Merge tag 'for-linus-20190706' of git://git.kernel.dk/linux-block
[linux/fpc-iii.git] / kernel / trace / trace_uprobe.c
blobb55906c77ce0602ad6d7f0c20225e99f288ddbb9
1 // SPDX-License-Identifier: GPL-2.0
2 /*
3 * uprobes-based tracing events
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
10 #include <linux/ctype.h>
11 #include <linux/module.h>
12 #include <linux/uaccess.h>
13 #include <linux/uprobes.h>
14 #include <linux/namei.h>
15 #include <linux/string.h>
16 #include <linux/rculist.h>
18 #include "trace_dynevent.h"
19 #include "trace_probe.h"
20 #include "trace_probe_tmpl.h"
22 #define UPROBE_EVENT_SYSTEM "uprobes"
24 struct uprobe_trace_entry_head {
25 struct trace_entry ent;
26 unsigned long vaddr[];
29 #define SIZEOF_TRACE_ENTRY(is_return) \
30 (sizeof(struct uprobe_trace_entry_head) + \
31 sizeof(unsigned long) * (is_return ? 2 : 1))
33 #define DATAOF_TRACE_ENTRY(entry, is_return) \
34 ((void*)(entry) + SIZEOF_TRACE_ENTRY(is_return))
36 struct trace_uprobe_filter {
37 rwlock_t rwlock;
38 int nr_systemwide;
39 struct list_head perf_events;
42 static int trace_uprobe_create(int argc, const char **argv);
43 static int trace_uprobe_show(struct seq_file *m, struct dyn_event *ev);
44 static int trace_uprobe_release(struct dyn_event *ev);
45 static bool trace_uprobe_is_busy(struct dyn_event *ev);
46 static bool trace_uprobe_match(const char *system, const char *event,
47 struct dyn_event *ev);
49 static struct dyn_event_operations trace_uprobe_ops = {
50 .create = trace_uprobe_create,
51 .show = trace_uprobe_show,
52 .is_busy = trace_uprobe_is_busy,
53 .free = trace_uprobe_release,
54 .match = trace_uprobe_match,
58 * uprobe event core functions
60 struct trace_uprobe {
61 struct dyn_event devent;
62 struct trace_uprobe_filter filter;
63 struct uprobe_consumer consumer;
64 struct path path;
65 struct inode *inode;
66 char *filename;
67 unsigned long offset;
68 unsigned long ref_ctr_offset;
69 unsigned long nhit;
70 struct trace_probe tp;
73 static bool is_trace_uprobe(struct dyn_event *ev)
75 return ev->ops == &trace_uprobe_ops;
78 static struct trace_uprobe *to_trace_uprobe(struct dyn_event *ev)
80 return container_of(ev, struct trace_uprobe, devent);
83 /**
84 * for_each_trace_uprobe - iterate over the trace_uprobe list
85 * @pos: the struct trace_uprobe * for each entry
86 * @dpos: the struct dyn_event * to use as a loop cursor
88 #define for_each_trace_uprobe(pos, dpos) \
89 for_each_dyn_event(dpos) \
90 if (is_trace_uprobe(dpos) && (pos = to_trace_uprobe(dpos)))
92 #define SIZEOF_TRACE_UPROBE(n) \
93 (offsetof(struct trace_uprobe, tp.args) + \
94 (sizeof(struct probe_arg) * (n)))
96 static int register_uprobe_event(struct trace_uprobe *tu);
97 static int unregister_uprobe_event(struct trace_uprobe *tu);
99 struct uprobe_dispatch_data {
100 struct trace_uprobe *tu;
101 unsigned long bp_addr;
104 static int uprobe_dispatcher(struct uprobe_consumer *con, struct pt_regs *regs);
105 static int uretprobe_dispatcher(struct uprobe_consumer *con,
106 unsigned long func, struct pt_regs *regs);
108 #ifdef CONFIG_STACK_GROWSUP
109 static unsigned long adjust_stack_addr(unsigned long addr, unsigned int n)
111 return addr - (n * sizeof(long));
113 #else
114 static unsigned long adjust_stack_addr(unsigned long addr, unsigned int n)
116 return addr + (n * sizeof(long));
118 #endif
120 static unsigned long get_user_stack_nth(struct pt_regs *regs, unsigned int n)
122 unsigned long ret;
123 unsigned long addr = user_stack_pointer(regs);
125 addr = adjust_stack_addr(addr, n);
127 if (copy_from_user(&ret, (void __force __user *) addr, sizeof(ret)))
128 return 0;
130 return ret;
134 * Uprobes-specific fetch functions
136 static nokprobe_inline int
137 probe_mem_read(void *dest, void *src, size_t size)
139 void __user *vaddr = (void __force __user *)src;
141 return copy_from_user(dest, vaddr, size) ? -EFAULT : 0;
144 * Fetch a null-terminated string. Caller MUST set *(u32 *)dest with max
145 * length and relative data location.
147 static nokprobe_inline int
148 fetch_store_string(unsigned long addr, void *dest, void *base)
150 long ret;
151 u32 loc = *(u32 *)dest;
152 int maxlen = get_loc_len(loc);
153 u8 *dst = get_loc_data(dest, base);
154 void __user *src = (void __force __user *) addr;
156 if (unlikely(!maxlen))
157 return -ENOMEM;
159 if (addr == FETCH_TOKEN_COMM)
160 ret = strlcpy(dst, current->comm, maxlen);
161 else
162 ret = strncpy_from_user(dst, src, maxlen);
163 if (ret >= 0) {
164 if (ret == maxlen)
165 dst[ret - 1] = '\0';
166 else
168 * Include the terminating null byte. In this case it
169 * was copied by strncpy_from_user but not accounted
170 * for in ret.
172 ret++;
173 *(u32 *)dest = make_data_loc(ret, (void *)dst - base);
176 return ret;
179 /* Return the length of string -- including null terminal byte */
180 static nokprobe_inline int
181 fetch_store_strlen(unsigned long addr)
183 int len;
184 void __user *vaddr = (void __force __user *) addr;
186 if (addr == FETCH_TOKEN_COMM)
187 len = strlen(current->comm) + 1;
188 else
189 len = strnlen_user(vaddr, MAX_STRING_SIZE);
191 return (len > MAX_STRING_SIZE) ? 0 : len;
194 static unsigned long translate_user_vaddr(unsigned long file_offset)
196 unsigned long base_addr;
197 struct uprobe_dispatch_data *udd;
199 udd = (void *) current->utask->vaddr;
201 base_addr = udd->bp_addr - udd->tu->offset;
202 return base_addr + file_offset;
205 /* Note that we don't verify it, since the code does not come from user space */
206 static int
207 process_fetch_insn(struct fetch_insn *code, struct pt_regs *regs, void *dest,
208 void *base)
210 unsigned long val;
212 /* 1st stage: get value from context */
213 switch (code->op) {
214 case FETCH_OP_REG:
215 val = regs_get_register(regs, code->param);
216 break;
217 case FETCH_OP_STACK:
218 val = get_user_stack_nth(regs, code->param);
219 break;
220 case FETCH_OP_STACKP:
221 val = user_stack_pointer(regs);
222 break;
223 case FETCH_OP_RETVAL:
224 val = regs_return_value(regs);
225 break;
226 case FETCH_OP_IMM:
227 val = code->immediate;
228 break;
229 case FETCH_OP_COMM:
230 val = FETCH_TOKEN_COMM;
231 break;
232 case FETCH_OP_FOFFS:
233 val = translate_user_vaddr(code->immediate);
234 break;
235 default:
236 return -EILSEQ;
238 code++;
240 return process_fetch_insn_bottom(code, val, dest, base);
242 NOKPROBE_SYMBOL(process_fetch_insn)
244 static inline void init_trace_uprobe_filter(struct trace_uprobe_filter *filter)
246 rwlock_init(&filter->rwlock);
247 filter->nr_systemwide = 0;
248 INIT_LIST_HEAD(&filter->perf_events);
251 static inline bool uprobe_filter_is_empty(struct trace_uprobe_filter *filter)
253 return !filter->nr_systemwide && list_empty(&filter->perf_events);
256 static inline bool is_ret_probe(struct trace_uprobe *tu)
258 return tu->consumer.ret_handler != NULL;
261 static bool trace_uprobe_is_busy(struct dyn_event *ev)
263 struct trace_uprobe *tu = to_trace_uprobe(ev);
265 return trace_probe_is_enabled(&tu->tp);
268 static bool trace_uprobe_match(const char *system, const char *event,
269 struct dyn_event *ev)
271 struct trace_uprobe *tu = to_trace_uprobe(ev);
273 return strcmp(trace_event_name(&tu->tp.call), event) == 0 &&
274 (!system || strcmp(tu->tp.call.class->system, system) == 0);
278 * Allocate new trace_uprobe and initialize it (including uprobes).
280 static struct trace_uprobe *
281 alloc_trace_uprobe(const char *group, const char *event, int nargs, bool is_ret)
283 struct trace_uprobe *tu;
285 if (!event || !group)
286 return ERR_PTR(-EINVAL);
288 tu = kzalloc(SIZEOF_TRACE_UPROBE(nargs), GFP_KERNEL);
289 if (!tu)
290 return ERR_PTR(-ENOMEM);
292 tu->tp.call.class = &tu->tp.class;
293 tu->tp.call.name = kstrdup(event, GFP_KERNEL);
294 if (!tu->tp.call.name)
295 goto error;
297 tu->tp.class.system = kstrdup(group, GFP_KERNEL);
298 if (!tu->tp.class.system)
299 goto error;
301 dyn_event_init(&tu->devent, &trace_uprobe_ops);
302 INIT_LIST_HEAD(&tu->tp.files);
303 tu->consumer.handler = uprobe_dispatcher;
304 if (is_ret)
305 tu->consumer.ret_handler = uretprobe_dispatcher;
306 init_trace_uprobe_filter(&tu->filter);
307 return tu;
309 error:
310 kfree(tu->tp.call.name);
311 kfree(tu);
313 return ERR_PTR(-ENOMEM);
316 static void free_trace_uprobe(struct trace_uprobe *tu)
318 int i;
320 if (!tu)
321 return;
323 for (i = 0; i < tu->tp.nr_args; i++)
324 traceprobe_free_probe_arg(&tu->tp.args[i]);
326 path_put(&tu->path);
327 kfree(tu->tp.call.class->system);
328 kfree(tu->tp.call.name);
329 kfree(tu->filename);
330 kfree(tu);
333 static struct trace_uprobe *find_probe_event(const char *event, const char *group)
335 struct dyn_event *pos;
336 struct trace_uprobe *tu;
338 for_each_trace_uprobe(tu, pos)
339 if (strcmp(trace_event_name(&tu->tp.call), event) == 0 &&
340 strcmp(tu->tp.call.class->system, group) == 0)
341 return tu;
343 return NULL;
346 /* Unregister a trace_uprobe and probe_event */
347 static int unregister_trace_uprobe(struct trace_uprobe *tu)
349 int ret;
351 ret = unregister_uprobe_event(tu);
352 if (ret)
353 return ret;
355 dyn_event_remove(&tu->devent);
356 free_trace_uprobe(tu);
357 return 0;
361 * Uprobe with multiple reference counter is not allowed. i.e.
362 * If inode and offset matches, reference counter offset *must*
363 * match as well. Though, there is one exception: If user is
364 * replacing old trace_uprobe with new one(same group/event),
365 * then we allow same uprobe with new reference counter as far
366 * as the new one does not conflict with any other existing
367 * ones.
369 static struct trace_uprobe *find_old_trace_uprobe(struct trace_uprobe *new)
371 struct dyn_event *pos;
372 struct trace_uprobe *tmp, *old = NULL;
373 struct inode *new_inode = d_real_inode(new->path.dentry);
375 old = find_probe_event(trace_event_name(&new->tp.call),
376 new->tp.call.class->system);
378 for_each_trace_uprobe(tmp, pos) {
379 if ((old ? old != tmp : true) &&
380 new_inode == d_real_inode(tmp->path.dentry) &&
381 new->offset == tmp->offset &&
382 new->ref_ctr_offset != tmp->ref_ctr_offset) {
383 pr_warn("Reference counter offset mismatch.");
384 return ERR_PTR(-EINVAL);
387 return old;
390 /* Register a trace_uprobe and probe_event */
391 static int register_trace_uprobe(struct trace_uprobe *tu)
393 struct trace_uprobe *old_tu;
394 int ret;
396 mutex_lock(&event_mutex);
398 /* register as an event */
399 old_tu = find_old_trace_uprobe(tu);
400 if (IS_ERR(old_tu)) {
401 ret = PTR_ERR(old_tu);
402 goto end;
405 if (old_tu) {
406 /* delete old event */
407 ret = unregister_trace_uprobe(old_tu);
408 if (ret)
409 goto end;
412 ret = register_uprobe_event(tu);
413 if (ret) {
414 pr_warn("Failed to register probe event(%d)\n", ret);
415 goto end;
418 dyn_event_add(&tu->devent);
420 end:
421 mutex_unlock(&event_mutex);
423 return ret;
427 * Argument syntax:
428 * - Add uprobe: p|r[:[GRP/]EVENT] PATH:OFFSET [FETCHARGS]
430 static int trace_uprobe_create(int argc, const char **argv)
432 struct trace_uprobe *tu;
433 const char *event = NULL, *group = UPROBE_EVENT_SYSTEM;
434 char *arg, *filename, *rctr, *rctr_end, *tmp;
435 char buf[MAX_EVENT_NAME_LEN];
436 struct path path;
437 unsigned long offset, ref_ctr_offset;
438 bool is_return = false;
439 int i, ret;
441 ret = 0;
442 ref_ctr_offset = 0;
444 switch (argv[0][0]) {
445 case 'r':
446 is_return = true;
447 break;
448 case 'p':
449 break;
450 default:
451 return -ECANCELED;
454 if (argc < 2)
455 return -ECANCELED;
457 if (argv[0][1] == ':')
458 event = &argv[0][2];
460 if (!strchr(argv[1], '/'))
461 return -ECANCELED;
463 filename = kstrdup(argv[1], GFP_KERNEL);
464 if (!filename)
465 return -ENOMEM;
467 /* Find the last occurrence, in case the path contains ':' too. */
468 arg = strrchr(filename, ':');
469 if (!arg || !isdigit(arg[1])) {
470 kfree(filename);
471 return -ECANCELED;
474 trace_probe_log_init("trace_uprobe", argc, argv);
475 trace_probe_log_set_index(1); /* filename is the 2nd argument */
477 *arg++ = '\0';
478 ret = kern_path(filename, LOOKUP_FOLLOW, &path);
479 if (ret) {
480 trace_probe_log_err(0, FILE_NOT_FOUND);
481 kfree(filename);
482 trace_probe_log_clear();
483 return ret;
485 if (!d_is_reg(path.dentry)) {
486 trace_probe_log_err(0, NO_REGULAR_FILE);
487 ret = -EINVAL;
488 goto fail_address_parse;
491 /* Parse reference counter offset if specified. */
492 rctr = strchr(arg, '(');
493 if (rctr) {
494 rctr_end = strchr(rctr, ')');
495 if (!rctr_end) {
496 ret = -EINVAL;
497 rctr_end = rctr + strlen(rctr);
498 trace_probe_log_err(rctr_end - filename,
499 REFCNT_OPEN_BRACE);
500 goto fail_address_parse;
501 } else if (rctr_end[1] != '\0') {
502 ret = -EINVAL;
503 trace_probe_log_err(rctr_end + 1 - filename,
504 BAD_REFCNT_SUFFIX);
505 goto fail_address_parse;
508 *rctr++ = '\0';
509 *rctr_end = '\0';
510 ret = kstrtoul(rctr, 0, &ref_ctr_offset);
511 if (ret) {
512 trace_probe_log_err(rctr - filename, BAD_REFCNT);
513 goto fail_address_parse;
517 /* Parse uprobe offset. */
518 ret = kstrtoul(arg, 0, &offset);
519 if (ret) {
520 trace_probe_log_err(arg - filename, BAD_UPROBE_OFFS);
521 goto fail_address_parse;
524 /* setup a probe */
525 trace_probe_log_set_index(0);
526 if (event) {
527 ret = traceprobe_parse_event_name(&event, &group, buf,
528 event - argv[0]);
529 if (ret)
530 goto fail_address_parse;
531 } else {
532 char *tail;
533 char *ptr;
535 tail = kstrdup(kbasename(filename), GFP_KERNEL);
536 if (!tail) {
537 ret = -ENOMEM;
538 goto fail_address_parse;
541 ptr = strpbrk(tail, ".-_");
542 if (ptr)
543 *ptr = '\0';
545 snprintf(buf, MAX_EVENT_NAME_LEN, "%c_%s_0x%lx", 'p', tail, offset);
546 event = buf;
547 kfree(tail);
550 argc -= 2;
551 argv += 2;
553 tu = alloc_trace_uprobe(group, event, argc, is_return);
554 if (IS_ERR(tu)) {
555 ret = PTR_ERR(tu);
556 /* This must return -ENOMEM otherwise there is a bug */
557 WARN_ON_ONCE(ret != -ENOMEM);
558 goto fail_address_parse;
560 tu->offset = offset;
561 tu->ref_ctr_offset = ref_ctr_offset;
562 tu->path = path;
563 tu->filename = filename;
565 /* parse arguments */
566 for (i = 0; i < argc && i < MAX_TRACE_ARGS; i++) {
567 tmp = kstrdup(argv[i], GFP_KERNEL);
568 if (!tmp) {
569 ret = -ENOMEM;
570 goto error;
573 trace_probe_log_set_index(i + 2);
574 ret = traceprobe_parse_probe_arg(&tu->tp, i, tmp,
575 is_return ? TPARG_FL_RETURN : 0);
576 kfree(tmp);
577 if (ret)
578 goto error;
581 ret = register_trace_uprobe(tu);
582 if (!ret)
583 goto out;
585 error:
586 free_trace_uprobe(tu);
587 out:
588 trace_probe_log_clear();
589 return ret;
591 fail_address_parse:
592 trace_probe_log_clear();
593 path_put(&path);
594 kfree(filename);
596 return ret;
599 static int create_or_delete_trace_uprobe(int argc, char **argv)
601 int ret;
603 if (argv[0][0] == '-')
604 return dyn_event_release(argc, argv, &trace_uprobe_ops);
606 ret = trace_uprobe_create(argc, (const char **)argv);
607 return ret == -ECANCELED ? -EINVAL : ret;
610 static int trace_uprobe_release(struct dyn_event *ev)
612 struct trace_uprobe *tu = to_trace_uprobe(ev);
614 return unregister_trace_uprobe(tu);
617 /* Probes listing interfaces */
618 static int trace_uprobe_show(struct seq_file *m, struct dyn_event *ev)
620 struct trace_uprobe *tu = to_trace_uprobe(ev);
621 char c = is_ret_probe(tu) ? 'r' : 'p';
622 int i;
624 seq_printf(m, "%c:%s/%s %s:0x%0*lx", c, tu->tp.call.class->system,
625 trace_event_name(&tu->tp.call), tu->filename,
626 (int)(sizeof(void *) * 2), tu->offset);
628 if (tu->ref_ctr_offset)
629 seq_printf(m, "(0x%lx)", tu->ref_ctr_offset);
631 for (i = 0; i < tu->tp.nr_args; i++)
632 seq_printf(m, " %s=%s", tu->tp.args[i].name, tu->tp.args[i].comm);
634 seq_putc(m, '\n');
635 return 0;
638 static int probes_seq_show(struct seq_file *m, void *v)
640 struct dyn_event *ev = v;
642 if (!is_trace_uprobe(ev))
643 return 0;
645 return trace_uprobe_show(m, ev);
648 static const struct seq_operations probes_seq_op = {
649 .start = dyn_event_seq_start,
650 .next = dyn_event_seq_next,
651 .stop = dyn_event_seq_stop,
652 .show = probes_seq_show
655 static int probes_open(struct inode *inode, struct file *file)
657 int ret;
659 if ((file->f_mode & FMODE_WRITE) && (file->f_flags & O_TRUNC)) {
660 ret = dyn_events_release_all(&trace_uprobe_ops);
661 if (ret)
662 return ret;
665 return seq_open(file, &probes_seq_op);
668 static ssize_t probes_write(struct file *file, const char __user *buffer,
669 size_t count, loff_t *ppos)
671 return trace_parse_run_command(file, buffer, count, ppos,
672 create_or_delete_trace_uprobe);
675 static const struct file_operations uprobe_events_ops = {
676 .owner = THIS_MODULE,
677 .open = probes_open,
678 .read = seq_read,
679 .llseek = seq_lseek,
680 .release = seq_release,
681 .write = probes_write,
684 /* Probes profiling interfaces */
685 static int probes_profile_seq_show(struct seq_file *m, void *v)
687 struct dyn_event *ev = v;
688 struct trace_uprobe *tu;
690 if (!is_trace_uprobe(ev))
691 return 0;
693 tu = to_trace_uprobe(ev);
694 seq_printf(m, " %s %-44s %15lu\n", tu->filename,
695 trace_event_name(&tu->tp.call), tu->nhit);
696 return 0;
699 static const struct seq_operations profile_seq_op = {
700 .start = dyn_event_seq_start,
701 .next = dyn_event_seq_next,
702 .stop = dyn_event_seq_stop,
703 .show = probes_profile_seq_show
706 static int profile_open(struct inode *inode, struct file *file)
708 return seq_open(file, &profile_seq_op);
711 static const struct file_operations uprobe_profile_ops = {
712 .owner = THIS_MODULE,
713 .open = profile_open,
714 .read = seq_read,
715 .llseek = seq_lseek,
716 .release = seq_release,
719 struct uprobe_cpu_buffer {
720 struct mutex mutex;
721 void *buf;
723 static struct uprobe_cpu_buffer __percpu *uprobe_cpu_buffer;
724 static int uprobe_buffer_refcnt;
726 static int uprobe_buffer_init(void)
728 int cpu, err_cpu;
730 uprobe_cpu_buffer = alloc_percpu(struct uprobe_cpu_buffer);
731 if (uprobe_cpu_buffer == NULL)
732 return -ENOMEM;
734 for_each_possible_cpu(cpu) {
735 struct page *p = alloc_pages_node(cpu_to_node(cpu),
736 GFP_KERNEL, 0);
737 if (p == NULL) {
738 err_cpu = cpu;
739 goto err;
741 per_cpu_ptr(uprobe_cpu_buffer, cpu)->buf = page_address(p);
742 mutex_init(&per_cpu_ptr(uprobe_cpu_buffer, cpu)->mutex);
745 return 0;
747 err:
748 for_each_possible_cpu(cpu) {
749 if (cpu == err_cpu)
750 break;
751 free_page((unsigned long)per_cpu_ptr(uprobe_cpu_buffer, cpu)->buf);
754 free_percpu(uprobe_cpu_buffer);
755 return -ENOMEM;
758 static int uprobe_buffer_enable(void)
760 int ret = 0;
762 BUG_ON(!mutex_is_locked(&event_mutex));
764 if (uprobe_buffer_refcnt++ == 0) {
765 ret = uprobe_buffer_init();
766 if (ret < 0)
767 uprobe_buffer_refcnt--;
770 return ret;
773 static void uprobe_buffer_disable(void)
775 int cpu;
777 BUG_ON(!mutex_is_locked(&event_mutex));
779 if (--uprobe_buffer_refcnt == 0) {
780 for_each_possible_cpu(cpu)
781 free_page((unsigned long)per_cpu_ptr(uprobe_cpu_buffer,
782 cpu)->buf);
784 free_percpu(uprobe_cpu_buffer);
785 uprobe_cpu_buffer = NULL;
789 static struct uprobe_cpu_buffer *uprobe_buffer_get(void)
791 struct uprobe_cpu_buffer *ucb;
792 int cpu;
794 cpu = raw_smp_processor_id();
795 ucb = per_cpu_ptr(uprobe_cpu_buffer, cpu);
798 * Use per-cpu buffers for fastest access, but we might migrate
799 * so the mutex makes sure we have sole access to it.
801 mutex_lock(&ucb->mutex);
803 return ucb;
806 static void uprobe_buffer_put(struct uprobe_cpu_buffer *ucb)
808 mutex_unlock(&ucb->mutex);
811 static void __uprobe_trace_func(struct trace_uprobe *tu,
812 unsigned long func, struct pt_regs *regs,
813 struct uprobe_cpu_buffer *ucb, int dsize,
814 struct trace_event_file *trace_file)
816 struct uprobe_trace_entry_head *entry;
817 struct ring_buffer_event *event;
818 struct ring_buffer *buffer;
819 void *data;
820 int size, esize;
821 struct trace_event_call *call = &tu->tp.call;
823 WARN_ON(call != trace_file->event_call);
825 if (WARN_ON_ONCE(tu->tp.size + dsize > PAGE_SIZE))
826 return;
828 if (trace_trigger_soft_disabled(trace_file))
829 return;
831 esize = SIZEOF_TRACE_ENTRY(is_ret_probe(tu));
832 size = esize + tu->tp.size + dsize;
833 event = trace_event_buffer_lock_reserve(&buffer, trace_file,
834 call->event.type, size, 0, 0);
835 if (!event)
836 return;
838 entry = ring_buffer_event_data(event);
839 if (is_ret_probe(tu)) {
840 entry->vaddr[0] = func;
841 entry->vaddr[1] = instruction_pointer(regs);
842 data = DATAOF_TRACE_ENTRY(entry, true);
843 } else {
844 entry->vaddr[0] = instruction_pointer(regs);
845 data = DATAOF_TRACE_ENTRY(entry, false);
848 memcpy(data, ucb->buf, tu->tp.size + dsize);
850 event_trigger_unlock_commit(trace_file, buffer, event, entry, 0, 0);
853 /* uprobe handler */
854 static int uprobe_trace_func(struct trace_uprobe *tu, struct pt_regs *regs,
855 struct uprobe_cpu_buffer *ucb, int dsize)
857 struct event_file_link *link;
859 if (is_ret_probe(tu))
860 return 0;
862 rcu_read_lock();
863 list_for_each_entry_rcu(link, &tu->tp.files, list)
864 __uprobe_trace_func(tu, 0, regs, ucb, dsize, link->file);
865 rcu_read_unlock();
867 return 0;
870 static void uretprobe_trace_func(struct trace_uprobe *tu, unsigned long func,
871 struct pt_regs *regs,
872 struct uprobe_cpu_buffer *ucb, int dsize)
874 struct event_file_link *link;
876 rcu_read_lock();
877 list_for_each_entry_rcu(link, &tu->tp.files, list)
878 __uprobe_trace_func(tu, func, regs, ucb, dsize, link->file);
879 rcu_read_unlock();
882 /* Event entry printers */
883 static enum print_line_t
884 print_uprobe_event(struct trace_iterator *iter, int flags, struct trace_event *event)
886 struct uprobe_trace_entry_head *entry;
887 struct trace_seq *s = &iter->seq;
888 struct trace_uprobe *tu;
889 u8 *data;
891 entry = (struct uprobe_trace_entry_head *)iter->ent;
892 tu = container_of(event, struct trace_uprobe, tp.call.event);
894 if (is_ret_probe(tu)) {
895 trace_seq_printf(s, "%s: (0x%lx <- 0x%lx)",
896 trace_event_name(&tu->tp.call),
897 entry->vaddr[1], entry->vaddr[0]);
898 data = DATAOF_TRACE_ENTRY(entry, true);
899 } else {
900 trace_seq_printf(s, "%s: (0x%lx)",
901 trace_event_name(&tu->tp.call),
902 entry->vaddr[0]);
903 data = DATAOF_TRACE_ENTRY(entry, false);
906 if (print_probe_args(s, tu->tp.args, tu->tp.nr_args, data, entry) < 0)
907 goto out;
909 trace_seq_putc(s, '\n');
911 out:
912 return trace_handle_return(s);
915 typedef bool (*filter_func_t)(struct uprobe_consumer *self,
916 enum uprobe_filter_ctx ctx,
917 struct mm_struct *mm);
919 static int
920 probe_event_enable(struct trace_uprobe *tu, struct trace_event_file *file,
921 filter_func_t filter)
923 bool enabled = trace_probe_is_enabled(&tu->tp);
924 struct event_file_link *link = NULL;
925 int ret;
927 if (file) {
928 if (tu->tp.flags & TP_FLAG_PROFILE)
929 return -EINTR;
931 link = kmalloc(sizeof(*link), GFP_KERNEL);
932 if (!link)
933 return -ENOMEM;
935 link->file = file;
936 list_add_tail_rcu(&link->list, &tu->tp.files);
938 tu->tp.flags |= TP_FLAG_TRACE;
939 } else {
940 if (tu->tp.flags & TP_FLAG_TRACE)
941 return -EINTR;
943 tu->tp.flags |= TP_FLAG_PROFILE;
946 WARN_ON(!uprobe_filter_is_empty(&tu->filter));
948 if (enabled)
949 return 0;
951 ret = uprobe_buffer_enable();
952 if (ret)
953 goto err_flags;
955 tu->consumer.filter = filter;
956 tu->inode = d_real_inode(tu->path.dentry);
957 if (tu->ref_ctr_offset) {
958 ret = uprobe_register_refctr(tu->inode, tu->offset,
959 tu->ref_ctr_offset, &tu->consumer);
960 } else {
961 ret = uprobe_register(tu->inode, tu->offset, &tu->consumer);
964 if (ret)
965 goto err_buffer;
967 return 0;
969 err_buffer:
970 uprobe_buffer_disable();
972 err_flags:
973 if (file) {
974 list_del(&link->list);
975 kfree(link);
976 tu->tp.flags &= ~TP_FLAG_TRACE;
977 } else {
978 tu->tp.flags &= ~TP_FLAG_PROFILE;
980 return ret;
983 static void
984 probe_event_disable(struct trace_uprobe *tu, struct trace_event_file *file)
986 if (!trace_probe_is_enabled(&tu->tp))
987 return;
989 if (file) {
990 struct event_file_link *link;
992 link = find_event_file_link(&tu->tp, file);
993 if (!link)
994 return;
996 list_del_rcu(&link->list);
997 /* synchronize with u{,ret}probe_trace_func */
998 synchronize_rcu();
999 kfree(link);
1001 if (!list_empty(&tu->tp.files))
1002 return;
1005 WARN_ON(!uprobe_filter_is_empty(&tu->filter));
1007 uprobe_unregister(tu->inode, tu->offset, &tu->consumer);
1008 tu->inode = NULL;
1009 tu->tp.flags &= file ? ~TP_FLAG_TRACE : ~TP_FLAG_PROFILE;
1011 uprobe_buffer_disable();
1014 static int uprobe_event_define_fields(struct trace_event_call *event_call)
1016 int ret, size;
1017 struct uprobe_trace_entry_head field;
1018 struct trace_uprobe *tu = event_call->data;
1020 if (is_ret_probe(tu)) {
1021 DEFINE_FIELD(unsigned long, vaddr[0], FIELD_STRING_FUNC, 0);
1022 DEFINE_FIELD(unsigned long, vaddr[1], FIELD_STRING_RETIP, 0);
1023 size = SIZEOF_TRACE_ENTRY(true);
1024 } else {
1025 DEFINE_FIELD(unsigned long, vaddr[0], FIELD_STRING_IP, 0);
1026 size = SIZEOF_TRACE_ENTRY(false);
1029 return traceprobe_define_arg_fields(event_call, size, &tu->tp);
1032 #ifdef CONFIG_PERF_EVENTS
1033 static bool
1034 __uprobe_perf_filter(struct trace_uprobe_filter *filter, struct mm_struct *mm)
1036 struct perf_event *event;
1038 if (filter->nr_systemwide)
1039 return true;
1041 list_for_each_entry(event, &filter->perf_events, hw.tp_list) {
1042 if (event->hw.target->mm == mm)
1043 return true;
1046 return false;
1049 static inline bool
1050 uprobe_filter_event(struct trace_uprobe *tu, struct perf_event *event)
1052 return __uprobe_perf_filter(&tu->filter, event->hw.target->mm);
1055 static int uprobe_perf_close(struct trace_uprobe *tu, struct perf_event *event)
1057 bool done;
1059 write_lock(&tu->filter.rwlock);
1060 if (event->hw.target) {
1061 list_del(&event->hw.tp_list);
1062 done = tu->filter.nr_systemwide ||
1063 (event->hw.target->flags & PF_EXITING) ||
1064 uprobe_filter_event(tu, event);
1065 } else {
1066 tu->filter.nr_systemwide--;
1067 done = tu->filter.nr_systemwide;
1069 write_unlock(&tu->filter.rwlock);
1071 if (!done)
1072 return uprobe_apply(tu->inode, tu->offset, &tu->consumer, false);
1074 return 0;
1077 static int uprobe_perf_open(struct trace_uprobe *tu, struct perf_event *event)
1079 bool done;
1080 int err;
1082 write_lock(&tu->filter.rwlock);
1083 if (event->hw.target) {
1085 * event->parent != NULL means copy_process(), we can avoid
1086 * uprobe_apply(). current->mm must be probed and we can rely
1087 * on dup_mmap() which preserves the already installed bp's.
1089 * attr.enable_on_exec means that exec/mmap will install the
1090 * breakpoints we need.
1092 done = tu->filter.nr_systemwide ||
1093 event->parent || event->attr.enable_on_exec ||
1094 uprobe_filter_event(tu, event);
1095 list_add(&event->hw.tp_list, &tu->filter.perf_events);
1096 } else {
1097 done = tu->filter.nr_systemwide;
1098 tu->filter.nr_systemwide++;
1100 write_unlock(&tu->filter.rwlock);
1102 err = 0;
1103 if (!done) {
1104 err = uprobe_apply(tu->inode, tu->offset, &tu->consumer, true);
1105 if (err)
1106 uprobe_perf_close(tu, event);
1108 return err;
1111 static bool uprobe_perf_filter(struct uprobe_consumer *uc,
1112 enum uprobe_filter_ctx ctx, struct mm_struct *mm)
1114 struct trace_uprobe *tu;
1115 int ret;
1117 tu = container_of(uc, struct trace_uprobe, consumer);
1118 read_lock(&tu->filter.rwlock);
1119 ret = __uprobe_perf_filter(&tu->filter, mm);
1120 read_unlock(&tu->filter.rwlock);
1122 return ret;
1125 static void __uprobe_perf_func(struct trace_uprobe *tu,
1126 unsigned long func, struct pt_regs *regs,
1127 struct uprobe_cpu_buffer *ucb, int dsize)
1129 struct trace_event_call *call = &tu->tp.call;
1130 struct uprobe_trace_entry_head *entry;
1131 struct hlist_head *head;
1132 void *data;
1133 int size, esize;
1134 int rctx;
1136 if (bpf_prog_array_valid(call) && !trace_call_bpf(call, regs))
1137 return;
1139 esize = SIZEOF_TRACE_ENTRY(is_ret_probe(tu));
1141 size = esize + tu->tp.size + dsize;
1142 size = ALIGN(size + sizeof(u32), sizeof(u64)) - sizeof(u32);
1143 if (WARN_ONCE(size > PERF_MAX_TRACE_SIZE, "profile buffer not large enough"))
1144 return;
1146 preempt_disable();
1147 head = this_cpu_ptr(call->perf_events);
1148 if (hlist_empty(head))
1149 goto out;
1151 entry = perf_trace_buf_alloc(size, NULL, &rctx);
1152 if (!entry)
1153 goto out;
1155 if (is_ret_probe(tu)) {
1156 entry->vaddr[0] = func;
1157 entry->vaddr[1] = instruction_pointer(regs);
1158 data = DATAOF_TRACE_ENTRY(entry, true);
1159 } else {
1160 entry->vaddr[0] = instruction_pointer(regs);
1161 data = DATAOF_TRACE_ENTRY(entry, false);
1164 memcpy(data, ucb->buf, tu->tp.size + dsize);
1166 if (size - esize > tu->tp.size + dsize) {
1167 int len = tu->tp.size + dsize;
1169 memset(data + len, 0, size - esize - len);
1172 perf_trace_buf_submit(entry, size, rctx, call->event.type, 1, regs,
1173 head, NULL);
1174 out:
1175 preempt_enable();
1178 /* uprobe profile handler */
1179 static int uprobe_perf_func(struct trace_uprobe *tu, struct pt_regs *regs,
1180 struct uprobe_cpu_buffer *ucb, int dsize)
1182 if (!uprobe_perf_filter(&tu->consumer, 0, current->mm))
1183 return UPROBE_HANDLER_REMOVE;
1185 if (!is_ret_probe(tu))
1186 __uprobe_perf_func(tu, 0, regs, ucb, dsize);
1187 return 0;
1190 static void uretprobe_perf_func(struct trace_uprobe *tu, unsigned long func,
1191 struct pt_regs *regs,
1192 struct uprobe_cpu_buffer *ucb, int dsize)
1194 __uprobe_perf_func(tu, func, regs, ucb, dsize);
1197 int bpf_get_uprobe_info(const struct perf_event *event, u32 *fd_type,
1198 const char **filename, u64 *probe_offset,
1199 bool perf_type_tracepoint)
1201 const char *pevent = trace_event_name(event->tp_event);
1202 const char *group = event->tp_event->class->system;
1203 struct trace_uprobe *tu;
1205 if (perf_type_tracepoint)
1206 tu = find_probe_event(pevent, group);
1207 else
1208 tu = event->tp_event->data;
1209 if (!tu)
1210 return -EINVAL;
1212 *fd_type = is_ret_probe(tu) ? BPF_FD_TYPE_URETPROBE
1213 : BPF_FD_TYPE_UPROBE;
1214 *filename = tu->filename;
1215 *probe_offset = tu->offset;
1216 return 0;
1218 #endif /* CONFIG_PERF_EVENTS */
1220 static int
1221 trace_uprobe_register(struct trace_event_call *event, enum trace_reg type,
1222 void *data)
1224 struct trace_uprobe *tu = event->data;
1225 struct trace_event_file *file = data;
1227 switch (type) {
1228 case TRACE_REG_REGISTER:
1229 return probe_event_enable(tu, file, NULL);
1231 case TRACE_REG_UNREGISTER:
1232 probe_event_disable(tu, file);
1233 return 0;
1235 #ifdef CONFIG_PERF_EVENTS
1236 case TRACE_REG_PERF_REGISTER:
1237 return probe_event_enable(tu, NULL, uprobe_perf_filter);
1239 case TRACE_REG_PERF_UNREGISTER:
1240 probe_event_disable(tu, NULL);
1241 return 0;
1243 case TRACE_REG_PERF_OPEN:
1244 return uprobe_perf_open(tu, data);
1246 case TRACE_REG_PERF_CLOSE:
1247 return uprobe_perf_close(tu, data);
1249 #endif
1250 default:
1251 return 0;
1253 return 0;
1256 static int uprobe_dispatcher(struct uprobe_consumer *con, struct pt_regs *regs)
1258 struct trace_uprobe *tu;
1259 struct uprobe_dispatch_data udd;
1260 struct uprobe_cpu_buffer *ucb;
1261 int dsize, esize;
1262 int ret = 0;
1265 tu = container_of(con, struct trace_uprobe, consumer);
1266 tu->nhit++;
1268 udd.tu = tu;
1269 udd.bp_addr = instruction_pointer(regs);
1271 current->utask->vaddr = (unsigned long) &udd;
1273 if (WARN_ON_ONCE(!uprobe_cpu_buffer))
1274 return 0;
1276 dsize = __get_data_size(&tu->tp, regs);
1277 esize = SIZEOF_TRACE_ENTRY(is_ret_probe(tu));
1279 ucb = uprobe_buffer_get();
1280 store_trace_args(ucb->buf, &tu->tp, regs, esize, dsize);
1282 if (tu->tp.flags & TP_FLAG_TRACE)
1283 ret |= uprobe_trace_func(tu, regs, ucb, dsize);
1285 #ifdef CONFIG_PERF_EVENTS
1286 if (tu->tp.flags & TP_FLAG_PROFILE)
1287 ret |= uprobe_perf_func(tu, regs, ucb, dsize);
1288 #endif
1289 uprobe_buffer_put(ucb);
1290 return ret;
1293 static int uretprobe_dispatcher(struct uprobe_consumer *con,
1294 unsigned long func, struct pt_regs *regs)
1296 struct trace_uprobe *tu;
1297 struct uprobe_dispatch_data udd;
1298 struct uprobe_cpu_buffer *ucb;
1299 int dsize, esize;
1301 tu = container_of(con, struct trace_uprobe, consumer);
1303 udd.tu = tu;
1304 udd.bp_addr = func;
1306 current->utask->vaddr = (unsigned long) &udd;
1308 if (WARN_ON_ONCE(!uprobe_cpu_buffer))
1309 return 0;
1311 dsize = __get_data_size(&tu->tp, regs);
1312 esize = SIZEOF_TRACE_ENTRY(is_ret_probe(tu));
1314 ucb = uprobe_buffer_get();
1315 store_trace_args(ucb->buf, &tu->tp, regs, esize, dsize);
1317 if (tu->tp.flags & TP_FLAG_TRACE)
1318 uretprobe_trace_func(tu, func, regs, ucb, dsize);
1320 #ifdef CONFIG_PERF_EVENTS
1321 if (tu->tp.flags & TP_FLAG_PROFILE)
1322 uretprobe_perf_func(tu, func, regs, ucb, dsize);
1323 #endif
1324 uprobe_buffer_put(ucb);
1325 return 0;
1328 static struct trace_event_functions uprobe_funcs = {
1329 .trace = print_uprobe_event
1332 static inline void init_trace_event_call(struct trace_uprobe *tu,
1333 struct trace_event_call *call)
1335 INIT_LIST_HEAD(&call->class->fields);
1336 call->event.funcs = &uprobe_funcs;
1337 call->class->define_fields = uprobe_event_define_fields;
1339 call->flags = TRACE_EVENT_FL_UPROBE;
1340 call->class->reg = trace_uprobe_register;
1341 call->data = tu;
1344 static int register_uprobe_event(struct trace_uprobe *tu)
1346 struct trace_event_call *call = &tu->tp.call;
1347 int ret = 0;
1349 init_trace_event_call(tu, call);
1351 if (traceprobe_set_print_fmt(&tu->tp, is_ret_probe(tu)) < 0)
1352 return -ENOMEM;
1354 ret = register_trace_event(&call->event);
1355 if (!ret) {
1356 kfree(call->print_fmt);
1357 return -ENODEV;
1360 ret = trace_add_event_call(call);
1362 if (ret) {
1363 pr_info("Failed to register uprobe event: %s\n",
1364 trace_event_name(call));
1365 kfree(call->print_fmt);
1366 unregister_trace_event(&call->event);
1369 return ret;
1372 static int unregister_uprobe_event(struct trace_uprobe *tu)
1374 int ret;
1376 /* tu->event is unregistered in trace_remove_event_call() */
1377 ret = trace_remove_event_call(&tu->tp.call);
1378 if (ret)
1379 return ret;
1380 kfree(tu->tp.call.print_fmt);
1381 tu->tp.call.print_fmt = NULL;
1382 return 0;
1385 #ifdef CONFIG_PERF_EVENTS
1386 struct trace_event_call *
1387 create_local_trace_uprobe(char *name, unsigned long offs,
1388 unsigned long ref_ctr_offset, bool is_return)
1390 struct trace_uprobe *tu;
1391 struct path path;
1392 int ret;
1394 ret = kern_path(name, LOOKUP_FOLLOW, &path);
1395 if (ret)
1396 return ERR_PTR(ret);
1398 if (!d_is_reg(path.dentry)) {
1399 path_put(&path);
1400 return ERR_PTR(-EINVAL);
1404 * local trace_kprobes are not added to dyn_event, so they are never
1405 * searched in find_trace_kprobe(). Therefore, there is no concern of
1406 * duplicated name "DUMMY_EVENT" here.
1408 tu = alloc_trace_uprobe(UPROBE_EVENT_SYSTEM, "DUMMY_EVENT", 0,
1409 is_return);
1411 if (IS_ERR(tu)) {
1412 pr_info("Failed to allocate trace_uprobe.(%d)\n",
1413 (int)PTR_ERR(tu));
1414 path_put(&path);
1415 return ERR_CAST(tu);
1418 tu->offset = offs;
1419 tu->path = path;
1420 tu->ref_ctr_offset = ref_ctr_offset;
1421 tu->filename = kstrdup(name, GFP_KERNEL);
1422 init_trace_event_call(tu, &tu->tp.call);
1424 if (traceprobe_set_print_fmt(&tu->tp, is_ret_probe(tu)) < 0) {
1425 ret = -ENOMEM;
1426 goto error;
1429 return &tu->tp.call;
1430 error:
1431 free_trace_uprobe(tu);
1432 return ERR_PTR(ret);
1435 void destroy_local_trace_uprobe(struct trace_event_call *event_call)
1437 struct trace_uprobe *tu;
1439 tu = container_of(event_call, struct trace_uprobe, tp.call);
1441 kfree(tu->tp.call.print_fmt);
1442 tu->tp.call.print_fmt = NULL;
1444 free_trace_uprobe(tu);
1446 #endif /* CONFIG_PERF_EVENTS */
1448 /* Make a trace interface for controling probe points */
1449 static __init int init_uprobe_trace(void)
1451 struct dentry *d_tracer;
1452 int ret;
1454 ret = dyn_event_register(&trace_uprobe_ops);
1455 if (ret)
1456 return ret;
1458 d_tracer = tracing_init_dentry();
1459 if (IS_ERR(d_tracer))
1460 return 0;
1462 trace_create_file("uprobe_events", 0644, d_tracer,
1463 NULL, &uprobe_events_ops);
1464 /* Profile interface */
1465 trace_create_file("uprobe_profile", 0444, d_tracer,
1466 NULL, &uprobe_profile_ops);
1467 return 0;
1470 fs_initcall(init_uprobe_trace);