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;
225 /* 1st stage: get value from context */
228 val = regs_get_register(regs, code->param);
231 val = get_user_stack_nth(regs, code->param);
233 case FETCH_OP_STACKP:
234 val = user_stack_pointer(regs);
236 case FETCH_OP_RETVAL:
237 val = regs_return_value(regs);
240 val = FETCH_TOKEN_COMM;
243 val = translate_user_vaddr(code->immediate);
246 ret = process_common_fetch_insn(code, &val);
252 return process_fetch_insn_bottom(code, val, dest, base);
254 NOKPROBE_SYMBOL(process_fetch_insn)
256 static inline void init_trace_uprobe_filter(struct trace_uprobe_filter *filter)
258 rwlock_init(&filter->rwlock);
259 filter->nr_systemwide = 0;
260 INIT_LIST_HEAD(&filter->perf_events);
263 static inline bool uprobe_filter_is_empty(struct trace_uprobe_filter *filter)
265 return !filter->nr_systemwide && list_empty(&filter->perf_events);
268 static inline bool is_ret_probe(struct trace_uprobe *tu)
270 return tu->consumer.ret_handler != NULL;
273 static bool trace_uprobe_is_busy(struct dyn_event *ev)
275 struct trace_uprobe *tu = to_trace_uprobe(ev);
277 return trace_probe_is_enabled(&tu->tp);
280 static bool trace_uprobe_match_command_head(struct trace_uprobe *tu,
281 int argc, const char **argv)
283 char buf[MAX_ARGSTR_LEN + 1];
289 len = strlen(tu->filename);
290 if (strncmp(tu->filename, argv[0], len) || argv[0][len] != ':')
293 if (tu->ref_ctr_offset == 0)
294 snprintf(buf, sizeof(buf), "0x%0*lx",
295 (int)(sizeof(void *) * 2), tu->offset);
297 snprintf(buf, sizeof(buf), "0x%0*lx(0x%lx)",
298 (int)(sizeof(void *) * 2), tu->offset,
300 if (strcmp(buf, &argv[0][len + 1]))
305 return trace_probe_match_command_args(&tu->tp, argc, argv);
308 static bool trace_uprobe_match(const char *system, const char *event,
309 int argc, const char **argv, struct dyn_event *ev)
311 struct trace_uprobe *tu = to_trace_uprobe(ev);
313 return (event[0] == '\0' ||
314 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 inode *comp_inode = d_real_inode(comp->path.dentry);
415 list_for_each_entry(orig, &tpe->probes, tp.list) {
416 if (comp_inode != d_real_inode(orig->path.dentry) ||
417 comp->offset != orig->offset)
421 * trace_probe_compare_arg_type() ensured that nr_args and
422 * each argument name and type are same. Let's compare comm.
424 for (i = 0; i < orig->tp.nr_args; i++) {
425 if (strcmp(orig->tp.args[i].comm,
426 comp->tp.args[i].comm))
430 if (i == orig->tp.nr_args)
437 static int append_trace_uprobe(struct trace_uprobe *tu, struct trace_uprobe *to)
441 ret = trace_probe_compare_arg_type(&tu->tp, &to->tp);
443 /* Note that argument starts index = 2 */
444 trace_probe_log_set_index(ret + 1);
445 trace_probe_log_err(0, DIFF_ARG_TYPE);
448 if (trace_uprobe_has_same_uprobe(to, tu)) {
449 trace_probe_log_set_index(0);
450 trace_probe_log_err(0, SAME_PROBE);
454 /* Append to existing event */
455 ret = trace_probe_append(&tu->tp, &to->tp);
457 dyn_event_add(&tu->devent, trace_probe_event_call(&tu->tp));
463 * Uprobe with multiple reference counter is not allowed. i.e.
464 * If inode and offset matches, reference counter offset *must*
465 * match as well. Though, there is one exception: If user is
466 * replacing old trace_uprobe with new one(same group/event),
467 * then we allow same uprobe with new reference counter as far
468 * as the new one does not conflict with any other existing
471 static int validate_ref_ctr_offset(struct trace_uprobe *new)
473 struct dyn_event *pos;
474 struct trace_uprobe *tmp;
475 struct inode *new_inode = d_real_inode(new->path.dentry);
477 for_each_trace_uprobe(tmp, pos) {
478 if (new_inode == d_real_inode(tmp->path.dentry) &&
479 new->offset == tmp->offset &&
480 new->ref_ctr_offset != tmp->ref_ctr_offset) {
481 pr_warn("Reference counter offset mismatch.");
488 /* Register a trace_uprobe and probe_event */
489 static int register_trace_uprobe(struct trace_uprobe *tu)
491 struct trace_uprobe *old_tu;
494 mutex_lock(&event_mutex);
496 ret = validate_ref_ctr_offset(tu);
500 /* register as an event */
501 old_tu = find_probe_event(trace_probe_name(&tu->tp),
502 trace_probe_group_name(&tu->tp));
504 if (is_ret_probe(tu) != is_ret_probe(old_tu)) {
505 trace_probe_log_set_index(0);
506 trace_probe_log_err(0, DIFF_PROBE_TYPE);
509 ret = append_trace_uprobe(tu, old_tu);
514 ret = register_uprobe_event(tu);
516 if (ret == -EEXIST) {
517 trace_probe_log_set_index(0);
518 trace_probe_log_err(0, EVENT_EXIST);
520 pr_warn("Failed to register probe event(%d)\n", ret);
524 dyn_event_add(&tu->devent, trace_probe_event_call(&tu->tp));
527 mutex_unlock(&event_mutex);
534 * - Add uprobe: p|r[:[GRP/][EVENT]] PATH:OFFSET[%return][(REF)] [FETCHARGS]
536 static int __trace_uprobe_create(int argc, const char **argv)
538 struct trace_uprobe *tu;
539 const char *event = NULL, *group = UPROBE_EVENT_SYSTEM;
540 char *arg, *filename, *rctr, *rctr_end, *tmp;
541 char buf[MAX_EVENT_NAME_LEN];
542 char gbuf[MAX_EVENT_NAME_LEN];
543 enum probe_print_type ptype;
545 unsigned long offset, ref_ctr_offset;
546 bool is_return = false;
551 switch (argv[0][0]) {
564 if (argv[0][1] == ':')
567 if (!strchr(argv[1], '/'))
570 filename = kstrdup(argv[1], GFP_KERNEL);
574 /* Find the last occurrence, in case the path contains ':' too. */
575 arg = strrchr(filename, ':');
576 if (!arg || !isdigit(arg[1])) {
581 trace_probe_log_init("trace_uprobe", argc, argv);
582 trace_probe_log_set_index(1); /* filename is the 2nd argument */
585 ret = kern_path(filename, LOOKUP_FOLLOW, &path);
587 trace_probe_log_err(0, FILE_NOT_FOUND);
589 trace_probe_log_clear();
592 if (!d_is_reg(path.dentry)) {
593 trace_probe_log_err(0, NO_REGULAR_FILE);
595 goto fail_address_parse;
598 /* Parse reference counter offset if specified. */
599 rctr = strchr(arg, '(');
601 rctr_end = strchr(rctr, ')');
604 rctr_end = rctr + strlen(rctr);
605 trace_probe_log_err(rctr_end - filename,
607 goto fail_address_parse;
608 } else if (rctr_end[1] != '\0') {
610 trace_probe_log_err(rctr_end + 1 - filename,
612 goto fail_address_parse;
617 ret = kstrtoul(rctr, 0, &ref_ctr_offset);
619 trace_probe_log_err(rctr - filename, BAD_REFCNT);
620 goto fail_address_parse;
624 /* Check if there is %return suffix */
625 tmp = strchr(arg, '%');
627 if (!strcmp(tmp, "%return")) {
631 trace_probe_log_err(tmp - filename, BAD_ADDR_SUFFIX);
633 goto fail_address_parse;
637 /* Parse uprobe offset. */
638 ret = kstrtoul(arg, 0, &offset);
640 trace_probe_log_err(arg - filename, BAD_UPROBE_OFFS);
641 goto fail_address_parse;
645 trace_probe_log_set_index(0);
647 ret = traceprobe_parse_event_name(&event, &group, gbuf,
650 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) |
697 ptype = is_ret_probe(tu) ? PROBE_PRINT_RETURN : PROBE_PRINT_NORMAL;
698 ret = traceprobe_set_print_fmt(&tu->tp, ptype);
702 ret = register_trace_uprobe(tu);
707 free_trace_uprobe(tu);
709 trace_probe_log_clear();
713 trace_probe_log_clear();
720 int trace_uprobe_create(const char *raw_command)
722 return trace_probe_create(raw_command, __trace_uprobe_create);
725 static int create_or_delete_trace_uprobe(const char *raw_command)
729 if (raw_command[0] == '-')
730 return dyn_event_release(raw_command, &trace_uprobe_ops);
732 ret = trace_uprobe_create(raw_command);
733 return ret == -ECANCELED ? -EINVAL : ret;
736 static int trace_uprobe_release(struct dyn_event *ev)
738 struct trace_uprobe *tu = to_trace_uprobe(ev);
740 return unregister_trace_uprobe(tu);
743 /* Probes listing interfaces */
744 static int trace_uprobe_show(struct seq_file *m, struct dyn_event *ev)
746 struct trace_uprobe *tu = to_trace_uprobe(ev);
747 char c = is_ret_probe(tu) ? 'r' : 'p';
750 seq_printf(m, "%c:%s/%s %s:0x%0*lx", c, trace_probe_group_name(&tu->tp),
751 trace_probe_name(&tu->tp), tu->filename,
752 (int)(sizeof(void *) * 2), tu->offset);
754 if (tu->ref_ctr_offset)
755 seq_printf(m, "(0x%lx)", tu->ref_ctr_offset);
757 for (i = 0; i < tu->tp.nr_args; i++)
758 seq_printf(m, " %s=%s", tu->tp.args[i].name, tu->tp.args[i].comm);
764 static int probes_seq_show(struct seq_file *m, void *v)
766 struct dyn_event *ev = v;
768 if (!is_trace_uprobe(ev))
771 return trace_uprobe_show(m, ev);
774 static const struct seq_operations probes_seq_op = {
775 .start = dyn_event_seq_start,
776 .next = dyn_event_seq_next,
777 .stop = dyn_event_seq_stop,
778 .show = probes_seq_show
781 static int probes_open(struct inode *inode, struct file *file)
785 ret = security_locked_down(LOCKDOWN_TRACEFS);
789 if ((file->f_mode & FMODE_WRITE) && (file->f_flags & O_TRUNC)) {
790 ret = dyn_events_release_all(&trace_uprobe_ops);
795 return seq_open(file, &probes_seq_op);
798 static ssize_t probes_write(struct file *file, const char __user *buffer,
799 size_t count, loff_t *ppos)
801 return trace_parse_run_command(file, buffer, count, ppos,
802 create_or_delete_trace_uprobe);
805 static const struct file_operations uprobe_events_ops = {
806 .owner = THIS_MODULE,
810 .release = seq_release,
811 .write = probes_write,
814 /* Probes profiling interfaces */
815 static int probes_profile_seq_show(struct seq_file *m, void *v)
817 struct dyn_event *ev = v;
818 struct trace_uprobe *tu;
820 if (!is_trace_uprobe(ev))
823 tu = to_trace_uprobe(ev);
824 seq_printf(m, " %s %-44s %15lu\n", tu->filename,
825 trace_probe_name(&tu->tp), tu->nhit);
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)
840 ret = security_locked_down(LOCKDOWN_TRACEFS);
844 return seq_open(file, &profile_seq_op);
847 static const struct file_operations uprobe_profile_ops = {
848 .owner = THIS_MODULE,
849 .open = profile_open,
852 .release = seq_release,
855 struct uprobe_cpu_buffer {
859 static struct uprobe_cpu_buffer __percpu *uprobe_cpu_buffer;
860 static int uprobe_buffer_refcnt;
862 static int uprobe_buffer_init(void)
866 uprobe_cpu_buffer = alloc_percpu(struct uprobe_cpu_buffer);
867 if (uprobe_cpu_buffer == NULL)
870 for_each_possible_cpu(cpu) {
871 struct page *p = alloc_pages_node(cpu_to_node(cpu),
877 per_cpu_ptr(uprobe_cpu_buffer, cpu)->buf = page_address(p);
878 mutex_init(&per_cpu_ptr(uprobe_cpu_buffer, cpu)->mutex);
884 for_each_possible_cpu(cpu) {
887 free_page((unsigned long)per_cpu_ptr(uprobe_cpu_buffer, cpu)->buf);
890 free_percpu(uprobe_cpu_buffer);
894 static int uprobe_buffer_enable(void)
898 BUG_ON(!mutex_is_locked(&event_mutex));
900 if (uprobe_buffer_refcnt++ == 0) {
901 ret = uprobe_buffer_init();
903 uprobe_buffer_refcnt--;
909 static void uprobe_buffer_disable(void)
913 BUG_ON(!mutex_is_locked(&event_mutex));
915 if (--uprobe_buffer_refcnt == 0) {
916 for_each_possible_cpu(cpu)
917 free_page((unsigned long)per_cpu_ptr(uprobe_cpu_buffer,
920 free_percpu(uprobe_cpu_buffer);
921 uprobe_cpu_buffer = NULL;
925 static struct uprobe_cpu_buffer *uprobe_buffer_get(void)
927 struct uprobe_cpu_buffer *ucb;
930 cpu = raw_smp_processor_id();
931 ucb = per_cpu_ptr(uprobe_cpu_buffer, cpu);
934 * Use per-cpu buffers for fastest access, but we might migrate
935 * so the mutex makes sure we have sole access to it.
937 mutex_lock(&ucb->mutex);
942 static void uprobe_buffer_put(struct uprobe_cpu_buffer *ucb)
944 mutex_unlock(&ucb->mutex);
947 static void __uprobe_trace_func(struct trace_uprobe *tu,
948 unsigned long func, struct pt_regs *regs,
949 struct uprobe_cpu_buffer *ucb, int dsize,
950 struct trace_event_file *trace_file)
952 struct uprobe_trace_entry_head *entry;
953 struct trace_event_buffer fbuffer;
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 entry = trace_event_buffer_reserve(&fbuffer, trace_file, size);
972 if (is_ret_probe(tu)) {
973 entry->vaddr[0] = func;
974 entry->vaddr[1] = instruction_pointer(regs);
975 data = DATAOF_TRACE_ENTRY(entry, true);
977 entry->vaddr[0] = instruction_pointer(regs);
978 data = DATAOF_TRACE_ENTRY(entry, false);
981 memcpy(data, ucb->buf, tu->tp.size + dsize);
983 trace_event_buffer_commit(&fbuffer);
987 static int uprobe_trace_func(struct trace_uprobe *tu, struct pt_regs *regs,
988 struct uprobe_cpu_buffer *ucb, int dsize)
990 struct event_file_link *link;
992 if (is_ret_probe(tu))
996 trace_probe_for_each_link_rcu(link, &tu->tp)
997 __uprobe_trace_func(tu, 0, regs, ucb, dsize, link->file);
1003 static void uretprobe_trace_func(struct trace_uprobe *tu, unsigned long func,
1004 struct pt_regs *regs,
1005 struct uprobe_cpu_buffer *ucb, int dsize)
1007 struct event_file_link *link;
1010 trace_probe_for_each_link_rcu(link, &tu->tp)
1011 __uprobe_trace_func(tu, func, regs, ucb, dsize, link->file);
1015 /* Event entry printers */
1016 static enum print_line_t
1017 print_uprobe_event(struct trace_iterator *iter, int flags, struct trace_event *event)
1019 struct uprobe_trace_entry_head *entry;
1020 struct trace_seq *s = &iter->seq;
1021 struct trace_uprobe *tu;
1024 entry = (struct uprobe_trace_entry_head *)iter->ent;
1025 tu = trace_uprobe_primary_from_call(
1026 container_of(event, struct trace_event_call, event));
1030 if (is_ret_probe(tu)) {
1031 trace_seq_printf(s, "%s: (0x%lx <- 0x%lx)",
1032 trace_probe_name(&tu->tp),
1033 entry->vaddr[1], entry->vaddr[0]);
1034 data = DATAOF_TRACE_ENTRY(entry, true);
1036 trace_seq_printf(s, "%s: (0x%lx)",
1037 trace_probe_name(&tu->tp),
1039 data = DATAOF_TRACE_ENTRY(entry, false);
1042 if (trace_probe_print_args(s, tu->tp.args, tu->tp.nr_args, data, entry) < 0)
1045 trace_seq_putc(s, '\n');
1048 return trace_handle_return(s);
1051 typedef bool (*filter_func_t)(struct uprobe_consumer *self,
1052 enum uprobe_filter_ctx ctx,
1053 struct mm_struct *mm);
1055 static int trace_uprobe_enable(struct trace_uprobe *tu, filter_func_t filter)
1059 tu->consumer.filter = filter;
1060 tu->inode = d_real_inode(tu->path.dentry);
1062 if (tu->ref_ctr_offset)
1063 ret = uprobe_register_refctr(tu->inode, tu->offset,
1064 tu->ref_ctr_offset, &tu->consumer);
1066 ret = uprobe_register(tu->inode, tu->offset, &tu->consumer);
1074 static void __probe_event_disable(struct trace_probe *tp)
1076 struct trace_uprobe *tu;
1078 tu = container_of(tp, struct trace_uprobe, tp);
1079 WARN_ON(!uprobe_filter_is_empty(tu->tp.event->filter));
1081 list_for_each_entry(tu, trace_probe_probe_list(tp), tp.list) {
1085 uprobe_unregister(tu->inode, tu->offset, &tu->consumer);
1090 static int probe_event_enable(struct trace_event_call *call,
1091 struct trace_event_file *file, filter_func_t filter)
1093 struct trace_probe *tp;
1094 struct trace_uprobe *tu;
1098 tp = trace_probe_primary_from_call(call);
1099 if (WARN_ON_ONCE(!tp))
1101 enabled = trace_probe_is_enabled(tp);
1103 /* This may also change "enabled" state */
1105 if (trace_probe_test_flag(tp, TP_FLAG_PROFILE))
1108 ret = trace_probe_add_file(tp, file);
1112 if (trace_probe_test_flag(tp, TP_FLAG_TRACE))
1115 trace_probe_set_flag(tp, TP_FLAG_PROFILE);
1118 tu = container_of(tp, struct trace_uprobe, tp);
1119 WARN_ON(!uprobe_filter_is_empty(tu->tp.event->filter));
1124 ret = uprobe_buffer_enable();
1128 list_for_each_entry(tu, trace_probe_probe_list(tp), tp.list) {
1129 ret = trace_uprobe_enable(tu, filter);
1131 __probe_event_disable(tp);
1139 uprobe_buffer_disable();
1143 trace_probe_remove_file(tp, file);
1145 trace_probe_clear_flag(tp, TP_FLAG_PROFILE);
1150 static void probe_event_disable(struct trace_event_call *call,
1151 struct trace_event_file *file)
1153 struct trace_probe *tp;
1155 tp = trace_probe_primary_from_call(call);
1156 if (WARN_ON_ONCE(!tp))
1159 if (!trace_probe_is_enabled(tp))
1163 if (trace_probe_remove_file(tp, file) < 0)
1166 if (trace_probe_is_enabled(tp))
1169 trace_probe_clear_flag(tp, TP_FLAG_PROFILE);
1171 __probe_event_disable(tp);
1172 uprobe_buffer_disable();
1175 static int uprobe_event_define_fields(struct trace_event_call *event_call)
1178 struct uprobe_trace_entry_head field;
1179 struct trace_uprobe *tu;
1181 tu = trace_uprobe_primary_from_call(event_call);
1185 if (is_ret_probe(tu)) {
1186 DEFINE_FIELD(unsigned long, vaddr[0], FIELD_STRING_FUNC, 0);
1187 DEFINE_FIELD(unsigned long, vaddr[1], FIELD_STRING_RETIP, 0);
1188 size = SIZEOF_TRACE_ENTRY(true);
1190 DEFINE_FIELD(unsigned long, vaddr[0], FIELD_STRING_IP, 0);
1191 size = SIZEOF_TRACE_ENTRY(false);
1194 return traceprobe_define_arg_fields(event_call, size, &tu->tp);
1197 #ifdef CONFIG_PERF_EVENTS
1199 __uprobe_perf_filter(struct trace_uprobe_filter *filter, struct mm_struct *mm)
1201 struct perf_event *event;
1203 if (filter->nr_systemwide)
1206 list_for_each_entry(event, &filter->perf_events, hw.tp_list) {
1207 if (event->hw.target->mm == mm)
1215 trace_uprobe_filter_event(struct trace_uprobe_filter *filter,
1216 struct perf_event *event)
1218 return __uprobe_perf_filter(filter, event->hw.target->mm);
1221 static bool trace_uprobe_filter_remove(struct trace_uprobe_filter *filter,
1222 struct perf_event *event)
1226 write_lock(&filter->rwlock);
1227 if (event->hw.target) {
1228 list_del(&event->hw.tp_list);
1229 done = filter->nr_systemwide ||
1230 (event->hw.target->flags & PF_EXITING) ||
1231 trace_uprobe_filter_event(filter, event);
1233 filter->nr_systemwide--;
1234 done = filter->nr_systemwide;
1236 write_unlock(&filter->rwlock);
1241 /* This returns true if the filter always covers target mm */
1242 static bool trace_uprobe_filter_add(struct trace_uprobe_filter *filter,
1243 struct perf_event *event)
1247 write_lock(&filter->rwlock);
1248 if (event->hw.target) {
1250 * event->parent != NULL means copy_process(), we can avoid
1251 * uprobe_apply(). current->mm must be probed and we can rely
1252 * on dup_mmap() which preserves the already installed bp's.
1254 * attr.enable_on_exec means that exec/mmap will install the
1255 * breakpoints we need.
1257 done = filter->nr_systemwide ||
1258 event->parent || event->attr.enable_on_exec ||
1259 trace_uprobe_filter_event(filter, event);
1260 list_add(&event->hw.tp_list, &filter->perf_events);
1262 done = filter->nr_systemwide;
1263 filter->nr_systemwide++;
1265 write_unlock(&filter->rwlock);
1270 static int uprobe_perf_close(struct trace_event_call *call,
1271 struct perf_event *event)
1273 struct trace_probe *tp;
1274 struct trace_uprobe *tu;
1277 tp = trace_probe_primary_from_call(call);
1278 if (WARN_ON_ONCE(!tp))
1281 tu = container_of(tp, struct trace_uprobe, tp);
1282 if (trace_uprobe_filter_remove(tu->tp.event->filter, event))
1285 list_for_each_entry(tu, trace_probe_probe_list(tp), tp.list) {
1286 ret = uprobe_apply(tu->inode, tu->offset, &tu->consumer, false);
1294 static int uprobe_perf_open(struct trace_event_call *call,
1295 struct perf_event *event)
1297 struct trace_probe *tp;
1298 struct trace_uprobe *tu;
1301 tp = trace_probe_primary_from_call(call);
1302 if (WARN_ON_ONCE(!tp))
1305 tu = container_of(tp, struct trace_uprobe, tp);
1306 if (trace_uprobe_filter_add(tu->tp.event->filter, event))
1309 list_for_each_entry(tu, trace_probe_probe_list(tp), tp.list) {
1310 err = uprobe_apply(tu->inode, tu->offset, &tu->consumer, true);
1312 uprobe_perf_close(call, event);
1320 static bool uprobe_perf_filter(struct uprobe_consumer *uc,
1321 enum uprobe_filter_ctx ctx, struct mm_struct *mm)
1323 struct trace_uprobe_filter *filter;
1324 struct trace_uprobe *tu;
1327 tu = container_of(uc, struct trace_uprobe, consumer);
1328 filter = tu->tp.event->filter;
1330 read_lock(&filter->rwlock);
1331 ret = __uprobe_perf_filter(filter, mm);
1332 read_unlock(&filter->rwlock);
1337 static void __uprobe_perf_func(struct trace_uprobe *tu,
1338 unsigned long func, struct pt_regs *regs,
1339 struct uprobe_cpu_buffer *ucb, int dsize)
1341 struct trace_event_call *call = trace_probe_event_call(&tu->tp);
1342 struct uprobe_trace_entry_head *entry;
1343 struct hlist_head *head;
1348 #ifdef CONFIG_BPF_EVENTS
1349 if (bpf_prog_array_valid(call)) {
1352 ret = bpf_prog_run_array_sleepable(call->prog_array, regs, bpf_prog_run);
1356 #endif /* CONFIG_BPF_EVENTS */
1358 esize = SIZEOF_TRACE_ENTRY(is_ret_probe(tu));
1360 size = esize + tu->tp.size + dsize;
1361 size = ALIGN(size + sizeof(u32), sizeof(u64)) - sizeof(u32);
1362 if (WARN_ONCE(size > PERF_MAX_TRACE_SIZE, "profile buffer not large enough"))
1366 head = this_cpu_ptr(call->perf_events);
1367 if (hlist_empty(head))
1370 entry = perf_trace_buf_alloc(size, NULL, &rctx);
1374 if (is_ret_probe(tu)) {
1375 entry->vaddr[0] = func;
1376 entry->vaddr[1] = instruction_pointer(regs);
1377 data = DATAOF_TRACE_ENTRY(entry, true);
1379 entry->vaddr[0] = instruction_pointer(regs);
1380 data = DATAOF_TRACE_ENTRY(entry, false);
1383 memcpy(data, ucb->buf, tu->tp.size + dsize);
1385 if (size - esize > tu->tp.size + dsize) {
1386 int len = tu->tp.size + dsize;
1388 memset(data + len, 0, size - esize - len);
1391 perf_trace_buf_submit(entry, size, rctx, call->event.type, 1, regs,
1397 /* uprobe profile handler */
1398 static int uprobe_perf_func(struct trace_uprobe *tu, struct pt_regs *regs,
1399 struct uprobe_cpu_buffer *ucb, int dsize)
1401 if (!uprobe_perf_filter(&tu->consumer, 0, current->mm))
1402 return UPROBE_HANDLER_REMOVE;
1404 if (!is_ret_probe(tu))
1405 __uprobe_perf_func(tu, 0, regs, ucb, dsize);
1409 static void uretprobe_perf_func(struct trace_uprobe *tu, unsigned long func,
1410 struct pt_regs *regs,
1411 struct uprobe_cpu_buffer *ucb, int dsize)
1413 __uprobe_perf_func(tu, func, regs, ucb, dsize);
1416 int bpf_get_uprobe_info(const struct perf_event *event, u32 *fd_type,
1417 const char **filename, u64 *probe_offset,
1418 bool perf_type_tracepoint)
1420 const char *pevent = trace_event_name(event->tp_event);
1421 const char *group = event->tp_event->class->system;
1422 struct trace_uprobe *tu;
1424 if (perf_type_tracepoint)
1425 tu = find_probe_event(pevent, group);
1427 tu = trace_uprobe_primary_from_call(event->tp_event);
1431 *fd_type = is_ret_probe(tu) ? BPF_FD_TYPE_URETPROBE
1432 : BPF_FD_TYPE_UPROBE;
1433 *filename = tu->filename;
1434 *probe_offset = tu->offset;
1437 #endif /* CONFIG_PERF_EVENTS */
1440 trace_uprobe_register(struct trace_event_call *event, enum trace_reg type,
1443 struct trace_event_file *file = data;
1446 case TRACE_REG_REGISTER:
1447 return probe_event_enable(event, file, NULL);
1449 case TRACE_REG_UNREGISTER:
1450 probe_event_disable(event, file);
1453 #ifdef CONFIG_PERF_EVENTS
1454 case TRACE_REG_PERF_REGISTER:
1455 return probe_event_enable(event, NULL, uprobe_perf_filter);
1457 case TRACE_REG_PERF_UNREGISTER:
1458 probe_event_disable(event, NULL);
1461 case TRACE_REG_PERF_OPEN:
1462 return uprobe_perf_open(event, data);
1464 case TRACE_REG_PERF_CLOSE:
1465 return uprobe_perf_close(event, data);
1473 static int uprobe_dispatcher(struct uprobe_consumer *con, struct pt_regs *regs)
1475 struct trace_uprobe *tu;
1476 struct uprobe_dispatch_data udd;
1477 struct uprobe_cpu_buffer *ucb;
1482 tu = container_of(con, struct trace_uprobe, consumer);
1486 udd.bp_addr = instruction_pointer(regs);
1488 current->utask->vaddr = (unsigned long) &udd;
1490 if (WARN_ON_ONCE(!uprobe_cpu_buffer))
1493 dsize = __get_data_size(&tu->tp, regs);
1494 esize = SIZEOF_TRACE_ENTRY(is_ret_probe(tu));
1496 ucb = uprobe_buffer_get();
1497 store_trace_args(ucb->buf, &tu->tp, regs, esize, dsize);
1499 if (trace_probe_test_flag(&tu->tp, TP_FLAG_TRACE))
1500 ret |= uprobe_trace_func(tu, regs, ucb, dsize);
1502 #ifdef CONFIG_PERF_EVENTS
1503 if (trace_probe_test_flag(&tu->tp, TP_FLAG_PROFILE))
1504 ret |= uprobe_perf_func(tu, regs, ucb, dsize);
1506 uprobe_buffer_put(ucb);
1510 static int uretprobe_dispatcher(struct uprobe_consumer *con,
1511 unsigned long func, struct pt_regs *regs)
1513 struct trace_uprobe *tu;
1514 struct uprobe_dispatch_data udd;
1515 struct uprobe_cpu_buffer *ucb;
1518 tu = container_of(con, struct trace_uprobe, consumer);
1523 current->utask->vaddr = (unsigned long) &udd;
1525 if (WARN_ON_ONCE(!uprobe_cpu_buffer))
1528 dsize = __get_data_size(&tu->tp, regs);
1529 esize = SIZEOF_TRACE_ENTRY(is_ret_probe(tu));
1531 ucb = uprobe_buffer_get();
1532 store_trace_args(ucb->buf, &tu->tp, regs, esize, dsize);
1534 if (trace_probe_test_flag(&tu->tp, TP_FLAG_TRACE))
1535 uretprobe_trace_func(tu, func, regs, ucb, dsize);
1537 #ifdef CONFIG_PERF_EVENTS
1538 if (trace_probe_test_flag(&tu->tp, TP_FLAG_PROFILE))
1539 uretprobe_perf_func(tu, func, regs, ucb, dsize);
1541 uprobe_buffer_put(ucb);
1545 static struct trace_event_functions uprobe_funcs = {
1546 .trace = print_uprobe_event
1549 static struct trace_event_fields uprobe_fields_array[] = {
1550 { .type = TRACE_FUNCTION_TYPE,
1551 .define_fields = uprobe_event_define_fields },
1555 static inline void init_trace_event_call(struct trace_uprobe *tu)
1557 struct trace_event_call *call = trace_probe_event_call(&tu->tp);
1558 call->event.funcs = &uprobe_funcs;
1559 call->class->fields_array = uprobe_fields_array;
1561 call->flags = TRACE_EVENT_FL_UPROBE | TRACE_EVENT_FL_CAP_ANY;
1562 call->class->reg = trace_uprobe_register;
1565 static int register_uprobe_event(struct trace_uprobe *tu)
1567 init_trace_event_call(tu);
1569 return trace_probe_register_event_call(&tu->tp);
1572 static int unregister_uprobe_event(struct trace_uprobe *tu)
1574 return trace_probe_unregister_event_call(&tu->tp);
1577 #ifdef CONFIG_PERF_EVENTS
1578 struct trace_event_call *
1579 create_local_trace_uprobe(char *name, unsigned long offs,
1580 unsigned long ref_ctr_offset, bool is_return)
1582 enum probe_print_type ptype;
1583 struct trace_uprobe *tu;
1587 ret = kern_path(name, LOOKUP_FOLLOW, &path);
1589 return ERR_PTR(ret);
1591 if (!d_is_reg(path.dentry)) {
1593 return ERR_PTR(-EINVAL);
1597 * local trace_kprobes are not added to dyn_event, so they are never
1598 * searched in find_trace_kprobe(). Therefore, there is no concern of
1599 * duplicated name "DUMMY_EVENT" here.
1601 tu = alloc_trace_uprobe(UPROBE_EVENT_SYSTEM, "DUMMY_EVENT", 0,
1605 pr_info("Failed to allocate trace_uprobe.(%d)\n",
1608 return ERR_CAST(tu);
1613 tu->ref_ctr_offset = ref_ctr_offset;
1614 tu->filename = kstrdup(name, GFP_KERNEL);
1615 if (!tu->filename) {
1620 init_trace_event_call(tu);
1622 ptype = is_ret_probe(tu) ? PROBE_PRINT_RETURN : PROBE_PRINT_NORMAL;
1623 if (traceprobe_set_print_fmt(&tu->tp, ptype) < 0) {
1628 return trace_probe_event_call(&tu->tp);
1630 free_trace_uprobe(tu);
1631 return ERR_PTR(ret);
1634 void destroy_local_trace_uprobe(struct trace_event_call *event_call)
1636 struct trace_uprobe *tu;
1638 tu = trace_uprobe_primary_from_call(event_call);
1640 free_trace_uprobe(tu);
1642 #endif /* CONFIG_PERF_EVENTS */
1644 /* Make a trace interface for controlling probe points */
1645 static __init int init_uprobe_trace(void)
1649 ret = dyn_event_register(&trace_uprobe_ops);
1653 ret = tracing_init_dentry();
1657 trace_create_file("uprobe_events", TRACE_MODE_WRITE, NULL,
1658 NULL, &uprobe_events_ops);
1659 /* Profile interface */
1660 trace_create_file("uprobe_profile", TRACE_MODE_READ, NULL,
1661 NULL, &uprobe_profile_ops);
1665 fs_initcall(init_uprobe_trace);