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>
20 #define pr_fmt(fmt) "trace_uprobe: " fmt
22 #include <linux/module.h>
23 #include <linux/uaccess.h>
24 #include <linux/uprobes.h>
25 #include <linux/namei.h>
26 #include <linux/string.h>
27 #include <linux/rculist.h>
29 #include "trace_probe.h"
31 #define UPROBE_EVENT_SYSTEM "uprobes"
33 struct uprobe_trace_entry_head {
34 struct trace_entry ent;
35 unsigned long vaddr[];
38 #define SIZEOF_TRACE_ENTRY(is_return) \
39 (sizeof(struct uprobe_trace_entry_head) + \
40 sizeof(unsigned long) * (is_return ? 2 : 1))
42 #define DATAOF_TRACE_ENTRY(entry, is_return) \
43 ((void*)(entry) + SIZEOF_TRACE_ENTRY(is_return))
45 struct trace_uprobe_filter {
48 struct list_head perf_events;
52 * uprobe event core functions
55 struct list_head list;
56 struct trace_uprobe_filter filter;
57 struct uprobe_consumer consumer;
63 struct trace_probe tp;
66 #define SIZEOF_TRACE_UPROBE(n) \
67 (offsetof(struct trace_uprobe, tp.args) + \
68 (sizeof(struct probe_arg) * (n)))
70 static int register_uprobe_event(struct trace_uprobe *tu);
71 static int unregister_uprobe_event(struct trace_uprobe *tu);
73 static DEFINE_MUTEX(uprobe_lock);
74 static LIST_HEAD(uprobe_list);
76 struct uprobe_dispatch_data {
77 struct trace_uprobe *tu;
78 unsigned long bp_addr;
81 static int uprobe_dispatcher(struct uprobe_consumer *con, struct pt_regs *regs);
82 static int uretprobe_dispatcher(struct uprobe_consumer *con,
83 unsigned long func, struct pt_regs *regs);
85 #ifdef CONFIG_STACK_GROWSUP
86 static unsigned long adjust_stack_addr(unsigned long addr, unsigned int n)
88 return addr - (n * sizeof(long));
91 static unsigned long adjust_stack_addr(unsigned long addr, unsigned int n)
93 return addr + (n * sizeof(long));
97 static unsigned long get_user_stack_nth(struct pt_regs *regs, unsigned int n)
100 unsigned long addr = user_stack_pointer(regs);
102 addr = adjust_stack_addr(addr, n);
104 if (copy_from_user(&ret, (void __force __user *) addr, sizeof(ret)))
111 * Uprobes-specific fetch functions
113 #define DEFINE_FETCH_stack(type) \
114 static void FETCH_FUNC_NAME(stack, type)(struct pt_regs *regs, \
115 void *offset, void *dest) \
117 *(type *)dest = (type)get_user_stack_nth(regs, \
118 ((unsigned long)offset)); \
120 DEFINE_BASIC_FETCH_FUNCS(stack)
121 /* No string on the stack entry */
122 #define fetch_stack_string NULL
123 #define fetch_stack_string_size NULL
125 #define DEFINE_FETCH_memory(type) \
126 static void FETCH_FUNC_NAME(memory, type)(struct pt_regs *regs, \
127 void *addr, void *dest) \
130 void __user *vaddr = (void __force __user *) addr; \
132 if (copy_from_user(&retval, vaddr, sizeof(type))) \
135 *(type *) dest = retval; \
137 DEFINE_BASIC_FETCH_FUNCS(memory)
139 * Fetch a null-terminated string. Caller MUST set *(u32 *)dest with max
140 * length and relative data location.
142 static void FETCH_FUNC_NAME(memory, string)(struct pt_regs *regs,
143 void *addr, void *dest)
146 u32 rloc = *(u32 *)dest;
147 int maxlen = get_rloc_len(rloc);
148 u8 *dst = get_rloc_data(dest);
149 void __user *src = (void __force __user *) addr;
154 ret = strncpy_from_user(dst, src, maxlen);
159 * Include the terminating null byte. In this case it
160 * was copied by strncpy_from_user but not accounted
165 if (ret < 0) { /* Failed to fetch string */
166 ((u8 *)get_rloc_data(dest))[0] = '\0';
167 *(u32 *)dest = make_data_rloc(0, get_rloc_offs(rloc));
169 *(u32 *)dest = make_data_rloc(ret, get_rloc_offs(rloc));
173 static void FETCH_FUNC_NAME(memory, string_size)(struct pt_regs *regs,
174 void *addr, void *dest)
177 void __user *vaddr = (void __force __user *) addr;
179 len = strnlen_user(vaddr, MAX_STRING_SIZE);
181 if (len == 0 || len > MAX_STRING_SIZE) /* Failed to check length */
187 static unsigned long translate_user_vaddr(void *file_offset)
189 unsigned long base_addr;
190 struct uprobe_dispatch_data *udd;
192 udd = (void *) current->utask->vaddr;
194 base_addr = udd->bp_addr - udd->tu->offset;
195 return base_addr + (unsigned long)file_offset;
198 #define DEFINE_FETCH_file_offset(type) \
199 static void FETCH_FUNC_NAME(file_offset, type)(struct pt_regs *regs, \
200 void *offset, void *dest)\
202 void *vaddr = (void *)translate_user_vaddr(offset); \
204 FETCH_FUNC_NAME(memory, type)(regs, vaddr, dest); \
206 DEFINE_BASIC_FETCH_FUNCS(file_offset)
207 DEFINE_FETCH_file_offset(string)
208 DEFINE_FETCH_file_offset(string_size)
210 /* Fetch type information table */
211 static const struct fetch_type uprobes_fetch_type_table[] = {
213 [FETCH_TYPE_STRING] = __ASSIGN_FETCH_TYPE("string", string, string,
214 sizeof(u32), 1, "__data_loc char[]"),
215 [FETCH_TYPE_STRSIZE] = __ASSIGN_FETCH_TYPE("string_size", u32,
216 string_size, sizeof(u32), 0, "u32"),
218 ASSIGN_FETCH_TYPE(u8, u8, 0),
219 ASSIGN_FETCH_TYPE(u16, u16, 0),
220 ASSIGN_FETCH_TYPE(u32, u32, 0),
221 ASSIGN_FETCH_TYPE(u64, u64, 0),
222 ASSIGN_FETCH_TYPE(s8, u8, 1),
223 ASSIGN_FETCH_TYPE(s16, u16, 1),
224 ASSIGN_FETCH_TYPE(s32, u32, 1),
225 ASSIGN_FETCH_TYPE(s64, u64, 1),
226 ASSIGN_FETCH_TYPE_ALIAS(x8, u8, u8, 0),
227 ASSIGN_FETCH_TYPE_ALIAS(x16, u16, u16, 0),
228 ASSIGN_FETCH_TYPE_ALIAS(x32, u32, u32, 0),
229 ASSIGN_FETCH_TYPE_ALIAS(x64, u64, u64, 0),
231 ASSIGN_FETCH_TYPE_END
234 static inline void init_trace_uprobe_filter(struct trace_uprobe_filter *filter)
236 rwlock_init(&filter->rwlock);
237 filter->nr_systemwide = 0;
238 INIT_LIST_HEAD(&filter->perf_events);
241 static inline bool uprobe_filter_is_empty(struct trace_uprobe_filter *filter)
243 return !filter->nr_systemwide && list_empty(&filter->perf_events);
246 static inline bool is_ret_probe(struct trace_uprobe *tu)
248 return tu->consumer.ret_handler != NULL;
252 * Allocate new trace_uprobe and initialize it (including uprobes).
254 static struct trace_uprobe *
255 alloc_trace_uprobe(const char *group, const char *event, int nargs, bool is_ret)
257 struct trace_uprobe *tu;
259 if (!event || !is_good_name(event))
260 return ERR_PTR(-EINVAL);
262 if (!group || !is_good_name(group))
263 return ERR_PTR(-EINVAL);
265 tu = kzalloc(SIZEOF_TRACE_UPROBE(nargs), GFP_KERNEL);
267 return ERR_PTR(-ENOMEM);
269 tu->tp.call.class = &tu->tp.class;
270 tu->tp.call.name = kstrdup(event, GFP_KERNEL);
271 if (!tu->tp.call.name)
274 tu->tp.class.system = kstrdup(group, GFP_KERNEL);
275 if (!tu->tp.class.system)
278 INIT_LIST_HEAD(&tu->list);
279 INIT_LIST_HEAD(&tu->tp.files);
280 tu->consumer.handler = uprobe_dispatcher;
282 tu->consumer.ret_handler = uretprobe_dispatcher;
283 init_trace_uprobe_filter(&tu->filter);
287 kfree(tu->tp.call.name);
290 return ERR_PTR(-ENOMEM);
293 static void free_trace_uprobe(struct trace_uprobe *tu)
297 for (i = 0; i < tu->tp.nr_args; i++)
298 traceprobe_free_probe_arg(&tu->tp.args[i]);
301 kfree(tu->tp.call.class->system);
302 kfree(tu->tp.call.name);
307 static struct trace_uprobe *find_probe_event(const char *event, const char *group)
309 struct trace_uprobe *tu;
311 list_for_each_entry(tu, &uprobe_list, list)
312 if (strcmp(trace_event_name(&tu->tp.call), event) == 0 &&
313 strcmp(tu->tp.call.class->system, group) == 0)
319 /* Unregister a trace_uprobe and probe_event: call with locking uprobe_lock */
320 static int unregister_trace_uprobe(struct trace_uprobe *tu)
324 ret = unregister_uprobe_event(tu);
329 free_trace_uprobe(tu);
333 /* Register a trace_uprobe and probe_event */
334 static int register_trace_uprobe(struct trace_uprobe *tu)
336 struct trace_uprobe *old_tu;
339 mutex_lock(&uprobe_lock);
341 /* register as an event */
342 old_tu = find_probe_event(trace_event_name(&tu->tp.call),
343 tu->tp.call.class->system);
345 /* delete old event */
346 ret = unregister_trace_uprobe(old_tu);
351 ret = register_uprobe_event(tu);
353 pr_warn("Failed to register probe event(%d)\n", ret);
357 list_add_tail(&tu->list, &uprobe_list);
360 mutex_unlock(&uprobe_lock);
367 * - Add uprobe: p|r[:[GRP/]EVENT] PATH:OFFSET [FETCHARGS]
369 * - Remove uprobe: -:[GRP/]EVENT
371 static int create_trace_uprobe(int argc, char **argv)
373 struct trace_uprobe *tu;
374 char *arg, *event, *group, *filename;
375 char buf[MAX_EVENT_NAME_LEN];
377 unsigned long offset;
378 bool is_delete, is_return;
387 /* argc must be >= 1 */
388 if (argv[0][0] == '-')
390 else if (argv[0][0] == 'r')
392 else if (argv[0][0] != 'p') {
393 pr_info("Probe definition must be started with 'p', 'r' or '-'.\n");
397 if (argv[0][1] == ':') {
399 arg = strchr(event, '/');
406 if (strlen(group) == 0) {
407 pr_info("Group name is not specified\n");
411 if (strlen(event) == 0) {
412 pr_info("Event name is not specified\n");
417 group = UPROBE_EVENT_SYSTEM;
423 pr_info("Delete command needs an event name.\n");
426 mutex_lock(&uprobe_lock);
427 tu = find_probe_event(event, group);
430 mutex_unlock(&uprobe_lock);
431 pr_info("Event %s/%s doesn't exist.\n", group, event);
434 /* delete an event */
435 ret = unregister_trace_uprobe(tu);
436 mutex_unlock(&uprobe_lock);
441 pr_info("Probe point is not specified.\n");
444 /* Find the last occurrence, in case the path contains ':' too. */
445 arg = strrchr(argv[1], ':');
451 ret = kern_path(filename, LOOKUP_FOLLOW, &path);
455 if (!d_is_reg(path.dentry)) {
457 goto fail_address_parse;
460 ret = kstrtoul(arg, 0, &offset);
462 goto fail_address_parse;
472 tail = kstrdup(kbasename(filename), GFP_KERNEL);
475 goto fail_address_parse;
478 ptr = strpbrk(tail, ".-_");
482 snprintf(buf, MAX_EVENT_NAME_LEN, "%c_%s_0x%lx", 'p', tail, offset);
487 tu = alloc_trace_uprobe(group, event, argc, is_return);
489 pr_info("Failed to allocate trace_uprobe.(%d)\n", (int)PTR_ERR(tu));
491 goto fail_address_parse;
495 tu->filename = kstrdup(filename, GFP_KERNEL);
498 pr_info("Failed to allocate filename.\n");
503 /* parse arguments */
505 for (i = 0; i < argc && i < MAX_TRACE_ARGS; i++) {
506 struct probe_arg *parg = &tu->tp.args[i];
508 /* Increment count for freeing args in error case */
511 /* Parse argument name */
512 arg = strchr(argv[i], '=');
515 parg->name = kstrdup(argv[i], GFP_KERNEL);
518 /* If argument name is omitted, set "argN" */
519 snprintf(buf, MAX_EVENT_NAME_LEN, "arg%d", i + 1);
520 parg->name = kstrdup(buf, GFP_KERNEL);
524 pr_info("Failed to allocate argument[%d] name.\n", i);
529 if (!is_good_name(parg->name)) {
530 pr_info("Invalid argument[%d] name: %s\n", i, parg->name);
535 if (traceprobe_conflict_field_name(parg->name, tu->tp.args, i)) {
536 pr_info("Argument[%d] name '%s' conflicts with "
537 "another field.\n", i, argv[i]);
542 /* Parse fetch argument */
543 ret = traceprobe_parse_probe_arg(arg, &tu->tp.size, parg,
545 uprobes_fetch_type_table);
547 pr_info("Parse error at argument[%d]. (%d)\n", i, ret);
552 ret = register_trace_uprobe(tu);
558 free_trace_uprobe(tu);
564 pr_info("Failed to parse address or file.\n");
569 static int cleanup_all_probes(void)
571 struct trace_uprobe *tu;
574 mutex_lock(&uprobe_lock);
575 while (!list_empty(&uprobe_list)) {
576 tu = list_entry(uprobe_list.next, struct trace_uprobe, list);
577 ret = unregister_trace_uprobe(tu);
581 mutex_unlock(&uprobe_lock);
585 /* Probes listing interfaces */
586 static void *probes_seq_start(struct seq_file *m, loff_t *pos)
588 mutex_lock(&uprobe_lock);
589 return seq_list_start(&uprobe_list, *pos);
592 static void *probes_seq_next(struct seq_file *m, void *v, loff_t *pos)
594 return seq_list_next(v, &uprobe_list, pos);
597 static void probes_seq_stop(struct seq_file *m, void *v)
599 mutex_unlock(&uprobe_lock);
602 static int probes_seq_show(struct seq_file *m, void *v)
604 struct trace_uprobe *tu = v;
605 char c = is_ret_probe(tu) ? 'r' : 'p';
608 seq_printf(m, "%c:%s/%s", c, tu->tp.call.class->system,
609 trace_event_name(&tu->tp.call));
610 seq_printf(m, " %s:", tu->filename);
612 /* Don't print "0x (null)" when offset is 0 */
614 seq_printf(m, "0x%p", (void *)tu->offset);
616 switch (sizeof(void *)) {
618 seq_printf(m, "0x00000000");
622 seq_printf(m, "0x0000000000000000");
627 for (i = 0; i < tu->tp.nr_args; i++)
628 seq_printf(m, " %s=%s", tu->tp.args[i].name, tu->tp.args[i].comm);
634 static const struct seq_operations probes_seq_op = {
635 .start = probes_seq_start,
636 .next = probes_seq_next,
637 .stop = probes_seq_stop,
638 .show = probes_seq_show
641 static int probes_open(struct inode *inode, struct file *file)
645 if ((file->f_mode & FMODE_WRITE) && (file->f_flags & O_TRUNC)) {
646 ret = cleanup_all_probes();
651 return seq_open(file, &probes_seq_op);
654 static ssize_t probes_write(struct file *file, const char __user *buffer,
655 size_t count, loff_t *ppos)
657 return traceprobe_probes_write(file, buffer, count, ppos, create_trace_uprobe);
660 static const struct file_operations uprobe_events_ops = {
661 .owner = THIS_MODULE,
665 .release = seq_release,
666 .write = probes_write,
669 /* Probes profiling interfaces */
670 static int probes_profile_seq_show(struct seq_file *m, void *v)
672 struct trace_uprobe *tu = v;
674 seq_printf(m, " %s %-44s %15lu\n", tu->filename,
675 trace_event_name(&tu->tp.call), tu->nhit);
679 static const struct seq_operations profile_seq_op = {
680 .start = probes_seq_start,
681 .next = probes_seq_next,
682 .stop = probes_seq_stop,
683 .show = probes_profile_seq_show
686 static int profile_open(struct inode *inode, struct file *file)
688 return seq_open(file, &profile_seq_op);
691 static const struct file_operations uprobe_profile_ops = {
692 .owner = THIS_MODULE,
693 .open = profile_open,
696 .release = seq_release,
699 struct uprobe_cpu_buffer {
703 static struct uprobe_cpu_buffer __percpu *uprobe_cpu_buffer;
704 static int uprobe_buffer_refcnt;
706 static int uprobe_buffer_init(void)
710 uprobe_cpu_buffer = alloc_percpu(struct uprobe_cpu_buffer);
711 if (uprobe_cpu_buffer == NULL)
714 for_each_possible_cpu(cpu) {
715 struct page *p = alloc_pages_node(cpu_to_node(cpu),
721 per_cpu_ptr(uprobe_cpu_buffer, cpu)->buf = page_address(p);
722 mutex_init(&per_cpu_ptr(uprobe_cpu_buffer, cpu)->mutex);
728 for_each_possible_cpu(cpu) {
731 free_page((unsigned long)per_cpu_ptr(uprobe_cpu_buffer, cpu)->buf);
734 free_percpu(uprobe_cpu_buffer);
738 static int uprobe_buffer_enable(void)
742 BUG_ON(!mutex_is_locked(&event_mutex));
744 if (uprobe_buffer_refcnt++ == 0) {
745 ret = uprobe_buffer_init();
747 uprobe_buffer_refcnt--;
753 static void uprobe_buffer_disable(void)
757 BUG_ON(!mutex_is_locked(&event_mutex));
759 if (--uprobe_buffer_refcnt == 0) {
760 for_each_possible_cpu(cpu)
761 free_page((unsigned long)per_cpu_ptr(uprobe_cpu_buffer,
764 free_percpu(uprobe_cpu_buffer);
765 uprobe_cpu_buffer = NULL;
769 static struct uprobe_cpu_buffer *uprobe_buffer_get(void)
771 struct uprobe_cpu_buffer *ucb;
774 cpu = raw_smp_processor_id();
775 ucb = per_cpu_ptr(uprobe_cpu_buffer, cpu);
778 * Use per-cpu buffers for fastest access, but we might migrate
779 * so the mutex makes sure we have sole access to it.
781 mutex_lock(&ucb->mutex);
786 static void uprobe_buffer_put(struct uprobe_cpu_buffer *ucb)
788 mutex_unlock(&ucb->mutex);
791 static void __uprobe_trace_func(struct trace_uprobe *tu,
792 unsigned long func, struct pt_regs *regs,
793 struct uprobe_cpu_buffer *ucb, int dsize,
794 struct trace_event_file *trace_file)
796 struct uprobe_trace_entry_head *entry;
797 struct ring_buffer_event *event;
798 struct ring_buffer *buffer;
801 struct trace_event_call *call = &tu->tp.call;
803 WARN_ON(call != trace_file->event_call);
805 if (WARN_ON_ONCE(tu->tp.size + dsize > PAGE_SIZE))
808 if (trace_trigger_soft_disabled(trace_file))
811 esize = SIZEOF_TRACE_ENTRY(is_ret_probe(tu));
812 size = esize + tu->tp.size + dsize;
813 event = trace_event_buffer_lock_reserve(&buffer, trace_file,
814 call->event.type, size, 0, 0);
818 entry = ring_buffer_event_data(event);
819 if (is_ret_probe(tu)) {
820 entry->vaddr[0] = func;
821 entry->vaddr[1] = instruction_pointer(regs);
822 data = DATAOF_TRACE_ENTRY(entry, true);
824 entry->vaddr[0] = instruction_pointer(regs);
825 data = DATAOF_TRACE_ENTRY(entry, false);
828 memcpy(data, ucb->buf, tu->tp.size + dsize);
830 event_trigger_unlock_commit(trace_file, buffer, event, entry, 0, 0);
834 static int uprobe_trace_func(struct trace_uprobe *tu, struct pt_regs *regs,
835 struct uprobe_cpu_buffer *ucb, int dsize)
837 struct event_file_link *link;
839 if (is_ret_probe(tu))
843 list_for_each_entry_rcu(link, &tu->tp.files, list)
844 __uprobe_trace_func(tu, 0, regs, ucb, dsize, link->file);
850 static void uretprobe_trace_func(struct trace_uprobe *tu, unsigned long func,
851 struct pt_regs *regs,
852 struct uprobe_cpu_buffer *ucb, int dsize)
854 struct event_file_link *link;
857 list_for_each_entry_rcu(link, &tu->tp.files, list)
858 __uprobe_trace_func(tu, func, regs, ucb, dsize, link->file);
862 /* Event entry printers */
863 static enum print_line_t
864 print_uprobe_event(struct trace_iterator *iter, int flags, struct trace_event *event)
866 struct uprobe_trace_entry_head *entry;
867 struct trace_seq *s = &iter->seq;
868 struct trace_uprobe *tu;
872 entry = (struct uprobe_trace_entry_head *)iter->ent;
873 tu = container_of(event, struct trace_uprobe, tp.call.event);
875 if (is_ret_probe(tu)) {
876 trace_seq_printf(s, "%s: (0x%lx <- 0x%lx)",
877 trace_event_name(&tu->tp.call),
878 entry->vaddr[1], entry->vaddr[0]);
879 data = DATAOF_TRACE_ENTRY(entry, true);
881 trace_seq_printf(s, "%s: (0x%lx)",
882 trace_event_name(&tu->tp.call),
884 data = DATAOF_TRACE_ENTRY(entry, false);
887 for (i = 0; i < tu->tp.nr_args; i++) {
888 struct probe_arg *parg = &tu->tp.args[i];
890 if (!parg->type->print(s, parg->name, data + parg->offset, entry))
894 trace_seq_putc(s, '\n');
897 return trace_handle_return(s);
900 typedef bool (*filter_func_t)(struct uprobe_consumer *self,
901 enum uprobe_filter_ctx ctx,
902 struct mm_struct *mm);
905 probe_event_enable(struct trace_uprobe *tu, struct trace_event_file *file,
906 filter_func_t filter)
908 bool enabled = trace_probe_is_enabled(&tu->tp);
909 struct event_file_link *link = NULL;
913 if (tu->tp.flags & TP_FLAG_PROFILE)
916 link = kmalloc(sizeof(*link), GFP_KERNEL);
921 list_add_tail_rcu(&link->list, &tu->tp.files);
923 tu->tp.flags |= TP_FLAG_TRACE;
925 if (tu->tp.flags & TP_FLAG_TRACE)
928 tu->tp.flags |= TP_FLAG_PROFILE;
931 WARN_ON(!uprobe_filter_is_empty(&tu->filter));
936 ret = uprobe_buffer_enable();
940 tu->consumer.filter = filter;
941 tu->inode = d_real_inode(tu->path.dentry);
942 ret = uprobe_register(tu->inode, tu->offset, &tu->consumer);
949 uprobe_buffer_disable();
953 list_del(&link->list);
955 tu->tp.flags &= ~TP_FLAG_TRACE;
957 tu->tp.flags &= ~TP_FLAG_PROFILE;
963 probe_event_disable(struct trace_uprobe *tu, struct trace_event_file *file)
965 if (!trace_probe_is_enabled(&tu->tp))
969 struct event_file_link *link;
971 link = find_event_file_link(&tu->tp, file);
975 list_del_rcu(&link->list);
976 /* synchronize with u{,ret}probe_trace_func */
980 if (!list_empty(&tu->tp.files))
984 WARN_ON(!uprobe_filter_is_empty(&tu->filter));
986 uprobe_unregister(tu->inode, tu->offset, &tu->consumer);
988 tu->tp.flags &= file ? ~TP_FLAG_TRACE : ~TP_FLAG_PROFILE;
990 uprobe_buffer_disable();
993 static int uprobe_event_define_fields(struct trace_event_call *event_call)
996 struct uprobe_trace_entry_head field;
997 struct trace_uprobe *tu = event_call->data;
999 if (is_ret_probe(tu)) {
1000 DEFINE_FIELD(unsigned long, vaddr[0], FIELD_STRING_FUNC, 0);
1001 DEFINE_FIELD(unsigned long, vaddr[1], FIELD_STRING_RETIP, 0);
1002 size = SIZEOF_TRACE_ENTRY(true);
1004 DEFINE_FIELD(unsigned long, vaddr[0], FIELD_STRING_IP, 0);
1005 size = SIZEOF_TRACE_ENTRY(false);
1007 /* Set argument names as fields */
1008 for (i = 0; i < tu->tp.nr_args; i++) {
1009 struct probe_arg *parg = &tu->tp.args[i];
1011 ret = trace_define_field(event_call, parg->type->fmttype,
1012 parg->name, size + parg->offset,
1013 parg->type->size, parg->type->is_signed,
1022 #ifdef CONFIG_PERF_EVENTS
1024 __uprobe_perf_filter(struct trace_uprobe_filter *filter, struct mm_struct *mm)
1026 struct perf_event *event;
1028 if (filter->nr_systemwide)
1031 list_for_each_entry(event, &filter->perf_events, hw.tp_list) {
1032 if (event->hw.target->mm == mm)
1040 uprobe_filter_event(struct trace_uprobe *tu, struct perf_event *event)
1042 return __uprobe_perf_filter(&tu->filter, event->hw.target->mm);
1045 static int uprobe_perf_close(struct trace_uprobe *tu, struct perf_event *event)
1049 write_lock(&tu->filter.rwlock);
1050 if (event->hw.target) {
1051 list_del(&event->hw.tp_list);
1052 done = tu->filter.nr_systemwide ||
1053 (event->hw.target->flags & PF_EXITING) ||
1054 uprobe_filter_event(tu, event);
1056 tu->filter.nr_systemwide--;
1057 done = tu->filter.nr_systemwide;
1059 write_unlock(&tu->filter.rwlock);
1062 return uprobe_apply(tu->inode, tu->offset, &tu->consumer, false);
1067 static int uprobe_perf_open(struct trace_uprobe *tu, struct perf_event *event)
1072 write_lock(&tu->filter.rwlock);
1073 if (event->hw.target) {
1075 * event->parent != NULL means copy_process(), we can avoid
1076 * uprobe_apply(). current->mm must be probed and we can rely
1077 * on dup_mmap() which preserves the already installed bp's.
1079 * attr.enable_on_exec means that exec/mmap will install the
1080 * breakpoints we need.
1082 done = tu->filter.nr_systemwide ||
1083 event->parent || event->attr.enable_on_exec ||
1084 uprobe_filter_event(tu, event);
1085 list_add(&event->hw.tp_list, &tu->filter.perf_events);
1087 done = tu->filter.nr_systemwide;
1088 tu->filter.nr_systemwide++;
1090 write_unlock(&tu->filter.rwlock);
1094 err = uprobe_apply(tu->inode, tu->offset, &tu->consumer, true);
1096 uprobe_perf_close(tu, event);
1101 static bool uprobe_perf_filter(struct uprobe_consumer *uc,
1102 enum uprobe_filter_ctx ctx, struct mm_struct *mm)
1104 struct trace_uprobe *tu;
1107 tu = container_of(uc, struct trace_uprobe, consumer);
1108 read_lock(&tu->filter.rwlock);
1109 ret = __uprobe_perf_filter(&tu->filter, mm);
1110 read_unlock(&tu->filter.rwlock);
1115 static void __uprobe_perf_func(struct trace_uprobe *tu,
1116 unsigned long func, struct pt_regs *regs,
1117 struct uprobe_cpu_buffer *ucb, int dsize)
1119 struct trace_event_call *call = &tu->tp.call;
1120 struct uprobe_trace_entry_head *entry;
1121 struct bpf_prog *prog = call->prog;
1122 struct hlist_head *head;
1127 if (prog && !trace_call_bpf(prog, regs))
1130 esize = SIZEOF_TRACE_ENTRY(is_ret_probe(tu));
1132 size = esize + tu->tp.size + dsize;
1133 size = ALIGN(size + sizeof(u32), sizeof(u64)) - sizeof(u32);
1134 if (WARN_ONCE(size > PERF_MAX_TRACE_SIZE, "profile buffer not large enough"))
1138 head = this_cpu_ptr(call->perf_events);
1139 if (hlist_empty(head))
1142 entry = perf_trace_buf_alloc(size, NULL, &rctx);
1146 if (is_ret_probe(tu)) {
1147 entry->vaddr[0] = func;
1148 entry->vaddr[1] = instruction_pointer(regs);
1149 data = DATAOF_TRACE_ENTRY(entry, true);
1151 entry->vaddr[0] = instruction_pointer(regs);
1152 data = DATAOF_TRACE_ENTRY(entry, false);
1155 memcpy(data, ucb->buf, tu->tp.size + dsize);
1157 if (size - esize > tu->tp.size + dsize) {
1158 int len = tu->tp.size + dsize;
1160 memset(data + len, 0, size - esize - len);
1163 perf_trace_buf_submit(entry, size, rctx, call->event.type, 1, regs,
1169 /* uprobe profile handler */
1170 static int uprobe_perf_func(struct trace_uprobe *tu, struct pt_regs *regs,
1171 struct uprobe_cpu_buffer *ucb, int dsize)
1173 if (!uprobe_perf_filter(&tu->consumer, 0, current->mm))
1174 return UPROBE_HANDLER_REMOVE;
1176 if (!is_ret_probe(tu))
1177 __uprobe_perf_func(tu, 0, regs, ucb, dsize);
1181 static void uretprobe_perf_func(struct trace_uprobe *tu, unsigned long func,
1182 struct pt_regs *regs,
1183 struct uprobe_cpu_buffer *ucb, int dsize)
1185 __uprobe_perf_func(tu, func, regs, ucb, dsize);
1187 #endif /* CONFIG_PERF_EVENTS */
1190 trace_uprobe_register(struct trace_event_call *event, enum trace_reg type,
1193 struct trace_uprobe *tu = event->data;
1194 struct trace_event_file *file = data;
1197 case TRACE_REG_REGISTER:
1198 return probe_event_enable(tu, file, NULL);
1200 case TRACE_REG_UNREGISTER:
1201 probe_event_disable(tu, file);
1204 #ifdef CONFIG_PERF_EVENTS
1205 case TRACE_REG_PERF_REGISTER:
1206 return probe_event_enable(tu, NULL, uprobe_perf_filter);
1208 case TRACE_REG_PERF_UNREGISTER:
1209 probe_event_disable(tu, NULL);
1212 case TRACE_REG_PERF_OPEN:
1213 return uprobe_perf_open(tu, data);
1215 case TRACE_REG_PERF_CLOSE:
1216 return uprobe_perf_close(tu, data);
1225 static int uprobe_dispatcher(struct uprobe_consumer *con, struct pt_regs *regs)
1227 struct trace_uprobe *tu;
1228 struct uprobe_dispatch_data udd;
1229 struct uprobe_cpu_buffer *ucb;
1234 tu = container_of(con, struct trace_uprobe, consumer);
1238 udd.bp_addr = instruction_pointer(regs);
1240 current->utask->vaddr = (unsigned long) &udd;
1242 if (WARN_ON_ONCE(!uprobe_cpu_buffer))
1245 dsize = __get_data_size(&tu->tp, regs);
1246 esize = SIZEOF_TRACE_ENTRY(is_ret_probe(tu));
1248 ucb = uprobe_buffer_get();
1249 store_trace_args(esize, &tu->tp, regs, ucb->buf, dsize);
1251 if (tu->tp.flags & TP_FLAG_TRACE)
1252 ret |= uprobe_trace_func(tu, regs, ucb, dsize);
1254 #ifdef CONFIG_PERF_EVENTS
1255 if (tu->tp.flags & TP_FLAG_PROFILE)
1256 ret |= uprobe_perf_func(tu, regs, ucb, dsize);
1258 uprobe_buffer_put(ucb);
1262 static int uretprobe_dispatcher(struct uprobe_consumer *con,
1263 unsigned long func, struct pt_regs *regs)
1265 struct trace_uprobe *tu;
1266 struct uprobe_dispatch_data udd;
1267 struct uprobe_cpu_buffer *ucb;
1270 tu = container_of(con, struct trace_uprobe, consumer);
1275 current->utask->vaddr = (unsigned long) &udd;
1277 if (WARN_ON_ONCE(!uprobe_cpu_buffer))
1280 dsize = __get_data_size(&tu->tp, regs);
1281 esize = SIZEOF_TRACE_ENTRY(is_ret_probe(tu));
1283 ucb = uprobe_buffer_get();
1284 store_trace_args(esize, &tu->tp, regs, ucb->buf, dsize);
1286 if (tu->tp.flags & TP_FLAG_TRACE)
1287 uretprobe_trace_func(tu, func, regs, ucb, dsize);
1289 #ifdef CONFIG_PERF_EVENTS
1290 if (tu->tp.flags & TP_FLAG_PROFILE)
1291 uretprobe_perf_func(tu, func, regs, ucb, dsize);
1293 uprobe_buffer_put(ucb);
1297 static struct trace_event_functions uprobe_funcs = {
1298 .trace = print_uprobe_event
1301 static int register_uprobe_event(struct trace_uprobe *tu)
1303 struct trace_event_call *call = &tu->tp.call;
1306 /* Initialize trace_event_call */
1307 INIT_LIST_HEAD(&call->class->fields);
1308 call->event.funcs = &uprobe_funcs;
1309 call->class->define_fields = uprobe_event_define_fields;
1311 if (set_print_fmt(&tu->tp, is_ret_probe(tu)) < 0)
1314 ret = register_trace_event(&call->event);
1316 kfree(call->print_fmt);
1320 call->flags = TRACE_EVENT_FL_UPROBE;
1321 call->class->reg = trace_uprobe_register;
1323 ret = trace_add_event_call(call);
1326 pr_info("Failed to register uprobe event: %s\n",
1327 trace_event_name(call));
1328 kfree(call->print_fmt);
1329 unregister_trace_event(&call->event);
1335 static int unregister_uprobe_event(struct trace_uprobe *tu)
1339 /* tu->event is unregistered in trace_remove_event_call() */
1340 ret = trace_remove_event_call(&tu->tp.call);
1343 kfree(tu->tp.call.print_fmt);
1344 tu->tp.call.print_fmt = NULL;
1348 /* Make a trace interface for controling probe points */
1349 static __init int init_uprobe_trace(void)
1351 struct dentry *d_tracer;
1353 d_tracer = tracing_init_dentry();
1354 if (IS_ERR(d_tracer))
1357 trace_create_file("uprobe_events", 0644, d_tracer,
1358 NULL, &uprobe_events_ops);
1359 /* Profile interface */
1360 trace_create_file("uprobe_profile", 0444, d_tracer,
1361 NULL, &uprobe_profile_ops);
1365 fs_initcall(init_uprobe_trace);