1 // SPDX-License-Identifier: GPL-2.0
3 * Kprobes-based tracing events
5 * Created by Masami Hiramatsu <mhiramat@redhat.com>
8 #define pr_fmt(fmt) "trace_kprobe: " fmt
10 #include <linux/module.h>
11 #include <linux/uaccess.h>
12 #include <linux/rculist.h>
13 #include <linux/error-injection.h>
15 #include <asm/setup.h> /* for COMMAND_LINE_SIZE */
17 #include "trace_dynevent.h"
18 #include "trace_kprobe_selftest.h"
19 #include "trace_probe.h"
20 #include "trace_probe_tmpl.h"
22 #define KPROBE_EVENT_SYSTEM "kprobes"
23 #define KRETPROBE_MAXACTIVE_MAX 4096
24 #define MAX_KPROBE_CMDLINE_SIZE 1024
26 /* Kprobe early definition from command line */
27 static char kprobe_boot_events_buf[COMMAND_LINE_SIZE] __initdata;
28 static bool kprobe_boot_events_enabled __initdata;
30 static int __init set_kprobe_boot_events(char *str)
32 strlcpy(kprobe_boot_events_buf, str, COMMAND_LINE_SIZE);
35 __setup("kprobe_event=", set_kprobe_boot_events);
37 static int trace_kprobe_create(int argc, const char **argv);
38 static int trace_kprobe_show(struct seq_file *m, struct dyn_event *ev);
39 static int trace_kprobe_release(struct dyn_event *ev);
40 static bool trace_kprobe_is_busy(struct dyn_event *ev);
41 static bool trace_kprobe_match(const char *system, const char *event,
42 struct dyn_event *ev);
44 static struct dyn_event_operations trace_kprobe_ops = {
45 .create = trace_kprobe_create,
46 .show = trace_kprobe_show,
47 .is_busy = trace_kprobe_is_busy,
48 .free = trace_kprobe_release,
49 .match = trace_kprobe_match,
53 * Kprobe event core functions
56 struct dyn_event devent;
57 struct kretprobe rp; /* Use rp.kp for kprobe use */
58 unsigned long __percpu *nhit;
59 const char *symbol; /* symbol name */
60 struct trace_probe tp;
63 static bool is_trace_kprobe(struct dyn_event *ev)
65 return ev->ops == &trace_kprobe_ops;
68 static struct trace_kprobe *to_trace_kprobe(struct dyn_event *ev)
70 return container_of(ev, struct trace_kprobe, devent);
74 * for_each_trace_kprobe - iterate over the trace_kprobe list
75 * @pos: the struct trace_kprobe * for each entry
76 * @dpos: the struct dyn_event * to use as a loop cursor
78 #define for_each_trace_kprobe(pos, dpos) \
79 for_each_dyn_event(dpos) \
80 if (is_trace_kprobe(dpos) && (pos = to_trace_kprobe(dpos)))
82 #define SIZEOF_TRACE_KPROBE(n) \
83 (offsetof(struct trace_kprobe, tp.args) + \
84 (sizeof(struct probe_arg) * (n)))
86 static nokprobe_inline bool trace_kprobe_is_return(struct trace_kprobe *tk)
88 return tk->rp.handler != NULL;
91 static nokprobe_inline const char *trace_kprobe_symbol(struct trace_kprobe *tk)
93 return tk->symbol ? tk->symbol : "unknown";
96 static nokprobe_inline unsigned long trace_kprobe_offset(struct trace_kprobe *tk)
98 return tk->rp.kp.offset;
101 static nokprobe_inline bool trace_kprobe_has_gone(struct trace_kprobe *tk)
103 return !!(kprobe_gone(&tk->rp.kp));
106 static nokprobe_inline bool trace_kprobe_within_module(struct trace_kprobe *tk,
109 int len = strlen(mod->name);
110 const char *name = trace_kprobe_symbol(tk);
111 return strncmp(mod->name, name, len) == 0 && name[len] == ':';
114 static nokprobe_inline bool trace_kprobe_module_exist(struct trace_kprobe *tk)
121 p = strchr(tk->symbol, ':');
125 mutex_lock(&module_mutex);
126 ret = !!find_module(tk->symbol);
127 mutex_unlock(&module_mutex);
133 static bool trace_kprobe_is_busy(struct dyn_event *ev)
135 struct trace_kprobe *tk = to_trace_kprobe(ev);
137 return trace_probe_is_enabled(&tk->tp);
140 static bool trace_kprobe_match(const char *system, const char *event,
141 struct dyn_event *ev)
143 struct trace_kprobe *tk = to_trace_kprobe(ev);
145 return strcmp(trace_probe_name(&tk->tp), event) == 0 &&
146 (!system || strcmp(trace_probe_group_name(&tk->tp), system) == 0);
149 static nokprobe_inline unsigned long trace_kprobe_nhit(struct trace_kprobe *tk)
151 unsigned long nhit = 0;
154 for_each_possible_cpu(cpu)
155 nhit += *per_cpu_ptr(tk->nhit, cpu);
160 static nokprobe_inline bool trace_kprobe_is_registered(struct trace_kprobe *tk)
162 return !(list_empty(&tk->rp.kp.list) &&
163 hlist_unhashed(&tk->rp.kp.hlist));
166 /* Return 0 if it fails to find the symbol address */
167 static nokprobe_inline
168 unsigned long trace_kprobe_address(struct trace_kprobe *tk)
173 addr = (unsigned long)
174 kallsyms_lookup_name(trace_kprobe_symbol(tk));
176 addr += tk->rp.kp.offset;
178 addr = (unsigned long)tk->rp.kp.addr;
183 bool trace_kprobe_on_func_entry(struct trace_event_call *call)
185 struct trace_kprobe *tk = (struct trace_kprobe *)call->data;
187 return kprobe_on_func_entry(tk->rp.kp.addr,
188 tk->rp.kp.addr ? NULL : tk->rp.kp.symbol_name,
189 tk->rp.kp.addr ? 0 : tk->rp.kp.offset);
192 bool trace_kprobe_error_injectable(struct trace_event_call *call)
194 struct trace_kprobe *tk = (struct trace_kprobe *)call->data;
196 return within_error_injection_list(trace_kprobe_address(tk));
199 static int register_kprobe_event(struct trace_kprobe *tk);
200 static int unregister_kprobe_event(struct trace_kprobe *tk);
202 static int kprobe_dispatcher(struct kprobe *kp, struct pt_regs *regs);
203 static int kretprobe_dispatcher(struct kretprobe_instance *ri,
204 struct pt_regs *regs);
206 static void free_trace_kprobe(struct trace_kprobe *tk)
209 trace_probe_cleanup(&tk->tp);
211 free_percpu(tk->nhit);
217 * Allocate new trace_probe and initialize it (including kprobes).
219 static struct trace_kprobe *alloc_trace_kprobe(const char *group,
225 int nargs, bool is_return)
227 struct trace_kprobe *tk;
230 tk = kzalloc(SIZEOF_TRACE_KPROBE(nargs), GFP_KERNEL);
234 tk->nhit = alloc_percpu(unsigned long);
239 tk->symbol = kstrdup(symbol, GFP_KERNEL);
242 tk->rp.kp.symbol_name = tk->symbol;
243 tk->rp.kp.offset = offs;
245 tk->rp.kp.addr = addr;
248 tk->rp.handler = kretprobe_dispatcher;
250 tk->rp.kp.pre_handler = kprobe_dispatcher;
252 tk->rp.maxactive = maxactive;
253 INIT_HLIST_NODE(&tk->rp.kp.hlist);
254 INIT_LIST_HEAD(&tk->rp.kp.list);
256 ret = trace_probe_init(&tk->tp, event, group);
260 dyn_event_init(&tk->devent, &trace_kprobe_ops);
263 free_trace_kprobe(tk);
267 static struct trace_kprobe *find_trace_kprobe(const char *event,
270 struct dyn_event *pos;
271 struct trace_kprobe *tk;
273 for_each_trace_kprobe(tk, pos)
274 if (strcmp(trace_probe_name(&tk->tp), event) == 0 &&
275 strcmp(trace_probe_group_name(&tk->tp), group) == 0)
280 static inline int __enable_trace_kprobe(struct trace_kprobe *tk)
284 if (trace_kprobe_is_registered(tk) && !trace_kprobe_has_gone(tk)) {
285 if (trace_kprobe_is_return(tk))
286 ret = enable_kretprobe(&tk->rp);
288 ret = enable_kprobe(&tk->rp.kp);
296 * if the file is NULL, enable "perf" handler, or enable "trace" handler.
299 enable_trace_kprobe(struct trace_kprobe *tk, struct trace_event_file *file)
301 bool enabled = trace_probe_is_enabled(&tk->tp);
305 ret = trace_probe_add_file(&tk->tp, file);
309 trace_probe_set_flag(&tk->tp, TP_FLAG_PROFILE);
314 ret = __enable_trace_kprobe(tk);
317 trace_probe_remove_file(&tk->tp, file);
319 trace_probe_clear_flag(&tk->tp, TP_FLAG_PROFILE);
326 * Disable trace_probe
327 * if the file is NULL, disable "perf" handler, or disable "trace" handler.
330 disable_trace_kprobe(struct trace_kprobe *tk, struct trace_event_file *file)
332 struct trace_probe *tp = &tk->tp;
336 if (!trace_probe_get_file_link(tp, file))
338 if (!trace_probe_has_single_file(tp))
340 trace_probe_clear_flag(tp, TP_FLAG_TRACE);
342 trace_probe_clear_flag(tp, TP_FLAG_PROFILE);
344 if (!trace_probe_is_enabled(tp) && trace_kprobe_is_registered(tk)) {
345 if (trace_kprobe_is_return(tk))
346 disable_kretprobe(&tk->rp);
348 disable_kprobe(&tk->rp.kp);
354 * Synchronization is done in below function. For perf event,
355 * file == NULL and perf_trace_event_unreg() calls
356 * tracepoint_synchronize_unregister() to ensure synchronize
357 * event. We don't need to care about it.
359 trace_probe_remove_file(tp, file);
364 #if defined(CONFIG_KPROBES_ON_FTRACE) && \
365 !defined(CONFIG_KPROBE_EVENTS_ON_NOTRACE)
366 static bool within_notrace_func(struct trace_kprobe *tk)
368 unsigned long offset, size, addr;
370 addr = trace_kprobe_address(tk);
371 if (!addr || !kallsyms_lookup_size_offset(addr, &size, &offset))
374 /* Get the entry address of the target function */
378 * Since ftrace_location_range() does inclusive range check, we need
379 * to subtract 1 byte from the end address.
381 return !ftrace_location_range(addr, addr + size - 1);
384 #define within_notrace_func(tk) (false)
387 /* Internal register function - just handle k*probes and flags */
388 static int __register_trace_kprobe(struct trace_kprobe *tk)
392 if (trace_kprobe_is_registered(tk))
395 if (within_notrace_func(tk)) {
396 pr_warn("Could not probe notrace function %s\n",
397 trace_kprobe_symbol(tk));
401 for (i = 0; i < tk->tp.nr_args; i++) {
402 ret = traceprobe_update_arg(&tk->tp.args[i]);
407 /* Set/clear disabled flag according to tp->flag */
408 if (trace_probe_is_enabled(&tk->tp))
409 tk->rp.kp.flags &= ~KPROBE_FLAG_DISABLED;
411 tk->rp.kp.flags |= KPROBE_FLAG_DISABLED;
413 if (trace_kprobe_is_return(tk))
414 ret = register_kretprobe(&tk->rp);
416 ret = register_kprobe(&tk->rp.kp);
421 /* Internal unregister function - just handle k*probes and flags */
422 static void __unregister_trace_kprobe(struct trace_kprobe *tk)
424 if (trace_kprobe_is_registered(tk)) {
425 if (trace_kprobe_is_return(tk))
426 unregister_kretprobe(&tk->rp);
428 unregister_kprobe(&tk->rp.kp);
429 /* Cleanup kprobe for reuse and mark it unregistered */
430 INIT_HLIST_NODE(&tk->rp.kp.hlist);
431 INIT_LIST_HEAD(&tk->rp.kp.list);
432 if (tk->rp.kp.symbol_name)
433 tk->rp.kp.addr = NULL;
437 /* Unregister a trace_probe and probe_event */
438 static int unregister_trace_kprobe(struct trace_kprobe *tk)
440 /* Enabled event can not be unregistered */
441 if (trace_probe_is_enabled(&tk->tp))
444 /* Will fail if probe is being used by ftrace or perf */
445 if (unregister_kprobe_event(tk))
448 __unregister_trace_kprobe(tk);
449 dyn_event_remove(&tk->devent);
454 /* Register a trace_probe and probe_event */
455 static int register_trace_kprobe(struct trace_kprobe *tk)
457 struct trace_kprobe *old_tk;
460 mutex_lock(&event_mutex);
462 /* Delete old (same name) event if exist */
463 old_tk = find_trace_kprobe(trace_probe_name(&tk->tp),
464 trace_probe_group_name(&tk->tp));
466 ret = unregister_trace_kprobe(old_tk);
469 free_trace_kprobe(old_tk);
472 /* Register new event */
473 ret = register_kprobe_event(tk);
475 pr_warn("Failed to register probe event(%d)\n", ret);
479 /* Register k*probe */
480 ret = __register_trace_kprobe(tk);
481 if (ret == -ENOENT && !trace_kprobe_module_exist(tk)) {
482 pr_warn("This probe might be able to register after target module is loaded. Continue.\n");
487 unregister_kprobe_event(tk);
489 dyn_event_add(&tk->devent);
492 mutex_unlock(&event_mutex);
496 /* Module notifier call back, checking event on the module */
497 static int trace_kprobe_module_callback(struct notifier_block *nb,
498 unsigned long val, void *data)
500 struct module *mod = data;
501 struct dyn_event *pos;
502 struct trace_kprobe *tk;
505 if (val != MODULE_STATE_COMING)
508 /* Update probes on coming module */
509 mutex_lock(&event_mutex);
510 for_each_trace_kprobe(tk, pos) {
511 if (trace_kprobe_within_module(tk, mod)) {
512 /* Don't need to check busy - this should have gone. */
513 __unregister_trace_kprobe(tk);
514 ret = __register_trace_kprobe(tk);
516 pr_warn("Failed to re-register probe %s on %s: %d\n",
517 trace_probe_name(&tk->tp),
521 mutex_unlock(&event_mutex);
526 static struct notifier_block trace_kprobe_module_nb = {
527 .notifier_call = trace_kprobe_module_callback,
528 .priority = 1 /* Invoked after kprobe module callback */
531 /* Convert certain expected symbols into '_' when generating event names */
532 static inline void sanitize_event_name(char *name)
534 while (*name++ != '\0')
535 if (*name == ':' || *name == '.')
539 static int trace_kprobe_create(int argc, const char *argv[])
544 * p[:[GRP/]EVENT] [MOD:]KSYM[+OFFS]|KADDR [FETCHARGS]
546 * r[MAXACTIVE][:[GRP/]EVENT] [MOD:]KSYM[+0] [FETCHARGS]
548 * $retval : fetch return value
549 * $stack : fetch stack address
550 * $stackN : fetch Nth of stack (N:0-)
551 * $comm : fetch current task comm
552 * @ADDR : fetch memory at ADDR (ADDR should be in kernel)
553 * @SYM[+|-offs] : fetch memory at SYM +|- offs (SYM is a data symbol)
554 * %REG : fetch register REG
555 * Dereferencing memory fetch:
556 * +|-offs(ARG) : fetch memory at ARG +|- offs address.
557 * Alias name of args:
558 * NAME=FETCHARG : set NAME as alias of FETCHARG.
560 * FETCHARG:TYPE : use TYPE instead of unsigned long.
562 struct trace_kprobe *tk = NULL;
564 bool is_return = false;
565 char *symbol = NULL, *tmp = NULL;
566 const char *event = NULL, *group = KPROBE_EVENT_SYSTEM;
570 char buf[MAX_EVENT_NAME_LEN];
571 unsigned int flags = TPARG_FL_KERNEL;
573 switch (argv[0][0]) {
576 flags |= TPARG_FL_RETURN;
586 trace_probe_log_init("trace_kprobe", argc, argv);
588 event = strchr(&argv[0][1], ':');
592 if (isdigit(argv[0][1])) {
594 trace_probe_log_err(1, MAXACT_NO_KPROBE);
598 len = event - &argv[0][1] - 1;
600 len = strlen(&argv[0][1]);
601 if (len > MAX_EVENT_NAME_LEN - 1) {
602 trace_probe_log_err(1, BAD_MAXACT);
605 memcpy(buf, &argv[0][1], len);
607 ret = kstrtouint(buf, 0, &maxactive);
608 if (ret || !maxactive) {
609 trace_probe_log_err(1, BAD_MAXACT);
612 /* kretprobes instances are iterated over via a list. The
613 * maximum should stay reasonable.
615 if (maxactive > KRETPROBE_MAXACTIVE_MAX) {
616 trace_probe_log_err(1, MAXACT_TOO_BIG);
621 /* try to parse an address. if that fails, try to read the
622 * input as a symbol. */
623 if (kstrtoul(argv[1], 0, (unsigned long *)&addr)) {
624 trace_probe_log_set_index(1);
625 /* Check whether uprobe event specified */
626 if (strchr(argv[1], '/') && strchr(argv[1], ':')) {
630 /* a symbol specified */
631 symbol = kstrdup(argv[1], GFP_KERNEL);
634 /* TODO: support .init module functions */
635 ret = traceprobe_split_symbol_offset(symbol, &offset);
636 if (ret || offset < 0 || offset > UINT_MAX) {
637 trace_probe_log_err(0, BAD_PROBE_ADDR);
640 if (kprobe_on_func_entry(NULL, symbol, offset))
641 flags |= TPARG_FL_FENTRY;
642 if (offset && is_return && !(flags & TPARG_FL_FENTRY)) {
643 trace_probe_log_err(0, BAD_RETPROBE);
648 trace_probe_log_set_index(0);
650 ret = traceprobe_parse_event_name(&event, &group, buf,
655 /* Make a new event name */
657 snprintf(buf, MAX_EVENT_NAME_LEN, "%c_%s_%ld",
658 is_return ? 'r' : 'p', symbol, offset);
660 snprintf(buf, MAX_EVENT_NAME_LEN, "%c_0x%p",
661 is_return ? 'r' : 'p', addr);
662 sanitize_event_name(buf);
667 tk = alloc_trace_kprobe(group, event, addr, symbol, offset, maxactive,
668 argc - 2, is_return);
671 /* This must return -ENOMEM, else there is a bug */
672 WARN_ON_ONCE(ret != -ENOMEM);
673 goto out; /* We know tk is not allocated */
675 argc -= 2; argv += 2;
677 /* parse arguments */
678 for (i = 0; i < argc && i < MAX_TRACE_ARGS; i++) {
679 tmp = kstrdup(argv[i], GFP_KERNEL);
685 trace_probe_log_set_index(i + 2);
686 ret = traceprobe_parse_probe_arg(&tk->tp, i, tmp, flags);
689 goto error; /* This can be -ENOMEM */
692 ret = traceprobe_set_print_fmt(&tk->tp, is_return);
696 ret = register_trace_kprobe(tk);
698 trace_probe_log_set_index(1);
700 trace_probe_log_err(0, BAD_INSN_BNDRY);
701 else if (ret == -ENOENT)
702 trace_probe_log_err(0, BAD_PROBE_ADDR);
703 else if (ret != -ENOMEM)
704 trace_probe_log_err(0, FAIL_REG_PROBE);
709 trace_probe_log_clear();
716 free_trace_kprobe(tk);
720 static int create_or_delete_trace_kprobe(int argc, char **argv)
724 if (argv[0][0] == '-')
725 return dyn_event_release(argc, argv, &trace_kprobe_ops);
727 ret = trace_kprobe_create(argc, (const char **)argv);
728 return ret == -ECANCELED ? -EINVAL : ret;
731 static int trace_kprobe_release(struct dyn_event *ev)
733 struct trace_kprobe *tk = to_trace_kprobe(ev);
734 int ret = unregister_trace_kprobe(tk);
737 free_trace_kprobe(tk);
741 static int trace_kprobe_show(struct seq_file *m, struct dyn_event *ev)
743 struct trace_kprobe *tk = to_trace_kprobe(ev);
746 seq_putc(m, trace_kprobe_is_return(tk) ? 'r' : 'p');
747 seq_printf(m, ":%s/%s", trace_probe_group_name(&tk->tp),
748 trace_probe_name(&tk->tp));
751 seq_printf(m, " 0x%p", tk->rp.kp.addr);
752 else if (tk->rp.kp.offset)
753 seq_printf(m, " %s+%u", trace_kprobe_symbol(tk),
756 seq_printf(m, " %s", trace_kprobe_symbol(tk));
758 for (i = 0; i < tk->tp.nr_args; i++)
759 seq_printf(m, " %s=%s", tk->tp.args[i].name, tk->tp.args[i].comm);
765 static int probes_seq_show(struct seq_file *m, void *v)
767 struct dyn_event *ev = v;
769 if (!is_trace_kprobe(ev))
772 return trace_kprobe_show(m, ev);
775 static const struct seq_operations probes_seq_op = {
776 .start = dyn_event_seq_start,
777 .next = dyn_event_seq_next,
778 .stop = dyn_event_seq_stop,
779 .show = probes_seq_show
782 static int probes_open(struct inode *inode, struct file *file)
786 if ((file->f_mode & FMODE_WRITE) && (file->f_flags & O_TRUNC)) {
787 ret = dyn_events_release_all(&trace_kprobe_ops);
792 return seq_open(file, &probes_seq_op);
795 static ssize_t probes_write(struct file *file, const char __user *buffer,
796 size_t count, loff_t *ppos)
798 return trace_parse_run_command(file, buffer, count, ppos,
799 create_or_delete_trace_kprobe);
802 static const struct file_operations kprobe_events_ops = {
803 .owner = THIS_MODULE,
807 .release = seq_release,
808 .write = probes_write,
811 /* Probes profiling interfaces */
812 static int probes_profile_seq_show(struct seq_file *m, void *v)
814 struct dyn_event *ev = v;
815 struct trace_kprobe *tk;
817 if (!is_trace_kprobe(ev))
820 tk = to_trace_kprobe(ev);
821 seq_printf(m, " %-44s %15lu %15lu\n",
822 trace_probe_name(&tk->tp),
823 trace_kprobe_nhit(tk),
829 static const struct seq_operations profile_seq_op = {
830 .start = dyn_event_seq_start,
831 .next = dyn_event_seq_next,
832 .stop = dyn_event_seq_stop,
833 .show = probes_profile_seq_show
836 static int profile_open(struct inode *inode, struct file *file)
838 return seq_open(file, &profile_seq_op);
841 static const struct file_operations kprobe_profile_ops = {
842 .owner = THIS_MODULE,
843 .open = profile_open,
846 .release = seq_release,
849 /* Kprobe specific fetch functions */
851 /* Return the length of string -- including null terminal byte */
852 static nokprobe_inline int
853 fetch_store_strlen(unsigned long addr)
859 ret = probe_kernel_read(&c, (u8 *)addr + len, 1);
861 } while (c && ret == 0 && len < MAX_STRING_SIZE);
863 return (ret < 0) ? ret : len;
866 /* Return the length of string -- including null terminal byte */
867 static nokprobe_inline int
868 fetch_store_strlen_user(unsigned long addr)
870 const void __user *uaddr = (__force const void __user *)addr;
872 return strnlen_unsafe_user(uaddr, MAX_STRING_SIZE);
876 * Fetch a null-terminated string. Caller MUST set *(u32 *)buf with max
877 * length and relative data location.
879 static nokprobe_inline int
880 fetch_store_string(unsigned long addr, void *dest, void *base)
882 int maxlen = get_loc_len(*(u32 *)dest);
886 if (unlikely(!maxlen))
889 __dest = get_loc_data(dest, base);
892 * Try to get string again, since the string can be changed while
895 ret = strncpy_from_unsafe(__dest, (void *)addr, maxlen);
897 *(u32 *)dest = make_data_loc(ret, __dest - base);
903 * Fetch a null-terminated string from user. Caller MUST set *(u32 *)buf
904 * with max length and relative data location.
906 static nokprobe_inline int
907 fetch_store_string_user(unsigned long addr, void *dest, void *base)
909 const void __user *uaddr = (__force const void __user *)addr;
910 int maxlen = get_loc_len(*(u32 *)dest);
914 if (unlikely(!maxlen))
917 __dest = get_loc_data(dest, base);
919 ret = strncpy_from_unsafe_user(__dest, uaddr, maxlen);
921 *(u32 *)dest = make_data_loc(ret, __dest - base);
926 static nokprobe_inline int
927 probe_mem_read(void *dest, void *src, size_t size)
929 return probe_kernel_read(dest, src, size);
932 static nokprobe_inline int
933 probe_mem_read_user(void *dest, void *src, size_t size)
935 const void __user *uaddr = (__force const void __user *)src;
937 return probe_user_read(dest, uaddr, size);
940 /* Note that we don't verify it, since the code does not come from user space */
942 process_fetch_insn(struct fetch_insn *code, struct pt_regs *regs, void *dest,
948 /* 1st stage: get value from context */
951 val = regs_get_register(regs, code->param);
954 val = regs_get_kernel_stack_nth(regs, code->param);
956 case FETCH_OP_STACKP:
957 val = kernel_stack_pointer(regs);
959 case FETCH_OP_RETVAL:
960 val = regs_return_value(regs);
963 val = code->immediate;
966 val = (unsigned long)current->comm;
968 #ifdef CONFIG_HAVE_FUNCTION_ARG_ACCESS_API
970 val = regs_get_kernel_argument(regs, code->param);
973 case FETCH_NOP_SYMBOL: /* Ignore a place holder */
981 return process_fetch_insn_bottom(code, val, dest, base);
983 NOKPROBE_SYMBOL(process_fetch_insn)
986 static nokprobe_inline void
987 __kprobe_trace_func(struct trace_kprobe *tk, struct pt_regs *regs,
988 struct trace_event_file *trace_file)
990 struct kprobe_trace_entry_head *entry;
991 struct ring_buffer_event *event;
992 struct ring_buffer *buffer;
994 unsigned long irq_flags;
995 struct trace_event_call *call = trace_probe_event_call(&tk->tp);
997 WARN_ON(call != trace_file->event_call);
999 if (trace_trigger_soft_disabled(trace_file))
1002 local_save_flags(irq_flags);
1003 pc = preempt_count();
1005 dsize = __get_data_size(&tk->tp, regs);
1006 size = sizeof(*entry) + tk->tp.size + dsize;
1008 event = trace_event_buffer_lock_reserve(&buffer, trace_file,
1010 size, irq_flags, pc);
1014 entry = ring_buffer_event_data(event);
1015 entry->ip = (unsigned long)tk->rp.kp.addr;
1016 store_trace_args(&entry[1], &tk->tp, regs, sizeof(*entry), dsize);
1018 event_trigger_unlock_commit_regs(trace_file, buffer, event,
1019 entry, irq_flags, pc, regs);
1023 kprobe_trace_func(struct trace_kprobe *tk, struct pt_regs *regs)
1025 struct event_file_link *link;
1027 trace_probe_for_each_link_rcu(link, &tk->tp)
1028 __kprobe_trace_func(tk, regs, link->file);
1030 NOKPROBE_SYMBOL(kprobe_trace_func);
1032 /* Kretprobe handler */
1033 static nokprobe_inline void
1034 __kretprobe_trace_func(struct trace_kprobe *tk, struct kretprobe_instance *ri,
1035 struct pt_regs *regs,
1036 struct trace_event_file *trace_file)
1038 struct kretprobe_trace_entry_head *entry;
1039 struct ring_buffer_event *event;
1040 struct ring_buffer *buffer;
1041 int size, pc, dsize;
1042 unsigned long irq_flags;
1043 struct trace_event_call *call = trace_probe_event_call(&tk->tp);
1045 WARN_ON(call != trace_file->event_call);
1047 if (trace_trigger_soft_disabled(trace_file))
1050 local_save_flags(irq_flags);
1051 pc = preempt_count();
1053 dsize = __get_data_size(&tk->tp, regs);
1054 size = sizeof(*entry) + tk->tp.size + dsize;
1056 event = trace_event_buffer_lock_reserve(&buffer, trace_file,
1058 size, irq_flags, pc);
1062 entry = ring_buffer_event_data(event);
1063 entry->func = (unsigned long)tk->rp.kp.addr;
1064 entry->ret_ip = (unsigned long)ri->ret_addr;
1065 store_trace_args(&entry[1], &tk->tp, regs, sizeof(*entry), dsize);
1067 event_trigger_unlock_commit_regs(trace_file, buffer, event,
1068 entry, irq_flags, pc, regs);
1072 kretprobe_trace_func(struct trace_kprobe *tk, struct kretprobe_instance *ri,
1073 struct pt_regs *regs)
1075 struct event_file_link *link;
1077 trace_probe_for_each_link_rcu(link, &tk->tp)
1078 __kretprobe_trace_func(tk, ri, regs, link->file);
1080 NOKPROBE_SYMBOL(kretprobe_trace_func);
1082 /* Event entry printers */
1083 static enum print_line_t
1084 print_kprobe_event(struct trace_iterator *iter, int flags,
1085 struct trace_event *event)
1087 struct kprobe_trace_entry_head *field;
1088 struct trace_seq *s = &iter->seq;
1089 struct trace_probe *tp;
1091 field = (struct kprobe_trace_entry_head *)iter->ent;
1092 tp = container_of(event, struct trace_probe, call.event);
1094 trace_seq_printf(s, "%s: (", trace_probe_name(tp));
1096 if (!seq_print_ip_sym(s, field->ip, flags | TRACE_ITER_SYM_OFFSET))
1099 trace_seq_putc(s, ')');
1101 if (print_probe_args(s, tp->args, tp->nr_args,
1102 (u8 *)&field[1], field) < 0)
1105 trace_seq_putc(s, '\n');
1107 return trace_handle_return(s);
1110 static enum print_line_t
1111 print_kretprobe_event(struct trace_iterator *iter, int flags,
1112 struct trace_event *event)
1114 struct kretprobe_trace_entry_head *field;
1115 struct trace_seq *s = &iter->seq;
1116 struct trace_probe *tp;
1118 field = (struct kretprobe_trace_entry_head *)iter->ent;
1119 tp = container_of(event, struct trace_probe, call.event);
1121 trace_seq_printf(s, "%s: (", trace_probe_name(tp));
1123 if (!seq_print_ip_sym(s, field->ret_ip, flags | TRACE_ITER_SYM_OFFSET))
1126 trace_seq_puts(s, " <- ");
1128 if (!seq_print_ip_sym(s, field->func, flags & ~TRACE_ITER_SYM_OFFSET))
1131 trace_seq_putc(s, ')');
1133 if (print_probe_args(s, tp->args, tp->nr_args,
1134 (u8 *)&field[1], field) < 0)
1137 trace_seq_putc(s, '\n');
1140 return trace_handle_return(s);
1144 static int kprobe_event_define_fields(struct trace_event_call *event_call)
1147 struct kprobe_trace_entry_head field;
1148 struct trace_kprobe *tk = (struct trace_kprobe *)event_call->data;
1150 DEFINE_FIELD(unsigned long, ip, FIELD_STRING_IP, 0);
1152 return traceprobe_define_arg_fields(event_call, sizeof(field), &tk->tp);
1155 static int kretprobe_event_define_fields(struct trace_event_call *event_call)
1158 struct kretprobe_trace_entry_head field;
1159 struct trace_kprobe *tk = (struct trace_kprobe *)event_call->data;
1161 DEFINE_FIELD(unsigned long, func, FIELD_STRING_FUNC, 0);
1162 DEFINE_FIELD(unsigned long, ret_ip, FIELD_STRING_RETIP, 0);
1164 return traceprobe_define_arg_fields(event_call, sizeof(field), &tk->tp);
1167 #ifdef CONFIG_PERF_EVENTS
1169 /* Kprobe profile handler */
1171 kprobe_perf_func(struct trace_kprobe *tk, struct pt_regs *regs)
1173 struct trace_event_call *call = trace_probe_event_call(&tk->tp);
1174 struct kprobe_trace_entry_head *entry;
1175 struct hlist_head *head;
1176 int size, __size, dsize;
1179 if (bpf_prog_array_valid(call)) {
1180 unsigned long orig_ip = instruction_pointer(regs);
1183 ret = trace_call_bpf(call, regs);
1186 * We need to check and see if we modified the pc of the
1187 * pt_regs, and if so return 1 so that we don't do the
1190 if (orig_ip != instruction_pointer(regs))
1196 head = this_cpu_ptr(call->perf_events);
1197 if (hlist_empty(head))
1200 dsize = __get_data_size(&tk->tp, regs);
1201 __size = sizeof(*entry) + tk->tp.size + dsize;
1202 size = ALIGN(__size + sizeof(u32), sizeof(u64));
1203 size -= sizeof(u32);
1205 entry = perf_trace_buf_alloc(size, NULL, &rctx);
1209 entry->ip = (unsigned long)tk->rp.kp.addr;
1210 memset(&entry[1], 0, dsize);
1211 store_trace_args(&entry[1], &tk->tp, regs, sizeof(*entry), dsize);
1212 perf_trace_buf_submit(entry, size, rctx, call->event.type, 1, regs,
1216 NOKPROBE_SYMBOL(kprobe_perf_func);
1218 /* Kretprobe profile handler */
1220 kretprobe_perf_func(struct trace_kprobe *tk, struct kretprobe_instance *ri,
1221 struct pt_regs *regs)
1223 struct trace_event_call *call = trace_probe_event_call(&tk->tp);
1224 struct kretprobe_trace_entry_head *entry;
1225 struct hlist_head *head;
1226 int size, __size, dsize;
1229 if (bpf_prog_array_valid(call) && !trace_call_bpf(call, regs))
1232 head = this_cpu_ptr(call->perf_events);
1233 if (hlist_empty(head))
1236 dsize = __get_data_size(&tk->tp, regs);
1237 __size = sizeof(*entry) + tk->tp.size + dsize;
1238 size = ALIGN(__size + sizeof(u32), sizeof(u64));
1239 size -= sizeof(u32);
1241 entry = perf_trace_buf_alloc(size, NULL, &rctx);
1245 entry->func = (unsigned long)tk->rp.kp.addr;
1246 entry->ret_ip = (unsigned long)ri->ret_addr;
1247 store_trace_args(&entry[1], &tk->tp, regs, sizeof(*entry), dsize);
1248 perf_trace_buf_submit(entry, size, rctx, call->event.type, 1, regs,
1251 NOKPROBE_SYMBOL(kretprobe_perf_func);
1253 int bpf_get_kprobe_info(const struct perf_event *event, u32 *fd_type,
1254 const char **symbol, u64 *probe_offset,
1255 u64 *probe_addr, bool perf_type_tracepoint)
1257 const char *pevent = trace_event_name(event->tp_event);
1258 const char *group = event->tp_event->class->system;
1259 struct trace_kprobe *tk;
1261 if (perf_type_tracepoint)
1262 tk = find_trace_kprobe(pevent, group);
1264 tk = event->tp_event->data;
1268 *fd_type = trace_kprobe_is_return(tk) ? BPF_FD_TYPE_KRETPROBE
1269 : BPF_FD_TYPE_KPROBE;
1271 *symbol = tk->symbol;
1272 *probe_offset = tk->rp.kp.offset;
1277 *probe_addr = (unsigned long)tk->rp.kp.addr;
1281 #endif /* CONFIG_PERF_EVENTS */
1284 * called by perf_trace_init() or __ftrace_set_clr_event() under event_mutex.
1286 * kprobe_trace_self_tests_init() does enable_trace_probe/disable_trace_probe
1287 * lockless, but we can't race with this __init function.
1289 static int kprobe_register(struct trace_event_call *event,
1290 enum trace_reg type, void *data)
1292 struct trace_kprobe *tk = (struct trace_kprobe *)event->data;
1293 struct trace_event_file *file = data;
1296 case TRACE_REG_REGISTER:
1297 return enable_trace_kprobe(tk, file);
1298 case TRACE_REG_UNREGISTER:
1299 return disable_trace_kprobe(tk, file);
1301 #ifdef CONFIG_PERF_EVENTS
1302 case TRACE_REG_PERF_REGISTER:
1303 return enable_trace_kprobe(tk, NULL);
1304 case TRACE_REG_PERF_UNREGISTER:
1305 return disable_trace_kprobe(tk, NULL);
1306 case TRACE_REG_PERF_OPEN:
1307 case TRACE_REG_PERF_CLOSE:
1308 case TRACE_REG_PERF_ADD:
1309 case TRACE_REG_PERF_DEL:
1316 static int kprobe_dispatcher(struct kprobe *kp, struct pt_regs *regs)
1318 struct trace_kprobe *tk = container_of(kp, struct trace_kprobe, rp.kp);
1321 raw_cpu_inc(*tk->nhit);
1323 if (trace_probe_test_flag(&tk->tp, TP_FLAG_TRACE))
1324 kprobe_trace_func(tk, regs);
1325 #ifdef CONFIG_PERF_EVENTS
1326 if (trace_probe_test_flag(&tk->tp, TP_FLAG_PROFILE))
1327 ret = kprobe_perf_func(tk, regs);
1331 NOKPROBE_SYMBOL(kprobe_dispatcher);
1334 kretprobe_dispatcher(struct kretprobe_instance *ri, struct pt_regs *regs)
1336 struct trace_kprobe *tk = container_of(ri->rp, struct trace_kprobe, rp);
1338 raw_cpu_inc(*tk->nhit);
1340 if (trace_probe_test_flag(&tk->tp, TP_FLAG_TRACE))
1341 kretprobe_trace_func(tk, ri, regs);
1342 #ifdef CONFIG_PERF_EVENTS
1343 if (trace_probe_test_flag(&tk->tp, TP_FLAG_PROFILE))
1344 kretprobe_perf_func(tk, ri, regs);
1346 return 0; /* We don't tweek kernel, so just return 0 */
1348 NOKPROBE_SYMBOL(kretprobe_dispatcher);
1350 static struct trace_event_functions kretprobe_funcs = {
1351 .trace = print_kretprobe_event
1354 static struct trace_event_functions kprobe_funcs = {
1355 .trace = print_kprobe_event
1358 static inline void init_trace_event_call(struct trace_kprobe *tk)
1360 struct trace_event_call *call = trace_probe_event_call(&tk->tp);
1362 if (trace_kprobe_is_return(tk)) {
1363 call->event.funcs = &kretprobe_funcs;
1364 call->class->define_fields = kretprobe_event_define_fields;
1366 call->event.funcs = &kprobe_funcs;
1367 call->class->define_fields = kprobe_event_define_fields;
1370 call->flags = TRACE_EVENT_FL_KPROBE;
1371 call->class->reg = kprobe_register;
1375 static int register_kprobe_event(struct trace_kprobe *tk)
1377 init_trace_event_call(tk);
1379 return trace_probe_register_event_call(&tk->tp);
1382 static int unregister_kprobe_event(struct trace_kprobe *tk)
1384 return trace_probe_unregister_event_call(&tk->tp);
1387 #ifdef CONFIG_PERF_EVENTS
1388 /* create a trace_kprobe, but don't add it to global lists */
1389 struct trace_event_call *
1390 create_local_trace_kprobe(char *func, void *addr, unsigned long offs,
1393 struct trace_kprobe *tk;
1398 * local trace_kprobes are not added to dyn_event, so they are never
1399 * searched in find_trace_kprobe(). Therefore, there is no concern of
1400 * duplicated name here.
1402 event = func ? func : "DUMMY_EVENT";
1404 tk = alloc_trace_kprobe(KPROBE_EVENT_SYSTEM, event, (void *)addr, func,
1405 offs, 0 /* maxactive */, 0 /* nargs */,
1409 pr_info("Failed to allocate trace_probe.(%d)\n",
1411 return ERR_CAST(tk);
1414 init_trace_event_call(tk);
1416 if (traceprobe_set_print_fmt(&tk->tp, trace_kprobe_is_return(tk)) < 0) {
1421 ret = __register_trace_kprobe(tk);
1425 return trace_probe_event_call(&tk->tp);
1427 free_trace_kprobe(tk);
1428 return ERR_PTR(ret);
1431 void destroy_local_trace_kprobe(struct trace_event_call *event_call)
1433 struct trace_kprobe *tk;
1435 tk = container_of(event_call, struct trace_kprobe, tp.call);
1437 if (trace_probe_is_enabled(&tk->tp)) {
1442 __unregister_trace_kprobe(tk);
1444 free_trace_kprobe(tk);
1446 #endif /* CONFIG_PERF_EVENTS */
1448 static __init void enable_boot_kprobe_events(void)
1450 struct trace_array *tr = top_trace_array();
1451 struct trace_event_file *file;
1452 struct trace_kprobe *tk;
1453 struct dyn_event *pos;
1455 mutex_lock(&event_mutex);
1456 for_each_trace_kprobe(tk, pos) {
1457 list_for_each_entry(file, &tr->events, list)
1458 if (file->event_call == trace_probe_event_call(&tk->tp))
1459 trace_event_enable_disable(file, 1, 0);
1461 mutex_unlock(&event_mutex);
1464 static __init void setup_boot_kprobe_events(void)
1466 char *p, *cmd = kprobe_boot_events_buf;
1469 strreplace(kprobe_boot_events_buf, ',', ' ');
1471 while (cmd && *cmd != '\0') {
1472 p = strchr(cmd, ';');
1476 ret = trace_run_command(cmd, create_or_delete_trace_kprobe);
1478 pr_warn("Failed to add event(%d): %s\n", ret, cmd);
1480 kprobe_boot_events_enabled = true;
1485 enable_boot_kprobe_events();
1488 /* Make a tracefs interface for controlling probe points */
1489 static __init int init_kprobe_trace(void)
1491 struct dentry *d_tracer;
1492 struct dentry *entry;
1495 ret = dyn_event_register(&trace_kprobe_ops);
1499 if (register_module_notifier(&trace_kprobe_module_nb))
1502 d_tracer = tracing_init_dentry();
1503 if (IS_ERR(d_tracer))
1506 entry = tracefs_create_file("kprobe_events", 0644, d_tracer,
1507 NULL, &kprobe_events_ops);
1509 /* Event list interface */
1511 pr_warn("Could not create tracefs 'kprobe_events' entry\n");
1513 /* Profile interface */
1514 entry = tracefs_create_file("kprobe_profile", 0444, d_tracer,
1515 NULL, &kprobe_profile_ops);
1518 pr_warn("Could not create tracefs 'kprobe_profile' entry\n");
1520 setup_boot_kprobe_events();
1524 fs_initcall(init_kprobe_trace);
1527 #ifdef CONFIG_FTRACE_STARTUP_TEST
1528 static __init struct trace_event_file *
1529 find_trace_probe_file(struct trace_kprobe *tk, struct trace_array *tr)
1531 struct trace_event_file *file;
1533 list_for_each_entry(file, &tr->events, list)
1534 if (file->event_call == trace_probe_event_call(&tk->tp))
1541 * Nobody but us can call enable_trace_kprobe/disable_trace_kprobe at this
1542 * stage, we can do this lockless.
1544 static __init int kprobe_trace_self_tests_init(void)
1547 int (*target)(int, int, int, int, int, int);
1548 struct trace_kprobe *tk;
1549 struct trace_event_file *file;
1551 if (tracing_is_disabled())
1554 if (kprobe_boot_events_enabled) {
1555 pr_info("Skipping kprobe tests due to kprobe_event on cmdline\n");
1559 target = kprobe_trace_selftest_target;
1561 pr_info("Testing kprobe tracing: ");
1563 ret = trace_run_command("p:testprobe kprobe_trace_selftest_target $stack $stack0 +0($stack)",
1564 create_or_delete_trace_kprobe);
1565 if (WARN_ON_ONCE(ret)) {
1566 pr_warn("error on probing function entry.\n");
1569 /* Enable trace point */
1570 tk = find_trace_kprobe("testprobe", KPROBE_EVENT_SYSTEM);
1571 if (WARN_ON_ONCE(tk == NULL)) {
1572 pr_warn("error on getting new probe.\n");
1575 file = find_trace_probe_file(tk, top_trace_array());
1576 if (WARN_ON_ONCE(file == NULL)) {
1577 pr_warn("error on getting probe file.\n");
1580 enable_trace_kprobe(tk, file);
1584 ret = trace_run_command("r:testprobe2 kprobe_trace_selftest_target $retval",
1585 create_or_delete_trace_kprobe);
1586 if (WARN_ON_ONCE(ret)) {
1587 pr_warn("error on probing function return.\n");
1590 /* Enable trace point */
1591 tk = find_trace_kprobe("testprobe2", KPROBE_EVENT_SYSTEM);
1592 if (WARN_ON_ONCE(tk == NULL)) {
1593 pr_warn("error on getting 2nd new probe.\n");
1596 file = find_trace_probe_file(tk, top_trace_array());
1597 if (WARN_ON_ONCE(file == NULL)) {
1598 pr_warn("error on getting probe file.\n");
1601 enable_trace_kprobe(tk, file);
1608 ret = target(1, 2, 3, 4, 5, 6);
1611 * Not expecting an error here, the check is only to prevent the
1612 * optimizer from removing the call to target() as otherwise there
1613 * are no side-effects and the call is never performed.
1618 /* Disable trace points before removing it */
1619 tk = find_trace_kprobe("testprobe", KPROBE_EVENT_SYSTEM);
1620 if (WARN_ON_ONCE(tk == NULL)) {
1621 pr_warn("error on getting test probe.\n");
1624 if (trace_kprobe_nhit(tk) != 1) {
1625 pr_warn("incorrect number of testprobe hits\n");
1629 file = find_trace_probe_file(tk, top_trace_array());
1630 if (WARN_ON_ONCE(file == NULL)) {
1631 pr_warn("error on getting probe file.\n");
1634 disable_trace_kprobe(tk, file);
1637 tk = find_trace_kprobe("testprobe2", KPROBE_EVENT_SYSTEM);
1638 if (WARN_ON_ONCE(tk == NULL)) {
1639 pr_warn("error on getting 2nd test probe.\n");
1642 if (trace_kprobe_nhit(tk) != 1) {
1643 pr_warn("incorrect number of testprobe2 hits\n");
1647 file = find_trace_probe_file(tk, top_trace_array());
1648 if (WARN_ON_ONCE(file == NULL)) {
1649 pr_warn("error on getting probe file.\n");
1652 disable_trace_kprobe(tk, file);
1655 ret = trace_run_command("-:testprobe", create_or_delete_trace_kprobe);
1656 if (WARN_ON_ONCE(ret)) {
1657 pr_warn("error on deleting a probe.\n");
1661 ret = trace_run_command("-:testprobe2", create_or_delete_trace_kprobe);
1662 if (WARN_ON_ONCE(ret)) {
1663 pr_warn("error on deleting a probe.\n");
1668 ret = dyn_events_release_all(&trace_kprobe_ops);
1669 if (WARN_ON_ONCE(ret)) {
1670 pr_warn("error on cleaning up probes.\n");
1674 * Wait for the optimizer work to finish. Otherwise it might fiddle
1675 * with probes in already freed __init text.
1677 wait_for_kprobe_optimizer();
1679 pr_cont("NG: Some tests are failed. Please check them.\n");
1685 late_initcall(kprobe_trace_self_tests_init);