[FEATURE] add instrumentation US for x86
[kernel/swap-modules.git] / uprobe / swap_uprobes.c
1 /*
2  *  Dynamic Binary Instrumentation Module based on KProbes
3  *  modules/kprobe/dbi_uprobes.h
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) Samsung Electronics, 2006-2010
20  *
21  * 2008-2009    Alexey Gerenkov <a.gerenkov@samsung.com> User-Space
22  *              Probes initial implementation; Support x86/ARM/MIPS for both user and kernel spaces.
23  * 2010         Ekaterina Gorelkina <e.gorelkina@samsung.com>: redesign module for separating core and arch parts
24  *
25  */
26
27
28 #include "swap_uprobes.h"
29 #include "dbi_kdebug.h"
30
31 #include <asm/swap_uprobes.h>
32
33 #include <linux/hash.h>
34 #include <linux/mempolicy.h>
35 #include <linux/module.h>
36 #include <dbi_insn_slots.h>
37 #include <dbi_kprobes_deps.h>
38
39 enum {
40         UPROBE_HASH_BITS  = 10,
41         UPROBE_TABLE_SIZE = (1 << UPROBE_HASH_BITS)
42 };
43
44 struct hlist_head uprobe_insn_slot_table[UPROBE_TABLE_SIZE];
45 struct hlist_head uprobe_table[UPROBE_TABLE_SIZE];
46 struct hlist_head uprobe_insn_pages;
47
48 DEFINE_SPINLOCK(uretprobe_lock);        /* Protects uretprobe_inst_table */
49 static struct hlist_head uretprobe_inst_table[UPROBE_TABLE_SIZE];
50
51 #define DEBUG_PRINT_HASH_TABLE 0
52
53 #if DEBUG_PRINT_HASH_TABLE
54 void print_kprobe_hash_table(void)
55 {
56         int i;
57         struct hlist_head *head;
58         struct hlist_node *node;
59         struct kprobe *p;
60
61         // print uprobe table
62         for (i = 0; i < KPROBE_TABLE_SIZE; ++i) {
63                 head = &kprobe_table[i];
64                 swap_hlist_for_each_entry_rcu(p, node, head, is_hlist_arm) {
65                         printk("####### find K tgid=%u, addr=%x\n",
66                                         p->tgid, p->addr);
67                 }
68         }
69 }
70
71 void print_kretprobe_hash_table(void)
72 {
73         int i;
74         struct hlist_head *head;
75         struct hlist_node *node;
76         struct kprobe *p;
77
78         // print uprobe table
79         for (i = 0; i < KPROBE_TABLE_SIZE; ++i) {
80                 head = &kretprobe_inst_table[i];
81                 swap_hlist_for_each_entry_rcu(p, node, head, is_hlist_arm) {
82                         printk("####### find KR tgid=%u, addr=%x\n",
83                                         p->tgid, p->addr);
84                 }
85         }
86 }
87
88 void print_uprobe_hash_table(void)
89 {
90         int i;
91         struct hlist_head *head;
92         struct hlist_node *node;
93         struct kprobe *p;
94
95         // print uprobe table
96         for (i = 0; i < UPROBE_TABLE_SIZE; ++i) {
97                 head = &uprobe_insn_slot_table[i];
98                 swap_hlist_for_each_entry_rcu(p, node, head, is_hlist_arm) {
99                         printk("####### find U tgid=%u, addr=%x\n",
100                                         p->tgid, p->addr);
101                 }
102         }
103 }
104 #endif
105
106 /*
107  * Keep all fields in the uprobe consistent
108  */
109 static inline void copy_uprobe(struct kprobe *old_p, struct kprobe *p)
110 {
111         memcpy(&p->opcode, &old_p->opcode, sizeof(kprobe_opcode_t));
112         memcpy(&p->ainsn, &old_p->ainsn, sizeof(struct arch_specific_insn));
113         p->ss_addr = old_p->ss_addr;
114 #ifdef CONFIG_ARM
115         p->safe_arm = old_p->safe_arm;
116         p->safe_thumb = old_p->safe_thumb;
117 #endif
118 }
119
120 /*
121  * Aggregate handlers for multiple uprobes support - these handlers
122  * take care of invoking the individual uprobe handlers on p->list
123  */
124 static int aggr_pre_uhandler(struct kprobe *p, struct pt_regs *regs)
125 {
126         struct kprobe *kp;
127         int ret;
128
129         list_for_each_entry_rcu(kp, &p->list, list) {
130                 if (kp->pre_handler) {
131                         ret = kp->pre_handler(kp, regs);
132                         if (ret) {
133                                 return ret;
134                         }
135                 }
136         }
137
138         return 0;
139 }
140
141 static void aggr_post_uhandler(struct kprobe *p, struct pt_regs *regs, unsigned long flags)
142 {
143         struct kprobe *kp;
144
145         list_for_each_entry_rcu(kp, &p->list, list) {
146                 if (kp->post_handler) {
147                         kp->post_handler(kp, regs, flags);
148                 }
149         }
150 }
151
152 static int aggr_fault_uhandler(struct kprobe *p, struct pt_regs *regs, int trapnr)
153 {
154         return 0;
155 }
156
157 static int aggr_break_uhandler(struct kprobe *p, struct pt_regs *regs)
158 {
159         return 0;
160 }
161
162 /*
163  * Add the new probe to old_p->list. Fail if this is the
164  * second ujprobe at the address - two ujprobes can't coexist
165  */
166 static int add_new_uprobe(struct kprobe *old_p, struct kprobe *p)
167 {
168         if (p->break_handler) {
169                 if (old_p->break_handler) {
170                         return -EEXIST;
171                 }
172
173                 list_add_tail_rcu(&p->list, &old_p->list);
174                 old_p->break_handler = aggr_break_uhandler;
175         } else {
176                 list_add_rcu (&p->list, &old_p->list);
177         }
178
179         if (p->post_handler && !old_p->post_handler) {
180                 old_p->post_handler = aggr_post_uhandler;
181         }
182
183         return 0;
184 }
185
186 /*
187  * Fill in the required fields of the "manager uprobe". Replace the
188  * earlier uprobe in the hlist with the manager uprobe
189  */
190 static inline void add_aggr_uprobe(struct kprobe *ap, struct kprobe *p)
191 {
192         copy_uprobe(p, ap);
193
194         ap->addr = p->addr;
195         ap->pre_handler = aggr_pre_uhandler;
196         ap->fault_handler = aggr_fault_uhandler;
197
198         if (p->post_handler) {
199                 ap->post_handler = aggr_post_uhandler;
200         }
201
202         if (p->break_handler) {
203                 ap->break_handler = aggr_break_uhandler;
204         }
205
206         INIT_LIST_HEAD(&ap->list);
207         list_add_rcu(&p->list, &ap->list);
208
209         hlist_replace_rcu(&p->hlist, &ap->hlist);
210 }
211
212 /*
213  * This is the second or subsequent uprobe at the address - handle
214  * the intricacies
215  */
216 static int register_aggr_uprobe(struct kprobe *old_p, struct kprobe *p)
217 {
218         int ret = 0;
219         struct kprobe *ap;
220
221         if (old_p->pre_handler == aggr_pre_uhandler) {
222                 copy_uprobe(old_p, p);
223                 ret = add_new_uprobe(old_p, p);
224         } else {
225                 struct uprobe *uap = kzalloc(sizeof(*uap), GFP_KERNEL);
226                 if (!uap) {
227                         return -ENOMEM;
228                 }
229
230                 uap->task = kp2up(p)->task;
231                 ap = up2kp(uap);
232                 add_aggr_uprobe(ap, old_p);
233                 copy_uprobe(ap, p);
234                 ret = add_new_uprobe(ap, p);
235         }
236
237         return ret;
238 }
239
240 static void arm_uprobe(struct uprobe *p)
241 {
242         kprobe_opcode_t insn = BREAKPOINT_INSTRUCTION;
243         int ret = write_proc_vm_atomic(p->task, (unsigned long)p->kp.addr,
244                                        &insn, sizeof(insn));
245         if (!ret) {
246                 panic("arm_uprobe: failed to write memory "
247                       "tgid=%u addr=%p!\n", p->task->tgid, p->kp.addr);
248         }
249 }
250
251 void disarm_uprobe(struct kprobe *p, struct task_struct *task)
252 {
253         int ret = write_proc_vm_atomic(task, (unsigned long)p->addr,
254                                        &p->opcode, sizeof(p->opcode));
255         if (!ret) {
256                 panic("disarm_uprobe: failed to write memory "
257                       "tgid=%u, addr=%p!\n", task->tgid, p->addr);
258         }
259 }
260 EXPORT_SYMBOL_GPL(disarm_uprobe);
261
262 static void init_uprobes_insn_slots(void)
263 {
264         int i;
265         for (i = 0; i < UPROBE_TABLE_SIZE; ++i) {
266                 INIT_HLIST_HEAD(&uprobe_insn_slot_table[i]);
267         }
268 }
269
270 static void init_uprobe_table(void)
271 {
272         int i;
273         for (i = 0; i < UPROBE_TABLE_SIZE; ++i) {
274                 INIT_HLIST_HEAD(&uprobe_table[i]);
275         }
276 }
277
278 static void init_uretprobe_inst_table(void)
279 {
280         int i;
281         for (i = 0; i < UPROBE_TABLE_SIZE; ++i) {
282                 INIT_HLIST_HEAD (&uretprobe_inst_table[i]);
283         }
284 }
285
286 struct kprobe *get_ukprobe(void *addr, pid_t tgid)
287 {
288         struct hlist_head *head;
289         struct hlist_node *node;
290         struct kprobe *p;
291
292         head = &uprobe_table[hash_ptr(addr, UPROBE_HASH_BITS)];
293         swap_hlist_for_each_entry_rcu(p, node, head, hlist) {
294                 if (p->addr == addr && kp2up(p)->task->tgid == tgid) {
295                         return p;
296                 }
297         }
298
299         return NULL;
300 }
301
302 static void add_uprobe_table(struct kprobe *p)
303 {
304 #ifdef CONFIG_ARM
305         INIT_HLIST_NODE(&p->is_hlist_arm);
306         hlist_add_head_rcu(&p->is_hlist_arm, &uprobe_insn_slot_table[hash_ptr(p->ainsn.insn_arm, UPROBE_HASH_BITS)]);
307         INIT_HLIST_NODE(&p->is_hlist_thumb);
308         hlist_add_head_rcu(&p->is_hlist_thumb, &uprobe_insn_slot_table[hash_ptr(p->ainsn.insn_thumb, UPROBE_HASH_BITS)]);
309 #else /* CONFIG_ARM */
310         INIT_HLIST_NODE(&p->is_hlist);
311         hlist_add_head_rcu(&p->is_hlist, &uprobe_insn_slot_table[hash_ptr(p->ainsn.insn, UPROBE_HASH_BITS)]);
312 #endif /* CONFIG_ARM */
313 }
314
315 #ifdef CONFIG_ARM
316 static struct kprobe *get_ukprobe_bis_arm(void *addr, pid_t tgid)
317 {
318         struct hlist_head *head;
319         struct hlist_node *node;
320         struct kprobe *p;
321
322         /* TODO: test - two processes invokes instrumented function */
323         head = &uprobe_insn_slot_table[hash_ptr(addr, UPROBE_HASH_BITS)];
324         swap_hlist_for_each_entry_rcu(p, node, head, is_hlist_arm) {
325                 if (p->ainsn.insn == addr && kp2up(p)->task->tgid == tgid) {
326                         return p;
327                 }
328         }
329
330         return NULL;
331 }
332
333 static struct kprobe *get_ukprobe_bis_thumb(void *addr, pid_t tgid)
334 {
335         struct hlist_head *head;
336         struct hlist_node *node;
337         struct kprobe *p;
338
339         /* TODO: test - two processes invokes instrumented function */
340         head = &uprobe_insn_slot_table[hash_ptr(addr, UPROBE_HASH_BITS)];
341         swap_hlist_for_each_entry_rcu(p, node, head, is_hlist_thumb) {
342                 if (p->ainsn.insn == addr && kp2up(p)->task->tgid == tgid) {
343                         return p;
344                 }
345         }
346
347         return NULL;
348 }
349
350 struct kprobe *get_ukprobe_by_insn_slot(void *addr, pid_t tgid, struct pt_regs *regs)
351 {
352         return thumb_mode(regs) ?
353                         get_ukprobe_bis_thumb(addr - 0x1a, tgid) :
354                         get_ukprobe_bis_arm(addr - 4 * UPROBES_TRAMP_RET_BREAK_IDX, tgid);
355 }
356 #else /* CONFIG_ARM */
357 struct kprobe *get_ukprobe_by_insn_slot(void *addr, pid_t tgid, struct pt_regs *regs)
358 {
359         struct hlist_head *head;
360         struct hlist_node *node;
361         struct kprobe *p;
362
363         addr -= UPROBES_TRAMP_RET_BREAK_IDX;
364
365         /* TODO: test - two processes invokes instrumented function */
366         head = &uprobe_insn_slot_table[hash_ptr(addr, UPROBE_HASH_BITS)];
367         swap_hlist_for_each_entry_rcu(p, node, head, is_hlist) {
368                 if (p->ainsn.insn == addr && kp2up(p)->task->tgid == tgid) {
369                         return p;
370                 }
371         }
372
373         return NULL;
374 }
375 #endif /* CONFIG_ARM */
376
377
378 static void remove_uprobe(struct uprobe *up)
379 {
380         struct kprobe *p = &up->kp;
381         struct task_struct *task = up->task;
382
383 #ifdef CONFIG_ARM
384         free_insn_slot(up->sm, p->ainsn.insn_arm);
385         free_insn_slot(up->sm, p->ainsn.insn_thumb);
386 #else /* CONFIG_ARM */
387         free_insn_slot(up->sm, p->ainsn.insn);
388 #endif /* CONFIG_ARM */
389 }
390
391 static struct hlist_head *uretprobe_inst_table_head(void *hash_key)
392 {
393         return &uretprobe_inst_table[hash_ptr (hash_key, UPROBE_HASH_BITS)];
394 }
395
396 /* Called with uretprobe_lock held */
397 static void add_urp_inst(struct uretprobe_instance *ri)
398 {
399         /*
400          * Remove rp inst off the free list -
401          * Add it back when probed function returns
402          */
403         hlist_del(&ri->uflist);
404
405         /* Add rp inst onto table */
406         INIT_HLIST_NODE(&ri->hlist);
407         hlist_add_head(&ri->hlist, uretprobe_inst_table_head(ri->task->mm));
408
409         /* Also add this rp inst to the used list. */
410         INIT_HLIST_NODE(&ri->uflist);
411         hlist_add_head(&ri->uflist, &ri->rp->used_instances);
412 }
413
414 /* Called with uretprobe_lock held */
415 static void recycle_urp_inst(struct uretprobe_instance *ri)
416 {
417         if (ri->rp) {
418                 hlist_del(&ri->hlist);
419                 /* remove rp inst off the used list */
420                 hlist_del(&ri->uflist);
421                 /* put rp inst back onto the free list */
422                 INIT_HLIST_NODE(&ri->uflist);
423                 hlist_add_head(&ri->uflist, &ri->rp->free_instances);
424         }
425 }
426
427 /* Called with uretprobe_lock held */
428 static struct uretprobe_instance *get_used_urp_inst(struct uretprobe *rp)
429 {
430         struct hlist_node *node;
431         struct uretprobe_instance *ri;
432
433         swap_hlist_for_each_entry(ri, node, &rp->used_instances, uflist) {
434                 return ri;
435         }
436
437         return NULL;
438 }
439
440 /* Called with uretprobe_lock held */
441 struct uretprobe_instance *get_free_urp_inst_no_alloc(struct uretprobe *rp)
442 {
443         struct hlist_node *node;
444         struct uretprobe_instance *ri;
445
446         swap_hlist_for_each_entry(ri, node, &rp->free_instances, uflist) {
447                 return ri;
448         }
449
450         return NULL;
451 }
452
453 /* Called with uretprobe_lock held */
454 static void free_urp_inst(struct uretprobe *rp)
455 {
456         struct uretprobe_instance *ri;
457         while ((ri = get_free_urp_inst_no_alloc(rp)) != NULL) {
458                 hlist_del(&ri->uflist);
459                 kfree(ri);
460         }
461 }
462
463 #define COMMON_URP_NR 10
464
465 static int alloc_nodes_uretprobe(struct uretprobe *rp)
466 {
467         int alloc_nodes;
468         struct uretprobe_instance *inst;
469         int i;
470
471 #if 1//def CONFIG_PREEMPT
472         rp->maxactive += max(COMMON_URP_NR, 2 * NR_CPUS);
473 #else
474         rp->maxacpptive += NR_CPUS;
475 #endif
476         alloc_nodes = COMMON_URP_NR;
477
478         for (i = 0; i < alloc_nodes; ++i) {
479                 inst = kmalloc(sizeof(*inst), GFP_ATOMIC);
480                 if (inst == NULL) {
481                         free_urp_inst(rp);
482                         return -ENOMEM;
483                 }
484                 INIT_HLIST_NODE(&inst->uflist);
485                 hlist_add_head(&inst->uflist, &rp->free_instances);
486         }
487
488         return 0;
489 }
490
491 /* Called with uretprobe_lock held */
492 static struct uretprobe_instance *get_free_urp_inst(struct uretprobe *rp)
493 {
494         struct hlist_node *node;
495         struct uretprobe_instance *ri;
496
497         swap_hlist_for_each_entry(ri, node, &rp->free_instances, uflist) {
498                 return ri;
499         }
500
501         if (!alloc_nodes_uretprobe(rp)) {
502                 swap_hlist_for_each_entry(ri, node, &rp->free_instances, uflist) {
503                         return ri;
504                 }
505         }
506
507         return NULL;
508 }
509 // ===================================================================
510
511 int dbi_register_uprobe(struct uprobe *up)
512 {
513         int ret = 0;
514         struct kprobe *p, *old_p;
515
516         p = &up->kp;
517         if (!p->addr) {
518                 return -EINVAL;
519         }
520
521         DBPRINTF("p->addr = 0x%p p = 0x%p\n", p->addr, p);
522
523 // thumb address = address-1;
524 #if defined(CONFIG_ARM)
525         // TODO: must be corrected in 'bundle'
526         if ((unsigned long) p->addr & 0x01) {
527                 p->addr = (kprobe_opcode_t *)((unsigned long)p->addr & 0xfffffffe);
528         }
529 #endif
530
531         p->mod_refcounted = 0;
532         p->nmissed = 0;
533         INIT_LIST_HEAD(&p->list);
534 #ifdef KPROBES_PROFILE
535         p->start_tm.tv_sec = p->start_tm.tv_usec = 0;
536         p->hnd_tm_sum.tv_sec = p->hnd_tm_sum.tv_usec = 0;
537         p->count = 0;
538 #endif
539
540         // get the first item
541         old_p = get_ukprobe(p->addr, kp2up(p)->task->tgid);
542         if (old_p) {
543 #ifdef CONFIG_ARM
544                 p->safe_arm = old_p->safe_arm;
545                 p->safe_thumb = old_p->safe_thumb;
546 #endif
547                 ret = register_aggr_uprobe(old_p, p);
548                 if (!ret) {
549 //                      atomic_inc(&kprobe_count);
550                         add_uprobe_table(p);
551                 }
552                 DBPRINTF("goto out\n", ret);
553                 goto out;
554         }
555
556         ret = arch_prepare_uprobe(up, &uprobe_insn_pages);
557         if (ret) {
558                 DBPRINTF("goto out\n", ret);
559                 goto out;
560         }
561
562         DBPRINTF ("before out ret = 0x%x\n", ret);
563
564         // TODO: add uprobe (must be in function)
565         INIT_HLIST_NODE(&p->hlist);
566         hlist_add_head_rcu(&p->hlist, &uprobe_table[hash_ptr(p->addr, UPROBE_HASH_BITS)]);
567         add_uprobe_table(p);
568         arm_uprobe(up);
569
570 out:
571         DBPRINTF("out ret = 0x%x\n", ret);
572         return ret;
573 }
574
575 void dbi_unregister_uprobe(struct uprobe *up)
576 {
577         struct kprobe *p, *old_p, *list_p;
578         int cleanup_p;
579
580         p = &up->kp;
581         old_p = get_ukprobe(p->addr, kp2up(p)->task->tgid);
582         if (unlikely(!old_p)) {
583                 return;
584         }
585
586         if (p != old_p) {
587                 list_for_each_entry_rcu(list_p, &old_p->list, list) {
588                         if (list_p == p) {
589                                 /* uprobe p is a valid probe */
590                                 goto valid_p;
591                         }
592                 }
593
594                 return;
595         }
596
597 valid_p:
598         if ((old_p == p) || ((old_p->pre_handler == aggr_pre_uhandler) &&
599             (p->list.next == &old_p->list) && (p->list.prev == &old_p->list))) {
600                 /* Only probe on the hash list */
601                 disarm_uprobe(&up->kp, up->task);
602                 hlist_del_rcu(&old_p->hlist);
603                 cleanup_p = 1;
604         } else {
605                 list_del_rcu(&p->list);
606                 cleanup_p = 0;
607         }
608
609         if (cleanup_p) {
610                 if (p != old_p) {
611                         list_del_rcu(&p->list);
612                         kfree(old_p);
613                 }
614
615                 if (!in_atomic()) {
616                         synchronize_sched();
617                 }
618
619                 remove_uprobe(up);
620         } else {
621                 if (p->break_handler) {
622                         old_p->break_handler = NULL;
623                 }
624
625                 if (p->post_handler) {
626                         list_for_each_entry_rcu (list_p, &old_p->list, list) {
627                                 if (list_p->post_handler) {
628                                         cleanup_p = 2;
629                                         break;
630                                 }
631                         }
632
633                         if (cleanup_p == 0) {
634                                 old_p->post_handler = NULL;
635                         }
636                 }
637         }
638 }
639
640 int dbi_register_ujprobe(struct ujprobe *jp)
641 {
642         int ret = 0;
643
644         /* Todo: Verify probepoint is a function entry point */
645         jp->up.kp.pre_handler = setjmp_upre_handler;
646         jp->up.kp.break_handler = longjmp_break_uhandler;
647
648         ret = dbi_register_uprobe(&jp->up);
649
650         return ret;
651 }
652
653 void dbi_unregister_ujprobe(struct ujprobe *jp)
654 {
655         dbi_unregister_uprobe(&jp->up);
656         /*
657          * Here is an attempt to unregister even those probes that have not been
658          * installed (hence not added to the hlist).
659          * So if we try to delete them from the hlist we will get NULL pointer
660          * dereference error. That is why we check whether this node
661          * really belongs to the hlist.
662          */
663 #ifdef CONFIG_ARM
664         if (!(hlist_unhashed(&jp->up.kp.is_hlist_arm))) {
665                 hlist_del_rcu(&jp->up.kp.is_hlist_arm);
666         }
667         if (!(hlist_unhashed(&jp->up.kp.is_hlist_thumb))) {
668                 hlist_del_rcu(&jp->up.kp.is_hlist_thumb);
669         }
670 #else /* CONFIG_ARM */
671         if (!(hlist_unhashed(&jp->up.kp.is_hlist))) {
672                 hlist_del_rcu(&jp->up.kp.is_hlist);
673         }
674 #endif /* CONFIG_ARM */
675 }
676
677 int trampoline_uprobe_handler(struct kprobe *p, struct pt_regs *regs)
678 {
679         struct uretprobe_instance *ri = NULL;
680         struct hlist_head *head;
681         struct hlist_node *node, *tmp;
682         unsigned long flags, tramp_addr, orig_ret_addr = 0;
683
684         tramp_addr = arch_get_trampoline_addr(p, regs);
685         spin_lock_irqsave(&uretprobe_lock, flags);
686
687         head = uretprobe_inst_table_head(current->mm);
688
689         /*
690          * It is possible to have multiple instances associated with a given
691          * task either because an multiple functions in the call path
692          * have a return probe installed on them, and/or more then one
693          * return probe was registered for a target function.
694          *
695          * We can handle this because:
696          *     - instances are always inserted at the head of the list
697          *     - when multiple return probes are registered for the same
698          *       function, the first instance's ret_addr will point to the
699          *       real return address, and all the rest will point to
700          *       uretprobe_trampoline
701          */
702         swap_hlist_for_each_entry_safe(ri, node, tmp, head, hlist) {
703                 if (ri->task != current) {
704                         /* another task is sharing our hash bucket */
705                         continue;
706                 }
707
708                 if (ri->rp && ri->rp->handler) {
709                         ri->rp->handler(ri, regs, ri->rp->priv_arg);
710                 }
711
712                 orig_ret_addr = (unsigned long)ri->ret_addr;
713                 recycle_urp_inst(ri);
714
715                 if (orig_ret_addr != tramp_addr) {
716                         /*
717                          * This is the real return address. Any other
718                          * instances associated with this task are for
719                          * other calls deeper on the call stack
720                          */
721                         break;
722                 }
723         }
724
725         spin_unlock_irqrestore(&uretprobe_lock, flags);
726         arch_set_orig_ret_addr(orig_ret_addr, regs);
727
728         return 1;
729 }
730
731 static int pre_handler_uretprobe(struct kprobe *p, struct pt_regs *regs)
732 {
733         struct uprobe *up = container_of(p, struct uprobe, kp);
734         struct uretprobe *rp = container_of(up, struct uretprobe, up);
735         struct uretprobe_instance *ri;
736         unsigned long flags;
737
738         /* TODO: consider to only swap the RA after the last pre_handler fired */
739         spin_lock_irqsave(&uretprobe_lock, flags);
740
741         /* TODO: test - remove retprobe after func entry but before its exit */
742         if ((ri = get_free_urp_inst(rp)) != NULL) {
743                 ri->rp = rp;
744                 ri->task = current;
745
746                 arch_prepare_uretprobe(ri, regs);
747
748                 add_urp_inst(ri);
749         } else {
750                 ++rp->nmissed;
751         }
752
753         spin_unlock_irqrestore(&uretprobe_lock, flags);
754
755         return 0;
756 }
757
758 int dbi_register_uretprobe(struct uretprobe *rp)
759 {
760         int i, ret = 0;
761         struct uretprobe_instance *inst;
762
763         DBPRINTF ("START\n");
764
765         rp->up.kp.pre_handler = NULL;
766         rp->up.kp.post_handler = NULL;
767         rp->up.kp.fault_handler = NULL;
768         rp->up.kp.break_handler = NULL;
769
770         /* Establish function entry probe point */
771         ret = dbi_register_uprobe(&rp->up);
772         if (ret)
773                 return ret;
774
775         ret = arch_opcode_analysis_uretprobe(rp->up.kp.opcode);
776         if (ret)
777                 goto unregister;
778
779         /* Pre-allocate memory for max kretprobe instances */
780         if (rp->maxactive <= 0) {
781 #if 1//def CONFIG_PREEMPT
782                 rp->maxactive = max(10, 2 * NR_CPUS);
783 #else
784                 rp->maxactive = NR_CPUS;
785 #endif
786         }
787
788         INIT_HLIST_HEAD(&rp->used_instances);
789         INIT_HLIST_HEAD(&rp->free_instances);
790
791         for (i = 0; i < rp->maxactive; i++) {
792                 inst = kmalloc(sizeof(*inst), GFP_KERNEL);
793                 if (inst == NULL) {
794                         free_urp_inst(rp);
795                         ret = -ENOMEM;
796                         goto unregister;
797                 }
798
799                 INIT_HLIST_NODE(&inst->uflist);
800                 hlist_add_head(&inst->uflist, &rp->free_instances);
801         }
802
803         rp->nmissed = 0;
804         rp->up.kp.pre_handler = pre_handler_uretprobe;
805
806         return 0;
807
808 unregister:
809         dbi_unregister_uprobe(&rp->up);
810         return ret;
811 }
812
813 int dbi_disarm_urp_inst(struct uretprobe_instance *ri, struct task_struct *rm_task)
814 {
815         struct task_struct *task = rm_task ? rm_task : ri->task;
816         unsigned long *tramp;
817         unsigned long *sp = (unsigned long *)((long)ri->sp & ~1);
818         unsigned long *stack = sp - RETPROBE_STACK_DEPTH + 1;
819         unsigned long *found = NULL;
820         unsigned long *buf[RETPROBE_STACK_DEPTH];
821         int i, retval;
822
823         /* Understand function mode */
824         if ((long)ri->sp & 1) {
825                 tramp = (unsigned long *)
826                         ((unsigned long)ri->rp->up.kp.ainsn.insn + 0x1b);
827         } else {
828                 tramp = (unsigned long *)
829                         (ri->rp->up.kp.ainsn.insn + UPROBES_TRAMP_RET_BREAK_IDX);
830         }
831
832         retval = read_proc_vm_atomic(task, (unsigned long)stack, buf, sizeof(buf));
833         if (retval != sizeof(buf)) {
834                 printk("---> %s (%d/%d): failed to read stack from %08lx",
835                         task->comm, task->tgid, task->pid, (unsigned long)stack);
836                 retval = -EFAULT;
837                 goto out;
838         }
839
840         /* search the stack from the bottom */
841         for (i = RETPROBE_STACK_DEPTH - 1; i >= 0; i--) {
842                 if (buf[i] == tramp) {
843                         found = stack + i;
844                         break;
845                 }
846         }
847
848         if (found) {
849                 printk("---> %s (%d/%d): trampoline found at %08lx (%08lx /%+d) - %p\n",
850                                 task->comm, task->tgid, task->pid,
851                                 (unsigned long)found, (unsigned long)sp,
852                                 found - sp, ri->rp->up.kp.addr);
853                 retval = write_proc_vm_atomic(task, (unsigned long)found, &ri->ret_addr,
854                                 sizeof(ri->ret_addr));
855                 if (retval != sizeof(ri->ret_addr)) {
856                         printk("---> %s (%d/%d): failed to write value to %08lx",
857                                 task->comm, task->tgid, task->pid, (unsigned long)found);
858                         retval = -EFAULT;
859                 } else {
860                         retval = 0;
861                 }
862         } else {
863                 struct pt_regs *uregs = task_pt_regs(ri->task);
864                 unsigned long ra = dbi_get_ret_addr(uregs);
865                 if (ra == (unsigned long)tramp) {
866                         printk("---> %s (%d/%d): trampoline found at lr = %08lx - %p\n",
867                                         task->comm, task->tgid, task->pid, ra, ri->rp->up.kp.addr);
868                         dbi_set_ret_addr(uregs, (unsigned long)tramp);
869                         retval = 0;
870                 } else {
871                         printk("---> %s (%d/%d): trampoline NOT found at sp = %08lx, lr = %08lx - %p\n",
872                                         task->comm, task->tgid, task->pid,
873                                         (unsigned long)sp, ra, ri->rp->up.kp.addr);
874                         retval = -ENOENT;
875                 }
876         }
877
878 out:
879         return retval;
880 }
881
882 /* Called with uretprobe_lock held */
883 int dbi_disarm_urp_inst_for_task(struct task_struct *parent, struct task_struct *task)
884 {
885         struct uretprobe_instance *ri;
886         struct hlist_node *node, *tmp;
887         struct hlist_head *head = uretprobe_inst_table_head(parent->mm);
888
889         swap_hlist_for_each_entry_safe(ri, node, tmp, head, hlist) {
890                 if (parent == ri->task) {
891                         dbi_disarm_urp_inst(ri, task);
892                 }
893         }
894
895         return 0;
896 }
897 EXPORT_SYMBOL_GPL(dbi_disarm_urp_inst_for_task);
898
899 void dbi_unregister_uretprobe(struct uretprobe *rp)
900 {
901         unsigned long flags;
902         struct uretprobe_instance *ri;
903
904         spin_lock_irqsave (&uretprobe_lock, flags);
905
906         while ((ri = get_used_urp_inst(rp)) != NULL) {
907                 if (dbi_disarm_urp_inst(ri, NULL) != 0)
908                         /*panic*/printk("%s (%d/%d): cannot disarm urp instance (%08lx)\n",
909                                         ri->task->comm, ri->task->tgid, ri->task->pid,
910                                         (unsigned long)rp->up.kp.addr);
911                 recycle_urp_inst(ri);
912         }
913
914         if (hlist_empty(&rp->used_instances)) {
915                 struct kprobe *p = &rp->up.kp;
916 #ifdef CONFIG_ARM
917                 if (!(hlist_unhashed(&p->is_hlist_arm))) {
918                         hlist_del_rcu(&p->is_hlist_arm);
919                 }
920
921                 if (!(hlist_unhashed(&p->is_hlist_thumb))) {
922                         hlist_del_rcu(&p->is_hlist_thumb);
923                 }
924 #else /* CONFIG_ARM */
925                 if (!(hlist_unhashed(&p->is_hlist))) {
926                         hlist_del_rcu(&p->is_hlist);
927                 }
928 #endif /* CONFIG_ARM */
929         }
930
931         while ((ri = get_used_urp_inst(rp)) != NULL) {
932                 ri->rp = NULL;
933                 hlist_del(&ri->uflist);
934         }
935
936         spin_unlock_irqrestore(&uretprobe_lock, flags);
937         free_urp_inst(rp);
938
939         dbi_unregister_uprobe(&rp->up);
940 }
941
942 void dbi_unregister_all_uprobes(struct task_struct *task)
943 {
944         struct hlist_head *head;
945         struct hlist_node *node, *tnode;
946         struct kprobe *p;
947         int i;
948
949         for (i = 0; i < UPROBE_TABLE_SIZE; ++i) {
950                 head = &uprobe_table[i];
951                 swap_hlist_for_each_entry_safe(p, node, tnode, head, hlist) {
952                         if (kp2up(p)->task->tgid == task->tgid) {
953                                 struct uprobe *up = container_of(p, struct uprobe, kp);
954                                 printk("dbi_unregister_all_uprobes: delete uprobe at %p[%lx] for %s/%d\n",
955                                                 p->addr, (unsigned long)p->opcode, task->comm, task->pid);
956                                 dbi_unregister_uprobe(up);
957                         }
958                 }
959         }
960 }
961
962 void swap_ujprobe_return(void)
963 {
964         arch_ujprobe_return();
965 }
966 EXPORT_SYMBOL_GPL(swap_ujprobe_return);
967
968 static int __init init_uprobes(void)
969 {
970         init_uprobe_table();
971         init_uprobes_insn_slots();
972         init_uretprobe_inst_table();
973
974         return swap_arch_init_uprobes();
975 }
976
977 static void __exit exit_uprobes(void)
978 {
979         swap_arch_exit_uprobes();
980 }
981
982 EXPORT_SYMBOL_GPL(dbi_register_ujprobe);
983 EXPORT_SYMBOL_GPL(dbi_unregister_ujprobe);
984 EXPORT_SYMBOL_GPL(dbi_register_uretprobe);
985 EXPORT_SYMBOL_GPL(dbi_unregister_uretprobe);
986 EXPORT_SYMBOL_GPL(dbi_unregister_all_uprobes);
987
988 module_init(init_uprobes);
989 module_exit(exit_uprobes);
990
991 MODULE_LICENSE ("GPL");