1 // SPDX-License-Identifier: GPL-2.0
3 * uprobes-based tracing events
5 * Copyright (C) IBM Corporation, 2010-2012
6 * Author: Srikar Dronamraju <srikar@linux.vnet.ibm.com>
8 #define pr_fmt(fmt) "trace_uprobe: " fmt
10 #include <linux/security.h>
11 #include <linux/ctype.h>
12 #include <linux/module.h>
13 #include <linux/uaccess.h>
14 #include <linux/uprobes.h>
15 #include <linux/namei.h>
16 #include <linux/string.h>
17 #include <linux/rculist.h>
19 #include "trace_dynevent.h"
20 #include "trace_probe.h"
21 #include "trace_probe_tmpl.h"
23 #define UPROBE_EVENT_SYSTEM "uprobes"
25 struct uprobe_trace_entry_head {
26 struct trace_entry ent;
27 unsigned long vaddr[];
30 #define SIZEOF_TRACE_ENTRY(is_return) \
31 (sizeof(struct uprobe_trace_entry_head) + \
32 sizeof(unsigned long) * (is_return ? 2 : 1))
34 #define DATAOF_TRACE_ENTRY(entry, is_return) \
35 ((void*)(entry) + SIZEOF_TRACE_ENTRY(is_return))
37 static int trace_uprobe_create(const char *raw_command);
38 static int trace_uprobe_show(struct seq_file *m, struct dyn_event *ev);
39 static int trace_uprobe_release(struct dyn_event *ev);
40 static bool trace_uprobe_is_busy(struct dyn_event *ev);
41 static bool trace_uprobe_match(const char *system, const char *event,
42 int argc, const char **argv, struct dyn_event *ev);
44 static struct dyn_event_operations trace_uprobe_ops = {
45 .create = trace_uprobe_create,
46 .show = trace_uprobe_show,
47 .is_busy = trace_uprobe_is_busy,
48 .free = trace_uprobe_release,
49 .match = trace_uprobe_match,
53 * uprobe event core functions
56 struct dyn_event devent;
57 struct uprobe_consumer consumer;
62 unsigned long ref_ctr_offset;
64 struct trace_probe tp;
67 static bool is_trace_uprobe(struct dyn_event *ev)
69 return ev->ops == &trace_uprobe_ops;
72 static struct trace_uprobe *to_trace_uprobe(struct dyn_event *ev)
74 return container_of(ev, struct trace_uprobe, devent);
78 * for_each_trace_uprobe - iterate over the trace_uprobe list
79 * @pos: the struct trace_uprobe * for each entry
80 * @dpos: the struct dyn_event * to use as a loop cursor
82 #define for_each_trace_uprobe(pos, dpos) \
83 for_each_dyn_event(dpos) \
84 if (is_trace_uprobe(dpos) && (pos = to_trace_uprobe(dpos)))
86 static int register_uprobe_event(struct trace_uprobe *tu);
87 static int unregister_uprobe_event(struct trace_uprobe *tu);
89 struct uprobe_dispatch_data {
90 struct trace_uprobe *tu;
91 unsigned long bp_addr;
94 static int uprobe_dispatcher(struct uprobe_consumer *con, struct pt_regs *regs);
95 static int uretprobe_dispatcher(struct uprobe_consumer *con,
96 unsigned long func, struct pt_regs *regs);
98 #ifdef CONFIG_STACK_GROWSUP
99 static unsigned long adjust_stack_addr(unsigned long addr, unsigned int n)
101 return addr - (n * sizeof(long));
104 static unsigned long adjust_stack_addr(unsigned long addr, unsigned int n)
106 return addr + (n * sizeof(long));
110 static unsigned long get_user_stack_nth(struct pt_regs *regs, unsigned int n)
113 unsigned long addr = user_stack_pointer(regs);
115 addr = adjust_stack_addr(addr, n);
117 if (copy_from_user(&ret, (void __force __user *) addr, sizeof(ret)))
124 * Uprobes-specific fetch functions
126 static nokprobe_inline int
127 probe_mem_read(void *dest, void *src, size_t size)
129 void __user *vaddr = (void __force __user *)src;
131 return copy_from_user(dest, vaddr, size) ? -EFAULT : 0;
134 static nokprobe_inline int
135 probe_mem_read_user(void *dest, void *src, size_t size)
137 return probe_mem_read(dest, src, size);
141 * Fetch a null-terminated string. Caller MUST set *(u32 *)dest with max
142 * length and relative data location.
144 static nokprobe_inline int
145 fetch_store_string(unsigned long addr, void *dest, void *base)
148 u32 loc = *(u32 *)dest;
149 int maxlen = get_loc_len(loc);
150 u8 *dst = get_loc_data(dest, base);
151 void __user *src = (void __force __user *) addr;
153 if (unlikely(!maxlen))
156 if (addr == FETCH_TOKEN_COMM)
157 ret = strlcpy(dst, current->comm, maxlen);
159 ret = strncpy_from_user(dst, src, maxlen);
165 * Include the terminating null byte. In this case it
166 * was copied by strncpy_from_user but not accounted
170 *(u32 *)dest = make_data_loc(ret, (void *)dst - base);
176 static nokprobe_inline int
177 fetch_store_string_user(unsigned long addr, void *dest, void *base)
179 return fetch_store_string(addr, dest, base);
182 /* Return the length of string -- including null terminal byte */
183 static nokprobe_inline int
184 fetch_store_strlen(unsigned long addr)
187 void __user *vaddr = (void __force __user *) addr;
189 if (addr == FETCH_TOKEN_COMM)
190 len = strlen(current->comm) + 1;
192 len = strnlen_user(vaddr, MAX_STRING_SIZE);
194 return (len > MAX_STRING_SIZE) ? 0 : len;
197 static nokprobe_inline int
198 fetch_store_strlen_user(unsigned long addr)
200 return fetch_store_strlen(addr);
203 static unsigned long translate_user_vaddr(unsigned long file_offset)
205 unsigned long base_addr;
206 struct uprobe_dispatch_data *udd;
208 udd = (void *) current->utask->vaddr;
210 base_addr = udd->bp_addr - udd->tu->offset;
211 return base_addr + file_offset;
214 /* Note that we don't verify it, since the code does not come from user space */
216 process_fetch_insn(struct fetch_insn *code, void *rec, void *dest,
219 struct pt_regs *regs = rec;
222 /* 1st stage: get value from context */
225 val = regs_get_register(regs, code->param);
228 val = get_user_stack_nth(regs, code->param);
230 case FETCH_OP_STACKP:
231 val = user_stack_pointer(regs);
233 case FETCH_OP_RETVAL:
234 val = regs_return_value(regs);
237 val = code->immediate;
240 val = FETCH_TOKEN_COMM;
243 val = (unsigned long)code->data;
246 val = translate_user_vaddr(code->immediate);
253 return process_fetch_insn_bottom(code, val, dest, base);
255 NOKPROBE_SYMBOL(process_fetch_insn)
257 static inline void init_trace_uprobe_filter(struct trace_uprobe_filter *filter)
259 rwlock_init(&filter->rwlock);
260 filter->nr_systemwide = 0;
261 INIT_LIST_HEAD(&filter->perf_events);
264 static inline bool uprobe_filter_is_empty(struct trace_uprobe_filter *filter)
266 return !filter->nr_systemwide && list_empty(&filter->perf_events);
269 static inline bool is_ret_probe(struct trace_uprobe *tu)
271 return tu->consumer.ret_handler != NULL;
274 static bool trace_uprobe_is_busy(struct dyn_event *ev)
276 struct trace_uprobe *tu = to_trace_uprobe(ev);
278 return trace_probe_is_enabled(&tu->tp);
281 static bool trace_uprobe_match_command_head(struct trace_uprobe *tu,
282 int argc, const char **argv)
284 char buf[MAX_ARGSTR_LEN + 1];
290 len = strlen(tu->filename);
291 if (strncmp(tu->filename, argv[0], len) || argv[0][len] != ':')
294 if (tu->ref_ctr_offset == 0)
295 snprintf(buf, sizeof(buf), "0x%0*lx",
296 (int)(sizeof(void *) * 2), tu->offset);
298 snprintf(buf, sizeof(buf), "0x%0*lx(0x%lx)",
299 (int)(sizeof(void *) * 2), tu->offset,
301 if (strcmp(buf, &argv[0][len + 1]))
306 return trace_probe_match_command_args(&tu->tp, argc, argv);
309 static bool trace_uprobe_match(const char *system, const char *event,
310 int argc, const char **argv, struct dyn_event *ev)
312 struct trace_uprobe *tu = to_trace_uprobe(ev);
314 return strcmp(trace_probe_name(&tu->tp), event) == 0 &&
315 (!system || strcmp(trace_probe_group_name(&tu->tp), system) == 0) &&
316 trace_uprobe_match_command_head(tu, argc, argv);
319 static nokprobe_inline struct trace_uprobe *
320 trace_uprobe_primary_from_call(struct trace_event_call *call)
322 struct trace_probe *tp;
324 tp = trace_probe_primary_from_call(call);
325 if (WARN_ON_ONCE(!tp))
328 return container_of(tp, struct trace_uprobe, tp);
332 * Allocate new trace_uprobe and initialize it (including uprobes).
334 static struct trace_uprobe *
335 alloc_trace_uprobe(const char *group, const char *event, int nargs, bool is_ret)
337 struct trace_uprobe *tu;
340 tu = kzalloc(struct_size(tu, tp.args, nargs), GFP_KERNEL);
342 return ERR_PTR(-ENOMEM);
344 ret = trace_probe_init(&tu->tp, event, group, true);
348 dyn_event_init(&tu->devent, &trace_uprobe_ops);
349 tu->consumer.handler = uprobe_dispatcher;
351 tu->consumer.ret_handler = uretprobe_dispatcher;
352 init_trace_uprobe_filter(tu->tp.event->filter);
361 static void free_trace_uprobe(struct trace_uprobe *tu)
367 trace_probe_cleanup(&tu->tp);
372 static struct trace_uprobe *find_probe_event(const char *event, const char *group)
374 struct dyn_event *pos;
375 struct trace_uprobe *tu;
377 for_each_trace_uprobe(tu, pos)
378 if (strcmp(trace_probe_name(&tu->tp), event) == 0 &&
379 strcmp(trace_probe_group_name(&tu->tp), group) == 0)
385 /* Unregister a trace_uprobe and probe_event */
386 static int unregister_trace_uprobe(struct trace_uprobe *tu)
390 if (trace_probe_has_sibling(&tu->tp))
393 /* If there's a reference to the dynamic event */
394 if (trace_event_dyn_busy(trace_probe_event_call(&tu->tp)))
397 ret = unregister_uprobe_event(tu);
402 dyn_event_remove(&tu->devent);
403 trace_probe_unlink(&tu->tp);
404 free_trace_uprobe(tu);
408 static bool trace_uprobe_has_same_uprobe(struct trace_uprobe *orig,
409 struct trace_uprobe *comp)
411 struct trace_probe_event *tpe = orig->tp.event;
412 struct trace_probe *pos;
413 struct inode *comp_inode = d_real_inode(comp->path.dentry);
416 list_for_each_entry(pos, &tpe->probes, list) {
417 orig = container_of(pos, struct trace_uprobe, tp);
418 if (comp_inode != d_real_inode(orig->path.dentry) ||
419 comp->offset != orig->offset)
423 * trace_probe_compare_arg_type() ensured that nr_args and
424 * each argument name and type are same. Let's compare comm.
426 for (i = 0; i < orig->tp.nr_args; i++) {
427 if (strcmp(orig->tp.args[i].comm,
428 comp->tp.args[i].comm))
432 if (i == orig->tp.nr_args)
439 static int append_trace_uprobe(struct trace_uprobe *tu, struct trace_uprobe *to)
443 ret = trace_probe_compare_arg_type(&tu->tp, &to->tp);
445 /* Note that argument starts index = 2 */
446 trace_probe_log_set_index(ret + 1);
447 trace_probe_log_err(0, DIFF_ARG_TYPE);
450 if (trace_uprobe_has_same_uprobe(to, tu)) {
451 trace_probe_log_set_index(0);
452 trace_probe_log_err(0, SAME_PROBE);
456 /* Append to existing event */
457 ret = trace_probe_append(&tu->tp, &to->tp);
459 dyn_event_add(&tu->devent, trace_probe_event_call(&tu->tp));
465 * Uprobe with multiple reference counter is not allowed. i.e.
466 * If inode and offset matches, reference counter offset *must*
467 * match as well. Though, there is one exception: If user is
468 * replacing old trace_uprobe with new one(same group/event),
469 * then we allow same uprobe with new reference counter as far
470 * as the new one does not conflict with any other existing
473 static int validate_ref_ctr_offset(struct trace_uprobe *new)
475 struct dyn_event *pos;
476 struct trace_uprobe *tmp;
477 struct inode *new_inode = d_real_inode(new->path.dentry);
479 for_each_trace_uprobe(tmp, pos) {
480 if (new_inode == d_real_inode(tmp->path.dentry) &&
481 new->offset == tmp->offset &&
482 new->ref_ctr_offset != tmp->ref_ctr_offset) {
483 pr_warn("Reference counter offset mismatch.");
490 /* Register a trace_uprobe and probe_event */
491 static int register_trace_uprobe(struct trace_uprobe *tu)
493 struct trace_uprobe *old_tu;
496 mutex_lock(&event_mutex);
498 ret = validate_ref_ctr_offset(tu);
502 /* register as an event */
503 old_tu = find_probe_event(trace_probe_name(&tu->tp),
504 trace_probe_group_name(&tu->tp));
506 if (is_ret_probe(tu) != is_ret_probe(old_tu)) {
507 trace_probe_log_set_index(0);
508 trace_probe_log_err(0, DIFF_PROBE_TYPE);
511 ret = append_trace_uprobe(tu, old_tu);
516 ret = register_uprobe_event(tu);
518 if (ret == -EEXIST) {
519 trace_probe_log_set_index(0);
520 trace_probe_log_err(0, EVENT_EXIST);
522 pr_warn("Failed to register probe event(%d)\n", ret);
526 dyn_event_add(&tu->devent, trace_probe_event_call(&tu->tp));
529 mutex_unlock(&event_mutex);
536 * - Add uprobe: p|r[:[GRP/]EVENT] PATH:OFFSET[%return][(REF)] [FETCHARGS]
538 static int __trace_uprobe_create(int argc, const char **argv)
540 struct trace_uprobe *tu;
541 const char *event = NULL, *group = UPROBE_EVENT_SYSTEM;
542 char *arg, *filename, *rctr, *rctr_end, *tmp;
543 char buf[MAX_EVENT_NAME_LEN];
544 enum probe_print_type ptype;
546 unsigned long offset, ref_ctr_offset;
547 bool is_return = false;
553 switch (argv[0][0]) {
566 if (argv[0][1] == ':')
569 if (!strchr(argv[1], '/'))
572 filename = kstrdup(argv[1], GFP_KERNEL);
576 /* Find the last occurrence, in case the path contains ':' too. */
577 arg = strrchr(filename, ':');
578 if (!arg || !isdigit(arg[1])) {
583 trace_probe_log_init("trace_uprobe", argc, argv);
584 trace_probe_log_set_index(1); /* filename is the 2nd argument */
587 ret = kern_path(filename, LOOKUP_FOLLOW, &path);
589 trace_probe_log_err(0, FILE_NOT_FOUND);
591 trace_probe_log_clear();
594 if (!d_is_reg(path.dentry)) {
595 trace_probe_log_err(0, NO_REGULAR_FILE);
597 goto fail_address_parse;
600 /* Parse reference counter offset if specified. */
601 rctr = strchr(arg, '(');
603 rctr_end = strchr(rctr, ')');
606 rctr_end = rctr + strlen(rctr);
607 trace_probe_log_err(rctr_end - filename,
609 goto fail_address_parse;
610 } else if (rctr_end[1] != '\0') {
612 trace_probe_log_err(rctr_end + 1 - filename,
614 goto fail_address_parse;
619 ret = kstrtoul(rctr, 0, &ref_ctr_offset);
621 trace_probe_log_err(rctr - filename, BAD_REFCNT);
622 goto fail_address_parse;
626 /* Check if there is %return suffix */
627 tmp = strchr(arg, '%');
629 if (!strcmp(tmp, "%return")) {
633 trace_probe_log_err(tmp - filename, BAD_ADDR_SUFFIX);
635 goto fail_address_parse;
639 /* Parse uprobe offset. */
640 ret = kstrtoul(arg, 0, &offset);
642 trace_probe_log_err(arg - filename, BAD_UPROBE_OFFS);
643 goto fail_address_parse;
647 trace_probe_log_set_index(0);
649 ret = traceprobe_parse_event_name(&event, &group, buf,
652 goto fail_address_parse;
657 tail = kstrdup(kbasename(filename), GFP_KERNEL);
660 goto fail_address_parse;
663 ptr = strpbrk(tail, ".-_");
667 snprintf(buf, MAX_EVENT_NAME_LEN, "%c_%s_0x%lx", 'p', tail, offset);
675 tu = alloc_trace_uprobe(group, event, argc, is_return);
678 /* This must return -ENOMEM otherwise there is a bug */
679 WARN_ON_ONCE(ret != -ENOMEM);
680 goto fail_address_parse;
683 tu->ref_ctr_offset = ref_ctr_offset;
685 tu->filename = filename;
687 /* parse arguments */
688 for (i = 0; i < argc && i < MAX_TRACE_ARGS; i++) {
689 trace_probe_log_set_index(i + 2);
690 ret = traceprobe_parse_probe_arg(&tu->tp, i, argv[i],
691 is_return ? TPARG_FL_RETURN : 0);
696 ptype = is_ret_probe(tu) ? PROBE_PRINT_RETURN : PROBE_PRINT_NORMAL;
697 ret = traceprobe_set_print_fmt(&tu->tp, ptype);
701 ret = register_trace_uprobe(tu);
706 free_trace_uprobe(tu);
708 trace_probe_log_clear();
712 trace_probe_log_clear();
719 int trace_uprobe_create(const char *raw_command)
721 return trace_probe_create(raw_command, __trace_uprobe_create);
724 static int create_or_delete_trace_uprobe(const char *raw_command)
728 if (raw_command[0] == '-')
729 return dyn_event_release(raw_command, &trace_uprobe_ops);
731 ret = trace_uprobe_create(raw_command);
732 return ret == -ECANCELED ? -EINVAL : ret;
735 static int trace_uprobe_release(struct dyn_event *ev)
737 struct trace_uprobe *tu = to_trace_uprobe(ev);
739 return unregister_trace_uprobe(tu);
742 /* Probes listing interfaces */
743 static int trace_uprobe_show(struct seq_file *m, struct dyn_event *ev)
745 struct trace_uprobe *tu = to_trace_uprobe(ev);
746 char c = is_ret_probe(tu) ? 'r' : 'p';
749 seq_printf(m, "%c:%s/%s %s:0x%0*lx", c, trace_probe_group_name(&tu->tp),
750 trace_probe_name(&tu->tp), tu->filename,
751 (int)(sizeof(void *) * 2), tu->offset);
753 if (tu->ref_ctr_offset)
754 seq_printf(m, "(0x%lx)", tu->ref_ctr_offset);
756 for (i = 0; i < tu->tp.nr_args; i++)
757 seq_printf(m, " %s=%s", tu->tp.args[i].name, tu->tp.args[i].comm);
763 static int probes_seq_show(struct seq_file *m, void *v)
765 struct dyn_event *ev = v;
767 if (!is_trace_uprobe(ev))
770 return trace_uprobe_show(m, ev);
773 static const struct seq_operations probes_seq_op = {
774 .start = dyn_event_seq_start,
775 .next = dyn_event_seq_next,
776 .stop = dyn_event_seq_stop,
777 .show = probes_seq_show
780 static int probes_open(struct inode *inode, struct file *file)
784 ret = security_locked_down(LOCKDOWN_TRACEFS);
788 if ((file->f_mode & FMODE_WRITE) && (file->f_flags & O_TRUNC)) {
789 ret = dyn_events_release_all(&trace_uprobe_ops);
794 return seq_open(file, &probes_seq_op);
797 static ssize_t probes_write(struct file *file, const char __user *buffer,
798 size_t count, loff_t *ppos)
800 return trace_parse_run_command(file, buffer, count, ppos,
801 create_or_delete_trace_uprobe);
804 static const struct file_operations uprobe_events_ops = {
805 .owner = THIS_MODULE,
809 .release = seq_release,
810 .write = probes_write,
813 /* Probes profiling interfaces */
814 static int probes_profile_seq_show(struct seq_file *m, void *v)
816 struct dyn_event *ev = v;
817 struct trace_uprobe *tu;
819 if (!is_trace_uprobe(ev))
822 tu = to_trace_uprobe(ev);
823 seq_printf(m, " %s %-44s %15lu\n", tu->filename,
824 trace_probe_name(&tu->tp), tu->nhit);
828 static const struct seq_operations profile_seq_op = {
829 .start = dyn_event_seq_start,
830 .next = dyn_event_seq_next,
831 .stop = dyn_event_seq_stop,
832 .show = probes_profile_seq_show
835 static int profile_open(struct inode *inode, struct file *file)
839 ret = security_locked_down(LOCKDOWN_TRACEFS);
843 return seq_open(file, &profile_seq_op);
846 static const struct file_operations uprobe_profile_ops = {
847 .owner = THIS_MODULE,
848 .open = profile_open,
851 .release = seq_release,
854 struct uprobe_cpu_buffer {
858 static struct uprobe_cpu_buffer __percpu *uprobe_cpu_buffer;
859 static int uprobe_buffer_refcnt;
861 static int uprobe_buffer_init(void)
865 uprobe_cpu_buffer = alloc_percpu(struct uprobe_cpu_buffer);
866 if (uprobe_cpu_buffer == NULL)
869 for_each_possible_cpu(cpu) {
870 struct page *p = alloc_pages_node(cpu_to_node(cpu),
876 per_cpu_ptr(uprobe_cpu_buffer, cpu)->buf = page_address(p);
877 mutex_init(&per_cpu_ptr(uprobe_cpu_buffer, cpu)->mutex);
883 for_each_possible_cpu(cpu) {
886 free_page((unsigned long)per_cpu_ptr(uprobe_cpu_buffer, cpu)->buf);
889 free_percpu(uprobe_cpu_buffer);
893 static int uprobe_buffer_enable(void)
897 BUG_ON(!mutex_is_locked(&event_mutex));
899 if (uprobe_buffer_refcnt++ == 0) {
900 ret = uprobe_buffer_init();
902 uprobe_buffer_refcnt--;
908 static void uprobe_buffer_disable(void)
912 BUG_ON(!mutex_is_locked(&event_mutex));
914 if (--uprobe_buffer_refcnt == 0) {
915 for_each_possible_cpu(cpu)
916 free_page((unsigned long)per_cpu_ptr(uprobe_cpu_buffer,
919 free_percpu(uprobe_cpu_buffer);
920 uprobe_cpu_buffer = NULL;
924 static struct uprobe_cpu_buffer *uprobe_buffer_get(void)
926 struct uprobe_cpu_buffer *ucb;
929 cpu = raw_smp_processor_id();
930 ucb = per_cpu_ptr(uprobe_cpu_buffer, cpu);
933 * Use per-cpu buffers for fastest access, but we might migrate
934 * so the mutex makes sure we have sole access to it.
936 mutex_lock(&ucb->mutex);
941 static void uprobe_buffer_put(struct uprobe_cpu_buffer *ucb)
943 mutex_unlock(&ucb->mutex);
946 static void __uprobe_trace_func(struct trace_uprobe *tu,
947 unsigned long func, struct pt_regs *regs,
948 struct uprobe_cpu_buffer *ucb, int dsize,
949 struct trace_event_file *trace_file)
951 struct uprobe_trace_entry_head *entry;
952 struct trace_buffer *buffer;
953 struct ring_buffer_event *event;
956 struct trace_event_call *call = trace_probe_event_call(&tu->tp);
958 WARN_ON(call != trace_file->event_call);
960 if (WARN_ON_ONCE(tu->tp.size + dsize > PAGE_SIZE))
963 if (trace_trigger_soft_disabled(trace_file))
966 esize = SIZEOF_TRACE_ENTRY(is_ret_probe(tu));
967 size = esize + tu->tp.size + dsize;
968 event = trace_event_buffer_lock_reserve(&buffer, trace_file,
969 call->event.type, size, 0);
973 entry = ring_buffer_event_data(event);
974 if (is_ret_probe(tu)) {
975 entry->vaddr[0] = func;
976 entry->vaddr[1] = instruction_pointer(regs);
977 data = DATAOF_TRACE_ENTRY(entry, true);
979 entry->vaddr[0] = instruction_pointer(regs);
980 data = DATAOF_TRACE_ENTRY(entry, false);
983 memcpy(data, ucb->buf, tu->tp.size + dsize);
985 event_trigger_unlock_commit(trace_file, buffer, event, entry, 0);
989 static int uprobe_trace_func(struct trace_uprobe *tu, struct pt_regs *regs,
990 struct uprobe_cpu_buffer *ucb, int dsize)
992 struct event_file_link *link;
994 if (is_ret_probe(tu))
998 trace_probe_for_each_link_rcu(link, &tu->tp)
999 __uprobe_trace_func(tu, 0, regs, ucb, dsize, link->file);
1005 static void uretprobe_trace_func(struct trace_uprobe *tu, unsigned long func,
1006 struct pt_regs *regs,
1007 struct uprobe_cpu_buffer *ucb, int dsize)
1009 struct event_file_link *link;
1012 trace_probe_for_each_link_rcu(link, &tu->tp)
1013 __uprobe_trace_func(tu, func, regs, ucb, dsize, link->file);
1017 /* Event entry printers */
1018 static enum print_line_t
1019 print_uprobe_event(struct trace_iterator *iter, int flags, struct trace_event *event)
1021 struct uprobe_trace_entry_head *entry;
1022 struct trace_seq *s = &iter->seq;
1023 struct trace_uprobe *tu;
1026 entry = (struct uprobe_trace_entry_head *)iter->ent;
1027 tu = trace_uprobe_primary_from_call(
1028 container_of(event, struct trace_event_call, event));
1032 if (is_ret_probe(tu)) {
1033 trace_seq_printf(s, "%s: (0x%lx <- 0x%lx)",
1034 trace_probe_name(&tu->tp),
1035 entry->vaddr[1], entry->vaddr[0]);
1036 data = DATAOF_TRACE_ENTRY(entry, true);
1038 trace_seq_printf(s, "%s: (0x%lx)",
1039 trace_probe_name(&tu->tp),
1041 data = DATAOF_TRACE_ENTRY(entry, false);
1044 if (print_probe_args(s, tu->tp.args, tu->tp.nr_args, data, entry) < 0)
1047 trace_seq_putc(s, '\n');
1050 return trace_handle_return(s);
1053 typedef bool (*filter_func_t)(struct uprobe_consumer *self,
1054 enum uprobe_filter_ctx ctx,
1055 struct mm_struct *mm);
1057 static int trace_uprobe_enable(struct trace_uprobe *tu, filter_func_t filter)
1061 tu->consumer.filter = filter;
1062 tu->inode = d_real_inode(tu->path.dentry);
1064 if (tu->ref_ctr_offset)
1065 ret = uprobe_register_refctr(tu->inode, tu->offset,
1066 tu->ref_ctr_offset, &tu->consumer);
1068 ret = uprobe_register(tu->inode, tu->offset, &tu->consumer);
1076 static void __probe_event_disable(struct trace_probe *tp)
1078 struct trace_probe *pos;
1079 struct trace_uprobe *tu;
1081 tu = container_of(tp, struct trace_uprobe, tp);
1082 WARN_ON(!uprobe_filter_is_empty(tu->tp.event->filter));
1084 list_for_each_entry(pos, trace_probe_probe_list(tp), list) {
1085 tu = container_of(pos, struct trace_uprobe, tp);
1089 uprobe_unregister(tu->inode, tu->offset, &tu->consumer);
1094 static int probe_event_enable(struct trace_event_call *call,
1095 struct trace_event_file *file, filter_func_t filter)
1097 struct trace_probe *pos, *tp;
1098 struct trace_uprobe *tu;
1102 tp = trace_probe_primary_from_call(call);
1103 if (WARN_ON_ONCE(!tp))
1105 enabled = trace_probe_is_enabled(tp);
1107 /* This may also change "enabled" state */
1109 if (trace_probe_test_flag(tp, TP_FLAG_PROFILE))
1112 ret = trace_probe_add_file(tp, file);
1116 if (trace_probe_test_flag(tp, TP_FLAG_TRACE))
1119 trace_probe_set_flag(tp, TP_FLAG_PROFILE);
1122 tu = container_of(tp, struct trace_uprobe, tp);
1123 WARN_ON(!uprobe_filter_is_empty(tu->tp.event->filter));
1128 ret = uprobe_buffer_enable();
1132 list_for_each_entry(pos, trace_probe_probe_list(tp), list) {
1133 tu = container_of(pos, struct trace_uprobe, tp);
1134 ret = trace_uprobe_enable(tu, filter);
1136 __probe_event_disable(tp);
1144 uprobe_buffer_disable();
1148 trace_probe_remove_file(tp, file);
1150 trace_probe_clear_flag(tp, TP_FLAG_PROFILE);
1155 static void probe_event_disable(struct trace_event_call *call,
1156 struct trace_event_file *file)
1158 struct trace_probe *tp;
1160 tp = trace_probe_primary_from_call(call);
1161 if (WARN_ON_ONCE(!tp))
1164 if (!trace_probe_is_enabled(tp))
1168 if (trace_probe_remove_file(tp, file) < 0)
1171 if (trace_probe_is_enabled(tp))
1174 trace_probe_clear_flag(tp, TP_FLAG_PROFILE);
1176 __probe_event_disable(tp);
1177 uprobe_buffer_disable();
1180 static int uprobe_event_define_fields(struct trace_event_call *event_call)
1183 struct uprobe_trace_entry_head field;
1184 struct trace_uprobe *tu;
1186 tu = trace_uprobe_primary_from_call(event_call);
1190 if (is_ret_probe(tu)) {
1191 DEFINE_FIELD(unsigned long, vaddr[0], FIELD_STRING_FUNC, 0);
1192 DEFINE_FIELD(unsigned long, vaddr[1], FIELD_STRING_RETIP, 0);
1193 size = SIZEOF_TRACE_ENTRY(true);
1195 DEFINE_FIELD(unsigned long, vaddr[0], FIELD_STRING_IP, 0);
1196 size = SIZEOF_TRACE_ENTRY(false);
1199 return traceprobe_define_arg_fields(event_call, size, &tu->tp);
1202 #ifdef CONFIG_PERF_EVENTS
1204 __uprobe_perf_filter(struct trace_uprobe_filter *filter, struct mm_struct *mm)
1206 struct perf_event *event;
1208 if (filter->nr_systemwide)
1211 list_for_each_entry(event, &filter->perf_events, hw.tp_list) {
1212 if (event->hw.target->mm == mm)
1220 trace_uprobe_filter_event(struct trace_uprobe_filter *filter,
1221 struct perf_event *event)
1223 return __uprobe_perf_filter(filter, event->hw.target->mm);
1226 static bool trace_uprobe_filter_remove(struct trace_uprobe_filter *filter,
1227 struct perf_event *event)
1231 write_lock(&filter->rwlock);
1232 if (event->hw.target) {
1233 list_del(&event->hw.tp_list);
1234 done = filter->nr_systemwide ||
1235 (event->hw.target->flags & PF_EXITING) ||
1236 trace_uprobe_filter_event(filter, event);
1238 filter->nr_systemwide--;
1239 done = filter->nr_systemwide;
1241 write_unlock(&filter->rwlock);
1246 /* This returns true if the filter always covers target mm */
1247 static bool trace_uprobe_filter_add(struct trace_uprobe_filter *filter,
1248 struct perf_event *event)
1252 write_lock(&filter->rwlock);
1253 if (event->hw.target) {
1255 * event->parent != NULL means copy_process(), we can avoid
1256 * uprobe_apply(). current->mm must be probed and we can rely
1257 * on dup_mmap() which preserves the already installed bp's.
1259 * attr.enable_on_exec means that exec/mmap will install the
1260 * breakpoints we need.
1262 done = filter->nr_systemwide ||
1263 event->parent || event->attr.enable_on_exec ||
1264 trace_uprobe_filter_event(filter, event);
1265 list_add(&event->hw.tp_list, &filter->perf_events);
1267 done = filter->nr_systemwide;
1268 filter->nr_systemwide++;
1270 write_unlock(&filter->rwlock);
1275 static int uprobe_perf_close(struct trace_event_call *call,
1276 struct perf_event *event)
1278 struct trace_probe *pos, *tp;
1279 struct trace_uprobe *tu;
1282 tp = trace_probe_primary_from_call(call);
1283 if (WARN_ON_ONCE(!tp))
1286 tu = container_of(tp, struct trace_uprobe, tp);
1287 if (trace_uprobe_filter_remove(tu->tp.event->filter, event))
1290 list_for_each_entry(pos, trace_probe_probe_list(tp), list) {
1291 tu = container_of(pos, struct trace_uprobe, tp);
1292 ret = uprobe_apply(tu->inode, tu->offset, &tu->consumer, false);
1300 static int uprobe_perf_open(struct trace_event_call *call,
1301 struct perf_event *event)
1303 struct trace_probe *pos, *tp;
1304 struct trace_uprobe *tu;
1307 tp = trace_probe_primary_from_call(call);
1308 if (WARN_ON_ONCE(!tp))
1311 tu = container_of(tp, struct trace_uprobe, tp);
1312 if (trace_uprobe_filter_add(tu->tp.event->filter, event))
1315 list_for_each_entry(pos, trace_probe_probe_list(tp), list) {
1316 err = uprobe_apply(tu->inode, tu->offset, &tu->consumer, true);
1318 uprobe_perf_close(call, event);
1326 static bool uprobe_perf_filter(struct uprobe_consumer *uc,
1327 enum uprobe_filter_ctx ctx, struct mm_struct *mm)
1329 struct trace_uprobe_filter *filter;
1330 struct trace_uprobe *tu;
1333 tu = container_of(uc, struct trace_uprobe, consumer);
1334 filter = tu->tp.event->filter;
1336 read_lock(&filter->rwlock);
1337 ret = __uprobe_perf_filter(filter, mm);
1338 read_unlock(&filter->rwlock);
1343 static void __uprobe_perf_func(struct trace_uprobe *tu,
1344 unsigned long func, struct pt_regs *regs,
1345 struct uprobe_cpu_buffer *ucb, int dsize)
1347 struct trace_event_call *call = trace_probe_event_call(&tu->tp);
1348 struct uprobe_trace_entry_head *entry;
1349 struct hlist_head *head;
1354 if (bpf_prog_array_valid(call)) {
1358 ret = trace_call_bpf(call, regs);
1364 esize = SIZEOF_TRACE_ENTRY(is_ret_probe(tu));
1366 size = esize + tu->tp.size + dsize;
1367 size = ALIGN(size + sizeof(u32), sizeof(u64)) - sizeof(u32);
1368 if (WARN_ONCE(size > PERF_MAX_TRACE_SIZE, "profile buffer not large enough"))
1372 head = this_cpu_ptr(call->perf_events);
1373 if (hlist_empty(head))
1376 entry = perf_trace_buf_alloc(size, NULL, &rctx);
1380 if (is_ret_probe(tu)) {
1381 entry->vaddr[0] = func;
1382 entry->vaddr[1] = instruction_pointer(regs);
1383 data = DATAOF_TRACE_ENTRY(entry, true);
1385 entry->vaddr[0] = instruction_pointer(regs);
1386 data = DATAOF_TRACE_ENTRY(entry, false);
1389 memcpy(data, ucb->buf, tu->tp.size + dsize);
1391 if (size - esize > tu->tp.size + dsize) {
1392 int len = tu->tp.size + dsize;
1394 memset(data + len, 0, size - esize - len);
1397 perf_trace_buf_submit(entry, size, rctx, call->event.type, 1, regs,
1403 /* uprobe profile handler */
1404 static int uprobe_perf_func(struct trace_uprobe *tu, struct pt_regs *regs,
1405 struct uprobe_cpu_buffer *ucb, int dsize)
1407 if (!uprobe_perf_filter(&tu->consumer, 0, current->mm))
1408 return UPROBE_HANDLER_REMOVE;
1410 if (!is_ret_probe(tu))
1411 __uprobe_perf_func(tu, 0, regs, ucb, dsize);
1415 static void uretprobe_perf_func(struct trace_uprobe *tu, unsigned long func,
1416 struct pt_regs *regs,
1417 struct uprobe_cpu_buffer *ucb, int dsize)
1419 __uprobe_perf_func(tu, func, regs, ucb, dsize);
1422 int bpf_get_uprobe_info(const struct perf_event *event, u32 *fd_type,
1423 const char **filename, u64 *probe_offset,
1424 bool perf_type_tracepoint)
1426 const char *pevent = trace_event_name(event->tp_event);
1427 const char *group = event->tp_event->class->system;
1428 struct trace_uprobe *tu;
1430 if (perf_type_tracepoint)
1431 tu = find_probe_event(pevent, group);
1433 tu = trace_uprobe_primary_from_call(event->tp_event);
1437 *fd_type = is_ret_probe(tu) ? BPF_FD_TYPE_URETPROBE
1438 : BPF_FD_TYPE_UPROBE;
1439 *filename = tu->filename;
1440 *probe_offset = tu->offset;
1443 #endif /* CONFIG_PERF_EVENTS */
1446 trace_uprobe_register(struct trace_event_call *event, enum trace_reg type,
1449 struct trace_event_file *file = data;
1452 case TRACE_REG_REGISTER:
1453 return probe_event_enable(event, file, NULL);
1455 case TRACE_REG_UNREGISTER:
1456 probe_event_disable(event, file);
1459 #ifdef CONFIG_PERF_EVENTS
1460 case TRACE_REG_PERF_REGISTER:
1461 return probe_event_enable(event, NULL, uprobe_perf_filter);
1463 case TRACE_REG_PERF_UNREGISTER:
1464 probe_event_disable(event, NULL);
1467 case TRACE_REG_PERF_OPEN:
1468 return uprobe_perf_open(event, data);
1470 case TRACE_REG_PERF_CLOSE:
1471 return uprobe_perf_close(event, data);
1479 static int uprobe_dispatcher(struct uprobe_consumer *con, struct pt_regs *regs)
1481 struct trace_uprobe *tu;
1482 struct uprobe_dispatch_data udd;
1483 struct uprobe_cpu_buffer *ucb;
1488 tu = container_of(con, struct trace_uprobe, consumer);
1492 udd.bp_addr = instruction_pointer(regs);
1494 current->utask->vaddr = (unsigned long) &udd;
1496 if (WARN_ON_ONCE(!uprobe_cpu_buffer))
1499 dsize = __get_data_size(&tu->tp, regs);
1500 esize = SIZEOF_TRACE_ENTRY(is_ret_probe(tu));
1502 ucb = uprobe_buffer_get();
1503 store_trace_args(ucb->buf, &tu->tp, regs, esize, dsize);
1505 if (trace_probe_test_flag(&tu->tp, TP_FLAG_TRACE))
1506 ret |= uprobe_trace_func(tu, regs, ucb, dsize);
1508 #ifdef CONFIG_PERF_EVENTS
1509 if (trace_probe_test_flag(&tu->tp, TP_FLAG_PROFILE))
1510 ret |= uprobe_perf_func(tu, regs, ucb, dsize);
1512 uprobe_buffer_put(ucb);
1516 static int uretprobe_dispatcher(struct uprobe_consumer *con,
1517 unsigned long func, struct pt_regs *regs)
1519 struct trace_uprobe *tu;
1520 struct uprobe_dispatch_data udd;
1521 struct uprobe_cpu_buffer *ucb;
1524 tu = container_of(con, struct trace_uprobe, consumer);
1529 current->utask->vaddr = (unsigned long) &udd;
1531 if (WARN_ON_ONCE(!uprobe_cpu_buffer))
1534 dsize = __get_data_size(&tu->tp, regs);
1535 esize = SIZEOF_TRACE_ENTRY(is_ret_probe(tu));
1537 ucb = uprobe_buffer_get();
1538 store_trace_args(ucb->buf, &tu->tp, regs, esize, dsize);
1540 if (trace_probe_test_flag(&tu->tp, TP_FLAG_TRACE))
1541 uretprobe_trace_func(tu, func, regs, ucb, dsize);
1543 #ifdef CONFIG_PERF_EVENTS
1544 if (trace_probe_test_flag(&tu->tp, TP_FLAG_PROFILE))
1545 uretprobe_perf_func(tu, func, regs, ucb, dsize);
1547 uprobe_buffer_put(ucb);
1551 static struct trace_event_functions uprobe_funcs = {
1552 .trace = print_uprobe_event
1555 static struct trace_event_fields uprobe_fields_array[] = {
1556 { .type = TRACE_FUNCTION_TYPE,
1557 .define_fields = uprobe_event_define_fields },
1561 static inline void init_trace_event_call(struct trace_uprobe *tu)
1563 struct trace_event_call *call = trace_probe_event_call(&tu->tp);
1564 call->event.funcs = &uprobe_funcs;
1565 call->class->fields_array = uprobe_fields_array;
1567 call->flags = TRACE_EVENT_FL_UPROBE | TRACE_EVENT_FL_CAP_ANY;
1568 call->class->reg = trace_uprobe_register;
1571 static int register_uprobe_event(struct trace_uprobe *tu)
1573 init_trace_event_call(tu);
1575 return trace_probe_register_event_call(&tu->tp);
1578 static int unregister_uprobe_event(struct trace_uprobe *tu)
1580 return trace_probe_unregister_event_call(&tu->tp);
1583 #ifdef CONFIG_PERF_EVENTS
1584 struct trace_event_call *
1585 create_local_trace_uprobe(char *name, unsigned long offs,
1586 unsigned long ref_ctr_offset, bool is_return)
1588 enum probe_print_type ptype;
1589 struct trace_uprobe *tu;
1593 ret = kern_path(name, LOOKUP_FOLLOW, &path);
1595 return ERR_PTR(ret);
1597 if (!d_is_reg(path.dentry)) {
1599 return ERR_PTR(-EINVAL);
1603 * local trace_kprobes are not added to dyn_event, so they are never
1604 * searched in find_trace_kprobe(). Therefore, there is no concern of
1605 * duplicated name "DUMMY_EVENT" here.
1607 tu = alloc_trace_uprobe(UPROBE_EVENT_SYSTEM, "DUMMY_EVENT", 0,
1611 pr_info("Failed to allocate trace_uprobe.(%d)\n",
1614 return ERR_CAST(tu);
1619 tu->ref_ctr_offset = ref_ctr_offset;
1620 tu->filename = kstrdup(name, GFP_KERNEL);
1621 init_trace_event_call(tu);
1623 ptype = is_ret_probe(tu) ? PROBE_PRINT_RETURN : PROBE_PRINT_NORMAL;
1624 if (traceprobe_set_print_fmt(&tu->tp, ptype) < 0) {
1629 return trace_probe_event_call(&tu->tp);
1631 free_trace_uprobe(tu);
1632 return ERR_PTR(ret);
1635 void destroy_local_trace_uprobe(struct trace_event_call *event_call)
1637 struct trace_uprobe *tu;
1639 tu = trace_uprobe_primary_from_call(event_call);
1641 free_trace_uprobe(tu);
1643 #endif /* CONFIG_PERF_EVENTS */
1645 /* Make a trace interface for controlling probe points */
1646 static __init int init_uprobe_trace(void)
1650 ret = dyn_event_register(&trace_uprobe_ops);
1654 ret = tracing_init_dentry();
1658 trace_create_file("uprobe_events", 0644, NULL,
1659 NULL, &uprobe_events_ops);
1660 /* Profile interface */
1661 trace_create_file("uprobe_profile", 0444, NULL,
1662 NULL, &uprobe_profile_ops);
1666 fs_initcall(init_uprobe_trace);