2 * uprobes-based tracing events
4 * This program is free software; you can redistribute it and/or modify
5 * it under the terms of the GNU General Public License version 2 as
6 * published by the Free Software Foundation.
8 * This program is distributed in the hope that it will be useful,
9 * but WITHOUT ANY WARRANTY; without even the implied warranty of
10 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
11 * GNU General Public License for more details.
13 * You should have received a copy of the GNU General Public License
14 * along with this program; if not, write to the Free Software
15 * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA
17 * Copyright (C) IBM Corporation, 2010-2012
18 * Author: Srikar Dronamraju <srikar@linux.vnet.ibm.com>
21 #include <linux/module.h>
22 #include <linux/uaccess.h>
23 #include <linux/uprobes.h>
24 #include <linux/namei.h>
25 #include <linux/string.h>
27 #include "trace_probe.h"
29 #define UPROBE_EVENT_SYSTEM "uprobes"
31 struct uprobe_trace_entry_head {
32 struct trace_entry ent;
33 unsigned long vaddr[];
36 #define SIZEOF_TRACE_ENTRY(is_return) \
37 (sizeof(struct uprobe_trace_entry_head) + \
38 sizeof(unsigned long) * (is_return ? 2 : 1))
40 #define DATAOF_TRACE_ENTRY(entry, is_return) \
41 ((void*)(entry) + SIZEOF_TRACE_ENTRY(is_return))
43 struct trace_uprobe_filter {
46 struct list_head perf_events;
50 * uprobe event core functions
53 struct list_head list;
54 struct trace_uprobe_filter filter;
55 struct uprobe_consumer consumer;
60 struct trace_probe tp;
63 #define SIZEOF_TRACE_UPROBE(n) \
64 (offsetof(struct trace_uprobe, tp.args) + \
65 (sizeof(struct probe_arg) * (n)))
67 static int register_uprobe_event(struct trace_uprobe *tu);
68 static int unregister_uprobe_event(struct trace_uprobe *tu);
70 static DEFINE_MUTEX(uprobe_lock);
71 static LIST_HEAD(uprobe_list);
73 struct uprobe_dispatch_data {
74 struct trace_uprobe *tu;
75 unsigned long bp_addr;
78 static int uprobe_dispatcher(struct uprobe_consumer *con, struct pt_regs *regs);
79 static int uretprobe_dispatcher(struct uprobe_consumer *con,
80 unsigned long func, struct pt_regs *regs);
82 #ifdef CONFIG_STACK_GROWSUP
83 static unsigned long adjust_stack_addr(unsigned long addr, unsigned int n)
85 return addr - (n * sizeof(long));
88 static unsigned long adjust_stack_addr(unsigned long addr, unsigned int n)
90 return addr + (n * sizeof(long));
94 static unsigned long get_user_stack_nth(struct pt_regs *regs, unsigned int n)
97 unsigned long addr = user_stack_pointer(regs);
99 addr = adjust_stack_addr(addr, n);
101 if (copy_from_user(&ret, (void __force __user *) addr, sizeof(ret)))
108 * Uprobes-specific fetch functions
110 #define DEFINE_FETCH_stack(type) \
111 static void FETCH_FUNC_NAME(stack, type)(struct pt_regs *regs, \
112 void *offset, void *dest) \
114 *(type *)dest = (type)get_user_stack_nth(regs, \
115 ((unsigned long)offset)); \
117 DEFINE_BASIC_FETCH_FUNCS(stack)
118 /* No string on the stack entry */
119 #define fetch_stack_string NULL
120 #define fetch_stack_string_size NULL
122 #define DEFINE_FETCH_memory(type) \
123 static void FETCH_FUNC_NAME(memory, type)(struct pt_regs *regs, \
124 void *addr, void *dest) \
127 void __user *vaddr = (void __force __user *) addr; \
129 if (copy_from_user(&retval, vaddr, sizeof(type))) \
132 *(type *) dest = retval; \
134 DEFINE_BASIC_FETCH_FUNCS(memory)
136 * Fetch a null-terminated string. Caller MUST set *(u32 *)dest with max
137 * length and relative data location.
139 static void FETCH_FUNC_NAME(memory, string)(struct pt_regs *regs,
140 void *addr, void *dest)
143 u32 rloc = *(u32 *)dest;
144 int maxlen = get_rloc_len(rloc);
145 u8 *dst = get_rloc_data(dest);
146 void __user *src = (void __force __user *) addr;
151 ret = strncpy_from_user(dst, src, maxlen);
153 if (ret < 0) { /* Failed to fetch string */
154 ((u8 *)get_rloc_data(dest))[0] = '\0';
155 *(u32 *)dest = make_data_rloc(0, get_rloc_offs(rloc));
157 *(u32 *)dest = make_data_rloc(ret, get_rloc_offs(rloc));
161 static void FETCH_FUNC_NAME(memory, string_size)(struct pt_regs *regs,
162 void *addr, void *dest)
165 void __user *vaddr = (void __force __user *) addr;
167 len = strnlen_user(vaddr, MAX_STRING_SIZE);
169 if (len == 0 || len > MAX_STRING_SIZE) /* Failed to check length */
175 static unsigned long translate_user_vaddr(void *file_offset)
177 unsigned long base_addr;
178 struct uprobe_dispatch_data *udd;
180 udd = (void *) current->utask->vaddr;
182 base_addr = udd->bp_addr - udd->tu->offset;
183 return base_addr + (unsigned long)file_offset;
186 #define DEFINE_FETCH_file_offset(type) \
187 static void FETCH_FUNC_NAME(file_offset, type)(struct pt_regs *regs, \
188 void *offset, void *dest)\
190 void *vaddr = (void *)translate_user_vaddr(offset); \
192 FETCH_FUNC_NAME(memory, type)(regs, vaddr, dest); \
194 DEFINE_BASIC_FETCH_FUNCS(file_offset)
195 DEFINE_FETCH_file_offset(string)
196 DEFINE_FETCH_file_offset(string_size)
198 /* Fetch type information table */
199 static const struct fetch_type uprobes_fetch_type_table[] = {
201 [FETCH_TYPE_STRING] = __ASSIGN_FETCH_TYPE("string", string, string,
202 sizeof(u32), 1, "__data_loc char[]"),
203 [FETCH_TYPE_STRSIZE] = __ASSIGN_FETCH_TYPE("string_size", u32,
204 string_size, sizeof(u32), 0, "u32"),
206 ASSIGN_FETCH_TYPE(u8, u8, 0),
207 ASSIGN_FETCH_TYPE(u16, u16, 0),
208 ASSIGN_FETCH_TYPE(u32, u32, 0),
209 ASSIGN_FETCH_TYPE(u64, u64, 0),
210 ASSIGN_FETCH_TYPE(s8, u8, 1),
211 ASSIGN_FETCH_TYPE(s16, u16, 1),
212 ASSIGN_FETCH_TYPE(s32, u32, 1),
213 ASSIGN_FETCH_TYPE(s64, u64, 1),
214 ASSIGN_FETCH_TYPE_ALIAS(x8, u8, u8, 0),
215 ASSIGN_FETCH_TYPE_ALIAS(x16, u16, u16, 0),
216 ASSIGN_FETCH_TYPE_ALIAS(x32, u32, u32, 0),
217 ASSIGN_FETCH_TYPE_ALIAS(x64, u64, u64, 0),
219 ASSIGN_FETCH_TYPE_END
222 static inline void init_trace_uprobe_filter(struct trace_uprobe_filter *filter)
224 rwlock_init(&filter->rwlock);
225 filter->nr_systemwide = 0;
226 INIT_LIST_HEAD(&filter->perf_events);
229 static inline bool uprobe_filter_is_empty(struct trace_uprobe_filter *filter)
231 return !filter->nr_systemwide && list_empty(&filter->perf_events);
234 static inline bool is_ret_probe(struct trace_uprobe *tu)
236 return tu->consumer.ret_handler != NULL;
240 * Allocate new trace_uprobe and initialize it (including uprobes).
242 static struct trace_uprobe *
243 alloc_trace_uprobe(const char *group, const char *event, int nargs, bool is_ret)
245 struct trace_uprobe *tu;
247 if (!event || !is_good_name(event))
248 return ERR_PTR(-EINVAL);
250 if (!group || !is_good_name(group))
251 return ERR_PTR(-EINVAL);
253 tu = kzalloc(SIZEOF_TRACE_UPROBE(nargs), GFP_KERNEL);
255 return ERR_PTR(-ENOMEM);
257 tu->tp.call.class = &tu->tp.class;
258 tu->tp.call.name = kstrdup(event, GFP_KERNEL);
259 if (!tu->tp.call.name)
262 tu->tp.class.system = kstrdup(group, GFP_KERNEL);
263 if (!tu->tp.class.system)
266 INIT_LIST_HEAD(&tu->list);
267 INIT_LIST_HEAD(&tu->tp.files);
268 tu->consumer.handler = uprobe_dispatcher;
270 tu->consumer.ret_handler = uretprobe_dispatcher;
271 init_trace_uprobe_filter(&tu->filter);
275 kfree(tu->tp.call.name);
278 return ERR_PTR(-ENOMEM);
281 static void free_trace_uprobe(struct trace_uprobe *tu)
285 for (i = 0; i < tu->tp.nr_args; i++)
286 traceprobe_free_probe_arg(&tu->tp.args[i]);
289 kfree(tu->tp.call.class->system);
290 kfree(tu->tp.call.name);
295 static struct trace_uprobe *find_probe_event(const char *event, const char *group)
297 struct trace_uprobe *tu;
299 list_for_each_entry(tu, &uprobe_list, list)
300 if (strcmp(trace_event_name(&tu->tp.call), event) == 0 &&
301 strcmp(tu->tp.call.class->system, group) == 0)
307 /* Unregister a trace_uprobe and probe_event: call with locking uprobe_lock */
308 static int unregister_trace_uprobe(struct trace_uprobe *tu)
312 ret = unregister_uprobe_event(tu);
317 free_trace_uprobe(tu);
321 /* Register a trace_uprobe and probe_event */
322 static int register_trace_uprobe(struct trace_uprobe *tu)
324 struct trace_uprobe *old_tu;
327 mutex_lock(&uprobe_lock);
329 /* register as an event */
330 old_tu = find_probe_event(trace_event_name(&tu->tp.call),
331 tu->tp.call.class->system);
333 /* delete old event */
334 ret = unregister_trace_uprobe(old_tu);
339 ret = register_uprobe_event(tu);
341 pr_warn("Failed to register probe event(%d)\n", ret);
345 list_add_tail(&tu->list, &uprobe_list);
348 mutex_unlock(&uprobe_lock);
355 * - Add uprobe: p|r[:[GRP/]EVENT] PATH:OFFSET [FETCHARGS]
357 * - Remove uprobe: -:[GRP/]EVENT
359 static int create_trace_uprobe(int argc, char **argv)
361 struct trace_uprobe *tu;
363 char *arg, *event, *group, *filename;
364 char buf[MAX_EVENT_NAME_LEN];
366 unsigned long offset;
367 bool is_delete, is_return;
377 /* argc must be >= 1 */
378 if (argv[0][0] == '-')
380 else if (argv[0][0] == 'r')
382 else if (argv[0][0] != 'p') {
383 pr_info("Probe definition must be started with 'p', 'r' or '-'.\n");
387 if (argv[0][1] == ':') {
389 arg = strchr(event, '/');
396 if (strlen(group) == 0) {
397 pr_info("Group name is not specified\n");
401 if (strlen(event) == 0) {
402 pr_info("Event name is not specified\n");
407 group = UPROBE_EVENT_SYSTEM;
413 pr_info("Delete command needs an event name.\n");
416 mutex_lock(&uprobe_lock);
417 tu = find_probe_event(event, group);
420 mutex_unlock(&uprobe_lock);
421 pr_info("Event %s/%s doesn't exist.\n", group, event);
424 /* delete an event */
425 ret = unregister_trace_uprobe(tu);
426 mutex_unlock(&uprobe_lock);
431 pr_info("Probe point is not specified.\n");
434 if (isdigit(argv[1][0])) {
435 pr_info("probe point must be have a filename.\n");
438 arg = strchr(argv[1], ':');
441 goto fail_address_parse;
446 ret = kern_path(filename, LOOKUP_FOLLOW, &path);
448 goto fail_address_parse;
450 inode = igrab(d_inode(path.dentry));
453 if (!inode || !S_ISREG(inode->i_mode)) {
455 goto fail_address_parse;
458 ret = kstrtoul(arg, 0, &offset);
460 goto fail_address_parse;
470 tail = kstrdup(kbasename(filename), GFP_KERNEL);
473 goto fail_address_parse;
476 ptr = strpbrk(tail, ".-_");
480 snprintf(buf, MAX_EVENT_NAME_LEN, "%c_%s_0x%lx", 'p', tail, offset);
485 tu = alloc_trace_uprobe(group, event, argc, is_return);
487 pr_info("Failed to allocate trace_uprobe.(%d)\n", (int)PTR_ERR(tu));
489 goto fail_address_parse;
493 tu->filename = kstrdup(filename, GFP_KERNEL);
496 pr_info("Failed to allocate filename.\n");
501 /* parse arguments */
503 for (i = 0; i < argc && i < MAX_TRACE_ARGS; i++) {
504 struct probe_arg *parg = &tu->tp.args[i];
506 /* Increment count for freeing args in error case */
509 /* Parse argument name */
510 arg = strchr(argv[i], '=');
513 parg->name = kstrdup(argv[i], GFP_KERNEL);
516 /* If argument name is omitted, set "argN" */
517 snprintf(buf, MAX_EVENT_NAME_LEN, "arg%d", i + 1);
518 parg->name = kstrdup(buf, GFP_KERNEL);
522 pr_info("Failed to allocate argument[%d] name.\n", i);
527 if (!is_good_name(parg->name)) {
528 pr_info("Invalid argument[%d] name: %s\n", i, parg->name);
533 if (traceprobe_conflict_field_name(parg->name, tu->tp.args, i)) {
534 pr_info("Argument[%d] name '%s' conflicts with "
535 "another field.\n", i, argv[i]);
540 /* Parse fetch argument */
541 ret = traceprobe_parse_probe_arg(arg, &tu->tp.size, parg,
543 uprobes_fetch_type_table);
545 pr_info("Parse error at argument[%d]. (%d)\n", i, ret);
550 ret = register_trace_uprobe(tu);
556 free_trace_uprobe(tu);
562 pr_info("Failed to parse address or file.\n");
567 static int cleanup_all_probes(void)
569 struct trace_uprobe *tu;
572 mutex_lock(&uprobe_lock);
573 while (!list_empty(&uprobe_list)) {
574 tu = list_entry(uprobe_list.next, struct trace_uprobe, list);
575 ret = unregister_trace_uprobe(tu);
579 mutex_unlock(&uprobe_lock);
583 /* Probes listing interfaces */
584 static void *probes_seq_start(struct seq_file *m, loff_t *pos)
586 mutex_lock(&uprobe_lock);
587 return seq_list_start(&uprobe_list, *pos);
590 static void *probes_seq_next(struct seq_file *m, void *v, loff_t *pos)
592 return seq_list_next(v, &uprobe_list, pos);
595 static void probes_seq_stop(struct seq_file *m, void *v)
597 mutex_unlock(&uprobe_lock);
600 static int probes_seq_show(struct seq_file *m, void *v)
602 struct trace_uprobe *tu = v;
603 char c = is_ret_probe(tu) ? 'r' : 'p';
606 seq_printf(m, "%c:%s/%s", c, tu->tp.call.class->system,
607 trace_event_name(&tu->tp.call));
608 seq_printf(m, " %s:", tu->filename);
610 /* Don't print "0x (null)" when offset is 0 */
612 seq_printf(m, "0x%p", (void *)tu->offset);
614 switch (sizeof(void *)) {
616 seq_printf(m, "0x00000000");
620 seq_printf(m, "0x0000000000000000");
625 for (i = 0; i < tu->tp.nr_args; i++)
626 seq_printf(m, " %s=%s", tu->tp.args[i].name, tu->tp.args[i].comm);
632 static const struct seq_operations probes_seq_op = {
633 .start = probes_seq_start,
634 .next = probes_seq_next,
635 .stop = probes_seq_stop,
636 .show = probes_seq_show
639 static int probes_open(struct inode *inode, struct file *file)
643 if ((file->f_mode & FMODE_WRITE) && (file->f_flags & O_TRUNC)) {
644 ret = cleanup_all_probes();
649 return seq_open(file, &probes_seq_op);
652 static ssize_t probes_write(struct file *file, const char __user *buffer,
653 size_t count, loff_t *ppos)
655 return traceprobe_probes_write(file, buffer, count, ppos, create_trace_uprobe);
658 static const struct file_operations uprobe_events_ops = {
659 .owner = THIS_MODULE,
663 .release = seq_release,
664 .write = probes_write,
667 /* Probes profiling interfaces */
668 static int probes_profile_seq_show(struct seq_file *m, void *v)
670 struct trace_uprobe *tu = v;
672 seq_printf(m, " %s %-44s %15lu\n", tu->filename,
673 trace_event_name(&tu->tp.call), tu->nhit);
677 static const struct seq_operations profile_seq_op = {
678 .start = probes_seq_start,
679 .next = probes_seq_next,
680 .stop = probes_seq_stop,
681 .show = probes_profile_seq_show
684 static int profile_open(struct inode *inode, struct file *file)
686 return seq_open(file, &profile_seq_op);
689 static const struct file_operations uprobe_profile_ops = {
690 .owner = THIS_MODULE,
691 .open = profile_open,
694 .release = seq_release,
697 struct uprobe_cpu_buffer {
701 static struct uprobe_cpu_buffer __percpu *uprobe_cpu_buffer;
702 static int uprobe_buffer_refcnt;
704 static int uprobe_buffer_init(void)
708 uprobe_cpu_buffer = alloc_percpu(struct uprobe_cpu_buffer);
709 if (uprobe_cpu_buffer == NULL)
712 for_each_possible_cpu(cpu) {
713 struct page *p = alloc_pages_node(cpu_to_node(cpu),
719 per_cpu_ptr(uprobe_cpu_buffer, cpu)->buf = page_address(p);
720 mutex_init(&per_cpu_ptr(uprobe_cpu_buffer, cpu)->mutex);
726 for_each_possible_cpu(cpu) {
729 free_page((unsigned long)per_cpu_ptr(uprobe_cpu_buffer, cpu)->buf);
732 free_percpu(uprobe_cpu_buffer);
736 static int uprobe_buffer_enable(void)
740 BUG_ON(!mutex_is_locked(&event_mutex));
742 if (uprobe_buffer_refcnt++ == 0) {
743 ret = uprobe_buffer_init();
745 uprobe_buffer_refcnt--;
751 static void uprobe_buffer_disable(void)
755 BUG_ON(!mutex_is_locked(&event_mutex));
757 if (--uprobe_buffer_refcnt == 0) {
758 for_each_possible_cpu(cpu)
759 free_page((unsigned long)per_cpu_ptr(uprobe_cpu_buffer,
762 free_percpu(uprobe_cpu_buffer);
763 uprobe_cpu_buffer = NULL;
767 static struct uprobe_cpu_buffer *uprobe_buffer_get(void)
769 struct uprobe_cpu_buffer *ucb;
772 cpu = raw_smp_processor_id();
773 ucb = per_cpu_ptr(uprobe_cpu_buffer, cpu);
776 * Use per-cpu buffers for fastest access, but we might migrate
777 * so the mutex makes sure we have sole access to it.
779 mutex_lock(&ucb->mutex);
784 static void uprobe_buffer_put(struct uprobe_cpu_buffer *ucb)
786 mutex_unlock(&ucb->mutex);
789 static void __uprobe_trace_func(struct trace_uprobe *tu,
790 unsigned long func, struct pt_regs *regs,
791 struct uprobe_cpu_buffer *ucb, int dsize,
792 struct trace_event_file *trace_file)
794 struct uprobe_trace_entry_head *entry;
795 struct ring_buffer_event *event;
796 struct ring_buffer *buffer;
799 struct trace_event_call *call = &tu->tp.call;
801 WARN_ON(call != trace_file->event_call);
803 if (WARN_ON_ONCE(tu->tp.size + dsize > PAGE_SIZE))
806 if (trace_trigger_soft_disabled(trace_file))
809 esize = SIZEOF_TRACE_ENTRY(is_ret_probe(tu));
810 size = esize + tu->tp.size + dsize;
811 event = trace_event_buffer_lock_reserve(&buffer, trace_file,
812 call->event.type, size, 0, 0);
816 entry = ring_buffer_event_data(event);
817 if (is_ret_probe(tu)) {
818 entry->vaddr[0] = func;
819 entry->vaddr[1] = instruction_pointer(regs);
820 data = DATAOF_TRACE_ENTRY(entry, true);
822 entry->vaddr[0] = instruction_pointer(regs);
823 data = DATAOF_TRACE_ENTRY(entry, false);
826 memcpy(data, ucb->buf, tu->tp.size + dsize);
828 event_trigger_unlock_commit(trace_file, buffer, event, entry, 0, 0);
832 static int uprobe_trace_func(struct trace_uprobe *tu, struct pt_regs *regs,
833 struct uprobe_cpu_buffer *ucb, int dsize)
835 struct event_file_link *link;
837 if (is_ret_probe(tu))
841 list_for_each_entry_rcu(link, &tu->tp.files, list)
842 __uprobe_trace_func(tu, 0, regs, ucb, dsize, link->file);
848 static void uretprobe_trace_func(struct trace_uprobe *tu, unsigned long func,
849 struct pt_regs *regs,
850 struct uprobe_cpu_buffer *ucb, int dsize)
852 struct event_file_link *link;
855 list_for_each_entry_rcu(link, &tu->tp.files, list)
856 __uprobe_trace_func(tu, func, regs, ucb, dsize, link->file);
860 /* Event entry printers */
861 static enum print_line_t
862 print_uprobe_event(struct trace_iterator *iter, int flags, struct trace_event *event)
864 struct uprobe_trace_entry_head *entry;
865 struct trace_seq *s = &iter->seq;
866 struct trace_uprobe *tu;
870 entry = (struct uprobe_trace_entry_head *)iter->ent;
871 tu = container_of(event, struct trace_uprobe, tp.call.event);
873 if (is_ret_probe(tu)) {
874 trace_seq_printf(s, "%s: (0x%lx <- 0x%lx)",
875 trace_event_name(&tu->tp.call),
876 entry->vaddr[1], entry->vaddr[0]);
877 data = DATAOF_TRACE_ENTRY(entry, true);
879 trace_seq_printf(s, "%s: (0x%lx)",
880 trace_event_name(&tu->tp.call),
882 data = DATAOF_TRACE_ENTRY(entry, false);
885 for (i = 0; i < tu->tp.nr_args; i++) {
886 struct probe_arg *parg = &tu->tp.args[i];
888 if (!parg->type->print(s, parg->name, data + parg->offset, entry))
892 trace_seq_putc(s, '\n');
895 return trace_handle_return(s);
898 typedef bool (*filter_func_t)(struct uprobe_consumer *self,
899 enum uprobe_filter_ctx ctx,
900 struct mm_struct *mm);
903 probe_event_enable(struct trace_uprobe *tu, struct trace_event_file *file,
904 filter_func_t filter)
906 bool enabled = trace_probe_is_enabled(&tu->tp);
907 struct event_file_link *link = NULL;
911 if (tu->tp.flags & TP_FLAG_PROFILE)
914 link = kmalloc(sizeof(*link), GFP_KERNEL);
919 list_add_tail_rcu(&link->list, &tu->tp.files);
921 tu->tp.flags |= TP_FLAG_TRACE;
923 if (tu->tp.flags & TP_FLAG_TRACE)
926 tu->tp.flags |= TP_FLAG_PROFILE;
929 WARN_ON(!uprobe_filter_is_empty(&tu->filter));
934 ret = uprobe_buffer_enable();
938 tu->consumer.filter = filter;
939 ret = uprobe_register(tu->inode, tu->offset, &tu->consumer);
946 uprobe_buffer_disable();
950 list_del(&link->list);
952 tu->tp.flags &= ~TP_FLAG_TRACE;
954 tu->tp.flags &= ~TP_FLAG_PROFILE;
960 probe_event_disable(struct trace_uprobe *tu, struct trace_event_file *file)
962 if (!trace_probe_is_enabled(&tu->tp))
966 struct event_file_link *link;
968 link = find_event_file_link(&tu->tp, file);
972 list_del_rcu(&link->list);
973 /* synchronize with u{,ret}probe_trace_func */
977 if (!list_empty(&tu->tp.files))
981 WARN_ON(!uprobe_filter_is_empty(&tu->filter));
983 uprobe_unregister(tu->inode, tu->offset, &tu->consumer);
984 tu->tp.flags &= file ? ~TP_FLAG_TRACE : ~TP_FLAG_PROFILE;
986 uprobe_buffer_disable();
989 static int uprobe_event_define_fields(struct trace_event_call *event_call)
992 struct uprobe_trace_entry_head field;
993 struct trace_uprobe *tu = event_call->data;
995 if (is_ret_probe(tu)) {
996 DEFINE_FIELD(unsigned long, vaddr[0], FIELD_STRING_FUNC, 0);
997 DEFINE_FIELD(unsigned long, vaddr[1], FIELD_STRING_RETIP, 0);
998 size = SIZEOF_TRACE_ENTRY(true);
1000 DEFINE_FIELD(unsigned long, vaddr[0], FIELD_STRING_IP, 0);
1001 size = SIZEOF_TRACE_ENTRY(false);
1003 /* Set argument names as fields */
1004 for (i = 0; i < tu->tp.nr_args; i++) {
1005 struct probe_arg *parg = &tu->tp.args[i];
1007 ret = trace_define_field(event_call, parg->type->fmttype,
1008 parg->name, size + parg->offset,
1009 parg->type->size, parg->type->is_signed,
1018 #ifdef CONFIG_PERF_EVENTS
1020 __uprobe_perf_filter(struct trace_uprobe_filter *filter, struct mm_struct *mm)
1022 struct perf_event *event;
1024 if (filter->nr_systemwide)
1027 list_for_each_entry(event, &filter->perf_events, hw.tp_list) {
1028 if (event->hw.target->mm == mm)
1036 uprobe_filter_event(struct trace_uprobe *tu, struct perf_event *event)
1038 return __uprobe_perf_filter(&tu->filter, event->hw.target->mm);
1041 static int uprobe_perf_close(struct trace_uprobe *tu, struct perf_event *event)
1045 write_lock(&tu->filter.rwlock);
1046 if (event->hw.target) {
1047 list_del(&event->hw.tp_list);
1048 done = tu->filter.nr_systemwide ||
1049 (event->hw.target->flags & PF_EXITING) ||
1050 uprobe_filter_event(tu, event);
1052 tu->filter.nr_systemwide--;
1053 done = tu->filter.nr_systemwide;
1055 write_unlock(&tu->filter.rwlock);
1058 return uprobe_apply(tu->inode, tu->offset, &tu->consumer, false);
1063 static int uprobe_perf_open(struct trace_uprobe *tu, struct perf_event *event)
1068 write_lock(&tu->filter.rwlock);
1069 if (event->hw.target) {
1071 * event->parent != NULL means copy_process(), we can avoid
1072 * uprobe_apply(). current->mm must be probed and we can rely
1073 * on dup_mmap() which preserves the already installed bp's.
1075 * attr.enable_on_exec means that exec/mmap will install the
1076 * breakpoints we need.
1078 done = tu->filter.nr_systemwide ||
1079 event->parent || event->attr.enable_on_exec ||
1080 uprobe_filter_event(tu, event);
1081 list_add(&event->hw.tp_list, &tu->filter.perf_events);
1083 done = tu->filter.nr_systemwide;
1084 tu->filter.nr_systemwide++;
1086 write_unlock(&tu->filter.rwlock);
1090 err = uprobe_apply(tu->inode, tu->offset, &tu->consumer, true);
1092 uprobe_perf_close(tu, event);
1097 static bool uprobe_perf_filter(struct uprobe_consumer *uc,
1098 enum uprobe_filter_ctx ctx, struct mm_struct *mm)
1100 struct trace_uprobe *tu;
1103 tu = container_of(uc, struct trace_uprobe, consumer);
1104 read_lock(&tu->filter.rwlock);
1105 ret = __uprobe_perf_filter(&tu->filter, mm);
1106 read_unlock(&tu->filter.rwlock);
1111 static void __uprobe_perf_func(struct trace_uprobe *tu,
1112 unsigned long func, struct pt_regs *regs,
1113 struct uprobe_cpu_buffer *ucb, int dsize)
1115 struct trace_event_call *call = &tu->tp.call;
1116 struct uprobe_trace_entry_head *entry;
1117 struct bpf_prog *prog = call->prog;
1118 struct hlist_head *head;
1123 if (prog && !trace_call_bpf(prog, regs))
1126 esize = SIZEOF_TRACE_ENTRY(is_ret_probe(tu));
1128 size = esize + tu->tp.size + dsize;
1129 size = ALIGN(size + sizeof(u32), sizeof(u64)) - sizeof(u32);
1130 if (WARN_ONCE(size > PERF_MAX_TRACE_SIZE, "profile buffer not large enough"))
1134 head = this_cpu_ptr(call->perf_events);
1135 if (hlist_empty(head))
1138 entry = perf_trace_buf_alloc(size, NULL, &rctx);
1142 if (is_ret_probe(tu)) {
1143 entry->vaddr[0] = func;
1144 entry->vaddr[1] = instruction_pointer(regs);
1145 data = DATAOF_TRACE_ENTRY(entry, true);
1147 entry->vaddr[0] = instruction_pointer(regs);
1148 data = DATAOF_TRACE_ENTRY(entry, false);
1151 memcpy(data, ucb->buf, tu->tp.size + dsize);
1153 if (size - esize > tu->tp.size + dsize) {
1154 int len = tu->tp.size + dsize;
1156 memset(data + len, 0, size - esize - len);
1159 perf_trace_buf_submit(entry, size, rctx, call->event.type, 1, regs,
1165 /* uprobe profile handler */
1166 static int uprobe_perf_func(struct trace_uprobe *tu, struct pt_regs *regs,
1167 struct uprobe_cpu_buffer *ucb, int dsize)
1169 if (!uprobe_perf_filter(&tu->consumer, 0, current->mm))
1170 return UPROBE_HANDLER_REMOVE;
1172 if (!is_ret_probe(tu))
1173 __uprobe_perf_func(tu, 0, regs, ucb, dsize);
1177 static void uretprobe_perf_func(struct trace_uprobe *tu, unsigned long func,
1178 struct pt_regs *regs,
1179 struct uprobe_cpu_buffer *ucb, int dsize)
1181 __uprobe_perf_func(tu, func, regs, ucb, dsize);
1183 #endif /* CONFIG_PERF_EVENTS */
1186 trace_uprobe_register(struct trace_event_call *event, enum trace_reg type,
1189 struct trace_uprobe *tu = event->data;
1190 struct trace_event_file *file = data;
1193 case TRACE_REG_REGISTER:
1194 return probe_event_enable(tu, file, NULL);
1196 case TRACE_REG_UNREGISTER:
1197 probe_event_disable(tu, file);
1200 #ifdef CONFIG_PERF_EVENTS
1201 case TRACE_REG_PERF_REGISTER:
1202 return probe_event_enable(tu, NULL, uprobe_perf_filter);
1204 case TRACE_REG_PERF_UNREGISTER:
1205 probe_event_disable(tu, NULL);
1208 case TRACE_REG_PERF_OPEN:
1209 return uprobe_perf_open(tu, data);
1211 case TRACE_REG_PERF_CLOSE:
1212 return uprobe_perf_close(tu, data);
1221 static int uprobe_dispatcher(struct uprobe_consumer *con, struct pt_regs *regs)
1223 struct trace_uprobe *tu;
1224 struct uprobe_dispatch_data udd;
1225 struct uprobe_cpu_buffer *ucb;
1230 tu = container_of(con, struct trace_uprobe, consumer);
1234 udd.bp_addr = instruction_pointer(regs);
1236 current->utask->vaddr = (unsigned long) &udd;
1238 if (WARN_ON_ONCE(!uprobe_cpu_buffer))
1241 dsize = __get_data_size(&tu->tp, regs);
1242 esize = SIZEOF_TRACE_ENTRY(is_ret_probe(tu));
1244 ucb = uprobe_buffer_get();
1245 store_trace_args(esize, &tu->tp, regs, ucb->buf, dsize);
1247 if (tu->tp.flags & TP_FLAG_TRACE)
1248 ret |= uprobe_trace_func(tu, regs, ucb, dsize);
1250 #ifdef CONFIG_PERF_EVENTS
1251 if (tu->tp.flags & TP_FLAG_PROFILE)
1252 ret |= uprobe_perf_func(tu, regs, ucb, dsize);
1254 uprobe_buffer_put(ucb);
1258 static int uretprobe_dispatcher(struct uprobe_consumer *con,
1259 unsigned long func, struct pt_regs *regs)
1261 struct trace_uprobe *tu;
1262 struct uprobe_dispatch_data udd;
1263 struct uprobe_cpu_buffer *ucb;
1266 tu = container_of(con, struct trace_uprobe, consumer);
1271 current->utask->vaddr = (unsigned long) &udd;
1273 if (WARN_ON_ONCE(!uprobe_cpu_buffer))
1276 dsize = __get_data_size(&tu->tp, regs);
1277 esize = SIZEOF_TRACE_ENTRY(is_ret_probe(tu));
1279 ucb = uprobe_buffer_get();
1280 store_trace_args(esize, &tu->tp, regs, ucb->buf, dsize);
1282 if (tu->tp.flags & TP_FLAG_TRACE)
1283 uretprobe_trace_func(tu, func, regs, ucb, dsize);
1285 #ifdef CONFIG_PERF_EVENTS
1286 if (tu->tp.flags & TP_FLAG_PROFILE)
1287 uretprobe_perf_func(tu, func, regs, ucb, dsize);
1289 uprobe_buffer_put(ucb);
1293 static struct trace_event_functions uprobe_funcs = {
1294 .trace = print_uprobe_event
1297 static int register_uprobe_event(struct trace_uprobe *tu)
1299 struct trace_event_call *call = &tu->tp.call;
1302 /* Initialize trace_event_call */
1303 INIT_LIST_HEAD(&call->class->fields);
1304 call->event.funcs = &uprobe_funcs;
1305 call->class->define_fields = uprobe_event_define_fields;
1307 if (set_print_fmt(&tu->tp, is_ret_probe(tu)) < 0)
1310 ret = register_trace_event(&call->event);
1312 kfree(call->print_fmt);
1316 call->flags = TRACE_EVENT_FL_UPROBE;
1317 call->class->reg = trace_uprobe_register;
1319 ret = trace_add_event_call(call);
1322 pr_info("Failed to register uprobe event: %s\n",
1323 trace_event_name(call));
1324 kfree(call->print_fmt);
1325 unregister_trace_event(&call->event);
1331 static int unregister_uprobe_event(struct trace_uprobe *tu)
1335 /* tu->event is unregistered in trace_remove_event_call() */
1336 ret = trace_remove_event_call(&tu->tp.call);
1339 kfree(tu->tp.call.print_fmt);
1340 tu->tp.call.print_fmt = NULL;
1344 /* Make a trace interface for controling probe points */
1345 static __init int init_uprobe_trace(void)
1347 struct dentry *d_tracer;
1349 d_tracer = tracing_init_dentry();
1350 if (IS_ERR(d_tracer))
1353 trace_create_file("uprobe_events", 0644, d_tracer,
1354 NULL, &uprobe_events_ops);
1355 /* Profile interface */
1356 trace_create_file("uprobe_profile", 0444, d_tracer,
1357 NULL, &uprobe_profile_ops);
1361 fs_initcall(init_uprobe_trace);