Loading...
1/*
2 * uprobes-based tracing events
3 *
4 * This program is free software; you can redistribute it and/or modify
5 * it under the terms of the GNU General Public License version 2 as
6 * published by the Free Software Foundation.
7 *
8 * This program is distributed in the hope that it will be useful,
9 * but WITHOUT ANY WARRANTY; without even the implied warranty of
10 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
11 * GNU General Public License for more details.
12 *
13 * You should have received a copy of the GNU General Public License
14 * along with this program; if not, write to the Free Software
15 * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA
16 *
17 * Copyright (C) IBM Corporation, 2010-2012
18 * Author: Srikar Dronamraju <srikar@linux.vnet.ibm.com>
19 */
20
21#include <linux/module.h>
22#include <linux/uaccess.h>
23#include <linux/uprobes.h>
24#include <linux/namei.h>
25#include <linux/string.h>
26
27#include "trace_probe.h"
28
29#define UPROBE_EVENT_SYSTEM "uprobes"
30
31struct uprobe_trace_entry_head {
32 struct trace_entry ent;
33 unsigned long vaddr[];
34};
35
36#define SIZEOF_TRACE_ENTRY(is_return) \
37 (sizeof(struct uprobe_trace_entry_head) + \
38 sizeof(unsigned long) * (is_return ? 2 : 1))
39
40#define DATAOF_TRACE_ENTRY(entry, is_return) \
41 ((void*)(entry) + SIZEOF_TRACE_ENTRY(is_return))
42
43struct trace_uprobe_filter {
44 rwlock_t rwlock;
45 int nr_systemwide;
46 struct list_head perf_events;
47};
48
49/*
50 * uprobe event core functions
51 */
52struct trace_uprobe {
53 struct list_head list;
54 struct trace_uprobe_filter filter;
55 struct uprobe_consumer consumer;
56 struct inode *inode;
57 char *filename;
58 unsigned long offset;
59 unsigned long nhit;
60 struct trace_probe tp;
61};
62
63#define SIZEOF_TRACE_UPROBE(n) \
64 (offsetof(struct trace_uprobe, tp.args) + \
65 (sizeof(struct probe_arg) * (n)))
66
67static int register_uprobe_event(struct trace_uprobe *tu);
68static int unregister_uprobe_event(struct trace_uprobe *tu);
69
70static DEFINE_MUTEX(uprobe_lock);
71static LIST_HEAD(uprobe_list);
72
73struct uprobe_dispatch_data {
74 struct trace_uprobe *tu;
75 unsigned long bp_addr;
76};
77
78static int uprobe_dispatcher(struct uprobe_consumer *con, struct pt_regs *regs);
79static int uretprobe_dispatcher(struct uprobe_consumer *con,
80 unsigned long func, struct pt_regs *regs);
81
82#ifdef CONFIG_STACK_GROWSUP
83static unsigned long adjust_stack_addr(unsigned long addr, unsigned int n)
84{
85 return addr - (n * sizeof(long));
86}
87#else
88static unsigned long adjust_stack_addr(unsigned long addr, unsigned int n)
89{
90 return addr + (n * sizeof(long));
91}
92#endif
93
94static unsigned long get_user_stack_nth(struct pt_regs *regs, unsigned int n)
95{
96 unsigned long ret;
97 unsigned long addr = user_stack_pointer(regs);
98
99 addr = adjust_stack_addr(addr, n);
100
101 if (copy_from_user(&ret, (void __force __user *) addr, sizeof(ret)))
102 return 0;
103
104 return ret;
105}
106
107/*
108 * Uprobes-specific fetch functions
109 */
110#define DEFINE_FETCH_stack(type) \
111static void FETCH_FUNC_NAME(stack, type)(struct pt_regs *regs, \
112 void *offset, void *dest) \
113{ \
114 *(type *)dest = (type)get_user_stack_nth(regs, \
115 ((unsigned long)offset)); \
116}
117DEFINE_BASIC_FETCH_FUNCS(stack)
118/* No string on the stack entry */
119#define fetch_stack_string NULL
120#define fetch_stack_string_size NULL
121
122#define DEFINE_FETCH_memory(type) \
123static void FETCH_FUNC_NAME(memory, type)(struct pt_regs *regs, \
124 void *addr, void *dest) \
125{ \
126 type retval; \
127 void __user *vaddr = (void __force __user *) addr; \
128 \
129 if (copy_from_user(&retval, vaddr, sizeof(type))) \
130 *(type *)dest = 0; \
131 else \
132 *(type *) dest = retval; \
133}
134DEFINE_BASIC_FETCH_FUNCS(memory)
135/*
136 * Fetch a null-terminated string. Caller MUST set *(u32 *)dest with max
137 * length and relative data location.
138 */
139static void FETCH_FUNC_NAME(memory, string)(struct pt_regs *regs,
140 void *addr, void *dest)
141{
142 long ret;
143 u32 rloc = *(u32 *)dest;
144 int maxlen = get_rloc_len(rloc);
145 u8 *dst = get_rloc_data(dest);
146 void __user *src = (void __force __user *) addr;
147
148 if (!maxlen)
149 return;
150
151 ret = strncpy_from_user(dst, src, maxlen);
152
153 if (ret < 0) { /* Failed to fetch string */
154 ((u8 *)get_rloc_data(dest))[0] = '\0';
155 *(u32 *)dest = make_data_rloc(0, get_rloc_offs(rloc));
156 } else {
157 *(u32 *)dest = make_data_rloc(ret, get_rloc_offs(rloc));
158 }
159}
160
161static void FETCH_FUNC_NAME(memory, string_size)(struct pt_regs *regs,
162 void *addr, void *dest)
163{
164 int len;
165 void __user *vaddr = (void __force __user *) addr;
166
167 len = strnlen_user(vaddr, MAX_STRING_SIZE);
168
169 if (len == 0 || len > MAX_STRING_SIZE) /* Failed to check length */
170 *(u32 *)dest = 0;
171 else
172 *(u32 *)dest = len;
173}
174
175static unsigned long translate_user_vaddr(void *file_offset)
176{
177 unsigned long base_addr;
178 struct uprobe_dispatch_data *udd;
179
180 udd = (void *) current->utask->vaddr;
181
182 base_addr = udd->bp_addr - udd->tu->offset;
183 return base_addr + (unsigned long)file_offset;
184}
185
186#define DEFINE_FETCH_file_offset(type) \
187static void FETCH_FUNC_NAME(file_offset, type)(struct pt_regs *regs, \
188 void *offset, void *dest)\
189{ \
190 void *vaddr = (void *)translate_user_vaddr(offset); \
191 \
192 FETCH_FUNC_NAME(memory, type)(regs, vaddr, dest); \
193}
194DEFINE_BASIC_FETCH_FUNCS(file_offset)
195DEFINE_FETCH_file_offset(string)
196DEFINE_FETCH_file_offset(string_size)
197
198/* Fetch type information table */
199static const struct fetch_type uprobes_fetch_type_table[] = {
200 /* Special types */
201 [FETCH_TYPE_STRING] = __ASSIGN_FETCH_TYPE("string", string, string,
202 sizeof(u32), 1, "__data_loc char[]"),
203 [FETCH_TYPE_STRSIZE] = __ASSIGN_FETCH_TYPE("string_size", u32,
204 string_size, sizeof(u32), 0, "u32"),
205 /* Basic types */
206 ASSIGN_FETCH_TYPE(u8, u8, 0),
207 ASSIGN_FETCH_TYPE(u16, u16, 0),
208 ASSIGN_FETCH_TYPE(u32, u32, 0),
209 ASSIGN_FETCH_TYPE(u64, u64, 0),
210 ASSIGN_FETCH_TYPE(s8, u8, 1),
211 ASSIGN_FETCH_TYPE(s16, u16, 1),
212 ASSIGN_FETCH_TYPE(s32, u32, 1),
213 ASSIGN_FETCH_TYPE(s64, u64, 1),
214
215 ASSIGN_FETCH_TYPE_END
216};
217
218static inline void init_trace_uprobe_filter(struct trace_uprobe_filter *filter)
219{
220 rwlock_init(&filter->rwlock);
221 filter->nr_systemwide = 0;
222 INIT_LIST_HEAD(&filter->perf_events);
223}
224
225static inline bool uprobe_filter_is_empty(struct trace_uprobe_filter *filter)
226{
227 return !filter->nr_systemwide && list_empty(&filter->perf_events);
228}
229
230static inline bool is_ret_probe(struct trace_uprobe *tu)
231{
232 return tu->consumer.ret_handler != NULL;
233}
234
235/*
236 * Allocate new trace_uprobe and initialize it (including uprobes).
237 */
238static struct trace_uprobe *
239alloc_trace_uprobe(const char *group, const char *event, int nargs, bool is_ret)
240{
241 struct trace_uprobe *tu;
242
243 if (!event || !is_good_name(event))
244 return ERR_PTR(-EINVAL);
245
246 if (!group || !is_good_name(group))
247 return ERR_PTR(-EINVAL);
248
249 tu = kzalloc(SIZEOF_TRACE_UPROBE(nargs), GFP_KERNEL);
250 if (!tu)
251 return ERR_PTR(-ENOMEM);
252
253 tu->tp.call.class = &tu->tp.class;
254 tu->tp.call.name = kstrdup(event, GFP_KERNEL);
255 if (!tu->tp.call.name)
256 goto error;
257
258 tu->tp.class.system = kstrdup(group, GFP_KERNEL);
259 if (!tu->tp.class.system)
260 goto error;
261
262 INIT_LIST_HEAD(&tu->list);
263 INIT_LIST_HEAD(&tu->tp.files);
264 tu->consumer.handler = uprobe_dispatcher;
265 if (is_ret)
266 tu->consumer.ret_handler = uretprobe_dispatcher;
267 init_trace_uprobe_filter(&tu->filter);
268 return tu;
269
270error:
271 kfree(tu->tp.call.name);
272 kfree(tu);
273
274 return ERR_PTR(-ENOMEM);
275}
276
277static void free_trace_uprobe(struct trace_uprobe *tu)
278{
279 int i;
280
281 for (i = 0; i < tu->tp.nr_args; i++)
282 traceprobe_free_probe_arg(&tu->tp.args[i]);
283
284 iput(tu->inode);
285 kfree(tu->tp.call.class->system);
286 kfree(tu->tp.call.name);
287 kfree(tu->filename);
288 kfree(tu);
289}
290
291static struct trace_uprobe *find_probe_event(const char *event, const char *group)
292{
293 struct trace_uprobe *tu;
294
295 list_for_each_entry(tu, &uprobe_list, list)
296 if (strcmp(trace_event_name(&tu->tp.call), event) == 0 &&
297 strcmp(tu->tp.call.class->system, group) == 0)
298 return tu;
299
300 return NULL;
301}
302
303/* Unregister a trace_uprobe and probe_event: call with locking uprobe_lock */
304static int unregister_trace_uprobe(struct trace_uprobe *tu)
305{
306 int ret;
307
308 ret = unregister_uprobe_event(tu);
309 if (ret)
310 return ret;
311
312 list_del(&tu->list);
313 free_trace_uprobe(tu);
314 return 0;
315}
316
317/* Register a trace_uprobe and probe_event */
318static int register_trace_uprobe(struct trace_uprobe *tu)
319{
320 struct trace_uprobe *old_tu;
321 int ret;
322
323 mutex_lock(&uprobe_lock);
324
325 /* register as an event */
326 old_tu = find_probe_event(trace_event_name(&tu->tp.call),
327 tu->tp.call.class->system);
328 if (old_tu) {
329 /* delete old event */
330 ret = unregister_trace_uprobe(old_tu);
331 if (ret)
332 goto end;
333 }
334
335 ret = register_uprobe_event(tu);
336 if (ret) {
337 pr_warn("Failed to register probe event(%d)\n", ret);
338 goto end;
339 }
340
341 list_add_tail(&tu->list, &uprobe_list);
342
343end:
344 mutex_unlock(&uprobe_lock);
345
346 return ret;
347}
348
349/*
350 * Argument syntax:
351 * - Add uprobe: p|r[:[GRP/]EVENT] PATH:OFFSET [FETCHARGS]
352 *
353 * - Remove uprobe: -:[GRP/]EVENT
354 */
355static int create_trace_uprobe(int argc, char **argv)
356{
357 struct trace_uprobe *tu;
358 struct inode *inode;
359 char *arg, *event, *group, *filename;
360 char buf[MAX_EVENT_NAME_LEN];
361 struct path path;
362 unsigned long offset;
363 bool is_delete, is_return;
364 int i, ret;
365
366 inode = NULL;
367 ret = 0;
368 is_delete = false;
369 is_return = false;
370 event = NULL;
371 group = NULL;
372
373 /* argc must be >= 1 */
374 if (argv[0][0] == '-')
375 is_delete = true;
376 else if (argv[0][0] == 'r')
377 is_return = true;
378 else if (argv[0][0] != 'p') {
379 pr_info("Probe definition must be started with 'p', 'r' or '-'.\n");
380 return -EINVAL;
381 }
382
383 if (argv[0][1] == ':') {
384 event = &argv[0][2];
385 arg = strchr(event, '/');
386
387 if (arg) {
388 group = event;
389 event = arg + 1;
390 event[-1] = '\0';
391
392 if (strlen(group) == 0) {
393 pr_info("Group name is not specified\n");
394 return -EINVAL;
395 }
396 }
397 if (strlen(event) == 0) {
398 pr_info("Event name is not specified\n");
399 return -EINVAL;
400 }
401 }
402 if (!group)
403 group = UPROBE_EVENT_SYSTEM;
404
405 if (is_delete) {
406 int ret;
407
408 if (!event) {
409 pr_info("Delete command needs an event name.\n");
410 return -EINVAL;
411 }
412 mutex_lock(&uprobe_lock);
413 tu = find_probe_event(event, group);
414
415 if (!tu) {
416 mutex_unlock(&uprobe_lock);
417 pr_info("Event %s/%s doesn't exist.\n", group, event);
418 return -ENOENT;
419 }
420 /* delete an event */
421 ret = unregister_trace_uprobe(tu);
422 mutex_unlock(&uprobe_lock);
423 return ret;
424 }
425
426 if (argc < 2) {
427 pr_info("Probe point is not specified.\n");
428 return -EINVAL;
429 }
430 if (isdigit(argv[1][0])) {
431 pr_info("probe point must be have a filename.\n");
432 return -EINVAL;
433 }
434 arg = strchr(argv[1], ':');
435 if (!arg) {
436 ret = -EINVAL;
437 goto fail_address_parse;
438 }
439
440 *arg++ = '\0';
441 filename = argv[1];
442 ret = kern_path(filename, LOOKUP_FOLLOW, &path);
443 if (ret)
444 goto fail_address_parse;
445
446 inode = igrab(d_inode(path.dentry));
447 path_put(&path);
448
449 if (!inode || !S_ISREG(inode->i_mode)) {
450 ret = -EINVAL;
451 goto fail_address_parse;
452 }
453
454 ret = kstrtoul(arg, 0, &offset);
455 if (ret)
456 goto fail_address_parse;
457
458 argc -= 2;
459 argv += 2;
460
461 /* setup a probe */
462 if (!event) {
463 char *tail;
464 char *ptr;
465
466 tail = kstrdup(kbasename(filename), GFP_KERNEL);
467 if (!tail) {
468 ret = -ENOMEM;
469 goto fail_address_parse;
470 }
471
472 ptr = strpbrk(tail, ".-_");
473 if (ptr)
474 *ptr = '\0';
475
476 snprintf(buf, MAX_EVENT_NAME_LEN, "%c_%s_0x%lx", 'p', tail, offset);
477 event = buf;
478 kfree(tail);
479 }
480
481 tu = alloc_trace_uprobe(group, event, argc, is_return);
482 if (IS_ERR(tu)) {
483 pr_info("Failed to allocate trace_uprobe.(%d)\n", (int)PTR_ERR(tu));
484 ret = PTR_ERR(tu);
485 goto fail_address_parse;
486 }
487 tu->offset = offset;
488 tu->inode = inode;
489 tu->filename = kstrdup(filename, GFP_KERNEL);
490
491 if (!tu->filename) {
492 pr_info("Failed to allocate filename.\n");
493 ret = -ENOMEM;
494 goto error;
495 }
496
497 /* parse arguments */
498 ret = 0;
499 for (i = 0; i < argc && i < MAX_TRACE_ARGS; i++) {
500 struct probe_arg *parg = &tu->tp.args[i];
501
502 /* Increment count for freeing args in error case */
503 tu->tp.nr_args++;
504
505 /* Parse argument name */
506 arg = strchr(argv[i], '=');
507 if (arg) {
508 *arg++ = '\0';
509 parg->name = kstrdup(argv[i], GFP_KERNEL);
510 } else {
511 arg = argv[i];
512 /* If argument name is omitted, set "argN" */
513 snprintf(buf, MAX_EVENT_NAME_LEN, "arg%d", i + 1);
514 parg->name = kstrdup(buf, GFP_KERNEL);
515 }
516
517 if (!parg->name) {
518 pr_info("Failed to allocate argument[%d] name.\n", i);
519 ret = -ENOMEM;
520 goto error;
521 }
522
523 if (!is_good_name(parg->name)) {
524 pr_info("Invalid argument[%d] name: %s\n", i, parg->name);
525 ret = -EINVAL;
526 goto error;
527 }
528
529 if (traceprobe_conflict_field_name(parg->name, tu->tp.args, i)) {
530 pr_info("Argument[%d] name '%s' conflicts with "
531 "another field.\n", i, argv[i]);
532 ret = -EINVAL;
533 goto error;
534 }
535
536 /* Parse fetch argument */
537 ret = traceprobe_parse_probe_arg(arg, &tu->tp.size, parg,
538 is_return, false,
539 uprobes_fetch_type_table);
540 if (ret) {
541 pr_info("Parse error at argument[%d]. (%d)\n", i, ret);
542 goto error;
543 }
544 }
545
546 ret = register_trace_uprobe(tu);
547 if (ret)
548 goto error;
549 return 0;
550
551error:
552 free_trace_uprobe(tu);
553 return ret;
554
555fail_address_parse:
556 iput(inode);
557
558 pr_info("Failed to parse address or file.\n");
559
560 return ret;
561}
562
563static int cleanup_all_probes(void)
564{
565 struct trace_uprobe *tu;
566 int ret = 0;
567
568 mutex_lock(&uprobe_lock);
569 while (!list_empty(&uprobe_list)) {
570 tu = list_entry(uprobe_list.next, struct trace_uprobe, list);
571 ret = unregister_trace_uprobe(tu);
572 if (ret)
573 break;
574 }
575 mutex_unlock(&uprobe_lock);
576 return ret;
577}
578
579/* Probes listing interfaces */
580static void *probes_seq_start(struct seq_file *m, loff_t *pos)
581{
582 mutex_lock(&uprobe_lock);
583 return seq_list_start(&uprobe_list, *pos);
584}
585
586static void *probes_seq_next(struct seq_file *m, void *v, loff_t *pos)
587{
588 return seq_list_next(v, &uprobe_list, pos);
589}
590
591static void probes_seq_stop(struct seq_file *m, void *v)
592{
593 mutex_unlock(&uprobe_lock);
594}
595
596static int probes_seq_show(struct seq_file *m, void *v)
597{
598 struct trace_uprobe *tu = v;
599 char c = is_ret_probe(tu) ? 'r' : 'p';
600 int i;
601
602 seq_printf(m, "%c:%s/%s", c, tu->tp.call.class->system,
603 trace_event_name(&tu->tp.call));
604 seq_printf(m, " %s:", tu->filename);
605
606 /* Don't print "0x (null)" when offset is 0 */
607 if (tu->offset) {
608 seq_printf(m, "0x%p", (void *)tu->offset);
609 } else {
610 switch (sizeof(void *)) {
611 case 4:
612 seq_printf(m, "0x00000000");
613 break;
614 case 8:
615 default:
616 seq_printf(m, "0x0000000000000000");
617 break;
618 }
619 }
620
621 for (i = 0; i < tu->tp.nr_args; i++)
622 seq_printf(m, " %s=%s", tu->tp.args[i].name, tu->tp.args[i].comm);
623
624 seq_putc(m, '\n');
625 return 0;
626}
627
628static const struct seq_operations probes_seq_op = {
629 .start = probes_seq_start,
630 .next = probes_seq_next,
631 .stop = probes_seq_stop,
632 .show = probes_seq_show
633};
634
635static int probes_open(struct inode *inode, struct file *file)
636{
637 int ret;
638
639 if ((file->f_mode & FMODE_WRITE) && (file->f_flags & O_TRUNC)) {
640 ret = cleanup_all_probes();
641 if (ret)
642 return ret;
643 }
644
645 return seq_open(file, &probes_seq_op);
646}
647
648static ssize_t probes_write(struct file *file, const char __user *buffer,
649 size_t count, loff_t *ppos)
650{
651 return traceprobe_probes_write(file, buffer, count, ppos, create_trace_uprobe);
652}
653
654static const struct file_operations uprobe_events_ops = {
655 .owner = THIS_MODULE,
656 .open = probes_open,
657 .read = seq_read,
658 .llseek = seq_lseek,
659 .release = seq_release,
660 .write = probes_write,
661};
662
663/* Probes profiling interfaces */
664static int probes_profile_seq_show(struct seq_file *m, void *v)
665{
666 struct trace_uprobe *tu = v;
667
668 seq_printf(m, " %s %-44s %15lu\n", tu->filename,
669 trace_event_name(&tu->tp.call), tu->nhit);
670 return 0;
671}
672
673static const struct seq_operations profile_seq_op = {
674 .start = probes_seq_start,
675 .next = probes_seq_next,
676 .stop = probes_seq_stop,
677 .show = probes_profile_seq_show
678};
679
680static int profile_open(struct inode *inode, struct file *file)
681{
682 return seq_open(file, &profile_seq_op);
683}
684
685static const struct file_operations uprobe_profile_ops = {
686 .owner = THIS_MODULE,
687 .open = profile_open,
688 .read = seq_read,
689 .llseek = seq_lseek,
690 .release = seq_release,
691};
692
693struct uprobe_cpu_buffer {
694 struct mutex mutex;
695 void *buf;
696};
697static struct uprobe_cpu_buffer __percpu *uprobe_cpu_buffer;
698static int uprobe_buffer_refcnt;
699
700static int uprobe_buffer_init(void)
701{
702 int cpu, err_cpu;
703
704 uprobe_cpu_buffer = alloc_percpu(struct uprobe_cpu_buffer);
705 if (uprobe_cpu_buffer == NULL)
706 return -ENOMEM;
707
708 for_each_possible_cpu(cpu) {
709 struct page *p = alloc_pages_node(cpu_to_node(cpu),
710 GFP_KERNEL, 0);
711 if (p == NULL) {
712 err_cpu = cpu;
713 goto err;
714 }
715 per_cpu_ptr(uprobe_cpu_buffer, cpu)->buf = page_address(p);
716 mutex_init(&per_cpu_ptr(uprobe_cpu_buffer, cpu)->mutex);
717 }
718
719 return 0;
720
721err:
722 for_each_possible_cpu(cpu) {
723 if (cpu == err_cpu)
724 break;
725 free_page((unsigned long)per_cpu_ptr(uprobe_cpu_buffer, cpu)->buf);
726 }
727
728 free_percpu(uprobe_cpu_buffer);
729 return -ENOMEM;
730}
731
732static int uprobe_buffer_enable(void)
733{
734 int ret = 0;
735
736 BUG_ON(!mutex_is_locked(&event_mutex));
737
738 if (uprobe_buffer_refcnt++ == 0) {
739 ret = uprobe_buffer_init();
740 if (ret < 0)
741 uprobe_buffer_refcnt--;
742 }
743
744 return ret;
745}
746
747static void uprobe_buffer_disable(void)
748{
749 int cpu;
750
751 BUG_ON(!mutex_is_locked(&event_mutex));
752
753 if (--uprobe_buffer_refcnt == 0) {
754 for_each_possible_cpu(cpu)
755 free_page((unsigned long)per_cpu_ptr(uprobe_cpu_buffer,
756 cpu)->buf);
757
758 free_percpu(uprobe_cpu_buffer);
759 uprobe_cpu_buffer = NULL;
760 }
761}
762
763static struct uprobe_cpu_buffer *uprobe_buffer_get(void)
764{
765 struct uprobe_cpu_buffer *ucb;
766 int cpu;
767
768 cpu = raw_smp_processor_id();
769 ucb = per_cpu_ptr(uprobe_cpu_buffer, cpu);
770
771 /*
772 * Use per-cpu buffers for fastest access, but we might migrate
773 * so the mutex makes sure we have sole access to it.
774 */
775 mutex_lock(&ucb->mutex);
776
777 return ucb;
778}
779
780static void uprobe_buffer_put(struct uprobe_cpu_buffer *ucb)
781{
782 mutex_unlock(&ucb->mutex);
783}
784
785static void __uprobe_trace_func(struct trace_uprobe *tu,
786 unsigned long func, struct pt_regs *regs,
787 struct uprobe_cpu_buffer *ucb, int dsize,
788 struct trace_event_file *trace_file)
789{
790 struct uprobe_trace_entry_head *entry;
791 struct ring_buffer_event *event;
792 struct ring_buffer *buffer;
793 void *data;
794 int size, esize;
795 struct trace_event_call *call = &tu->tp.call;
796
797 WARN_ON(call != trace_file->event_call);
798
799 if (WARN_ON_ONCE(tu->tp.size + dsize > PAGE_SIZE))
800 return;
801
802 if (trace_trigger_soft_disabled(trace_file))
803 return;
804
805 esize = SIZEOF_TRACE_ENTRY(is_ret_probe(tu));
806 size = esize + tu->tp.size + dsize;
807 event = trace_event_buffer_lock_reserve(&buffer, trace_file,
808 call->event.type, size, 0, 0);
809 if (!event)
810 return;
811
812 entry = ring_buffer_event_data(event);
813 if (is_ret_probe(tu)) {
814 entry->vaddr[0] = func;
815 entry->vaddr[1] = instruction_pointer(regs);
816 data = DATAOF_TRACE_ENTRY(entry, true);
817 } else {
818 entry->vaddr[0] = instruction_pointer(regs);
819 data = DATAOF_TRACE_ENTRY(entry, false);
820 }
821
822 memcpy(data, ucb->buf, tu->tp.size + dsize);
823
824 event_trigger_unlock_commit(trace_file, buffer, event, entry, 0, 0);
825}
826
827/* uprobe handler */
828static int uprobe_trace_func(struct trace_uprobe *tu, struct pt_regs *regs,
829 struct uprobe_cpu_buffer *ucb, int dsize)
830{
831 struct event_file_link *link;
832
833 if (is_ret_probe(tu))
834 return 0;
835
836 rcu_read_lock();
837 list_for_each_entry_rcu(link, &tu->tp.files, list)
838 __uprobe_trace_func(tu, 0, regs, ucb, dsize, link->file);
839 rcu_read_unlock();
840
841 return 0;
842}
843
844static void uretprobe_trace_func(struct trace_uprobe *tu, unsigned long func,
845 struct pt_regs *regs,
846 struct uprobe_cpu_buffer *ucb, int dsize)
847{
848 struct event_file_link *link;
849
850 rcu_read_lock();
851 list_for_each_entry_rcu(link, &tu->tp.files, list)
852 __uprobe_trace_func(tu, func, regs, ucb, dsize, link->file);
853 rcu_read_unlock();
854}
855
856/* Event entry printers */
857static enum print_line_t
858print_uprobe_event(struct trace_iterator *iter, int flags, struct trace_event *event)
859{
860 struct uprobe_trace_entry_head *entry;
861 struct trace_seq *s = &iter->seq;
862 struct trace_uprobe *tu;
863 u8 *data;
864 int i;
865
866 entry = (struct uprobe_trace_entry_head *)iter->ent;
867 tu = container_of(event, struct trace_uprobe, tp.call.event);
868
869 if (is_ret_probe(tu)) {
870 trace_seq_printf(s, "%s: (0x%lx <- 0x%lx)",
871 trace_event_name(&tu->tp.call),
872 entry->vaddr[1], entry->vaddr[0]);
873 data = DATAOF_TRACE_ENTRY(entry, true);
874 } else {
875 trace_seq_printf(s, "%s: (0x%lx)",
876 trace_event_name(&tu->tp.call),
877 entry->vaddr[0]);
878 data = DATAOF_TRACE_ENTRY(entry, false);
879 }
880
881 for (i = 0; i < tu->tp.nr_args; i++) {
882 struct probe_arg *parg = &tu->tp.args[i];
883
884 if (!parg->type->print(s, parg->name, data + parg->offset, entry))
885 goto out;
886 }
887
888 trace_seq_putc(s, '\n');
889
890 out:
891 return trace_handle_return(s);
892}
893
894typedef bool (*filter_func_t)(struct uprobe_consumer *self,
895 enum uprobe_filter_ctx ctx,
896 struct mm_struct *mm);
897
898static int
899probe_event_enable(struct trace_uprobe *tu, struct trace_event_file *file,
900 filter_func_t filter)
901{
902 bool enabled = trace_probe_is_enabled(&tu->tp);
903 struct event_file_link *link = NULL;
904 int ret;
905
906 if (file) {
907 if (tu->tp.flags & TP_FLAG_PROFILE)
908 return -EINTR;
909
910 link = kmalloc(sizeof(*link), GFP_KERNEL);
911 if (!link)
912 return -ENOMEM;
913
914 link->file = file;
915 list_add_tail_rcu(&link->list, &tu->tp.files);
916
917 tu->tp.flags |= TP_FLAG_TRACE;
918 } else {
919 if (tu->tp.flags & TP_FLAG_TRACE)
920 return -EINTR;
921
922 tu->tp.flags |= TP_FLAG_PROFILE;
923 }
924
925 WARN_ON(!uprobe_filter_is_empty(&tu->filter));
926
927 if (enabled)
928 return 0;
929
930 ret = uprobe_buffer_enable();
931 if (ret)
932 goto err_flags;
933
934 tu->consumer.filter = filter;
935 ret = uprobe_register(tu->inode, tu->offset, &tu->consumer);
936 if (ret)
937 goto err_buffer;
938
939 return 0;
940
941 err_buffer:
942 uprobe_buffer_disable();
943
944 err_flags:
945 if (file) {
946 list_del(&link->list);
947 kfree(link);
948 tu->tp.flags &= ~TP_FLAG_TRACE;
949 } else {
950 tu->tp.flags &= ~TP_FLAG_PROFILE;
951 }
952 return ret;
953}
954
955static void
956probe_event_disable(struct trace_uprobe *tu, struct trace_event_file *file)
957{
958 if (!trace_probe_is_enabled(&tu->tp))
959 return;
960
961 if (file) {
962 struct event_file_link *link;
963
964 link = find_event_file_link(&tu->tp, file);
965 if (!link)
966 return;
967
968 list_del_rcu(&link->list);
969 /* synchronize with u{,ret}probe_trace_func */
970 synchronize_sched();
971 kfree(link);
972
973 if (!list_empty(&tu->tp.files))
974 return;
975 }
976
977 WARN_ON(!uprobe_filter_is_empty(&tu->filter));
978
979 uprobe_unregister(tu->inode, tu->offset, &tu->consumer);
980 tu->tp.flags &= file ? ~TP_FLAG_TRACE : ~TP_FLAG_PROFILE;
981
982 uprobe_buffer_disable();
983}
984
985static int uprobe_event_define_fields(struct trace_event_call *event_call)
986{
987 int ret, i, size;
988 struct uprobe_trace_entry_head field;
989 struct trace_uprobe *tu = event_call->data;
990
991 if (is_ret_probe(tu)) {
992 DEFINE_FIELD(unsigned long, vaddr[0], FIELD_STRING_FUNC, 0);
993 DEFINE_FIELD(unsigned long, vaddr[1], FIELD_STRING_RETIP, 0);
994 size = SIZEOF_TRACE_ENTRY(true);
995 } else {
996 DEFINE_FIELD(unsigned long, vaddr[0], FIELD_STRING_IP, 0);
997 size = SIZEOF_TRACE_ENTRY(false);
998 }
999 /* Set argument names as fields */
1000 for (i = 0; i < tu->tp.nr_args; i++) {
1001 struct probe_arg *parg = &tu->tp.args[i];
1002
1003 ret = trace_define_field(event_call, parg->type->fmttype,
1004 parg->name, size + parg->offset,
1005 parg->type->size, parg->type->is_signed,
1006 FILTER_OTHER);
1007
1008 if (ret)
1009 return ret;
1010 }
1011 return 0;
1012}
1013
1014#ifdef CONFIG_PERF_EVENTS
1015static bool
1016__uprobe_perf_filter(struct trace_uprobe_filter *filter, struct mm_struct *mm)
1017{
1018 struct perf_event *event;
1019
1020 if (filter->nr_systemwide)
1021 return true;
1022
1023 list_for_each_entry(event, &filter->perf_events, hw.tp_list) {
1024 if (event->hw.target->mm == mm)
1025 return true;
1026 }
1027
1028 return false;
1029}
1030
1031static inline bool
1032uprobe_filter_event(struct trace_uprobe *tu, struct perf_event *event)
1033{
1034 return __uprobe_perf_filter(&tu->filter, event->hw.target->mm);
1035}
1036
1037static int uprobe_perf_close(struct trace_uprobe *tu, struct perf_event *event)
1038{
1039 bool done;
1040
1041 write_lock(&tu->filter.rwlock);
1042 if (event->hw.target) {
1043 list_del(&event->hw.tp_list);
1044 done = tu->filter.nr_systemwide ||
1045 (event->hw.target->flags & PF_EXITING) ||
1046 uprobe_filter_event(tu, event);
1047 } else {
1048 tu->filter.nr_systemwide--;
1049 done = tu->filter.nr_systemwide;
1050 }
1051 write_unlock(&tu->filter.rwlock);
1052
1053 if (!done)
1054 return uprobe_apply(tu->inode, tu->offset, &tu->consumer, false);
1055
1056 return 0;
1057}
1058
1059static int uprobe_perf_open(struct trace_uprobe *tu, struct perf_event *event)
1060{
1061 bool done;
1062 int err;
1063
1064 write_lock(&tu->filter.rwlock);
1065 if (event->hw.target) {
1066 /*
1067 * event->parent != NULL means copy_process(), we can avoid
1068 * uprobe_apply(). current->mm must be probed and we can rely
1069 * on dup_mmap() which preserves the already installed bp's.
1070 *
1071 * attr.enable_on_exec means that exec/mmap will install the
1072 * breakpoints we need.
1073 */
1074 done = tu->filter.nr_systemwide ||
1075 event->parent || event->attr.enable_on_exec ||
1076 uprobe_filter_event(tu, event);
1077 list_add(&event->hw.tp_list, &tu->filter.perf_events);
1078 } else {
1079 done = tu->filter.nr_systemwide;
1080 tu->filter.nr_systemwide++;
1081 }
1082 write_unlock(&tu->filter.rwlock);
1083
1084 err = 0;
1085 if (!done) {
1086 err = uprobe_apply(tu->inode, tu->offset, &tu->consumer, true);
1087 if (err)
1088 uprobe_perf_close(tu, event);
1089 }
1090 return err;
1091}
1092
1093static bool uprobe_perf_filter(struct uprobe_consumer *uc,
1094 enum uprobe_filter_ctx ctx, struct mm_struct *mm)
1095{
1096 struct trace_uprobe *tu;
1097 int ret;
1098
1099 tu = container_of(uc, struct trace_uprobe, consumer);
1100 read_lock(&tu->filter.rwlock);
1101 ret = __uprobe_perf_filter(&tu->filter, mm);
1102 read_unlock(&tu->filter.rwlock);
1103
1104 return ret;
1105}
1106
1107static void __uprobe_perf_func(struct trace_uprobe *tu,
1108 unsigned long func, struct pt_regs *regs,
1109 struct uprobe_cpu_buffer *ucb, int dsize)
1110{
1111 struct trace_event_call *call = &tu->tp.call;
1112 struct uprobe_trace_entry_head *entry;
1113 struct bpf_prog *prog = call->prog;
1114 struct hlist_head *head;
1115 void *data;
1116 int size, esize;
1117 int rctx;
1118
1119 if (prog && !trace_call_bpf(prog, regs))
1120 return;
1121
1122 esize = SIZEOF_TRACE_ENTRY(is_ret_probe(tu));
1123
1124 size = esize + tu->tp.size + dsize;
1125 size = ALIGN(size + sizeof(u32), sizeof(u64)) - sizeof(u32);
1126 if (WARN_ONCE(size > PERF_MAX_TRACE_SIZE, "profile buffer not large enough"))
1127 return;
1128
1129 preempt_disable();
1130 head = this_cpu_ptr(call->perf_events);
1131 if (hlist_empty(head))
1132 goto out;
1133
1134 entry = perf_trace_buf_prepare(size, call->event.type, NULL, &rctx);
1135 if (!entry)
1136 goto out;
1137
1138 if (is_ret_probe(tu)) {
1139 entry->vaddr[0] = func;
1140 entry->vaddr[1] = instruction_pointer(regs);
1141 data = DATAOF_TRACE_ENTRY(entry, true);
1142 } else {
1143 entry->vaddr[0] = instruction_pointer(regs);
1144 data = DATAOF_TRACE_ENTRY(entry, false);
1145 }
1146
1147 memcpy(data, ucb->buf, tu->tp.size + dsize);
1148
1149 if (size - esize > tu->tp.size + dsize) {
1150 int len = tu->tp.size + dsize;
1151
1152 memset(data + len, 0, size - esize - len);
1153 }
1154
1155 perf_trace_buf_submit(entry, size, rctx, 0, 1, regs, head, NULL);
1156 out:
1157 preempt_enable();
1158}
1159
1160/* uprobe profile handler */
1161static int uprobe_perf_func(struct trace_uprobe *tu, struct pt_regs *regs,
1162 struct uprobe_cpu_buffer *ucb, int dsize)
1163{
1164 if (!uprobe_perf_filter(&tu->consumer, 0, current->mm))
1165 return UPROBE_HANDLER_REMOVE;
1166
1167 if (!is_ret_probe(tu))
1168 __uprobe_perf_func(tu, 0, regs, ucb, dsize);
1169 return 0;
1170}
1171
1172static void uretprobe_perf_func(struct trace_uprobe *tu, unsigned long func,
1173 struct pt_regs *regs,
1174 struct uprobe_cpu_buffer *ucb, int dsize)
1175{
1176 __uprobe_perf_func(tu, func, regs, ucb, dsize);
1177}
1178#endif /* CONFIG_PERF_EVENTS */
1179
1180static int
1181trace_uprobe_register(struct trace_event_call *event, enum trace_reg type,
1182 void *data)
1183{
1184 struct trace_uprobe *tu = event->data;
1185 struct trace_event_file *file = data;
1186
1187 switch (type) {
1188 case TRACE_REG_REGISTER:
1189 return probe_event_enable(tu, file, NULL);
1190
1191 case TRACE_REG_UNREGISTER:
1192 probe_event_disable(tu, file);
1193 return 0;
1194
1195#ifdef CONFIG_PERF_EVENTS
1196 case TRACE_REG_PERF_REGISTER:
1197 return probe_event_enable(tu, NULL, uprobe_perf_filter);
1198
1199 case TRACE_REG_PERF_UNREGISTER:
1200 probe_event_disable(tu, NULL);
1201 return 0;
1202
1203 case TRACE_REG_PERF_OPEN:
1204 return uprobe_perf_open(tu, data);
1205
1206 case TRACE_REG_PERF_CLOSE:
1207 return uprobe_perf_close(tu, data);
1208
1209#endif
1210 default:
1211 return 0;
1212 }
1213 return 0;
1214}
1215
1216static int uprobe_dispatcher(struct uprobe_consumer *con, struct pt_regs *regs)
1217{
1218 struct trace_uprobe *tu;
1219 struct uprobe_dispatch_data udd;
1220 struct uprobe_cpu_buffer *ucb;
1221 int dsize, esize;
1222 int ret = 0;
1223
1224
1225 tu = container_of(con, struct trace_uprobe, consumer);
1226 tu->nhit++;
1227
1228 udd.tu = tu;
1229 udd.bp_addr = instruction_pointer(regs);
1230
1231 current->utask->vaddr = (unsigned long) &udd;
1232
1233 if (WARN_ON_ONCE(!uprobe_cpu_buffer))
1234 return 0;
1235
1236 dsize = __get_data_size(&tu->tp, regs);
1237 esize = SIZEOF_TRACE_ENTRY(is_ret_probe(tu));
1238
1239 ucb = uprobe_buffer_get();
1240 store_trace_args(esize, &tu->tp, regs, ucb->buf, dsize);
1241
1242 if (tu->tp.flags & TP_FLAG_TRACE)
1243 ret |= uprobe_trace_func(tu, regs, ucb, dsize);
1244
1245#ifdef CONFIG_PERF_EVENTS
1246 if (tu->tp.flags & TP_FLAG_PROFILE)
1247 ret |= uprobe_perf_func(tu, regs, ucb, dsize);
1248#endif
1249 uprobe_buffer_put(ucb);
1250 return ret;
1251}
1252
1253static int uretprobe_dispatcher(struct uprobe_consumer *con,
1254 unsigned long func, struct pt_regs *regs)
1255{
1256 struct trace_uprobe *tu;
1257 struct uprobe_dispatch_data udd;
1258 struct uprobe_cpu_buffer *ucb;
1259 int dsize, esize;
1260
1261 tu = container_of(con, struct trace_uprobe, consumer);
1262
1263 udd.tu = tu;
1264 udd.bp_addr = func;
1265
1266 current->utask->vaddr = (unsigned long) &udd;
1267
1268 if (WARN_ON_ONCE(!uprobe_cpu_buffer))
1269 return 0;
1270
1271 dsize = __get_data_size(&tu->tp, regs);
1272 esize = SIZEOF_TRACE_ENTRY(is_ret_probe(tu));
1273
1274 ucb = uprobe_buffer_get();
1275 store_trace_args(esize, &tu->tp, regs, ucb->buf, dsize);
1276
1277 if (tu->tp.flags & TP_FLAG_TRACE)
1278 uretprobe_trace_func(tu, func, regs, ucb, dsize);
1279
1280#ifdef CONFIG_PERF_EVENTS
1281 if (tu->tp.flags & TP_FLAG_PROFILE)
1282 uretprobe_perf_func(tu, func, regs, ucb, dsize);
1283#endif
1284 uprobe_buffer_put(ucb);
1285 return 0;
1286}
1287
1288static struct trace_event_functions uprobe_funcs = {
1289 .trace = print_uprobe_event
1290};
1291
1292static int register_uprobe_event(struct trace_uprobe *tu)
1293{
1294 struct trace_event_call *call = &tu->tp.call;
1295 int ret;
1296
1297 /* Initialize trace_event_call */
1298 INIT_LIST_HEAD(&call->class->fields);
1299 call->event.funcs = &uprobe_funcs;
1300 call->class->define_fields = uprobe_event_define_fields;
1301
1302 if (set_print_fmt(&tu->tp, is_ret_probe(tu)) < 0)
1303 return -ENOMEM;
1304
1305 ret = register_trace_event(&call->event);
1306 if (!ret) {
1307 kfree(call->print_fmt);
1308 return -ENODEV;
1309 }
1310
1311 call->flags = TRACE_EVENT_FL_UPROBE;
1312 call->class->reg = trace_uprobe_register;
1313 call->data = tu;
1314 ret = trace_add_event_call(call);
1315
1316 if (ret) {
1317 pr_info("Failed to register uprobe event: %s\n",
1318 trace_event_name(call));
1319 kfree(call->print_fmt);
1320 unregister_trace_event(&call->event);
1321 }
1322
1323 return ret;
1324}
1325
1326static int unregister_uprobe_event(struct trace_uprobe *tu)
1327{
1328 int ret;
1329
1330 /* tu->event is unregistered in trace_remove_event_call() */
1331 ret = trace_remove_event_call(&tu->tp.call);
1332 if (ret)
1333 return ret;
1334 kfree(tu->tp.call.print_fmt);
1335 tu->tp.call.print_fmt = NULL;
1336 return 0;
1337}
1338
1339/* Make a trace interface for controling probe points */
1340static __init int init_uprobe_trace(void)
1341{
1342 struct dentry *d_tracer;
1343
1344 d_tracer = tracing_init_dentry();
1345 if (IS_ERR(d_tracer))
1346 return 0;
1347
1348 trace_create_file("uprobe_events", 0644, d_tracer,
1349 NULL, &uprobe_events_ops);
1350 /* Profile interface */
1351 trace_create_file("uprobe_profile", 0444, d_tracer,
1352 NULL, &uprobe_profile_ops);
1353 return 0;
1354}
1355
1356fs_initcall(init_uprobe_trace);
1// SPDX-License-Identifier: GPL-2.0
2/*
3 * uprobes-based tracing events
4 *
5 * Copyright (C) IBM Corporation, 2010-2012
6 * Author: Srikar Dronamraju <srikar@linux.vnet.ibm.com>
7 */
8#define pr_fmt(fmt) "trace_uprobe: " fmt
9
10#include <linux/security.h>
11#include <linux/ctype.h>
12#include <linux/module.h>
13#include <linux/uaccess.h>
14#include <linux/uprobes.h>
15#include <linux/namei.h>
16#include <linux/string.h>
17#include <linux/rculist.h>
18
19#include "trace_dynevent.h"
20#include "trace_probe.h"
21#include "trace_probe_tmpl.h"
22
23#define UPROBE_EVENT_SYSTEM "uprobes"
24
25struct uprobe_trace_entry_head {
26 struct trace_entry ent;
27 unsigned long vaddr[];
28};
29
30#define SIZEOF_TRACE_ENTRY(is_return) \
31 (sizeof(struct uprobe_trace_entry_head) + \
32 sizeof(unsigned long) * (is_return ? 2 : 1))
33
34#define DATAOF_TRACE_ENTRY(entry, is_return) \
35 ((void*)(entry) + SIZEOF_TRACE_ENTRY(is_return))
36
37struct trace_uprobe_filter {
38 rwlock_t rwlock;
39 int nr_systemwide;
40 struct list_head perf_events;
41};
42
43static int trace_uprobe_create(int argc, const char **argv);
44static int trace_uprobe_show(struct seq_file *m, struct dyn_event *ev);
45static int trace_uprobe_release(struct dyn_event *ev);
46static bool trace_uprobe_is_busy(struct dyn_event *ev);
47static bool trace_uprobe_match(const char *system, const char *event,
48 int argc, const char **argv, struct dyn_event *ev);
49
50static struct dyn_event_operations trace_uprobe_ops = {
51 .create = trace_uprobe_create,
52 .show = trace_uprobe_show,
53 .is_busy = trace_uprobe_is_busy,
54 .free = trace_uprobe_release,
55 .match = trace_uprobe_match,
56};
57
58/*
59 * uprobe event core functions
60 */
61struct trace_uprobe {
62 struct dyn_event devent;
63 struct trace_uprobe_filter filter;
64 struct uprobe_consumer consumer;
65 struct path path;
66 struct inode *inode;
67 char *filename;
68 unsigned long offset;
69 unsigned long ref_ctr_offset;
70 unsigned long nhit;
71 struct trace_probe tp;
72};
73
74static bool is_trace_uprobe(struct dyn_event *ev)
75{
76 return ev->ops == &trace_uprobe_ops;
77}
78
79static struct trace_uprobe *to_trace_uprobe(struct dyn_event *ev)
80{
81 return container_of(ev, struct trace_uprobe, devent);
82}
83
84/**
85 * for_each_trace_uprobe - iterate over the trace_uprobe list
86 * @pos: the struct trace_uprobe * for each entry
87 * @dpos: the struct dyn_event * to use as a loop cursor
88 */
89#define for_each_trace_uprobe(pos, dpos) \
90 for_each_dyn_event(dpos) \
91 if (is_trace_uprobe(dpos) && (pos = to_trace_uprobe(dpos)))
92
93#define SIZEOF_TRACE_UPROBE(n) \
94 (offsetof(struct trace_uprobe, tp.args) + \
95 (sizeof(struct probe_arg) * (n)))
96
97static int register_uprobe_event(struct trace_uprobe *tu);
98static int unregister_uprobe_event(struct trace_uprobe *tu);
99
100struct uprobe_dispatch_data {
101 struct trace_uprobe *tu;
102 unsigned long bp_addr;
103};
104
105static int uprobe_dispatcher(struct uprobe_consumer *con, struct pt_regs *regs);
106static int uretprobe_dispatcher(struct uprobe_consumer *con,
107 unsigned long func, struct pt_regs *regs);
108
109#ifdef CONFIG_STACK_GROWSUP
110static unsigned long adjust_stack_addr(unsigned long addr, unsigned int n)
111{
112 return addr - (n * sizeof(long));
113}
114#else
115static unsigned long adjust_stack_addr(unsigned long addr, unsigned int n)
116{
117 return addr + (n * sizeof(long));
118}
119#endif
120
121static unsigned long get_user_stack_nth(struct pt_regs *regs, unsigned int n)
122{
123 unsigned long ret;
124 unsigned long addr = user_stack_pointer(regs);
125
126 addr = adjust_stack_addr(addr, n);
127
128 if (copy_from_user(&ret, (void __force __user *) addr, sizeof(ret)))
129 return 0;
130
131 return ret;
132}
133
134/*
135 * Uprobes-specific fetch functions
136 */
137static nokprobe_inline int
138probe_mem_read(void *dest, void *src, size_t size)
139{
140 void __user *vaddr = (void __force __user *)src;
141
142 return copy_from_user(dest, vaddr, size) ? -EFAULT : 0;
143}
144
145static nokprobe_inline int
146probe_mem_read_user(void *dest, void *src, size_t size)
147{
148 return probe_mem_read(dest, src, size);
149}
150
151/*
152 * Fetch a null-terminated string. Caller MUST set *(u32 *)dest with max
153 * length and relative data location.
154 */
155static nokprobe_inline int
156fetch_store_string(unsigned long addr, void *dest, void *base)
157{
158 long ret;
159 u32 loc = *(u32 *)dest;
160 int maxlen = get_loc_len(loc);
161 u8 *dst = get_loc_data(dest, base);
162 void __user *src = (void __force __user *) addr;
163
164 if (unlikely(!maxlen))
165 return -ENOMEM;
166
167 if (addr == FETCH_TOKEN_COMM)
168 ret = strlcpy(dst, current->comm, maxlen);
169 else
170 ret = strncpy_from_user(dst, src, maxlen);
171 if (ret >= 0) {
172 if (ret == maxlen)
173 dst[ret - 1] = '\0';
174 else
175 /*
176 * Include the terminating null byte. In this case it
177 * was copied by strncpy_from_user but not accounted
178 * for in ret.
179 */
180 ret++;
181 *(u32 *)dest = make_data_loc(ret, (void *)dst - base);
182 }
183
184 return ret;
185}
186
187static nokprobe_inline int
188fetch_store_string_user(unsigned long addr, void *dest, void *base)
189{
190 return fetch_store_string(addr, dest, base);
191}
192
193/* Return the length of string -- including null terminal byte */
194static nokprobe_inline int
195fetch_store_strlen(unsigned long addr)
196{
197 int len;
198 void __user *vaddr = (void __force __user *) addr;
199
200 if (addr == FETCH_TOKEN_COMM)
201 len = strlen(current->comm) + 1;
202 else
203 len = strnlen_user(vaddr, MAX_STRING_SIZE);
204
205 return (len > MAX_STRING_SIZE) ? 0 : len;
206}
207
208static nokprobe_inline int
209fetch_store_strlen_user(unsigned long addr)
210{
211 return fetch_store_strlen(addr);
212}
213
214static unsigned long translate_user_vaddr(unsigned long file_offset)
215{
216 unsigned long base_addr;
217 struct uprobe_dispatch_data *udd;
218
219 udd = (void *) current->utask->vaddr;
220
221 base_addr = udd->bp_addr - udd->tu->offset;
222 return base_addr + file_offset;
223}
224
225/* Note that we don't verify it, since the code does not come from user space */
226static int
227process_fetch_insn(struct fetch_insn *code, struct pt_regs *regs, void *dest,
228 void *base)
229{
230 unsigned long val;
231
232 /* 1st stage: get value from context */
233 switch (code->op) {
234 case FETCH_OP_REG:
235 val = regs_get_register(regs, code->param);
236 break;
237 case FETCH_OP_STACK:
238 val = get_user_stack_nth(regs, code->param);
239 break;
240 case FETCH_OP_STACKP:
241 val = user_stack_pointer(regs);
242 break;
243 case FETCH_OP_RETVAL:
244 val = regs_return_value(regs);
245 break;
246 case FETCH_OP_IMM:
247 val = code->immediate;
248 break;
249 case FETCH_OP_COMM:
250 val = FETCH_TOKEN_COMM;
251 break;
252 case FETCH_OP_DATA:
253 val = (unsigned long)code->data;
254 break;
255 case FETCH_OP_FOFFS:
256 val = translate_user_vaddr(code->immediate);
257 break;
258 default:
259 return -EILSEQ;
260 }
261 code++;
262
263 return process_fetch_insn_bottom(code, val, dest, base);
264}
265NOKPROBE_SYMBOL(process_fetch_insn)
266
267static inline void init_trace_uprobe_filter(struct trace_uprobe_filter *filter)
268{
269 rwlock_init(&filter->rwlock);
270 filter->nr_systemwide = 0;
271 INIT_LIST_HEAD(&filter->perf_events);
272}
273
274static inline bool uprobe_filter_is_empty(struct trace_uprobe_filter *filter)
275{
276 return !filter->nr_systemwide && list_empty(&filter->perf_events);
277}
278
279static inline bool is_ret_probe(struct trace_uprobe *tu)
280{
281 return tu->consumer.ret_handler != NULL;
282}
283
284static bool trace_uprobe_is_busy(struct dyn_event *ev)
285{
286 struct trace_uprobe *tu = to_trace_uprobe(ev);
287
288 return trace_probe_is_enabled(&tu->tp);
289}
290
291static bool trace_uprobe_match_command_head(struct trace_uprobe *tu,
292 int argc, const char **argv)
293{
294 char buf[MAX_ARGSTR_LEN + 1];
295 int len;
296
297 if (!argc)
298 return true;
299
300 len = strlen(tu->filename);
301 if (strncmp(tu->filename, argv[0], len) || argv[0][len] != ':')
302 return false;
303
304 if (tu->ref_ctr_offset == 0)
305 snprintf(buf, sizeof(buf), "0x%0*lx",
306 (int)(sizeof(void *) * 2), tu->offset);
307 else
308 snprintf(buf, sizeof(buf), "0x%0*lx(0x%lx)",
309 (int)(sizeof(void *) * 2), tu->offset,
310 tu->ref_ctr_offset);
311 if (strcmp(buf, &argv[0][len + 1]))
312 return false;
313
314 argc--; argv++;
315
316 return trace_probe_match_command_args(&tu->tp, argc, argv);
317}
318
319static bool trace_uprobe_match(const char *system, const char *event,
320 int argc, const char **argv, struct dyn_event *ev)
321{
322 struct trace_uprobe *tu = to_trace_uprobe(ev);
323
324 return strcmp(trace_probe_name(&tu->tp), event) == 0 &&
325 (!system || strcmp(trace_probe_group_name(&tu->tp), system) == 0) &&
326 trace_uprobe_match_command_head(tu, argc, argv);
327}
328
329static nokprobe_inline struct trace_uprobe *
330trace_uprobe_primary_from_call(struct trace_event_call *call)
331{
332 struct trace_probe *tp;
333
334 tp = trace_probe_primary_from_call(call);
335 if (WARN_ON_ONCE(!tp))
336 return NULL;
337
338 return container_of(tp, struct trace_uprobe, tp);
339}
340
341/*
342 * Allocate new trace_uprobe and initialize it (including uprobes).
343 */
344static struct trace_uprobe *
345alloc_trace_uprobe(const char *group, const char *event, int nargs, bool is_ret)
346{
347 struct trace_uprobe *tu;
348 int ret;
349
350 tu = kzalloc(SIZEOF_TRACE_UPROBE(nargs), GFP_KERNEL);
351 if (!tu)
352 return ERR_PTR(-ENOMEM);
353
354 ret = trace_probe_init(&tu->tp, event, group);
355 if (ret < 0)
356 goto error;
357
358 dyn_event_init(&tu->devent, &trace_uprobe_ops);
359 tu->consumer.handler = uprobe_dispatcher;
360 if (is_ret)
361 tu->consumer.ret_handler = uretprobe_dispatcher;
362 init_trace_uprobe_filter(&tu->filter);
363 return tu;
364
365error:
366 kfree(tu);
367
368 return ERR_PTR(ret);
369}
370
371static void free_trace_uprobe(struct trace_uprobe *tu)
372{
373 if (!tu)
374 return;
375
376 path_put(&tu->path);
377 trace_probe_cleanup(&tu->tp);
378 kfree(tu->filename);
379 kfree(tu);
380}
381
382static struct trace_uprobe *find_probe_event(const char *event, const char *group)
383{
384 struct dyn_event *pos;
385 struct trace_uprobe *tu;
386
387 for_each_trace_uprobe(tu, pos)
388 if (strcmp(trace_probe_name(&tu->tp), event) == 0 &&
389 strcmp(trace_probe_group_name(&tu->tp), group) == 0)
390 return tu;
391
392 return NULL;
393}
394
395/* Unregister a trace_uprobe and probe_event */
396static int unregister_trace_uprobe(struct trace_uprobe *tu)
397{
398 int ret;
399
400 if (trace_probe_has_sibling(&tu->tp))
401 goto unreg;
402
403 ret = unregister_uprobe_event(tu);
404 if (ret)
405 return ret;
406
407unreg:
408 dyn_event_remove(&tu->devent);
409 trace_probe_unlink(&tu->tp);
410 free_trace_uprobe(tu);
411 return 0;
412}
413
414static bool trace_uprobe_has_same_uprobe(struct trace_uprobe *orig,
415 struct trace_uprobe *comp)
416{
417 struct trace_probe_event *tpe = orig->tp.event;
418 struct trace_probe *pos;
419 struct inode *comp_inode = d_real_inode(comp->path.dentry);
420 int i;
421
422 list_for_each_entry(pos, &tpe->probes, list) {
423 orig = container_of(pos, struct trace_uprobe, tp);
424 if (comp_inode != d_real_inode(orig->path.dentry) ||
425 comp->offset != orig->offset)
426 continue;
427
428 /*
429 * trace_probe_compare_arg_type() ensured that nr_args and
430 * each argument name and type are same. Let's compare comm.
431 */
432 for (i = 0; i < orig->tp.nr_args; i++) {
433 if (strcmp(orig->tp.args[i].comm,
434 comp->tp.args[i].comm))
435 break;
436 }
437
438 if (i == orig->tp.nr_args)
439 return true;
440 }
441
442 return false;
443}
444
445static int append_trace_uprobe(struct trace_uprobe *tu, struct trace_uprobe *to)
446{
447 int ret;
448
449 ret = trace_probe_compare_arg_type(&tu->tp, &to->tp);
450 if (ret) {
451 /* Note that argument starts index = 2 */
452 trace_probe_log_set_index(ret + 1);
453 trace_probe_log_err(0, DIFF_ARG_TYPE);
454 return -EEXIST;
455 }
456 if (trace_uprobe_has_same_uprobe(to, tu)) {
457 trace_probe_log_set_index(0);
458 trace_probe_log_err(0, SAME_PROBE);
459 return -EEXIST;
460 }
461
462 /* Append to existing event */
463 ret = trace_probe_append(&tu->tp, &to->tp);
464 if (!ret)
465 dyn_event_add(&tu->devent);
466
467 return ret;
468}
469
470/*
471 * Uprobe with multiple reference counter is not allowed. i.e.
472 * If inode and offset matches, reference counter offset *must*
473 * match as well. Though, there is one exception: If user is
474 * replacing old trace_uprobe with new one(same group/event),
475 * then we allow same uprobe with new reference counter as far
476 * as the new one does not conflict with any other existing
477 * ones.
478 */
479static int validate_ref_ctr_offset(struct trace_uprobe *new)
480{
481 struct dyn_event *pos;
482 struct trace_uprobe *tmp;
483 struct inode *new_inode = d_real_inode(new->path.dentry);
484
485 for_each_trace_uprobe(tmp, pos) {
486 if (new_inode == d_real_inode(tmp->path.dentry) &&
487 new->offset == tmp->offset &&
488 new->ref_ctr_offset != tmp->ref_ctr_offset) {
489 pr_warn("Reference counter offset mismatch.");
490 return -EINVAL;
491 }
492 }
493 return 0;
494}
495
496/* Register a trace_uprobe and probe_event */
497static int register_trace_uprobe(struct trace_uprobe *tu)
498{
499 struct trace_uprobe *old_tu;
500 int ret;
501
502 mutex_lock(&event_mutex);
503
504 ret = validate_ref_ctr_offset(tu);
505 if (ret)
506 goto end;
507
508 /* register as an event */
509 old_tu = find_probe_event(trace_probe_name(&tu->tp),
510 trace_probe_group_name(&tu->tp));
511 if (old_tu) {
512 if (is_ret_probe(tu) != is_ret_probe(old_tu)) {
513 trace_probe_log_set_index(0);
514 trace_probe_log_err(0, DIFF_PROBE_TYPE);
515 ret = -EEXIST;
516 } else {
517 ret = append_trace_uprobe(tu, old_tu);
518 }
519 goto end;
520 }
521
522 ret = register_uprobe_event(tu);
523 if (ret) {
524 pr_warn("Failed to register probe event(%d)\n", ret);
525 goto end;
526 }
527
528 dyn_event_add(&tu->devent);
529
530end:
531 mutex_unlock(&event_mutex);
532
533 return ret;
534}
535
536/*
537 * Argument syntax:
538 * - Add uprobe: p|r[:[GRP/]EVENT] PATH:OFFSET [FETCHARGS]
539 */
540static int trace_uprobe_create(int argc, const char **argv)
541{
542 struct trace_uprobe *tu;
543 const char *event = NULL, *group = UPROBE_EVENT_SYSTEM;
544 char *arg, *filename, *rctr, *rctr_end, *tmp;
545 char buf[MAX_EVENT_NAME_LEN];
546 struct path path;
547 unsigned long offset, ref_ctr_offset;
548 bool is_return = false;
549 int i, ret;
550
551 ret = 0;
552 ref_ctr_offset = 0;
553
554 switch (argv[0][0]) {
555 case 'r':
556 is_return = true;
557 break;
558 case 'p':
559 break;
560 default:
561 return -ECANCELED;
562 }
563
564 if (argc < 2)
565 return -ECANCELED;
566
567 if (argv[0][1] == ':')
568 event = &argv[0][2];
569
570 if (!strchr(argv[1], '/'))
571 return -ECANCELED;
572
573 filename = kstrdup(argv[1], GFP_KERNEL);
574 if (!filename)
575 return -ENOMEM;
576
577 /* Find the last occurrence, in case the path contains ':' too. */
578 arg = strrchr(filename, ':');
579 if (!arg || !isdigit(arg[1])) {
580 kfree(filename);
581 return -ECANCELED;
582 }
583
584 trace_probe_log_init("trace_uprobe", argc, argv);
585 trace_probe_log_set_index(1); /* filename is the 2nd argument */
586
587 *arg++ = '\0';
588 ret = kern_path(filename, LOOKUP_FOLLOW, &path);
589 if (ret) {
590 trace_probe_log_err(0, FILE_NOT_FOUND);
591 kfree(filename);
592 trace_probe_log_clear();
593 return ret;
594 }
595 if (!d_is_reg(path.dentry)) {
596 trace_probe_log_err(0, NO_REGULAR_FILE);
597 ret = -EINVAL;
598 goto fail_address_parse;
599 }
600
601 /* Parse reference counter offset if specified. */
602 rctr = strchr(arg, '(');
603 if (rctr) {
604 rctr_end = strchr(rctr, ')');
605 if (!rctr_end) {
606 ret = -EINVAL;
607 rctr_end = rctr + strlen(rctr);
608 trace_probe_log_err(rctr_end - filename,
609 REFCNT_OPEN_BRACE);
610 goto fail_address_parse;
611 } else if (rctr_end[1] != '\0') {
612 ret = -EINVAL;
613 trace_probe_log_err(rctr_end + 1 - filename,
614 BAD_REFCNT_SUFFIX);
615 goto fail_address_parse;
616 }
617
618 *rctr++ = '\0';
619 *rctr_end = '\0';
620 ret = kstrtoul(rctr, 0, &ref_ctr_offset);
621 if (ret) {
622 trace_probe_log_err(rctr - filename, BAD_REFCNT);
623 goto fail_address_parse;
624 }
625 }
626
627 /* Parse uprobe offset. */
628 ret = kstrtoul(arg, 0, &offset);
629 if (ret) {
630 trace_probe_log_err(arg - filename, BAD_UPROBE_OFFS);
631 goto fail_address_parse;
632 }
633
634 /* setup a probe */
635 trace_probe_log_set_index(0);
636 if (event) {
637 ret = traceprobe_parse_event_name(&event, &group, buf,
638 event - argv[0]);
639 if (ret)
640 goto fail_address_parse;
641 } else {
642 char *tail;
643 char *ptr;
644
645 tail = kstrdup(kbasename(filename), GFP_KERNEL);
646 if (!tail) {
647 ret = -ENOMEM;
648 goto fail_address_parse;
649 }
650
651 ptr = strpbrk(tail, ".-_");
652 if (ptr)
653 *ptr = '\0';
654
655 snprintf(buf, MAX_EVENT_NAME_LEN, "%c_%s_0x%lx", 'p', tail, offset);
656 event = buf;
657 kfree(tail);
658 }
659
660 argc -= 2;
661 argv += 2;
662
663 tu = alloc_trace_uprobe(group, event, argc, is_return);
664 if (IS_ERR(tu)) {
665 ret = PTR_ERR(tu);
666 /* This must return -ENOMEM otherwise there is a bug */
667 WARN_ON_ONCE(ret != -ENOMEM);
668 goto fail_address_parse;
669 }
670 tu->offset = offset;
671 tu->ref_ctr_offset = ref_ctr_offset;
672 tu->path = path;
673 tu->filename = filename;
674
675 /* parse arguments */
676 for (i = 0; i < argc && i < MAX_TRACE_ARGS; i++) {
677 tmp = kstrdup(argv[i], GFP_KERNEL);
678 if (!tmp) {
679 ret = -ENOMEM;
680 goto error;
681 }
682
683 trace_probe_log_set_index(i + 2);
684 ret = traceprobe_parse_probe_arg(&tu->tp, i, tmp,
685 is_return ? TPARG_FL_RETURN : 0);
686 kfree(tmp);
687 if (ret)
688 goto error;
689 }
690
691 ret = traceprobe_set_print_fmt(&tu->tp, is_ret_probe(tu));
692 if (ret < 0)
693 goto error;
694
695 ret = register_trace_uprobe(tu);
696 if (!ret)
697 goto out;
698
699error:
700 free_trace_uprobe(tu);
701out:
702 trace_probe_log_clear();
703 return ret;
704
705fail_address_parse:
706 trace_probe_log_clear();
707 path_put(&path);
708 kfree(filename);
709
710 return ret;
711}
712
713static int create_or_delete_trace_uprobe(int argc, char **argv)
714{
715 int ret;
716
717 if (argv[0][0] == '-')
718 return dyn_event_release(argc, argv, &trace_uprobe_ops);
719
720 ret = trace_uprobe_create(argc, (const char **)argv);
721 return ret == -ECANCELED ? -EINVAL : ret;
722}
723
724static int trace_uprobe_release(struct dyn_event *ev)
725{
726 struct trace_uprobe *tu = to_trace_uprobe(ev);
727
728 return unregister_trace_uprobe(tu);
729}
730
731/* Probes listing interfaces */
732static int trace_uprobe_show(struct seq_file *m, struct dyn_event *ev)
733{
734 struct trace_uprobe *tu = to_trace_uprobe(ev);
735 char c = is_ret_probe(tu) ? 'r' : 'p';
736 int i;
737
738 seq_printf(m, "%c:%s/%s %s:0x%0*lx", c, trace_probe_group_name(&tu->tp),
739 trace_probe_name(&tu->tp), tu->filename,
740 (int)(sizeof(void *) * 2), tu->offset);
741
742 if (tu->ref_ctr_offset)
743 seq_printf(m, "(0x%lx)", tu->ref_ctr_offset);
744
745 for (i = 0; i < tu->tp.nr_args; i++)
746 seq_printf(m, " %s=%s", tu->tp.args[i].name, tu->tp.args[i].comm);
747
748 seq_putc(m, '\n');
749 return 0;
750}
751
752static int probes_seq_show(struct seq_file *m, void *v)
753{
754 struct dyn_event *ev = v;
755
756 if (!is_trace_uprobe(ev))
757 return 0;
758
759 return trace_uprobe_show(m, ev);
760}
761
762static const struct seq_operations probes_seq_op = {
763 .start = dyn_event_seq_start,
764 .next = dyn_event_seq_next,
765 .stop = dyn_event_seq_stop,
766 .show = probes_seq_show
767};
768
769static int probes_open(struct inode *inode, struct file *file)
770{
771 int ret;
772
773 ret = security_locked_down(LOCKDOWN_TRACEFS);
774 if (ret)
775 return ret;
776
777 if ((file->f_mode & FMODE_WRITE) && (file->f_flags & O_TRUNC)) {
778 ret = dyn_events_release_all(&trace_uprobe_ops);
779 if (ret)
780 return ret;
781 }
782
783 return seq_open(file, &probes_seq_op);
784}
785
786static ssize_t probes_write(struct file *file, const char __user *buffer,
787 size_t count, loff_t *ppos)
788{
789 return trace_parse_run_command(file, buffer, count, ppos,
790 create_or_delete_trace_uprobe);
791}
792
793static const struct file_operations uprobe_events_ops = {
794 .owner = THIS_MODULE,
795 .open = probes_open,
796 .read = seq_read,
797 .llseek = seq_lseek,
798 .release = seq_release,
799 .write = probes_write,
800};
801
802/* Probes profiling interfaces */
803static int probes_profile_seq_show(struct seq_file *m, void *v)
804{
805 struct dyn_event *ev = v;
806 struct trace_uprobe *tu;
807
808 if (!is_trace_uprobe(ev))
809 return 0;
810
811 tu = to_trace_uprobe(ev);
812 seq_printf(m, " %s %-44s %15lu\n", tu->filename,
813 trace_probe_name(&tu->tp), tu->nhit);
814 return 0;
815}
816
817static const struct seq_operations profile_seq_op = {
818 .start = dyn_event_seq_start,
819 .next = dyn_event_seq_next,
820 .stop = dyn_event_seq_stop,
821 .show = probes_profile_seq_show
822};
823
824static int profile_open(struct inode *inode, struct file *file)
825{
826 int ret;
827
828 ret = security_locked_down(LOCKDOWN_TRACEFS);
829 if (ret)
830 return ret;
831
832 return seq_open(file, &profile_seq_op);
833}
834
835static const struct file_operations uprobe_profile_ops = {
836 .owner = THIS_MODULE,
837 .open = profile_open,
838 .read = seq_read,
839 .llseek = seq_lseek,
840 .release = seq_release,
841};
842
843struct uprobe_cpu_buffer {
844 struct mutex mutex;
845 void *buf;
846};
847static struct uprobe_cpu_buffer __percpu *uprobe_cpu_buffer;
848static int uprobe_buffer_refcnt;
849
850static int uprobe_buffer_init(void)
851{
852 int cpu, err_cpu;
853
854 uprobe_cpu_buffer = alloc_percpu(struct uprobe_cpu_buffer);
855 if (uprobe_cpu_buffer == NULL)
856 return -ENOMEM;
857
858 for_each_possible_cpu(cpu) {
859 struct page *p = alloc_pages_node(cpu_to_node(cpu),
860 GFP_KERNEL, 0);
861 if (p == NULL) {
862 err_cpu = cpu;
863 goto err;
864 }
865 per_cpu_ptr(uprobe_cpu_buffer, cpu)->buf = page_address(p);
866 mutex_init(&per_cpu_ptr(uprobe_cpu_buffer, cpu)->mutex);
867 }
868
869 return 0;
870
871err:
872 for_each_possible_cpu(cpu) {
873 if (cpu == err_cpu)
874 break;
875 free_page((unsigned long)per_cpu_ptr(uprobe_cpu_buffer, cpu)->buf);
876 }
877
878 free_percpu(uprobe_cpu_buffer);
879 return -ENOMEM;
880}
881
882static int uprobe_buffer_enable(void)
883{
884 int ret = 0;
885
886 BUG_ON(!mutex_is_locked(&event_mutex));
887
888 if (uprobe_buffer_refcnt++ == 0) {
889 ret = uprobe_buffer_init();
890 if (ret < 0)
891 uprobe_buffer_refcnt--;
892 }
893
894 return ret;
895}
896
897static void uprobe_buffer_disable(void)
898{
899 int cpu;
900
901 BUG_ON(!mutex_is_locked(&event_mutex));
902
903 if (--uprobe_buffer_refcnt == 0) {
904 for_each_possible_cpu(cpu)
905 free_page((unsigned long)per_cpu_ptr(uprobe_cpu_buffer,
906 cpu)->buf);
907
908 free_percpu(uprobe_cpu_buffer);
909 uprobe_cpu_buffer = NULL;
910 }
911}
912
913static struct uprobe_cpu_buffer *uprobe_buffer_get(void)
914{
915 struct uprobe_cpu_buffer *ucb;
916 int cpu;
917
918 cpu = raw_smp_processor_id();
919 ucb = per_cpu_ptr(uprobe_cpu_buffer, cpu);
920
921 /*
922 * Use per-cpu buffers for fastest access, but we might migrate
923 * so the mutex makes sure we have sole access to it.
924 */
925 mutex_lock(&ucb->mutex);
926
927 return ucb;
928}
929
930static void uprobe_buffer_put(struct uprobe_cpu_buffer *ucb)
931{
932 mutex_unlock(&ucb->mutex);
933}
934
935static void __uprobe_trace_func(struct trace_uprobe *tu,
936 unsigned long func, struct pt_regs *regs,
937 struct uprobe_cpu_buffer *ucb, int dsize,
938 struct trace_event_file *trace_file)
939{
940 struct uprobe_trace_entry_head *entry;
941 struct ring_buffer_event *event;
942 struct ring_buffer *buffer;
943 void *data;
944 int size, esize;
945 struct trace_event_call *call = trace_probe_event_call(&tu->tp);
946
947 WARN_ON(call != trace_file->event_call);
948
949 if (WARN_ON_ONCE(tu->tp.size + dsize > PAGE_SIZE))
950 return;
951
952 if (trace_trigger_soft_disabled(trace_file))
953 return;
954
955 esize = SIZEOF_TRACE_ENTRY(is_ret_probe(tu));
956 size = esize + tu->tp.size + dsize;
957 event = trace_event_buffer_lock_reserve(&buffer, trace_file,
958 call->event.type, size, 0, 0);
959 if (!event)
960 return;
961
962 entry = ring_buffer_event_data(event);
963 if (is_ret_probe(tu)) {
964 entry->vaddr[0] = func;
965 entry->vaddr[1] = instruction_pointer(regs);
966 data = DATAOF_TRACE_ENTRY(entry, true);
967 } else {
968 entry->vaddr[0] = instruction_pointer(regs);
969 data = DATAOF_TRACE_ENTRY(entry, false);
970 }
971
972 memcpy(data, ucb->buf, tu->tp.size + dsize);
973
974 event_trigger_unlock_commit(trace_file, buffer, event, entry, 0, 0);
975}
976
977/* uprobe handler */
978static int uprobe_trace_func(struct trace_uprobe *tu, struct pt_regs *regs,
979 struct uprobe_cpu_buffer *ucb, int dsize)
980{
981 struct event_file_link *link;
982
983 if (is_ret_probe(tu))
984 return 0;
985
986 rcu_read_lock();
987 trace_probe_for_each_link_rcu(link, &tu->tp)
988 __uprobe_trace_func(tu, 0, regs, ucb, dsize, link->file);
989 rcu_read_unlock();
990
991 return 0;
992}
993
994static void uretprobe_trace_func(struct trace_uprobe *tu, unsigned long func,
995 struct pt_regs *regs,
996 struct uprobe_cpu_buffer *ucb, int dsize)
997{
998 struct event_file_link *link;
999
1000 rcu_read_lock();
1001 trace_probe_for_each_link_rcu(link, &tu->tp)
1002 __uprobe_trace_func(tu, func, regs, ucb, dsize, link->file);
1003 rcu_read_unlock();
1004}
1005
1006/* Event entry printers */
1007static enum print_line_t
1008print_uprobe_event(struct trace_iterator *iter, int flags, struct trace_event *event)
1009{
1010 struct uprobe_trace_entry_head *entry;
1011 struct trace_seq *s = &iter->seq;
1012 struct trace_uprobe *tu;
1013 u8 *data;
1014
1015 entry = (struct uprobe_trace_entry_head *)iter->ent;
1016 tu = trace_uprobe_primary_from_call(
1017 container_of(event, struct trace_event_call, event));
1018 if (unlikely(!tu))
1019 goto out;
1020
1021 if (is_ret_probe(tu)) {
1022 trace_seq_printf(s, "%s: (0x%lx <- 0x%lx)",
1023 trace_probe_name(&tu->tp),
1024 entry->vaddr[1], entry->vaddr[0]);
1025 data = DATAOF_TRACE_ENTRY(entry, true);
1026 } else {
1027 trace_seq_printf(s, "%s: (0x%lx)",
1028 trace_probe_name(&tu->tp),
1029 entry->vaddr[0]);
1030 data = DATAOF_TRACE_ENTRY(entry, false);
1031 }
1032
1033 if (print_probe_args(s, tu->tp.args, tu->tp.nr_args, data, entry) < 0)
1034 goto out;
1035
1036 trace_seq_putc(s, '\n');
1037
1038 out:
1039 return trace_handle_return(s);
1040}
1041
1042typedef bool (*filter_func_t)(struct uprobe_consumer *self,
1043 enum uprobe_filter_ctx ctx,
1044 struct mm_struct *mm);
1045
1046static int trace_uprobe_enable(struct trace_uprobe *tu, filter_func_t filter)
1047{
1048 int ret;
1049
1050 tu->consumer.filter = filter;
1051 tu->inode = d_real_inode(tu->path.dentry);
1052
1053 if (tu->ref_ctr_offset)
1054 ret = uprobe_register_refctr(tu->inode, tu->offset,
1055 tu->ref_ctr_offset, &tu->consumer);
1056 else
1057 ret = uprobe_register(tu->inode, tu->offset, &tu->consumer);
1058
1059 if (ret)
1060 tu->inode = NULL;
1061
1062 return ret;
1063}
1064
1065static void __probe_event_disable(struct trace_probe *tp)
1066{
1067 struct trace_probe *pos;
1068 struct trace_uprobe *tu;
1069
1070 list_for_each_entry(pos, trace_probe_probe_list(tp), list) {
1071 tu = container_of(pos, struct trace_uprobe, tp);
1072 if (!tu->inode)
1073 continue;
1074
1075 WARN_ON(!uprobe_filter_is_empty(&tu->filter));
1076
1077 uprobe_unregister(tu->inode, tu->offset, &tu->consumer);
1078 tu->inode = NULL;
1079 }
1080}
1081
1082static int probe_event_enable(struct trace_event_call *call,
1083 struct trace_event_file *file, filter_func_t filter)
1084{
1085 struct trace_probe *pos, *tp;
1086 struct trace_uprobe *tu;
1087 bool enabled;
1088 int ret;
1089
1090 tp = trace_probe_primary_from_call(call);
1091 if (WARN_ON_ONCE(!tp))
1092 return -ENODEV;
1093 enabled = trace_probe_is_enabled(tp);
1094
1095 /* This may also change "enabled" state */
1096 if (file) {
1097 if (trace_probe_test_flag(tp, TP_FLAG_PROFILE))
1098 return -EINTR;
1099
1100 ret = trace_probe_add_file(tp, file);
1101 if (ret < 0)
1102 return ret;
1103 } else {
1104 if (trace_probe_test_flag(tp, TP_FLAG_TRACE))
1105 return -EINTR;
1106
1107 trace_probe_set_flag(tp, TP_FLAG_PROFILE);
1108 }
1109
1110 tu = container_of(tp, struct trace_uprobe, tp);
1111 WARN_ON(!uprobe_filter_is_empty(&tu->filter));
1112
1113 if (enabled)
1114 return 0;
1115
1116 ret = uprobe_buffer_enable();
1117 if (ret)
1118 goto err_flags;
1119
1120 list_for_each_entry(pos, trace_probe_probe_list(tp), list) {
1121 tu = container_of(pos, struct trace_uprobe, tp);
1122 ret = trace_uprobe_enable(tu, filter);
1123 if (ret) {
1124 __probe_event_disable(tp);
1125 goto err_buffer;
1126 }
1127 }
1128
1129 return 0;
1130
1131 err_buffer:
1132 uprobe_buffer_disable();
1133
1134 err_flags:
1135 if (file)
1136 trace_probe_remove_file(tp, file);
1137 else
1138 trace_probe_clear_flag(tp, TP_FLAG_PROFILE);
1139
1140 return ret;
1141}
1142
1143static void probe_event_disable(struct trace_event_call *call,
1144 struct trace_event_file *file)
1145{
1146 struct trace_probe *tp;
1147
1148 tp = trace_probe_primary_from_call(call);
1149 if (WARN_ON_ONCE(!tp))
1150 return;
1151
1152 if (!trace_probe_is_enabled(tp))
1153 return;
1154
1155 if (file) {
1156 if (trace_probe_remove_file(tp, file) < 0)
1157 return;
1158
1159 if (trace_probe_is_enabled(tp))
1160 return;
1161 } else
1162 trace_probe_clear_flag(tp, TP_FLAG_PROFILE);
1163
1164 __probe_event_disable(tp);
1165 uprobe_buffer_disable();
1166}
1167
1168static int uprobe_event_define_fields(struct trace_event_call *event_call)
1169{
1170 int ret, size;
1171 struct uprobe_trace_entry_head field;
1172 struct trace_uprobe *tu;
1173
1174 tu = trace_uprobe_primary_from_call(event_call);
1175 if (unlikely(!tu))
1176 return -ENODEV;
1177
1178 if (is_ret_probe(tu)) {
1179 DEFINE_FIELD(unsigned long, vaddr[0], FIELD_STRING_FUNC, 0);
1180 DEFINE_FIELD(unsigned long, vaddr[1], FIELD_STRING_RETIP, 0);
1181 size = SIZEOF_TRACE_ENTRY(true);
1182 } else {
1183 DEFINE_FIELD(unsigned long, vaddr[0], FIELD_STRING_IP, 0);
1184 size = SIZEOF_TRACE_ENTRY(false);
1185 }
1186
1187 return traceprobe_define_arg_fields(event_call, size, &tu->tp);
1188}
1189
1190#ifdef CONFIG_PERF_EVENTS
1191static bool
1192__uprobe_perf_filter(struct trace_uprobe_filter *filter, struct mm_struct *mm)
1193{
1194 struct perf_event *event;
1195
1196 if (filter->nr_systemwide)
1197 return true;
1198
1199 list_for_each_entry(event, &filter->perf_events, hw.tp_list) {
1200 if (event->hw.target->mm == mm)
1201 return true;
1202 }
1203
1204 return false;
1205}
1206
1207static inline bool
1208uprobe_filter_event(struct trace_uprobe *tu, struct perf_event *event)
1209{
1210 return __uprobe_perf_filter(&tu->filter, event->hw.target->mm);
1211}
1212
1213static int uprobe_perf_close(struct trace_uprobe *tu, struct perf_event *event)
1214{
1215 bool done;
1216
1217 write_lock(&tu->filter.rwlock);
1218 if (event->hw.target) {
1219 list_del(&event->hw.tp_list);
1220 done = tu->filter.nr_systemwide ||
1221 (event->hw.target->flags & PF_EXITING) ||
1222 uprobe_filter_event(tu, event);
1223 } else {
1224 tu->filter.nr_systemwide--;
1225 done = tu->filter.nr_systemwide;
1226 }
1227 write_unlock(&tu->filter.rwlock);
1228
1229 if (!done)
1230 return uprobe_apply(tu->inode, tu->offset, &tu->consumer, false);
1231
1232 return 0;
1233}
1234
1235static int uprobe_perf_open(struct trace_uprobe *tu, struct perf_event *event)
1236{
1237 bool done;
1238 int err;
1239
1240 write_lock(&tu->filter.rwlock);
1241 if (event->hw.target) {
1242 /*
1243 * event->parent != NULL means copy_process(), we can avoid
1244 * uprobe_apply(). current->mm must be probed and we can rely
1245 * on dup_mmap() which preserves the already installed bp's.
1246 *
1247 * attr.enable_on_exec means that exec/mmap will install the
1248 * breakpoints we need.
1249 */
1250 done = tu->filter.nr_systemwide ||
1251 event->parent || event->attr.enable_on_exec ||
1252 uprobe_filter_event(tu, event);
1253 list_add(&event->hw.tp_list, &tu->filter.perf_events);
1254 } else {
1255 done = tu->filter.nr_systemwide;
1256 tu->filter.nr_systemwide++;
1257 }
1258 write_unlock(&tu->filter.rwlock);
1259
1260 err = 0;
1261 if (!done) {
1262 err = uprobe_apply(tu->inode, tu->offset, &tu->consumer, true);
1263 if (err)
1264 uprobe_perf_close(tu, event);
1265 }
1266 return err;
1267}
1268
1269static int uprobe_perf_multi_call(struct trace_event_call *call,
1270 struct perf_event *event,
1271 int (*op)(struct trace_uprobe *tu, struct perf_event *event))
1272{
1273 struct trace_probe *pos, *tp;
1274 struct trace_uprobe *tu;
1275 int ret = 0;
1276
1277 tp = trace_probe_primary_from_call(call);
1278 if (WARN_ON_ONCE(!tp))
1279 return -ENODEV;
1280
1281 list_for_each_entry(pos, trace_probe_probe_list(tp), list) {
1282 tu = container_of(pos, struct trace_uprobe, tp);
1283 ret = op(tu, event);
1284 if (ret)
1285 break;
1286 }
1287
1288 return ret;
1289}
1290static bool uprobe_perf_filter(struct uprobe_consumer *uc,
1291 enum uprobe_filter_ctx ctx, struct mm_struct *mm)
1292{
1293 struct trace_uprobe *tu;
1294 int ret;
1295
1296 tu = container_of(uc, struct trace_uprobe, consumer);
1297 read_lock(&tu->filter.rwlock);
1298 ret = __uprobe_perf_filter(&tu->filter, mm);
1299 read_unlock(&tu->filter.rwlock);
1300
1301 return ret;
1302}
1303
1304static void __uprobe_perf_func(struct trace_uprobe *tu,
1305 unsigned long func, struct pt_regs *regs,
1306 struct uprobe_cpu_buffer *ucb, int dsize)
1307{
1308 struct trace_event_call *call = trace_probe_event_call(&tu->tp);
1309 struct uprobe_trace_entry_head *entry;
1310 struct hlist_head *head;
1311 void *data;
1312 int size, esize;
1313 int rctx;
1314
1315 if (bpf_prog_array_valid(call) && !trace_call_bpf(call, regs))
1316 return;
1317
1318 esize = SIZEOF_TRACE_ENTRY(is_ret_probe(tu));
1319
1320 size = esize + tu->tp.size + dsize;
1321 size = ALIGN(size + sizeof(u32), sizeof(u64)) - sizeof(u32);
1322 if (WARN_ONCE(size > PERF_MAX_TRACE_SIZE, "profile buffer not large enough"))
1323 return;
1324
1325 preempt_disable();
1326 head = this_cpu_ptr(call->perf_events);
1327 if (hlist_empty(head))
1328 goto out;
1329
1330 entry = perf_trace_buf_alloc(size, NULL, &rctx);
1331 if (!entry)
1332 goto out;
1333
1334 if (is_ret_probe(tu)) {
1335 entry->vaddr[0] = func;
1336 entry->vaddr[1] = instruction_pointer(regs);
1337 data = DATAOF_TRACE_ENTRY(entry, true);
1338 } else {
1339 entry->vaddr[0] = instruction_pointer(regs);
1340 data = DATAOF_TRACE_ENTRY(entry, false);
1341 }
1342
1343 memcpy(data, ucb->buf, tu->tp.size + dsize);
1344
1345 if (size - esize > tu->tp.size + dsize) {
1346 int len = tu->tp.size + dsize;
1347
1348 memset(data + len, 0, size - esize - len);
1349 }
1350
1351 perf_trace_buf_submit(entry, size, rctx, call->event.type, 1, regs,
1352 head, NULL);
1353 out:
1354 preempt_enable();
1355}
1356
1357/* uprobe profile handler */
1358static int uprobe_perf_func(struct trace_uprobe *tu, struct pt_regs *regs,
1359 struct uprobe_cpu_buffer *ucb, int dsize)
1360{
1361 if (!uprobe_perf_filter(&tu->consumer, 0, current->mm))
1362 return UPROBE_HANDLER_REMOVE;
1363
1364 if (!is_ret_probe(tu))
1365 __uprobe_perf_func(tu, 0, regs, ucb, dsize);
1366 return 0;
1367}
1368
1369static void uretprobe_perf_func(struct trace_uprobe *tu, unsigned long func,
1370 struct pt_regs *regs,
1371 struct uprobe_cpu_buffer *ucb, int dsize)
1372{
1373 __uprobe_perf_func(tu, func, regs, ucb, dsize);
1374}
1375
1376int bpf_get_uprobe_info(const struct perf_event *event, u32 *fd_type,
1377 const char **filename, u64 *probe_offset,
1378 bool perf_type_tracepoint)
1379{
1380 const char *pevent = trace_event_name(event->tp_event);
1381 const char *group = event->tp_event->class->system;
1382 struct trace_uprobe *tu;
1383
1384 if (perf_type_tracepoint)
1385 tu = find_probe_event(pevent, group);
1386 else
1387 tu = event->tp_event->data;
1388 if (!tu)
1389 return -EINVAL;
1390
1391 *fd_type = is_ret_probe(tu) ? BPF_FD_TYPE_URETPROBE
1392 : BPF_FD_TYPE_UPROBE;
1393 *filename = tu->filename;
1394 *probe_offset = tu->offset;
1395 return 0;
1396}
1397#endif /* CONFIG_PERF_EVENTS */
1398
1399static int
1400trace_uprobe_register(struct trace_event_call *event, enum trace_reg type,
1401 void *data)
1402{
1403 struct trace_event_file *file = data;
1404
1405 switch (type) {
1406 case TRACE_REG_REGISTER:
1407 return probe_event_enable(event, file, NULL);
1408
1409 case TRACE_REG_UNREGISTER:
1410 probe_event_disable(event, file);
1411 return 0;
1412
1413#ifdef CONFIG_PERF_EVENTS
1414 case TRACE_REG_PERF_REGISTER:
1415 return probe_event_enable(event, NULL, uprobe_perf_filter);
1416
1417 case TRACE_REG_PERF_UNREGISTER:
1418 probe_event_disable(event, NULL);
1419 return 0;
1420
1421 case TRACE_REG_PERF_OPEN:
1422 return uprobe_perf_multi_call(event, data, uprobe_perf_open);
1423
1424 case TRACE_REG_PERF_CLOSE:
1425 return uprobe_perf_multi_call(event, data, uprobe_perf_close);
1426
1427#endif
1428 default:
1429 return 0;
1430 }
1431 return 0;
1432}
1433
1434static int uprobe_dispatcher(struct uprobe_consumer *con, struct pt_regs *regs)
1435{
1436 struct trace_uprobe *tu;
1437 struct uprobe_dispatch_data udd;
1438 struct uprobe_cpu_buffer *ucb;
1439 int dsize, esize;
1440 int ret = 0;
1441
1442
1443 tu = container_of(con, struct trace_uprobe, consumer);
1444 tu->nhit++;
1445
1446 udd.tu = tu;
1447 udd.bp_addr = instruction_pointer(regs);
1448
1449 current->utask->vaddr = (unsigned long) &udd;
1450
1451 if (WARN_ON_ONCE(!uprobe_cpu_buffer))
1452 return 0;
1453
1454 dsize = __get_data_size(&tu->tp, regs);
1455 esize = SIZEOF_TRACE_ENTRY(is_ret_probe(tu));
1456
1457 ucb = uprobe_buffer_get();
1458 store_trace_args(ucb->buf, &tu->tp, regs, esize, dsize);
1459
1460 if (trace_probe_test_flag(&tu->tp, TP_FLAG_TRACE))
1461 ret |= uprobe_trace_func(tu, regs, ucb, dsize);
1462
1463#ifdef CONFIG_PERF_EVENTS
1464 if (trace_probe_test_flag(&tu->tp, TP_FLAG_PROFILE))
1465 ret |= uprobe_perf_func(tu, regs, ucb, dsize);
1466#endif
1467 uprobe_buffer_put(ucb);
1468 return ret;
1469}
1470
1471static int uretprobe_dispatcher(struct uprobe_consumer *con,
1472 unsigned long func, struct pt_regs *regs)
1473{
1474 struct trace_uprobe *tu;
1475 struct uprobe_dispatch_data udd;
1476 struct uprobe_cpu_buffer *ucb;
1477 int dsize, esize;
1478
1479 tu = container_of(con, struct trace_uprobe, consumer);
1480
1481 udd.tu = tu;
1482 udd.bp_addr = func;
1483
1484 current->utask->vaddr = (unsigned long) &udd;
1485
1486 if (WARN_ON_ONCE(!uprobe_cpu_buffer))
1487 return 0;
1488
1489 dsize = __get_data_size(&tu->tp, regs);
1490 esize = SIZEOF_TRACE_ENTRY(is_ret_probe(tu));
1491
1492 ucb = uprobe_buffer_get();
1493 store_trace_args(ucb->buf, &tu->tp, regs, esize, dsize);
1494
1495 if (trace_probe_test_flag(&tu->tp, TP_FLAG_TRACE))
1496 uretprobe_trace_func(tu, func, regs, ucb, dsize);
1497
1498#ifdef CONFIG_PERF_EVENTS
1499 if (trace_probe_test_flag(&tu->tp, TP_FLAG_PROFILE))
1500 uretprobe_perf_func(tu, func, regs, ucb, dsize);
1501#endif
1502 uprobe_buffer_put(ucb);
1503 return 0;
1504}
1505
1506static struct trace_event_functions uprobe_funcs = {
1507 .trace = print_uprobe_event
1508};
1509
1510static inline void init_trace_event_call(struct trace_uprobe *tu)
1511{
1512 struct trace_event_call *call = trace_probe_event_call(&tu->tp);
1513
1514 call->event.funcs = &uprobe_funcs;
1515 call->class->define_fields = uprobe_event_define_fields;
1516
1517 call->flags = TRACE_EVENT_FL_UPROBE | TRACE_EVENT_FL_CAP_ANY;
1518 call->class->reg = trace_uprobe_register;
1519}
1520
1521static int register_uprobe_event(struct trace_uprobe *tu)
1522{
1523 init_trace_event_call(tu);
1524
1525 return trace_probe_register_event_call(&tu->tp);
1526}
1527
1528static int unregister_uprobe_event(struct trace_uprobe *tu)
1529{
1530 return trace_probe_unregister_event_call(&tu->tp);
1531}
1532
1533#ifdef CONFIG_PERF_EVENTS
1534struct trace_event_call *
1535create_local_trace_uprobe(char *name, unsigned long offs,
1536 unsigned long ref_ctr_offset, bool is_return)
1537{
1538 struct trace_uprobe *tu;
1539 struct path path;
1540 int ret;
1541
1542 ret = kern_path(name, LOOKUP_FOLLOW, &path);
1543 if (ret)
1544 return ERR_PTR(ret);
1545
1546 if (!d_is_reg(path.dentry)) {
1547 path_put(&path);
1548 return ERR_PTR(-EINVAL);
1549 }
1550
1551 /*
1552 * local trace_kprobes are not added to dyn_event, so they are never
1553 * searched in find_trace_kprobe(). Therefore, there is no concern of
1554 * duplicated name "DUMMY_EVENT" here.
1555 */
1556 tu = alloc_trace_uprobe(UPROBE_EVENT_SYSTEM, "DUMMY_EVENT", 0,
1557 is_return);
1558
1559 if (IS_ERR(tu)) {
1560 pr_info("Failed to allocate trace_uprobe.(%d)\n",
1561 (int)PTR_ERR(tu));
1562 path_put(&path);
1563 return ERR_CAST(tu);
1564 }
1565
1566 tu->offset = offs;
1567 tu->path = path;
1568 tu->ref_ctr_offset = ref_ctr_offset;
1569 tu->filename = kstrdup(name, GFP_KERNEL);
1570 init_trace_event_call(tu);
1571
1572 if (traceprobe_set_print_fmt(&tu->tp, is_ret_probe(tu)) < 0) {
1573 ret = -ENOMEM;
1574 goto error;
1575 }
1576
1577 return trace_probe_event_call(&tu->tp);
1578error:
1579 free_trace_uprobe(tu);
1580 return ERR_PTR(ret);
1581}
1582
1583void destroy_local_trace_uprobe(struct trace_event_call *event_call)
1584{
1585 struct trace_uprobe *tu;
1586
1587 tu = trace_uprobe_primary_from_call(event_call);
1588
1589 free_trace_uprobe(tu);
1590}
1591#endif /* CONFIG_PERF_EVENTS */
1592
1593/* Make a trace interface for controling probe points */
1594static __init int init_uprobe_trace(void)
1595{
1596 struct dentry *d_tracer;
1597 int ret;
1598
1599 ret = dyn_event_register(&trace_uprobe_ops);
1600 if (ret)
1601 return ret;
1602
1603 d_tracer = tracing_init_dentry();
1604 if (IS_ERR(d_tracer))
1605 return 0;
1606
1607 trace_create_file("uprobe_events", 0644, d_tracer,
1608 NULL, &uprobe_events_ops);
1609 /* Profile interface */
1610 trace_create_file("uprobe_profile", 0444, d_tracer,
1611 NULL, &uprobe_profile_ops);
1612 return 0;
1613}
1614
1615fs_initcall(init_uprobe_trace);