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> | |
1eeb66a1 | 45 | #include <linux/kdebug.h> |
4460fdad | 46 | #include <linux/memory.h> |
bf8f6e5b | 47 | |
d0aaff97 | 48 | #include <asm-generic/sections.h> |
1da177e4 LT |
49 | #include <asm/cacheflush.h> |
50 | #include <asm/errno.h> | |
bf8f6e5b | 51 | #include <asm/uaccess.h> |
1da177e4 LT |
52 | |
53 | #define KPROBE_HASH_BITS 6 | |
54 | #define KPROBE_TABLE_SIZE (1 << KPROBE_HASH_BITS) | |
55 | ||
3a872d89 AM |
56 | |
57 | /* | |
58 | * Some oddball architectures like 64bit powerpc have function descriptors | |
59 | * so this must be overridable. | |
60 | */ | |
61 | #ifndef kprobe_lookup_name | |
62 | #define kprobe_lookup_name(name, addr) \ | |
63 | addr = ((kprobe_opcode_t *)(kallsyms_lookup_name(name))) | |
64 | #endif | |
65 | ||
ef53d9c5 | 66 | static int kprobes_initialized; |
1da177e4 | 67 | static struct hlist_head kprobe_table[KPROBE_TABLE_SIZE]; |
b94cce92 | 68 | static struct hlist_head kretprobe_inst_table[KPROBE_TABLE_SIZE]; |
1da177e4 | 69 | |
bf8f6e5b AM |
70 | /* NOTE: change this value only with kprobe_mutex held */ |
71 | static bool kprobe_enabled; | |
72 | ||
12941560 | 73 | static DEFINE_MUTEX(kprobe_mutex); /* Protects kprobe_table */ |
e6584523 | 74 | static DEFINE_PER_CPU(struct kprobe *, kprobe_instance) = NULL; |
ef53d9c5 | 75 | static struct { |
7e036d04 | 76 | spinlock_t lock ____cacheline_aligned_in_smp; |
ef53d9c5 S |
77 | } kretprobe_table_locks[KPROBE_TABLE_SIZE]; |
78 | ||
79 | static spinlock_t *kretprobe_table_lock_ptr(unsigned long hash) | |
80 | { | |
81 | return &(kretprobe_table_locks[hash].lock); | |
82 | } | |
1da177e4 | 83 | |
3d8d996e SD |
84 | /* |
85 | * Normally, functions that we'd want to prohibit kprobes in, are marked | |
86 | * __kprobes. But, there are cases where such functions already belong to | |
87 | * a different section (__sched for preempt_schedule) | |
88 | * | |
89 | * For such cases, we now have a blacklist | |
90 | */ | |
544304b2 | 91 | static struct kprobe_blackpoint kprobe_blacklist[] = { |
3d8d996e SD |
92 | {"preempt_schedule",}, |
93 | {NULL} /* Terminator */ | |
94 | }; | |
95 | ||
2d14e39d | 96 | #ifdef __ARCH_WANT_KPROBES_INSN_SLOT |
9ec4b1f3 AM |
97 | /* |
98 | * kprobe->ainsn.insn points to the copy of the instruction to be | |
99 | * single-stepped. x86_64, POWER4 and above have no-exec support and | |
100 | * stepping on the instruction on a vmalloced/kmalloced/data page | |
101 | * is a recipe for disaster | |
102 | */ | |
103 | #define INSNS_PER_PAGE (PAGE_SIZE/(MAX_INSN_SIZE * sizeof(kprobe_opcode_t))) | |
104 | ||
105 | struct kprobe_insn_page { | |
106 | struct hlist_node hlist; | |
107 | kprobe_opcode_t *insns; /* Page of instruction slots */ | |
108 | char slot_used[INSNS_PER_PAGE]; | |
109 | int nused; | |
b4c6c34a | 110 | int ngarbage; |
9ec4b1f3 AM |
111 | }; |
112 | ||
ab40c5c6 MH |
113 | enum kprobe_slot_state { |
114 | SLOT_CLEAN = 0, | |
115 | SLOT_DIRTY = 1, | |
116 | SLOT_USED = 2, | |
117 | }; | |
118 | ||
12941560 | 119 | static DEFINE_MUTEX(kprobe_insn_mutex); /* Protects kprobe_insn_pages */ |
9ec4b1f3 | 120 | static struct hlist_head kprobe_insn_pages; |
b4c6c34a MH |
121 | static int kprobe_garbage_slots; |
122 | static int collect_garbage_slots(void); | |
123 | ||
124 | static int __kprobes check_safety(void) | |
125 | { | |
126 | int ret = 0; | |
5a4ccaf3 | 127 | #if defined(CONFIG_PREEMPT) && defined(CONFIG_FREEZER) |
b4c6c34a MH |
128 | ret = freeze_processes(); |
129 | if (ret == 0) { | |
130 | struct task_struct *p, *q; | |
131 | do_each_thread(p, q) { | |
132 | if (p != current && p->state == TASK_RUNNING && | |
133 | p->pid != 0) { | |
134 | printk("Check failed: %s is running\n",p->comm); | |
135 | ret = -1; | |
136 | goto loop_end; | |
137 | } | |
138 | } while_each_thread(p, q); | |
139 | } | |
140 | loop_end: | |
141 | thaw_processes(); | |
142 | #else | |
143 | synchronize_sched(); | |
144 | #endif | |
145 | return ret; | |
146 | } | |
9ec4b1f3 AM |
147 | |
148 | /** | |
12941560 | 149 | * __get_insn_slot() - Find a slot on an executable page for an instruction. |
9ec4b1f3 AM |
150 | * We allocate an executable page if there's no room on existing ones. |
151 | */ | |
12941560 | 152 | static kprobe_opcode_t __kprobes *__get_insn_slot(void) |
9ec4b1f3 AM |
153 | { |
154 | struct kprobe_insn_page *kip; | |
155 | struct hlist_node *pos; | |
156 | ||
6f716acd | 157 | retry: |
b0bb5016 | 158 | hlist_for_each_entry(kip, pos, &kprobe_insn_pages, hlist) { |
9ec4b1f3 AM |
159 | if (kip->nused < INSNS_PER_PAGE) { |
160 | int i; | |
161 | for (i = 0; i < INSNS_PER_PAGE; i++) { | |
ab40c5c6 MH |
162 | if (kip->slot_used[i] == SLOT_CLEAN) { |
163 | kip->slot_used[i] = SLOT_USED; | |
9ec4b1f3 AM |
164 | kip->nused++; |
165 | return kip->insns + (i * MAX_INSN_SIZE); | |
166 | } | |
167 | } | |
168 | /* Surprise! No unused slots. Fix kip->nused. */ | |
169 | kip->nused = INSNS_PER_PAGE; | |
170 | } | |
171 | } | |
172 | ||
b4c6c34a MH |
173 | /* If there are any garbage slots, collect it and try again. */ |
174 | if (kprobe_garbage_slots && collect_garbage_slots() == 0) { | |
175 | goto retry; | |
176 | } | |
177 | /* All out of space. Need to allocate a new page. Use slot 0. */ | |
9ec4b1f3 | 178 | kip = kmalloc(sizeof(struct kprobe_insn_page), GFP_KERNEL); |
6f716acd | 179 | if (!kip) |
9ec4b1f3 | 180 | return NULL; |
9ec4b1f3 AM |
181 | |
182 | /* | |
183 | * Use module_alloc so this page is within +/- 2GB of where the | |
184 | * kernel image and loaded module images reside. This is required | |
185 | * so x86_64 can correctly handle the %rip-relative fixups. | |
186 | */ | |
187 | kip->insns = module_alloc(PAGE_SIZE); | |
188 | if (!kip->insns) { | |
189 | kfree(kip); | |
190 | return NULL; | |
191 | } | |
192 | INIT_HLIST_NODE(&kip->hlist); | |
193 | hlist_add_head(&kip->hlist, &kprobe_insn_pages); | |
ab40c5c6 MH |
194 | memset(kip->slot_used, SLOT_CLEAN, INSNS_PER_PAGE); |
195 | kip->slot_used[0] = SLOT_USED; | |
9ec4b1f3 | 196 | kip->nused = 1; |
b4c6c34a | 197 | kip->ngarbage = 0; |
9ec4b1f3 AM |
198 | return kip->insns; |
199 | } | |
200 | ||
12941560 MH |
201 | kprobe_opcode_t __kprobes *get_insn_slot(void) |
202 | { | |
203 | kprobe_opcode_t *ret; | |
204 | mutex_lock(&kprobe_insn_mutex); | |
205 | ret = __get_insn_slot(); | |
206 | mutex_unlock(&kprobe_insn_mutex); | |
207 | return ret; | |
208 | } | |
209 | ||
b4c6c34a MH |
210 | /* Return 1 if all garbages are collected, otherwise 0. */ |
211 | static int __kprobes collect_one_slot(struct kprobe_insn_page *kip, int idx) | |
212 | { | |
ab40c5c6 | 213 | kip->slot_used[idx] = SLOT_CLEAN; |
b4c6c34a MH |
214 | kip->nused--; |
215 | if (kip->nused == 0) { | |
216 | /* | |
217 | * Page is no longer in use. Free it unless | |
218 | * it's the last one. We keep the last one | |
219 | * so as not to have to set it up again the | |
220 | * next time somebody inserts a probe. | |
221 | */ | |
222 | hlist_del(&kip->hlist); | |
223 | if (hlist_empty(&kprobe_insn_pages)) { | |
224 | INIT_HLIST_NODE(&kip->hlist); | |
225 | hlist_add_head(&kip->hlist, | |
226 | &kprobe_insn_pages); | |
227 | } else { | |
228 | module_free(NULL, kip->insns); | |
229 | kfree(kip); | |
230 | } | |
231 | return 1; | |
232 | } | |
233 | return 0; | |
234 | } | |
235 | ||
236 | static int __kprobes collect_garbage_slots(void) | |
237 | { | |
238 | struct kprobe_insn_page *kip; | |
239 | struct hlist_node *pos, *next; | |
12941560 | 240 | int safety; |
b4c6c34a MH |
241 | |
242 | /* Ensure no-one is preepmted on the garbages */ | |
12941560 MH |
243 | mutex_unlock(&kprobe_insn_mutex); |
244 | safety = check_safety(); | |
245 | mutex_lock(&kprobe_insn_mutex); | |
246 | if (safety != 0) | |
b4c6c34a MH |
247 | return -EAGAIN; |
248 | ||
b0bb5016 | 249 | hlist_for_each_entry_safe(kip, pos, next, &kprobe_insn_pages, hlist) { |
b4c6c34a | 250 | int i; |
b4c6c34a MH |
251 | if (kip->ngarbage == 0) |
252 | continue; | |
253 | kip->ngarbage = 0; /* we will collect all garbages */ | |
254 | for (i = 0; i < INSNS_PER_PAGE; i++) { | |
ab40c5c6 | 255 | if (kip->slot_used[i] == SLOT_DIRTY && |
b4c6c34a MH |
256 | collect_one_slot(kip, i)) |
257 | break; | |
258 | } | |
259 | } | |
260 | kprobe_garbage_slots = 0; | |
261 | return 0; | |
262 | } | |
263 | ||
264 | void __kprobes free_insn_slot(kprobe_opcode_t * slot, int dirty) | |
9ec4b1f3 AM |
265 | { |
266 | struct kprobe_insn_page *kip; | |
267 | struct hlist_node *pos; | |
268 | ||
12941560 | 269 | mutex_lock(&kprobe_insn_mutex); |
b0bb5016 | 270 | hlist_for_each_entry(kip, pos, &kprobe_insn_pages, hlist) { |
9ec4b1f3 AM |
271 | if (kip->insns <= slot && |
272 | slot < kip->insns + (INSNS_PER_PAGE * MAX_INSN_SIZE)) { | |
273 | int i = (slot - kip->insns) / MAX_INSN_SIZE; | |
b4c6c34a | 274 | if (dirty) { |
ab40c5c6 | 275 | kip->slot_used[i] = SLOT_DIRTY; |
b4c6c34a MH |
276 | kip->ngarbage++; |
277 | } else { | |
278 | collect_one_slot(kip, i); | |
9ec4b1f3 | 279 | } |
b4c6c34a | 280 | break; |
9ec4b1f3 AM |
281 | } |
282 | } | |
6f716acd CH |
283 | |
284 | if (dirty && ++kprobe_garbage_slots > INSNS_PER_PAGE) | |
b4c6c34a | 285 | collect_garbage_slots(); |
12941560 MH |
286 | |
287 | mutex_unlock(&kprobe_insn_mutex); | |
9ec4b1f3 | 288 | } |
2d14e39d | 289 | #endif |
9ec4b1f3 | 290 | |
e6584523 AM |
291 | /* We have preemption disabled.. so it is safe to use __ versions */ |
292 | static inline void set_kprobe_instance(struct kprobe *kp) | |
293 | { | |
294 | __get_cpu_var(kprobe_instance) = kp; | |
295 | } | |
296 | ||
297 | static inline void reset_kprobe_instance(void) | |
298 | { | |
299 | __get_cpu_var(kprobe_instance) = NULL; | |
300 | } | |
301 | ||
3516a460 AM |
302 | /* |
303 | * This routine is called either: | |
49a2a1b8 | 304 | * - under the kprobe_mutex - during kprobe_[un]register() |
3516a460 | 305 | * OR |
d217d545 | 306 | * - with preemption disabled - from arch/xxx/kernel/kprobes.c |
3516a460 | 307 | */ |
d0aaff97 | 308 | struct kprobe __kprobes *get_kprobe(void *addr) |
1da177e4 LT |
309 | { |
310 | struct hlist_head *head; | |
311 | struct hlist_node *node; | |
3516a460 | 312 | struct kprobe *p; |
1da177e4 LT |
313 | |
314 | head = &kprobe_table[hash_ptr(addr, KPROBE_HASH_BITS)]; | |
3516a460 | 315 | hlist_for_each_entry_rcu(p, node, head, hlist) { |
1da177e4 LT |
316 | if (p->addr == addr) |
317 | return p; | |
318 | } | |
319 | return NULL; | |
320 | } | |
321 | ||
64f562c6 AM |
322 | /* |
323 | * Aggregate handlers for multiple kprobes support - these handlers | |
324 | * take care of invoking the individual kprobe handlers on p->list | |
325 | */ | |
d0aaff97 | 326 | static int __kprobes aggr_pre_handler(struct kprobe *p, struct pt_regs *regs) |
64f562c6 AM |
327 | { |
328 | struct kprobe *kp; | |
329 | ||
3516a460 | 330 | list_for_each_entry_rcu(kp, &p->list, list) { |
e8386a0c | 331 | if (kp->pre_handler && !kprobe_gone(kp)) { |
e6584523 | 332 | set_kprobe_instance(kp); |
8b0914ea PP |
333 | if (kp->pre_handler(kp, regs)) |
334 | return 1; | |
64f562c6 | 335 | } |
e6584523 | 336 | reset_kprobe_instance(); |
64f562c6 AM |
337 | } |
338 | return 0; | |
339 | } | |
340 | ||
d0aaff97 PP |
341 | static void __kprobes aggr_post_handler(struct kprobe *p, struct pt_regs *regs, |
342 | unsigned long flags) | |
64f562c6 AM |
343 | { |
344 | struct kprobe *kp; | |
345 | ||
3516a460 | 346 | list_for_each_entry_rcu(kp, &p->list, list) { |
e8386a0c | 347 | if (kp->post_handler && !kprobe_gone(kp)) { |
e6584523 | 348 | set_kprobe_instance(kp); |
64f562c6 | 349 | kp->post_handler(kp, regs, flags); |
e6584523 | 350 | reset_kprobe_instance(); |
64f562c6 AM |
351 | } |
352 | } | |
64f562c6 AM |
353 | } |
354 | ||
d0aaff97 PP |
355 | static int __kprobes aggr_fault_handler(struct kprobe *p, struct pt_regs *regs, |
356 | int trapnr) | |
64f562c6 | 357 | { |
e6584523 AM |
358 | struct kprobe *cur = __get_cpu_var(kprobe_instance); |
359 | ||
64f562c6 AM |
360 | /* |
361 | * if we faulted "during" the execution of a user specified | |
362 | * probe handler, invoke just that probe's fault handler | |
363 | */ | |
e6584523 AM |
364 | if (cur && cur->fault_handler) { |
365 | if (cur->fault_handler(cur, regs, trapnr)) | |
64f562c6 AM |
366 | return 1; |
367 | } | |
368 | return 0; | |
369 | } | |
370 | ||
d0aaff97 | 371 | static int __kprobes aggr_break_handler(struct kprobe *p, struct pt_regs *regs) |
8b0914ea | 372 | { |
e6584523 AM |
373 | struct kprobe *cur = __get_cpu_var(kprobe_instance); |
374 | int ret = 0; | |
375 | ||
376 | if (cur && cur->break_handler) { | |
377 | if (cur->break_handler(cur, regs)) | |
378 | ret = 1; | |
8b0914ea | 379 | } |
e6584523 AM |
380 | reset_kprobe_instance(); |
381 | return ret; | |
8b0914ea PP |
382 | } |
383 | ||
bf8d5c52 KA |
384 | /* Walks the list and increments nmissed count for multiprobe case */ |
385 | void __kprobes kprobes_inc_nmissed_count(struct kprobe *p) | |
386 | { | |
387 | struct kprobe *kp; | |
388 | if (p->pre_handler != aggr_pre_handler) { | |
389 | p->nmissed++; | |
390 | } else { | |
391 | list_for_each_entry_rcu(kp, &p->list, list) | |
392 | kp->nmissed++; | |
393 | } | |
394 | return; | |
395 | } | |
396 | ||
99219a3f | 397 | void __kprobes recycle_rp_inst(struct kretprobe_instance *ri, |
398 | struct hlist_head *head) | |
b94cce92 | 399 | { |
ef53d9c5 S |
400 | struct kretprobe *rp = ri->rp; |
401 | ||
b94cce92 HN |
402 | /* remove rp inst off the rprobe_inst_table */ |
403 | hlist_del(&ri->hlist); | |
ef53d9c5 S |
404 | INIT_HLIST_NODE(&ri->hlist); |
405 | if (likely(rp)) { | |
406 | spin_lock(&rp->lock); | |
407 | hlist_add_head(&ri->hlist, &rp->free_instances); | |
408 | spin_unlock(&rp->lock); | |
b94cce92 HN |
409 | } else |
410 | /* Unregistering */ | |
99219a3f | 411 | hlist_add_head(&ri->hlist, head); |
b94cce92 HN |
412 | } |
413 | ||
017c39bd | 414 | void __kprobes kretprobe_hash_lock(struct task_struct *tsk, |
ef53d9c5 S |
415 | struct hlist_head **head, unsigned long *flags) |
416 | { | |
417 | unsigned long hash = hash_ptr(tsk, KPROBE_HASH_BITS); | |
418 | spinlock_t *hlist_lock; | |
419 | ||
420 | *head = &kretprobe_inst_table[hash]; | |
421 | hlist_lock = kretprobe_table_lock_ptr(hash); | |
422 | spin_lock_irqsave(hlist_lock, *flags); | |
423 | } | |
424 | ||
017c39bd MH |
425 | static void __kprobes kretprobe_table_lock(unsigned long hash, |
426 | unsigned long *flags) | |
b94cce92 | 427 | { |
ef53d9c5 S |
428 | spinlock_t *hlist_lock = kretprobe_table_lock_ptr(hash); |
429 | spin_lock_irqsave(hlist_lock, *flags); | |
430 | } | |
431 | ||
017c39bd MH |
432 | void __kprobes kretprobe_hash_unlock(struct task_struct *tsk, |
433 | unsigned long *flags) | |
ef53d9c5 S |
434 | { |
435 | unsigned long hash = hash_ptr(tsk, KPROBE_HASH_BITS); | |
436 | spinlock_t *hlist_lock; | |
437 | ||
438 | hlist_lock = kretprobe_table_lock_ptr(hash); | |
439 | spin_unlock_irqrestore(hlist_lock, *flags); | |
440 | } | |
441 | ||
017c39bd | 442 | void __kprobes kretprobe_table_unlock(unsigned long hash, unsigned long *flags) |
ef53d9c5 S |
443 | { |
444 | spinlock_t *hlist_lock = kretprobe_table_lock_ptr(hash); | |
445 | spin_unlock_irqrestore(hlist_lock, *flags); | |
b94cce92 HN |
446 | } |
447 | ||
b94cce92 | 448 | /* |
c6fd91f0 | 449 | * This function is called from finish_task_switch when task tk becomes dead, |
450 | * so that we can recycle any function-return probe instances associated | |
451 | * with this task. These left over instances represent probed functions | |
452 | * that have been called but will never return. | |
b94cce92 | 453 | */ |
d0aaff97 | 454 | void __kprobes kprobe_flush_task(struct task_struct *tk) |
b94cce92 | 455 | { |
62c27be0 | 456 | struct kretprobe_instance *ri; |
99219a3f | 457 | struct hlist_head *head, empty_rp; |
802eae7c | 458 | struct hlist_node *node, *tmp; |
ef53d9c5 | 459 | unsigned long hash, flags = 0; |
802eae7c | 460 | |
ef53d9c5 S |
461 | if (unlikely(!kprobes_initialized)) |
462 | /* Early boot. kretprobe_table_locks not yet initialized. */ | |
463 | return; | |
464 | ||
465 | hash = hash_ptr(tk, KPROBE_HASH_BITS); | |
466 | head = &kretprobe_inst_table[hash]; | |
467 | kretprobe_table_lock(hash, &flags); | |
62c27be0 | 468 | hlist_for_each_entry_safe(ri, node, tmp, head, hlist) { |
469 | if (ri->task == tk) | |
99219a3f | 470 | recycle_rp_inst(ri, &empty_rp); |
62c27be0 | 471 | } |
ef53d9c5 S |
472 | kretprobe_table_unlock(hash, &flags); |
473 | INIT_HLIST_HEAD(&empty_rp); | |
99219a3f | 474 | hlist_for_each_entry_safe(ri, node, tmp, &empty_rp, hlist) { |
475 | hlist_del(&ri->hlist); | |
476 | kfree(ri); | |
477 | } | |
b94cce92 HN |
478 | } |
479 | ||
b94cce92 HN |
480 | static inline void free_rp_inst(struct kretprobe *rp) |
481 | { | |
482 | struct kretprobe_instance *ri; | |
4c4308cb CH |
483 | struct hlist_node *pos, *next; |
484 | ||
ef53d9c5 S |
485 | hlist_for_each_entry_safe(ri, pos, next, &rp->free_instances, hlist) { |
486 | hlist_del(&ri->hlist); | |
b94cce92 HN |
487 | kfree(ri); |
488 | } | |
489 | } | |
490 | ||
4a296e07 MH |
491 | static void __kprobes cleanup_rp_inst(struct kretprobe *rp) |
492 | { | |
ef53d9c5 | 493 | unsigned long flags, hash; |
4a296e07 MH |
494 | struct kretprobe_instance *ri; |
495 | struct hlist_node *pos, *next; | |
ef53d9c5 S |
496 | struct hlist_head *head; |
497 | ||
4a296e07 | 498 | /* No race here */ |
ef53d9c5 S |
499 | for (hash = 0; hash < KPROBE_TABLE_SIZE; hash++) { |
500 | kretprobe_table_lock(hash, &flags); | |
501 | head = &kretprobe_inst_table[hash]; | |
502 | hlist_for_each_entry_safe(ri, pos, next, head, hlist) { | |
503 | if (ri->rp == rp) | |
504 | ri->rp = NULL; | |
505 | } | |
506 | kretprobe_table_unlock(hash, &flags); | |
4a296e07 | 507 | } |
4a296e07 MH |
508 | free_rp_inst(rp); |
509 | } | |
510 | ||
8b0914ea PP |
511 | /* |
512 | * Keep all fields in the kprobe consistent | |
513 | */ | |
514 | static inline void copy_kprobe(struct kprobe *old_p, struct kprobe *p) | |
515 | { | |
516 | memcpy(&p->opcode, &old_p->opcode, sizeof(kprobe_opcode_t)); | |
517 | memcpy(&p->ainsn, &old_p->ainsn, sizeof(struct arch_specific_insn)); | |
518 | } | |
519 | ||
520 | /* | |
b918e5e6 | 521 | * Add the new probe to ap->list. Fail if this is the |
8b0914ea PP |
522 | * second jprobe at the address - two jprobes can't coexist |
523 | */ | |
b918e5e6 | 524 | static int __kprobes add_new_kprobe(struct kprobe *ap, struct kprobe *p) |
8b0914ea | 525 | { |
8b0914ea | 526 | if (p->break_handler) { |
b918e5e6 | 527 | if (ap->break_handler) |
36721656 | 528 | return -EEXIST; |
b918e5e6 MH |
529 | list_add_tail_rcu(&p->list, &ap->list); |
530 | ap->break_handler = aggr_break_handler; | |
8b0914ea | 531 | } else |
b918e5e6 MH |
532 | list_add_rcu(&p->list, &ap->list); |
533 | if (p->post_handler && !ap->post_handler) | |
534 | ap->post_handler = aggr_post_handler; | |
8b0914ea PP |
535 | return 0; |
536 | } | |
537 | ||
64f562c6 AM |
538 | /* |
539 | * Fill in the required fields of the "manager kprobe". Replace the | |
540 | * earlier kprobe in the hlist with the manager kprobe | |
541 | */ | |
542 | static inline void add_aggr_kprobe(struct kprobe *ap, struct kprobe *p) | |
543 | { | |
8b0914ea | 544 | copy_kprobe(p, ap); |
a9ad965e | 545 | flush_insn_slot(ap); |
64f562c6 | 546 | ap->addr = p->addr; |
b918e5e6 | 547 | ap->flags = p->flags; |
64f562c6 | 548 | ap->pre_handler = aggr_pre_handler; |
64f562c6 | 549 | ap->fault_handler = aggr_fault_handler; |
e8386a0c MH |
550 | /* We don't care the kprobe which has gone. */ |
551 | if (p->post_handler && !kprobe_gone(p)) | |
36721656 | 552 | ap->post_handler = aggr_post_handler; |
e8386a0c | 553 | if (p->break_handler && !kprobe_gone(p)) |
36721656 | 554 | ap->break_handler = aggr_break_handler; |
64f562c6 AM |
555 | |
556 | INIT_LIST_HEAD(&ap->list); | |
3516a460 | 557 | list_add_rcu(&p->list, &ap->list); |
64f562c6 | 558 | |
adad0f33 | 559 | hlist_replace_rcu(&p->hlist, &ap->hlist); |
64f562c6 AM |
560 | } |
561 | ||
562 | /* | |
563 | * This is the second or subsequent kprobe at the address - handle | |
564 | * the intricacies | |
64f562c6 | 565 | */ |
d0aaff97 PP |
566 | static int __kprobes register_aggr_kprobe(struct kprobe *old_p, |
567 | struct kprobe *p) | |
64f562c6 AM |
568 | { |
569 | int ret = 0; | |
b918e5e6 | 570 | struct kprobe *ap = old_p; |
64f562c6 | 571 | |
b918e5e6 MH |
572 | if (old_p->pre_handler != aggr_pre_handler) { |
573 | /* If old_p is not an aggr_probe, create new aggr_kprobe. */ | |
574 | ap = kzalloc(sizeof(struct kprobe), GFP_KERNEL); | |
575 | if (!ap) | |
576 | return -ENOMEM; | |
577 | add_aggr_kprobe(ap, old_p); | |
578 | } | |
579 | ||
580 | if (kprobe_gone(ap)) { | |
e8386a0c MH |
581 | /* |
582 | * Attempting to insert new probe at the same location that | |
583 | * had a probe in the module vaddr area which already | |
584 | * freed. So, the instruction slot has already been | |
585 | * released. We need a new slot for the new probe. | |
586 | */ | |
b918e5e6 | 587 | ret = arch_prepare_kprobe(ap); |
e8386a0c | 588 | if (ret) |
b918e5e6 MH |
589 | /* |
590 | * Even if fail to allocate new slot, don't need to | |
591 | * free aggr_probe. It will be used next time, or | |
592 | * freed by unregister_kprobe. | |
593 | */ | |
e8386a0c | 594 | return ret; |
b918e5e6 MH |
595 | /* Clear gone flag to prevent allocating new slot again. */ |
596 | ap->flags &= ~KPROBE_FLAG_GONE; | |
e8386a0c MH |
597 | /* |
598 | * If the old_p has gone, its breakpoint has been disarmed. | |
599 | * We have to arm it again after preparing real kprobes. | |
600 | */ | |
e8386a0c MH |
601 | if (kprobe_enabled) |
602 | arch_arm_kprobe(ap); | |
603 | } | |
b918e5e6 MH |
604 | |
605 | copy_kprobe(ap, p); | |
606 | return add_new_kprobe(ap, p); | |
64f562c6 AM |
607 | } |
608 | ||
d0aaff97 PP |
609 | static int __kprobes in_kprobes_functions(unsigned long addr) |
610 | { | |
3d8d996e SD |
611 | struct kprobe_blackpoint *kb; |
612 | ||
6f716acd CH |
613 | if (addr >= (unsigned long)__kprobes_text_start && |
614 | addr < (unsigned long)__kprobes_text_end) | |
d0aaff97 | 615 | return -EINVAL; |
3d8d996e SD |
616 | /* |
617 | * If there exists a kprobe_blacklist, verify and | |
618 | * fail any probe registration in the prohibited area | |
619 | */ | |
620 | for (kb = kprobe_blacklist; kb->name != NULL; kb++) { | |
621 | if (kb->start_addr) { | |
622 | if (addr >= kb->start_addr && | |
623 | addr < (kb->start_addr + kb->range)) | |
624 | return -EINVAL; | |
625 | } | |
626 | } | |
d0aaff97 PP |
627 | return 0; |
628 | } | |
629 | ||
b2a5cd69 MH |
630 | /* |
631 | * If we have a symbol_name argument, look it up and add the offset field | |
632 | * to it. This way, we can specify a relative address to a symbol. | |
633 | */ | |
634 | static kprobe_opcode_t __kprobes *kprobe_addr(struct kprobe *p) | |
635 | { | |
636 | kprobe_opcode_t *addr = p->addr; | |
637 | if (p->symbol_name) { | |
638 | if (addr) | |
639 | return NULL; | |
640 | kprobe_lookup_name(p->symbol_name, addr); | |
641 | } | |
642 | ||
643 | if (!addr) | |
644 | return NULL; | |
645 | return (kprobe_opcode_t *)(((char *)addr) + p->offset); | |
646 | } | |
647 | ||
49ad2fd7 | 648 | int __kprobes register_kprobe(struct kprobe *p) |
1da177e4 LT |
649 | { |
650 | int ret = 0; | |
64f562c6 | 651 | struct kprobe *old_p; |
df019b1d | 652 | struct module *probed_mod; |
b2a5cd69 | 653 | kprobe_opcode_t *addr; |
b3e55c72 | 654 | |
b2a5cd69 MH |
655 | addr = kprobe_addr(p); |
656 | if (!addr) | |
3a872d89 | 657 | return -EINVAL; |
b2a5cd69 | 658 | p->addr = addr; |
3a872d89 | 659 | |
a189d035 MH |
660 | preempt_disable(); |
661 | if (!__kernel_text_address((unsigned long) p->addr) || | |
662 | in_kprobes_functions((unsigned long) p->addr)) { | |
663 | preempt_enable(); | |
b3e55c72 | 664 | return -EINVAL; |
a189d035 | 665 | } |
b3e55c72 | 666 | |
e8386a0c | 667 | p->flags = 0; |
6f716acd CH |
668 | /* |
669 | * Check if are we probing a module. | |
670 | */ | |
a189d035 | 671 | probed_mod = __module_text_address((unsigned long) p->addr); |
6f716acd | 672 | if (probed_mod) { |
6f716acd | 673 | /* |
e8386a0c MH |
674 | * We must hold a refcount of the probed module while updating |
675 | * its code to prohibit unexpected unloading. | |
df019b1d | 676 | */ |
49ad2fd7 MH |
677 | if (unlikely(!try_module_get(probed_mod))) { |
678 | preempt_enable(); | |
679 | return -EINVAL; | |
680 | } | |
f24659d9 MH |
681 | /* |
682 | * If the module freed .init.text, we couldn't insert | |
683 | * kprobes in there. | |
684 | */ | |
685 | if (within_module_init((unsigned long)p->addr, probed_mod) && | |
686 | probed_mod->state != MODULE_STATE_COMING) { | |
687 | module_put(probed_mod); | |
688 | preempt_enable(); | |
689 | return -EINVAL; | |
690 | } | |
df019b1d | 691 | } |
a189d035 | 692 | preempt_enable(); |
1da177e4 | 693 | |
3516a460 | 694 | p->nmissed = 0; |
9861668f | 695 | INIT_LIST_HEAD(&p->list); |
7a7d1cf9 | 696 | mutex_lock(&kprobe_mutex); |
64f562c6 AM |
697 | old_p = get_kprobe(p->addr); |
698 | if (old_p) { | |
699 | ret = register_aggr_kprobe(old_p, p); | |
1da177e4 LT |
700 | goto out; |
701 | } | |
1da177e4 | 702 | |
4460fdad | 703 | mutex_lock(&text_mutex); |
6f716acd CH |
704 | ret = arch_prepare_kprobe(p); |
705 | if (ret) | |
4460fdad | 706 | goto out_unlock_text; |
49a2a1b8 | 707 | |
64f562c6 | 708 | INIT_HLIST_NODE(&p->hlist); |
3516a460 | 709 | hlist_add_head_rcu(&p->hlist, |
1da177e4 LT |
710 | &kprobe_table[hash_ptr(p->addr, KPROBE_HASH_BITS)]); |
711 | ||
74a0b576 | 712 | if (kprobe_enabled) |
bf8f6e5b | 713 | arch_arm_kprobe(p); |
74a0b576 | 714 | |
4460fdad MD |
715 | out_unlock_text: |
716 | mutex_unlock(&text_mutex); | |
1da177e4 | 717 | out: |
7a7d1cf9 | 718 | mutex_unlock(&kprobe_mutex); |
49a2a1b8 | 719 | |
e8386a0c | 720 | if (probed_mod) |
df019b1d | 721 | module_put(probed_mod); |
e8386a0c | 722 | |
1da177e4 LT |
723 | return ret; |
724 | } | |
725 | ||
9861668f MH |
726 | /* |
727 | * Unregister a kprobe without a scheduler synchronization. | |
728 | */ | |
729 | static int __kprobes __unregister_kprobe_top(struct kprobe *p) | |
1da177e4 | 730 | { |
f709b122 | 731 | struct kprobe *old_p, *list_p; |
64f562c6 | 732 | |
64f562c6 | 733 | old_p = get_kprobe(p->addr); |
9861668f MH |
734 | if (unlikely(!old_p)) |
735 | return -EINVAL; | |
736 | ||
f709b122 KA |
737 | if (p != old_p) { |
738 | list_for_each_entry_rcu(list_p, &old_p->list, list) | |
739 | if (list_p == p) | |
740 | /* kprobe p is a valid probe */ | |
741 | goto valid_p; | |
9861668f | 742 | return -EINVAL; |
f709b122 KA |
743 | } |
744 | valid_p: | |
6f716acd CH |
745 | if (old_p == p || |
746 | (old_p->pre_handler == aggr_pre_handler && | |
9861668f | 747 | list_is_singular(&old_p->list))) { |
bf8f6e5b AM |
748 | /* |
749 | * Only probe on the hash list. Disarm only if kprobes are | |
e8386a0c MH |
750 | * enabled and not gone - otherwise, the breakpoint would |
751 | * already have been removed. We save on flushing icache. | |
bf8f6e5b | 752 | */ |
4460fdad MD |
753 | if (kprobe_enabled && !kprobe_gone(old_p)) { |
754 | mutex_lock(&text_mutex); | |
bf8f6e5b | 755 | arch_disarm_kprobe(p); |
4460fdad MD |
756 | mutex_unlock(&text_mutex); |
757 | } | |
49a2a1b8 | 758 | hlist_del_rcu(&old_p->hlist); |
49a2a1b8 | 759 | } else { |
e8386a0c | 760 | if (p->break_handler && !kprobe_gone(p)) |
9861668f | 761 | old_p->break_handler = NULL; |
e8386a0c | 762 | if (p->post_handler && !kprobe_gone(p)) { |
9861668f MH |
763 | list_for_each_entry_rcu(list_p, &old_p->list, list) { |
764 | if ((list_p != p) && (list_p->post_handler)) | |
765 | goto noclean; | |
766 | } | |
767 | old_p->post_handler = NULL; | |
768 | } | |
769 | noclean: | |
49a2a1b8 | 770 | list_del_rcu(&p->list); |
49a2a1b8 | 771 | } |
9861668f MH |
772 | return 0; |
773 | } | |
3516a460 | 774 | |
9861668f MH |
775 | static void __kprobes __unregister_kprobe_bottom(struct kprobe *p) |
776 | { | |
9861668f | 777 | struct kprobe *old_p; |
b3e55c72 | 778 | |
e8386a0c | 779 | if (list_empty(&p->list)) |
0498b635 | 780 | arch_remove_kprobe(p); |
e8386a0c MH |
781 | else if (list_is_singular(&p->list)) { |
782 | /* "p" is the last child of an aggr_kprobe */ | |
783 | old_p = list_entry(p->list.next, struct kprobe, list); | |
784 | list_del(&p->list); | |
785 | arch_remove_kprobe(old_p); | |
786 | kfree(old_p); | |
9861668f MH |
787 | } |
788 | } | |
789 | ||
49ad2fd7 | 790 | int __kprobes register_kprobes(struct kprobe **kps, int num) |
9861668f MH |
791 | { |
792 | int i, ret = 0; | |
793 | ||
794 | if (num <= 0) | |
795 | return -EINVAL; | |
796 | for (i = 0; i < num; i++) { | |
49ad2fd7 | 797 | ret = register_kprobe(kps[i]); |
67dddaad MH |
798 | if (ret < 0) { |
799 | if (i > 0) | |
800 | unregister_kprobes(kps, i); | |
9861668f | 801 | break; |
36721656 | 802 | } |
49a2a1b8 | 803 | } |
9861668f MH |
804 | return ret; |
805 | } | |
806 | ||
9861668f MH |
807 | void __kprobes unregister_kprobe(struct kprobe *p) |
808 | { | |
809 | unregister_kprobes(&p, 1); | |
810 | } | |
811 | ||
9861668f MH |
812 | void __kprobes unregister_kprobes(struct kprobe **kps, int num) |
813 | { | |
814 | int i; | |
815 | ||
816 | if (num <= 0) | |
817 | return; | |
818 | mutex_lock(&kprobe_mutex); | |
819 | for (i = 0; i < num; i++) | |
820 | if (__unregister_kprobe_top(kps[i]) < 0) | |
821 | kps[i]->addr = NULL; | |
822 | mutex_unlock(&kprobe_mutex); | |
823 | ||
824 | synchronize_sched(); | |
825 | for (i = 0; i < num; i++) | |
826 | if (kps[i]->addr) | |
827 | __unregister_kprobe_bottom(kps[i]); | |
1da177e4 LT |
828 | } |
829 | ||
830 | static struct notifier_block kprobe_exceptions_nb = { | |
3d5631e0 AK |
831 | .notifier_call = kprobe_exceptions_notify, |
832 | .priority = 0x7fffffff /* we need to be notified first */ | |
833 | }; | |
834 | ||
3d7e3382 ME |
835 | unsigned long __weak arch_deref_entry_point(void *entry) |
836 | { | |
837 | return (unsigned long)entry; | |
838 | } | |
1da177e4 | 839 | |
49ad2fd7 | 840 | int __kprobes register_jprobes(struct jprobe **jps, int num) |
1da177e4 | 841 | { |
26b31c19 MH |
842 | struct jprobe *jp; |
843 | int ret = 0, i; | |
3d7e3382 | 844 | |
26b31c19 | 845 | if (num <= 0) |
3d7e3382 | 846 | return -EINVAL; |
26b31c19 MH |
847 | for (i = 0; i < num; i++) { |
848 | unsigned long addr; | |
849 | jp = jps[i]; | |
850 | addr = arch_deref_entry_point(jp->entry); | |
851 | ||
852 | if (!kernel_text_address(addr)) | |
853 | ret = -EINVAL; | |
854 | else { | |
855 | /* Todo: Verify probepoint is a function entry point */ | |
856 | jp->kp.pre_handler = setjmp_pre_handler; | |
857 | jp->kp.break_handler = longjmp_break_handler; | |
49ad2fd7 | 858 | ret = register_kprobe(&jp->kp); |
26b31c19 | 859 | } |
67dddaad MH |
860 | if (ret < 0) { |
861 | if (i > 0) | |
862 | unregister_jprobes(jps, i); | |
26b31c19 MH |
863 | break; |
864 | } | |
865 | } | |
866 | return ret; | |
867 | } | |
3d7e3382 | 868 | |
26b31c19 MH |
869 | int __kprobes register_jprobe(struct jprobe *jp) |
870 | { | |
49ad2fd7 | 871 | return register_jprobes(&jp, 1); |
1da177e4 LT |
872 | } |
873 | ||
d0aaff97 | 874 | void __kprobes unregister_jprobe(struct jprobe *jp) |
1da177e4 | 875 | { |
26b31c19 MH |
876 | unregister_jprobes(&jp, 1); |
877 | } | |
878 | ||
26b31c19 MH |
879 | void __kprobes unregister_jprobes(struct jprobe **jps, int num) |
880 | { | |
881 | int i; | |
882 | ||
883 | if (num <= 0) | |
884 | return; | |
885 | mutex_lock(&kprobe_mutex); | |
886 | for (i = 0; i < num; i++) | |
887 | if (__unregister_kprobe_top(&jps[i]->kp) < 0) | |
888 | jps[i]->kp.addr = NULL; | |
889 | mutex_unlock(&kprobe_mutex); | |
890 | ||
891 | synchronize_sched(); | |
892 | for (i = 0; i < num; i++) { | |
893 | if (jps[i]->kp.addr) | |
894 | __unregister_kprobe_bottom(&jps[i]->kp); | |
895 | } | |
1da177e4 LT |
896 | } |
897 | ||
9edddaa2 | 898 | #ifdef CONFIG_KRETPROBES |
e65cefe8 AB |
899 | /* |
900 | * This kprobe pre_handler is registered with every kretprobe. When probe | |
901 | * hits it will set up the return probe. | |
902 | */ | |
903 | static int __kprobes pre_handler_kretprobe(struct kprobe *p, | |
904 | struct pt_regs *regs) | |
905 | { | |
906 | struct kretprobe *rp = container_of(p, struct kretprobe, kp); | |
ef53d9c5 S |
907 | unsigned long hash, flags = 0; |
908 | struct kretprobe_instance *ri; | |
e65cefe8 AB |
909 | |
910 | /*TODO: consider to only swap the RA after the last pre_handler fired */ | |
ef53d9c5 S |
911 | hash = hash_ptr(current, KPROBE_HASH_BITS); |
912 | spin_lock_irqsave(&rp->lock, flags); | |
4c4308cb | 913 | if (!hlist_empty(&rp->free_instances)) { |
4c4308cb | 914 | ri = hlist_entry(rp->free_instances.first, |
ef53d9c5 S |
915 | struct kretprobe_instance, hlist); |
916 | hlist_del(&ri->hlist); | |
917 | spin_unlock_irqrestore(&rp->lock, flags); | |
918 | ||
4c4308cb CH |
919 | ri->rp = rp; |
920 | ri->task = current; | |
f47cd9b5 | 921 | |
f02b8624 | 922 | if (rp->entry_handler && rp->entry_handler(ri, regs)) |
f47cd9b5 | 923 | return 0; |
f47cd9b5 | 924 | |
4c4308cb CH |
925 | arch_prepare_kretprobe(ri, regs); |
926 | ||
927 | /* XXX(hch): why is there no hlist_move_head? */ | |
ef53d9c5 S |
928 | INIT_HLIST_NODE(&ri->hlist); |
929 | kretprobe_table_lock(hash, &flags); | |
930 | hlist_add_head(&ri->hlist, &kretprobe_inst_table[hash]); | |
931 | kretprobe_table_unlock(hash, &flags); | |
932 | } else { | |
4c4308cb | 933 | rp->nmissed++; |
ef53d9c5 S |
934 | spin_unlock_irqrestore(&rp->lock, flags); |
935 | } | |
e65cefe8 AB |
936 | return 0; |
937 | } | |
938 | ||
49ad2fd7 | 939 | int __kprobes register_kretprobe(struct kretprobe *rp) |
b94cce92 HN |
940 | { |
941 | int ret = 0; | |
942 | struct kretprobe_instance *inst; | |
943 | int i; | |
b2a5cd69 | 944 | void *addr; |
f438d914 MH |
945 | |
946 | if (kretprobe_blacklist_size) { | |
b2a5cd69 MH |
947 | addr = kprobe_addr(&rp->kp); |
948 | if (!addr) | |
949 | return -EINVAL; | |
f438d914 MH |
950 | |
951 | for (i = 0; kretprobe_blacklist[i].name != NULL; i++) { | |
952 | if (kretprobe_blacklist[i].addr == addr) | |
953 | return -EINVAL; | |
954 | } | |
955 | } | |
b94cce92 HN |
956 | |
957 | rp->kp.pre_handler = pre_handler_kretprobe; | |
7522a842 AM |
958 | rp->kp.post_handler = NULL; |
959 | rp->kp.fault_handler = NULL; | |
960 | rp->kp.break_handler = NULL; | |
b94cce92 HN |
961 | |
962 | /* Pre-allocate memory for max kretprobe instances */ | |
963 | if (rp->maxactive <= 0) { | |
964 | #ifdef CONFIG_PREEMPT | |
965 | rp->maxactive = max(10, 2 * NR_CPUS); | |
966 | #else | |
967 | rp->maxactive = NR_CPUS; | |
968 | #endif | |
969 | } | |
ef53d9c5 | 970 | spin_lock_init(&rp->lock); |
b94cce92 HN |
971 | INIT_HLIST_HEAD(&rp->free_instances); |
972 | for (i = 0; i < rp->maxactive; i++) { | |
f47cd9b5 AS |
973 | inst = kmalloc(sizeof(struct kretprobe_instance) + |
974 | rp->data_size, GFP_KERNEL); | |
b94cce92 HN |
975 | if (inst == NULL) { |
976 | free_rp_inst(rp); | |
977 | return -ENOMEM; | |
978 | } | |
ef53d9c5 S |
979 | INIT_HLIST_NODE(&inst->hlist); |
980 | hlist_add_head(&inst->hlist, &rp->free_instances); | |
b94cce92 HN |
981 | } |
982 | ||
983 | rp->nmissed = 0; | |
984 | /* Establish function entry probe point */ | |
49ad2fd7 | 985 | ret = register_kprobe(&rp->kp); |
4a296e07 | 986 | if (ret != 0) |
b94cce92 HN |
987 | free_rp_inst(rp); |
988 | return ret; | |
989 | } | |
990 | ||
49ad2fd7 | 991 | int __kprobes register_kretprobes(struct kretprobe **rps, int num) |
4a296e07 MH |
992 | { |
993 | int ret = 0, i; | |
994 | ||
995 | if (num <= 0) | |
996 | return -EINVAL; | |
997 | for (i = 0; i < num; i++) { | |
49ad2fd7 | 998 | ret = register_kretprobe(rps[i]); |
67dddaad MH |
999 | if (ret < 0) { |
1000 | if (i > 0) | |
1001 | unregister_kretprobes(rps, i); | |
4a296e07 MH |
1002 | break; |
1003 | } | |
1004 | } | |
1005 | return ret; | |
1006 | } | |
1007 | ||
4a296e07 MH |
1008 | void __kprobes unregister_kretprobe(struct kretprobe *rp) |
1009 | { | |
1010 | unregister_kretprobes(&rp, 1); | |
1011 | } | |
1012 | ||
4a296e07 MH |
1013 | void __kprobes unregister_kretprobes(struct kretprobe **rps, int num) |
1014 | { | |
1015 | int i; | |
1016 | ||
1017 | if (num <= 0) | |
1018 | return; | |
1019 | mutex_lock(&kprobe_mutex); | |
1020 | for (i = 0; i < num; i++) | |
1021 | if (__unregister_kprobe_top(&rps[i]->kp) < 0) | |
1022 | rps[i]->kp.addr = NULL; | |
1023 | mutex_unlock(&kprobe_mutex); | |
1024 | ||
1025 | synchronize_sched(); | |
1026 | for (i = 0; i < num; i++) { | |
1027 | if (rps[i]->kp.addr) { | |
1028 | __unregister_kprobe_bottom(&rps[i]->kp); | |
1029 | cleanup_rp_inst(rps[i]); | |
1030 | } | |
1031 | } | |
1032 | } | |
1033 | ||
9edddaa2 | 1034 | #else /* CONFIG_KRETPROBES */ |
d0aaff97 | 1035 | int __kprobes register_kretprobe(struct kretprobe *rp) |
b94cce92 HN |
1036 | { |
1037 | return -ENOSYS; | |
1038 | } | |
1039 | ||
4a296e07 | 1040 | int __kprobes register_kretprobes(struct kretprobe **rps, int num) |
346fd59b | 1041 | { |
4a296e07 | 1042 | return -ENOSYS; |
346fd59b | 1043 | } |
d0aaff97 | 1044 | void __kprobes unregister_kretprobe(struct kretprobe *rp) |
b94cce92 | 1045 | { |
4a296e07 | 1046 | } |
b94cce92 | 1047 | |
4a296e07 MH |
1048 | void __kprobes unregister_kretprobes(struct kretprobe **rps, int num) |
1049 | { | |
1050 | } | |
4c4308cb | 1051 | |
4a296e07 MH |
1052 | static int __kprobes pre_handler_kretprobe(struct kprobe *p, |
1053 | struct pt_regs *regs) | |
1054 | { | |
1055 | return 0; | |
b94cce92 HN |
1056 | } |
1057 | ||
4a296e07 MH |
1058 | #endif /* CONFIG_KRETPROBES */ |
1059 | ||
e8386a0c MH |
1060 | /* Set the kprobe gone and remove its instruction buffer. */ |
1061 | static void __kprobes kill_kprobe(struct kprobe *p) | |
1062 | { | |
1063 | struct kprobe *kp; | |
1064 | p->flags |= KPROBE_FLAG_GONE; | |
1065 | if (p->pre_handler == aggr_pre_handler) { | |
1066 | /* | |
1067 | * If this is an aggr_kprobe, we have to list all the | |
1068 | * chained probes and mark them GONE. | |
1069 | */ | |
1070 | list_for_each_entry_rcu(kp, &p->list, list) | |
1071 | kp->flags |= KPROBE_FLAG_GONE; | |
1072 | p->post_handler = NULL; | |
1073 | p->break_handler = NULL; | |
1074 | } | |
1075 | /* | |
1076 | * Here, we can remove insn_slot safely, because no thread calls | |
1077 | * the original probed function (which will be freed soon) any more. | |
1078 | */ | |
1079 | arch_remove_kprobe(p); | |
1080 | } | |
1081 | ||
1082 | /* Module notifier call back, checking kprobes on the module */ | |
1083 | static int __kprobes kprobes_module_callback(struct notifier_block *nb, | |
1084 | unsigned long val, void *data) | |
1085 | { | |
1086 | struct module *mod = data; | |
1087 | struct hlist_head *head; | |
1088 | struct hlist_node *node; | |
1089 | struct kprobe *p; | |
1090 | unsigned int i; | |
f24659d9 | 1091 | int checkcore = (val == MODULE_STATE_GOING); |
e8386a0c | 1092 | |
f24659d9 | 1093 | if (val != MODULE_STATE_GOING && val != MODULE_STATE_LIVE) |
e8386a0c MH |
1094 | return NOTIFY_DONE; |
1095 | ||
1096 | /* | |
f24659d9 MH |
1097 | * When MODULE_STATE_GOING was notified, both of module .text and |
1098 | * .init.text sections would be freed. When MODULE_STATE_LIVE was | |
1099 | * notified, only .init.text section would be freed. We need to | |
1100 | * disable kprobes which have been inserted in the sections. | |
e8386a0c MH |
1101 | */ |
1102 | mutex_lock(&kprobe_mutex); | |
1103 | for (i = 0; i < KPROBE_TABLE_SIZE; i++) { | |
1104 | head = &kprobe_table[i]; | |
1105 | hlist_for_each_entry_rcu(p, node, head, hlist) | |
f24659d9 MH |
1106 | if (within_module_init((unsigned long)p->addr, mod) || |
1107 | (checkcore && | |
1108 | within_module_core((unsigned long)p->addr, mod))) { | |
e8386a0c MH |
1109 | /* |
1110 | * The vaddr this probe is installed will soon | |
1111 | * be vfreed buy not synced to disk. Hence, | |
1112 | * disarming the breakpoint isn't needed. | |
1113 | */ | |
1114 | kill_kprobe(p); | |
1115 | } | |
1116 | } | |
1117 | mutex_unlock(&kprobe_mutex); | |
1118 | return NOTIFY_DONE; | |
1119 | } | |
1120 | ||
1121 | static struct notifier_block kprobe_module_nb = { | |
1122 | .notifier_call = kprobes_module_callback, | |
1123 | .priority = 0 | |
1124 | }; | |
1125 | ||
1da177e4 LT |
1126 | static int __init init_kprobes(void) |
1127 | { | |
1128 | int i, err = 0; | |
3d8d996e SD |
1129 | unsigned long offset = 0, size = 0; |
1130 | char *modname, namebuf[128]; | |
1131 | const char *symbol_name; | |
1132 | void *addr; | |
1133 | struct kprobe_blackpoint *kb; | |
1da177e4 LT |
1134 | |
1135 | /* FIXME allocate the probe table, currently defined statically */ | |
1136 | /* initialize all list heads */ | |
b94cce92 | 1137 | for (i = 0; i < KPROBE_TABLE_SIZE; i++) { |
1da177e4 | 1138 | INIT_HLIST_HEAD(&kprobe_table[i]); |
b94cce92 | 1139 | INIT_HLIST_HEAD(&kretprobe_inst_table[i]); |
ef53d9c5 | 1140 | spin_lock_init(&(kretprobe_table_locks[i].lock)); |
b94cce92 | 1141 | } |
1da177e4 | 1142 | |
3d8d996e SD |
1143 | /* |
1144 | * Lookup and populate the kprobe_blacklist. | |
1145 | * | |
1146 | * Unlike the kretprobe blacklist, we'll need to determine | |
1147 | * the range of addresses that belong to the said functions, | |
1148 | * since a kprobe need not necessarily be at the beginning | |
1149 | * of a function. | |
1150 | */ | |
1151 | for (kb = kprobe_blacklist; kb->name != NULL; kb++) { | |
1152 | kprobe_lookup_name(kb->name, addr); | |
1153 | if (!addr) | |
1154 | continue; | |
1155 | ||
1156 | kb->start_addr = (unsigned long)addr; | |
1157 | symbol_name = kallsyms_lookup(kb->start_addr, | |
1158 | &size, &offset, &modname, namebuf); | |
1159 | if (!symbol_name) | |
1160 | kb->range = 0; | |
1161 | else | |
1162 | kb->range = size; | |
1163 | } | |
1164 | ||
f438d914 MH |
1165 | if (kretprobe_blacklist_size) { |
1166 | /* lookup the function address from its name */ | |
1167 | for (i = 0; kretprobe_blacklist[i].name != NULL; i++) { | |
1168 | kprobe_lookup_name(kretprobe_blacklist[i].name, | |
1169 | kretprobe_blacklist[i].addr); | |
1170 | if (!kretprobe_blacklist[i].addr) | |
1171 | printk("kretprobe: lookup failed: %s\n", | |
1172 | kretprobe_blacklist[i].name); | |
1173 | } | |
1174 | } | |
1175 | ||
bf8f6e5b AM |
1176 | /* By default, kprobes are enabled */ |
1177 | kprobe_enabled = true; | |
1178 | ||
6772926b | 1179 | err = arch_init_kprobes(); |
802eae7c RL |
1180 | if (!err) |
1181 | err = register_die_notifier(&kprobe_exceptions_nb); | |
e8386a0c MH |
1182 | if (!err) |
1183 | err = register_module_notifier(&kprobe_module_nb); | |
1184 | ||
ef53d9c5 | 1185 | kprobes_initialized = (err == 0); |
802eae7c | 1186 | |
8c1c9356 AM |
1187 | if (!err) |
1188 | init_test_probes(); | |
1da177e4 LT |
1189 | return err; |
1190 | } | |
1191 | ||
346fd59b SD |
1192 | #ifdef CONFIG_DEBUG_FS |
1193 | static void __kprobes report_probe(struct seq_file *pi, struct kprobe *p, | |
bf8f6e5b | 1194 | const char *sym, int offset,char *modname) |
346fd59b SD |
1195 | { |
1196 | char *kprobe_type; | |
1197 | ||
1198 | if (p->pre_handler == pre_handler_kretprobe) | |
1199 | kprobe_type = "r"; | |
1200 | else if (p->pre_handler == setjmp_pre_handler) | |
1201 | kprobe_type = "j"; | |
1202 | else | |
1203 | kprobe_type = "k"; | |
1204 | if (sym) | |
e8386a0c MH |
1205 | seq_printf(pi, "%p %s %s+0x%x %s %s\n", p->addr, kprobe_type, |
1206 | sym, offset, (modname ? modname : " "), | |
1207 | (kprobe_gone(p) ? "[GONE]" : "")); | |
346fd59b | 1208 | else |
e8386a0c MH |
1209 | seq_printf(pi, "%p %s %p %s\n", p->addr, kprobe_type, p->addr, |
1210 | (kprobe_gone(p) ? "[GONE]" : "")); | |
346fd59b SD |
1211 | } |
1212 | ||
1213 | static void __kprobes *kprobe_seq_start(struct seq_file *f, loff_t *pos) | |
1214 | { | |
1215 | return (*pos < KPROBE_TABLE_SIZE) ? pos : NULL; | |
1216 | } | |
1217 | ||
1218 | static void __kprobes *kprobe_seq_next(struct seq_file *f, void *v, loff_t *pos) | |
1219 | { | |
1220 | (*pos)++; | |
1221 | if (*pos >= KPROBE_TABLE_SIZE) | |
1222 | return NULL; | |
1223 | return pos; | |
1224 | } | |
1225 | ||
1226 | static void __kprobes kprobe_seq_stop(struct seq_file *f, void *v) | |
1227 | { | |
1228 | /* Nothing to do */ | |
1229 | } | |
1230 | ||
1231 | static int __kprobes show_kprobe_addr(struct seq_file *pi, void *v) | |
1232 | { | |
1233 | struct hlist_head *head; | |
1234 | struct hlist_node *node; | |
1235 | struct kprobe *p, *kp; | |
1236 | const char *sym = NULL; | |
1237 | unsigned int i = *(loff_t *) v; | |
ffb45122 | 1238 | unsigned long offset = 0; |
346fd59b SD |
1239 | char *modname, namebuf[128]; |
1240 | ||
1241 | head = &kprobe_table[i]; | |
1242 | preempt_disable(); | |
1243 | hlist_for_each_entry_rcu(p, node, head, hlist) { | |
ffb45122 | 1244 | sym = kallsyms_lookup((unsigned long)p->addr, NULL, |
346fd59b SD |
1245 | &offset, &modname, namebuf); |
1246 | if (p->pre_handler == aggr_pre_handler) { | |
1247 | list_for_each_entry_rcu(kp, &p->list, list) | |
1248 | report_probe(pi, kp, sym, offset, modname); | |
1249 | } else | |
1250 | report_probe(pi, p, sym, offset, modname); | |
1251 | } | |
1252 | preempt_enable(); | |
1253 | return 0; | |
1254 | } | |
1255 | ||
1256 | static struct seq_operations kprobes_seq_ops = { | |
1257 | .start = kprobe_seq_start, | |
1258 | .next = kprobe_seq_next, | |
1259 | .stop = kprobe_seq_stop, | |
1260 | .show = show_kprobe_addr | |
1261 | }; | |
1262 | ||
1263 | static int __kprobes kprobes_open(struct inode *inode, struct file *filp) | |
1264 | { | |
1265 | return seq_open(filp, &kprobes_seq_ops); | |
1266 | } | |
1267 | ||
1268 | static struct file_operations debugfs_kprobes_operations = { | |
1269 | .open = kprobes_open, | |
1270 | .read = seq_read, | |
1271 | .llseek = seq_lseek, | |
1272 | .release = seq_release, | |
1273 | }; | |
1274 | ||
bf8f6e5b AM |
1275 | static void __kprobes enable_all_kprobes(void) |
1276 | { | |
1277 | struct hlist_head *head; | |
1278 | struct hlist_node *node; | |
1279 | struct kprobe *p; | |
1280 | unsigned int i; | |
1281 | ||
1282 | mutex_lock(&kprobe_mutex); | |
1283 | ||
1284 | /* If kprobes are already enabled, just return */ | |
1285 | if (kprobe_enabled) | |
1286 | goto already_enabled; | |
1287 | ||
4460fdad | 1288 | mutex_lock(&text_mutex); |
bf8f6e5b AM |
1289 | for (i = 0; i < KPROBE_TABLE_SIZE; i++) { |
1290 | head = &kprobe_table[i]; | |
1291 | hlist_for_each_entry_rcu(p, node, head, hlist) | |
e8386a0c MH |
1292 | if (!kprobe_gone(p)) |
1293 | arch_arm_kprobe(p); | |
bf8f6e5b | 1294 | } |
4460fdad | 1295 | mutex_unlock(&text_mutex); |
bf8f6e5b AM |
1296 | |
1297 | kprobe_enabled = true; | |
1298 | printk(KERN_INFO "Kprobes globally enabled\n"); | |
1299 | ||
1300 | already_enabled: | |
1301 | mutex_unlock(&kprobe_mutex); | |
1302 | return; | |
1303 | } | |
1304 | ||
1305 | static void __kprobes disable_all_kprobes(void) | |
1306 | { | |
1307 | struct hlist_head *head; | |
1308 | struct hlist_node *node; | |
1309 | struct kprobe *p; | |
1310 | unsigned int i; | |
1311 | ||
1312 | mutex_lock(&kprobe_mutex); | |
1313 | ||
1314 | /* If kprobes are already disabled, just return */ | |
1315 | if (!kprobe_enabled) | |
1316 | goto already_disabled; | |
1317 | ||
1318 | kprobe_enabled = false; | |
1319 | printk(KERN_INFO "Kprobes globally disabled\n"); | |
4460fdad | 1320 | mutex_lock(&text_mutex); |
bf8f6e5b AM |
1321 | for (i = 0; i < KPROBE_TABLE_SIZE; i++) { |
1322 | head = &kprobe_table[i]; | |
1323 | hlist_for_each_entry_rcu(p, node, head, hlist) { | |
e8386a0c | 1324 | if (!arch_trampoline_kprobe(p) && !kprobe_gone(p)) |
bf8f6e5b AM |
1325 | arch_disarm_kprobe(p); |
1326 | } | |
1327 | } | |
1328 | ||
4460fdad | 1329 | mutex_unlock(&text_mutex); |
bf8f6e5b AM |
1330 | mutex_unlock(&kprobe_mutex); |
1331 | /* Allow all currently running kprobes to complete */ | |
1332 | synchronize_sched(); | |
74a0b576 | 1333 | return; |
bf8f6e5b AM |
1334 | |
1335 | already_disabled: | |
1336 | mutex_unlock(&kprobe_mutex); | |
1337 | return; | |
1338 | } | |
1339 | ||
1340 | /* | |
1341 | * XXX: The debugfs bool file interface doesn't allow for callbacks | |
1342 | * when the bool state is switched. We can reuse that facility when | |
1343 | * available | |
1344 | */ | |
1345 | static ssize_t read_enabled_file_bool(struct file *file, | |
1346 | char __user *user_buf, size_t count, loff_t *ppos) | |
1347 | { | |
1348 | char buf[3]; | |
1349 | ||
1350 | if (kprobe_enabled) | |
1351 | buf[0] = '1'; | |
1352 | else | |
1353 | buf[0] = '0'; | |
1354 | buf[1] = '\n'; | |
1355 | buf[2] = 0x00; | |
1356 | return simple_read_from_buffer(user_buf, count, ppos, buf, 2); | |
1357 | } | |
1358 | ||
1359 | static ssize_t write_enabled_file_bool(struct file *file, | |
1360 | const char __user *user_buf, size_t count, loff_t *ppos) | |
1361 | { | |
1362 | char buf[32]; | |
1363 | int buf_size; | |
1364 | ||
1365 | buf_size = min(count, (sizeof(buf)-1)); | |
1366 | if (copy_from_user(buf, user_buf, buf_size)) | |
1367 | return -EFAULT; | |
1368 | ||
1369 | switch (buf[0]) { | |
1370 | case 'y': | |
1371 | case 'Y': | |
1372 | case '1': | |
1373 | enable_all_kprobes(); | |
1374 | break; | |
1375 | case 'n': | |
1376 | case 'N': | |
1377 | case '0': | |
1378 | disable_all_kprobes(); | |
1379 | break; | |
1380 | } | |
1381 | ||
1382 | return count; | |
1383 | } | |
1384 | ||
1385 | static struct file_operations fops_kp = { | |
1386 | .read = read_enabled_file_bool, | |
1387 | .write = write_enabled_file_bool, | |
1388 | }; | |
1389 | ||
346fd59b SD |
1390 | static int __kprobes debugfs_kprobe_init(void) |
1391 | { | |
1392 | struct dentry *dir, *file; | |
bf8f6e5b | 1393 | unsigned int value = 1; |
346fd59b SD |
1394 | |
1395 | dir = debugfs_create_dir("kprobes", NULL); | |
1396 | if (!dir) | |
1397 | return -ENOMEM; | |
1398 | ||
e3869792 | 1399 | file = debugfs_create_file("list", 0444, dir, NULL, |
346fd59b SD |
1400 | &debugfs_kprobes_operations); |
1401 | if (!file) { | |
1402 | debugfs_remove(dir); | |
1403 | return -ENOMEM; | |
1404 | } | |
1405 | ||
bf8f6e5b AM |
1406 | file = debugfs_create_file("enabled", 0600, dir, |
1407 | &value, &fops_kp); | |
1408 | if (!file) { | |
1409 | debugfs_remove(dir); | |
1410 | return -ENOMEM; | |
1411 | } | |
1412 | ||
346fd59b SD |
1413 | return 0; |
1414 | } | |
1415 | ||
1416 | late_initcall(debugfs_kprobe_init); | |
1417 | #endif /* CONFIG_DEBUG_FS */ | |
1418 | ||
1419 | module_init(init_kprobes); | |
1da177e4 LT |
1420 | |
1421 | EXPORT_SYMBOL_GPL(register_kprobe); | |
1422 | EXPORT_SYMBOL_GPL(unregister_kprobe); | |
9861668f MH |
1423 | EXPORT_SYMBOL_GPL(register_kprobes); |
1424 | EXPORT_SYMBOL_GPL(unregister_kprobes); | |
1da177e4 LT |
1425 | EXPORT_SYMBOL_GPL(register_jprobe); |
1426 | EXPORT_SYMBOL_GPL(unregister_jprobe); | |
26b31c19 MH |
1427 | EXPORT_SYMBOL_GPL(register_jprobes); |
1428 | EXPORT_SYMBOL_GPL(unregister_jprobes); | |
1da177e4 | 1429 | EXPORT_SYMBOL_GPL(jprobe_return); |
b94cce92 HN |
1430 | EXPORT_SYMBOL_GPL(register_kretprobe); |
1431 | EXPORT_SYMBOL_GPL(unregister_kretprobe); | |
4a296e07 MH |
1432 | EXPORT_SYMBOL_GPL(register_kretprobes); |
1433 | EXPORT_SYMBOL_GPL(unregister_kretprobes); |