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/bpf-cgroup.h>
11 #include <linux/security.h>
12 #include <linux/ctype.h>
13 #include <linux/module.h>
14 #include <linux/uaccess.h>
15 #include <linux/uprobes.h>
16 #include <linux/namei.h>
17 #include <linux/string.h>
18 #include <linux/rculist.h>
19 #include <linux/filter.h>
21 #include "trace_dynevent.h"
22 #include "trace_probe.h"
23 #include "trace_probe_tmpl.h"
25 #define UPROBE_EVENT_SYSTEM "uprobes"
27 struct uprobe_trace_entry_head {
28 struct trace_entry ent;
29 unsigned long vaddr[];
32 #define SIZEOF_TRACE_ENTRY(is_return) \
33 (sizeof(struct uprobe_trace_entry_head) + \
34 sizeof(unsigned long) * (is_return ? 2 : 1))
36 #define DATAOF_TRACE_ENTRY(entry, is_return) \
37 ((void*)(entry) + SIZEOF_TRACE_ENTRY(is_return))
39 static int trace_uprobe_create(const char *raw_command);
40 static int trace_uprobe_show(struct seq_file *m, struct dyn_event *ev);
41 static int trace_uprobe_release(struct dyn_event *ev);
42 static bool trace_uprobe_is_busy(struct dyn_event *ev);
43 static bool trace_uprobe_match(const char *system, const char *event,
44 int argc, const char **argv, struct dyn_event *ev);
46 static struct dyn_event_operations trace_uprobe_ops = {
47 .create = trace_uprobe_create,
48 .show = trace_uprobe_show,
49 .is_busy = trace_uprobe_is_busy,
50 .free = trace_uprobe_release,
51 .match = trace_uprobe_match,
55 * uprobe event core functions
58 struct dyn_event devent;
59 struct uprobe_consumer consumer;
64 unsigned long ref_ctr_offset;
66 struct trace_probe tp;
69 static bool is_trace_uprobe(struct dyn_event *ev)
71 return ev->ops == &trace_uprobe_ops;
74 static struct trace_uprobe *to_trace_uprobe(struct dyn_event *ev)
76 return container_of(ev, struct trace_uprobe, devent);
80 * for_each_trace_uprobe - iterate over the trace_uprobe list
81 * @pos: the struct trace_uprobe * for each entry
82 * @dpos: the struct dyn_event * to use as a loop cursor
84 #define for_each_trace_uprobe(pos, dpos) \
85 for_each_dyn_event(dpos) \
86 if (is_trace_uprobe(dpos) && (pos = to_trace_uprobe(dpos)))
88 static int register_uprobe_event(struct trace_uprobe *tu);
89 static int unregister_uprobe_event(struct trace_uprobe *tu);
91 struct uprobe_dispatch_data {
92 struct trace_uprobe *tu;
93 unsigned long bp_addr;
96 static int uprobe_dispatcher(struct uprobe_consumer *con, struct pt_regs *regs);
97 static int uretprobe_dispatcher(struct uprobe_consumer *con,
98 unsigned long func, struct pt_regs *regs);
100 #ifdef CONFIG_STACK_GROWSUP
101 static unsigned long adjust_stack_addr(unsigned long addr, unsigned int n)
103 return addr - (n * sizeof(long));
106 static unsigned long adjust_stack_addr(unsigned long addr, unsigned int n)
108 return addr + (n * sizeof(long));
112 static unsigned long get_user_stack_nth(struct pt_regs *regs, unsigned int n)
115 unsigned long addr = user_stack_pointer(regs);
117 addr = adjust_stack_addr(addr, n);
119 if (copy_from_user(&ret, (void __force __user *) addr, sizeof(ret)))
126 * Uprobes-specific fetch functions
128 static nokprobe_inline int
129 probe_mem_read(void *dest, void *src, size_t size)
131 void __user *vaddr = (void __force __user *)src;
133 return copy_from_user(dest, vaddr, size) ? -EFAULT : 0;
136 static nokprobe_inline int
137 probe_mem_read_user(void *dest, void *src, size_t size)
139 return probe_mem_read(dest, src, size);
143 * Fetch a null-terminated string. Caller MUST set *(u32 *)dest with max
144 * length and relative data location.
146 static nokprobe_inline int
147 fetch_store_string(unsigned long addr, void *dest, void *base)
150 u32 loc = *(u32 *)dest;
151 int maxlen = get_loc_len(loc);
152 u8 *dst = get_loc_data(dest, base);
153 void __user *src = (void __force __user *) addr;
155 if (unlikely(!maxlen))
158 if (addr == FETCH_TOKEN_COMM)
159 ret = strlcpy(dst, current->comm, maxlen);
161 ret = strncpy_from_user(dst, src, maxlen);
167 * Include the terminating null byte. In this case it
168 * was copied by strncpy_from_user but not accounted
172 *(u32 *)dest = make_data_loc(ret, (void *)dst - base);
178 static nokprobe_inline int
179 fetch_store_string_user(unsigned long addr, void *dest, void *base)
181 return fetch_store_string(addr, dest, base);
184 /* Return the length of string -- including null terminal byte */
185 static nokprobe_inline int
186 fetch_store_strlen(unsigned long addr)
189 void __user *vaddr = (void __force __user *) addr;
191 if (addr == FETCH_TOKEN_COMM)
192 len = strlen(current->comm) + 1;
194 len = strnlen_user(vaddr, MAX_STRING_SIZE);
196 return (len > MAX_STRING_SIZE) ? 0 : len;
199 static nokprobe_inline int
200 fetch_store_strlen_user(unsigned long addr)
202 return fetch_store_strlen(addr);
205 static unsigned long translate_user_vaddr(unsigned long file_offset)
207 unsigned long base_addr;
208 struct uprobe_dispatch_data *udd;
210 udd = (void *) current->utask->vaddr;
212 base_addr = udd->bp_addr - udd->tu->offset;
213 return base_addr + file_offset;
216 /* Note that we don't verify it, since the code does not come from user space */
218 process_fetch_insn(struct fetch_insn *code, void *rec, void *dest,
221 struct pt_regs *regs = rec;
224 /* 1st stage: get value from context */
227 val = regs_get_register(regs, code->param);
230 val = get_user_stack_nth(regs, code->param);
232 case FETCH_OP_STACKP:
233 val = user_stack_pointer(regs);
235 case FETCH_OP_RETVAL:
236 val = regs_return_value(regs);
239 val = code->immediate;
242 val = FETCH_TOKEN_COMM;
245 val = (unsigned long)code->data;
248 val = translate_user_vaddr(code->immediate);
255 return process_fetch_insn_bottom(code, val, dest, base);
257 NOKPROBE_SYMBOL(process_fetch_insn)
259 static inline void init_trace_uprobe_filter(struct trace_uprobe_filter *filter)
261 rwlock_init(&filter->rwlock);
262 filter->nr_systemwide = 0;
263 INIT_LIST_HEAD(&filter->perf_events);
266 static inline bool uprobe_filter_is_empty(struct trace_uprobe_filter *filter)
268 return !filter->nr_systemwide && list_empty(&filter->perf_events);
271 static inline bool is_ret_probe(struct trace_uprobe *tu)
273 return tu->consumer.ret_handler != NULL;
276 static bool trace_uprobe_is_busy(struct dyn_event *ev)
278 struct trace_uprobe *tu = to_trace_uprobe(ev);
280 return trace_probe_is_enabled(&tu->tp);
283 static bool trace_uprobe_match_command_head(struct trace_uprobe *tu,
284 int argc, const char **argv)
286 char buf[MAX_ARGSTR_LEN + 1];
292 len = strlen(tu->filename);
293 if (strncmp(tu->filename, argv[0], len) || argv[0][len] != ':')
296 if (tu->ref_ctr_offset == 0)
297 snprintf(buf, sizeof(buf), "0x%0*lx",
298 (int)(sizeof(void *) * 2), tu->offset);
300 snprintf(buf, sizeof(buf), "0x%0*lx(0x%lx)",
301 (int)(sizeof(void *) * 2), tu->offset,
303 if (strcmp(buf, &argv[0][len + 1]))
308 return trace_probe_match_command_args(&tu->tp, argc, argv);
311 static bool trace_uprobe_match(const char *system, const char *event,
312 int argc, const char **argv, struct dyn_event *ev)
314 struct trace_uprobe *tu = to_trace_uprobe(ev);
316 return strcmp(trace_probe_name(&tu->tp), event) == 0 &&
317 (!system || strcmp(trace_probe_group_name(&tu->tp), system) == 0) &&
318 trace_uprobe_match_command_head(tu, argc, argv);
321 static nokprobe_inline struct trace_uprobe *
322 trace_uprobe_primary_from_call(struct trace_event_call *call)
324 struct trace_probe *tp;
326 tp = trace_probe_primary_from_call(call);
327 if (WARN_ON_ONCE(!tp))
330 return container_of(tp, struct trace_uprobe, tp);
334 * Allocate new trace_uprobe and initialize it (including uprobes).
336 static struct trace_uprobe *
337 alloc_trace_uprobe(const char *group, const char *event, int nargs, bool is_ret)
339 struct trace_uprobe *tu;
342 tu = kzalloc(struct_size(tu, tp.args, nargs), GFP_KERNEL);
344 return ERR_PTR(-ENOMEM);
346 ret = trace_probe_init(&tu->tp, event, group, true);
350 dyn_event_init(&tu->devent, &trace_uprobe_ops);
351 tu->consumer.handler = uprobe_dispatcher;
353 tu->consumer.ret_handler = uretprobe_dispatcher;
354 init_trace_uprobe_filter(tu->tp.event->filter);
363 static void free_trace_uprobe(struct trace_uprobe *tu)
369 trace_probe_cleanup(&tu->tp);
374 static struct trace_uprobe *find_probe_event(const char *event, const char *group)
376 struct dyn_event *pos;
377 struct trace_uprobe *tu;
379 for_each_trace_uprobe(tu, pos)
380 if (strcmp(trace_probe_name(&tu->tp), event) == 0 &&
381 strcmp(trace_probe_group_name(&tu->tp), group) == 0)
387 /* Unregister a trace_uprobe and probe_event */
388 static int unregister_trace_uprobe(struct trace_uprobe *tu)
392 if (trace_probe_has_sibling(&tu->tp))
395 /* If there's a reference to the dynamic event */
396 if (trace_event_dyn_busy(trace_probe_event_call(&tu->tp)))
399 ret = unregister_uprobe_event(tu);
404 dyn_event_remove(&tu->devent);
405 trace_probe_unlink(&tu->tp);
406 free_trace_uprobe(tu);
410 static bool trace_uprobe_has_same_uprobe(struct trace_uprobe *orig,
411 struct trace_uprobe *comp)
413 struct trace_probe_event *tpe = orig->tp.event;
414 struct inode *comp_inode = d_real_inode(comp->path.dentry);
417 list_for_each_entry(orig, &tpe->probes, tp.list) {
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;
552 switch (argv[0][0]) {
565 if (argv[0][1] == ':')
568 if (!strchr(argv[1], '/'))
571 filename = kstrdup(argv[1], GFP_KERNEL);
575 /* Find the last occurrence, in case the path contains ':' too. */
576 arg = strrchr(filename, ':');
577 if (!arg || !isdigit(arg[1])) {
582 trace_probe_log_init("trace_uprobe", argc, argv);
583 trace_probe_log_set_index(1); /* filename is the 2nd argument */
586 ret = kern_path(filename, LOOKUP_FOLLOW, &path);
588 trace_probe_log_err(0, FILE_NOT_FOUND);
590 trace_probe_log_clear();
593 if (!d_is_reg(path.dentry)) {
594 trace_probe_log_err(0, NO_REGULAR_FILE);
596 goto fail_address_parse;
599 /* Parse reference counter offset if specified. */
600 rctr = strchr(arg, '(');
602 rctr_end = strchr(rctr, ')');
605 rctr_end = rctr + strlen(rctr);
606 trace_probe_log_err(rctr_end - filename,
608 goto fail_address_parse;
609 } else if (rctr_end[1] != '\0') {
611 trace_probe_log_err(rctr_end + 1 - filename,
613 goto fail_address_parse;
618 ret = kstrtoul(rctr, 0, &ref_ctr_offset);
620 trace_probe_log_err(rctr - filename, BAD_REFCNT);
621 goto fail_address_parse;
625 /* Check if there is %return suffix */
626 tmp = strchr(arg, '%');
628 if (!strcmp(tmp, "%return")) {
632 trace_probe_log_err(tmp - filename, BAD_ADDR_SUFFIX);
634 goto fail_address_parse;
638 /* Parse uprobe offset. */
639 ret = kstrtoul(arg, 0, &offset);
641 trace_probe_log_err(arg - filename, BAD_UPROBE_OFFS);
642 goto fail_address_parse;
646 trace_probe_log_set_index(0);
648 ret = traceprobe_parse_event_name(&event, &group, buf,
651 goto fail_address_parse;
656 tail = kstrdup(kbasename(filename), GFP_KERNEL);
659 goto fail_address_parse;
662 ptr = strpbrk(tail, ".-_");
666 snprintf(buf, MAX_EVENT_NAME_LEN, "%c_%s_0x%lx", 'p', tail, offset);
674 tu = alloc_trace_uprobe(group, event, argc, is_return);
677 /* This must return -ENOMEM otherwise there is a bug */
678 WARN_ON_ONCE(ret != -ENOMEM);
679 goto fail_address_parse;
682 tu->ref_ctr_offset = ref_ctr_offset;
684 tu->filename = filename;
686 /* parse arguments */
687 for (i = 0; i < argc && i < MAX_TRACE_ARGS; i++) {
688 trace_probe_log_set_index(i + 2);
689 ret = traceprobe_parse_probe_arg(&tu->tp, i, argv[i],
690 is_return ? TPARG_FL_RETURN : 0);
695 ptype = is_ret_probe(tu) ? PROBE_PRINT_RETURN : PROBE_PRINT_NORMAL;
696 ret = traceprobe_set_print_fmt(&tu->tp, ptype);
700 ret = register_trace_uprobe(tu);
705 free_trace_uprobe(tu);
707 trace_probe_log_clear();
711 trace_probe_log_clear();
718 int trace_uprobe_create(const char *raw_command)
720 return trace_probe_create(raw_command, __trace_uprobe_create);
723 static int create_or_delete_trace_uprobe(const char *raw_command)
727 if (raw_command[0] == '-')
728 return dyn_event_release(raw_command, &trace_uprobe_ops);
730 ret = trace_uprobe_create(raw_command);
731 return ret == -ECANCELED ? -EINVAL : ret;
734 static int trace_uprobe_release(struct dyn_event *ev)
736 struct trace_uprobe *tu = to_trace_uprobe(ev);
738 return unregister_trace_uprobe(tu);
741 /* Probes listing interfaces */
742 static int trace_uprobe_show(struct seq_file *m, struct dyn_event *ev)
744 struct trace_uprobe *tu = to_trace_uprobe(ev);
745 char c = is_ret_probe(tu) ? 'r' : 'p';
748 seq_printf(m, "%c:%s/%s %s:0x%0*lx", c, trace_probe_group_name(&tu->tp),
749 trace_probe_name(&tu->tp), tu->filename,
750 (int)(sizeof(void *) * 2), tu->offset);
752 if (tu->ref_ctr_offset)
753 seq_printf(m, "(0x%lx)", tu->ref_ctr_offset);
755 for (i = 0; i < tu->tp.nr_args; i++)
756 seq_printf(m, " %s=%s", tu->tp.args[i].name, tu->tp.args[i].comm);
762 static int probes_seq_show(struct seq_file *m, void *v)
764 struct dyn_event *ev = v;
766 if (!is_trace_uprobe(ev))
769 return trace_uprobe_show(m, ev);
772 static const struct seq_operations probes_seq_op = {
773 .start = dyn_event_seq_start,
774 .next = dyn_event_seq_next,
775 .stop = dyn_event_seq_stop,
776 .show = probes_seq_show
779 static int probes_open(struct inode *inode, struct file *file)
783 ret = security_locked_down(LOCKDOWN_TRACEFS);
787 if ((file->f_mode & FMODE_WRITE) && (file->f_flags & O_TRUNC)) {
788 ret = dyn_events_release_all(&trace_uprobe_ops);
793 return seq_open(file, &probes_seq_op);
796 static ssize_t probes_write(struct file *file, const char __user *buffer,
797 size_t count, loff_t *ppos)
799 return trace_parse_run_command(file, buffer, count, ppos,
800 create_or_delete_trace_uprobe);
803 static const struct file_operations uprobe_events_ops = {
804 .owner = THIS_MODULE,
808 .release = seq_release,
809 .write = probes_write,
812 /* Probes profiling interfaces */
813 static int probes_profile_seq_show(struct seq_file *m, void *v)
815 struct dyn_event *ev = v;
816 struct trace_uprobe *tu;
818 if (!is_trace_uprobe(ev))
821 tu = to_trace_uprobe(ev);
822 seq_printf(m, " %s %-44s %15lu\n", tu->filename,
823 trace_probe_name(&tu->tp), tu->nhit);
827 static const struct seq_operations profile_seq_op = {
828 .start = dyn_event_seq_start,
829 .next = dyn_event_seq_next,
830 .stop = dyn_event_seq_stop,
831 .show = probes_profile_seq_show
834 static int profile_open(struct inode *inode, struct file *file)
838 ret = security_locked_down(LOCKDOWN_TRACEFS);
842 return seq_open(file, &profile_seq_op);
845 static const struct file_operations uprobe_profile_ops = {
846 .owner = THIS_MODULE,
847 .open = profile_open,
850 .release = seq_release,
853 struct uprobe_cpu_buffer {
857 static struct uprobe_cpu_buffer __percpu *uprobe_cpu_buffer;
858 static int uprobe_buffer_refcnt;
860 static int uprobe_buffer_init(void)
864 uprobe_cpu_buffer = alloc_percpu(struct uprobe_cpu_buffer);
865 if (uprobe_cpu_buffer == NULL)
868 for_each_possible_cpu(cpu) {
869 struct page *p = alloc_pages_node(cpu_to_node(cpu),
875 per_cpu_ptr(uprobe_cpu_buffer, cpu)->buf = page_address(p);
876 mutex_init(&per_cpu_ptr(uprobe_cpu_buffer, cpu)->mutex);
882 for_each_possible_cpu(cpu) {
885 free_page((unsigned long)per_cpu_ptr(uprobe_cpu_buffer, cpu)->buf);
888 free_percpu(uprobe_cpu_buffer);
892 static int uprobe_buffer_enable(void)
896 BUG_ON(!mutex_is_locked(&event_mutex));
898 if (uprobe_buffer_refcnt++ == 0) {
899 ret = uprobe_buffer_init();
901 uprobe_buffer_refcnt--;
907 static void uprobe_buffer_disable(void)
911 BUG_ON(!mutex_is_locked(&event_mutex));
913 if (--uprobe_buffer_refcnt == 0) {
914 for_each_possible_cpu(cpu)
915 free_page((unsigned long)per_cpu_ptr(uprobe_cpu_buffer,
918 free_percpu(uprobe_cpu_buffer);
919 uprobe_cpu_buffer = NULL;
923 static struct uprobe_cpu_buffer *uprobe_buffer_get(void)
925 struct uprobe_cpu_buffer *ucb;
928 cpu = raw_smp_processor_id();
929 ucb = per_cpu_ptr(uprobe_cpu_buffer, cpu);
932 * Use per-cpu buffers for fastest access, but we might migrate
933 * so the mutex makes sure we have sole access to it.
935 mutex_lock(&ucb->mutex);
940 static void uprobe_buffer_put(struct uprobe_cpu_buffer *ucb)
942 mutex_unlock(&ucb->mutex);
945 static void __uprobe_trace_func(struct trace_uprobe *tu,
946 unsigned long func, struct pt_regs *regs,
947 struct uprobe_cpu_buffer *ucb, int dsize,
948 struct trace_event_file *trace_file)
950 struct uprobe_trace_entry_head *entry;
951 struct trace_event_buffer fbuffer;
954 struct trace_event_call *call = trace_probe_event_call(&tu->tp);
956 WARN_ON(call != trace_file->event_call);
958 if (WARN_ON_ONCE(tu->tp.size + dsize > PAGE_SIZE))
961 if (trace_trigger_soft_disabled(trace_file))
964 esize = SIZEOF_TRACE_ENTRY(is_ret_probe(tu));
965 size = esize + tu->tp.size + dsize;
966 entry = trace_event_buffer_reserve(&fbuffer, trace_file, size);
970 if (is_ret_probe(tu)) {
971 entry->vaddr[0] = func;
972 entry->vaddr[1] = instruction_pointer(regs);
973 data = DATAOF_TRACE_ENTRY(entry, true);
975 entry->vaddr[0] = instruction_pointer(regs);
976 data = DATAOF_TRACE_ENTRY(entry, false);
979 memcpy(data, ucb->buf, tu->tp.size + dsize);
981 trace_event_buffer_commit(&fbuffer);
985 static int uprobe_trace_func(struct trace_uprobe *tu, struct pt_regs *regs,
986 struct uprobe_cpu_buffer *ucb, int dsize)
988 struct event_file_link *link;
990 if (is_ret_probe(tu))
994 trace_probe_for_each_link_rcu(link, &tu->tp)
995 __uprobe_trace_func(tu, 0, regs, ucb, dsize, link->file);
1001 static void uretprobe_trace_func(struct trace_uprobe *tu, unsigned long func,
1002 struct pt_regs *regs,
1003 struct uprobe_cpu_buffer *ucb, int dsize)
1005 struct event_file_link *link;
1008 trace_probe_for_each_link_rcu(link, &tu->tp)
1009 __uprobe_trace_func(tu, func, regs, ucb, dsize, link->file);
1013 /* Event entry printers */
1014 static enum print_line_t
1015 print_uprobe_event(struct trace_iterator *iter, int flags, struct trace_event *event)
1017 struct uprobe_trace_entry_head *entry;
1018 struct trace_seq *s = &iter->seq;
1019 struct trace_uprobe *tu;
1022 entry = (struct uprobe_trace_entry_head *)iter->ent;
1023 tu = trace_uprobe_primary_from_call(
1024 container_of(event, struct trace_event_call, event));
1028 if (is_ret_probe(tu)) {
1029 trace_seq_printf(s, "%s: (0x%lx <- 0x%lx)",
1030 trace_probe_name(&tu->tp),
1031 entry->vaddr[1], entry->vaddr[0]);
1032 data = DATAOF_TRACE_ENTRY(entry, true);
1034 trace_seq_printf(s, "%s: (0x%lx)",
1035 trace_probe_name(&tu->tp),
1037 data = DATAOF_TRACE_ENTRY(entry, false);
1040 if (print_probe_args(s, tu->tp.args, tu->tp.nr_args, data, entry) < 0)
1043 trace_seq_putc(s, '\n');
1046 return trace_handle_return(s);
1049 typedef bool (*filter_func_t)(struct uprobe_consumer *self,
1050 enum uprobe_filter_ctx ctx,
1051 struct mm_struct *mm);
1053 static int trace_uprobe_enable(struct trace_uprobe *tu, filter_func_t filter)
1057 tu->consumer.filter = filter;
1058 tu->inode = d_real_inode(tu->path.dentry);
1060 if (tu->ref_ctr_offset)
1061 ret = uprobe_register_refctr(tu->inode, tu->offset,
1062 tu->ref_ctr_offset, &tu->consumer);
1064 ret = uprobe_register(tu->inode, tu->offset, &tu->consumer);
1072 static void __probe_event_disable(struct trace_probe *tp)
1074 struct trace_uprobe *tu;
1076 tu = container_of(tp, struct trace_uprobe, tp);
1077 WARN_ON(!uprobe_filter_is_empty(tu->tp.event->filter));
1079 list_for_each_entry(tu, trace_probe_probe_list(tp), tp.list) {
1083 uprobe_unregister(tu->inode, tu->offset, &tu->consumer);
1088 static int probe_event_enable(struct trace_event_call *call,
1089 struct trace_event_file *file, filter_func_t filter)
1091 struct trace_probe *tp;
1092 struct trace_uprobe *tu;
1096 tp = trace_probe_primary_from_call(call);
1097 if (WARN_ON_ONCE(!tp))
1099 enabled = trace_probe_is_enabled(tp);
1101 /* This may also change "enabled" state */
1103 if (trace_probe_test_flag(tp, TP_FLAG_PROFILE))
1106 ret = trace_probe_add_file(tp, file);
1110 if (trace_probe_test_flag(tp, TP_FLAG_TRACE))
1113 trace_probe_set_flag(tp, TP_FLAG_PROFILE);
1116 tu = container_of(tp, struct trace_uprobe, tp);
1117 WARN_ON(!uprobe_filter_is_empty(tu->tp.event->filter));
1122 ret = uprobe_buffer_enable();
1126 list_for_each_entry(tu, trace_probe_probe_list(tp), tp.list) {
1127 ret = trace_uprobe_enable(tu, filter);
1129 __probe_event_disable(tp);
1137 uprobe_buffer_disable();
1141 trace_probe_remove_file(tp, file);
1143 trace_probe_clear_flag(tp, TP_FLAG_PROFILE);
1148 static void probe_event_disable(struct trace_event_call *call,
1149 struct trace_event_file *file)
1151 struct trace_probe *tp;
1153 tp = trace_probe_primary_from_call(call);
1154 if (WARN_ON_ONCE(!tp))
1157 if (!trace_probe_is_enabled(tp))
1161 if (trace_probe_remove_file(tp, file) < 0)
1164 if (trace_probe_is_enabled(tp))
1167 trace_probe_clear_flag(tp, TP_FLAG_PROFILE);
1169 __probe_event_disable(tp);
1170 uprobe_buffer_disable();
1173 static int uprobe_event_define_fields(struct trace_event_call *event_call)
1176 struct uprobe_trace_entry_head field;
1177 struct trace_uprobe *tu;
1179 tu = trace_uprobe_primary_from_call(event_call);
1183 if (is_ret_probe(tu)) {
1184 DEFINE_FIELD(unsigned long, vaddr[0], FIELD_STRING_FUNC, 0);
1185 DEFINE_FIELD(unsigned long, vaddr[1], FIELD_STRING_RETIP, 0);
1186 size = SIZEOF_TRACE_ENTRY(true);
1188 DEFINE_FIELD(unsigned long, vaddr[0], FIELD_STRING_IP, 0);
1189 size = SIZEOF_TRACE_ENTRY(false);
1192 return traceprobe_define_arg_fields(event_call, size, &tu->tp);
1195 #ifdef CONFIG_PERF_EVENTS
1197 __uprobe_perf_filter(struct trace_uprobe_filter *filter, struct mm_struct *mm)
1199 struct perf_event *event;
1201 if (filter->nr_systemwide)
1204 list_for_each_entry(event, &filter->perf_events, hw.tp_list) {
1205 if (event->hw.target->mm == mm)
1213 trace_uprobe_filter_event(struct trace_uprobe_filter *filter,
1214 struct perf_event *event)
1216 return __uprobe_perf_filter(filter, event->hw.target->mm);
1219 static bool trace_uprobe_filter_remove(struct trace_uprobe_filter *filter,
1220 struct perf_event *event)
1224 write_lock(&filter->rwlock);
1225 if (event->hw.target) {
1226 list_del(&event->hw.tp_list);
1227 done = filter->nr_systemwide ||
1228 (event->hw.target->flags & PF_EXITING) ||
1229 trace_uprobe_filter_event(filter, event);
1231 filter->nr_systemwide--;
1232 done = filter->nr_systemwide;
1234 write_unlock(&filter->rwlock);
1239 /* This returns true if the filter always covers target mm */
1240 static bool trace_uprobe_filter_add(struct trace_uprobe_filter *filter,
1241 struct perf_event *event)
1245 write_lock(&filter->rwlock);
1246 if (event->hw.target) {
1248 * event->parent != NULL means copy_process(), we can avoid
1249 * uprobe_apply(). current->mm must be probed and we can rely
1250 * on dup_mmap() which preserves the already installed bp's.
1252 * attr.enable_on_exec means that exec/mmap will install the
1253 * breakpoints we need.
1255 done = filter->nr_systemwide ||
1256 event->parent || event->attr.enable_on_exec ||
1257 trace_uprobe_filter_event(filter, event);
1258 list_add(&event->hw.tp_list, &filter->perf_events);
1260 done = filter->nr_systemwide;
1261 filter->nr_systemwide++;
1263 write_unlock(&filter->rwlock);
1268 static int uprobe_perf_close(struct trace_event_call *call,
1269 struct perf_event *event)
1271 struct trace_probe *tp;
1272 struct trace_uprobe *tu;
1275 tp = trace_probe_primary_from_call(call);
1276 if (WARN_ON_ONCE(!tp))
1279 tu = container_of(tp, struct trace_uprobe, tp);
1280 if (trace_uprobe_filter_remove(tu->tp.event->filter, event))
1283 list_for_each_entry(tu, trace_probe_probe_list(tp), tp.list) {
1284 ret = uprobe_apply(tu->inode, tu->offset, &tu->consumer, false);
1292 static int uprobe_perf_open(struct trace_event_call *call,
1293 struct perf_event *event)
1295 struct trace_probe *tp;
1296 struct trace_uprobe *tu;
1299 tp = trace_probe_primary_from_call(call);
1300 if (WARN_ON_ONCE(!tp))
1303 tu = container_of(tp, struct trace_uprobe, tp);
1304 if (trace_uprobe_filter_add(tu->tp.event->filter, event))
1307 list_for_each_entry(tu, trace_probe_probe_list(tp), tp.list) {
1308 err = uprobe_apply(tu->inode, tu->offset, &tu->consumer, true);
1310 uprobe_perf_close(call, event);
1318 static bool uprobe_perf_filter(struct uprobe_consumer *uc,
1319 enum uprobe_filter_ctx ctx, struct mm_struct *mm)
1321 struct trace_uprobe_filter *filter;
1322 struct trace_uprobe *tu;
1325 tu = container_of(uc, struct trace_uprobe, consumer);
1326 filter = tu->tp.event->filter;
1328 read_lock(&filter->rwlock);
1329 ret = __uprobe_perf_filter(filter, mm);
1330 read_unlock(&filter->rwlock);
1335 static void __uprobe_perf_func(struct trace_uprobe *tu,
1336 unsigned long func, struct pt_regs *regs,
1337 struct uprobe_cpu_buffer *ucb, int dsize)
1339 struct trace_event_call *call = trace_probe_event_call(&tu->tp);
1340 struct uprobe_trace_entry_head *entry;
1341 struct hlist_head *head;
1346 #ifdef CONFIG_BPF_EVENTS
1347 if (bpf_prog_array_valid(call)) {
1350 ret = bpf_prog_run_array_sleepable(call->prog_array, regs, bpf_prog_run);
1354 #endif /* CONFIG_BPF_EVENTS */
1356 esize = SIZEOF_TRACE_ENTRY(is_ret_probe(tu));
1358 size = esize + tu->tp.size + dsize;
1359 size = ALIGN(size + sizeof(u32), sizeof(u64)) - sizeof(u32);
1360 if (WARN_ONCE(size > PERF_MAX_TRACE_SIZE, "profile buffer not large enough"))
1364 head = this_cpu_ptr(call->perf_events);
1365 if (hlist_empty(head))
1368 entry = perf_trace_buf_alloc(size, NULL, &rctx);
1372 if (is_ret_probe(tu)) {
1373 entry->vaddr[0] = func;
1374 entry->vaddr[1] = instruction_pointer(regs);
1375 data = DATAOF_TRACE_ENTRY(entry, true);
1377 entry->vaddr[0] = instruction_pointer(regs);
1378 data = DATAOF_TRACE_ENTRY(entry, false);
1381 memcpy(data, ucb->buf, tu->tp.size + dsize);
1383 if (size - esize > tu->tp.size + dsize) {
1384 int len = tu->tp.size + dsize;
1386 memset(data + len, 0, size - esize - len);
1389 perf_trace_buf_submit(entry, size, rctx, call->event.type, 1, regs,
1395 /* uprobe profile handler */
1396 static int uprobe_perf_func(struct trace_uprobe *tu, struct pt_regs *regs,
1397 struct uprobe_cpu_buffer *ucb, int dsize)
1399 if (!uprobe_perf_filter(&tu->consumer, 0, current->mm))
1400 return UPROBE_HANDLER_REMOVE;
1402 if (!is_ret_probe(tu))
1403 __uprobe_perf_func(tu, 0, regs, ucb, dsize);
1407 static void uretprobe_perf_func(struct trace_uprobe *tu, unsigned long func,
1408 struct pt_regs *regs,
1409 struct uprobe_cpu_buffer *ucb, int dsize)
1411 __uprobe_perf_func(tu, func, regs, ucb, dsize);
1414 int bpf_get_uprobe_info(const struct perf_event *event, u32 *fd_type,
1415 const char **filename, u64 *probe_offset,
1416 bool perf_type_tracepoint)
1418 const char *pevent = trace_event_name(event->tp_event);
1419 const char *group = event->tp_event->class->system;
1420 struct trace_uprobe *tu;
1422 if (perf_type_tracepoint)
1423 tu = find_probe_event(pevent, group);
1425 tu = trace_uprobe_primary_from_call(event->tp_event);
1429 *fd_type = is_ret_probe(tu) ? BPF_FD_TYPE_URETPROBE
1430 : BPF_FD_TYPE_UPROBE;
1431 *filename = tu->filename;
1432 *probe_offset = tu->offset;
1435 #endif /* CONFIG_PERF_EVENTS */
1438 trace_uprobe_register(struct trace_event_call *event, enum trace_reg type,
1441 struct trace_event_file *file = data;
1444 case TRACE_REG_REGISTER:
1445 return probe_event_enable(event, file, NULL);
1447 case TRACE_REG_UNREGISTER:
1448 probe_event_disable(event, file);
1451 #ifdef CONFIG_PERF_EVENTS
1452 case TRACE_REG_PERF_REGISTER:
1453 return probe_event_enable(event, NULL, uprobe_perf_filter);
1455 case TRACE_REG_PERF_UNREGISTER:
1456 probe_event_disable(event, NULL);
1459 case TRACE_REG_PERF_OPEN:
1460 return uprobe_perf_open(event, data);
1462 case TRACE_REG_PERF_CLOSE:
1463 return uprobe_perf_close(event, data);
1471 static int uprobe_dispatcher(struct uprobe_consumer *con, struct pt_regs *regs)
1473 struct trace_uprobe *tu;
1474 struct uprobe_dispatch_data udd;
1475 struct uprobe_cpu_buffer *ucb;
1480 tu = container_of(con, struct trace_uprobe, consumer);
1484 udd.bp_addr = instruction_pointer(regs);
1486 current->utask->vaddr = (unsigned long) &udd;
1488 if (WARN_ON_ONCE(!uprobe_cpu_buffer))
1491 dsize = __get_data_size(&tu->tp, regs);
1492 esize = SIZEOF_TRACE_ENTRY(is_ret_probe(tu));
1494 ucb = uprobe_buffer_get();
1495 store_trace_args(ucb->buf, &tu->tp, regs, esize, dsize);
1497 if (trace_probe_test_flag(&tu->tp, TP_FLAG_TRACE))
1498 ret |= uprobe_trace_func(tu, regs, ucb, dsize);
1500 #ifdef CONFIG_PERF_EVENTS
1501 if (trace_probe_test_flag(&tu->tp, TP_FLAG_PROFILE))
1502 ret |= uprobe_perf_func(tu, regs, ucb, dsize);
1504 uprobe_buffer_put(ucb);
1508 static int uretprobe_dispatcher(struct uprobe_consumer *con,
1509 unsigned long func, struct pt_regs *regs)
1511 struct trace_uprobe *tu;
1512 struct uprobe_dispatch_data udd;
1513 struct uprobe_cpu_buffer *ucb;
1516 tu = container_of(con, struct trace_uprobe, consumer);
1521 current->utask->vaddr = (unsigned long) &udd;
1523 if (WARN_ON_ONCE(!uprobe_cpu_buffer))
1526 dsize = __get_data_size(&tu->tp, regs);
1527 esize = SIZEOF_TRACE_ENTRY(is_ret_probe(tu));
1529 ucb = uprobe_buffer_get();
1530 store_trace_args(ucb->buf, &tu->tp, regs, esize, dsize);
1532 if (trace_probe_test_flag(&tu->tp, TP_FLAG_TRACE))
1533 uretprobe_trace_func(tu, func, regs, ucb, dsize);
1535 #ifdef CONFIG_PERF_EVENTS
1536 if (trace_probe_test_flag(&tu->tp, TP_FLAG_PROFILE))
1537 uretprobe_perf_func(tu, func, regs, ucb, dsize);
1539 uprobe_buffer_put(ucb);
1543 static struct trace_event_functions uprobe_funcs = {
1544 .trace = print_uprobe_event
1547 static struct trace_event_fields uprobe_fields_array[] = {
1548 { .type = TRACE_FUNCTION_TYPE,
1549 .define_fields = uprobe_event_define_fields },
1553 static inline void init_trace_event_call(struct trace_uprobe *tu)
1555 struct trace_event_call *call = trace_probe_event_call(&tu->tp);
1556 call->event.funcs = &uprobe_funcs;
1557 call->class->fields_array = uprobe_fields_array;
1559 call->flags = TRACE_EVENT_FL_UPROBE | TRACE_EVENT_FL_CAP_ANY;
1560 call->class->reg = trace_uprobe_register;
1563 static int register_uprobe_event(struct trace_uprobe *tu)
1565 init_trace_event_call(tu);
1567 return trace_probe_register_event_call(&tu->tp);
1570 static int unregister_uprobe_event(struct trace_uprobe *tu)
1572 return trace_probe_unregister_event_call(&tu->tp);
1575 #ifdef CONFIG_PERF_EVENTS
1576 struct trace_event_call *
1577 create_local_trace_uprobe(char *name, unsigned long offs,
1578 unsigned long ref_ctr_offset, bool is_return)
1580 enum probe_print_type ptype;
1581 struct trace_uprobe *tu;
1585 ret = kern_path(name, LOOKUP_FOLLOW, &path);
1587 return ERR_PTR(ret);
1589 if (!d_is_reg(path.dentry)) {
1591 return ERR_PTR(-EINVAL);
1595 * local trace_kprobes are not added to dyn_event, so they are never
1596 * searched in find_trace_kprobe(). Therefore, there is no concern of
1597 * duplicated name "DUMMY_EVENT" here.
1599 tu = alloc_trace_uprobe(UPROBE_EVENT_SYSTEM, "DUMMY_EVENT", 0,
1603 pr_info("Failed to allocate trace_uprobe.(%d)\n",
1606 return ERR_CAST(tu);
1611 tu->ref_ctr_offset = ref_ctr_offset;
1612 tu->filename = kstrdup(name, GFP_KERNEL);
1613 if (!tu->filename) {
1618 init_trace_event_call(tu);
1620 ptype = is_ret_probe(tu) ? PROBE_PRINT_RETURN : PROBE_PRINT_NORMAL;
1621 if (traceprobe_set_print_fmt(&tu->tp, ptype) < 0) {
1626 return trace_probe_event_call(&tu->tp);
1628 free_trace_uprobe(tu);
1629 return ERR_PTR(ret);
1632 void destroy_local_trace_uprobe(struct trace_event_call *event_call)
1634 struct trace_uprobe *tu;
1636 tu = trace_uprobe_primary_from_call(event_call);
1638 free_trace_uprobe(tu);
1640 #endif /* CONFIG_PERF_EVENTS */
1642 /* Make a trace interface for controlling probe points */
1643 static __init int init_uprobe_trace(void)
1647 ret = dyn_event_register(&trace_uprobe_ops);
1651 ret = tracing_init_dentry();
1655 trace_create_file("uprobe_events", TRACE_MODE_WRITE, NULL,
1656 NULL, &uprobe_events_ops);
1657 /* Profile interface */
1658 trace_create_file("uprobe_profile", TRACE_MODE_READ, NULL,
1659 NULL, &uprobe_profile_ops);
1663 fs_initcall(init_uprobe_trace);