need) kprobes listed on unoptimizing_list. | |
508 | */ | |
509 | static __kprobes void do_unoptimize_kprobes(struct list_head *free_list) | |
510 | { | |
511 | struct optimized_kprobe *op, *tmp; | |
512 | ||
513 | /* Unoptimization must be done anytime */ | |
514 | if (list_empty(&unoptimizing_list)) | |
515 | return; | |
516 | ||
517 | /* Ditto to do_optimize_kprobes */ | |
518 | get_online_cpus(); | |
519 | mutex_lock(&text_mutex); | |
520 | list_for_each_entry_safe(op, tmp, &unoptimizing_list, list) { | |
521 | /* Unoptimize kprobes */ | |
522 | arch_unoptimize_kprobe(op); | |
523 | /* Disarm probes if marked disabled */ | |
524 | if (kprobe_disabled(&op->kp)) | |
525 | arch_disarm_kprobe(&op->kp); | |
526 | if (kprobe_unused(&op->kp)) { | |
527 | /* | |
528 | * Remove unused probes from hash list. After waiting | |
529 | * for synchronization, these probes are reclaimed. | |
530 | * (reclaiming is done by do_free_cleaned_kprobes.) | |
531 | */ | |
532 | hlist_del_rcu(&op->kp.hlist); | |
533 | /* Move only unused probes on free_list */ | |
534 | list_move(&op->list, free_list); | |
535 | } else | |
536 | list_del_init(&op->list); | |
537 | } | |
538 | mutex_unlock(&text_mutex); | |
539 | put_online_cpus(); | |
540 | } | |
541 | ||
542 | /* Reclaim all kprobes on the free_list */ | |
543 | static __kprobes void do_free_cleaned_kprobes(struct list_head *free_list) | |
544 | { | |
545 | struct optimized_kprobe *op, *tmp; | |
546 | ||
547 | list_for_each_entry_safe(op, tmp, free_list, list) { | |
548 | BUG_ON(!kprobe_unused(&op->kp)); | |
549 | list_del_init(&op->list); | |
550 | free_aggr_kprobe(&op->kp); | |
551 | } | |
552 | } | |
553 | ||
554 | /* Start optimizer after OPTIMIZE_DELAY passed */ | |
555 | static __kprobes void kick_kprobe_optimizer(void) | |
556 | { | |
557 | if (!delayed_work_pending(&optimizing_work)) | |
558 | schedule_delayed_work(&optimizing_work, OPTIMIZE_DELAY); | |
559 | } | |
560 | ||
61f4e13f MH |
561 | /* Kprobe jump optimizer */ |
562 | static __kprobes void kprobe_optimizer(struct work_struct *work) | |
563 | { | |
6274de49 MH |
564 | LIST_HEAD(free_list); |
565 | ||
61f4e13f MH |
566 | /* Lock modules while optimizing kprobes */ |
567 | mutex_lock(&module_mutex); | |
568 | mutex_lock(&kprobe_mutex); | |
61f4e13f MH |
569 | |
570 | /* | |
6274de49 MH |
571 | * Step 1: Unoptimize kprobes and collect cleaned (unused and disarmed) |
572 | * kprobes before waiting for quiesence period. | |
573 | */ | |
574 | do_unoptimize_kprobes(&free_list); | |
575 | ||
576 | /* | |
577 | * Step 2: Wait for quiesence period to ensure all running interrupts | |
61f4e13f MH |
578 | * are done. Because optprobe may modify multiple instructions |
579 | * there is a chance that Nth instruction is interrupted. In that | |
580 | * case, running interrupt can return to 2nd-Nth byte of jump | |
581 | * instruction. This wait is for avoiding it. | |
582 | */ | |
583 | synchronize_sched(); | |
584 | ||
6274de49 | 585 | /* Step 3: Optimize kprobes after quiesence period */ |
61f4e13f | 586 | do_optimize_kprobes(); |
6274de49 MH |
587 | |
588 | /* Step 4: Free cleaned kprobes after quiesence period */ | |
589 | do_free_cleaned_kprobes(&free_list); | |
590 | ||
afd66255 MH |
591 | mutex_unlock(&kprobe_mutex); |
592 | mutex_unlock(&module_mutex); | |
6274de49 | 593 | |
cd7ebe22 MH |
594 | /* Step 5: Kick optimizer again if needed */ |
595 | if (!list_empty(&optimizing_list)) | |
596 | kick_kprobe_optimizer(); | |
597 | else | |
598 | /* Wake up all waiters */ | |
599 | complete_all(&optimizer_comp); | |
6274de49 MH |
600 | } |
601 | ||
602 | /* Wait for completing optimization and unoptimization */ | |
603 | static __kprobes void wait_for_kprobe_optimizer(void) | |
604 | { | |
605 | if (delayed_work_pending(&optimizing_work)) | |
606 | wait_for_completion(&optimizer_comp); | |
afd66255 MH |
607 | } |
608 | ||
609 | /* Optimize kprobe if p is ready to be optimized */ | |
610 | static __kprobes void optimize_kprobe(struct kprobe *p) | |
611 | { | |
612 | struct optimized_kprobe *op; | |
613 | ||
614 | /* Check if the kprobe is disabled or not ready for optimization. */ | |
b2be84df | 615 | if (!kprobe_optready(p) || !kprobes_allow_optimization || |
afd66255 MH |
616 | (kprobe_disabled(p) || kprobes_all_disarmed)) |
617 | return; | |
618 | ||
619 | /* Both of break_handler and post_handler are not supported. */ | |
620 | if (p->break_handler || p->post_handler) | |
621 | return; | |
622 | ||
623 | op = container_of(p, struct optimized_kprobe, kp); | |
624 | ||
625 | /* Check there is no other kprobes at the optimized instructions */ | |
626 | if (arch_check_optimized_kprobe(op) < 0) | |
627 | return; | |
628 | ||
629 | /* Check if it is already optimized. */ | |
630 | if (op->kp.flags & KPROBE_FLAG_OPTIMIZED) | |
631 | return; | |
afd66255 | 632 | op->kp.flags |= KPROBE_FLAG_OPTIMIZED; |
6274de49 MH |
633 | |
634 | if (!list_empty(&op->list)) | |
635 | /* This is under unoptimizing. Just dequeue the probe */ | |
636 | list_del_init(&op->list); | |
637 | else { | |
638 | list_add(&op->list, &optimizing_list); | |
639 | kick_kprobe_optimizer(); | |
640 | } | |
641 | } | |
642 | ||
643 | /* Short cut to direct unoptimizing */ | |
644 | static __kprobes void force_unoptimize_kprobe(struct optimized_kprobe *op) | |
645 | { | |
646 | get_online_cpus(); | |
647 | arch_unoptimize_kprobe(op); | |
648 | put_online_cpus(); | |
649 | if (kprobe_disabled(&op->kp)) | |
650 | arch_disarm_kprobe(&op->kp); | |
afd66255 MH |
651 | } |
652 | ||
653 | /* Unoptimize a kprobe if p is optimized */ | |
6274de49 | 654 | static __kprobes void unoptimize_kprobe(struct kprobe *p, bool force) |
afd66255 MH |
655 | { |
656 | struct optimized_kprobe *op; | |
657 | ||
6274de49 MH |
658 | if (!kprobe_aggrprobe(p) || kprobe_disarmed(p)) |
659 | return; /* This is not an optprobe nor optimized */ | |
660 | ||
661 | op = container_of(p, struct optimized_kprobe, kp); | |
662 | if (!kprobe_optimized(p)) { | |
663 | /* Unoptimized or unoptimizing case */ | |
664 | if (force && !list_empty(&op->list)) { | |
665 | /* | |
666 | * Only if this is unoptimizing kprobe and forced, | |
667 | * forcibly unoptimize it. (No need to unoptimize | |
668 | * unoptimized kprobe again :) | |
669 | */ | |
afd66255 | 670 | list_del_init(&op->list); |
6274de49 MH |
671 | force_unoptimize_kprobe(op); |
672 | } | |
673 | return; | |
674 | } | |
675 | ||
676 | op->kp.flags &= ~KPROBE_FLAG_OPTIMIZED; | |
677 | if (!list_empty(&op->list)) { | |
678 | /* Dequeue from the optimization queue */ | |
679 | list_del_init(&op->list); | |
680 | return; | |
681 | } | |
682 | /* Optimized kprobe case */ | |
683 | if (force) | |
684 | /* Forcibly update the code: this is a special case */ | |
685 | force_unoptimize_kprobe(op); | |
686 | else { | |
687 | list_add(&op->list, &unoptimizing_list); | |
688 | kick_kprobe_optimizer(); | |
afd66255 MH |
689 | } |
690 | } | |
691 | ||
0490cd1f MH |
692 | /* Cancel unoptimizing for reusing */ |
693 | static void reuse_unused_kprobe(struct kprobe *ap) | |
694 | { | |
695 | struct optimized_kprobe *op; | |
696 | ||
697 | BUG_ON(!kprobe_unused(ap)); | |
698 | /* | |
699 | * Unused kprobe MUST be on the way of delayed unoptimizing (means | |
700 | * there is still a relative jump) and disabled. | |
701 | */ | |
702 | op = container_of(ap, struct optimized_kprobe, kp); | |
703 | if (unlikely(list_empty(&op->list))) | |
704 | printk(KERN_WARNING "Warning: found a stray unused " | |
705 | "aggrprobe@%p\n", ap->addr); | |
706 | /* Enable the probe again */ | |
707 | ap->flags &= ~KPROBE_FLAG_DISABLED; | |
708 | /* Optimize it again (remove from op->list) */ | |
709 | BUG_ON(!kprobe_optready(ap)); | |
710 | optimize_kprobe(ap); | |
711 | } | |
712 | ||
afd66255 MH |
713 | /* Remove optimized instructions */ |
714 | static void __kprobes kill_optimized_kprobe(struct kprobe *p) | |
715 | { | |
716 | struct optimized_kprobe *op; | |
717 | ||
718 | op = container_of(p, struct optimized_kprobe, kp); | |
6274de49 MH |
719 | if (!list_empty(&op->list)) |
720 | /* Dequeue from the (un)optimization queue */ | |
afd66255 | 721 | list_del_init(&op->list); |
6274de49 MH |
722 | |
723 | op->kp.flags &= ~KPROBE_FLAG_OPTIMIZED; | |
724 | /* Don't touch the code, because it is already freed. */ | |
afd66255 MH |
725 | arch_remove_optimized_kprobe(op); |
726 | } | |
727 | ||
728 | /* Try to prepare optimized instructions */ | |
729 | static __kprobes void prepare_optimized_kprobe(struct kprobe *p) | |
730 | { | |
731 | struct optimized_kprobe *op; | |
732 | ||
733 | op = container_of(p, struct optimized_kprobe, kp); | |
734 | arch_prepare_optimized_kprobe(op); | |
735 | } | |
736 | ||
afd66255 MH |
737 | /* Allocate new optimized_kprobe and try to prepare optimized instructions */ |
738 | static __kprobes struct kprobe *alloc_aggr_kprobe(struct kprobe *p) | |
739 | { | |
740 | struct optimized_kprobe *op; | |
741 | ||
742 | op = kzalloc(sizeof(struct optimized_kprobe), GFP_KERNEL); | |
743 | if (!op) | |
744 | return NULL; | |
745 | ||
746 | INIT_LIST_HEAD(&op->list); | |
747 | op->kp.addr = p->addr; | |
748 | arch_prepare_optimized_kprobe(op); | |
749 | ||
750 | return &op->kp; | |
751 | } | |
752 | ||
753 | static void __kprobes init_aggr_kprobe(struct kprobe *ap, struct kprobe *p); | |
754 | ||
755 | /* | |
756 | * Prepare an optimized_kprobe and optimize it | |
757 | * NOTE: p must be a normal registered kprobe | |
758 | */ | |
759 | static __kprobes void try_to_optimize_kprobe(struct kprobe *p) | |
760 | { | |
761 | struct kprobe *ap; | |
762 | struct optimized_kprobe *op; | |
763 | ||
764 | ap = alloc_aggr_kprobe(p); | |
765 | if (!ap) | |
766 | return; | |
767 | ||
768 | op = container_of(ap, struct optimized_kprobe, kp); | |
769 | if (!arch_prepared_optinsn(&op->optinsn)) { | |
770 | /* If failed to setup optimizing, fallback to kprobe */ | |
6274de49 MH |
771 | arch_remove_optimized_kprobe(op); |
772 | kfree(op); | |
afd66255 MH |
773 | return; |
774 | } | |
775 | ||
776 | init_aggr_kprobe(ap, p); | |
777 | optimize_kprobe(ap); | |
778 | } | |
779 | ||
b2be84df | 780 | #ifdef CONFIG_SYSCTL |
43948f50 | 781 | /* This should be called with kprobe_mutex locked */ |
b2be84df MH |
782 | static void __kprobes optimize_all_kprobes(void) |
783 | { | |
784 | struct hlist_head *head; | |
785 | struct hlist_node *node; | |
786 | struct kprobe *p; | |
787 | unsigned int i; | |
788 | ||
789 | /* If optimization is already allowed, just return */ | |
790 | if (kprobes_allow_optimization) | |
791 | return; | |
792 | ||
793 | kprobes_allow_optimization = true; | |
b2be84df MH |
794 | for (i = 0; i < KPROBE_TABLE_SIZE; i++) { |
795 | head = &kprobe_table[i]; | |
796 | hlist_for_each_entry_rcu(p, node, head, hlist) | |
797 | if (!kprobe_disabled(p)) | |
798 | optimize_kprobe(p); | |
799 | } | |
b2be84df MH |
800 | printk(KERN_INFO "Kprobes globally optimized\n"); |
801 | } | |
802 | ||
43948f50 | 803 | /* This should be called with kprobe_mutex locked */ |
b2be84df MH |
804 | static void __kprobes unoptimize_all_kprobes(void) |
805 | { | |
806 | struct hlist_head *head; | |
807 | struct hlist_node *node; | |
808 | struct kprobe *p; | |
809 | unsigned int i; | |
810 | ||
811 | /* If optimization is already prohibited, just return */ | |
812 | if (!kprobes_allow_optimization) | |
813 | return; | |
814 | ||
815 | kprobes_allow_optimization = false; | |
b2be84df MH |
816 | for (i = 0; i < KPROBE_TABLE_SIZE; i++) { |
817 | head = &kprobe_table[i]; | |
818 | hlist_for_each_entry_rcu(p, node, head, hlist) { | |
819 | if (!kprobe_disabled(p)) | |
6274de49 | 820 | unoptimize_kprobe(p, false); |
b2be84df MH |
821 | } |
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 | ||
828 | int sysctl_kprobes_optimization; | |
829 | int proc_kprobes_optimization_handler(struct ctl_table *table, int write, | |
830 | void __user *buffer, size_t *length, | |
831 | loff_t *ppos) | |
832 | { | |
833 | int ret; | |
834 | ||
835 | mutex_lock(&kprobe_mutex); | |
836 | sysctl_kprobes_optimization = kprobes_allow_optimization ? 1 : 0; | |
837 | ret = proc_dointvec_minmax(table, write, buffer, length, ppos); | |
838 | ||
839 | if (sysctl_kprobes_optimization) | |
840 | optimize_all_kprobes(); | |
841 | else | |
842 | unoptimize_all_kprobes(); | |
843 | mutex_unlock(&kprobe_mutex); | |
844 | ||
845 | return ret; | |
846 | } | |
847 | #endif /* CONFIG_SYSCTL */ | |
848 | ||
6274de49 | 849 | /* Put a breakpoint for a probe. Must be called with text_mutex locked */ |
afd66255 MH |
850 | static void __kprobes __arm_kprobe(struct kprobe *p) |
851 | { | |
6d8e40a8 | 852 | struct kprobe *_p; |
afd66255 MH |
853 | |
854 | /* Check collision with other optimized kprobes */ | |
6d8e40a8 MH |
855 | _p = get_optimized_kprobe((unsigned long)p->addr); |
856 | if (unlikely(_p)) | |
6274de49 MH |
857 | /* Fallback to unoptimized kprobe */ |
858 | unoptimize_kprobe(_p, true); | |
afd66255 MH |
859 | |
860 | arch_arm_kprobe(p); | |
861 | optimize_kprobe(p); /* Try to optimize (add kprobe to a list) */ | |
862 | } | |
863 | ||
6274de49 MH |
864 | /* Remove the breakpoint of a probe. Must be called with text_mutex locked */ |
865 | static void __kprobes __disarm_kprobe(struct kprobe *p, bool reopt) | |
afd66255 | 866 | { |
6d8e40a8 | 867 | struct kprobe *_p; |
afd66255 | 868 | |
6274de49 | 869 | unoptimize_kprobe(p, false); /* Try to unoptimize */ |
afd66255 | 870 | |
6274de49 MH |
871 | if (!kprobe_queued(p)) { |
872 | arch_disarm_kprobe(p); | |
873 | /* If another kprobe was blocked, optimize it. */ | |
874 | _p = get_optimized_kprobe((unsigned long)p->addr); | |
875 | if (unlikely(_p) && reopt) | |
876 | optimize_kprobe(_p); | |
877 | } | |
878 | /* TODO: reoptimize others after unoptimized this probe */ | |
afd66255 MH |
879 | } |
880 | ||
881 | #else /* !CONFIG_OPTPROBES */ | |
882 | ||
883 | #define optimize_kprobe(p) do {} while (0) | |
6274de49 | 884 | #define unoptimize_kprobe(p, f) do {} while (0) |
afd66255 MH |
885 | #define kill_optimized_kprobe(p) do {} while (0) |
886 | #define prepare_optimized_kprobe(p) do {} while (0) | |
887 | #define try_to_optimize_kprobe(p) do {} while (0) | |
888 | #define __arm_kprobe(p) arch_arm_kprobe(p) | |
6274de49 MH |
889 | #define __disarm_kprobe(p, o) arch_disarm_kprobe(p) |
890 | #define kprobe_disarmed(p) kprobe_disabled(p) | |
891 | #define wait_for_kprobe_optimizer() do {} while (0) | |
afd66255 | 892 | |
0490cd1f MH |
893 | /* There should be no unused kprobes can be reused without optimization */ |
894 | static void reuse_unused_kprobe(struct kprobe *ap) | |
895 | { | |
896 | printk(KERN_ERR "Error: There should be no unused kprobe here.\n"); | |
897 | BUG_ON(kprobe_unused(ap)); | |
898 | } | |
899 | ||
afd66255 MH |
900 | static __kprobes void free_aggr_kprobe(struct kprobe *p) |
901 | { | |
6274de49 | 902 | arch_remove_kprobe(p); |
afd66255 MH |
903 | kfree(p); |
904 | } | |
905 | ||
906 | static __kprobes struct kprobe *alloc_aggr_kprobe(struct kprobe *p) | |
907 | { | |
908 | return kzalloc(sizeof(struct kprobe), GFP_KERNEL); | |
909 | } | |
910 | #endif /* CONFIG_OPTPROBES */ | |
911 | ||
201517a7 MH |
912 | /* Arm a kprobe with text_mutex */ |
913 | static void __kprobes arm_kprobe(struct kprobe *kp) | |
914 | { | |
afd66255 MH |
915 | /* |
916 | * Here, since __arm_kprobe() doesn't use stop_machine(), | |
917 | * this doesn't cause deadlock on text_mutex. So, we don't | |
918 | * need get_online_cpus(). | |
919 | */ | |
201517a7 | 920 | mutex_lock(&text_mutex); |
afd66255 | 921 | __arm_kprobe(kp); |
201517a7 MH |
922 | mutex_unlock(&text_mutex); |
923 | } | |
924 | ||
925 | /* Disarm a kprobe with text_mutex */ | |
926 | static void __kprobes disarm_kprobe(struct kprobe *kp) | |
927 | { | |
6274de49 | 928 | /* Ditto */ |
201517a7 | 929 | mutex_lock(&text_mutex); |
6274de49 | 930 | __disarm_kprobe(kp, true); |
201517a7 MH |
931 | mutex_unlock(&text_mutex); |
932 | } | |
933 | ||
64f562c6 AM |
934 | /* |
935 | * Aggregate handlers for multiple kprobes support - these handlers | |
936 | * take care of invoking the individual kprobe handlers on p->list | |
937 | */ | |
d0aaff97 | 938 | static int __kprobes aggr_pre_handler(struct kprobe *p, struct pt_regs *regs) |
64f562c6 AM |
939 | { |
940 | struct kprobe *kp; | |
941 | ||
3516a460 | 942 | list_for_each_entry_rcu(kp, &p->list, list) { |
de5bd88d | 943 | if (kp->pre_handler && likely(!kprobe_disabled(kp))) { |
e6584523 | 944 | set_kprobe_instance(kp); |
8b0914ea PP |
945 | if (kp->pre_handler(kp, regs)) |
946 | return 1; | |
64f562c6 | 947 | } |
e6584523 | 948 | reset_kprobe_instance(); |
64f562c6 AM |
949 | } |
950 | return 0; | |
951 | } | |
952 | ||
d0aaff97 PP |
953 | static void __kprobes aggr_post_handler(struct kprobe *p, struct pt_regs *regs, |
954 | unsigned long flags) | |
64f562c6 AM |
955 | { |
956 | struct kprobe *kp; | |
957 | ||
3516a460 | 958 | list_for_each_entry_rcu(kp, &p->list, list) { |
de5bd88d | 959 | if (kp->post_handler && likely(!kprobe_disabled(kp))) { |
e6584523 | 960 | set_kprobe_instance(kp); |
64f562c6 | 961 | kp->post_handler(kp, regs, flags); |
e6584523 | 962 | reset_kprobe_instance(); |
64f562c6 AM |
963 | } |
964 | } | |
64f562c6 AM |
965 | } |
966 | ||
d0aaff97 PP |
967 | static int __kprobes aggr_fault_handler(struct kprobe *p, struct pt_regs *regs, |
968 | int trapnr) | |
64f562c6 | 969 | { |
e6584523 AM |
970 | struct kprobe *cur = __get_cpu_var(kprobe_instance); |
971 | ||
64f562c6 AM |
972 | /* |
973 | * if we faulted "during" the execution of a user specified | |
974 | * probe handler, invoke just that probe's fault handler | |
975 | */ | |
e6584523 AM |
976 | if (cur && cur->fault_handler) { |
977 | if (cur->fault_handler(cur, regs, trapnr)) | |
64f562c6 AM |
978 | return 1; |
979 | } | |
980 | return 0; | |
981 | } | |
982 | ||
d0aaff97 | 983 | static int __kprobes aggr_break_handler(struct kprobe *p, struct pt_regs *regs) |
8b0914ea | 984 | { |
e6584523 AM |
985 | struct kprobe *cur = __get_cpu_var(kprobe_instance); |
986 | int ret = 0; | |
987 | ||
988 | if (cur && cur->break_handler) { | |
989 | if (cur->break_handler(cur, regs)) | |
990 | ret = 1; | |
8b0914ea | 991 | } |
e6584523 AM |
992 | reset_kprobe_instance(); |
993 | return ret; | |
8b0914ea PP |
994 | } |
995 | ||
bf8d5c52 KA |
996 | /* Walks the list and increments nmissed count for multiprobe case */ |
997 | void __kprobes kprobes_inc_nmissed_count(struct kprobe *p) | |
998 | { | |
999 | struct kprobe *kp; | |
afd66255 | 1000 | if (!kprobe_aggrprobe(p)) { |
bf8d5c52 KA |
1001 | p->nmissed++; |
1002 | } else { | |
1003 | list_for_each_entry_rcu(kp, &p->list, list) | |
1004 | kp->nmissed++; | |
1005 | } | |
1006 | return; | |
1007 | } | |
1008 | ||
99219a3f | 1009 | void __kprobes recycle_rp_inst(struct kretprobe_instance *ri, |
1010 | struct hlist_head *head) | |
b94cce92 | 1011 | { |
ef53d9c5 S |
1012 | struct kretprobe *rp = ri->rp; |
1013 | ||
b94cce92 HN |
1014 | /* remove rp inst off the rprobe_inst_table */ |
1015 | hlist_del(&ri->hlist); | |
ef53d9c5 S |
1016 | INIT_HLIST_NODE(&ri->hlist); |
1017 | if (likely(rp)) { | |
1018 | spin_lock(&rp->lock); | |
1019 | hlist_add_head(&ri->hlist, &rp->free_instances); | |
1020 | spin_unlock(&rp->lock); | |
b94cce92 HN |
1021 | } else |
1022 | /* Unregistering */ | |
99219a3f | 1023 | hlist_add_head(&ri->hlist, head); |
b94cce92 HN |
1024 | } |
1025 | ||
017c39bd | 1026 | void __kprobes kretprobe_hash_lock(struct task_struct *tsk, |
ef53d9c5 | 1027 | struct hlist_head **head, unsigned long *flags) |
635c17c2 | 1028 | __acquires(hlist_lock) |
ef53d9c5 S |
1029 | { |
1030 | unsigned long hash = hash_ptr(tsk, KPROBE_HASH_BITS); | |
1031 | spinlock_t *hlist_lock; | |
1032 | ||
1033 | *head = &kretprobe_inst_table[hash]; | |
1034 | hlist_lock = kretprobe_table_lock_ptr(hash); | |
1035 | spin_lock_irqsave(hlist_lock, *flags); | |
1036 | } | |
1037 | ||
017c39bd MH |
1038 | static void __kprobes kretprobe_table_lock(unsigned long hash, |
1039 | unsigned long *flags) | |
635c17c2 | 1040 | __acquires(hlist_lock) |
b94cce92 | 1041 | { |
ef53d9c5 S |
1042 | spinlock_t *hlist_lock = kretprobe_table_lock_ptr(hash); |
1043 | spin_lock_irqsave(hlist_lock, *flags); | |
1044 | } | |
1045 | ||
017c39bd MH |
1046 | void __kprobes kretprobe_hash_unlock(struct task_struct *tsk, |
1047 | unsigned long *flags) | |
635c17c2 | 1048 | __releases(hlist_lock) |
ef53d9c5 S |
1049 | { |
1050 | unsigned long hash = hash_ptr(tsk, KPROBE_HASH_BITS); | |
1051 | spinlock_t *hlist_lock; | |
1052 | ||
1053 | hlist_lock = kretprobe_table_lock_ptr(hash); | |
1054 | spin_unlock_irqrestore(hlist_lock, *flags); | |
1055 | } | |
1056 | ||
6376b229 NK |
1057 | static void __kprobes kretprobe_table_unlock(unsigned long hash, |
1058 | unsigned long *flags) | |
635c17c2 | 1059 | __releases(hlist_lock) |
ef53d9c5 S |
1060 | { |
1061 | spinlock_t *hlist_lock = kretprobe_table_lock_ptr(hash); | |
1062 | spin_unlock_irqrestore(hlist_lock, *flags); | |
b94cce92 HN |
1063 | } |
1064 | ||
b94cce92 | 1065 | /* |
c6fd91f0 | 1066 | * This function is called from finish_task_switch when task tk becomes dead, |
1067 | * so that we can recycle any function-return probe instances associated | |
1068 | * with this task. These left over instances represent probed functions | |
1069 | * that have been called but will never return. | |
b94cce92 | 1070 | */ |
d0aaff97 | 1071 | void __kprobes kprobe_flush_task(struct task_struct *tk) |
b94cce92 | 1072 | { |
62c27be0 | 1073 | struct kretprobe_instance *ri; |
99219a3f | 1074 | struct hlist_head *head, empty_rp; |
802eae7c | 1075 | struct hlist_node *node, *tmp; |
ef53d9c5 | 1076 | unsigned long hash, flags = 0; |
802eae7c | 1077 | |
ef53d9c5 S |
1078 | if (unlikely(!kprobes_initialized)) |
1079 | /* Early boot. kretprobe_table_locks not yet initialized. */ | |
1080 | return; | |
1081 | ||
1082 | hash = hash_ptr(tk, KPROBE_HASH_BITS); | |
1083 | head = &kretprobe_inst_table[hash]; | |
1084 | kretprobe_table_lock(hash, &flags); | |
62c27be0 | 1085 | hlist_for_each_entry_safe(ri, node, tmp, head, hlist) { |
1086 | if (ri->task == tk) | |
99219a3f | 1087 | recycle_rp_inst(ri, &empty_rp); |
62c27be0 | 1088 | } |
ef53d9c5 S |
1089 | kretprobe_table_unlock(hash, &flags); |
1090 | INIT_HLIST_HEAD(&empty_rp); | |
99219a3f | 1091 | hlist_for_each_entry_safe(ri, node, tmp, &empty_rp, hlist) { |
1092 | hlist_del(&ri->hlist); | |
1093 | kfree(ri); | |
1094 | } | |
b94cce92 HN |
1095 | } |
1096 | ||
b94cce92 HN |
1097 | static inline void free_rp_inst(struct kretprobe *rp) |
1098 | { | |
1099 | struct kretprobe_instance *ri; | |
4c4308cb CH |
1100 | struct hlist_node *pos, *next; |
1101 | ||
ef53d9c5 S |
1102 | hlist_for_each_entry_safe(ri, pos, next, &rp->free_instances, hlist) { |
1103 | hlist_del(&ri->hlist); | |
b94cce92 HN |
1104 | kfree(ri); |
1105 | } | |
1106 | } | |
1107 | ||
4a296e07 MH |
1108 | static void __kprobes cleanup_rp_inst(struct kretprobe *rp) |
1109 | { | |
ef53d9c5 | 1110 | unsigned long flags, hash; |
4a296e07 MH |
1111 | struct kretprobe_instance *ri; |
1112 | struct hlist_node *pos, *next; | |
ef53d9c5 S |
1113 | struct hlist_head *head; |
1114 | ||
4a296e07 | 1115 | /* No race here */ |
ef53d9c5 S |
1116 | for (hash = 0; hash < KPROBE_TABLE_SIZE; hash++) { |
1117 | kretprobe_table_lock(hash, &flags); | |
1118 | head = &kretprobe_inst_table[hash]; | |
1119 | hlist_for_each_entry_safe(ri, pos, next, head, hlist) { | |
1120 | if (ri->rp == rp) | |
1121 | ri->rp = NULL; | |
1122 | } | |
1123 | kretprobe_table_unlock(hash, &flags); | |
4a296e07 | 1124 | } |
4a296e07 MH |
1125 | free_rp_inst(rp); |
1126 | } | |
1127 | ||
8b0914ea | 1128 | /* |
b918e5e6 | 1129 | * Add the new probe to ap->list. Fail if this is the |
8b0914ea PP |
1130 | * second jprobe at the address - two jprobes can't coexist |
1131 | */ | |
b918e5e6 | 1132 | static int __kprobes add_new_kprobe(struct kprobe *ap, struct kprobe *p) |
8b0914ea | 1133 | { |
de5bd88d | 1134 | BUG_ON(kprobe_gone(ap) || kprobe_gone(p)); |
afd66255 MH |
1135 | |
1136 | if (p->break_handler || p->post_handler) | |
6274de49 | 1137 | unoptimize_kprobe(ap, true); /* Fall back to normal kprobe */ |
afd66255 | 1138 | |
8b0914ea | 1139 | if (p->break_handler) { |
b918e5e6 | 1140 | if (ap->break_handler) |
36721656 | 1141 | return -EEXIST; |
b918e5e6 MH |
1142 | list_add_tail_rcu(&p->list, &ap->list); |
1143 | ap->break_handler = aggr_break_handler; | |
8b0914ea | 1144 | } else |
b918e5e6 MH |
1145 | list_add_rcu(&p->list, &ap->list); |
1146 | if (p->post_handler && !ap->post_handler) | |
1147 | ap->post_handler = aggr_post_handler; | |
de5bd88d MH |
1148 | |
1149 | if (kprobe_disabled(ap) && !kprobe_disabled(p)) { | |
1150 | ap->flags &= ~KPROBE_FLAG_DISABLED; | |
1151 | if (!kprobes_all_disarmed) | |
1152 | /* Arm the breakpoint again. */ | |
afd66255 | 1153 | __arm_kprobe(ap); |
de5bd88d | 1154 | } |
8b0914ea PP |
1155 | return 0; |
1156 | } | |
1157 | ||
64f562c6 AM |
1158 | /* |
1159 | * Fill in the required fields of the "manager kprobe". Replace the | |
1160 | * earlier kprobe in the hlist with the manager kprobe | |
1161 | */ | |
afd66255 | 1162 | static void __kprobes init_aggr_kprobe(struct kprobe *ap, struct kprobe *p) |
64f562c6 | 1163 | { |
afd66255 | 1164 | /* Copy p's insn slot to ap */ |
8b0914ea | 1165 | copy_kprobe(p, ap); |
a9ad965e | 1166 | flush_insn_slot(ap); |
64f562c6 | 1167 | ap->addr = p->addr; |
afd66255 | 1168 | ap->flags = p->flags & ~KPROBE_FLAG_OPTIMIZED; |
64f562c6 | 1169 | ap->pre_handler = aggr_pre_handler; |
64f562c6 | 1170 | ap->fault_handler = aggr_fault_handler; |
e8386a0c MH |
1171 | /* We don't care the kprobe which has gone. */ |
1172 | if (p->post_handler && !kprobe_gone(p)) | |
36721656 | 1173 | ap->post_handler = aggr_post_handler; |
e8386a0c | 1174 | if (p->break_handler && !kprobe_gone(p)) |
36721656 | 1175 | ap->break_handler = aggr_break_handler; |
64f562c6 AM |
1176 | |
1177 | INIT_LIST_HEAD(&ap->list); | |
afd66255 | 1178 | INIT_HLIST_NODE(&ap->hlist); |
64f562c6 | 1179 | |
afd66255 | 1180 | list_add_rcu(&p->list, &ap->list); |
adad0f33 | 1181 | hlist_replace_rcu(&p->hlist, &ap->hlist); |
64f562c6 AM |
1182 | } |
1183 | ||
1184 | /* | |
1185 | * This is the second or subsequent kprobe at the address - handle | |
1186 | * the intricacies | |
64f562c6 | 1187 | */ |
6d8e40a8 | 1188 | static int __kprobes register_aggr_kprobe(struct kprobe *orig_p, |
d0aaff97 | 1189 | struct kprobe *p) |
64f562c6 AM |
1190 | { |
1191 | int ret = 0; | |
6d8e40a8 | 1192 | struct kprobe *ap = orig_p; |
64f562c6 | 1193 | |
6d8e40a8 MH |
1194 | if (!kprobe_aggrprobe(orig_p)) { |
1195 | /* If orig_p is not an aggr_kprobe, create new aggr_kprobe. */ | |
1196 | ap = alloc_aggr_kprobe(orig_p); | |
b918e5e6 MH |
1197 | if (!ap) |
1198 | return -ENOMEM; | |
6d8e40a8 | 1199 | init_aggr_kprobe(ap, orig_p); |
6274de49 | 1200 | } else if (kprobe_unused(ap)) |
0490cd1f MH |
1201 | /* This probe is going to die. Rescue it */ |
1202 | reuse_unused_kprobe(ap); | |
b918e5e6 MH |
1203 | |
1204 | if (kprobe_gone(ap)) { | |
e8386a0c MH |
1205 | /* |
1206 | * Attempting to insert new probe at the same location that | |
1207 | * had a probe in the module vaddr area which already | |
1208 | * freed. So, the instruction slot has already been | |
1209 | * released. We need a new slot for the new probe. | |
1210 | */ | |
b918e5e6 | 1211 | ret = arch_prepare_kprobe(ap); |
e8386a0c | 1212 | if (ret) |
b918e5e6 MH |
1213 | /* |
1214 | * Even if fail to allocate new slot, don't need to | |
1215 | * free aggr_probe. It will be used next time, or | |
1216 | * freed by unregister_kprobe. | |
1217 | */ | |
e8386a0c | 1218 | return ret; |
de5bd88d | 1219 | |
afd66255 MH |
1220 | /* Prepare optimized instructions if possible. */ |
1221 | prepare_optimized_kprobe(ap); | |
1222 | ||
e8386a0c | 1223 | /* |
de5bd88d MH |
1224 | * Clear gone flag to prevent allocating new slot again, and |
1225 | * set disabled flag because it is not armed yet. | |
e8386a0c | 1226 | */ |
de5bd88d MH |
1227 | ap->flags = (ap->flags & ~KPROBE_FLAG_GONE) |
1228 | | KPROBE_FLAG_DISABLED; | |
e8386a0c | 1229 | } |
b918e5e6 | 1230 | |
afd66255 | 1231 | /* Copy ap's insn slot to p */ |
b918e5e6 MH |
1232 | copy_kprobe(ap, p); |
1233 | return add_new_kprobe(ap, p); | |
64f562c6 AM |
1234 | } |
1235 | ||
d0aaff97 PP |
1236 | static int __kprobes in_kprobes_functions(unsigned long addr) |
1237 | { | |
3d8d996e SD |
1238 | struct kprobe_blackpoint *kb; |
1239 | ||
6f716acd CH |
1240 | if (addr >= (unsigned long)__kprobes_text_start && |
1241 | addr < (unsigned long)__kprobes_text_end) | |
d0aaff97 | 1242 | return -EINVAL; |
3d8d996e SD |
1243 | /* |
1244 | * If there exists a kprobe_blacklist, verify and | |
1245 | * fail any probe registration in the prohibited area | |
1246 | */ | |
1247 | for (kb = kprobe_blacklist; kb->name != NULL; kb++) { | |
1248 | if (kb->start_addr) { | |
1249 | if (addr >= kb->start_addr && | |
1250 | addr < (kb->start_addr + kb->range)) | |
1251 | return -EINVAL; | |
1252 | } | |
1253 | } | |
d0aaff97 PP |
1254 | return 0; |
1255 | } | |
1256 | ||
b2a5cd69 MH |
1257 | /* |
1258 | * If we have a symbol_name argument, look it up and add the offset field | |
1259 | * to it. This way, we can specify a relative address to a symbol. | |
1260 | */ | |
1261 | static kprobe_opcode_t __kprobes *kprobe_addr(struct kprobe *p) | |
1262 | { | |
1263 | kprobe_opcode_t *addr = p->addr; | |
1264 | if (p->symbol_name) { | |
1265 | if (addr) | |
1266 | return NULL; | |
1267 | kprobe_lookup_name(p->symbol_name, addr); | |
1268 | } | |
1269 | ||
1270 | if (!addr) | |
1271 | return NULL; | |
1272 | return (kprobe_opcode_t *)(((char *)addr) + p->offset); | |
1273 | } | |
1274 | ||
1f0ab409 AM |
1275 | /* Check passed kprobe is valid and return kprobe in kprobe_table. */ |
1276 | static struct kprobe * __kprobes __get_valid_kprobe(struct kprobe *p) | |
1277 | { | |
6d8e40a8 | 1278 | struct kprobe *ap, *list_p; |
1f0ab409 | 1279 | |
6d8e40a8 MH |
1280 | ap = get_kprobe(p->addr); |
1281 | if (unlikely(!ap)) | |
1f0ab409 AM |
1282 | return NULL; |
1283 | ||
6d8e40a8 MH |
1284 | if (p != ap) { |
1285 | list_for_each_entry_rcu(list_p, &ap->list, list) | |
1f0ab409 AM |
1286 | if (list_p == p) |
1287 | /* kprobe p is a valid probe */ | |
1288 | goto valid; | |
1289 | return NULL; | |
1290 | } | |
1291 | valid: | |
6d8e40a8 | 1292 | return ap; |
1f0ab409 AM |
1293 | } |
1294 | ||
1295 | /* Return error if the kprobe is being re-registered */ | |
1296 | static inline int check_kprobe_rereg(struct kprobe *p) | |
1297 | { | |
1298 | int ret = 0; | |
1f0ab409 AM |
1299 | |
1300 | mutex_lock(&kprobe_mutex); | |
6d8e40a8 | 1301 | if (__get_valid_kprobe(p)) |
1f0ab409 AM |
1302 | ret = -EINVAL; |
1303 | mutex_unlock(&kprobe_mutex); | |
6d8e40a8 | 1304 | |
1f0ab409 AM |
1305 | return ret; |
1306 | } | |
1307 | ||
49ad2fd7 | 1308 | int __kprobes register_kprobe(struct kprobe *p) |
1da177e4 LT |
1309 | { |
1310 | int ret = 0; | |
64f562c6 | 1311 | struct kprobe *old_p; |
df019b1d | 1312 | struct module *probed_mod; |
b2a5cd69 | 1313 | kprobe_opcode_t *addr; |
b3e55c72 | 1314 | |
b2a5cd69 MH |
1315 | addr = kprobe_addr(p); |
1316 | if (!addr) | |
3a872d89 | 1317 | return -EINVAL; |
b2a5cd69 | 1318 | p->addr = addr; |
3a872d89 | 1319 | |
1f0ab409 AM |
1320 | ret = check_kprobe_rereg(p); |
1321 | if (ret) | |
1322 | return ret; | |
1323 | ||
91bad2f8 | 1324 | jump_label_lock(); |
de31c3ca | 1325 | preempt_disable(); |
ec30c5f3 | 1326 | if (!kernel_text_address((unsigned long) p->addr) || |
4554dbcb | 1327 | in_kprobes_functions((unsigned long) p->addr) || |
4c3ef6d7 | 1328 | ftrace_text_reserved(p->addr, p->addr) || |
de31c3ca SR |
1329 | jump_label_text_reserved(p->addr, p->addr)) |
1330 | goto fail_with_jump_label; | |
b3e55c72 | 1331 | |
de5bd88d MH |
1332 | /* User can pass only KPROBE_FLAG_DISABLED to register_kprobe */ |
1333 | p->flags &= KPROBE_FLAG_DISABLED; | |
1334 | ||
6f716acd CH |
1335 | /* |
1336 | * Check if are we probing a module. | |
1337 | */ | |
a189d035 | 1338 | probed_mod = __module_text_address((unsigned long) p->addr); |
6f716acd | 1339 | if (probed_mod) { |
6f716acd | 1340 | /* |
e8386a0c MH |
1341 | * We must hold a refcount of the probed module while updating |
1342 | * its code to prohibit unexpected unloading. | |
df019b1d | 1343 | */ |
de31c3ca SR |
1344 | if (unlikely(!try_module_get(probed_mod))) |
1345 | goto fail_with_jump_label; | |
1346 | ||
f24659d9 MH |
1347 | /* |
1348 | * If the module freed .init.text, we couldn't insert | |
1349 | * kprobes in there. | |
1350 | */ | |
1351 | if (within_module_init((unsigned long)p->addr, probed_mod) && | |
1352 | probed_mod->state != MODULE_STATE_COMING) { | |
1353 | module_put(probed_mod); | |
de31c3ca | 1354 | goto fail_with_jump_label; |
f24659d9 | 1355 | } |
df019b1d | 1356 | } |
a189d035 | 1357 | preempt_enable(); |
de31c3ca | 1358 | jump_label_unlock(); |
1da177e4 | 1359 | |
3516a460 | 1360 | p->nmissed = 0; |
9861668f | 1361 | INIT_LIST_HEAD(&p->list); |
7a7d1cf9 | 1362 | mutex_lock(&kprobe_mutex); |
afd66255 | 1363 | |
91bad2f8 JB |
1364 | jump_label_lock(); /* needed to call jump_label_text_reserved() */ |
1365 | ||
afd66255 MH |
1366 | get_online_cpus(); /* For avoiding text_mutex deadlock. */ |
1367 | mutex_lock(&text_mutex); | |
1368 | ||
64f562c6 AM |
1369 | old_p = get_kprobe(p->addr); |
1370 | if (old_p) { | |
afd66255 | 1371 | /* Since this may unoptimize old_p, locking text_mutex. */ |
64f562c6 | 1372 | ret = register_aggr_kprobe(old_p, p); |
1da177e4 LT |
1373 | goto out; |
1374 | } | |
1da177e4 | 1375 | |
6f716acd CH |
1376 | ret = arch_prepare_kprobe(p); |
1377 | if (ret) | |
afd66255 | 1378 | goto out; |
49a2a1b8 | 1379 | |
64f562c6 | 1380 | INIT_HLIST_NODE(&p->hlist); |
3516a460 | 1381 | hlist_add_head_rcu(&p->hlist, |
1da177e4 LT |
1382 | &kprobe_table[hash_ptr(p->addr, KPROBE_HASH_BITS)]); |
1383 | ||
de5bd88d | 1384 | if (!kprobes_all_disarmed && !kprobe_disabled(p)) |
afd66255 MH |
1385 | __arm_kprobe(p); |
1386 | ||
1387 | /* Try to optimize kprobe */ | |
1388 | try_to_optimize_kprobe(p); | |
74a0b576 | 1389 | |
1da177e4 | 1390 | out: |
afd66255 MH |
1391 | mutex_unlock(&text_mutex); |
1392 | put_online_cpus(); | |
91bad2f8 | 1393 | jump_label_unlock(); |
7a7d1cf9 | 1394 | mutex_unlock(&kprobe_mutex); |
49a2a1b8 | 1395 | |
e8386a0c | 1396 | if (probed_mod) |
df019b1d | 1397 | module_put(probed_mod); |
e8386a0c | 1398 | |
1da177e4 | 1399 | return ret; |
de31c3ca SR |
1400 | |
1401 | fail_with_jump_label: | |
1402 | preempt_enable(); | |
1403 | jump_label_unlock(); | |
1404 | return -EINVAL; | |
1da177e4 | 1405 | } |
99081ab5 | 1406 | EXPORT_SYMBOL_GPL(register_kprobe); |
1da177e4 | 1407 | |
6f0f1dd7 MH |
1408 | /* Check if all probes on the aggrprobe are disabled */ |
1409 | static int __kprobes aggr_kprobe_disabled(struct kprobe *ap) | |
1410 | { | |
1411 | struct kprobe *kp; | |
1412 | ||
1413 | list_for_each_entry_rcu(kp, &ap->list, list) | |
1414 | if (!kprobe_disabled(kp)) | |
1415 | /* | |
1416 | * There is an active probe on the list. | |
1417 | * We can't disable this ap. | |
1418 | */ | |
1419 | return 0; | |
1420 | ||
1421 | return 1; | |
1422 | } | |
1423 | ||
1424 | /* Disable one kprobe: Make sure called under kprobe_mutex is locked */ | |
1425 | static struct kprobe *__kprobes __disable_kprobe(struct kprobe *p) | |
1426 | { | |
1427 | struct kprobe *orig_p; | |
1428 | ||
1429 | /* Get an original kprobe for return */ | |
1430 | orig_p = __get_valid_kprobe(p); | |
1431 | if (unlikely(orig_p == NULL)) | |
1432 | return NULL; | |
1433 | ||
1434 | if (!kprobe_disabled(p)) { | |
1435 | /* Disable probe if it is a child probe */ | |
1436 | if (p != orig_p) | |
1437 | p->flags |= KPROBE_FLAG_DISABLED; | |
1438 | ||
1439 | /* Try to disarm and disable this/parent probe */ | |
1440 | if (p == orig_p || aggr_kprobe_disabled(orig_p)) { | |
1441 | disarm_kprobe(orig_p); | |
1442 | orig_p->flags |= KPROBE_FLAG_DISABLED; | |
1443 | } | |
1444 | } | |
1445 | ||
1446 | return orig_p; | |
1447 | } | |
1448 | ||
de5bd88d MH |
1449 | /* |
1450 | * Unregister a kprobe without a scheduler synchronization. | |
1451 | */ | |
1452 | static int __kprobes __unregister_kprobe_top(struct kprobe *p) | |
1453 | { | |
6d8e40a8 | 1454 | struct kprobe *ap, *list_p; |
de5bd88d | 1455 | |
6f0f1dd7 MH |
1456 | /* Disable kprobe. This will disarm it if needed. */ |
1457 | ap = __disable_kprobe(p); | |
6d8e40a8 | 1458 | if (ap == NULL) |
de5bd88d MH |
1459 | return -EINVAL; |
1460 | ||
6f0f1dd7 | 1461 | if (ap == p) |
bf8f6e5b | 1462 | /* |
6f0f1dd7 MH |
1463 | * This probe is an independent(and non-optimized) kprobe |
1464 | * (not an aggrprobe). Remove from the hash list. | |
bf8f6e5b | 1465 | */ |
6f0f1dd7 MH |
1466 | goto disarmed; |
1467 | ||
1468 | /* Following process expects this probe is an aggrprobe */ | |
1469 | WARN_ON(!kprobe_aggrprobe(ap)); | |
1470 | ||
6274de49 MH |
1471 | if (list_is_singular(&ap->list) && kprobe_disarmed(ap)) |
1472 | /* | |
1473 | * !disarmed could be happen if the probe is under delayed | |
1474 | * unoptimizing. | |
1475 | */ | |
6f0f1dd7 MH |
1476 | goto disarmed; |
1477 | else { | |
1478 | /* If disabling probe has special handlers, update aggrprobe */ | |
e8386a0c | 1479 | if (p->break_handler && !kprobe_gone(p)) |
6d8e40a8 | 1480 | ap->break_handler = NULL; |
e8386a0c | 1481 | if (p->post_handler && !kprobe_gone(p)) { |
6d8e40a8 | 1482 | list_for_each_entry_rcu(list_p, &ap->list, list) { |
9861668f MH |
1483 | if ((list_p != p) && (list_p->post_handler)) |
1484 | goto noclean; | |
1485 | } | |
6d8e40a8 | 1486 | ap->post_handler = NULL; |
9861668f MH |
1487 | } |
1488 | noclean: | |
6f0f1dd7 MH |
1489 | /* |
1490 | * Remove from the aggrprobe: this path will do nothing in | |
1491 | * __unregister_kprobe_bottom(). | |
1492 | */ | |
49a2a1b8 | 1493 | list_del_rcu(&p->list); |
6f0f1dd7 MH |
1494 | if (!kprobe_disabled(ap) && !kprobes_all_disarmed) |
1495 | /* | |
1496 | * Try to optimize this probe again, because post | |
1497 | * handler may have been changed. | |
1498 | */ | |
1499 | optimize_kprobe(ap); | |
49a2a1b8 | 1500 | } |
9861668f | 1501 | return 0; |
6f0f1dd7 MH |
1502 | |
1503 | disarmed: | |
6274de49 | 1504 | BUG_ON(!kprobe_disarmed(ap)); |
6f0f1dd7 MH |
1505 | hlist_del_rcu(&ap->hlist); |
1506 | return 0; | |
9861668f | 1507 | } |
3516a460 | 1508 | |
9861668f MH |
1509 | static void __kprobes __unregister_kprobe_bottom(struct kprobe *p) |
1510 | { | |
6d8e40a8 | 1511 | struct kprobe *ap; |
b3e55c72 | 1512 | |
e8386a0c | 1513 | if (list_empty(&p->list)) |
6274de49 | 1514 | /* This is an independent kprobe */ |
0498b635 | 1515 | arch_remove_kprobe(p); |
e8386a0c | 1516 | else if (list_is_singular(&p->list)) { |
6274de49 | 1517 | /* This is the last child of an aggrprobe */ |
6d8e40a8 | 1518 | ap = list_entry(p->list.next, struct kprobe, list); |
e8386a0c | 1519 | list_del(&p->list); |
6d8e40a8 | 1520 | free_aggr_kprobe(ap); |
9861668f | 1521 | } |
6274de49 | 1522 | /* Otherwise, do nothing. */ |
9861668f MH |
1523 | } |
1524 | ||
49ad2fd7 | 1525 | int __kprobes register_kprobes(struct kprobe **kps, int num) |
9861668f MH |
1526 | { |
1527 | int i, ret = 0; | |
1528 | ||
1529 | if (num <= 0) | |
1530 | return -EINVAL; | |
1531 | for (i = 0; i < num; i++) { | |
49ad2fd7 | 1532 | ret = register_kprobe(kps[i]); |
67dddaad MH |
1533 | if (ret < 0) { |
1534 | if (i > 0) | |
1535 | unregister_kprobes(kps, i); | |
9861668f | 1536 | break; |
36721656 | 1537 | } |
49a2a1b8 | 1538 | } |
9861668f MH |
1539 | return ret; |
1540 | } | |
99081ab5 | 1541 | EXPORT_SYMBOL_GPL(register_kprobes); |
9861668f | 1542 | |
9861668f MH |
1543 | void __kprobes unregister_kprobe(struct kprobe *p) |
1544 | { | |
1545 | unregister_kprobes(&p, 1); | |
1546 | } | |
99081ab5 | 1547 | EXPORT_SYMBOL_GPL(unregister_kprobe); |
9861668f | 1548 | |
9861668f MH |
1549 | void __kprobes unregister_kprobes(struct kprobe **kps, int num) |
1550 | { | |
1551 | int i; | |
1552 | ||
1553 | if (num <= 0) | |
1554 | return; | |
1555 | mutex_lock(&kprobe_mutex); | |
1556 | for (i = 0; i < num; i++) | |
1557 | if (__unregister_kprobe_top(kps[i]) < 0) | |
1558 | kps[i]->addr = NULL; | |
1559 | mutex_unlock(&kprobe_mutex); | |
1560 | ||
1561 | synchronize_sched(); | |
1562 | for (i = 0; i < num; i++) | |
1563 | if (kps[i]->addr) | |
1564 | __unregister_kprobe_bottom(kps[i]); | |
1da177e4 | 1565 | } |
99081ab5 | 1566 | EXPORT_SYMBOL_GPL(unregister_kprobes); |
1da177e4 LT |
1567 | |
1568 | static struct notifier_block kprobe_exceptions_nb = { | |
3d5631e0 AK |
1569 | .notifier_call = kprobe_exceptions_notify, |
1570 | .priority = 0x7fffffff /* we need to be notified first */ | |
1571 | }; | |
1572 | ||
3d7e3382 ME |
1573 | unsigned long __weak arch_deref_entry_point(void *entry) |
1574 | { | |
1575 | return (unsigned long)entry; | |
1576 | } | |
1da177e4 | 1577 | |
49ad2fd7 | 1578 | int __kprobes register_jprobes(struct jprobe **jps, int num) |
1da177e4 | 1579 | { |
26b31c19 MH |
1580 | struct jprobe *jp; |
1581 | int ret = 0, i; | |
3d7e3382 | 1582 | |
26b31c19 | 1583 | if (num <= 0) |
3d7e3382 | 1584 | return -EINVAL; |
26b31c19 | 1585 | for (i = 0; i < num; i++) { |
05662bdb | 1586 | unsigned long addr, offset; |
26b31c19 MH |
1587 | jp = jps[i]; |
1588 | addr = arch_deref_entry_point(jp->entry); | |
1589 | ||
05662bdb NK |
1590 | /* Verify probepoint is a function entry point */ |
1591 | if (kallsyms_lookup_size_offset(addr, NULL, &offset) && | |
1592 | offset == 0) { | |
1593 | jp->kp.pre_handler = setjmp_pre_handler; | |
1594 | jp->kp.break_handler = longjmp_break_handler; | |
1595 | ret = register_kprobe(&jp->kp); | |
1596 | } else | |
1597 | ret = -EINVAL; | |
edbaadbe | 1598 | |
67dddaad MH |
1599 | if (ret < 0) { |
1600 | if (i > 0) | |
1601 | unregister_jprobes(jps, i); | |
26b31c19 MH |
1602 | break; |
1603 | } | |
1604 | } | |
1605 | return ret; | |
1606 | } | |
99081ab5 | 1607 | EXPORT_SYMBOL_GPL(register_jprobes); |
3d7e3382 | 1608 | |
26b31c19 MH |
1609 | int __kprobes register_jprobe(struct jprobe *jp) |
1610 | { | |
49ad2fd7 | 1611 | return register_jprobes(&jp, 1); |
1da177e4 | 1612 | } |
99081ab5 | 1613 | EXPORT_SYMBOL_GPL(register_jprobe); |
1da177e4 | 1614 | |
d0aaff97 | 1615 | void __kprobes unregister_jprobe(struct jprobe *jp) |
1da177e4 | 1616 | { |
26b31c19 MH |
1617 | unregister_jprobes(&jp, 1); |
1618 | } | |
99081ab5 | 1619 | EXPORT_SYMBOL_GPL(unregister_jprobe); |
26b31c19 | 1620 | |
26b31c19 MH |
1621 | void __kprobes unregister_jprobes(struct jprobe **jps, int num) |
1622 | { | |
1623 | int i; | |
1624 | ||
1625 | if (num <= 0) | |
1626 | return; | |
1627 | mutex_lock(&kprobe_mutex); | |
1628 | for (i = 0; i < num; i++) | |
1629 | if (__unregister_kprobe_top(&jps[i]->kp) < 0) | |
1630 | jps[i]->kp.addr = NULL; | |
1631 | mutex_unlock(&kprobe_mutex); | |
1632 | ||
1633 | synchronize_sched(); | |
1634 | for (i = 0; i < num; i++) { | |
1635 | if (jps[i]->kp.addr) | |
1636 | __unregister_kprobe_bottom(&jps[i]->kp); | |
1637 | } | |
1da177e4 | 1638 | } |
99081ab5 | 1639 | EXPORT_SYMBOL_GPL(unregister_jprobes); |
1da177e4 | 1640 | |
9edddaa2 | 1641 | #ifdef CONFIG_KRETPROBES |
e65cefe8 AB |
1642 | /* |
1643 | * This kprobe pre_handler is registered with every kretprobe. When probe | |
1644 | * hits it will set up the return probe. | |
1645 | */ | |
1646 | static int __kprobes pre_handler_kretprobe(struct kprobe *p, | |
1647 | struct pt_regs *regs) | |
1648 | { | |
1649 | struct kretprobe *rp = container_of(p, struct kretprobe, kp); | |
ef53d9c5 S |
1650 | unsigned long hash, flags = 0; |
1651 | struct kretprobe_instance *ri; | |
e65cefe8 AB |
1652 | |
1653 | /*TODO: consider to only swap the RA after the last pre_handler fired */ | |
ef53d9c5 S |
1654 | hash = hash_ptr(current, KPROBE_HASH_BITS); |
1655 | spin_lock_irqsave(&rp->lock, flags); | |
4c4308cb | 1656 | if (!hlist_empty(&rp->free_instances)) { |
4c4308cb | 1657 | ri = hlist_entry(rp->free_instances.first, |
ef53d9c5 S |
1658 | struct kretprobe_instance, hlist); |
1659 | hlist_del(&ri->hlist); | |
1660 | spin_unlock_irqrestore(&rp->lock, flags); | |
1661 | ||
4c4308cb CH |
1662 | ri->rp = rp; |
1663 | ri->task = current; | |
f47cd9b5 | 1664 | |
f02b8624 | 1665 | if (rp->entry_handler && rp->entry_handler(ri, regs)) |
f47cd9b5 | 1666 | return 0; |
f47cd9b5 | 1667 | |
4c4308cb CH |
1668 | arch_prepare_kretprobe(ri, regs); |
1669 | ||
1670 | /* XXX(hch): why is there no hlist_move_head? */ | |
ef53d9c5 S |
1671 | INIT_HLIST_NODE(&ri->hlist); |
1672 | kretprobe_table_lock(hash, &flags); | |
1673 | hlist_add_head(&ri->hlist, &kretprobe_inst_table[hash]); | |
1674 | kretprobe_table_unlock(hash, &flags); | |
1675 | } else { | |
4c4308cb | 1676 | rp->nmissed++; |
ef53d9c5 S |
1677 | spin_unlock_irqrestore(&rp->lock, flags); |
1678 | } | |
e65cefe8 AB |
1679 | return 0; |
1680 | } | |
1681 | ||
49ad2fd7 | 1682 | int __kprobes register_kretprobe(struct kretprobe *rp) |
b94cce92 HN |
1683 | { |
1684 | int ret = 0; | |
1685 | struct kretprobe_instance *inst; | |
1686 | int i; | |
b2a5cd69 | 1687 | void *addr; |
f438d914 MH |
1688 | |
1689 | if (kretprobe_blacklist_size) { | |
b2a5cd69 MH |
1690 | addr = kprobe_addr(&rp->kp); |
1691 | if (!addr) | |
1692 | return -EINVAL; | |
f438d914 MH |
1693 | |
1694 | for (i = 0; kretprobe_blacklist[i].name != NULL; i++) { | |
1695 | if (kretprobe_blacklist[i].addr == addr) | |
1696 | return -EINVAL; | |
1697 | } | |
1698 | } | |
b94cce92 HN |
1699 | |
1700 | rp->kp.pre_handler = pre_handler_kretprobe; | |
7522a842 AM |
1701 | rp->kp.post_handler = NULL; |
1702 | rp->kp.fault_handler = NULL; | |
1703 | rp->kp.break_handler = NULL; | |
b94cce92 HN |
1704 | |
1705 | /* Pre-allocate memory for max kretprobe instances */ | |
1706 | if (rp->maxactive <= 0) { | |
1707 | #ifdef CONFIG_PREEMPT | |
c2ef6661 | 1708 | rp->maxactive = max_t(unsigned int, 10, 2*num_possible_cpus()); |
b94cce92 | 1709 | #else |
4dae560f | 1710 | rp->maxactive = num_possible_cpus(); |
b94cce92 HN |
1711 | #endif |
1712 | } | |
ef53d9c5 | 1713 | spin_lock_init(&rp->lock); |
b94cce92 HN |
1714 | INIT_HLIST_HEAD(&rp->free_instances); |
1715 | for (i = 0; i < rp->maxactive; i++) { | |
f47cd9b5 AS |
1716 | inst = kmalloc(sizeof(struct kretprobe_instance) + |
1717 | rp->data_size, GFP_KERNEL); | |
b94cce92 HN |
1718 | if (inst == NULL) { |
1719 | free_rp_inst(rp); | |
1720 | return -ENOMEM; | |
1721 | } | |
ef53d9c5 S |
1722 | INIT_HLIST_NODE(&inst->hlist); |
1723 | hlist_add_head(&inst->hlist, &rp->free_instances); | |
b94cce92 HN |
1724 | } |
1725 | ||
1726 | rp->nmissed = 0; | |
1727 | /* Establish function entry probe point */ | |
49ad2fd7 | 1728 | ret = register_kprobe(&rp->kp); |
4a296e07 | 1729 | if (ret != 0) |
b94cce92 HN |
1730 | free_rp_inst(rp); |
1731 | return ret; | |
1732 | } | |
99081ab5 | 1733 | EXPORT_SYMBOL_GPL(register_kretprobe); |
b94cce92 | 1734 | |
49ad2fd7 | 1735 | int __kprobes register_kretprobes(struct kretprobe **rps, int num) |
4a296e07 MH |
1736 | { |
1737 | int ret = 0, i; | |
1738 | ||
1739 | if (num <= 0) | |
1740 | return -EINVAL; | |
1741 | for (i = 0; i < num; i++) { | |
49ad2fd7 | 1742 | ret = register_kretprobe(rps[i]); |
67dddaad MH |
1743 | if (ret < 0) { |
1744 | if (i > 0) | |
1745 | unregister_kretprobes(rps, i); | |
4a296e07 MH |
1746 | break; |
1747 | } | |
1748 | } | |
1749 | return ret; | |
1750 | } | |
99081ab5 | 1751 | EXPORT_SYMBOL_GPL(register_kretprobes); |
4a296e07 | 1752 | |
4a296e07 MH |
1753 | void __kprobes unregister_kretprobe(struct kretprobe *rp) |
1754 | { | |
1755 | unregister_kretprobes(&rp, 1); | |
1756 | } | |
99081ab5 | 1757 | EXPORT_SYMBOL_GPL(unregister_kretprobe); |
4a296e07 | 1758 | |
4a296e07 MH |
1759 | void __kprobes unregister_kretprobes(struct kretprobe **rps, int num) |
1760 | { | |
1761 | int i; | |
1762 | ||
1763 | if (num <= 0) | |
1764 | return; | |
1765 | mutex_lock(&kprobe_mutex); | |
1766 | for (i = 0; i < num; i++) | |
1767 | if (__unregister_kprobe_top(&rps[i]->kp) < 0) | |
1768 | rps[i]->kp.addr = NULL; | |
1769 | mutex_unlock(&kprobe_mutex); | |
1770 | ||
1771 | synchronize_sched(); | |
1772 | for (i = 0; i < num; i++) { | |
1773 | if (rps[i]->kp.addr) { | |
1774 | __unregister_kprobe_bottom(&rps[i]->kp); | |
1775 | cleanup_rp_inst(rps[i]); | |
1776 | } | |
1777 | } | |
1778 | } | |
99081ab5 | 1779 | EXPORT_SYMBOL_GPL(unregister_kretprobes); |
4a296e07 | 1780 | |
9edddaa2 | 1781 | #else /* CONFIG_KRETPROBES */ |
d0aaff97 | 1782 | int __kprobes register_kretprobe(struct kretprobe *rp) |
b94cce92 HN |
1783 | { |
1784 | return -ENOSYS; | |
1785 | } | |
99081ab5 | 1786 | EXPORT_SYMBOL_GPL(register_kretprobe); |
b94cce92 | 1787 | |
4a296e07 | 1788 | int __kprobes register_kretprobes(struct kretprobe **rps, int num) |
346fd59b | 1789 | { |
4a296e07 | 1790 | return -ENOSYS; |
346fd59b | 1791 | } |
99081ab5 MH |
1792 | EXPORT_SYMBOL_GPL(register_kretprobes); |
1793 | ||
d0aaff97 | 1794 | void __kprobes unregister_kretprobe(struct kretprobe *rp) |
b94cce92 | 1795 | { |
4a296e07 | 1796 | } |
99081ab5 | 1797 | EXPORT_SYMBOL_GPL(unregister_kretprobe); |
b94cce92 | 1798 | |
4a296e07 MH |
1799 | void __kprobes unregister_kretprobes(struct kretprobe **rps, int num) |
1800 | { | |
1801 | } | |
99081ab5 | 1802 | EXPORT_SYMBOL_GPL(unregister_kretprobes); |
4c4308cb | 1803 | |
4a296e07 MH |
1804 | static int __kprobes pre_handler_kretprobe(struct kprobe *p, |
1805 | struct pt_regs *regs) | |
1806 | { | |
1807 | return 0; | |
b94cce92 HN |
1808 | } |
1809 | ||
4a296e07 MH |
1810 | #endif /* CONFIG_KRETPROBES */ |
1811 | ||
e8386a0c MH |
1812 | /* Set the kprobe gone and remove its instruction buffer. */ |
1813 | static void __kprobes kill_kprobe(struct kprobe *p) | |
1814 | { | |
1815 | struct kprobe *kp; | |
de5bd88d | 1816 | |
e8386a0c | 1817 | p->flags |= KPROBE_FLAG_GONE; |
afd66255 | 1818 | if (kprobe_aggrprobe(p)) { |
e8386a0c MH |
1819 | /* |
1820 | * If this is an aggr_kprobe, we have to list all the | |
1821 | * chained probes and mark them GONE. | |
1822 | */ | |
1823 | list_for_each_entry_rcu(kp, &p->list, list) | |
1824 | kp->flags |= KPROBE_FLAG_GONE; | |
1825 | p->post_handler = NULL; | |
1826 | p->break_handler = NULL; | |
afd66255 | 1827 | kill_optimized_kprobe(p); |
e8386a0c MH |
1828 | } |
1829 | /* | |
1830 | * Here, we can remove insn_slot safely, because no thread calls | |
1831 | * the original probed function (which will be freed soon) any more. | |
1832 | */ | |
1833 | arch_remove_kprobe(p); | |
1834 | } | |
1835 | ||
c0614829 MH |
1836 | /* Disable one kprobe */ |
1837 | int __kprobes disable_kprobe(struct kprobe *kp) | |
1838 | { | |
1839 | int ret = 0; | |
c0614829 MH |
1840 | |
1841 | mutex_lock(&kprobe_mutex); | |
1842 | ||
6f0f1dd7 MH |
1843 | /* Disable this kprobe */ |
1844 | if (__disable_kprobe(kp) == NULL) | |
c0614829 | 1845 | ret = -EINVAL; |
c0614829 | 1846 | |
c0614829 MH |
1847 | mutex_unlock(&kprobe_mutex); |
1848 | return ret; | |
1849 | } | |
1850 | EXPORT_SYMBOL_GPL(disable_kprobe); | |
1851 | ||
1852 | /* Enable one kprobe */ | |
1853 | int __kprobes enable_kprobe(struct kprobe *kp) | |
1854 | { | |
1855 | int ret = 0; | |
1856 | struct kprobe *p; | |
1857 | ||
1858 | mutex_lock(&kprobe_mutex); | |
1859 | ||
1860 | /* Check whether specified probe is valid. */ | |
1861 | p = __get_valid_kprobe(kp); | |
1862 | if (unlikely(p == NULL)) { | |
1863 | ret = -EINVAL; | |
1864 | goto out; | |
1865 | } | |
1866 | ||
1867 | if (kprobe_gone(kp)) { | |
1868 | /* This kprobe has gone, we couldn't enable it. */ | |
1869 | ret = -EINVAL; | |
1870 | goto out; | |
1871 | } | |
1872 | ||
1873 | if (p != kp) | |
1874 | kp->flags &= ~KPROBE_FLAG_DISABLED; | |
1875 | ||
1876 | if (!kprobes_all_disarmed && kprobe_disabled(p)) { | |
1877 | p->flags &= ~KPROBE_FLAG_DISABLED; | |
1878 | arm_kprobe(p); | |
1879 | } | |
1880 | out: | |
1881 | mutex_unlock(&kprobe_mutex); | |
1882 | return ret; | |
1883 | } | |
1884 | EXPORT_SYMBOL_GPL(enable_kprobe); | |
1885 | ||
24851d24 FW |
1886 | void __kprobes dump_kprobe(struct kprobe *kp) |
1887 | { | |
1888 | printk(KERN_WARNING "Dumping kprobe:\n"); | |
1889 | printk(KERN_WARNING "Name: %s\nAddress: %p\nOffset: %x\n", | |
1890 | kp->symbol_name, kp->addr, kp->offset); | |
1891 | } | |
1892 | ||
e8386a0c MH |
1893 | /* Module notifier call back, checking kprobes on the module */ |
1894 | static int __kprobes kprobes_module_callback(struct notifier_block *nb, | |
1895 | unsigned long val, void *data) | |
1896 | { | |
1897 | struct module *mod = data; | |
1898 | struct hlist_head *head; | |
1899 | struct hlist_node *node; | |
1900 | struct kprobe *p; | |
1901 | unsigned int i; | |
f24659d9 | 1902 | int checkcore = (val == MODULE_STATE_GOING); |
e8386a0c | 1903 | |
f24659d9 | 1904 | if (val != MODULE_STATE_GOING && val != MODULE_STATE_LIVE) |
e8386a0c MH |
1905 | return NOTIFY_DONE; |
1906 | ||
1907 | /* | |
f24659d9 MH |
1908 | * When MODULE_STATE_GOING was notified, both of module .text and |
1909 | * .init.text sections would be freed. When MODULE_STATE_LIVE was | |
1910 | * notified, only .init.text section would be freed. We need to | |
1911 | * disable kprobes which have been inserted in the sections. | |
e8386a0c MH |
1912 | */ |
1913 | mutex_lock(&kprobe_mutex); | |
1914 | for (i = 0; i < KPROBE_TABLE_SIZE; i++) { | |
1915 | head = &kprobe_table[i]; | |
1916 | hlist_for_each_entry_rcu(p, node, head, hlist) | |
f24659d9 MH |
1917 | if (within_module_init((unsigned long)p->addr, mod) || |
1918 | (checkcore && | |
1919 | within_module_core((unsigned long)p->addr, mod))) { | |
e8386a0c MH |
1920 | /* |
1921 | * The vaddr this probe is installed will soon | |
1922 | * be vfreed buy not synced to disk. Hence, | |
1923 | * disarming the breakpoint isn't needed. | |
1924 | */ | |
1925 | kill_kprobe(p); | |
1926 | } | |
1927 | } | |
1928 | mutex_unlock(&kprobe_mutex); | |
1929 | return NOTIFY_DONE; | |
1930 | } | |
1931 | ||
1932 | static struct notifier_block kprobe_module_nb = { | |
1933 | .notifier_call = kprobes_module_callback, | |
1934 | .priority = 0 | |
1935 | }; | |
1936 | ||
1da177e4 LT |
1937 | static int __init init_kprobes(void) |
1938 | { | |
1939 | int i, err = 0; | |
3d8d996e SD |
1940 | unsigned long offset = 0, size = 0; |
1941 | char *modname, namebuf[128]; | |
1942 | const char *symbol_name; | |
1943 | void *addr; | |
1944 | struct kprobe_blackpoint *kb; | |
1da177e4 LT |
1945 | |
1946 | /* FIXME allocate the probe table, currently defined statically */ | |
1947 | /* initialize all list heads */ | |
b94cce92 | 1948 | for (i = 0; i < KPROBE_TABLE_SIZE; i++) { |
1da177e4 | 1949 | INIT_HLIST_HEAD(&kprobe_table[i]); |
b94cce92 | 1950 | INIT_HLIST_HEAD(&kretprobe_inst_table[i]); |
ef53d9c5 | 1951 | spin_lock_init(&(kretprobe_table_locks[i].lock)); |
b94cce92 | 1952 | } |
1da177e4 | 1953 | |
3d8d996e SD |
1954 | /* |
1955 | * Lookup and populate the kprobe_blacklist. | |
1956 | * | |
1957 | * Unlike the kretprobe blacklist, we'll need to determine | |
1958 | * the range of addresses that belong to the said functions, | |
1959 | * since a kprobe need not necessarily be at the beginning | |
1960 | * of a function. | |
1961 | */ | |
1962 | for (kb = kprobe_blacklist; kb->name != NULL; kb++) { | |
1963 | kprobe_lookup_name(kb->name, addr); | |
1964 | if (!addr) | |
1965 | continue; | |
1966 | ||
1967 | kb->start_addr = (unsigned long)addr; | |
1968 | symbol_name = kallsyms_lookup(kb->start_addr, | |
1969 | &size, &offset, &modname, namebuf); | |
1970 | if (!symbol_name) | |
1971 | kb->range = 0; | |
1972 | else | |
1973 | kb->range = size; | |
1974 | } | |
1975 | ||
f438d914 MH |
1976 | if (kretprobe_blacklist_size) { |
1977 | /* lookup the function address from its name */ | |
1978 | for (i = 0; kretprobe_blacklist[i].name != NULL; i++) { | |
1979 | kprobe_lookup_name(kretprobe_blacklist[i].name, | |
1980 | kretprobe_blacklist[i].addr); | |
1981 | if (!kretprobe_blacklist[i].addr) | |
1982 | printk("kretprobe: lookup failed: %s\n", | |
1983 | kretprobe_blacklist[i].name); | |
1984 | } | |
1985 | } | |
1986 | ||
b2be84df MH |
1987 | #if defined(CONFIG_OPTPROBES) |
1988 | #if defined(__ARCH_WANT_KPROBES_INSN_SLOT) | |
afd66255 MH |
1989 | /* Init kprobe_optinsn_slots */ |
1990 | kprobe_optinsn_slots.insn_size = MAX_OPTINSN_SIZE; | |
1991 | #endif | |
b2be84df MH |
1992 | /* By default, kprobes can be optimized */ |
1993 | kprobes_allow_optimization = true; | |
1994 | #endif | |
afd66255 | 1995 | |
e579abeb MH |
1996 | /* By default, kprobes are armed */ |
1997 | kprobes_all_disarmed = false; | |
bf8f6e5b | 1998 | |
6772926b | 1999 | err = arch_init_kprobes(); |
802eae7c RL |
2000 | if (!err) |
2001 | err = register_die_notifier(&kprobe_exceptions_nb); | |
e8386a0c MH |
2002 | if (!err) |
2003 | err = register_module_notifier(&kprobe_module_nb); | |
2004 | ||
ef53d9c5 | 2005 | kprobes_initialized = (err == 0); |
802eae7c | 2006 | |
8c1c9356 AM |
2007 | if (!err) |
2008 | init_test_probes(); | |
1da177e4 LT |
2009 | return err; |
2010 | } | |
2011 | ||
346fd59b SD |
2012 | #ifdef CONFIG_DEBUG_FS |
2013 | static void __kprobes report_probe(struct seq_file *pi, struct kprobe *p, | |
afd66255 | 2014 | const char *sym, int offset, char *modname, struct kprobe *pp) |
346fd59b SD |
2015 | { |
2016 | char *kprobe_type; | |
2017 | ||
2018 | if (p->pre_handler == pre_handler_kretprobe) | |
2019 | kprobe_type = "r"; | |
2020 | else if (p->pre_handler == setjmp_pre_handler) | |
2021 | kprobe_type = "j"; | |
2022 | else | |
2023 | kprobe_type = "k"; | |
afd66255 | 2024 | |
346fd59b | 2025 | if (sym) |
afd66255 | 2026 | seq_printf(pi, "%p %s %s+0x%x %s ", |
de5bd88d | 2027 | p->addr, kprobe_type, sym, offset, |
afd66255 | 2028 | (modname ? modname : " ")); |
346fd59b | 2029 | else |
afd66255 MH |
2030 | seq_printf(pi, "%p %s %p ", |
2031 | p->addr, kprobe_type, p->addr); | |
2032 | ||
2033 | if (!pp) | |
2034 | pp = p; | |
2035 | seq_printf(pi, "%s%s%s\n", | |
2036 | (kprobe_gone(p) ? "[GONE]" : ""), | |
2037 | ((kprobe_disabled(p) && !kprobe_gone(p)) ? "[DISABLED]" : ""), | |
2038 | (kprobe_optimized(pp) ? "[OPTIMIZED]" : "")); | |
346fd59b SD |
2039 | } |
2040 | ||
2041 | static void __kprobes *kprobe_seq_start(struct seq_file *f, loff_t *pos) | |
2042 | { | |
2043 | return (*pos < KPROBE_TABLE_SIZE) ? pos : NULL; | |
2044 | } | |
2045 | ||
2046 | static void __kprobes *kprobe_seq_next(struct seq_file *f, void *v, loff_t *pos) | |
2047 | { | |
2048 | (*pos)++; | |
2049 | if (*pos >= KPROBE_TABLE_SIZE) | |
2050 | return NULL; | |
2051 | return pos; | |
2052 | } | |
2053 | ||
2054 | static void __kprobes kprobe_seq_stop(struct seq_file *f, void *v) | |
2055 | { | |
2056 | /* Nothing to do */ | |
2057 | } | |
2058 | ||
2059 | static int __kprobes show_kprobe_addr(struct seq_file *pi, void *v) | |
2060 | { | |
2061 | struct hlist_head *head; | |
2062 | struct hlist_node *node; | |
2063 | struct kprobe *p, *kp; | |
2064 | const char *sym = NULL; | |
2065 | unsigned int i = *(loff_t *) v; | |
ffb45122 | 2066 | unsigned long offset = 0; |
346fd59b SD |
2067 | char *modname, namebuf[128]; |
2068 | ||
2069 | head = &kprobe_table[i]; | |
2070 | preempt_disable(); | |
2071 | hlist_for_each_entry_rcu(p, node, head, hlist) { | |
ffb45122 | 2072 | sym = kallsyms_lookup((unsigned long)p->addr, NULL, |
346fd59b | 2073 | &offset, &modname, namebuf); |
afd66255 | 2074 | if (kprobe_aggrprobe(p)) { |
346fd59b | 2075 | list_for_each_entry_rcu(kp, &p->list, list) |
afd66255 | 2076 | report_probe(pi, kp, sym, offset, modname, p); |
346fd59b | 2077 | } else |
afd66255 | 2078 | report_probe(pi, p, sym, offset, modname, NULL); |
346fd59b SD |
2079 | } |
2080 | preempt_enable(); | |
2081 | return 0; | |
2082 | } | |
2083 | ||
88e9d34c | 2084 | static const struct seq_operations kprobes_seq_ops = { |
346fd59b SD |
2085 | .start = kprobe_seq_start, |
2086 | .next = kprobe_seq_next, | |
2087 | .stop = kprobe_seq_stop, | |
2088 | .show = show_kprobe_addr | |
2089 | }; | |
2090 | ||
2091 | static int __kprobes kprobes_open(struct inode *inode, struct file *filp) | |
2092 | { | |
2093 | return seq_open(filp, &kprobes_seq_ops); | |
2094 | } | |
2095 | ||
828c0950 | 2096 | static const struct file_operations debugfs_kprobes_operations = { |
346fd59b SD |
2097 | .open = kprobes_open, |
2098 | .read = seq_read, | |
2099 | .llseek = seq_lseek, | |
2100 | .release = seq_release, | |
2101 | }; | |
2102 | ||
e579abeb | 2103 | static void __kprobes arm_all_kprobes(void) |
bf8f6e5b AM |
2104 | { |
2105 | struct hlist_head *head; | |
2106 | struct hlist_node *node; | |
2107 | struct kprobe *p; | |
2108 | unsigned int i; | |
2109 | ||
2110 | mutex_lock(&kprobe_mutex); | |
2111 | ||
e579abeb MH |
2112 | /* If kprobes are armed, just return */ |
2113 | if (!kprobes_all_disarmed) | |
bf8f6e5b AM |
2114 | goto already_enabled; |
2115 | ||
afd66255 | 2116 | /* Arming kprobes doesn't optimize kprobe itself */ |
4460fdad | 2117 | mutex_lock(&text_mutex); |
bf8f6e5b AM |
2118 | for (i = 0; i < KPROBE_TABLE_SIZE; i++) { |
2119 | head = &kprobe_table[i]; | |
2120 | hlist_for_each_entry_rcu(p, node, head, hlist) | |
de5bd88d | 2121 | if (!kprobe_disabled(p)) |
afd66255 | 2122 | __arm_kprobe(p); |
bf8f6e5b | 2123 | } |
4460fdad | 2124 | mutex_unlock(&text_mutex); |
bf8f6e5b | 2125 | |
e579abeb | 2126 | kprobes_all_disarmed = false; |
bf8f6e5b AM |
2127 | printk(KERN_INFO "Kprobes globally enabled\n"); |
2128 | ||
2129 | already_enabled: | |
2130 | mutex_unlock(&kprobe_mutex); | |
2131 | return; | |
2132 | } | |
2133 | ||
e579abeb | 2134 | static void __kprobes disarm_all_kprobes(void) |
bf8f6e5b AM |
2135 | { |
2136 | struct hlist_head *head; | |
2137 | struct hlist_node *node; | |
2138 | struct kprobe *p; | |
2139 | unsigned int i; | |
2140 | ||
2141 | mutex_lock(&kprobe_mutex); | |
2142 | ||
e579abeb | 2143 | /* If kprobes are already disarmed, just return */ |
6274de49 MH |
2144 | if (kprobes_all_disarmed) { |
2145 | mutex_unlock(&kprobe_mutex); | |
2146 | return; | |
2147 | } | |
bf8f6e5b | 2148 | |
e579abeb | 2149 | kprobes_all_disarmed = true; |
bf8f6e5b | 2150 | printk(KERN_INFO "Kprobes globally disabled\n"); |
afd66255 | 2151 | |
4460fdad | 2152 | mutex_lock(&text_mutex); |
bf8f6e5b AM |
2153 | for (i = 0; i < KPROBE_TABLE_SIZE; i++) { |
2154 | head = &kprobe_table[i]; | |
2155 | hlist_for_each_entry_rcu(p, node, head, hlist) { | |
de5bd88d | 2156 | if (!arch_trampoline_kprobe(p) && !kprobe_disabled(p)) |
6274de49 | 2157 | __disarm_kprobe(p, false); |
bf8f6e5b AM |
2158 | } |
2159 | } | |
4460fdad | 2160 | mutex_unlock(&text_mutex); |
bf8f6e5b | 2161 | mutex_unlock(&kprobe_mutex); |
bf8f6e5b | 2162 | |
6274de49 MH |
2163 | /* Wait for disarming all kprobes by optimizer */ |
2164 | wait_for_kprobe_optimizer(); | |
bf8f6e5b AM |
2165 | } |
2166 | ||
2167 | /* | |
2168 | * XXX: The debugfs bool file interface doesn't allow for callbacks | |
2169 | * when the bool state is switched. We can reuse that facility when | |
2170 | * available | |
2171 | */ | |
2172 | static ssize_t read_enabled_file_bool(struct file *file, | |
2173 | char __user *user_buf, size_t count, loff_t *ppos) | |
2174 | { | |
2175 | char buf[3]; | |
2176 | ||
e579abeb | 2177 | if (!kprobes_all_disarmed) |
bf8f6e5b AM |
2178 | buf[0] = '1'; |
2179 | else | |
2180 | buf[0] = '0'; | |
2181 | buf[1] = '\n'; | |
2182 | buf[2] = 0x00; | |
2183 | return simple_read_from_buffer(user_buf, count, ppos, buf, 2); | |
2184 | } | |
2185 | ||
2186 | static ssize_t write_enabled_file_bool(struct file *file, | |
2187 | const char __user *user_buf, size_t count, loff_t *ppos) | |
2188 | { | |
2189 | char buf[32]; | |
2190 | int buf_size; | |
2191 | ||
2192 | buf_size = min(count, (sizeof(buf)-1)); | |
2193 | if (copy_from_user(buf, user_buf, buf_size)) | |
2194 | return -EFAULT; | |
2195 | ||
2196 | switch (buf[0]) { | |
2197 | case 'y': | |
2198 | case 'Y': | |
2199 | case '1': | |
e579abeb | 2200 | arm_all_kprobes(); |
bf8f6e5b AM |
2201 | break; |
2202 | case 'n': | |
2203 | case 'N': | |
2204 | case '0': | |
e579abeb | 2205 | disarm_all_kprobes(); |
bf8f6e5b AM |
2206 | break; |
2207 | } | |
2208 | ||
2209 | return count; | |
2210 | } | |
2211 | ||
828c0950 | 2212 | static const struct file_operations fops_kp = { |
bf8f6e5b AM |
2213 | .read = read_enabled_file_bool, |
2214 | .write = write_enabled_file_bool, | |
6038f373 | 2215 | .llseek = default_llseek, |
bf8f6e5b AM |
2216 | }; |
2217 | ||
346fd59b SD |
2218 | static int __kprobes debugfs_kprobe_init(void) |
2219 | { | |
2220 | struct dentry *dir, *file; | |
bf8f6e5b | 2221 | unsigned int value = 1; |
346fd59b SD |
2222 | |
2223 | dir = debugfs_create_dir("kprobes", NULL); | |
2224 | if (!dir) | |
2225 | return -ENOMEM; | |
2226 | ||
e3869792 | 2227 | file = debugfs_create_file("list", 0444, dir, NULL, |
346fd59b SD |
2228 | &debugfs_kprobes_operations); |
2229 | if (!file) { | |
2230 | debugfs_remove(dir); | |
2231 | return -ENOMEM; | |
2232 | } | |
2233 | ||
bf8f6e5b AM |
2234 | file = debugfs_create_file("enabled", 0600, dir, |
2235 | &value, &fops_kp); | |
2236 | if (!file) { | |
2237 | debugfs_remove(dir); | |
2238 | return -ENOMEM; | |
2239 | } | |
2240 | ||
346fd59b SD |
2241 | return 0; |
2242 | } | |
2243 | ||
2244 | late_initcall(debugfs_kprobe_init); | |
2245 | #endif /* CONFIG_DEBUG_FS */ | |
2246 | ||
2247 | module_init(init_kprobes); | |
1da177e4 | 2248 | |
99081ab5 | 2249 | /* defined in arch/.../kernel/kprobes.c */ |
1da177e4 | 2250 | EXPORT_SYMBOL_GPL(jprobe_return); |