]>
Commit | Line | Data |
---|---|---|
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. | |
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. | |
33 | */ | |
34 | #include <linux/kprobes.h> | |
35 | #include <linux/hash.h> | |
36 | #include <linux/init.h> | |
37 | #include <linux/slab.h> | |
38 | #include <linux/module.h> | |
39 | #include <linux/moduleloader.h> | |
40 | #include <linux/kallsyms.h> | |
41 | #include <linux/freezer.h> | |
42 | #include <linux/seq_file.h> | |
43 | #include <linux/debugfs.h> | |
44 | #include <asm-generic/sections.h> | |
45 | #include <asm/cacheflush.h> | |
46 | #include <asm/errno.h> | |
47 | #include <asm/kdebug.h> | |
48 | ||
49 | #define KPROBE_HASH_BITS 6 | |
50 | #define KPROBE_TABLE_SIZE (1 << KPROBE_HASH_BITS) | |
51 | ||
52 | ||
53 | /* | |
54 | * Some oddball architectures like 64bit powerpc have function descriptors | |
55 | * so this must be overridable. | |
56 | */ | |
57 | #ifndef kprobe_lookup_name | |
58 | #define kprobe_lookup_name(name, addr) \ | |
59 | addr = ((kprobe_opcode_t *)(kallsyms_lookup_name(name))) | |
60 | #endif | |
61 | ||
62 | static struct hlist_head kprobe_table[KPROBE_TABLE_SIZE]; | |
63 | static struct hlist_head kretprobe_inst_table[KPROBE_TABLE_SIZE]; | |
64 | static atomic_t kprobe_count; | |
65 | ||
66 | DEFINE_MUTEX(kprobe_mutex); /* Protects kprobe_table */ | |
67 | DEFINE_SPINLOCK(kretprobe_lock); /* Protects kretprobe_inst_table */ | |
68 | static DEFINE_PER_CPU(struct kprobe *, kprobe_instance) = NULL; | |
69 | ||
70 | static struct notifier_block kprobe_page_fault_nb = { | |
71 | .notifier_call = kprobe_exceptions_notify, | |
72 | .priority = 0x7fffffff /* we need to notified first */ | |
73 | }; | |
74 | ||
75 | #ifdef __ARCH_WANT_KPROBES_INSN_SLOT | |
76 | /* | |
77 | * kprobe->ainsn.insn points to the copy of the instruction to be | |
78 | * single-stepped. x86_64, POWER4 and above have no-exec support and | |
79 | * stepping on the instruction on a vmalloced/kmalloced/data page | |
80 | * is a recipe for disaster | |
81 | */ | |
82 | #define INSNS_PER_PAGE (PAGE_SIZE/(MAX_INSN_SIZE * sizeof(kprobe_opcode_t))) | |
83 | ||
84 | struct kprobe_insn_page { | |
85 | struct hlist_node hlist; | |
86 | kprobe_opcode_t *insns; /* Page of instruction slots */ | |
87 | char slot_used[INSNS_PER_PAGE]; | |
88 | int nused; | |
89 | int ngarbage; | |
90 | }; | |
91 | ||
92 | enum kprobe_slot_state { | |
93 | SLOT_CLEAN = 0, | |
94 | SLOT_DIRTY = 1, | |
95 | SLOT_USED = 2, | |
96 | }; | |
97 | ||
98 | static struct hlist_head kprobe_insn_pages; | |
99 | static int kprobe_garbage_slots; | |
100 | static int collect_garbage_slots(void); | |
101 | ||
102 | static int __kprobes check_safety(void) | |
103 | { | |
104 | int ret = 0; | |
105 | #if defined(CONFIG_PREEMPT) && defined(CONFIG_PM) | |
106 | ret = freeze_processes(); | |
107 | if (ret == 0) { | |
108 | struct task_struct *p, *q; | |
109 | do_each_thread(p, q) { | |
110 | if (p != current && p->state == TASK_RUNNING && | |
111 | p->pid != 0) { | |
112 | printk("Check failed: %s is running\n",p->comm); | |
113 | ret = -1; | |
114 | goto loop_end; | |
115 | } | |
116 | } while_each_thread(p, q); | |
117 | } | |
118 | loop_end: | |
119 | thaw_processes(); | |
120 | #else | |
121 | synchronize_sched(); | |
122 | #endif | |
123 | return ret; | |
124 | } | |
125 | ||
126 | /** | |
127 | * get_insn_slot() - Find a slot on an executable page for an instruction. | |
128 | * We allocate an executable page if there's no room on existing ones. | |
129 | */ | |
130 | kprobe_opcode_t __kprobes *get_insn_slot(void) | |
131 | { | |
132 | struct kprobe_insn_page *kip; | |
133 | struct hlist_node *pos; | |
134 | ||
135 | retry: | |
136 | hlist_for_each(pos, &kprobe_insn_pages) { | |
137 | kip = hlist_entry(pos, struct kprobe_insn_page, hlist); | |
138 | if (kip->nused < INSNS_PER_PAGE) { | |
139 | int i; | |
140 | for (i = 0; i < INSNS_PER_PAGE; i++) { | |
141 | if (kip->slot_used[i] == SLOT_CLEAN) { | |
142 | kip->slot_used[i] = SLOT_USED; | |
143 | kip->nused++; | |
144 | return kip->insns + (i * MAX_INSN_SIZE); | |
145 | } | |
146 | } | |
147 | /* Surprise! No unused slots. Fix kip->nused. */ | |
148 | kip->nused = INSNS_PER_PAGE; | |
149 | } | |
150 | } | |
151 | ||
152 | /* If there are any garbage slots, collect it and try again. */ | |
153 | if (kprobe_garbage_slots && collect_garbage_slots() == 0) { | |
154 | goto retry; | |
155 | } | |
156 | /* All out of space. Need to allocate a new page. Use slot 0. */ | |
157 | kip = kmalloc(sizeof(struct kprobe_insn_page), GFP_KERNEL); | |
158 | if (!kip) { | |
159 | return NULL; | |
160 | } | |
161 | ||
162 | /* | |
163 | * Use module_alloc so this page is within +/- 2GB of where the | |
164 | * kernel image and loaded module images reside. This is required | |
165 | * so x86_64 can correctly handle the %rip-relative fixups. | |
166 | */ | |
167 | kip->insns = module_alloc(PAGE_SIZE); | |
168 | if (!kip->insns) { | |
169 | kfree(kip); | |
170 | return NULL; | |
171 | } | |
172 | INIT_HLIST_NODE(&kip->hlist); | |
173 | hlist_add_head(&kip->hlist, &kprobe_insn_pages); | |
174 | memset(kip->slot_used, SLOT_CLEAN, INSNS_PER_PAGE); | |
175 | kip->slot_used[0] = SLOT_USED; | |
176 | kip->nused = 1; | |
177 | kip->ngarbage = 0; | |
178 | return kip->insns; | |
179 | } | |
180 | ||
181 | /* Return 1 if all garbages are collected, otherwise 0. */ | |
182 | static int __kprobes collect_one_slot(struct kprobe_insn_page *kip, int idx) | |
183 | { | |
184 | kip->slot_used[idx] = SLOT_CLEAN; | |
185 | kip->nused--; | |
186 | if (kip->nused == 0) { | |
187 | /* | |
188 | * Page is no longer in use. Free it unless | |
189 | * it's the last one. We keep the last one | |
190 | * so as not to have to set it up again the | |
191 | * next time somebody inserts a probe. | |
192 | */ | |
193 | hlist_del(&kip->hlist); | |
194 | if (hlist_empty(&kprobe_insn_pages)) { | |
195 | INIT_HLIST_NODE(&kip->hlist); | |
196 | hlist_add_head(&kip->hlist, | |
197 | &kprobe_insn_pages); | |
198 | } else { | |
199 | module_free(NULL, kip->insns); | |
200 | kfree(kip); | |
201 | } | |
202 | return 1; | |
203 | } | |
204 | return 0; | |
205 | } | |
206 | ||
207 | static int __kprobes collect_garbage_slots(void) | |
208 | { | |
209 | struct kprobe_insn_page *kip; | |
210 | struct hlist_node *pos, *next; | |
211 | ||
212 | /* Ensure no-one is preepmted on the garbages */ | |
213 | if (check_safety() != 0) | |
214 | return -EAGAIN; | |
215 | ||
216 | hlist_for_each_safe(pos, next, &kprobe_insn_pages) { | |
217 | int i; | |
218 | kip = hlist_entry(pos, struct kprobe_insn_page, hlist); | |
219 | if (kip->ngarbage == 0) | |
220 | continue; | |
221 | kip->ngarbage = 0; /* we will collect all garbages */ | |
222 | for (i = 0; i < INSNS_PER_PAGE; i++) { | |
223 | if (kip->slot_used[i] == SLOT_DIRTY && | |
224 | collect_one_slot(kip, i)) | |
225 | break; | |
226 | } | |
227 | } | |
228 | kprobe_garbage_slots = 0; | |
229 | return 0; | |
230 | } | |
231 | ||
232 | void __kprobes free_insn_slot(kprobe_opcode_t * slot, int dirty) | |
233 | { | |
234 | struct kprobe_insn_page *kip; | |
235 | struct hlist_node *pos; | |
236 | ||
237 | hlist_for_each(pos, &kprobe_insn_pages) { | |
238 | kip = hlist_entry(pos, struct kprobe_insn_page, hlist); | |
239 | if (kip->insns <= slot && | |
240 | slot < kip->insns + (INSNS_PER_PAGE * MAX_INSN_SIZE)) { | |
241 | int i = (slot - kip->insns) / MAX_INSN_SIZE; | |
242 | if (dirty) { | |
243 | kip->slot_used[i] = SLOT_DIRTY; | |
244 | kip->ngarbage++; | |
245 | } else { | |
246 | collect_one_slot(kip, i); | |
247 | } | |
248 | break; | |
249 | } | |
250 | } | |
251 | if (dirty && (++kprobe_garbage_slots > INSNS_PER_PAGE)) { | |
252 | collect_garbage_slots(); | |
253 | } | |
254 | } | |
255 | #endif | |
256 | ||
257 | /* We have preemption disabled.. so it is safe to use __ versions */ | |
258 | static inline void set_kprobe_instance(struct kprobe *kp) | |
259 | { | |
260 | __get_cpu_var(kprobe_instance) = kp; | |
261 | } | |
262 | ||
263 | static inline void reset_kprobe_instance(void) | |
264 | { | |
265 | __get_cpu_var(kprobe_instance) = NULL; | |
266 | } | |
267 | ||
268 | /* | |
269 | * This routine is called either: | |
270 | * - under the kprobe_mutex - during kprobe_[un]register() | |
271 | * OR | |
272 | * - with preemption disabled - from arch/xxx/kernel/kprobes.c | |
273 | */ | |
274 | struct kprobe __kprobes *get_kprobe(void *addr) | |
275 | { | |
276 | struct hlist_head *head; | |
277 | struct hlist_node *node; | |
278 | struct kprobe *p; | |
279 | ||
280 | head = &kprobe_table[hash_ptr(addr, KPROBE_HASH_BITS)]; | |
281 | hlist_for_each_entry_rcu(p, node, head, hlist) { | |
282 | if (p->addr == addr) | |
283 | return p; | |
284 | } | |
285 | return NULL; | |
286 | } | |
287 | ||
288 | /* | |
289 | * Aggregate handlers for multiple kprobes support - these handlers | |
290 | * take care of invoking the individual kprobe handlers on p->list | |
291 | */ | |
292 | static int __kprobes aggr_pre_handler(struct kprobe *p, struct pt_regs *regs) | |
293 | { | |
294 | struct kprobe *kp; | |
295 | ||
296 | list_for_each_entry_rcu(kp, &p->list, list) { | |
297 | if (kp->pre_handler) { | |
298 | set_kprobe_instance(kp); | |
299 | if (kp->pre_handler(kp, regs)) | |
300 | return 1; | |
301 | } | |
302 | reset_kprobe_instance(); | |
303 | } | |
304 | return 0; | |
305 | } | |
306 | ||
307 | static void __kprobes aggr_post_handler(struct kprobe *p, struct pt_regs *regs, | |
308 | unsigned long flags) | |
309 | { | |
310 | struct kprobe *kp; | |
311 | ||
312 | list_for_each_entry_rcu(kp, &p->list, list) { | |
313 | if (kp->post_handler) { | |
314 | set_kprobe_instance(kp); | |
315 | kp->post_handler(kp, regs, flags); | |
316 | reset_kprobe_instance(); | |
317 | } | |
318 | } | |
319 | return; | |
320 | } | |
321 | ||
322 | static int __kprobes aggr_fault_handler(struct kprobe *p, struct pt_regs *regs, | |
323 | int trapnr) | |
324 | { | |
325 | struct kprobe *cur = __get_cpu_var(kprobe_instance); | |
326 | ||
327 | /* | |
328 | * if we faulted "during" the execution of a user specified | |
329 | * probe handler, invoke just that probe's fault handler | |
330 | */ | |
331 | if (cur && cur->fault_handler) { | |
332 | if (cur->fault_handler(cur, regs, trapnr)) | |
333 | return 1; | |
334 | } | |
335 | return 0; | |
336 | } | |
337 | ||
338 | static int __kprobes aggr_break_handler(struct kprobe *p, struct pt_regs *regs) | |
339 | { | |
340 | struct kprobe *cur = __get_cpu_var(kprobe_instance); | |
341 | int ret = 0; | |
342 | ||
343 | if (cur && cur->break_handler) { | |
344 | if (cur->break_handler(cur, regs)) | |
345 | ret = 1; | |
346 | } | |
347 | reset_kprobe_instance(); | |
348 | return ret; | |
349 | } | |
350 | ||
351 | /* Walks the list and increments nmissed count for multiprobe case */ | |
352 | void __kprobes kprobes_inc_nmissed_count(struct kprobe *p) | |
353 | { | |
354 | struct kprobe *kp; | |
355 | if (p->pre_handler != aggr_pre_handler) { | |
356 | p->nmissed++; | |
357 | } else { | |
358 | list_for_each_entry_rcu(kp, &p->list, list) | |
359 | kp->nmissed++; | |
360 | } | |
361 | return; | |
362 | } | |
363 | ||
364 | /* Called with kretprobe_lock held */ | |
365 | struct kretprobe_instance __kprobes *get_free_rp_inst(struct kretprobe *rp) | |
366 | { | |
367 | struct hlist_node *node; | |
368 | struct kretprobe_instance *ri; | |
369 | hlist_for_each_entry(ri, node, &rp->free_instances, uflist) | |
370 | return ri; | |
371 | return NULL; | |
372 | } | |
373 | ||
374 | /* Called with kretprobe_lock held */ | |
375 | static struct kretprobe_instance __kprobes *get_used_rp_inst(struct kretprobe | |
376 | *rp) | |
377 | { | |
378 | struct hlist_node *node; | |
379 | struct kretprobe_instance *ri; | |
380 | hlist_for_each_entry(ri, node, &rp->used_instances, uflist) | |
381 | return ri; | |
382 | return NULL; | |
383 | } | |
384 | ||
385 | /* Called with kretprobe_lock held */ | |
386 | void __kprobes add_rp_inst(struct kretprobe_instance *ri) | |
387 | { | |
388 | /* | |
389 | * Remove rp inst off the free list - | |
390 | * Add it back when probed function returns | |
391 | */ | |
392 | hlist_del(&ri->uflist); | |
393 | ||
394 | /* Add rp inst onto table */ | |
395 | INIT_HLIST_NODE(&ri->hlist); | |
396 | hlist_add_head(&ri->hlist, | |
397 | &kretprobe_inst_table[hash_ptr(ri->task, KPROBE_HASH_BITS)]); | |
398 | ||
399 | /* Also add this rp inst to the used list. */ | |
400 | INIT_HLIST_NODE(&ri->uflist); | |
401 | hlist_add_head(&ri->uflist, &ri->rp->used_instances); | |
402 | } | |
403 | ||
404 | /* Called with kretprobe_lock held */ | |
405 | void __kprobes recycle_rp_inst(struct kretprobe_instance *ri, | |
406 | struct hlist_head *head) | |
407 | { | |
408 | /* remove rp inst off the rprobe_inst_table */ | |
409 | hlist_del(&ri->hlist); | |
410 | if (ri->rp) { | |
411 | /* remove rp inst off the used list */ | |
412 | hlist_del(&ri->uflist); | |
413 | /* put rp inst back onto the free list */ | |
414 | INIT_HLIST_NODE(&ri->uflist); | |
415 | hlist_add_head(&ri->uflist, &ri->rp->free_instances); | |
416 | } else | |
417 | /* Unregistering */ | |
418 | hlist_add_head(&ri->hlist, head); | |
419 | } | |
420 | ||
421 | struct hlist_head __kprobes *kretprobe_inst_table_head(struct task_struct *tsk) | |
422 | { | |
423 | return &kretprobe_inst_table[hash_ptr(tsk, KPROBE_HASH_BITS)]; | |
424 | } | |
425 | ||
426 | /* | |
427 | * This function is called from finish_task_switch when task tk becomes dead, | |
428 | * so that we can recycle any function-return probe instances associated | |
429 | * with this task. These left over instances represent probed functions | |
430 | * that have been called but will never return. | |
431 | */ | |
432 | void __kprobes kprobe_flush_task(struct task_struct *tk) | |
433 | { | |
434 | struct kretprobe_instance *ri; | |
435 | struct hlist_head *head, empty_rp; | |
436 | struct hlist_node *node, *tmp; | |
437 | unsigned long flags = 0; | |
438 | ||
439 | INIT_HLIST_HEAD(&empty_rp); | |
440 | spin_lock_irqsave(&kretprobe_lock, flags); | |
441 | head = kretprobe_inst_table_head(tk); | |
442 | hlist_for_each_entry_safe(ri, node, tmp, head, hlist) { | |
443 | if (ri->task == tk) | |
444 | recycle_rp_inst(ri, &empty_rp); | |
445 | } | |
446 | spin_unlock_irqrestore(&kretprobe_lock, flags); | |
447 | ||
448 | hlist_for_each_entry_safe(ri, node, tmp, &empty_rp, hlist) { | |
449 | hlist_del(&ri->hlist); | |
450 | kfree(ri); | |
451 | } | |
452 | } | |
453 | ||
454 | static inline void free_rp_inst(struct kretprobe *rp) | |
455 | { | |
456 | struct kretprobe_instance *ri; | |
457 | while ((ri = get_free_rp_inst(rp)) != NULL) { | |
458 | hlist_del(&ri->uflist); | |
459 | kfree(ri); | |
460 | } | |
461 | } | |
462 | ||
463 | /* | |
464 | * Keep all fields in the kprobe consistent | |
465 | */ | |
466 | static inline void copy_kprobe(struct kprobe *old_p, struct kprobe *p) | |
467 | { | |
468 | memcpy(&p->opcode, &old_p->opcode, sizeof(kprobe_opcode_t)); | |
469 | memcpy(&p->ainsn, &old_p->ainsn, sizeof(struct arch_specific_insn)); | |
470 | } | |
471 | ||
472 | /* | |
473 | * Add the new probe to old_p->list. Fail if this is the | |
474 | * second jprobe at the address - two jprobes can't coexist | |
475 | */ | |
476 | static int __kprobes add_new_kprobe(struct kprobe *old_p, struct kprobe *p) | |
477 | { | |
478 | if (p->break_handler) { | |
479 | if (old_p->break_handler) | |
480 | return -EEXIST; | |
481 | list_add_tail_rcu(&p->list, &old_p->list); | |
482 | old_p->break_handler = aggr_break_handler; | |
483 | } else | |
484 | list_add_rcu(&p->list, &old_p->list); | |
485 | if (p->post_handler && !old_p->post_handler) | |
486 | old_p->post_handler = aggr_post_handler; | |
487 | return 0; | |
488 | } | |
489 | ||
490 | /* | |
491 | * Fill in the required fields of the "manager kprobe". Replace the | |
492 | * earlier kprobe in the hlist with the manager kprobe | |
493 | */ | |
494 | static inline void add_aggr_kprobe(struct kprobe *ap, struct kprobe *p) | |
495 | { | |
496 | copy_kprobe(p, ap); | |
497 | flush_insn_slot(ap); | |
498 | ap->addr = p->addr; | |
499 | ap->pre_handler = aggr_pre_handler; | |
500 | ap->fault_handler = aggr_fault_handler; | |
501 | if (p->post_handler) | |
502 | ap->post_handler = aggr_post_handler; | |
503 | if (p->break_handler) | |
504 | ap->break_handler = aggr_break_handler; | |
505 | ||
506 | INIT_LIST_HEAD(&ap->list); | |
507 | list_add_rcu(&p->list, &ap->list); | |
508 | ||
509 | hlist_replace_rcu(&p->hlist, &ap->hlist); | |
510 | } | |
511 | ||
512 | /* | |
513 | * This is the second or subsequent kprobe at the address - handle | |
514 | * the intricacies | |
515 | */ | |
516 | static int __kprobes register_aggr_kprobe(struct kprobe *old_p, | |
517 | struct kprobe *p) | |
518 | { | |
519 | int ret = 0; | |
520 | struct kprobe *ap; | |
521 | ||
522 | if (old_p->pre_handler == aggr_pre_handler) { | |
523 | copy_kprobe(old_p, p); | |
524 | ret = add_new_kprobe(old_p, p); | |
525 | } else { | |
526 | ap = kzalloc(sizeof(struct kprobe), GFP_KERNEL); | |
527 | if (!ap) | |
528 | return -ENOMEM; | |
529 | add_aggr_kprobe(ap, old_p); | |
530 | copy_kprobe(ap, p); | |
531 | ret = add_new_kprobe(ap, p); | |
532 | } | |
533 | return ret; | |
534 | } | |
535 | ||
536 | static int __kprobes in_kprobes_functions(unsigned long addr) | |
537 | { | |
538 | if (addr >= (unsigned long)__kprobes_text_start | |
539 | && addr < (unsigned long)__kprobes_text_end) | |
540 | return -EINVAL; | |
541 | return 0; | |
542 | } | |
543 | ||
544 | static int __kprobes __register_kprobe(struct kprobe *p, | |
545 | unsigned long called_from) | |
546 | { | |
547 | int ret = 0; | |
548 | struct kprobe *old_p; | |
549 | struct module *probed_mod; | |
550 | ||
551 | /* | |
552 | * If we have a symbol_name argument look it up, | |
553 | * and add it to the address. That way the addr | |
554 | * field can either be global or relative to a symbol. | |
555 | */ | |
556 | if (p->symbol_name) { | |
557 | if (p->addr) | |
558 | return -EINVAL; | |
559 | kprobe_lookup_name(p->symbol_name, p->addr); | |
560 | } | |
561 | ||
562 | if (!p->addr) | |
563 | return -EINVAL; | |
564 | p->addr = (kprobe_opcode_t *)(((char *)p->addr)+ p->offset); | |
565 | ||
566 | if ((!kernel_text_address((unsigned long) p->addr)) || | |
567 | in_kprobes_functions((unsigned long) p->addr)) | |
568 | return -EINVAL; | |
569 | ||
570 | p->mod_refcounted = 0; | |
571 | /* Check are we probing a module */ | |
572 | if ((probed_mod = module_text_address((unsigned long) p->addr))) { | |
573 | struct module *calling_mod = module_text_address(called_from); | |
574 | /* We must allow modules to probe themself and | |
575 | * in this case avoid incrementing the module refcount, | |
576 | * so as to allow unloading of self probing modules. | |
577 | */ | |
578 | if (calling_mod && (calling_mod != probed_mod)) { | |
579 | if (unlikely(!try_module_get(probed_mod))) | |
580 | return -EINVAL; | |
581 | p->mod_refcounted = 1; | |
582 | } else | |
583 | probed_mod = NULL; | |
584 | } | |
585 | ||
586 | p->nmissed = 0; | |
587 | mutex_lock(&kprobe_mutex); | |
588 | old_p = get_kprobe(p->addr); | |
589 | if (old_p) { | |
590 | ret = register_aggr_kprobe(old_p, p); | |
591 | if (!ret) | |
592 | atomic_inc(&kprobe_count); | |
593 | goto out; | |
594 | } | |
595 | ||
596 | if ((ret = arch_prepare_kprobe(p)) != 0) | |
597 | goto out; | |
598 | ||
599 | INIT_HLIST_NODE(&p->hlist); | |
600 | hlist_add_head_rcu(&p->hlist, | |
601 | &kprobe_table[hash_ptr(p->addr, KPROBE_HASH_BITS)]); | |
602 | ||
603 | if (atomic_add_return(1, &kprobe_count) == \ | |
604 | (ARCH_INACTIVE_KPROBE_COUNT + 1)) | |
605 | register_page_fault_notifier(&kprobe_page_fault_nb); | |
606 | ||
607 | arch_arm_kprobe(p); | |
608 | ||
609 | out: | |
610 | mutex_unlock(&kprobe_mutex); | |
611 | ||
612 | if (ret && probed_mod) | |
613 | module_put(probed_mod); | |
614 | return ret; | |
615 | } | |
616 | ||
617 | int __kprobes register_kprobe(struct kprobe *p) | |
618 | { | |
619 | return __register_kprobe(p, | |
620 | (unsigned long)__builtin_return_address(0)); | |
621 | } | |
622 | ||
623 | void __kprobes unregister_kprobe(struct kprobe *p) | |
624 | { | |
625 | struct module *mod; | |
626 | struct kprobe *old_p, *list_p; | |
627 | int cleanup_p; | |
628 | ||
629 | mutex_lock(&kprobe_mutex); | |
630 | old_p = get_kprobe(p->addr); | |
631 | if (unlikely(!old_p)) { | |
632 | mutex_unlock(&kprobe_mutex); | |
633 | return; | |
634 | } | |
635 | if (p != old_p) { | |
636 | list_for_each_entry_rcu(list_p, &old_p->list, list) | |
637 | if (list_p == p) | |
638 | /* kprobe p is a valid probe */ | |
639 | goto valid_p; | |
640 | mutex_unlock(&kprobe_mutex); | |
641 | return; | |
642 | } | |
643 | valid_p: | |
644 | if ((old_p == p) || ((old_p->pre_handler == aggr_pre_handler) && | |
645 | (p->list.next == &old_p->list) && | |
646 | (p->list.prev == &old_p->list))) { | |
647 | /* Only probe on the hash list */ | |
648 | arch_disarm_kprobe(p); | |
649 | hlist_del_rcu(&old_p->hlist); | |
650 | cleanup_p = 1; | |
651 | } else { | |
652 | list_del_rcu(&p->list); | |
653 | cleanup_p = 0; | |
654 | } | |
655 | ||
656 | mutex_unlock(&kprobe_mutex); | |
657 | ||
658 | synchronize_sched(); | |
659 | if (p->mod_refcounted && | |
660 | (mod = module_text_address((unsigned long)p->addr))) | |
661 | module_put(mod); | |
662 | ||
663 | if (cleanup_p) { | |
664 | if (p != old_p) { | |
665 | list_del_rcu(&p->list); | |
666 | kfree(old_p); | |
667 | } | |
668 | arch_remove_kprobe(p); | |
669 | } else { | |
670 | mutex_lock(&kprobe_mutex); | |
671 | if (p->break_handler) | |
672 | old_p->break_handler = NULL; | |
673 | if (p->post_handler){ | |
674 | list_for_each_entry_rcu(list_p, &old_p->list, list){ | |
675 | if (list_p->post_handler){ | |
676 | cleanup_p = 2; | |
677 | break; | |
678 | } | |
679 | } | |
680 | if (cleanup_p == 0) | |
681 | old_p->post_handler = NULL; | |
682 | } | |
683 | mutex_unlock(&kprobe_mutex); | |
684 | } | |
685 | ||
686 | /* Call unregister_page_fault_notifier() | |
687 | * if no probes are active | |
688 | */ | |
689 | mutex_lock(&kprobe_mutex); | |
690 | if (atomic_add_return(-1, &kprobe_count) == \ | |
691 | ARCH_INACTIVE_KPROBE_COUNT) | |
692 | unregister_page_fault_notifier(&kprobe_page_fault_nb); | |
693 | mutex_unlock(&kprobe_mutex); | |
694 | return; | |
695 | } | |
696 | ||
697 | static struct notifier_block kprobe_exceptions_nb = { | |
698 | .notifier_call = kprobe_exceptions_notify, | |
699 | .priority = 0x7fffffff /* we need to be notified first */ | |
700 | }; | |
701 | ||
702 | ||
703 | int __kprobes register_jprobe(struct jprobe *jp) | |
704 | { | |
705 | /* Todo: Verify probepoint is a function entry point */ | |
706 | jp->kp.pre_handler = setjmp_pre_handler; | |
707 | jp->kp.break_handler = longjmp_break_handler; | |
708 | ||
709 | return __register_kprobe(&jp->kp, | |
710 | (unsigned long)__builtin_return_address(0)); | |
711 | } | |
712 | ||
713 | void __kprobes unregister_jprobe(struct jprobe *jp) | |
714 | { | |
715 | unregister_kprobe(&jp->kp); | |
716 | } | |
717 | ||
718 | #ifdef ARCH_SUPPORTS_KRETPROBES | |
719 | ||
720 | /* | |
721 | * This kprobe pre_handler is registered with every kretprobe. When probe | |
722 | * hits it will set up the return probe. | |
723 | */ | |
724 | static int __kprobes pre_handler_kretprobe(struct kprobe *p, | |
725 | struct pt_regs *regs) | |
726 | { | |
727 | struct kretprobe *rp = container_of(p, struct kretprobe, kp); | |
728 | unsigned long flags = 0; | |
729 | ||
730 | /*TODO: consider to only swap the RA after the last pre_handler fired */ | |
731 | spin_lock_irqsave(&kretprobe_lock, flags); | |
732 | arch_prepare_kretprobe(rp, regs); | |
733 | spin_unlock_irqrestore(&kretprobe_lock, flags); | |
734 | return 0; | |
735 | } | |
736 | ||
737 | int __kprobes register_kretprobe(struct kretprobe *rp) | |
738 | { | |
739 | int ret = 0; | |
740 | struct kretprobe_instance *inst; | |
741 | int i; | |
742 | ||
743 | rp->kp.pre_handler = pre_handler_kretprobe; | |
744 | rp->kp.post_handler = NULL; | |
745 | rp->kp.fault_handler = NULL; | |
746 | rp->kp.break_handler = NULL; | |
747 | ||
748 | /* Pre-allocate memory for max kretprobe instances */ | |
749 | if (rp->maxactive <= 0) { | |
750 | #ifdef CONFIG_PREEMPT | |
751 | rp->maxactive = max(10, 2 * NR_CPUS); | |
752 | #else | |
753 | rp->maxactive = NR_CPUS; | |
754 | #endif | |
755 | } | |
756 | INIT_HLIST_HEAD(&rp->used_instances); | |
757 | INIT_HLIST_HEAD(&rp->free_instances); | |
758 | for (i = 0; i < rp->maxactive; i++) { | |
759 | inst = kmalloc(sizeof(struct kretprobe_instance), GFP_KERNEL); | |
760 | if (inst == NULL) { | |
761 | free_rp_inst(rp); | |
762 | return -ENOMEM; | |
763 | } | |
764 | INIT_HLIST_NODE(&inst->uflist); | |
765 | hlist_add_head(&inst->uflist, &rp->free_instances); | |
766 | } | |
767 | ||
768 | rp->nmissed = 0; | |
769 | /* Establish function entry probe point */ | |
770 | if ((ret = __register_kprobe(&rp->kp, | |
771 | (unsigned long)__builtin_return_address(0))) != 0) | |
772 | free_rp_inst(rp); | |
773 | return ret; | |
774 | } | |
775 | ||
776 | #else /* ARCH_SUPPORTS_KRETPROBES */ | |
777 | ||
778 | int __kprobes register_kretprobe(struct kretprobe *rp) | |
779 | { | |
780 | return -ENOSYS; | |
781 | } | |
782 | ||
783 | static int __kprobes pre_handler_kretprobe(struct kprobe *p, | |
784 | struct pt_regs *regs) | |
785 | { | |
786 | return 0; | |
787 | } | |
788 | ||
789 | #endif /* ARCH_SUPPORTS_KRETPROBES */ | |
790 | ||
791 | void __kprobes unregister_kretprobe(struct kretprobe *rp) | |
792 | { | |
793 | unsigned long flags; | |
794 | struct kretprobe_instance *ri; | |
795 | ||
796 | unregister_kprobe(&rp->kp); | |
797 | /* No race here */ | |
798 | spin_lock_irqsave(&kretprobe_lock, flags); | |
799 | while ((ri = get_used_rp_inst(rp)) != NULL) { | |
800 | ri->rp = NULL; | |
801 | hlist_del(&ri->uflist); | |
802 | } | |
803 | spin_unlock_irqrestore(&kretprobe_lock, flags); | |
804 | free_rp_inst(rp); | |
805 | } | |
806 | ||
807 | static int __init init_kprobes(void) | |
808 | { | |
809 | int i, err = 0; | |
810 | ||
811 | /* FIXME allocate the probe table, currently defined statically */ | |
812 | /* initialize all list heads */ | |
813 | for (i = 0; i < KPROBE_TABLE_SIZE; i++) { | |
814 | INIT_HLIST_HEAD(&kprobe_table[i]); | |
815 | INIT_HLIST_HEAD(&kretprobe_inst_table[i]); | |
816 | } | |
817 | atomic_set(&kprobe_count, 0); | |
818 | ||
819 | err = arch_init_kprobes(); | |
820 | if (!err) | |
821 | err = register_die_notifier(&kprobe_exceptions_nb); | |
822 | ||
823 | return err; | |
824 | } | |
825 | ||
826 | #ifdef CONFIG_DEBUG_FS | |
827 | static void __kprobes report_probe(struct seq_file *pi, struct kprobe *p, | |
828 | const char *sym, int offset,char *modname) | |
829 | { | |
830 | char *kprobe_type; | |
831 | ||
832 | if (p->pre_handler == pre_handler_kretprobe) | |
833 | kprobe_type = "r"; | |
834 | else if (p->pre_handler == setjmp_pre_handler) | |
835 | kprobe_type = "j"; | |
836 | else | |
837 | kprobe_type = "k"; | |
838 | if (sym) | |
839 | seq_printf(pi, "%p %s %s+0x%x %s\n", p->addr, kprobe_type, | |
840 | sym, offset, (modname ? modname : " ")); | |
841 | else | |
842 | seq_printf(pi, "%p %s %p\n", p->addr, kprobe_type, p->addr); | |
843 | } | |
844 | ||
845 | static void __kprobes *kprobe_seq_start(struct seq_file *f, loff_t *pos) | |
846 | { | |
847 | return (*pos < KPROBE_TABLE_SIZE) ? pos : NULL; | |
848 | } | |
849 | ||
850 | static void __kprobes *kprobe_seq_next(struct seq_file *f, void *v, loff_t *pos) | |
851 | { | |
852 | (*pos)++; | |
853 | if (*pos >= KPROBE_TABLE_SIZE) | |
854 | return NULL; | |
855 | return pos; | |
856 | } | |
857 | ||
858 | static void __kprobes kprobe_seq_stop(struct seq_file *f, void *v) | |
859 | { | |
860 | /* Nothing to do */ | |
861 | } | |
862 | ||
863 | static int __kprobes show_kprobe_addr(struct seq_file *pi, void *v) | |
864 | { | |
865 | struct hlist_head *head; | |
866 | struct hlist_node *node; | |
867 | struct kprobe *p, *kp; | |
868 | const char *sym = NULL; | |
869 | unsigned int i = *(loff_t *) v; | |
870 | unsigned long size, offset = 0; | |
871 | char *modname, namebuf[128]; | |
872 | ||
873 | head = &kprobe_table[i]; | |
874 | preempt_disable(); | |
875 | hlist_for_each_entry_rcu(p, node, head, hlist) { | |
876 | sym = kallsyms_lookup((unsigned long)p->addr, &size, | |
877 | &offset, &modname, namebuf); | |
878 | if (p->pre_handler == aggr_pre_handler) { | |
879 | list_for_each_entry_rcu(kp, &p->list, list) | |
880 | report_probe(pi, kp, sym, offset, modname); | |
881 | } else | |
882 | report_probe(pi, p, sym, offset, modname); | |
883 | } | |
884 | preempt_enable(); | |
885 | return 0; | |
886 | } | |
887 | ||
888 | static struct seq_operations kprobes_seq_ops = { | |
889 | .start = kprobe_seq_start, | |
890 | .next = kprobe_seq_next, | |
891 | .stop = kprobe_seq_stop, | |
892 | .show = show_kprobe_addr | |
893 | }; | |
894 | ||
895 | static int __kprobes kprobes_open(struct inode *inode, struct file *filp) | |
896 | { | |
897 | return seq_open(filp, &kprobes_seq_ops); | |
898 | } | |
899 | ||
900 | static struct file_operations debugfs_kprobes_operations = { | |
901 | .open = kprobes_open, | |
902 | .read = seq_read, | |
903 | .llseek = seq_lseek, | |
904 | .release = seq_release, | |
905 | }; | |
906 | ||
907 | static int __kprobes debugfs_kprobe_init(void) | |
908 | { | |
909 | struct dentry *dir, *file; | |
910 | ||
911 | dir = debugfs_create_dir("kprobes", NULL); | |
912 | if (!dir) | |
913 | return -ENOMEM; | |
914 | ||
915 | file = debugfs_create_file("list", 0444, dir , 0 , | |
916 | &debugfs_kprobes_operations); | |
917 | if (!file) { | |
918 | debugfs_remove(dir); | |
919 | return -ENOMEM; | |
920 | } | |
921 | ||
922 | return 0; | |
923 | } | |
924 | ||
925 | late_initcall(debugfs_kprobe_init); | |
926 | #endif /* CONFIG_DEBUG_FS */ | |
927 | ||
928 | module_init(init_kprobes); | |
929 | ||
930 | EXPORT_SYMBOL_GPL(register_kprobe); | |
931 | EXPORT_SYMBOL_GPL(unregister_kprobe); | |
932 | EXPORT_SYMBOL_GPL(register_jprobe); | |
933 | EXPORT_SYMBOL_GPL(unregister_jprobe); | |
934 | EXPORT_SYMBOL_GPL(jprobe_return); | |
935 | EXPORT_SYMBOL_GPL(register_kretprobe); | |
936 | EXPORT_SYMBOL_GPL(unregister_kretprobe); |