Loading...
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);
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
37static int trace_uprobe_create(const char *raw_command);
38static int trace_uprobe_show(struct seq_file *m, struct dyn_event *ev);
39static int trace_uprobe_release(struct dyn_event *ev);
40static bool trace_uprobe_is_busy(struct dyn_event *ev);
41static bool trace_uprobe_match(const char *system, const char *event,
42 int argc, const char **argv, struct dyn_event *ev);
43
44static struct dyn_event_operations trace_uprobe_ops = {
45 .create = trace_uprobe_create,
46 .show = trace_uprobe_show,
47 .is_busy = trace_uprobe_is_busy,
48 .free = trace_uprobe_release,
49 .match = trace_uprobe_match,
50};
51
52/*
53 * uprobe event core functions
54 */
55struct trace_uprobe {
56 struct dyn_event devent;
57 struct uprobe_consumer consumer;
58 struct path path;
59 struct inode *inode;
60 char *filename;
61 unsigned long offset;
62 unsigned long ref_ctr_offset;
63 unsigned long nhit;
64 struct trace_probe tp;
65};
66
67static bool is_trace_uprobe(struct dyn_event *ev)
68{
69 return ev->ops == &trace_uprobe_ops;
70}
71
72static struct trace_uprobe *to_trace_uprobe(struct dyn_event *ev)
73{
74 return container_of(ev, struct trace_uprobe, devent);
75}
76
77/**
78 * for_each_trace_uprobe - iterate over the trace_uprobe list
79 * @pos: the struct trace_uprobe * for each entry
80 * @dpos: the struct dyn_event * to use as a loop cursor
81 */
82#define for_each_trace_uprobe(pos, dpos) \
83 for_each_dyn_event(dpos) \
84 if (is_trace_uprobe(dpos) && (pos = to_trace_uprobe(dpos)))
85
86#define SIZEOF_TRACE_UPROBE(n) \
87 (offsetof(struct trace_uprobe, tp.args) + \
88 (sizeof(struct probe_arg) * (n)))
89
90static int register_uprobe_event(struct trace_uprobe *tu);
91static int unregister_uprobe_event(struct trace_uprobe *tu);
92
93struct uprobe_dispatch_data {
94 struct trace_uprobe *tu;
95 unsigned long bp_addr;
96};
97
98static int uprobe_dispatcher(struct uprobe_consumer *con, struct pt_regs *regs);
99static int uretprobe_dispatcher(struct uprobe_consumer *con,
100 unsigned long func, struct pt_regs *regs);
101
102#ifdef CONFIG_STACK_GROWSUP
103static unsigned long adjust_stack_addr(unsigned long addr, unsigned int n)
104{
105 return addr - (n * sizeof(long));
106}
107#else
108static unsigned long adjust_stack_addr(unsigned long addr, unsigned int n)
109{
110 return addr + (n * sizeof(long));
111}
112#endif
113
114static unsigned long get_user_stack_nth(struct pt_regs *regs, unsigned int n)
115{
116 unsigned long ret;
117 unsigned long addr = user_stack_pointer(regs);
118
119 addr = adjust_stack_addr(addr, n);
120
121 if (copy_from_user(&ret, (void __force __user *) addr, sizeof(ret)))
122 return 0;
123
124 return ret;
125}
126
127/*
128 * Uprobes-specific fetch functions
129 */
130static nokprobe_inline int
131probe_mem_read(void *dest, void *src, size_t size)
132{
133 void __user *vaddr = (void __force __user *)src;
134
135 return copy_from_user(dest, vaddr, size) ? -EFAULT : 0;
136}
137
138static nokprobe_inline int
139probe_mem_read_user(void *dest, void *src, size_t size)
140{
141 return probe_mem_read(dest, src, size);
142}
143
144/*
145 * Fetch a null-terminated string. Caller MUST set *(u32 *)dest with max
146 * length and relative data location.
147 */
148static nokprobe_inline int
149fetch_store_string(unsigned long addr, void *dest, void *base)
150{
151 long ret;
152 u32 loc = *(u32 *)dest;
153 int maxlen = get_loc_len(loc);
154 u8 *dst = get_loc_data(dest, base);
155 void __user *src = (void __force __user *) addr;
156
157 if (unlikely(!maxlen))
158 return -ENOMEM;
159
160 if (addr == FETCH_TOKEN_COMM)
161 ret = strlcpy(dst, current->comm, maxlen);
162 else
163 ret = strncpy_from_user(dst, src, maxlen);
164 if (ret >= 0) {
165 if (ret == maxlen)
166 dst[ret - 1] = '\0';
167 else
168 /*
169 * Include the terminating null byte. In this case it
170 * was copied by strncpy_from_user but not accounted
171 * for in ret.
172 */
173 ret++;
174 *(u32 *)dest = make_data_loc(ret, (void *)dst - base);
175 }
176
177 return ret;
178}
179
180static nokprobe_inline int
181fetch_store_string_user(unsigned long addr, void *dest, void *base)
182{
183 return fetch_store_string(addr, dest, base);
184}
185
186/* Return the length of string -- including null terminal byte */
187static nokprobe_inline int
188fetch_store_strlen(unsigned long addr)
189{
190 int len;
191 void __user *vaddr = (void __force __user *) addr;
192
193 if (addr == FETCH_TOKEN_COMM)
194 len = strlen(current->comm) + 1;
195 else
196 len = strnlen_user(vaddr, MAX_STRING_SIZE);
197
198 return (len > MAX_STRING_SIZE) ? 0 : len;
199}
200
201static nokprobe_inline int
202fetch_store_strlen_user(unsigned long addr)
203{
204 return fetch_store_strlen(addr);
205}
206
207static unsigned long translate_user_vaddr(unsigned long file_offset)
208{
209 unsigned long base_addr;
210 struct uprobe_dispatch_data *udd;
211
212 udd = (void *) current->utask->vaddr;
213
214 base_addr = udd->bp_addr - udd->tu->offset;
215 return base_addr + file_offset;
216}
217
218/* Note that we don't verify it, since the code does not come from user space */
219static int
220process_fetch_insn(struct fetch_insn *code, struct pt_regs *regs, void *dest,
221 void *base)
222{
223 unsigned long val;
224
225 /* 1st stage: get value from context */
226 switch (code->op) {
227 case FETCH_OP_REG:
228 val = regs_get_register(regs, code->param);
229 break;
230 case FETCH_OP_STACK:
231 val = get_user_stack_nth(regs, code->param);
232 break;
233 case FETCH_OP_STACKP:
234 val = user_stack_pointer(regs);
235 break;
236 case FETCH_OP_RETVAL:
237 val = regs_return_value(regs);
238 break;
239 case FETCH_OP_IMM:
240 val = code->immediate;
241 break;
242 case FETCH_OP_COMM:
243 val = FETCH_TOKEN_COMM;
244 break;
245 case FETCH_OP_DATA:
246 val = (unsigned long)code->data;
247 break;
248 case FETCH_OP_FOFFS:
249 val = translate_user_vaddr(code->immediate);
250 break;
251 default:
252 return -EILSEQ;
253 }
254 code++;
255
256 return process_fetch_insn_bottom(code, val, dest, base);
257}
258NOKPROBE_SYMBOL(process_fetch_insn)
259
260static inline void init_trace_uprobe_filter(struct trace_uprobe_filter *filter)
261{
262 rwlock_init(&filter->rwlock);
263 filter->nr_systemwide = 0;
264 INIT_LIST_HEAD(&filter->perf_events);
265}
266
267static inline bool uprobe_filter_is_empty(struct trace_uprobe_filter *filter)
268{
269 return !filter->nr_systemwide && list_empty(&filter->perf_events);
270}
271
272static inline bool is_ret_probe(struct trace_uprobe *tu)
273{
274 return tu->consumer.ret_handler != NULL;
275}
276
277static bool trace_uprobe_is_busy(struct dyn_event *ev)
278{
279 struct trace_uprobe *tu = to_trace_uprobe(ev);
280
281 return trace_probe_is_enabled(&tu->tp);
282}
283
284static bool trace_uprobe_match_command_head(struct trace_uprobe *tu,
285 int argc, const char **argv)
286{
287 char buf[MAX_ARGSTR_LEN + 1];
288 int len;
289
290 if (!argc)
291 return true;
292
293 len = strlen(tu->filename);
294 if (strncmp(tu->filename, argv[0], len) || argv[0][len] != ':')
295 return false;
296
297 if (tu->ref_ctr_offset == 0)
298 snprintf(buf, sizeof(buf), "0x%0*lx",
299 (int)(sizeof(void *) * 2), tu->offset);
300 else
301 snprintf(buf, sizeof(buf), "0x%0*lx(0x%lx)",
302 (int)(sizeof(void *) * 2), tu->offset,
303 tu->ref_ctr_offset);
304 if (strcmp(buf, &argv[0][len + 1]))
305 return false;
306
307 argc--; argv++;
308
309 return trace_probe_match_command_args(&tu->tp, argc, argv);
310}
311
312static bool trace_uprobe_match(const char *system, const char *event,
313 int argc, const char **argv, struct dyn_event *ev)
314{
315 struct trace_uprobe *tu = to_trace_uprobe(ev);
316
317 return strcmp(trace_probe_name(&tu->tp), event) == 0 &&
318 (!system || strcmp(trace_probe_group_name(&tu->tp), system) == 0) &&
319 trace_uprobe_match_command_head(tu, argc, argv);
320}
321
322static nokprobe_inline struct trace_uprobe *
323trace_uprobe_primary_from_call(struct trace_event_call *call)
324{
325 struct trace_probe *tp;
326
327 tp = trace_probe_primary_from_call(call);
328 if (WARN_ON_ONCE(!tp))
329 return NULL;
330
331 return container_of(tp, struct trace_uprobe, tp);
332}
333
334/*
335 * Allocate new trace_uprobe and initialize it (including uprobes).
336 */
337static struct trace_uprobe *
338alloc_trace_uprobe(const char *group, const char *event, int nargs, bool is_ret)
339{
340 struct trace_uprobe *tu;
341 int ret;
342
343 tu = kzalloc(SIZEOF_TRACE_UPROBE(nargs), GFP_KERNEL);
344 if (!tu)
345 return ERR_PTR(-ENOMEM);
346
347 ret = trace_probe_init(&tu->tp, event, group, true);
348 if (ret < 0)
349 goto error;
350
351 dyn_event_init(&tu->devent, &trace_uprobe_ops);
352 tu->consumer.handler = uprobe_dispatcher;
353 if (is_ret)
354 tu->consumer.ret_handler = uretprobe_dispatcher;
355 init_trace_uprobe_filter(tu->tp.event->filter);
356 return tu;
357
358error:
359 kfree(tu);
360
361 return ERR_PTR(ret);
362}
363
364static void free_trace_uprobe(struct trace_uprobe *tu)
365{
366 if (!tu)
367 return;
368
369 path_put(&tu->path);
370 trace_probe_cleanup(&tu->tp);
371 kfree(tu->filename);
372 kfree(tu);
373}
374
375static struct trace_uprobe *find_probe_event(const char *event, const char *group)
376{
377 struct dyn_event *pos;
378 struct trace_uprobe *tu;
379
380 for_each_trace_uprobe(tu, pos)
381 if (strcmp(trace_probe_name(&tu->tp), event) == 0 &&
382 strcmp(trace_probe_group_name(&tu->tp), group) == 0)
383 return tu;
384
385 return NULL;
386}
387
388/* Unregister a trace_uprobe and probe_event */
389static int unregister_trace_uprobe(struct trace_uprobe *tu)
390{
391 int ret;
392
393 if (trace_probe_has_sibling(&tu->tp))
394 goto unreg;
395
396 ret = unregister_uprobe_event(tu);
397 if (ret)
398 return ret;
399
400unreg:
401 dyn_event_remove(&tu->devent);
402 trace_probe_unlink(&tu->tp);
403 free_trace_uprobe(tu);
404 return 0;
405}
406
407static bool trace_uprobe_has_same_uprobe(struct trace_uprobe *orig,
408 struct trace_uprobe *comp)
409{
410 struct trace_probe_event *tpe = orig->tp.event;
411 struct trace_probe *pos;
412 struct inode *comp_inode = d_real_inode(comp->path.dentry);
413 int i;
414
415 list_for_each_entry(pos, &tpe->probes, list) {
416 orig = container_of(pos, struct trace_uprobe, tp);
417 if (comp_inode != d_real_inode(orig->path.dentry) ||
418 comp->offset != orig->offset)
419 continue;
420
421 /*
422 * trace_probe_compare_arg_type() ensured that nr_args and
423 * each argument name and type are same. Let's compare comm.
424 */
425 for (i = 0; i < orig->tp.nr_args; i++) {
426 if (strcmp(orig->tp.args[i].comm,
427 comp->tp.args[i].comm))
428 break;
429 }
430
431 if (i == orig->tp.nr_args)
432 return true;
433 }
434
435 return false;
436}
437
438static int append_trace_uprobe(struct trace_uprobe *tu, struct trace_uprobe *to)
439{
440 int ret;
441
442 ret = trace_probe_compare_arg_type(&tu->tp, &to->tp);
443 if (ret) {
444 /* Note that argument starts index = 2 */
445 trace_probe_log_set_index(ret + 1);
446 trace_probe_log_err(0, DIFF_ARG_TYPE);
447 return -EEXIST;
448 }
449 if (trace_uprobe_has_same_uprobe(to, tu)) {
450 trace_probe_log_set_index(0);
451 trace_probe_log_err(0, SAME_PROBE);
452 return -EEXIST;
453 }
454
455 /* Append to existing event */
456 ret = trace_probe_append(&tu->tp, &to->tp);
457 if (!ret)
458 dyn_event_add(&tu->devent);
459
460 return ret;
461}
462
463/*
464 * Uprobe with multiple reference counter is not allowed. i.e.
465 * If inode and offset matches, reference counter offset *must*
466 * match as well. Though, there is one exception: If user is
467 * replacing old trace_uprobe with new one(same group/event),
468 * then we allow same uprobe with new reference counter as far
469 * as the new one does not conflict with any other existing
470 * ones.
471 */
472static int validate_ref_ctr_offset(struct trace_uprobe *new)
473{
474 struct dyn_event *pos;
475 struct trace_uprobe *tmp;
476 struct inode *new_inode = d_real_inode(new->path.dentry);
477
478 for_each_trace_uprobe(tmp, pos) {
479 if (new_inode == d_real_inode(tmp->path.dentry) &&
480 new->offset == tmp->offset &&
481 new->ref_ctr_offset != tmp->ref_ctr_offset) {
482 pr_warn("Reference counter offset mismatch.");
483 return -EINVAL;
484 }
485 }
486 return 0;
487}
488
489/* Register a trace_uprobe and probe_event */
490static int register_trace_uprobe(struct trace_uprobe *tu)
491{
492 struct trace_uprobe *old_tu;
493 int ret;
494
495 mutex_lock(&event_mutex);
496
497 ret = validate_ref_ctr_offset(tu);
498 if (ret)
499 goto end;
500
501 /* register as an event */
502 old_tu = find_probe_event(trace_probe_name(&tu->tp),
503 trace_probe_group_name(&tu->tp));
504 if (old_tu) {
505 if (is_ret_probe(tu) != is_ret_probe(old_tu)) {
506 trace_probe_log_set_index(0);
507 trace_probe_log_err(0, DIFF_PROBE_TYPE);
508 ret = -EEXIST;
509 } else {
510 ret = append_trace_uprobe(tu, old_tu);
511 }
512 goto end;
513 }
514
515 ret = register_uprobe_event(tu);
516 if (ret) {
517 if (ret == -EEXIST) {
518 trace_probe_log_set_index(0);
519 trace_probe_log_err(0, EVENT_EXIST);
520 } else
521 pr_warn("Failed to register probe event(%d)\n", ret);
522 goto end;
523 }
524
525 dyn_event_add(&tu->devent);
526
527end:
528 mutex_unlock(&event_mutex);
529
530 return ret;
531}
532
533/*
534 * Argument syntax:
535 * - Add uprobe: p|r[:[GRP/]EVENT] PATH:OFFSET[%return][(REF)] [FETCHARGS]
536 */
537static int __trace_uprobe_create(int argc, const char **argv)
538{
539 struct trace_uprobe *tu;
540 const char *event = NULL, *group = UPROBE_EVENT_SYSTEM;
541 char *arg, *filename, *rctr, *rctr_end, *tmp;
542 char buf[MAX_EVENT_NAME_LEN];
543 struct path path;
544 unsigned long offset, ref_ctr_offset;
545 bool is_return = false;
546 int i, ret;
547
548 ret = 0;
549 ref_ctr_offset = 0;
550
551 switch (argv[0][0]) {
552 case 'r':
553 is_return = true;
554 break;
555 case 'p':
556 break;
557 default:
558 return -ECANCELED;
559 }
560
561 if (argc < 2)
562 return -ECANCELED;
563
564 if (argv[0][1] == ':')
565 event = &argv[0][2];
566
567 if (!strchr(argv[1], '/'))
568 return -ECANCELED;
569
570 filename = kstrdup(argv[1], GFP_KERNEL);
571 if (!filename)
572 return -ENOMEM;
573
574 /* Find the last occurrence, in case the path contains ':' too. */
575 arg = strrchr(filename, ':');
576 if (!arg || !isdigit(arg[1])) {
577 kfree(filename);
578 return -ECANCELED;
579 }
580
581 trace_probe_log_init("trace_uprobe", argc, argv);
582 trace_probe_log_set_index(1); /* filename is the 2nd argument */
583
584 *arg++ = '\0';
585 ret = kern_path(filename, LOOKUP_FOLLOW, &path);
586 if (ret) {
587 trace_probe_log_err(0, FILE_NOT_FOUND);
588 kfree(filename);
589 trace_probe_log_clear();
590 return ret;
591 }
592 if (!d_is_reg(path.dentry)) {
593 trace_probe_log_err(0, NO_REGULAR_FILE);
594 ret = -EINVAL;
595 goto fail_address_parse;
596 }
597
598 /* Parse reference counter offset if specified. */
599 rctr = strchr(arg, '(');
600 if (rctr) {
601 rctr_end = strchr(rctr, ')');
602 if (!rctr_end) {
603 ret = -EINVAL;
604 rctr_end = rctr + strlen(rctr);
605 trace_probe_log_err(rctr_end - filename,
606 REFCNT_OPEN_BRACE);
607 goto fail_address_parse;
608 } else if (rctr_end[1] != '\0') {
609 ret = -EINVAL;
610 trace_probe_log_err(rctr_end + 1 - filename,
611 BAD_REFCNT_SUFFIX);
612 goto fail_address_parse;
613 }
614
615 *rctr++ = '\0';
616 *rctr_end = '\0';
617 ret = kstrtoul(rctr, 0, &ref_ctr_offset);
618 if (ret) {
619 trace_probe_log_err(rctr - filename, BAD_REFCNT);
620 goto fail_address_parse;
621 }
622 }
623
624 /* Check if there is %return suffix */
625 tmp = strchr(arg, '%');
626 if (tmp) {
627 if (!strcmp(tmp, "%return")) {
628 *tmp = '\0';
629 is_return = true;
630 } else {
631 trace_probe_log_err(tmp - filename, BAD_ADDR_SUFFIX);
632 ret = -EINVAL;
633 goto fail_address_parse;
634 }
635 }
636
637 /* Parse uprobe offset. */
638 ret = kstrtoul(arg, 0, &offset);
639 if (ret) {
640 trace_probe_log_err(arg - filename, BAD_UPROBE_OFFS);
641 goto fail_address_parse;
642 }
643
644 /* setup a probe */
645 trace_probe_log_set_index(0);
646 if (event) {
647 ret = traceprobe_parse_event_name(&event, &group, buf,
648 event - argv[0]);
649 if (ret)
650 goto fail_address_parse;
651 } else {
652 char *tail;
653 char *ptr;
654
655 tail = kstrdup(kbasename(filename), GFP_KERNEL);
656 if (!tail) {
657 ret = -ENOMEM;
658 goto fail_address_parse;
659 }
660
661 ptr = strpbrk(tail, ".-_");
662 if (ptr)
663 *ptr = '\0';
664
665 snprintf(buf, MAX_EVENT_NAME_LEN, "%c_%s_0x%lx", 'p', tail, offset);
666 event = buf;
667 kfree(tail);
668 }
669
670 argc -= 2;
671 argv += 2;
672
673 tu = alloc_trace_uprobe(group, event, argc, is_return);
674 if (IS_ERR(tu)) {
675 ret = PTR_ERR(tu);
676 /* This must return -ENOMEM otherwise there is a bug */
677 WARN_ON_ONCE(ret != -ENOMEM);
678 goto fail_address_parse;
679 }
680 tu->offset = offset;
681 tu->ref_ctr_offset = ref_ctr_offset;
682 tu->path = path;
683 tu->filename = filename;
684
685 /* parse arguments */
686 for (i = 0; i < argc && i < MAX_TRACE_ARGS; i++) {
687 tmp = kstrdup(argv[i], GFP_KERNEL);
688 if (!tmp) {
689 ret = -ENOMEM;
690 goto error;
691 }
692
693 trace_probe_log_set_index(i + 2);
694 ret = traceprobe_parse_probe_arg(&tu->tp, i, tmp,
695 is_return ? TPARG_FL_RETURN : 0);
696 kfree(tmp);
697 if (ret)
698 goto error;
699 }
700
701 ret = traceprobe_set_print_fmt(&tu->tp, is_ret_probe(tu));
702 if (ret < 0)
703 goto error;
704
705 ret = register_trace_uprobe(tu);
706 if (!ret)
707 goto out;
708
709error:
710 free_trace_uprobe(tu);
711out:
712 trace_probe_log_clear();
713 return ret;
714
715fail_address_parse:
716 trace_probe_log_clear();
717 path_put(&path);
718 kfree(filename);
719
720 return ret;
721}
722
723int trace_uprobe_create(const char *raw_command)
724{
725 return trace_probe_create(raw_command, __trace_uprobe_create);
726}
727
728static int create_or_delete_trace_uprobe(const char *raw_command)
729{
730 int ret;
731
732 if (raw_command[0] == '-')
733 return dyn_event_release(raw_command, &trace_uprobe_ops);
734
735 ret = trace_uprobe_create(raw_command);
736 return ret == -ECANCELED ? -EINVAL : ret;
737}
738
739static int trace_uprobe_release(struct dyn_event *ev)
740{
741 struct trace_uprobe *tu = to_trace_uprobe(ev);
742
743 return unregister_trace_uprobe(tu);
744}
745
746/* Probes listing interfaces */
747static int trace_uprobe_show(struct seq_file *m, struct dyn_event *ev)
748{
749 struct trace_uprobe *tu = to_trace_uprobe(ev);
750 char c = is_ret_probe(tu) ? 'r' : 'p';
751 int i;
752
753 seq_printf(m, "%c:%s/%s %s:0x%0*lx", c, trace_probe_group_name(&tu->tp),
754 trace_probe_name(&tu->tp), tu->filename,
755 (int)(sizeof(void *) * 2), tu->offset);
756
757 if (tu->ref_ctr_offset)
758 seq_printf(m, "(0x%lx)", tu->ref_ctr_offset);
759
760 for (i = 0; i < tu->tp.nr_args; i++)
761 seq_printf(m, " %s=%s", tu->tp.args[i].name, tu->tp.args[i].comm);
762
763 seq_putc(m, '\n');
764 return 0;
765}
766
767static int probes_seq_show(struct seq_file *m, void *v)
768{
769 struct dyn_event *ev = v;
770
771 if (!is_trace_uprobe(ev))
772 return 0;
773
774 return trace_uprobe_show(m, ev);
775}
776
777static const struct seq_operations probes_seq_op = {
778 .start = dyn_event_seq_start,
779 .next = dyn_event_seq_next,
780 .stop = dyn_event_seq_stop,
781 .show = probes_seq_show
782};
783
784static int probes_open(struct inode *inode, struct file *file)
785{
786 int ret;
787
788 ret = security_locked_down(LOCKDOWN_TRACEFS);
789 if (ret)
790 return ret;
791
792 if ((file->f_mode & FMODE_WRITE) && (file->f_flags & O_TRUNC)) {
793 ret = dyn_events_release_all(&trace_uprobe_ops);
794 if (ret)
795 return ret;
796 }
797
798 return seq_open(file, &probes_seq_op);
799}
800
801static ssize_t probes_write(struct file *file, const char __user *buffer,
802 size_t count, loff_t *ppos)
803{
804 return trace_parse_run_command(file, buffer, count, ppos,
805 create_or_delete_trace_uprobe);
806}
807
808static const struct file_operations uprobe_events_ops = {
809 .owner = THIS_MODULE,
810 .open = probes_open,
811 .read = seq_read,
812 .llseek = seq_lseek,
813 .release = seq_release,
814 .write = probes_write,
815};
816
817/* Probes profiling interfaces */
818static int probes_profile_seq_show(struct seq_file *m, void *v)
819{
820 struct dyn_event *ev = v;
821 struct trace_uprobe *tu;
822
823 if (!is_trace_uprobe(ev))
824 return 0;
825
826 tu = to_trace_uprobe(ev);
827 seq_printf(m, " %s %-44s %15lu\n", tu->filename,
828 trace_probe_name(&tu->tp), tu->nhit);
829 return 0;
830}
831
832static const struct seq_operations profile_seq_op = {
833 .start = dyn_event_seq_start,
834 .next = dyn_event_seq_next,
835 .stop = dyn_event_seq_stop,
836 .show = probes_profile_seq_show
837};
838
839static int profile_open(struct inode *inode, struct file *file)
840{
841 int ret;
842
843 ret = security_locked_down(LOCKDOWN_TRACEFS);
844 if (ret)
845 return ret;
846
847 return seq_open(file, &profile_seq_op);
848}
849
850static const struct file_operations uprobe_profile_ops = {
851 .owner = THIS_MODULE,
852 .open = profile_open,
853 .read = seq_read,
854 .llseek = seq_lseek,
855 .release = seq_release,
856};
857
858struct uprobe_cpu_buffer {
859 struct mutex mutex;
860 void *buf;
861};
862static struct uprobe_cpu_buffer __percpu *uprobe_cpu_buffer;
863static int uprobe_buffer_refcnt;
864
865static int uprobe_buffer_init(void)
866{
867 int cpu, err_cpu;
868
869 uprobe_cpu_buffer = alloc_percpu(struct uprobe_cpu_buffer);
870 if (uprobe_cpu_buffer == NULL)
871 return -ENOMEM;
872
873 for_each_possible_cpu(cpu) {
874 struct page *p = alloc_pages_node(cpu_to_node(cpu),
875 GFP_KERNEL, 0);
876 if (p == NULL) {
877 err_cpu = cpu;
878 goto err;
879 }
880 per_cpu_ptr(uprobe_cpu_buffer, cpu)->buf = page_address(p);
881 mutex_init(&per_cpu_ptr(uprobe_cpu_buffer, cpu)->mutex);
882 }
883
884 return 0;
885
886err:
887 for_each_possible_cpu(cpu) {
888 if (cpu == err_cpu)
889 break;
890 free_page((unsigned long)per_cpu_ptr(uprobe_cpu_buffer, cpu)->buf);
891 }
892
893 free_percpu(uprobe_cpu_buffer);
894 return -ENOMEM;
895}
896
897static int uprobe_buffer_enable(void)
898{
899 int ret = 0;
900
901 BUG_ON(!mutex_is_locked(&event_mutex));
902
903 if (uprobe_buffer_refcnt++ == 0) {
904 ret = uprobe_buffer_init();
905 if (ret < 0)
906 uprobe_buffer_refcnt--;
907 }
908
909 return ret;
910}
911
912static void uprobe_buffer_disable(void)
913{
914 int cpu;
915
916 BUG_ON(!mutex_is_locked(&event_mutex));
917
918 if (--uprobe_buffer_refcnt == 0) {
919 for_each_possible_cpu(cpu)
920 free_page((unsigned long)per_cpu_ptr(uprobe_cpu_buffer,
921 cpu)->buf);
922
923 free_percpu(uprobe_cpu_buffer);
924 uprobe_cpu_buffer = NULL;
925 }
926}
927
928static struct uprobe_cpu_buffer *uprobe_buffer_get(void)
929{
930 struct uprobe_cpu_buffer *ucb;
931 int cpu;
932
933 cpu = raw_smp_processor_id();
934 ucb = per_cpu_ptr(uprobe_cpu_buffer, cpu);
935
936 /*
937 * Use per-cpu buffers for fastest access, but we might migrate
938 * so the mutex makes sure we have sole access to it.
939 */
940 mutex_lock(&ucb->mutex);
941
942 return ucb;
943}
944
945static void uprobe_buffer_put(struct uprobe_cpu_buffer *ucb)
946{
947 mutex_unlock(&ucb->mutex);
948}
949
950static void __uprobe_trace_func(struct trace_uprobe *tu,
951 unsigned long func, struct pt_regs *regs,
952 struct uprobe_cpu_buffer *ucb, int dsize,
953 struct trace_event_file *trace_file)
954{
955 struct uprobe_trace_entry_head *entry;
956 struct trace_buffer *buffer;
957 struct ring_buffer_event *event;
958 void *data;
959 int size, esize;
960 struct trace_event_call *call = trace_probe_event_call(&tu->tp);
961
962 WARN_ON(call != trace_file->event_call);
963
964 if (WARN_ON_ONCE(tu->tp.size + dsize > PAGE_SIZE))
965 return;
966
967 if (trace_trigger_soft_disabled(trace_file))
968 return;
969
970 esize = SIZEOF_TRACE_ENTRY(is_ret_probe(tu));
971 size = esize + tu->tp.size + dsize;
972 event = trace_event_buffer_lock_reserve(&buffer, trace_file,
973 call->event.type, size, 0);
974 if (!event)
975 return;
976
977 entry = ring_buffer_event_data(event);
978 if (is_ret_probe(tu)) {
979 entry->vaddr[0] = func;
980 entry->vaddr[1] = instruction_pointer(regs);
981 data = DATAOF_TRACE_ENTRY(entry, true);
982 } else {
983 entry->vaddr[0] = instruction_pointer(regs);
984 data = DATAOF_TRACE_ENTRY(entry, false);
985 }
986
987 memcpy(data, ucb->buf, tu->tp.size + dsize);
988
989 event_trigger_unlock_commit(trace_file, buffer, event, entry, 0);
990}
991
992/* uprobe handler */
993static int uprobe_trace_func(struct trace_uprobe *tu, struct pt_regs *regs,
994 struct uprobe_cpu_buffer *ucb, int dsize)
995{
996 struct event_file_link *link;
997
998 if (is_ret_probe(tu))
999 return 0;
1000
1001 rcu_read_lock();
1002 trace_probe_for_each_link_rcu(link, &tu->tp)
1003 __uprobe_trace_func(tu, 0, regs, ucb, dsize, link->file);
1004 rcu_read_unlock();
1005
1006 return 0;
1007}
1008
1009static void uretprobe_trace_func(struct trace_uprobe *tu, unsigned long func,
1010 struct pt_regs *regs,
1011 struct uprobe_cpu_buffer *ucb, int dsize)
1012{
1013 struct event_file_link *link;
1014
1015 rcu_read_lock();
1016 trace_probe_for_each_link_rcu(link, &tu->tp)
1017 __uprobe_trace_func(tu, func, regs, ucb, dsize, link->file);
1018 rcu_read_unlock();
1019}
1020
1021/* Event entry printers */
1022static enum print_line_t
1023print_uprobe_event(struct trace_iterator *iter, int flags, struct trace_event *event)
1024{
1025 struct uprobe_trace_entry_head *entry;
1026 struct trace_seq *s = &iter->seq;
1027 struct trace_uprobe *tu;
1028 u8 *data;
1029
1030 entry = (struct uprobe_trace_entry_head *)iter->ent;
1031 tu = trace_uprobe_primary_from_call(
1032 container_of(event, struct trace_event_call, event));
1033 if (unlikely(!tu))
1034 goto out;
1035
1036 if (is_ret_probe(tu)) {
1037 trace_seq_printf(s, "%s: (0x%lx <- 0x%lx)",
1038 trace_probe_name(&tu->tp),
1039 entry->vaddr[1], entry->vaddr[0]);
1040 data = DATAOF_TRACE_ENTRY(entry, true);
1041 } else {
1042 trace_seq_printf(s, "%s: (0x%lx)",
1043 trace_probe_name(&tu->tp),
1044 entry->vaddr[0]);
1045 data = DATAOF_TRACE_ENTRY(entry, false);
1046 }
1047
1048 if (print_probe_args(s, tu->tp.args, tu->tp.nr_args, data, entry) < 0)
1049 goto out;
1050
1051 trace_seq_putc(s, '\n');
1052
1053 out:
1054 return trace_handle_return(s);
1055}
1056
1057typedef bool (*filter_func_t)(struct uprobe_consumer *self,
1058 enum uprobe_filter_ctx ctx,
1059 struct mm_struct *mm);
1060
1061static int trace_uprobe_enable(struct trace_uprobe *tu, filter_func_t filter)
1062{
1063 int ret;
1064
1065 tu->consumer.filter = filter;
1066 tu->inode = d_real_inode(tu->path.dentry);
1067
1068 if (tu->ref_ctr_offset)
1069 ret = uprobe_register_refctr(tu->inode, tu->offset,
1070 tu->ref_ctr_offset, &tu->consumer);
1071 else
1072 ret = uprobe_register(tu->inode, tu->offset, &tu->consumer);
1073
1074 if (ret)
1075 tu->inode = NULL;
1076
1077 return ret;
1078}
1079
1080static void __probe_event_disable(struct trace_probe *tp)
1081{
1082 struct trace_probe *pos;
1083 struct trace_uprobe *tu;
1084
1085 tu = container_of(tp, struct trace_uprobe, tp);
1086 WARN_ON(!uprobe_filter_is_empty(tu->tp.event->filter));
1087
1088 list_for_each_entry(pos, trace_probe_probe_list(tp), list) {
1089 tu = container_of(pos, struct trace_uprobe, tp);
1090 if (!tu->inode)
1091 continue;
1092
1093 uprobe_unregister(tu->inode, tu->offset, &tu->consumer);
1094 tu->inode = NULL;
1095 }
1096}
1097
1098static int probe_event_enable(struct trace_event_call *call,
1099 struct trace_event_file *file, filter_func_t filter)
1100{
1101 struct trace_probe *pos, *tp;
1102 struct trace_uprobe *tu;
1103 bool enabled;
1104 int ret;
1105
1106 tp = trace_probe_primary_from_call(call);
1107 if (WARN_ON_ONCE(!tp))
1108 return -ENODEV;
1109 enabled = trace_probe_is_enabled(tp);
1110
1111 /* This may also change "enabled" state */
1112 if (file) {
1113 if (trace_probe_test_flag(tp, TP_FLAG_PROFILE))
1114 return -EINTR;
1115
1116 ret = trace_probe_add_file(tp, file);
1117 if (ret < 0)
1118 return ret;
1119 } else {
1120 if (trace_probe_test_flag(tp, TP_FLAG_TRACE))
1121 return -EINTR;
1122
1123 trace_probe_set_flag(tp, TP_FLAG_PROFILE);
1124 }
1125
1126 tu = container_of(tp, struct trace_uprobe, tp);
1127 WARN_ON(!uprobe_filter_is_empty(tu->tp.event->filter));
1128
1129 if (enabled)
1130 return 0;
1131
1132 ret = uprobe_buffer_enable();
1133 if (ret)
1134 goto err_flags;
1135
1136 list_for_each_entry(pos, trace_probe_probe_list(tp), list) {
1137 tu = container_of(pos, struct trace_uprobe, tp);
1138 ret = trace_uprobe_enable(tu, filter);
1139 if (ret) {
1140 __probe_event_disable(tp);
1141 goto err_buffer;
1142 }
1143 }
1144
1145 return 0;
1146
1147 err_buffer:
1148 uprobe_buffer_disable();
1149
1150 err_flags:
1151 if (file)
1152 trace_probe_remove_file(tp, file);
1153 else
1154 trace_probe_clear_flag(tp, TP_FLAG_PROFILE);
1155
1156 return ret;
1157}
1158
1159static void probe_event_disable(struct trace_event_call *call,
1160 struct trace_event_file *file)
1161{
1162 struct trace_probe *tp;
1163
1164 tp = trace_probe_primary_from_call(call);
1165 if (WARN_ON_ONCE(!tp))
1166 return;
1167
1168 if (!trace_probe_is_enabled(tp))
1169 return;
1170
1171 if (file) {
1172 if (trace_probe_remove_file(tp, file) < 0)
1173 return;
1174
1175 if (trace_probe_is_enabled(tp))
1176 return;
1177 } else
1178 trace_probe_clear_flag(tp, TP_FLAG_PROFILE);
1179
1180 __probe_event_disable(tp);
1181 uprobe_buffer_disable();
1182}
1183
1184static int uprobe_event_define_fields(struct trace_event_call *event_call)
1185{
1186 int ret, size;
1187 struct uprobe_trace_entry_head field;
1188 struct trace_uprobe *tu;
1189
1190 tu = trace_uprobe_primary_from_call(event_call);
1191 if (unlikely(!tu))
1192 return -ENODEV;
1193
1194 if (is_ret_probe(tu)) {
1195 DEFINE_FIELD(unsigned long, vaddr[0], FIELD_STRING_FUNC, 0);
1196 DEFINE_FIELD(unsigned long, vaddr[1], FIELD_STRING_RETIP, 0);
1197 size = SIZEOF_TRACE_ENTRY(true);
1198 } else {
1199 DEFINE_FIELD(unsigned long, vaddr[0], FIELD_STRING_IP, 0);
1200 size = SIZEOF_TRACE_ENTRY(false);
1201 }
1202
1203 return traceprobe_define_arg_fields(event_call, size, &tu->tp);
1204}
1205
1206#ifdef CONFIG_PERF_EVENTS
1207static bool
1208__uprobe_perf_filter(struct trace_uprobe_filter *filter, struct mm_struct *mm)
1209{
1210 struct perf_event *event;
1211
1212 if (filter->nr_systemwide)
1213 return true;
1214
1215 list_for_each_entry(event, &filter->perf_events, hw.tp_list) {
1216 if (event->hw.target->mm == mm)
1217 return true;
1218 }
1219
1220 return false;
1221}
1222
1223static inline bool
1224trace_uprobe_filter_event(struct trace_uprobe_filter *filter,
1225 struct perf_event *event)
1226{
1227 return __uprobe_perf_filter(filter, event->hw.target->mm);
1228}
1229
1230static bool trace_uprobe_filter_remove(struct trace_uprobe_filter *filter,
1231 struct perf_event *event)
1232{
1233 bool done;
1234
1235 write_lock(&filter->rwlock);
1236 if (event->hw.target) {
1237 list_del(&event->hw.tp_list);
1238 done = filter->nr_systemwide ||
1239 (event->hw.target->flags & PF_EXITING) ||
1240 trace_uprobe_filter_event(filter, event);
1241 } else {
1242 filter->nr_systemwide--;
1243 done = filter->nr_systemwide;
1244 }
1245 write_unlock(&filter->rwlock);
1246
1247 return done;
1248}
1249
1250/* This returns true if the filter always covers target mm */
1251static bool trace_uprobe_filter_add(struct trace_uprobe_filter *filter,
1252 struct perf_event *event)
1253{
1254 bool done;
1255
1256 write_lock(&filter->rwlock);
1257 if (event->hw.target) {
1258 /*
1259 * event->parent != NULL means copy_process(), we can avoid
1260 * uprobe_apply(). current->mm must be probed and we can rely
1261 * on dup_mmap() which preserves the already installed bp's.
1262 *
1263 * attr.enable_on_exec means that exec/mmap will install the
1264 * breakpoints we need.
1265 */
1266 done = filter->nr_systemwide ||
1267 event->parent || event->attr.enable_on_exec ||
1268 trace_uprobe_filter_event(filter, event);
1269 list_add(&event->hw.tp_list, &filter->perf_events);
1270 } else {
1271 done = filter->nr_systemwide;
1272 filter->nr_systemwide++;
1273 }
1274 write_unlock(&filter->rwlock);
1275
1276 return done;
1277}
1278
1279static int uprobe_perf_close(struct trace_event_call *call,
1280 struct perf_event *event)
1281{
1282 struct trace_probe *pos, *tp;
1283 struct trace_uprobe *tu;
1284 int ret = 0;
1285
1286 tp = trace_probe_primary_from_call(call);
1287 if (WARN_ON_ONCE(!tp))
1288 return -ENODEV;
1289
1290 tu = container_of(tp, struct trace_uprobe, tp);
1291 if (trace_uprobe_filter_remove(tu->tp.event->filter, event))
1292 return 0;
1293
1294 list_for_each_entry(pos, trace_probe_probe_list(tp), list) {
1295 tu = container_of(pos, struct trace_uprobe, tp);
1296 ret = uprobe_apply(tu->inode, tu->offset, &tu->consumer, false);
1297 if (ret)
1298 break;
1299 }
1300
1301 return ret;
1302}
1303
1304static int uprobe_perf_open(struct trace_event_call *call,
1305 struct perf_event *event)
1306{
1307 struct trace_probe *pos, *tp;
1308 struct trace_uprobe *tu;
1309 int err = 0;
1310
1311 tp = trace_probe_primary_from_call(call);
1312 if (WARN_ON_ONCE(!tp))
1313 return -ENODEV;
1314
1315 tu = container_of(tp, struct trace_uprobe, tp);
1316 if (trace_uprobe_filter_add(tu->tp.event->filter, event))
1317 return 0;
1318
1319 list_for_each_entry(pos, trace_probe_probe_list(tp), list) {
1320 err = uprobe_apply(tu->inode, tu->offset, &tu->consumer, true);
1321 if (err) {
1322 uprobe_perf_close(call, event);
1323 break;
1324 }
1325 }
1326
1327 return err;
1328}
1329
1330static bool uprobe_perf_filter(struct uprobe_consumer *uc,
1331 enum uprobe_filter_ctx ctx, struct mm_struct *mm)
1332{
1333 struct trace_uprobe_filter *filter;
1334 struct trace_uprobe *tu;
1335 int ret;
1336
1337 tu = container_of(uc, struct trace_uprobe, consumer);
1338 filter = tu->tp.event->filter;
1339
1340 read_lock(&filter->rwlock);
1341 ret = __uprobe_perf_filter(filter, mm);
1342 read_unlock(&filter->rwlock);
1343
1344 return ret;
1345}
1346
1347static void __uprobe_perf_func(struct trace_uprobe *tu,
1348 unsigned long func, struct pt_regs *regs,
1349 struct uprobe_cpu_buffer *ucb, int dsize)
1350{
1351 struct trace_event_call *call = trace_probe_event_call(&tu->tp);
1352 struct uprobe_trace_entry_head *entry;
1353 struct hlist_head *head;
1354 void *data;
1355 int size, esize;
1356 int rctx;
1357
1358 if (bpf_prog_array_valid(call)) {
1359 u32 ret;
1360
1361 preempt_disable();
1362 ret = trace_call_bpf(call, regs);
1363 preempt_enable();
1364 if (!ret)
1365 return;
1366 }
1367
1368 esize = SIZEOF_TRACE_ENTRY(is_ret_probe(tu));
1369
1370 size = esize + tu->tp.size + dsize;
1371 size = ALIGN(size + sizeof(u32), sizeof(u64)) - sizeof(u32);
1372 if (WARN_ONCE(size > PERF_MAX_TRACE_SIZE, "profile buffer not large enough"))
1373 return;
1374
1375 preempt_disable();
1376 head = this_cpu_ptr(call->perf_events);
1377 if (hlist_empty(head))
1378 goto out;
1379
1380 entry = perf_trace_buf_alloc(size, NULL, &rctx);
1381 if (!entry)
1382 goto out;
1383
1384 if (is_ret_probe(tu)) {
1385 entry->vaddr[0] = func;
1386 entry->vaddr[1] = instruction_pointer(regs);
1387 data = DATAOF_TRACE_ENTRY(entry, true);
1388 } else {
1389 entry->vaddr[0] = instruction_pointer(regs);
1390 data = DATAOF_TRACE_ENTRY(entry, false);
1391 }
1392
1393 memcpy(data, ucb->buf, tu->tp.size + dsize);
1394
1395 if (size - esize > tu->tp.size + dsize) {
1396 int len = tu->tp.size + dsize;
1397
1398 memset(data + len, 0, size - esize - len);
1399 }
1400
1401 perf_trace_buf_submit(entry, size, rctx, call->event.type, 1, regs,
1402 head, NULL);
1403 out:
1404 preempt_enable();
1405}
1406
1407/* uprobe profile handler */
1408static int uprobe_perf_func(struct trace_uprobe *tu, struct pt_regs *regs,
1409 struct uprobe_cpu_buffer *ucb, int dsize)
1410{
1411 if (!uprobe_perf_filter(&tu->consumer, 0, current->mm))
1412 return UPROBE_HANDLER_REMOVE;
1413
1414 if (!is_ret_probe(tu))
1415 __uprobe_perf_func(tu, 0, regs, ucb, dsize);
1416 return 0;
1417}
1418
1419static void uretprobe_perf_func(struct trace_uprobe *tu, unsigned long func,
1420 struct pt_regs *regs,
1421 struct uprobe_cpu_buffer *ucb, int dsize)
1422{
1423 __uprobe_perf_func(tu, func, regs, ucb, dsize);
1424}
1425
1426int bpf_get_uprobe_info(const struct perf_event *event, u32 *fd_type,
1427 const char **filename, u64 *probe_offset,
1428 bool perf_type_tracepoint)
1429{
1430 const char *pevent = trace_event_name(event->tp_event);
1431 const char *group = event->tp_event->class->system;
1432 struct trace_uprobe *tu;
1433
1434 if (perf_type_tracepoint)
1435 tu = find_probe_event(pevent, group);
1436 else
1437 tu = trace_uprobe_primary_from_call(event->tp_event);
1438 if (!tu)
1439 return -EINVAL;
1440
1441 *fd_type = is_ret_probe(tu) ? BPF_FD_TYPE_URETPROBE
1442 : BPF_FD_TYPE_UPROBE;
1443 *filename = tu->filename;
1444 *probe_offset = tu->offset;
1445 return 0;
1446}
1447#endif /* CONFIG_PERF_EVENTS */
1448
1449static int
1450trace_uprobe_register(struct trace_event_call *event, enum trace_reg type,
1451 void *data)
1452{
1453 struct trace_event_file *file = data;
1454
1455 switch (type) {
1456 case TRACE_REG_REGISTER:
1457 return probe_event_enable(event, file, NULL);
1458
1459 case TRACE_REG_UNREGISTER:
1460 probe_event_disable(event, file);
1461 return 0;
1462
1463#ifdef CONFIG_PERF_EVENTS
1464 case TRACE_REG_PERF_REGISTER:
1465 return probe_event_enable(event, NULL, uprobe_perf_filter);
1466
1467 case TRACE_REG_PERF_UNREGISTER:
1468 probe_event_disable(event, NULL);
1469 return 0;
1470
1471 case TRACE_REG_PERF_OPEN:
1472 return uprobe_perf_open(event, data);
1473
1474 case TRACE_REG_PERF_CLOSE:
1475 return uprobe_perf_close(event, data);
1476
1477#endif
1478 default:
1479 return 0;
1480 }
1481}
1482
1483static int uprobe_dispatcher(struct uprobe_consumer *con, struct pt_regs *regs)
1484{
1485 struct trace_uprobe *tu;
1486 struct uprobe_dispatch_data udd;
1487 struct uprobe_cpu_buffer *ucb;
1488 int dsize, esize;
1489 int ret = 0;
1490
1491
1492 tu = container_of(con, struct trace_uprobe, consumer);
1493 tu->nhit++;
1494
1495 udd.tu = tu;
1496 udd.bp_addr = instruction_pointer(regs);
1497
1498 current->utask->vaddr = (unsigned long) &udd;
1499
1500 if (WARN_ON_ONCE(!uprobe_cpu_buffer))
1501 return 0;
1502
1503 dsize = __get_data_size(&tu->tp, regs);
1504 esize = SIZEOF_TRACE_ENTRY(is_ret_probe(tu));
1505
1506 ucb = uprobe_buffer_get();
1507 store_trace_args(ucb->buf, &tu->tp, regs, esize, dsize);
1508
1509 if (trace_probe_test_flag(&tu->tp, TP_FLAG_TRACE))
1510 ret |= uprobe_trace_func(tu, regs, ucb, dsize);
1511
1512#ifdef CONFIG_PERF_EVENTS
1513 if (trace_probe_test_flag(&tu->tp, TP_FLAG_PROFILE))
1514 ret |= uprobe_perf_func(tu, regs, ucb, dsize);
1515#endif
1516 uprobe_buffer_put(ucb);
1517 return ret;
1518}
1519
1520static int uretprobe_dispatcher(struct uprobe_consumer *con,
1521 unsigned long func, struct pt_regs *regs)
1522{
1523 struct trace_uprobe *tu;
1524 struct uprobe_dispatch_data udd;
1525 struct uprobe_cpu_buffer *ucb;
1526 int dsize, esize;
1527
1528 tu = container_of(con, struct trace_uprobe, consumer);
1529
1530 udd.tu = tu;
1531 udd.bp_addr = func;
1532
1533 current->utask->vaddr = (unsigned long) &udd;
1534
1535 if (WARN_ON_ONCE(!uprobe_cpu_buffer))
1536 return 0;
1537
1538 dsize = __get_data_size(&tu->tp, regs);
1539 esize = SIZEOF_TRACE_ENTRY(is_ret_probe(tu));
1540
1541 ucb = uprobe_buffer_get();
1542 store_trace_args(ucb->buf, &tu->tp, regs, esize, dsize);
1543
1544 if (trace_probe_test_flag(&tu->tp, TP_FLAG_TRACE))
1545 uretprobe_trace_func(tu, func, regs, ucb, dsize);
1546
1547#ifdef CONFIG_PERF_EVENTS
1548 if (trace_probe_test_flag(&tu->tp, TP_FLAG_PROFILE))
1549 uretprobe_perf_func(tu, func, regs, ucb, dsize);
1550#endif
1551 uprobe_buffer_put(ucb);
1552 return 0;
1553}
1554
1555static struct trace_event_functions uprobe_funcs = {
1556 .trace = print_uprobe_event
1557};
1558
1559static struct trace_event_fields uprobe_fields_array[] = {
1560 { .type = TRACE_FUNCTION_TYPE,
1561 .define_fields = uprobe_event_define_fields },
1562 {}
1563};
1564
1565static inline void init_trace_event_call(struct trace_uprobe *tu)
1566{
1567 struct trace_event_call *call = trace_probe_event_call(&tu->tp);
1568 call->event.funcs = &uprobe_funcs;
1569 call->class->fields_array = uprobe_fields_array;
1570
1571 call->flags = TRACE_EVENT_FL_UPROBE | TRACE_EVENT_FL_CAP_ANY;
1572 call->class->reg = trace_uprobe_register;
1573}
1574
1575static int register_uprobe_event(struct trace_uprobe *tu)
1576{
1577 init_trace_event_call(tu);
1578
1579 return trace_probe_register_event_call(&tu->tp);
1580}
1581
1582static int unregister_uprobe_event(struct trace_uprobe *tu)
1583{
1584 return trace_probe_unregister_event_call(&tu->tp);
1585}
1586
1587#ifdef CONFIG_PERF_EVENTS
1588struct trace_event_call *
1589create_local_trace_uprobe(char *name, unsigned long offs,
1590 unsigned long ref_ctr_offset, bool is_return)
1591{
1592 struct trace_uprobe *tu;
1593 struct path path;
1594 int ret;
1595
1596 ret = kern_path(name, LOOKUP_FOLLOW, &path);
1597 if (ret)
1598 return ERR_PTR(ret);
1599
1600 if (!d_is_reg(path.dentry)) {
1601 path_put(&path);
1602 return ERR_PTR(-EINVAL);
1603 }
1604
1605 /*
1606 * local trace_kprobes are not added to dyn_event, so they are never
1607 * searched in find_trace_kprobe(). Therefore, there is no concern of
1608 * duplicated name "DUMMY_EVENT" here.
1609 */
1610 tu = alloc_trace_uprobe(UPROBE_EVENT_SYSTEM, "DUMMY_EVENT", 0,
1611 is_return);
1612
1613 if (IS_ERR(tu)) {
1614 pr_info("Failed to allocate trace_uprobe.(%d)\n",
1615 (int)PTR_ERR(tu));
1616 path_put(&path);
1617 return ERR_CAST(tu);
1618 }
1619
1620 tu->offset = offs;
1621 tu->path = path;
1622 tu->ref_ctr_offset = ref_ctr_offset;
1623 tu->filename = kstrdup(name, GFP_KERNEL);
1624 init_trace_event_call(tu);
1625
1626 if (traceprobe_set_print_fmt(&tu->tp, is_ret_probe(tu)) < 0) {
1627 ret = -ENOMEM;
1628 goto error;
1629 }
1630
1631 return trace_probe_event_call(&tu->tp);
1632error:
1633 free_trace_uprobe(tu);
1634 return ERR_PTR(ret);
1635}
1636
1637void destroy_local_trace_uprobe(struct trace_event_call *event_call)
1638{
1639 struct trace_uprobe *tu;
1640
1641 tu = trace_uprobe_primary_from_call(event_call);
1642
1643 free_trace_uprobe(tu);
1644}
1645#endif /* CONFIG_PERF_EVENTS */
1646
1647/* Make a trace interface for controlling probe points */
1648static __init int init_uprobe_trace(void)
1649{
1650 int ret;
1651
1652 ret = dyn_event_register(&trace_uprobe_ops);
1653 if (ret)
1654 return ret;
1655
1656 ret = tracing_init_dentry();
1657 if (ret)
1658 return 0;
1659
1660 trace_create_file("uprobe_events", 0644, NULL,
1661 NULL, &uprobe_events_ops);
1662 /* Profile interface */
1663 trace_create_file("uprobe_profile", 0444, NULL,
1664 NULL, &uprobe_profile_ops);
1665 return 0;
1666}
1667
1668fs_initcall(init_uprobe_trace);