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 #define SIZEOF_TRACE_UPROBE(n) \
87 (offsetof(struct trace_uprobe, tp.args) + \
88 (sizeof(struct probe_arg) * (n)))
90 static int register_uprobe_event(struct trace_uprobe *tu);
91 static int unregister_uprobe_event(struct trace_uprobe *tu);
93 struct uprobe_dispatch_data {
94 struct trace_uprobe *tu;
95 unsigned long bp_addr;
98 static int uprobe_dispatcher(struct uprobe_consumer *con, struct pt_regs *regs);
99 static int uretprobe_dispatcher(struct uprobe_consumer *con,
100 unsigned long func, struct pt_regs *regs);
102 #ifdef CONFIG_STACK_GROWSUP
103 static unsigned long adjust_stack_addr(unsigned long addr, unsigned int n)
105 return addr - (n * sizeof(long));
108 static unsigned long adjust_stack_addr(unsigned long addr, unsigned int n)
110 return addr + (n * sizeof(long));
114 static unsigned long get_user_stack_nth(struct pt_regs *regs, unsigned int n)
117 unsigned long addr = user_stack_pointer(regs);
119 addr = adjust_stack_addr(addr, n);
121 if (copy_from_user(&ret, (void __force __user *) addr, sizeof(ret)))
128 * Uprobes-specific fetch functions
130 static nokprobe_inline int
131 probe_mem_read(void *dest, void *src, size_t size)
133 void __user *vaddr = (void __force __user *)src;
135 return copy_from_user(dest, vaddr, size) ? -EFAULT : 0;
138 static nokprobe_inline int
139 probe_mem_read_user(void *dest, void *src, size_t size)
141 return probe_mem_read(dest, src, size);
145 * Fetch a null-terminated string. Caller MUST set *(u32 *)dest with max
146 * length and relative data location.
148 static nokprobe_inline int
149 fetch_store_string(unsigned long addr, void *dest, void *base)
152 u32 loc = *(u32 *)dest;
153 int maxlen = get_loc_len(loc);
154 u8 *dst = get_loc_data(dest, base);
155 void __user *src = (void __force __user *) addr;
157 if (unlikely(!maxlen))
160 if (addr == FETCH_TOKEN_COMM)
161 ret = strlcpy(dst, current->comm, maxlen);
163 ret = strncpy_from_user(dst, src, maxlen);
169 * Include the terminating null byte. In this case it
170 * was copied by strncpy_from_user but not accounted
174 *(u32 *)dest = make_data_loc(ret, (void *)dst - base);
180 static nokprobe_inline int
181 fetch_store_string_user(unsigned long addr, void *dest, void *base)
183 return fetch_store_string(addr, dest, base);
186 /* Return the length of string -- including null terminal byte */
187 static nokprobe_inline int
188 fetch_store_strlen(unsigned long addr)
191 void __user *vaddr = (void __force __user *) addr;
193 if (addr == FETCH_TOKEN_COMM)
194 len = strlen(current->comm) + 1;
196 len = strnlen_user(vaddr, MAX_STRING_SIZE);
198 return (len > MAX_STRING_SIZE) ? 0 : len;
201 static nokprobe_inline int
202 fetch_store_strlen_user(unsigned long addr)
204 return fetch_store_strlen(addr);
207 static unsigned long translate_user_vaddr(unsigned long file_offset)
209 unsigned long base_addr;
210 struct uprobe_dispatch_data *udd;
212 udd = (void *) current->utask->vaddr;
214 base_addr = udd->bp_addr - udd->tu->offset;
215 return base_addr + file_offset;
218 /* Note that we don't verify it, since the code does not come from user space */
220 process_fetch_insn(struct fetch_insn *code, struct pt_regs *regs, void *dest,
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 = code->immediate;
243 val = FETCH_TOKEN_COMM;
246 val = (unsigned long)code->data;
249 val = translate_user_vaddr(code->immediate);
256 return process_fetch_insn_bottom(code, val, dest, base);
258 NOKPROBE_SYMBOL(process_fetch_insn)
260 static inline void init_trace_uprobe_filter(struct trace_uprobe_filter *filter)
262 rwlock_init(&filter->rwlock);
263 filter->nr_systemwide = 0;
264 INIT_LIST_HEAD(&filter->perf_events);
267 static inline bool uprobe_filter_is_empty(struct trace_uprobe_filter *filter)
269 return !filter->nr_systemwide && list_empty(&filter->perf_events);
272 static inline bool is_ret_probe(struct trace_uprobe *tu)
274 return tu->consumer.ret_handler != NULL;
277 static bool trace_uprobe_is_busy(struct dyn_event *ev)
279 struct trace_uprobe *tu = to_trace_uprobe(ev);
281 return trace_probe_is_enabled(&tu->tp);
284 static bool trace_uprobe_match_command_head(struct trace_uprobe *tu,
285 int argc, const char **argv)
287 char buf[MAX_ARGSTR_LEN + 1];
293 len = strlen(tu->filename);
294 if (strncmp(tu->filename, argv[0], len) || argv[0][len] != ':')
297 if (tu->ref_ctr_offset == 0)
298 snprintf(buf, sizeof(buf), "0x%0*lx",
299 (int)(sizeof(void *) * 2), tu->offset);
301 snprintf(buf, sizeof(buf), "0x%0*lx(0x%lx)",
302 (int)(sizeof(void *) * 2), tu->offset,
304 if (strcmp(buf, &argv[0][len + 1]))
309 return trace_probe_match_command_args(&tu->tp, argc, argv);
312 static bool trace_uprobe_match(const char *system, const char *event,
313 int argc, const char **argv, struct dyn_event *ev)
315 struct trace_uprobe *tu = to_trace_uprobe(ev);
317 return strcmp(trace_probe_name(&tu->tp), event) == 0 &&
318 (!system || strcmp(trace_probe_group_name(&tu->tp), system) == 0) &&
319 trace_uprobe_match_command_head(tu, argc, argv);
322 static nokprobe_inline struct trace_uprobe *
323 trace_uprobe_primary_from_call(struct trace_event_call *call)
325 struct trace_probe *tp;
327 tp = trace_probe_primary_from_call(call);
328 if (WARN_ON_ONCE(!tp))
331 return container_of(tp, struct trace_uprobe, tp);
335 * Allocate new trace_uprobe and initialize it (including uprobes).
337 static struct trace_uprobe *
338 alloc_trace_uprobe(const char *group, const char *event, int nargs, bool is_ret)
340 struct trace_uprobe *tu;
343 tu = kzalloc(SIZEOF_TRACE_UPROBE(nargs), GFP_KERNEL);
345 return ERR_PTR(-ENOMEM);
347 ret = trace_probe_init(&tu->tp, event, group, true);
351 dyn_event_init(&tu->devent, &trace_uprobe_ops);
352 tu->consumer.handler = uprobe_dispatcher;
354 tu->consumer.ret_handler = uretprobe_dispatcher;
355 init_trace_uprobe_filter(tu->tp.event->filter);
364 static void free_trace_uprobe(struct trace_uprobe *tu)
370 trace_probe_cleanup(&tu->tp);
375 static struct trace_uprobe *find_probe_event(const char *event, const char *group)
377 struct dyn_event *pos;
378 struct trace_uprobe *tu;
380 for_each_trace_uprobe(tu, pos)
381 if (strcmp(trace_probe_name(&tu->tp), event) == 0 &&
382 strcmp(trace_probe_group_name(&tu->tp), group) == 0)
388 /* Unregister a trace_uprobe and probe_event */
389 static int unregister_trace_uprobe(struct trace_uprobe *tu)
393 if (trace_probe_has_sibling(&tu->tp))
396 /* If there's a reference to the dynamic event */
397 if (trace_event_dyn_busy(trace_probe_event_call(&tu->tp)))
400 ret = unregister_uprobe_event(tu);
405 dyn_event_remove(&tu->devent);
406 trace_probe_unlink(&tu->tp);
407 free_trace_uprobe(tu);
411 static bool trace_uprobe_has_same_uprobe(struct trace_uprobe *orig,
412 struct trace_uprobe *comp)
414 struct trace_probe_event *tpe = orig->tp.event;
415 struct trace_probe *pos;
416 struct inode *comp_inode = d_real_inode(comp->path.dentry);
419 list_for_each_entry(pos, &tpe->probes, list) {
420 orig = container_of(pos, struct trace_uprobe, tp);
421 if (comp_inode != d_real_inode(orig->path.dentry) ||
422 comp->offset != orig->offset)
426 * trace_probe_compare_arg_type() ensured that nr_args and
427 * each argument name and type are same. Let's compare comm.
429 for (i = 0; i < orig->tp.nr_args; i++) {
430 if (strcmp(orig->tp.args[i].comm,
431 comp->tp.args[i].comm))
435 if (i == orig->tp.nr_args)
442 static int append_trace_uprobe(struct trace_uprobe *tu, struct trace_uprobe *to)
446 ret = trace_probe_compare_arg_type(&tu->tp, &to->tp);
448 /* Note that argument starts index = 2 */
449 trace_probe_log_set_index(ret + 1);
450 trace_probe_log_err(0, DIFF_ARG_TYPE);
453 if (trace_uprobe_has_same_uprobe(to, tu)) {
454 trace_probe_log_set_index(0);
455 trace_probe_log_err(0, SAME_PROBE);
459 /* Append to existing event */
460 ret = trace_probe_append(&tu->tp, &to->tp);
462 dyn_event_add(&tu->devent, trace_probe_event_call(&tu->tp));
468 * Uprobe with multiple reference counter is not allowed. i.e.
469 * If inode and offset matches, reference counter offset *must*
470 * match as well. Though, there is one exception: If user is
471 * replacing old trace_uprobe with new one(same group/event),
472 * then we allow same uprobe with new reference counter as far
473 * as the new one does not conflict with any other existing
476 static int validate_ref_ctr_offset(struct trace_uprobe *new)
478 struct dyn_event *pos;
479 struct trace_uprobe *tmp;
480 struct inode *new_inode = d_real_inode(new->path.dentry);
482 for_each_trace_uprobe(tmp, pos) {
483 if (new_inode == d_real_inode(tmp->path.dentry) &&
484 new->offset == tmp->offset &&
485 new->ref_ctr_offset != tmp->ref_ctr_offset) {
486 pr_warn("Reference counter offset mismatch.");
493 /* Register a trace_uprobe and probe_event */
494 static int register_trace_uprobe(struct trace_uprobe *tu)
496 struct trace_uprobe *old_tu;
499 mutex_lock(&event_mutex);
501 ret = validate_ref_ctr_offset(tu);
505 /* register as an event */
506 old_tu = find_probe_event(trace_probe_name(&tu->tp),
507 trace_probe_group_name(&tu->tp));
509 if (is_ret_probe(tu) != is_ret_probe(old_tu)) {
510 trace_probe_log_set_index(0);
511 trace_probe_log_err(0, DIFF_PROBE_TYPE);
514 ret = append_trace_uprobe(tu, old_tu);
519 ret = register_uprobe_event(tu);
521 pr_warn("Failed to register probe event(%d)\n", ret);
525 dyn_event_add(&tu->devent, trace_probe_event_call(&tu->tp));
528 mutex_unlock(&event_mutex);
535 * - Add uprobe: p|r[:[GRP/]EVENT] PATH:OFFSET[%return][(REF)] [FETCHARGS]
537 static int __trace_uprobe_create(int argc, const char **argv)
539 struct trace_uprobe *tu;
540 const char *event = NULL, *group = UPROBE_EVENT_SYSTEM;
541 char *arg, *filename, *rctr, *rctr_end, *tmp;
542 char buf[MAX_EVENT_NAME_LEN];
544 unsigned long offset, ref_ctr_offset;
545 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, buf,
650 goto fail_address_parse;
655 tail = kstrdup(kbasename(filename), GFP_KERNEL);
658 goto fail_address_parse;
661 ptr = strpbrk(tail, ".-_");
665 snprintf(buf, MAX_EVENT_NAME_LEN, "%c_%s_0x%lx", 'p', tail, offset);
673 tu = alloc_trace_uprobe(group, event, argc, is_return);
676 /* This must return -ENOMEM otherwise there is a bug */
677 WARN_ON_ONCE(ret != -ENOMEM);
678 goto fail_address_parse;
681 tu->ref_ctr_offset = ref_ctr_offset;
683 tu->filename = filename;
685 /* parse arguments */
686 for (i = 0; i < argc && i < MAX_TRACE_ARGS; i++) {
687 trace_probe_log_set_index(i + 2);
688 ret = traceprobe_parse_probe_arg(&tu->tp, i, argv[i],
689 is_return ? TPARG_FL_RETURN : 0);
694 ret = traceprobe_set_print_fmt(&tu->tp, is_ret_probe(tu));
698 ret = register_trace_uprobe(tu);
703 free_trace_uprobe(tu);
705 trace_probe_log_clear();
709 trace_probe_log_clear();
716 int trace_uprobe_create(const char *raw_command)
718 return trace_probe_create(raw_command, __trace_uprobe_create);
721 static int create_or_delete_trace_uprobe(const char *raw_command)
725 if (raw_command[0] == '-')
726 return dyn_event_release(raw_command, &trace_uprobe_ops);
728 ret = trace_uprobe_create(raw_command);
729 return ret == -ECANCELED ? -EINVAL : ret;
732 static int trace_uprobe_release(struct dyn_event *ev)
734 struct trace_uprobe *tu = to_trace_uprobe(ev);
736 return unregister_trace_uprobe(tu);
739 /* Probes listing interfaces */
740 static int trace_uprobe_show(struct seq_file *m, struct dyn_event *ev)
742 struct trace_uprobe *tu = to_trace_uprobe(ev);
743 char c = is_ret_probe(tu) ? 'r' : 'p';
746 seq_printf(m, "%c:%s/%s %s:0x%0*lx", c, trace_probe_group_name(&tu->tp),
747 trace_probe_name(&tu->tp), tu->filename,
748 (int)(sizeof(void *) * 2), tu->offset);
750 if (tu->ref_ctr_offset)
751 seq_printf(m, "(0x%lx)", tu->ref_ctr_offset);
753 for (i = 0; i < tu->tp.nr_args; i++)
754 seq_printf(m, " %s=%s", tu->tp.args[i].name, tu->tp.args[i].comm);
760 static int probes_seq_show(struct seq_file *m, void *v)
762 struct dyn_event *ev = v;
764 if (!is_trace_uprobe(ev))
767 return trace_uprobe_show(m, ev);
770 static const struct seq_operations probes_seq_op = {
771 .start = dyn_event_seq_start,
772 .next = dyn_event_seq_next,
773 .stop = dyn_event_seq_stop,
774 .show = probes_seq_show
777 static int probes_open(struct inode *inode, struct file *file)
781 ret = security_locked_down(LOCKDOWN_TRACEFS);
785 if ((file->f_mode & FMODE_WRITE) && (file->f_flags & O_TRUNC)) {
786 ret = dyn_events_release_all(&trace_uprobe_ops);
791 return seq_open(file, &probes_seq_op);
794 static ssize_t probes_write(struct file *file, const char __user *buffer,
795 size_t count, loff_t *ppos)
797 return trace_parse_run_command(file, buffer, count, ppos,
798 create_or_delete_trace_uprobe);
801 static const struct file_operations uprobe_events_ops = {
802 .owner = THIS_MODULE,
806 .release = seq_release,
807 .write = probes_write,
810 /* Probes profiling interfaces */
811 static int probes_profile_seq_show(struct seq_file *m, void *v)
813 struct dyn_event *ev = v;
814 struct trace_uprobe *tu;
816 if (!is_trace_uprobe(ev))
819 tu = to_trace_uprobe(ev);
820 seq_printf(m, " %s %-44s %15lu\n", tu->filename,
821 trace_probe_name(&tu->tp), tu->nhit);
825 static const struct seq_operations profile_seq_op = {
826 .start = dyn_event_seq_start,
827 .next = dyn_event_seq_next,
828 .stop = dyn_event_seq_stop,
829 .show = probes_profile_seq_show
832 static int profile_open(struct inode *inode, struct file *file)
836 ret = security_locked_down(LOCKDOWN_TRACEFS);
840 return seq_open(file, &profile_seq_op);
843 static const struct file_operations uprobe_profile_ops = {
844 .owner = THIS_MODULE,
845 .open = profile_open,
848 .release = seq_release,
851 struct uprobe_cpu_buffer {
855 static struct uprobe_cpu_buffer __percpu *uprobe_cpu_buffer;
856 static int uprobe_buffer_refcnt;
858 static int uprobe_buffer_init(void)
862 uprobe_cpu_buffer = alloc_percpu(struct uprobe_cpu_buffer);
863 if (uprobe_cpu_buffer == NULL)
866 for_each_possible_cpu(cpu) {
867 struct page *p = alloc_pages_node(cpu_to_node(cpu),
873 per_cpu_ptr(uprobe_cpu_buffer, cpu)->buf = page_address(p);
874 mutex_init(&per_cpu_ptr(uprobe_cpu_buffer, cpu)->mutex);
880 for_each_possible_cpu(cpu) {
883 free_page((unsigned long)per_cpu_ptr(uprobe_cpu_buffer, cpu)->buf);
886 free_percpu(uprobe_cpu_buffer);
890 static int uprobe_buffer_enable(void)
894 BUG_ON(!mutex_is_locked(&event_mutex));
896 if (uprobe_buffer_refcnt++ == 0) {
897 ret = uprobe_buffer_init();
899 uprobe_buffer_refcnt--;
905 static void uprobe_buffer_disable(void)
909 BUG_ON(!mutex_is_locked(&event_mutex));
911 if (--uprobe_buffer_refcnt == 0) {
912 for_each_possible_cpu(cpu)
913 free_page((unsigned long)per_cpu_ptr(uprobe_cpu_buffer,
916 free_percpu(uprobe_cpu_buffer);
917 uprobe_cpu_buffer = NULL;
921 static struct uprobe_cpu_buffer *uprobe_buffer_get(void)
923 struct uprobe_cpu_buffer *ucb;
926 cpu = raw_smp_processor_id();
927 ucb = per_cpu_ptr(uprobe_cpu_buffer, cpu);
930 * Use per-cpu buffers for fastest access, but we might migrate
931 * so the mutex makes sure we have sole access to it.
933 mutex_lock(&ucb->mutex);
938 static void uprobe_buffer_put(struct uprobe_cpu_buffer *ucb)
940 mutex_unlock(&ucb->mutex);
943 static void __uprobe_trace_func(struct trace_uprobe *tu,
944 unsigned long func, struct pt_regs *regs,
945 struct uprobe_cpu_buffer *ucb, int dsize,
946 struct trace_event_file *trace_file)
948 struct uprobe_trace_entry_head *entry;
949 struct trace_buffer *buffer;
950 struct ring_buffer_event *event;
953 struct trace_event_call *call = trace_probe_event_call(&tu->tp);
955 WARN_ON(call != trace_file->event_call);
957 if (WARN_ON_ONCE(tu->tp.size + dsize > PAGE_SIZE))
960 if (trace_trigger_soft_disabled(trace_file))
963 esize = SIZEOF_TRACE_ENTRY(is_ret_probe(tu));
964 size = esize + tu->tp.size + dsize;
965 event = trace_event_buffer_lock_reserve(&buffer, trace_file,
966 call->event.type, size, 0);
970 entry = ring_buffer_event_data(event);
971 if (is_ret_probe(tu)) {
972 entry->vaddr[0] = func;
973 entry->vaddr[1] = instruction_pointer(regs);
974 data = DATAOF_TRACE_ENTRY(entry, true);
976 entry->vaddr[0] = instruction_pointer(regs);
977 data = DATAOF_TRACE_ENTRY(entry, false);
980 memcpy(data, ucb->buf, tu->tp.size + dsize);
982 event_trigger_unlock_commit(trace_file, buffer, event, entry, 0);
986 static int uprobe_trace_func(struct trace_uprobe *tu, struct pt_regs *regs,
987 struct uprobe_cpu_buffer *ucb, int dsize)
989 struct event_file_link *link;
991 if (is_ret_probe(tu))
995 trace_probe_for_each_link_rcu(link, &tu->tp)
996 __uprobe_trace_func(tu, 0, regs, ucb, dsize, link->file);
1002 static void uretprobe_trace_func(struct trace_uprobe *tu, unsigned long func,
1003 struct pt_regs *regs,
1004 struct uprobe_cpu_buffer *ucb, int dsize)
1006 struct event_file_link *link;
1009 trace_probe_for_each_link_rcu(link, &tu->tp)
1010 __uprobe_trace_func(tu, func, regs, ucb, dsize, link->file);
1014 /* Event entry printers */
1015 static enum print_line_t
1016 print_uprobe_event(struct trace_iterator *iter, int flags, struct trace_event *event)
1018 struct uprobe_trace_entry_head *entry;
1019 struct trace_seq *s = &iter->seq;
1020 struct trace_uprobe *tu;
1023 entry = (struct uprobe_trace_entry_head *)iter->ent;
1024 tu = trace_uprobe_primary_from_call(
1025 container_of(event, struct trace_event_call, event));
1029 if (is_ret_probe(tu)) {
1030 trace_seq_printf(s, "%s: (0x%lx <- 0x%lx)",
1031 trace_probe_name(&tu->tp),
1032 entry->vaddr[1], entry->vaddr[0]);
1033 data = DATAOF_TRACE_ENTRY(entry, true);
1035 trace_seq_printf(s, "%s: (0x%lx)",
1036 trace_probe_name(&tu->tp),
1038 data = DATAOF_TRACE_ENTRY(entry, false);
1041 if (print_probe_args(s, tu->tp.args, tu->tp.nr_args, data, entry) < 0)
1044 trace_seq_putc(s, '\n');
1047 return trace_handle_return(s);
1050 typedef bool (*filter_func_t)(struct uprobe_consumer *self,
1051 enum uprobe_filter_ctx ctx,
1052 struct mm_struct *mm);
1054 static int trace_uprobe_enable(struct trace_uprobe *tu, filter_func_t filter)
1058 tu->consumer.filter = filter;
1059 tu->inode = d_real_inode(tu->path.dentry);
1061 if (tu->ref_ctr_offset)
1062 ret = uprobe_register_refctr(tu->inode, tu->offset,
1063 tu->ref_ctr_offset, &tu->consumer);
1065 ret = uprobe_register(tu->inode, tu->offset, &tu->consumer);
1073 static void __probe_event_disable(struct trace_probe *tp)
1075 struct trace_probe *pos;
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(pos, trace_probe_probe_list(tp), list) {
1082 tu = container_of(pos, struct trace_uprobe, tp);
1086 uprobe_unregister(tu->inode, tu->offset, &tu->consumer);
1091 static int probe_event_enable(struct trace_event_call *call,
1092 struct trace_event_file *file, filter_func_t filter)
1094 struct trace_probe *pos, *tp;
1095 struct trace_uprobe *tu;
1099 tp = trace_probe_primary_from_call(call);
1100 if (WARN_ON_ONCE(!tp))
1102 enabled = trace_probe_is_enabled(tp);
1104 /* This may also change "enabled" state */
1106 if (trace_probe_test_flag(tp, TP_FLAG_PROFILE))
1109 ret = trace_probe_add_file(tp, file);
1113 if (trace_probe_test_flag(tp, TP_FLAG_TRACE))
1116 trace_probe_set_flag(tp, TP_FLAG_PROFILE);
1119 tu = container_of(tp, struct trace_uprobe, tp);
1120 WARN_ON(!uprobe_filter_is_empty(tu->tp.event->filter));
1125 ret = uprobe_buffer_enable();
1129 list_for_each_entry(pos, trace_probe_probe_list(tp), list) {
1130 tu = container_of(pos, struct trace_uprobe, tp);
1131 ret = trace_uprobe_enable(tu, filter);
1133 __probe_event_disable(tp);
1141 uprobe_buffer_disable();
1145 trace_probe_remove_file(tp, file);
1147 trace_probe_clear_flag(tp, TP_FLAG_PROFILE);
1152 static void probe_event_disable(struct trace_event_call *call,
1153 struct trace_event_file *file)
1155 struct trace_probe *tp;
1157 tp = trace_probe_primary_from_call(call);
1158 if (WARN_ON_ONCE(!tp))
1161 if (!trace_probe_is_enabled(tp))
1165 if (trace_probe_remove_file(tp, file) < 0)
1168 if (trace_probe_is_enabled(tp))
1171 trace_probe_clear_flag(tp, TP_FLAG_PROFILE);
1173 __probe_event_disable(tp);
1174 uprobe_buffer_disable();
1177 static int uprobe_event_define_fields(struct trace_event_call *event_call)
1180 struct uprobe_trace_entry_head field;
1181 struct trace_uprobe *tu;
1183 tu = trace_uprobe_primary_from_call(event_call);
1187 if (is_ret_probe(tu)) {
1188 DEFINE_FIELD(unsigned long, vaddr[0], FIELD_STRING_FUNC, 0);
1189 DEFINE_FIELD(unsigned long, vaddr[1], FIELD_STRING_RETIP, 0);
1190 size = SIZEOF_TRACE_ENTRY(true);
1192 DEFINE_FIELD(unsigned long, vaddr[0], FIELD_STRING_IP, 0);
1193 size = SIZEOF_TRACE_ENTRY(false);
1196 return traceprobe_define_arg_fields(event_call, size, &tu->tp);
1199 #ifdef CONFIG_PERF_EVENTS
1201 __uprobe_perf_filter(struct trace_uprobe_filter *filter, struct mm_struct *mm)
1203 struct perf_event *event;
1205 if (filter->nr_systemwide)
1208 list_for_each_entry(event, &filter->perf_events, hw.tp_list) {
1209 if (event->hw.target->mm == mm)
1217 trace_uprobe_filter_event(struct trace_uprobe_filter *filter,
1218 struct perf_event *event)
1220 return __uprobe_perf_filter(filter, event->hw.target->mm);
1223 static bool trace_uprobe_filter_remove(struct trace_uprobe_filter *filter,
1224 struct perf_event *event)
1228 write_lock(&filter->rwlock);
1229 if (event->hw.target) {
1230 list_del(&event->hw.tp_list);
1231 done = filter->nr_systemwide ||
1232 (event->hw.target->flags & PF_EXITING) ||
1233 trace_uprobe_filter_event(filter, event);
1235 filter->nr_systemwide--;
1236 done = filter->nr_systemwide;
1238 write_unlock(&filter->rwlock);
1243 /* This returns true if the filter always covers target mm */
1244 static bool trace_uprobe_filter_add(struct trace_uprobe_filter *filter,
1245 struct perf_event *event)
1249 write_lock(&filter->rwlock);
1250 if (event->hw.target) {
1252 * event->parent != NULL means copy_process(), we can avoid
1253 * uprobe_apply(). current->mm must be probed and we can rely
1254 * on dup_mmap() which preserves the already installed bp's.
1256 * attr.enable_on_exec means that exec/mmap will install the
1257 * breakpoints we need.
1259 done = filter->nr_systemwide ||
1260 event->parent || event->attr.enable_on_exec ||
1261 trace_uprobe_filter_event(filter, event);
1262 list_add(&event->hw.tp_list, &filter->perf_events);
1264 done = filter->nr_systemwide;
1265 filter->nr_systemwide++;
1267 write_unlock(&filter->rwlock);
1272 static int uprobe_perf_close(struct trace_event_call *call,
1273 struct perf_event *event)
1275 struct trace_probe *pos, *tp;
1276 struct trace_uprobe *tu;
1279 tp = trace_probe_primary_from_call(call);
1280 if (WARN_ON_ONCE(!tp))
1283 tu = container_of(tp, struct trace_uprobe, tp);
1284 if (trace_uprobe_filter_remove(tu->tp.event->filter, event))
1287 list_for_each_entry(pos, trace_probe_probe_list(tp), list) {
1288 tu = container_of(pos, struct trace_uprobe, tp);
1289 ret = uprobe_apply(tu->inode, tu->offset, &tu->consumer, false);
1297 static int uprobe_perf_open(struct trace_event_call *call,
1298 struct perf_event *event)
1300 struct trace_probe *pos, *tp;
1301 struct trace_uprobe *tu;
1304 tp = trace_probe_primary_from_call(call);
1305 if (WARN_ON_ONCE(!tp))
1308 tu = container_of(tp, struct trace_uprobe, tp);
1309 if (trace_uprobe_filter_add(tu->tp.event->filter, event))
1312 list_for_each_entry(pos, trace_probe_probe_list(tp), list) {
1313 err = uprobe_apply(tu->inode, tu->offset, &tu->consumer, true);
1315 uprobe_perf_close(call, event);
1323 static bool uprobe_perf_filter(struct uprobe_consumer *uc,
1324 enum uprobe_filter_ctx ctx, struct mm_struct *mm)
1326 struct trace_uprobe_filter *filter;
1327 struct trace_uprobe *tu;
1330 tu = container_of(uc, struct trace_uprobe, consumer);
1331 filter = tu->tp.event->filter;
1333 read_lock(&filter->rwlock);
1334 ret = __uprobe_perf_filter(filter, mm);
1335 read_unlock(&filter->rwlock);
1340 static void __uprobe_perf_func(struct trace_uprobe *tu,
1341 unsigned long func, struct pt_regs *regs,
1342 struct uprobe_cpu_buffer *ucb, int dsize)
1344 struct trace_event_call *call = trace_probe_event_call(&tu->tp);
1345 struct uprobe_trace_entry_head *entry;
1346 struct hlist_head *head;
1351 if (bpf_prog_array_valid(call)) {
1355 ret = trace_call_bpf(call, regs);
1361 esize = SIZEOF_TRACE_ENTRY(is_ret_probe(tu));
1363 size = esize + tu->tp.size + dsize;
1364 size = ALIGN(size + sizeof(u32), sizeof(u64)) - sizeof(u32);
1365 if (WARN_ONCE(size > PERF_MAX_TRACE_SIZE, "profile buffer not large enough"))
1369 head = this_cpu_ptr(call->perf_events);
1370 if (hlist_empty(head))
1373 entry = perf_trace_buf_alloc(size, NULL, &rctx);
1377 if (is_ret_probe(tu)) {
1378 entry->vaddr[0] = func;
1379 entry->vaddr[1] = instruction_pointer(regs);
1380 data = DATAOF_TRACE_ENTRY(entry, true);
1382 entry->vaddr[0] = instruction_pointer(regs);
1383 data = DATAOF_TRACE_ENTRY(entry, false);
1386 memcpy(data, ucb->buf, tu->tp.size + dsize);
1388 if (size - esize > tu->tp.size + dsize) {
1389 int len = tu->tp.size + dsize;
1391 memset(data + len, 0, size - esize - len);
1394 perf_trace_buf_submit(entry, size, rctx, call->event.type, 1, regs,
1400 /* uprobe profile handler */
1401 static int uprobe_perf_func(struct trace_uprobe *tu, struct pt_regs *regs,
1402 struct uprobe_cpu_buffer *ucb, int dsize)
1404 if (!uprobe_perf_filter(&tu->consumer, 0, current->mm))
1405 return UPROBE_HANDLER_REMOVE;
1407 if (!is_ret_probe(tu))
1408 __uprobe_perf_func(tu, 0, regs, ucb, dsize);
1412 static void uretprobe_perf_func(struct trace_uprobe *tu, unsigned long func,
1413 struct pt_regs *regs,
1414 struct uprobe_cpu_buffer *ucb, int dsize)
1416 __uprobe_perf_func(tu, func, regs, ucb, dsize);
1419 int bpf_get_uprobe_info(const struct perf_event *event, u32 *fd_type,
1420 const char **filename, u64 *probe_offset,
1421 bool perf_type_tracepoint)
1423 const char *pevent = trace_event_name(event->tp_event);
1424 const char *group = event->tp_event->class->system;
1425 struct trace_uprobe *tu;
1427 if (perf_type_tracepoint)
1428 tu = find_probe_event(pevent, group);
1430 tu = trace_uprobe_primary_from_call(event->tp_event);
1434 *fd_type = is_ret_probe(tu) ? BPF_FD_TYPE_URETPROBE
1435 : BPF_FD_TYPE_UPROBE;
1436 *filename = tu->filename;
1437 *probe_offset = tu->offset;
1440 #endif /* CONFIG_PERF_EVENTS */
1443 trace_uprobe_register(struct trace_event_call *event, enum trace_reg type,
1446 struct trace_event_file *file = data;
1449 case TRACE_REG_REGISTER:
1450 return probe_event_enable(event, file, NULL);
1452 case TRACE_REG_UNREGISTER:
1453 probe_event_disable(event, file);
1456 #ifdef CONFIG_PERF_EVENTS
1457 case TRACE_REG_PERF_REGISTER:
1458 return probe_event_enable(event, NULL, uprobe_perf_filter);
1460 case TRACE_REG_PERF_UNREGISTER:
1461 probe_event_disable(event, NULL);
1464 case TRACE_REG_PERF_OPEN:
1465 return uprobe_perf_open(event, data);
1467 case TRACE_REG_PERF_CLOSE:
1468 return uprobe_perf_close(event, data);
1476 static int uprobe_dispatcher(struct uprobe_consumer *con, struct pt_regs *regs)
1478 struct trace_uprobe *tu;
1479 struct uprobe_dispatch_data udd;
1480 struct uprobe_cpu_buffer *ucb;
1485 tu = container_of(con, struct trace_uprobe, consumer);
1489 udd.bp_addr = instruction_pointer(regs);
1491 current->utask->vaddr = (unsigned long) &udd;
1493 if (WARN_ON_ONCE(!uprobe_cpu_buffer))
1496 dsize = __get_data_size(&tu->tp, regs);
1497 esize = SIZEOF_TRACE_ENTRY(is_ret_probe(tu));
1499 ucb = uprobe_buffer_get();
1500 store_trace_args(ucb->buf, &tu->tp, regs, esize, dsize);
1502 if (trace_probe_test_flag(&tu->tp, TP_FLAG_TRACE))
1503 ret |= uprobe_trace_func(tu, regs, ucb, dsize);
1505 #ifdef CONFIG_PERF_EVENTS
1506 if (trace_probe_test_flag(&tu->tp, TP_FLAG_PROFILE))
1507 ret |= uprobe_perf_func(tu, regs, ucb, dsize);
1509 uprobe_buffer_put(ucb);
1513 static int uretprobe_dispatcher(struct uprobe_consumer *con,
1514 unsigned long func, struct pt_regs *regs)
1516 struct trace_uprobe *tu;
1517 struct uprobe_dispatch_data udd;
1518 struct uprobe_cpu_buffer *ucb;
1521 tu = container_of(con, struct trace_uprobe, consumer);
1526 current->utask->vaddr = (unsigned long) &udd;
1528 if (WARN_ON_ONCE(!uprobe_cpu_buffer))
1531 dsize = __get_data_size(&tu->tp, regs);
1532 esize = SIZEOF_TRACE_ENTRY(is_ret_probe(tu));
1534 ucb = uprobe_buffer_get();
1535 store_trace_args(ucb->buf, &tu->tp, regs, esize, dsize);
1537 if (trace_probe_test_flag(&tu->tp, TP_FLAG_TRACE))
1538 uretprobe_trace_func(tu, func, regs, ucb, dsize);
1540 #ifdef CONFIG_PERF_EVENTS
1541 if (trace_probe_test_flag(&tu->tp, TP_FLAG_PROFILE))
1542 uretprobe_perf_func(tu, func, regs, ucb, dsize);
1544 uprobe_buffer_put(ucb);
1548 static struct trace_event_functions uprobe_funcs = {
1549 .trace = print_uprobe_event
1552 static struct trace_event_fields uprobe_fields_array[] = {
1553 { .type = TRACE_FUNCTION_TYPE,
1554 .define_fields = uprobe_event_define_fields },
1558 static inline void init_trace_event_call(struct trace_uprobe *tu)
1560 struct trace_event_call *call = trace_probe_event_call(&tu->tp);
1561 call->event.funcs = &uprobe_funcs;
1562 call->class->fields_array = uprobe_fields_array;
1564 call->flags = TRACE_EVENT_FL_UPROBE | TRACE_EVENT_FL_CAP_ANY;
1565 call->class->reg = trace_uprobe_register;
1568 static int register_uprobe_event(struct trace_uprobe *tu)
1570 init_trace_event_call(tu);
1572 return trace_probe_register_event_call(&tu->tp);
1575 static int unregister_uprobe_event(struct trace_uprobe *tu)
1577 return trace_probe_unregister_event_call(&tu->tp);
1580 #ifdef CONFIG_PERF_EVENTS
1581 struct trace_event_call *
1582 create_local_trace_uprobe(char *name, unsigned long offs,
1583 unsigned long ref_ctr_offset, bool is_return)
1585 struct trace_uprobe *tu;
1589 ret = kern_path(name, LOOKUP_FOLLOW, &path);
1591 return ERR_PTR(ret);
1593 if (!d_is_reg(path.dentry)) {
1595 return ERR_PTR(-EINVAL);
1599 * local trace_kprobes are not added to dyn_event, so they are never
1600 * searched in find_trace_kprobe(). Therefore, there is no concern of
1601 * duplicated name "DUMMY_EVENT" here.
1603 tu = alloc_trace_uprobe(UPROBE_EVENT_SYSTEM, "DUMMY_EVENT", 0,
1607 pr_info("Failed to allocate trace_uprobe.(%d)\n",
1610 return ERR_CAST(tu);
1615 tu->ref_ctr_offset = ref_ctr_offset;
1616 tu->filename = kstrdup(name, GFP_KERNEL);
1617 init_trace_event_call(tu);
1619 if (traceprobe_set_print_fmt(&tu->tp, is_ret_probe(tu)) < 0) {
1624 return trace_probe_event_call(&tu->tp);
1626 free_trace_uprobe(tu);
1627 return ERR_PTR(ret);
1630 void destroy_local_trace_uprobe(struct trace_event_call *event_call)
1632 struct trace_uprobe *tu;
1634 tu = trace_uprobe_primary_from_call(event_call);
1636 free_trace_uprobe(tu);
1638 #endif /* CONFIG_PERF_EVENTS */
1640 /* Make a trace interface for controlling probe points */
1641 static __init int init_uprobe_trace(void)
1645 ret = dyn_event_register(&trace_uprobe_ops);
1649 ret = tracing_init_dentry();
1653 trace_create_file("uprobe_events", 0644, NULL,
1654 NULL, &uprobe_events_ops);
1655 /* Profile interface */
1656 trace_create_file("uprobe_profile", 0444, NULL,
1657 NULL, &uprobe_profile_ops);
1661 fs_initcall(init_uprobe_trace);