Merge tag 'for-linus' of git://git.kernel.org/pub/scm/virt/kvm/kvm
[linux-block.git] / kernel / trace / trace_uprobe.c
CommitLineData
bcea3f96 1// SPDX-License-Identifier: GPL-2.0
f3f096cf
SD
2/*
3 * uprobes-based tracing events
4 *
f3f096cf
SD
5 * Copyright (C) IBM Corporation, 2010-2012
6 * Author: Srikar Dronamraju <srikar@linux.vnet.ibm.com>
7 */
ea6eb5e7 8#define pr_fmt(fmt) "trace_uprobe: " fmt
f3f096cf 9
aef2feda 10#include <linux/bpf-cgroup.h>
17911ff3 11#include <linux/security.h>
0597c49c 12#include <linux/ctype.h>
f3f096cf
SD
13#include <linux/module.h>
14#include <linux/uaccess.h>
15#include <linux/uprobes.h>
16#include <linux/namei.h>
b2e902f0 17#include <linux/string.h>
b2d09103 18#include <linux/rculist.h>
8c7dcb84 19#include <linux/filter.h>
10cdb82a 20#include <linux/percpu.h>
f3f096cf 21
0597c49c 22#include "trace_dynevent.h"
f3f096cf 23#include "trace_probe.h"
53305928 24#include "trace_probe_tmpl.h"
f3f096cf
SD
25
26#define UPROBE_EVENT_SYSTEM "uprobes"
27
457d1772
ON
28struct uprobe_trace_entry_head {
29 struct trace_entry ent;
30 unsigned long vaddr[];
31};
32
33#define SIZEOF_TRACE_ENTRY(is_return) \
34 (sizeof(struct uprobe_trace_entry_head) + \
35 sizeof(unsigned long) * (is_return ? 2 : 1))
36
37#define DATAOF_TRACE_ENTRY(entry, is_return) \
38 ((void*)(entry) + SIZEOF_TRACE_ENTRY(is_return))
39
d262271d 40static int trace_uprobe_create(const char *raw_command);
0597c49c
MH
41static int trace_uprobe_show(struct seq_file *m, struct dyn_event *ev);
42static int trace_uprobe_release(struct dyn_event *ev);
43static bool trace_uprobe_is_busy(struct dyn_event *ev);
44static bool trace_uprobe_match(const char *system, const char *event,
30199137 45 int argc, const char **argv, struct dyn_event *ev);
0597c49c
MH
46
47static struct dyn_event_operations trace_uprobe_ops = {
48 .create = trace_uprobe_create,
49 .show = trace_uprobe_show,
50 .is_busy = trace_uprobe_is_busy,
51 .free = trace_uprobe_release,
52 .match = trace_uprobe_match,
53};
54
f3f096cf
SD
55/*
56 * uprobe event core functions
57 */
f3f096cf 58struct trace_uprobe {
0597c49c 59 struct dyn_event devent;
a932b738 60 struct uprobe_consumer consumer;
0c92c7a3 61 struct path path;
f3f096cf 62 char *filename;
3c83a9ad 63 struct uprobe *uprobe;
f3f096cf 64 unsigned long offset;
1cc33161 65 unsigned long ref_ctr_offset;
10cdb82a 66 unsigned long __percpu *nhits;
14577c39 67 struct trace_probe tp;
f3f096cf
SD
68};
69
0597c49c
MH
70static bool is_trace_uprobe(struct dyn_event *ev)
71{
72 return ev->ops == &trace_uprobe_ops;
73}
74
75static struct trace_uprobe *to_trace_uprobe(struct dyn_event *ev)
76{
77 return container_of(ev, struct trace_uprobe, devent);
78}
79
80/**
81 * for_each_trace_uprobe - iterate over the trace_uprobe list
82 * @pos: the struct trace_uprobe * for each entry
83 * @dpos: the struct dyn_event * to use as a loop cursor
84 */
85#define for_each_trace_uprobe(pos, dpos) \
86 for_each_dyn_event(dpos) \
87 if (is_trace_uprobe(dpos) && (pos = to_trace_uprobe(dpos)))
88
f3f096cf 89static int register_uprobe_event(struct trace_uprobe *tu);
c6c2401d 90static int unregister_uprobe_event(struct trace_uprobe *tu);
f3f096cf 91
f3f096cf 92static int uprobe_dispatcher(struct uprobe_consumer *con, struct pt_regs *regs);
c1ae5c75
ON
93static int uretprobe_dispatcher(struct uprobe_consumer *con,
94 unsigned long func, struct pt_regs *regs);
f3f096cf 95
3fd996a2
NK
96#ifdef CONFIG_STACK_GROWSUP
97static unsigned long adjust_stack_addr(unsigned long addr, unsigned int n)
98{
99 return addr - (n * sizeof(long));
100}
101#else
102static unsigned long adjust_stack_addr(unsigned long addr, unsigned int n)
103{
104 return addr + (n * sizeof(long));
105}
106#endif
107
108static unsigned long get_user_stack_nth(struct pt_regs *regs, unsigned int n)
109{
110 unsigned long ret;
111 unsigned long addr = user_stack_pointer(regs);
112
113 addr = adjust_stack_addr(addr, n);
114
115 if (copy_from_user(&ret, (void __force __user *) addr, sizeof(ret)))
116 return 0;
117
118 return ret;
119}
120
121/*
122 * Uprobes-specific fetch functions
123 */
53305928 124static nokprobe_inline int
9b960a38 125probe_mem_read(void *dest, void *src, size_t size)
53305928
MH
126{
127 void __user *vaddr = (void __force __user *)src;
128
f3f58935 129 return copy_from_user(dest, vaddr, size) ? -EFAULT : 0;
5baaa59e 130}
e65f7ae7
MH
131
132static nokprobe_inline int
133probe_mem_read_user(void *dest, void *src, size_t size)
134{
135 return probe_mem_read(dest, src, size);
136}
137
5baaa59e
NK
138/*
139 * Fetch a null-terminated string. Caller MUST set *(u32 *)dest with max
140 * length and relative data location.
141 */
9178412d
MH
142static nokprobe_inline int
143fetch_store_string(unsigned long addr, void *dest, void *base)
5baaa59e
NK
144{
145 long ret;
9178412d
MH
146 u32 loc = *(u32 *)dest;
147 int maxlen = get_loc_len(loc);
148 u8 *dst = get_loc_data(dest, base);
5baaa59e
NK
149 void __user *src = (void __force __user *) addr;
150
9178412d
MH
151 if (unlikely(!maxlen))
152 return -ENOMEM;
5baaa59e 153
4dd537ac 154 if (addr == FETCH_TOKEN_COMM)
8a3750ec 155 ret = strscpy(dst, current->comm, maxlen);
4dd537ac
MH
156 else
157 ret = strncpy_from_user(dst, src, maxlen);
9178412d
MH
158 if (ret >= 0) {
159 if (ret == maxlen)
160 dst[ret - 1] = '\0';
0722069a
AZ
161 else
162 /*
163 * Include the terminating null byte. In this case it
164 * was copied by strncpy_from_user but not accounted
165 * for in ret.
166 */
167 ret++;
9178412d 168 *(u32 *)dest = make_data_loc(ret, (void *)dst - base);
797311bc
MHG
169 } else
170 *(u32 *)dest = make_data_loc(0, (void *)dst - base);
9178412d
MH
171
172 return ret;
5baaa59e
NK
173}
174
88903c46
MH
175static nokprobe_inline int
176fetch_store_string_user(unsigned long addr, void *dest, void *base)
177{
178 return fetch_store_string(addr, dest, base);
179}
180
53305928 181/* Return the length of string -- including null terminal byte */
9178412d
MH
182static nokprobe_inline int
183fetch_store_strlen(unsigned long addr)
5baaa59e
NK
184{
185 int len;
186 void __user *vaddr = (void __force __user *) addr;
187
4dd537ac
MH
188 if (addr == FETCH_TOKEN_COMM)
189 len = strlen(current->comm) + 1;
190 else
191 len = strnlen_user(vaddr, MAX_STRING_SIZE);
5baaa59e 192
9178412d 193 return (len > MAX_STRING_SIZE) ? 0 : len;
5baaa59e 194}
3fd996a2 195
88903c46
MH
196static nokprobe_inline int
197fetch_store_strlen_user(unsigned long addr)
198{
199 return fetch_store_strlen(addr);
200}
201
53305928 202static unsigned long translate_user_vaddr(unsigned long file_offset)
b7e0bf34
NK
203{
204 unsigned long base_addr;
205 struct uprobe_dispatch_data *udd;
206
207 udd = (void *) current->utask->vaddr;
208
209 base_addr = udd->bp_addr - udd->tu->offset;
53305928 210 return base_addr + file_offset;
b7e0bf34 211}
b7e0bf34 212
53305928
MH
213/* Note that we don't verify it, since the code does not come from user space */
214static int
25f00e40
MHG
215process_fetch_insn(struct fetch_insn *code, void *rec, void *edata,
216 void *dest, void *base)
53305928 217{
8565a45d 218 struct pt_regs *regs = rec;
53305928 219 unsigned long val;
bd78acc8 220 int ret;
53305928
MH
221
222 /* 1st stage: get value from context */
223 switch (code->op) {
224 case FETCH_OP_REG:
225 val = regs_get_register(regs, code->param);
226 break;
227 case FETCH_OP_STACK:
228 val = get_user_stack_nth(regs, code->param);
229 break;
230 case FETCH_OP_STACKP:
231 val = user_stack_pointer(regs);
232 break;
233 case FETCH_OP_RETVAL:
234 val = regs_return_value(regs);
235 break;
4dd537ac
MH
236 case FETCH_OP_COMM:
237 val = FETCH_TOKEN_COMM;
238 break;
53305928
MH
239 case FETCH_OP_FOFFS:
240 val = translate_user_vaddr(code->immediate);
241 break;
242 default:
bd78acc8
SC
243 ret = process_common_fetch_insn(code, &val);
244 if (ret < 0)
245 return ret;
53305928
MH
246 }
247 code++;
248
9b960a38 249 return process_fetch_insn_bottom(code, val, dest, base);
53305928
MH
250}
251NOKPROBE_SYMBOL(process_fetch_insn)
252
736288ba
ON
253static inline void init_trace_uprobe_filter(struct trace_uprobe_filter *filter)
254{
255 rwlock_init(&filter->rwlock);
256 filter->nr_systemwide = 0;
257 INIT_LIST_HEAD(&filter->perf_events);
258}
259
260static inline bool uprobe_filter_is_empty(struct trace_uprobe_filter *filter)
261{
262 return !filter->nr_systemwide && list_empty(&filter->perf_events);
263}
264
c1ae5c75
ON
265static inline bool is_ret_probe(struct trace_uprobe *tu)
266{
267 return tu->consumer.ret_handler != NULL;
268}
269
0597c49c
MH
270static bool trace_uprobe_is_busy(struct dyn_event *ev)
271{
272 struct trace_uprobe *tu = to_trace_uprobe(ev);
273
274 return trace_probe_is_enabled(&tu->tp);
275}
276
ab10d69e
MH
277static bool trace_uprobe_match_command_head(struct trace_uprobe *tu,
278 int argc, const char **argv)
279{
280 char buf[MAX_ARGSTR_LEN + 1];
281 int len;
282
283 if (!argc)
284 return true;
285
286 len = strlen(tu->filename);
287 if (strncmp(tu->filename, argv[0], len) || argv[0][len] != ':')
288 return false;
289
290 if (tu->ref_ctr_offset == 0)
291 snprintf(buf, sizeof(buf), "0x%0*lx",
292 (int)(sizeof(void *) * 2), tu->offset);
293 else
294 snprintf(buf, sizeof(buf), "0x%0*lx(0x%lx)",
295 (int)(sizeof(void *) * 2), tu->offset,
296 tu->ref_ctr_offset);
297 if (strcmp(buf, &argv[0][len + 1]))
298 return false;
299
300 argc--; argv++;
301
302 return trace_probe_match_command_args(&tu->tp, argc, argv);
303}
304
0597c49c 305static bool trace_uprobe_match(const char *system, const char *event,
30199137 306 int argc, const char **argv, struct dyn_event *ev)
0597c49c
MH
307{
308 struct trace_uprobe *tu = to_trace_uprobe(ev);
309
95c104c3
LY
310 return (event[0] == '\0' ||
311 strcmp(trace_probe_name(&tu->tp), event) == 0) &&
ab10d69e
MH
312 (!system || strcmp(trace_probe_group_name(&tu->tp), system) == 0) &&
313 trace_uprobe_match_command_head(tu, argc, argv);
0597c49c
MH
314}
315
60d53e2c
MH
316static nokprobe_inline struct trace_uprobe *
317trace_uprobe_primary_from_call(struct trace_event_call *call)
318{
319 struct trace_probe *tp;
320
321 tp = trace_probe_primary_from_call(call);
322 if (WARN_ON_ONCE(!tp))
323 return NULL;
324
325 return container_of(tp, struct trace_uprobe, tp);
326}
327
f3f096cf
SD
328/*
329 * Allocate new trace_uprobe and initialize it (including uprobes).
330 */
331static struct trace_uprobe *
c1ae5c75 332alloc_trace_uprobe(const char *group, const char *event, int nargs, bool is_ret)
f3f096cf
SD
333{
334 struct trace_uprobe *tu;
455b2899 335 int ret;
f3f096cf 336
845cbf3e 337 tu = kzalloc(struct_size(tu, tp.args, nargs), GFP_KERNEL);
f3f096cf
SD
338 if (!tu)
339 return ERR_PTR(-ENOMEM);
340
10cdb82a
AN
341 tu->nhits = alloc_percpu(unsigned long);
342 if (!tu->nhits) {
343 ret = -ENOMEM;
344 goto error;
345 }
346
035ba760 347 ret = trace_probe_init(&tu->tp, event, group, true, nargs);
455b2899 348 if (ret < 0)
f3f096cf
SD
349 goto error;
350
0597c49c 351 dyn_event_init(&tu->devent, &trace_uprobe_ops);
a932b738 352 tu->consumer.handler = uprobe_dispatcher;
c1ae5c75
ON
353 if (is_ret)
354 tu->consumer.ret_handler = uretprobe_dispatcher;
b61387cb 355 init_trace_uprobe_filter(tu->tp.event->filter);
f3f096cf
SD
356 return tu;
357
358error:
10cdb82a 359 free_percpu(tu->nhits);
f3f096cf
SD
360 kfree(tu);
361
455b2899 362 return ERR_PTR(ret);
f3f096cf
SD
363}
364
365static void free_trace_uprobe(struct trace_uprobe *tu)
366{
0597c49c
MH
367 if (!tu)
368 return;
369
0c92c7a3 370 path_put(&tu->path);
455b2899 371 trace_probe_cleanup(&tu->tp);
f3f096cf 372 kfree(tu->filename);
10cdb82a 373 free_percpu(tu->nhits);
f3f096cf
SD
374 kfree(tu);
375}
376
377static struct trace_uprobe *find_probe_event(const char *event, const char *group)
378{
0597c49c 379 struct dyn_event *pos;
f3f096cf
SD
380 struct trace_uprobe *tu;
381
0597c49c 382 for_each_trace_uprobe(tu, pos)
b55ce203
MH
383 if (strcmp(trace_probe_name(&tu->tp), event) == 0 &&
384 strcmp(trace_probe_group_name(&tu->tp), group) == 0)
f3f096cf
SD
385 return tu;
386
387 return NULL;
388}
389
0597c49c 390/* Unregister a trace_uprobe and probe_event */
c6c2401d 391static int unregister_trace_uprobe(struct trace_uprobe *tu)
f3f096cf 392{
c6c2401d
SRRH
393 int ret;
394
41af3cf5
MH
395 if (trace_probe_has_sibling(&tu->tp))
396 goto unreg;
397
1d18538e
SRV
398 /* If there's a reference to the dynamic event */
399 if (trace_event_dyn_busy(trace_probe_event_call(&tu->tp)))
400 return -EBUSY;
401
c6c2401d
SRRH
402 ret = unregister_uprobe_event(tu);
403 if (ret)
404 return ret;
405
41af3cf5 406unreg:
0597c49c 407 dyn_event_remove(&tu->devent);
41af3cf5 408 trace_probe_unlink(&tu->tp);
f3f096cf 409 free_trace_uprobe(tu);
c6c2401d 410 return 0;
f3f096cf
SD
411}
412
fe60b0ce
MH
413static bool trace_uprobe_has_same_uprobe(struct trace_uprobe *orig,
414 struct trace_uprobe *comp)
415{
416 struct trace_probe_event *tpe = orig->tp.event;
fe60b0ce
MH
417 struct inode *comp_inode = d_real_inode(comp->path.dentry);
418 int i;
419
e161c6bf 420 list_for_each_entry(orig, &tpe->probes, tp.list) {
fe60b0ce
MH
421 if (comp_inode != d_real_inode(orig->path.dentry) ||
422 comp->offset != orig->offset)
423 continue;
424
425 /*
426 * trace_probe_compare_arg_type() ensured that nr_args and
427 * each argument name and type are same. Let's compare comm.
428 */
429 for (i = 0; i < orig->tp.nr_args; i++) {
430 if (strcmp(orig->tp.args[i].comm,
431 comp->tp.args[i].comm))
f8d7ab2b 432 break;
fe60b0ce
MH
433 }
434
f8d7ab2b
SD
435 if (i == orig->tp.nr_args)
436 return true;
fe60b0ce
MH
437 }
438
439 return false;
440}
441
41af3cf5
MH
442static int append_trace_uprobe(struct trace_uprobe *tu, struct trace_uprobe *to)
443{
444 int ret;
445
fe60b0ce
MH
446 ret = trace_probe_compare_arg_type(&tu->tp, &to->tp);
447 if (ret) {
448 /* Note that argument starts index = 2 */
449 trace_probe_log_set_index(ret + 1);
450 trace_probe_log_err(0, DIFF_ARG_TYPE);
451 return -EEXIST;
452 }
453 if (trace_uprobe_has_same_uprobe(to, tu)) {
454 trace_probe_log_set_index(0);
455 trace_probe_log_err(0, SAME_PROBE);
456 return -EEXIST;
457 }
458
41af3cf5
MH
459 /* Append to existing event */
460 ret = trace_probe_append(&tu->tp, &to->tp);
461 if (!ret)
8b0e6c74 462 dyn_event_add(&tu->devent, trace_probe_event_call(&tu->tp));
41af3cf5
MH
463
464 return ret;
465}
466
ccea8727
RB
467/*
468 * Uprobe with multiple reference counter is not allowed. i.e.
469 * If inode and offset matches, reference counter offset *must*
470 * match as well. Though, there is one exception: If user is
471 * replacing old trace_uprobe with new one(same group/event),
472 * then we allow same uprobe with new reference counter as far
473 * as the new one does not conflict with any other existing
474 * ones.
475 */
41af3cf5 476static int validate_ref_ctr_offset(struct trace_uprobe *new)
ccea8727 477{
0597c49c 478 struct dyn_event *pos;
41af3cf5 479 struct trace_uprobe *tmp;
ccea8727
RB
480 struct inode *new_inode = d_real_inode(new->path.dentry);
481
0597c49c 482 for_each_trace_uprobe(tmp, pos) {
41af3cf5 483 if (new_inode == d_real_inode(tmp->path.dentry) &&
ccea8727
RB
484 new->offset == tmp->offset &&
485 new->ref_ctr_offset != tmp->ref_ctr_offset) {
486 pr_warn("Reference counter offset mismatch.");
41af3cf5 487 return -EINVAL;
ccea8727
RB
488 }
489 }
41af3cf5 490 return 0;
ccea8727
RB
491}
492
f3f096cf
SD
493/* Register a trace_uprobe and probe_event */
494static int register_trace_uprobe(struct trace_uprobe *tu)
495{
14577c39 496 struct trace_uprobe *old_tu;
f3f096cf
SD
497 int ret;
498
0597c49c 499 mutex_lock(&event_mutex);
f3f096cf 500
41af3cf5
MH
501 ret = validate_ref_ctr_offset(tu);
502 if (ret)
ccea8727 503 goto end;
ccea8727 504
41af3cf5
MH
505 /* register as an event */
506 old_tu = find_probe_event(trace_probe_name(&tu->tp),
507 trace_probe_group_name(&tu->tp));
14577c39 508 if (old_tu) {
41af3cf5
MH
509 if (is_ret_probe(tu) != is_ret_probe(old_tu)) {
510 trace_probe_log_set_index(0);
511 trace_probe_log_err(0, DIFF_PROBE_TYPE);
512 ret = -EEXIST;
513 } else {
fe60b0ce 514 ret = append_trace_uprobe(tu, old_tu);
41af3cf5
MH
515 }
516 goto end;
c6c2401d 517 }
f3f096cf
SD
518
519 ret = register_uprobe_event(tu);
520 if (ret) {
8e242060
MH
521 if (ret == -EEXIST) {
522 trace_probe_log_set_index(0);
523 trace_probe_log_err(0, EVENT_EXIST);
524 } else
525 pr_warn("Failed to register probe event(%d)\n", ret);
f3f096cf
SD
526 goto end;
527 }
528
8b0e6c74 529 dyn_event_add(&tu->devent, trace_probe_event_call(&tu->tp));
f3f096cf
SD
530
531end:
0597c49c 532 mutex_unlock(&event_mutex);
f3f096cf
SD
533
534 return ret;
535}
536
537/*
538 * Argument syntax:
95c104c3 539 * - Add uprobe: p|r[:[GRP/][EVENT]] PATH:OFFSET[%return][(REF)] [FETCHARGS]
f3f096cf 540 */
d262271d 541static int __trace_uprobe_create(int argc, const char **argv)
f3f096cf
SD
542{
543 struct trace_uprobe *tu;
0597c49c
MH
544 const char *event = NULL, *group = UPROBE_EVENT_SYSTEM;
545 char *arg, *filename, *rctr, *rctr_end, *tmp;
f3f096cf 546 char buf[MAX_EVENT_NAME_LEN];
95c104c3 547 char gbuf[MAX_EVENT_NAME_LEN];
007517a0 548 enum probe_print_type ptype;
f3f096cf 549 struct path path;
1cc33161 550 unsigned long offset, ref_ctr_offset;
0597c49c 551 bool is_return = false;
f3f096cf
SD
552 int i, ret;
553
1cc33161 554 ref_ctr_offset = 0;
f3f096cf 555
f01098c7
ET
556 switch (argv[0][0]) {
557 case 'r':
4ee5a52e 558 is_return = true;
f01098c7
ET
559 break;
560 case 'p':
561 break;
562 default:
563 return -ECANCELED;
564 }
565
566 if (argc < 2)
0597c49c 567 return -ECANCELED;
f3f096cf 568
0597c49c 569 if (argv[0][1] == ':')
f3f096cf 570 event = &argv[0][2];
f3f096cf 571
0597c49c
MH
572 if (!strchr(argv[1], '/'))
573 return -ECANCELED;
f3f096cf 574
0597c49c
MH
575 filename = kstrdup(argv[1], GFP_KERNEL);
576 if (!filename)
577 return -ENOMEM;
f3f096cf 578
6496bb72 579 /* Find the last occurrence, in case the path contains ':' too. */
0597c49c
MH
580 arg = strrchr(filename, ':');
581 if (!arg || !isdigit(arg[1])) {
582 kfree(filename);
583 return -ECANCELED;
584 }
f3f096cf 585
ab105a4f
MH
586 trace_probe_log_init("trace_uprobe", argc, argv);
587 trace_probe_log_set_index(1); /* filename is the 2nd argument */
588
f3f096cf 589 *arg++ = '\0';
f3f096cf 590 ret = kern_path(filename, LOOKUP_FOLLOW, &path);
0597c49c 591 if (ret) {
ab105a4f 592 trace_probe_log_err(0, FILE_NOT_FOUND);
0597c49c 593 kfree(filename);
ab105a4f 594 trace_probe_log_clear();
0c92c7a3 595 return ret;
0597c49c 596 }
0c92c7a3 597 if (!d_is_reg(path.dentry)) {
ab105a4f 598 trace_probe_log_err(0, NO_REGULAR_FILE);
d24d7dbf
JZ
599 ret = -EINVAL;
600 goto fail_address_parse;
601 }
f3f096cf 602
1cc33161
RB
603 /* Parse reference counter offset if specified. */
604 rctr = strchr(arg, '(');
605 if (rctr) {
606 rctr_end = strchr(rctr, ')');
ab105a4f
MH
607 if (!rctr_end) {
608 ret = -EINVAL;
609 rctr_end = rctr + strlen(rctr);
610 trace_probe_log_err(rctr_end - filename,
611 REFCNT_OPEN_BRACE);
612 goto fail_address_parse;
613 } else if (rctr_end[1] != '\0') {
1cc33161 614 ret = -EINVAL;
ab105a4f
MH
615 trace_probe_log_err(rctr_end + 1 - filename,
616 BAD_REFCNT_SUFFIX);
1cc33161
RB
617 goto fail_address_parse;
618 }
619
620 *rctr++ = '\0';
621 *rctr_end = '\0';
622 ret = kstrtoul(rctr, 0, &ref_ctr_offset);
623 if (ret) {
ab105a4f 624 trace_probe_log_err(rctr - filename, BAD_REFCNT);
1cc33161
RB
625 goto fail_address_parse;
626 }
627 }
628
3dd3aae3
MH
629 /* Check if there is %return suffix */
630 tmp = strchr(arg, '%');
631 if (tmp) {
632 if (!strcmp(tmp, "%return")) {
633 *tmp = '\0';
634 is_return = true;
635 } else {
636 trace_probe_log_err(tmp - filename, BAD_ADDR_SUFFIX);
637 ret = -EINVAL;
638 goto fail_address_parse;
639 }
640 }
641
1cc33161 642 /* Parse uprobe offset. */
84d7ed79 643 ret = kstrtoul(arg, 0, &offset);
ab105a4f
MH
644 if (ret) {
645 trace_probe_log_err(arg - filename, BAD_UPROBE_OFFS);
84d7ed79 646 goto fail_address_parse;
ab105a4f 647 }
f3f096cf
SD
648
649 /* setup a probe */
ab105a4f 650 trace_probe_log_set_index(0);
0597c49c 651 if (event) {
95c104c3 652 ret = traceprobe_parse_event_name(&event, &group, gbuf,
ab105a4f 653 event - argv[0]);
0597c49c
MH
654 if (ret)
655 goto fail_address_parse;
95c104c3
LY
656 }
657
658 if (!event) {
b2e902f0 659 char *tail;
f3f096cf
SD
660 char *ptr;
661
b2e902f0
AS
662 tail = kstrdup(kbasename(filename), GFP_KERNEL);
663 if (!tail) {
f3f096cf
SD
664 ret = -ENOMEM;
665 goto fail_address_parse;
666 }
667
f3f096cf
SD
668 ptr = strpbrk(tail, ".-_");
669 if (ptr)
670 *ptr = '\0';
671
672 snprintf(buf, MAX_EVENT_NAME_LEN, "%c_%s_0x%lx", 'p', tail, offset);
673 event = buf;
674 kfree(tail);
675 }
676
ab105a4f
MH
677 argc -= 2;
678 argv += 2;
679
4ee5a52e 680 tu = alloc_trace_uprobe(group, event, argc, is_return);
f3f096cf 681 if (IS_ERR(tu)) {
f3f096cf 682 ret = PTR_ERR(tu);
a039480e
MH
683 /* This must return -ENOMEM otherwise there is a bug */
684 WARN_ON_ONCE(ret != -ENOMEM);
f3f096cf
SD
685 goto fail_address_parse;
686 }
687 tu->offset = offset;
1cc33161 688 tu->ref_ctr_offset = ref_ctr_offset;
0c92c7a3 689 tu->path = path;
0597c49c 690 tu->filename = filename;
f3f096cf
SD
691
692 /* parse arguments */
f3f096cf 693 for (i = 0; i < argc && i < MAX_TRACE_ARGS; i++) {
1b8b0cd7
MHG
694 struct traceprobe_parse_context ctx = {
695 .flags = (is_return ? TPARG_FL_RETURN : 0) | TPARG_FL_USER,
696 };
697
ab105a4f 698 trace_probe_log_set_index(i + 2);
1b8b0cd7 699 ret = traceprobe_parse_probe_arg(&tu->tp, i, argv[i], &ctx);
b1d1e904 700 traceprobe_finish_parse(&ctx);
d00bbea9 701 if (ret)
f3f096cf 702 goto error;
f3f096cf
SD
703 }
704
007517a0
SRV
705 ptype = is_ret_probe(tu) ? PROBE_PRINT_RETURN : PROBE_PRINT_NORMAL;
706 ret = traceprobe_set_print_fmt(&tu->tp, ptype);
b4d4b96b
MH
707 if (ret < 0)
708 goto error;
709
f3f096cf 710 ret = register_trace_uprobe(tu);
ab105a4f
MH
711 if (!ret)
712 goto out;
f3f096cf
SD
713
714error:
715 free_trace_uprobe(tu);
ab105a4f
MH
716out:
717 trace_probe_log_clear();
f3f096cf
SD
718 return ret;
719
720fail_address_parse:
ab105a4f 721 trace_probe_log_clear();
0c92c7a3 722 path_put(&path);
0597c49c 723 kfree(filename);
f3f096cf 724
f3f096cf
SD
725 return ret;
726}
727
d262271d
MH
728int trace_uprobe_create(const char *raw_command)
729{
730 return trace_probe_create(raw_command, __trace_uprobe_create);
731}
732
733static int create_or_delete_trace_uprobe(const char *raw_command)
f3f096cf 734{
0597c49c 735 int ret;
f3f096cf 736
d262271d
MH
737 if (raw_command[0] == '-')
738 return dyn_event_release(raw_command, &trace_uprobe_ops);
f3f096cf 739
d262271d 740 ret = trace_uprobe_create(raw_command);
0597c49c 741 return ret == -ECANCELED ? -EINVAL : ret;
f3f096cf
SD
742}
743
0597c49c 744static int trace_uprobe_release(struct dyn_event *ev)
f3f096cf 745{
0597c49c 746 struct trace_uprobe *tu = to_trace_uprobe(ev);
f3f096cf 747
0597c49c 748 return unregister_trace_uprobe(tu);
f3f096cf
SD
749}
750
0597c49c
MH
751/* Probes listing interfaces */
752static int trace_uprobe_show(struct seq_file *m, struct dyn_event *ev)
f3f096cf 753{
0597c49c 754 struct trace_uprobe *tu = to_trace_uprobe(ev);
3ede82dd 755 char c = is_ret_probe(tu) ? 'r' : 'p';
f3f096cf
SD
756 int i;
757
b55ce203
MH
758 seq_printf(m, "%c:%s/%s %s:0x%0*lx", c, trace_probe_group_name(&tu->tp),
759 trace_probe_name(&tu->tp), tu->filename,
a64b2c01 760 (int)(sizeof(void *) * 2), tu->offset);
f3f096cf 761
1cc33161
RB
762 if (tu->ref_ctr_offset)
763 seq_printf(m, "(0x%lx)", tu->ref_ctr_offset);
764
14577c39
NK
765 for (i = 0; i < tu->tp.nr_args; i++)
766 seq_printf(m, " %s=%s", tu->tp.args[i].name, tu->tp.args[i].comm);
f3f096cf 767
fa6f0cc7 768 seq_putc(m, '\n');
f3f096cf
SD
769 return 0;
770}
771
0597c49c
MH
772static int probes_seq_show(struct seq_file *m, void *v)
773{
774 struct dyn_event *ev = v;
775
776 if (!is_trace_uprobe(ev))
777 return 0;
778
779 return trace_uprobe_show(m, ev);
780}
781
f3f096cf 782static const struct seq_operations probes_seq_op = {
0597c49c
MH
783 .start = dyn_event_seq_start,
784 .next = dyn_event_seq_next,
785 .stop = dyn_event_seq_stop,
786 .show = probes_seq_show
f3f096cf
SD
787};
788
789static int probes_open(struct inode *inode, struct file *file)
790{
c6c2401d
SRRH
791 int ret;
792
17911ff3
SRV
793 ret = security_locked_down(LOCKDOWN_TRACEFS);
794 if (ret)
795 return ret;
796
c6c2401d 797 if ((file->f_mode & FMODE_WRITE) && (file->f_flags & O_TRUNC)) {
0597c49c 798 ret = dyn_events_release_all(&trace_uprobe_ops);
c6c2401d
SRRH
799 if (ret)
800 return ret;
801 }
f3f096cf
SD
802
803 return seq_open(file, &probes_seq_op);
804}
805
806static ssize_t probes_write(struct file *file, const char __user *buffer,
807 size_t count, loff_t *ppos)
808{
0597c49c
MH
809 return trace_parse_run_command(file, buffer, count, ppos,
810 create_or_delete_trace_uprobe);
f3f096cf
SD
811}
812
813static const struct file_operations uprobe_events_ops = {
814 .owner = THIS_MODULE,
815 .open = probes_open,
816 .read = seq_read,
817 .llseek = seq_lseek,
818 .release = seq_release,
819 .write = probes_write,
820};
821
822/* Probes profiling interfaces */
823static int probes_profile_seq_show(struct seq_file *m, void *v)
824{
0597c49c
MH
825 struct dyn_event *ev = v;
826 struct trace_uprobe *tu;
10cdb82a
AN
827 unsigned long nhits;
828 int cpu;
0597c49c
MH
829
830 if (!is_trace_uprobe(ev))
831 return 0;
f3f096cf 832
0597c49c 833 tu = to_trace_uprobe(ev);
10cdb82a
AN
834
835 nhits = 0;
836 for_each_possible_cpu(cpu) {
837 nhits += per_cpu(*tu->nhits, cpu);
838 }
839
de7b2973 840 seq_printf(m, " %s %-44s %15lu\n", tu->filename,
10cdb82a 841 trace_probe_name(&tu->tp), nhits);
f3f096cf
SD
842 return 0;
843}
844
845static const struct seq_operations profile_seq_op = {
0597c49c
MH
846 .start = dyn_event_seq_start,
847 .next = dyn_event_seq_next,
848 .stop = dyn_event_seq_stop,
f3f096cf
SD
849 .show = probes_profile_seq_show
850};
851
852static int profile_open(struct inode *inode, struct file *file)
853{
17911ff3
SRV
854 int ret;
855
856 ret = security_locked_down(LOCKDOWN_TRACEFS);
857 if (ret)
858 return ret;
859
f3f096cf
SD
860 return seq_open(file, &profile_seq_op);
861}
862
863static const struct file_operations uprobe_profile_ops = {
864 .owner = THIS_MODULE,
865 .open = profile_open,
866 .read = seq_read,
867 .llseek = seq_lseek,
868 .release = seq_release,
869};
870
dcad1a20
NK
871struct uprobe_cpu_buffer {
872 struct mutex mutex;
873 void *buf;
3eaea21b 874 int dsize;
dcad1a20
NK
875};
876static struct uprobe_cpu_buffer __percpu *uprobe_cpu_buffer;
877static int uprobe_buffer_refcnt;
878
879static int uprobe_buffer_init(void)
880{
881 int cpu, err_cpu;
882
883 uprobe_cpu_buffer = alloc_percpu(struct uprobe_cpu_buffer);
884 if (uprobe_cpu_buffer == NULL)
885 return -ENOMEM;
886
887 for_each_possible_cpu(cpu) {
888 struct page *p = alloc_pages_node(cpu_to_node(cpu),
889 GFP_KERNEL, 0);
890 if (p == NULL) {
891 err_cpu = cpu;
892 goto err;
893 }
894 per_cpu_ptr(uprobe_cpu_buffer, cpu)->buf = page_address(p);
895 mutex_init(&per_cpu_ptr(uprobe_cpu_buffer, cpu)->mutex);
896 }
897
898 return 0;
899
900err:
901 for_each_possible_cpu(cpu) {
902 if (cpu == err_cpu)
903 break;
904 free_page((unsigned long)per_cpu_ptr(uprobe_cpu_buffer, cpu)->buf);
905 }
906
907 free_percpu(uprobe_cpu_buffer);
908 return -ENOMEM;
909}
910
911static int uprobe_buffer_enable(void)
912{
913 int ret = 0;
914
915 BUG_ON(!mutex_is_locked(&event_mutex));
916
917 if (uprobe_buffer_refcnt++ == 0) {
918 ret = uprobe_buffer_init();
919 if (ret < 0)
920 uprobe_buffer_refcnt--;
921 }
922
923 return ret;
924}
925
926static void uprobe_buffer_disable(void)
927{
6ea6215f
J
928 int cpu;
929
dcad1a20
NK
930 BUG_ON(!mutex_is_locked(&event_mutex));
931
932 if (--uprobe_buffer_refcnt == 0) {
6ea6215f
J
933 for_each_possible_cpu(cpu)
934 free_page((unsigned long)per_cpu_ptr(uprobe_cpu_buffer,
935 cpu)->buf);
936
dcad1a20
NK
937 free_percpu(uprobe_cpu_buffer);
938 uprobe_cpu_buffer = NULL;
939 }
940}
941
942static struct uprobe_cpu_buffer *uprobe_buffer_get(void)
943{
944 struct uprobe_cpu_buffer *ucb;
945 int cpu;
946
947 cpu = raw_smp_processor_id();
948 ucb = per_cpu_ptr(uprobe_cpu_buffer, cpu);
949
950 /*
951 * Use per-cpu buffers for fastest access, but we might migrate
952 * so the mutex makes sure we have sole access to it.
953 */
954 mutex_lock(&ucb->mutex);
955
956 return ucb;
957}
958
959static void uprobe_buffer_put(struct uprobe_cpu_buffer *ucb)
960{
1b8f85de
AN
961 if (!ucb)
962 return;
dcad1a20
NK
963 mutex_unlock(&ucb->mutex);
964}
965
3eaea21b 966static struct uprobe_cpu_buffer *prepare_uprobe_buffer(struct trace_uprobe *tu,
1b8f85de
AN
967 struct pt_regs *regs,
968 struct uprobe_cpu_buffer **ucbp)
3eaea21b
AN
969{
970 struct uprobe_cpu_buffer *ucb;
971 int dsize, esize;
972
1b8f85de
AN
973 if (*ucbp)
974 return *ucbp;
975
3eaea21b
AN
976 esize = SIZEOF_TRACE_ENTRY(is_ret_probe(tu));
977 dsize = __get_data_size(&tu->tp, regs, NULL);
978
979 ucb = uprobe_buffer_get();
980 ucb->dsize = tu->tp.size + dsize;
981
982 store_trace_args(ucb->buf, &tu->tp, regs, NULL, esize, dsize);
983
1b8f85de 984 *ucbp = ucb;
3eaea21b
AN
985 return ucb;
986}
987
a43b9704 988static void __uprobe_trace_func(struct trace_uprobe *tu,
dd9fa555 989 unsigned long func, struct pt_regs *regs,
69964673 990 struct uprobe_cpu_buffer *ucb,
7f1d2f82 991 struct trace_event_file *trace_file)
f3f096cf
SD
992{
993 struct uprobe_trace_entry_head *entry;
b7d5eb26 994 struct trace_event_buffer fbuffer;
457d1772 995 void *data;
dd9fa555 996 int size, esize;
e3dc9f89 997 struct trace_event_call *call = trace_probe_event_call(&tu->tp);
f3f096cf 998
7f1d2f82 999 WARN_ON(call != trace_file->event_call);
70ed91c6 1000
3eaea21b 1001 if (WARN_ON_ONCE(ucb->dsize > PAGE_SIZE))
dcad1a20
NK
1002 return;
1003
09a5059a 1004 if (trace_trigger_soft_disabled(trace_file))
ca3b1620
NK
1005 return;
1006
dd9fa555 1007 esize = SIZEOF_TRACE_ENTRY(is_ret_probe(tu));
3eaea21b 1008 size = esize + ucb->dsize;
b7d5eb26
SRV
1009 entry = trace_event_buffer_reserve(&fbuffer, trace_file, size);
1010 if (!entry)
dd9fa555 1011 return;
f3f096cf 1012
393a736c
ON
1013 if (is_ret_probe(tu)) {
1014 entry->vaddr[0] = func;
1015 entry->vaddr[1] = instruction_pointer(regs);
1016 data = DATAOF_TRACE_ENTRY(entry, true);
1017 } else {
1018 entry->vaddr[0] = instruction_pointer(regs);
1019 data = DATAOF_TRACE_ENTRY(entry, false);
1020 }
1021
3eaea21b 1022 memcpy(data, ucb->buf, ucb->dsize);
f3f096cf 1023
b7d5eb26 1024 trace_event_buffer_commit(&fbuffer);
a51cc604 1025}
f42d24a1 1026
a51cc604 1027/* uprobe handler */
dd9fa555 1028static int uprobe_trace_func(struct trace_uprobe *tu, struct pt_regs *regs,
1b8f85de 1029 struct uprobe_cpu_buffer **ucbp)
a51cc604 1030{
70ed91c6 1031 struct event_file_link *link;
69964673 1032 struct uprobe_cpu_buffer *ucb;
70ed91c6
J
1033
1034 if (is_ret_probe(tu))
1035 return 0;
1036
69964673
AN
1037 ucb = prepare_uprobe_buffer(tu, regs, ucbp);
1038
70ed91c6 1039 rcu_read_lock();
b5f935ee 1040 trace_probe_for_each_link_rcu(link, &tu->tp)
69964673 1041 __uprobe_trace_func(tu, 0, regs, ucb, link->file);
70ed91c6
J
1042 rcu_read_unlock();
1043
f42d24a1 1044 return 0;
f3f096cf
SD
1045}
1046
c1ae5c75 1047static void uretprobe_trace_func(struct trace_uprobe *tu, unsigned long func,
dd9fa555 1048 struct pt_regs *regs,
1b8f85de 1049 struct uprobe_cpu_buffer **ucbp)
c1ae5c75 1050{
70ed91c6 1051 struct event_file_link *link;
69964673
AN
1052 struct uprobe_cpu_buffer *ucb;
1053
1054 ucb = prepare_uprobe_buffer(tu, regs, ucbp);
70ed91c6
J
1055
1056 rcu_read_lock();
b5f935ee 1057 trace_probe_for_each_link_rcu(link, &tu->tp)
69964673 1058 __uprobe_trace_func(tu, func, regs, ucb, link->file);
70ed91c6 1059 rcu_read_unlock();
c1ae5c75
ON
1060}
1061
f3f096cf
SD
1062/* Event entry printers */
1063static enum print_line_t
1064print_uprobe_event(struct trace_iterator *iter, int flags, struct trace_event *event)
1065{
457d1772 1066 struct uprobe_trace_entry_head *entry;
f3f096cf
SD
1067 struct trace_seq *s = &iter->seq;
1068 struct trace_uprobe *tu;
1069 u8 *data;
f3f096cf 1070
457d1772 1071 entry = (struct uprobe_trace_entry_head *)iter->ent;
60d53e2c
MH
1072 tu = trace_uprobe_primary_from_call(
1073 container_of(event, struct trace_event_call, event));
1074 if (unlikely(!tu))
1075 goto out;
f3f096cf 1076
3ede82dd 1077 if (is_ret_probe(tu)) {
8579a107 1078 trace_seq_printf(s, "%s: (0x%lx <- 0x%lx)",
b55ce203 1079 trace_probe_name(&tu->tp),
8579a107 1080 entry->vaddr[1], entry->vaddr[0]);
3ede82dd
ON
1081 data = DATAOF_TRACE_ENTRY(entry, true);
1082 } else {
8579a107 1083 trace_seq_printf(s, "%s: (0x%lx)",
b55ce203 1084 trace_probe_name(&tu->tp),
8579a107 1085 entry->vaddr[0]);
3ede82dd
ON
1086 data = DATAOF_TRACE_ENTRY(entry, false);
1087 }
f3f096cf 1088
196b6389 1089 if (trace_probe_print_args(s, tu->tp.args, tu->tp.nr_args, data, entry) < 0)
56de7630 1090 goto out;
f3f096cf 1091
8579a107 1092 trace_seq_putc(s, '\n');
f3f096cf 1093
8579a107
SRRH
1094 out:
1095 return trace_handle_return(s);
f3f096cf
SD
1096}
1097
59da880a 1098typedef bool (*filter_func_t)(struct uprobe_consumer *self, struct mm_struct *mm);
31ba3348 1099
60d53e2c 1100static int trace_uprobe_enable(struct trace_uprobe *tu, filter_func_t filter)
f3f096cf 1101{
3c83a9ad
ON
1102 struct inode *inode = d_real_inode(tu->path.dentry);
1103 struct uprobe *uprobe;
70ed91c6 1104
60d53e2c 1105 tu->consumer.filter = filter;
3c83a9ad
ON
1106 uprobe = uprobe_register(inode, tu->offset, tu->ref_ctr_offset, &tu->consumer);
1107 if (IS_ERR(uprobe))
1108 return PTR_ERR(uprobe);
60d53e2c 1109
3c83a9ad
ON
1110 tu->uprobe = uprobe;
1111 return 0;
60d53e2c
MH
1112}
1113
1114static void __probe_event_disable(struct trace_probe *tp)
1115{
60d53e2c 1116 struct trace_uprobe *tu;
04b01625 1117 bool sync = false;
60d53e2c 1118
99c9a923 1119 tu = container_of(tp, struct trace_uprobe, tp);
b61387cb 1120 WARN_ON(!uprobe_filter_is_empty(tu->tp.event->filter));
99c9a923 1121
e161c6bf 1122 list_for_each_entry(tu, trace_probe_probe_list(tp), tp.list) {
3c83a9ad 1123 if (!tu->uprobe)
60d53e2c
MH
1124 continue;
1125
04b01625
PZ
1126 uprobe_unregister_nosync(tu->uprobe, &tu->consumer);
1127 sync = true;
3c83a9ad 1128 tu->uprobe = NULL;
60d53e2c 1129 }
04b01625
PZ
1130 if (sync)
1131 uprobe_unregister_sync();
60d53e2c
MH
1132}
1133
1134static int probe_event_enable(struct trace_event_call *call,
1135 struct trace_event_file *file, filter_func_t filter)
1136{
e161c6bf 1137 struct trace_probe *tp;
60d53e2c
MH
1138 struct trace_uprobe *tu;
1139 bool enabled;
1140 int ret;
1141
1142 tp = trace_probe_primary_from_call(call);
1143 if (WARN_ON_ONCE(!tp))
1144 return -ENODEV;
1145 enabled = trace_probe_is_enabled(tp);
1146
1147 /* This may also change "enabled" state */
70ed91c6 1148 if (file) {
60d53e2c 1149 if (trace_probe_test_flag(tp, TP_FLAG_PROFILE))
48212542
ON
1150 return -EINTR;
1151
60d53e2c 1152 ret = trace_probe_add_file(tp, file);
b5f935ee
MH
1153 if (ret < 0)
1154 return ret;
48212542 1155 } else {
60d53e2c 1156 if (trace_probe_test_flag(tp, TP_FLAG_TRACE))
48212542
ON
1157 return -EINTR;
1158
60d53e2c 1159 trace_probe_set_flag(tp, TP_FLAG_PROFILE);
48212542 1160 }
f3f096cf 1161
60d53e2c 1162 tu = container_of(tp, struct trace_uprobe, tp);
b61387cb 1163 WARN_ON(!uprobe_filter_is_empty(tu->tp.event->filter));
736288ba 1164
70ed91c6
J
1165 if (enabled)
1166 return 0;
1167
fb6bab6a
ON
1168 ret = uprobe_buffer_enable();
1169 if (ret)
1170 goto err_flags;
1171
e161c6bf 1172 list_for_each_entry(tu, trace_probe_probe_list(tp), tp.list) {
60d53e2c
MH
1173 ret = trace_uprobe_enable(tu, filter);
1174 if (ret) {
1175 __probe_event_disable(tp);
1176 goto err_buffer;
1177 }
1cc33161
RB
1178 }
1179
fb6bab6a
ON
1180 return 0;
1181
1182 err_buffer:
1183 uprobe_buffer_disable();
f3f096cf 1184
fb6bab6a 1185 err_flags:
b5f935ee 1186 if (file)
60d53e2c 1187 trace_probe_remove_file(tp, file);
b5f935ee 1188 else
60d53e2c 1189 trace_probe_clear_flag(tp, TP_FLAG_PROFILE);
b5f935ee 1190
4161824f 1191 return ret;
f3f096cf
SD
1192}
1193
60d53e2c
MH
1194static void probe_event_disable(struct trace_event_call *call,
1195 struct trace_event_file *file)
f3f096cf 1196{
60d53e2c
MH
1197 struct trace_probe *tp;
1198
1199 tp = trace_probe_primary_from_call(call);
1200 if (WARN_ON_ONCE(!tp))
1201 return;
1202
1203 if (!trace_probe_is_enabled(tp))
f3f096cf
SD
1204 return;
1205
70ed91c6 1206 if (file) {
60d53e2c 1207 if (trace_probe_remove_file(tp, file) < 0)
70ed91c6
J
1208 return;
1209
60d53e2c 1210 if (trace_probe_is_enabled(tp))
70ed91c6 1211 return;
b5f935ee 1212 } else
60d53e2c 1213 trace_probe_clear_flag(tp, TP_FLAG_PROFILE);
dcad1a20 1214
60d53e2c 1215 __probe_event_disable(tp);
dcad1a20 1216 uprobe_buffer_disable();
f3f096cf
SD
1217}
1218
2425bcb9 1219static int uprobe_event_define_fields(struct trace_event_call *event_call)
f3f096cf 1220{
eeb07b06 1221 int ret, size;
f3f096cf 1222 struct uprobe_trace_entry_head field;
60d53e2c
MH
1223 struct trace_uprobe *tu;
1224
1225 tu = trace_uprobe_primary_from_call(event_call);
1226 if (unlikely(!tu))
1227 return -ENODEV;
f3f096cf 1228
4d1298e2
ON
1229 if (is_ret_probe(tu)) {
1230 DEFINE_FIELD(unsigned long, vaddr[0], FIELD_STRING_FUNC, 0);
1231 DEFINE_FIELD(unsigned long, vaddr[1], FIELD_STRING_RETIP, 0);
1232 size = SIZEOF_TRACE_ENTRY(true);
1233 } else {
1234 DEFINE_FIELD(unsigned long, vaddr[0], FIELD_STRING_IP, 0);
1235 size = SIZEOF_TRACE_ENTRY(false);
1236 }
f3f096cf 1237
eeb07b06 1238 return traceprobe_define_arg_fields(event_call, size, &tu->tp);
f3f096cf
SD
1239}
1240
f3f096cf 1241#ifdef CONFIG_PERF_EVENTS
31ba3348
ON
1242static bool
1243__uprobe_perf_filter(struct trace_uprobe_filter *filter, struct mm_struct *mm)
1244{
1245 struct perf_event *event;
1246
31ba3348 1247 list_for_each_entry(event, &filter->perf_events, hw.tp_list) {
50f16a8b 1248 if (event->hw.target->mm == mm)
31ba3348
ON
1249 return true;
1250 }
1251
1252 return false;
1253}
1254
b2fe8ba6 1255static inline bool
99c9a923
MH
1256trace_uprobe_filter_event(struct trace_uprobe_filter *filter,
1257 struct perf_event *event)
b2fe8ba6 1258{
99c9a923 1259 return __uprobe_perf_filter(filter, event->hw.target->mm);
b2fe8ba6
ON
1260}
1261
99c9a923
MH
1262static bool trace_uprobe_filter_remove(struct trace_uprobe_filter *filter,
1263 struct perf_event *event)
736288ba 1264{
b2fe8ba6
ON
1265 bool done;
1266
99c9a923 1267 write_lock(&filter->rwlock);
50f16a8b 1268 if (event->hw.target) {
ce5f36a5 1269 list_del(&event->hw.tp_list);
99c9a923 1270 done = filter->nr_systemwide ||
50f16a8b 1271 (event->hw.target->flags & PF_EXITING) ||
99c9a923 1272 trace_uprobe_filter_event(filter, event);
b2fe8ba6 1273 } else {
99c9a923
MH
1274 filter->nr_systemwide--;
1275 done = filter->nr_systemwide;
b2fe8ba6 1276 }
99c9a923 1277 write_unlock(&filter->rwlock);
31ba3348 1278
99c9a923 1279 return done;
736288ba
ON
1280}
1281
99c9a923
MH
1282/* This returns true if the filter always covers target mm */
1283static bool trace_uprobe_filter_add(struct trace_uprobe_filter *filter,
1284 struct perf_event *event)
736288ba 1285{
b2fe8ba6
ON
1286 bool done;
1287
99c9a923 1288 write_lock(&filter->rwlock);
50f16a8b 1289 if (event->hw.target) {
ce5f36a5
ON
1290 /*
1291 * event->parent != NULL means copy_process(), we can avoid
1292 * uprobe_apply(). current->mm must be probed and we can rely
1293 * on dup_mmap() which preserves the already installed bp's.
1294 *
1295 * attr.enable_on_exec means that exec/mmap will install the
1296 * breakpoints we need.
1297 */
99c9a923 1298 done = filter->nr_systemwide ||
ce5f36a5 1299 event->parent || event->attr.enable_on_exec ||
99c9a923
MH
1300 trace_uprobe_filter_event(filter, event);
1301 list_add(&event->hw.tp_list, &filter->perf_events);
b2fe8ba6 1302 } else {
99c9a923
MH
1303 done = filter->nr_systemwide;
1304 filter->nr_systemwide++;
b2fe8ba6 1305 }
99c9a923 1306 write_unlock(&filter->rwlock);
736288ba 1307
99c9a923 1308 return done;
736288ba
ON
1309}
1310
99c9a923
MH
1311static int uprobe_perf_close(struct trace_event_call *call,
1312 struct perf_event *event)
60d53e2c 1313{
e161c6bf 1314 struct trace_probe *tp;
60d53e2c
MH
1315 struct trace_uprobe *tu;
1316 int ret = 0;
1317
1318 tp = trace_probe_primary_from_call(call);
1319 if (WARN_ON_ONCE(!tp))
1320 return -ENODEV;
1321
99c9a923 1322 tu = container_of(tp, struct trace_uprobe, tp);
b61387cb 1323 if (trace_uprobe_filter_remove(tu->tp.event->filter, event))
99c9a923
MH
1324 return 0;
1325
e161c6bf 1326 list_for_each_entry(tu, trace_probe_probe_list(tp), tp.list) {
3c83a9ad 1327 ret = uprobe_apply(tu->uprobe, &tu->consumer, false);
60d53e2c
MH
1328 if (ret)
1329 break;
1330 }
1331
1332 return ret;
1333}
99c9a923
MH
1334
1335static int uprobe_perf_open(struct trace_event_call *call,
1336 struct perf_event *event)
1337{
e161c6bf 1338 struct trace_probe *tp;
99c9a923
MH
1339 struct trace_uprobe *tu;
1340 int err = 0;
1341
1342 tp = trace_probe_primary_from_call(call);
1343 if (WARN_ON_ONCE(!tp))
1344 return -ENODEV;
1345
1346 tu = container_of(tp, struct trace_uprobe, tp);
b61387cb 1347 if (trace_uprobe_filter_add(tu->tp.event->filter, event))
99c9a923
MH
1348 return 0;
1349
e161c6bf 1350 list_for_each_entry(tu, trace_probe_probe_list(tp), tp.list) {
3c83a9ad 1351 err = uprobe_apply(tu->uprobe, &tu->consumer, true);
99c9a923
MH
1352 if (err) {
1353 uprobe_perf_close(call, event);
1354 break;
1355 }
1356 }
1357
1358 return err;
1359}
1360
59da880a 1361static bool uprobe_perf_filter(struct uprobe_consumer *uc, struct mm_struct *mm)
31ba3348 1362{
99c9a923 1363 struct trace_uprobe_filter *filter;
31ba3348
ON
1364 struct trace_uprobe *tu;
1365 int ret;
1366
1367 tu = container_of(uc, struct trace_uprobe, consumer);
b61387cb 1368 filter = tu->tp.event->filter;
99c9a923 1369
cdf355cc
AN
1370 /*
1371 * speculative short-circuiting check to avoid unnecessarily taking
1372 * filter->rwlock below, if the uprobe has system-wide consumer
1373 */
1374 if (READ_ONCE(filter->nr_systemwide))
1375 return true;
1376
99c9a923
MH
1377 read_lock(&filter->rwlock);
1378 ret = __uprobe_perf_filter(filter, mm);
1379 read_unlock(&filter->rwlock);
31ba3348
ON
1380
1381 return ret;
1382}
1383
a43b9704 1384static void __uprobe_perf_func(struct trace_uprobe *tu,
dd9fa555 1385 unsigned long func, struct pt_regs *regs,
1b8f85de 1386 struct uprobe_cpu_buffer **ucbp)
f3f096cf 1387{
e3dc9f89 1388 struct trace_event_call *call = trace_probe_event_call(&tu->tp);
f3f096cf 1389 struct uprobe_trace_entry_head *entry;
1b8f85de 1390 struct uprobe_cpu_buffer *ucb;
f3f096cf 1391 struct hlist_head *head;
457d1772 1392 void *data;
dd9fa555 1393 int size, esize;
dcad1a20
NK
1394 int rctx;
1395
aca80dd9 1396#ifdef CONFIG_BPF_EVENTS
70ed0706
AS
1397 if (bpf_prog_array_valid(call)) {
1398 u32 ret;
1399
a3c485a5 1400 ret = bpf_prog_run_array_uprobe(call->prog_array, regs, bpf_prog_run);
70ed0706
AS
1401 if (!ret)
1402 return;
1403 }
aca80dd9 1404#endif /* CONFIG_BPF_EVENTS */
04a22fae 1405
dcad1a20 1406 esize = SIZEOF_TRACE_ENTRY(is_ret_probe(tu));
f3f096cf 1407
1b8f85de 1408 ucb = prepare_uprobe_buffer(tu, regs, ucbp);
3eaea21b 1409 size = esize + ucb->dsize;
dcad1a20
NK
1410 size = ALIGN(size + sizeof(u32), sizeof(u64)) - sizeof(u32);
1411 if (WARN_ONCE(size > PERF_MAX_TRACE_SIZE, "profile buffer not large enough"))
1412 return;
1413
f3f096cf 1414 preempt_disable();
515619f2
ON
1415 head = this_cpu_ptr(call->perf_events);
1416 if (hlist_empty(head))
1417 goto out;
1418
1e1dcd93 1419 entry = perf_trace_buf_alloc(size, NULL, &rctx);
f3f096cf
SD
1420 if (!entry)
1421 goto out;
1422
393a736c
ON
1423 if (is_ret_probe(tu)) {
1424 entry->vaddr[0] = func;
32520b2c 1425 entry->vaddr[1] = instruction_pointer(regs);
393a736c
ON
1426 data = DATAOF_TRACE_ENTRY(entry, true);
1427 } else {
32520b2c 1428 entry->vaddr[0] = instruction_pointer(regs);
393a736c
ON
1429 data = DATAOF_TRACE_ENTRY(entry, false);
1430 }
1431
3eaea21b 1432 memcpy(data, ucb->buf, ucb->dsize);
dcad1a20 1433
3eaea21b
AN
1434 if (size - esize > ucb->dsize)
1435 memset(data + ucb->dsize, 0, size - esize - ucb->dsize);
f3f096cf 1436
1e1dcd93 1437 perf_trace_buf_submit(entry, size, rctx, call->event.type, 1, regs,
8fd0fbbe 1438 head, NULL);
f3f096cf
SD
1439 out:
1440 preempt_enable();
a51cc604
ON
1441}
1442
1443/* uprobe profile handler */
dd9fa555 1444static int uprobe_perf_func(struct trace_uprobe *tu, struct pt_regs *regs,
1b8f85de 1445 struct uprobe_cpu_buffer **ucbp)
a51cc604 1446{
59da880a 1447 if (!uprobe_perf_filter(&tu->consumer, current->mm))
a51cc604
ON
1448 return UPROBE_HANDLER_REMOVE;
1449
393a736c 1450 if (!is_ret_probe(tu))
1b8f85de 1451 __uprobe_perf_func(tu, 0, regs, ucbp);
f42d24a1 1452 return 0;
f3f096cf 1453}
c1ae5c75
ON
1454
1455static void uretprobe_perf_func(struct trace_uprobe *tu, unsigned long func,
dd9fa555 1456 struct pt_regs *regs,
1b8f85de 1457 struct uprobe_cpu_buffer **ucbp)
c1ae5c75 1458{
1b8f85de 1459 __uprobe_perf_func(tu, func, regs, ucbp);
c1ae5c75 1460}
41bdc4b4
YS
1461
1462int bpf_get_uprobe_info(const struct perf_event *event, u32 *fd_type,
1463 const char **filename, u64 *probe_offset,
5125e757 1464 u64 *probe_addr, bool perf_type_tracepoint)
41bdc4b4
YS
1465{
1466 const char *pevent = trace_event_name(event->tp_event);
1467 const char *group = event->tp_event->class->system;
1468 struct trace_uprobe *tu;
1469
1470 if (perf_type_tracepoint)
1471 tu = find_probe_event(pevent, group);
1472 else
22d5bd68 1473 tu = trace_uprobe_primary_from_call(event->tp_event);
41bdc4b4
YS
1474 if (!tu)
1475 return -EINVAL;
1476
1477 *fd_type = is_ret_probe(tu) ? BPF_FD_TYPE_URETPROBE
1478 : BPF_FD_TYPE_UPROBE;
1479 *filename = tu->filename;
1480 *probe_offset = tu->offset;
5125e757 1481 *probe_addr = 0;
41bdc4b4
YS
1482 return 0;
1483}
f3f096cf
SD
1484#endif /* CONFIG_PERF_EVENTS */
1485
70ed91c6 1486static int
2425bcb9 1487trace_uprobe_register(struct trace_event_call *event, enum trace_reg type,
70ed91c6 1488 void *data)
f3f096cf 1489{
7f1d2f82 1490 struct trace_event_file *file = data;
f3f096cf
SD
1491
1492 switch (type) {
1493 case TRACE_REG_REGISTER:
60d53e2c 1494 return probe_event_enable(event, file, NULL);
f3f096cf
SD
1495
1496 case TRACE_REG_UNREGISTER:
60d53e2c 1497 probe_event_disable(event, file);
f3f096cf
SD
1498 return 0;
1499
1500#ifdef CONFIG_PERF_EVENTS
1501 case TRACE_REG_PERF_REGISTER:
60d53e2c 1502 return probe_event_enable(event, NULL, uprobe_perf_filter);
f3f096cf
SD
1503
1504 case TRACE_REG_PERF_UNREGISTER:
60d53e2c 1505 probe_event_disable(event, NULL);
f3f096cf 1506 return 0;
736288ba
ON
1507
1508 case TRACE_REG_PERF_OPEN:
99c9a923 1509 return uprobe_perf_open(event, data);
736288ba
ON
1510
1511 case TRACE_REG_PERF_CLOSE:
99c9a923 1512 return uprobe_perf_close(event, data);
736288ba 1513
f3f096cf
SD
1514#endif
1515 default:
1516 return 0;
1517 }
f3f096cf
SD
1518}
1519
1520static int uprobe_dispatcher(struct uprobe_consumer *con, struct pt_regs *regs)
1521{
f3f096cf 1522 struct trace_uprobe *tu;
b7e0bf34 1523 struct uprobe_dispatch_data udd;
1b8f85de 1524 struct uprobe_cpu_buffer *ucb = NULL;
f42d24a1 1525 int ret = 0;
f3f096cf 1526
a932b738 1527 tu = container_of(con, struct trace_uprobe, consumer);
10cdb82a
AN
1528
1529 this_cpu_inc(*tu->nhits);
f3f096cf 1530
b7e0bf34
NK
1531 udd.tu = tu;
1532 udd.bp_addr = instruction_pointer(regs);
1533
1534 current->utask->vaddr = (unsigned long) &udd;
1535
dd9fa555
NK
1536 if (WARN_ON_ONCE(!uprobe_cpu_buffer))
1537 return 0;
1538
747774d6 1539 if (trace_probe_test_flag(&tu->tp, TP_FLAG_TRACE))
1b8f85de 1540 ret |= uprobe_trace_func(tu, regs, &ucb);
f3f096cf
SD
1541
1542#ifdef CONFIG_PERF_EVENTS
747774d6 1543 if (trace_probe_test_flag(&tu->tp, TP_FLAG_PROFILE))
1b8f85de 1544 ret |= uprobe_perf_func(tu, regs, &ucb);
f3f096cf 1545#endif
dd9fa555 1546 uprobe_buffer_put(ucb);
f42d24a1 1547 return ret;
f3f096cf
SD
1548}
1549
c1ae5c75
ON
1550static int uretprobe_dispatcher(struct uprobe_consumer *con,
1551 unsigned long func, struct pt_regs *regs)
1552{
1553 struct trace_uprobe *tu;
b7e0bf34 1554 struct uprobe_dispatch_data udd;
1b8f85de 1555 struct uprobe_cpu_buffer *ucb = NULL;
c1ae5c75
ON
1556
1557 tu = container_of(con, struct trace_uprobe, consumer);
1558
b7e0bf34
NK
1559 udd.tu = tu;
1560 udd.bp_addr = func;
1561
1562 current->utask->vaddr = (unsigned long) &udd;
1563
dd9fa555
NK
1564 if (WARN_ON_ONCE(!uprobe_cpu_buffer))
1565 return 0;
1566
747774d6 1567 if (trace_probe_test_flag(&tu->tp, TP_FLAG_TRACE))
1b8f85de 1568 uretprobe_trace_func(tu, func, regs, &ucb);
c1ae5c75
ON
1569
1570#ifdef CONFIG_PERF_EVENTS
747774d6 1571 if (trace_probe_test_flag(&tu->tp, TP_FLAG_PROFILE))
1b8f85de 1572 uretprobe_perf_func(tu, func, regs, &ucb);
c1ae5c75 1573#endif
dd9fa555 1574 uprobe_buffer_put(ucb);
c1ae5c75
ON
1575 return 0;
1576}
1577
f3f096cf
SD
1578static struct trace_event_functions uprobe_funcs = {
1579 .trace = print_uprobe_event
1580};
1581
04ae87a5
PZ
1582static struct trace_event_fields uprobe_fields_array[] = {
1583 { .type = TRACE_FUNCTION_TYPE,
1584 .define_fields = uprobe_event_define_fields },
1585 {}
1586};
1587
e3dc9f89 1588static inline void init_trace_event_call(struct trace_uprobe *tu)
f3f096cf 1589{
e3dc9f89 1590 struct trace_event_call *call = trace_probe_event_call(&tu->tp);
f3f096cf 1591 call->event.funcs = &uprobe_funcs;
04ae87a5 1592 call->class->fields_array = uprobe_fields_array;
f3f096cf 1593
9fd2e48b 1594 call->flags = TRACE_EVENT_FL_UPROBE | TRACE_EVENT_FL_CAP_ANY;
33ea4b24 1595 call->class->reg = trace_uprobe_register;
33ea4b24
SL
1596}
1597
1598static int register_uprobe_event(struct trace_uprobe *tu)
1599{
e3dc9f89 1600 init_trace_event_call(tu);
f3f096cf 1601
46e5376d 1602 return trace_probe_register_event_call(&tu->tp);
f3f096cf
SD
1603}
1604
c6c2401d 1605static int unregister_uprobe_event(struct trace_uprobe *tu)
f3f096cf 1606{
46e5376d 1607 return trace_probe_unregister_event_call(&tu->tp);
f3f096cf
SD
1608}
1609
33ea4b24
SL
1610#ifdef CONFIG_PERF_EVENTS
1611struct trace_event_call *
a6ca88b2
SL
1612create_local_trace_uprobe(char *name, unsigned long offs,
1613 unsigned long ref_ctr_offset, bool is_return)
33ea4b24 1614{
007517a0 1615 enum probe_print_type ptype;
33ea4b24 1616 struct trace_uprobe *tu;
33ea4b24
SL
1617 struct path path;
1618 int ret;
1619
1620 ret = kern_path(name, LOOKUP_FOLLOW, &path);
1621 if (ret)
1622 return ERR_PTR(ret);
1623
0c92c7a3
SL
1624 if (!d_is_reg(path.dentry)) {
1625 path_put(&path);
33ea4b24
SL
1626 return ERR_PTR(-EINVAL);
1627 }
1628
1629 /*
0597c49c 1630 * local trace_kprobes are not added to dyn_event, so they are never
33ea4b24
SL
1631 * searched in find_trace_kprobe(). Therefore, there is no concern of
1632 * duplicated name "DUMMY_EVENT" here.
1633 */
1634 tu = alloc_trace_uprobe(UPROBE_EVENT_SYSTEM, "DUMMY_EVENT", 0,
1635 is_return);
1636
1637 if (IS_ERR(tu)) {
1638 pr_info("Failed to allocate trace_uprobe.(%d)\n",
1639 (int)PTR_ERR(tu));
0c92c7a3 1640 path_put(&path);
33ea4b24
SL
1641 return ERR_CAST(tu);
1642 }
1643
1644 tu->offset = offs;
0c92c7a3 1645 tu->path = path;
a6ca88b2 1646 tu->ref_ctr_offset = ref_ctr_offset;
33ea4b24 1647 tu->filename = kstrdup(name, GFP_KERNEL);
8c722424
XW
1648 if (!tu->filename) {
1649 ret = -ENOMEM;
1650 goto error;
1651 }
1652
e3dc9f89 1653 init_trace_event_call(tu);
33ea4b24 1654
007517a0
SRV
1655 ptype = is_ret_probe(tu) ? PROBE_PRINT_RETURN : PROBE_PRINT_NORMAL;
1656 if (traceprobe_set_print_fmt(&tu->tp, ptype) < 0) {
33ea4b24
SL
1657 ret = -ENOMEM;
1658 goto error;
1659 }
1660
e3dc9f89 1661 return trace_probe_event_call(&tu->tp);
33ea4b24
SL
1662error:
1663 free_trace_uprobe(tu);
1664 return ERR_PTR(ret);
1665}
1666
1667void destroy_local_trace_uprobe(struct trace_event_call *event_call)
1668{
1669 struct trace_uprobe *tu;
1670
60d53e2c 1671 tu = trace_uprobe_primary_from_call(event_call);
33ea4b24 1672
33ea4b24
SL
1673 free_trace_uprobe(tu);
1674}
1675#endif /* CONFIG_PERF_EVENTS */
1676
39bcdd6a 1677/* Make a trace interface for controlling probe points */
f3f096cf
SD
1678static __init int init_uprobe_trace(void)
1679{
0597c49c
MH
1680 int ret;
1681
1682 ret = dyn_event_register(&trace_uprobe_ops);
1683 if (ret)
1684 return ret;
f3f096cf 1685
22c36b18
WY
1686 ret = tracing_init_dentry();
1687 if (ret)
f3f096cf
SD
1688 return 0;
1689
21ccc9cd 1690 trace_create_file("uprobe_events", TRACE_MODE_WRITE, NULL,
f3f096cf
SD
1691 NULL, &uprobe_events_ops);
1692 /* Profile interface */
21ccc9cd 1693 trace_create_file("uprobe_profile", TRACE_MODE_READ, NULL,
f3f096cf
SD
1694 NULL, &uprobe_profile_ops);
1695 return 0;
1696}
1697
1698fs_initcall(init_uprobe_trace);