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(int argc
, const char **argv
);
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 ret
= unregister_uprobe_event(tu
);
401 dyn_event_remove(&tu
->devent
);
402 trace_probe_unlink(&tu
->tp
);
403 free_trace_uprobe(tu
);
407 static bool trace_uprobe_has_same_uprobe(struct trace_uprobe
*orig
,
408 struct trace_uprobe
*comp
)
410 struct trace_probe_event
*tpe
= orig
->tp
.event
;
411 struct trace_probe
*pos
;
412 struct inode
*comp_inode
= d_real_inode(comp
->path
.dentry
);
415 list_for_each_entry(pos
, &tpe
->probes
, list
) {
416 orig
= container_of(pos
, struct trace_uprobe
, tp
);
417 if (comp_inode
!= d_real_inode(orig
->path
.dentry
) ||
418 comp
->offset
!= orig
->offset
)
422 * trace_probe_compare_arg_type() ensured that nr_args and
423 * each argument name and type are same. Let's compare comm.
425 for (i
= 0; i
< orig
->tp
.nr_args
; i
++) {
426 if (strcmp(orig
->tp
.args
[i
].comm
,
427 comp
->tp
.args
[i
].comm
))
431 if (i
== orig
->tp
.nr_args
)
438 static int append_trace_uprobe(struct trace_uprobe
*tu
, struct trace_uprobe
*to
)
442 ret
= trace_probe_compare_arg_type(&tu
->tp
, &to
->tp
);
444 /* Note that argument starts index = 2 */
445 trace_probe_log_set_index(ret
+ 1);
446 trace_probe_log_err(0, DIFF_ARG_TYPE
);
449 if (trace_uprobe_has_same_uprobe(to
, tu
)) {
450 trace_probe_log_set_index(0);
451 trace_probe_log_err(0, SAME_PROBE
);
455 /* Append to existing event */
456 ret
= trace_probe_append(&tu
->tp
, &to
->tp
);
458 dyn_event_add(&tu
->devent
);
464 * Uprobe with multiple reference counter is not allowed. i.e.
465 * If inode and offset matches, reference counter offset *must*
466 * match as well. Though, there is one exception: If user is
467 * replacing old trace_uprobe with new one(same group/event),
468 * then we allow same uprobe with new reference counter as far
469 * as the new one does not conflict with any other existing
472 static int validate_ref_ctr_offset(struct trace_uprobe
*new)
474 struct dyn_event
*pos
;
475 struct trace_uprobe
*tmp
;
476 struct inode
*new_inode
= d_real_inode(new->path
.dentry
);
478 for_each_trace_uprobe(tmp
, pos
) {
479 if (new_inode
== d_real_inode(tmp
->path
.dentry
) &&
480 new->offset
== tmp
->offset
&&
481 new->ref_ctr_offset
!= tmp
->ref_ctr_offset
) {
482 pr_warn("Reference counter offset mismatch.");
489 /* Register a trace_uprobe and probe_event */
490 static int register_trace_uprobe(struct trace_uprobe
*tu
)
492 struct trace_uprobe
*old_tu
;
495 mutex_lock(&event_mutex
);
497 ret
= validate_ref_ctr_offset(tu
);
501 /* register as an event */
502 old_tu
= find_probe_event(trace_probe_name(&tu
->tp
),
503 trace_probe_group_name(&tu
->tp
));
505 if (is_ret_probe(tu
) != is_ret_probe(old_tu
)) {
506 trace_probe_log_set_index(0);
507 trace_probe_log_err(0, DIFF_PROBE_TYPE
);
510 ret
= append_trace_uprobe(tu
, old_tu
);
515 ret
= register_uprobe_event(tu
);
517 pr_warn("Failed to register probe event(%d)\n", ret
);
521 dyn_event_add(&tu
->devent
);
524 mutex_unlock(&event_mutex
);
531 * - Add uprobe: p|r[:[GRP/]EVENT] PATH:OFFSET [FETCHARGS]
533 static int trace_uprobe_create(int argc
, const char **argv
)
535 struct trace_uprobe
*tu
;
536 const char *event
= NULL
, *group
= UPROBE_EVENT_SYSTEM
;
537 char *arg
, *filename
, *rctr
, *rctr_end
, *tmp
;
538 char buf
[MAX_EVENT_NAME_LEN
];
540 unsigned long offset
, ref_ctr_offset
;
541 bool is_return
= false;
547 switch (argv
[0][0]) {
560 if (argv
[0][1] == ':')
563 if (!strchr(argv
[1], '/'))
566 filename
= kstrdup(argv
[1], GFP_KERNEL
);
570 /* Find the last occurrence, in case the path contains ':' too. */
571 arg
= strrchr(filename
, ':');
572 if (!arg
|| !isdigit(arg
[1])) {
577 trace_probe_log_init("trace_uprobe", argc
, argv
);
578 trace_probe_log_set_index(1); /* filename is the 2nd argument */
581 ret
= kern_path(filename
, LOOKUP_FOLLOW
, &path
);
583 trace_probe_log_err(0, FILE_NOT_FOUND
);
585 trace_probe_log_clear();
588 if (!d_is_reg(path
.dentry
)) {
589 trace_probe_log_err(0, NO_REGULAR_FILE
);
591 goto fail_address_parse
;
594 /* Parse reference counter offset if specified. */
595 rctr
= strchr(arg
, '(');
597 rctr_end
= strchr(rctr
, ')');
600 rctr_end
= rctr
+ strlen(rctr
);
601 trace_probe_log_err(rctr_end
- filename
,
603 goto fail_address_parse
;
604 } else if (rctr_end
[1] != '\0') {
606 trace_probe_log_err(rctr_end
+ 1 - filename
,
608 goto fail_address_parse
;
613 ret
= kstrtoul(rctr
, 0, &ref_ctr_offset
);
615 trace_probe_log_err(rctr
- filename
, BAD_REFCNT
);
616 goto fail_address_parse
;
620 /* Parse uprobe offset. */
621 ret
= kstrtoul(arg
, 0, &offset
);
623 trace_probe_log_err(arg
- filename
, BAD_UPROBE_OFFS
);
624 goto fail_address_parse
;
628 trace_probe_log_set_index(0);
630 ret
= traceprobe_parse_event_name(&event
, &group
, buf
,
633 goto fail_address_parse
;
638 tail
= kstrdup(kbasename(filename
), GFP_KERNEL
);
641 goto fail_address_parse
;
644 ptr
= strpbrk(tail
, ".-_");
648 snprintf(buf
, MAX_EVENT_NAME_LEN
, "%c_%s_0x%lx", 'p', tail
, offset
);
656 tu
= alloc_trace_uprobe(group
, event
, argc
, is_return
);
659 /* This must return -ENOMEM otherwise there is a bug */
660 WARN_ON_ONCE(ret
!= -ENOMEM
);
661 goto fail_address_parse
;
664 tu
->ref_ctr_offset
= ref_ctr_offset
;
666 tu
->filename
= filename
;
668 /* parse arguments */
669 for (i
= 0; i
< argc
&& i
< MAX_TRACE_ARGS
; i
++) {
670 tmp
= kstrdup(argv
[i
], GFP_KERNEL
);
676 trace_probe_log_set_index(i
+ 2);
677 ret
= traceprobe_parse_probe_arg(&tu
->tp
, i
, tmp
,
678 is_return
? TPARG_FL_RETURN
: 0);
684 ret
= traceprobe_set_print_fmt(&tu
->tp
, is_ret_probe(tu
));
688 ret
= register_trace_uprobe(tu
);
693 free_trace_uprobe(tu
);
695 trace_probe_log_clear();
699 trace_probe_log_clear();
706 static int create_or_delete_trace_uprobe(int argc
, char **argv
)
710 if (argv
[0][0] == '-')
711 return dyn_event_release(argc
, argv
, &trace_uprobe_ops
);
713 ret
= trace_uprobe_create(argc
, (const char **)argv
);
714 return ret
== -ECANCELED
? -EINVAL
: ret
;
717 static int trace_uprobe_release(struct dyn_event
*ev
)
719 struct trace_uprobe
*tu
= to_trace_uprobe(ev
);
721 return unregister_trace_uprobe(tu
);
724 /* Probes listing interfaces */
725 static int trace_uprobe_show(struct seq_file
*m
, struct dyn_event
*ev
)
727 struct trace_uprobe
*tu
= to_trace_uprobe(ev
);
728 char c
= is_ret_probe(tu
) ? 'r' : 'p';
731 seq_printf(m
, "%c:%s/%s %s:0x%0*lx", c
, trace_probe_group_name(&tu
->tp
),
732 trace_probe_name(&tu
->tp
), tu
->filename
,
733 (int)(sizeof(void *) * 2), tu
->offset
);
735 if (tu
->ref_ctr_offset
)
736 seq_printf(m
, "(0x%lx)", tu
->ref_ctr_offset
);
738 for (i
= 0; i
< tu
->tp
.nr_args
; i
++)
739 seq_printf(m
, " %s=%s", tu
->tp
.args
[i
].name
, tu
->tp
.args
[i
].comm
);
745 static int probes_seq_show(struct seq_file
*m
, void *v
)
747 struct dyn_event
*ev
= v
;
749 if (!is_trace_uprobe(ev
))
752 return trace_uprobe_show(m
, ev
);
755 static const struct seq_operations probes_seq_op
= {
756 .start
= dyn_event_seq_start
,
757 .next
= dyn_event_seq_next
,
758 .stop
= dyn_event_seq_stop
,
759 .show
= probes_seq_show
762 static int probes_open(struct inode
*inode
, struct file
*file
)
766 ret
= security_locked_down(LOCKDOWN_TRACEFS
);
770 if ((file
->f_mode
& FMODE_WRITE
) && (file
->f_flags
& O_TRUNC
)) {
771 ret
= dyn_events_release_all(&trace_uprobe_ops
);
776 return seq_open(file
, &probes_seq_op
);
779 static ssize_t
probes_write(struct file
*file
, const char __user
*buffer
,
780 size_t count
, loff_t
*ppos
)
782 return trace_parse_run_command(file
, buffer
, count
, ppos
,
783 create_or_delete_trace_uprobe
);
786 static const struct file_operations uprobe_events_ops
= {
787 .owner
= THIS_MODULE
,
791 .release
= seq_release
,
792 .write
= probes_write
,
795 /* Probes profiling interfaces */
796 static int probes_profile_seq_show(struct seq_file
*m
, void *v
)
798 struct dyn_event
*ev
= v
;
799 struct trace_uprobe
*tu
;
801 if (!is_trace_uprobe(ev
))
804 tu
= to_trace_uprobe(ev
);
805 seq_printf(m
, " %s %-44s %15lu\n", tu
->filename
,
806 trace_probe_name(&tu
->tp
), tu
->nhit
);
810 static const struct seq_operations profile_seq_op
= {
811 .start
= dyn_event_seq_start
,
812 .next
= dyn_event_seq_next
,
813 .stop
= dyn_event_seq_stop
,
814 .show
= probes_profile_seq_show
817 static int profile_open(struct inode
*inode
, struct file
*file
)
821 ret
= security_locked_down(LOCKDOWN_TRACEFS
);
825 return seq_open(file
, &profile_seq_op
);
828 static const struct file_operations uprobe_profile_ops
= {
829 .owner
= THIS_MODULE
,
830 .open
= profile_open
,
833 .release
= seq_release
,
836 struct uprobe_cpu_buffer
{
840 static struct uprobe_cpu_buffer __percpu
*uprobe_cpu_buffer
;
841 static int uprobe_buffer_refcnt
;
843 static int uprobe_buffer_init(void)
847 uprobe_cpu_buffer
= alloc_percpu(struct uprobe_cpu_buffer
);
848 if (uprobe_cpu_buffer
== NULL
)
851 for_each_possible_cpu(cpu
) {
852 struct page
*p
= alloc_pages_node(cpu_to_node(cpu
),
858 per_cpu_ptr(uprobe_cpu_buffer
, cpu
)->buf
= page_address(p
);
859 mutex_init(&per_cpu_ptr(uprobe_cpu_buffer
, cpu
)->mutex
);
865 for_each_possible_cpu(cpu
) {
868 free_page((unsigned long)per_cpu_ptr(uprobe_cpu_buffer
, cpu
)->buf
);
871 free_percpu(uprobe_cpu_buffer
);
875 static int uprobe_buffer_enable(void)
879 BUG_ON(!mutex_is_locked(&event_mutex
));
881 if (uprobe_buffer_refcnt
++ == 0) {
882 ret
= uprobe_buffer_init();
884 uprobe_buffer_refcnt
--;
890 static void uprobe_buffer_disable(void)
894 BUG_ON(!mutex_is_locked(&event_mutex
));
896 if (--uprobe_buffer_refcnt
== 0) {
897 for_each_possible_cpu(cpu
)
898 free_page((unsigned long)per_cpu_ptr(uprobe_cpu_buffer
,
901 free_percpu(uprobe_cpu_buffer
);
902 uprobe_cpu_buffer
= NULL
;
906 static struct uprobe_cpu_buffer
*uprobe_buffer_get(void)
908 struct uprobe_cpu_buffer
*ucb
;
911 cpu
= raw_smp_processor_id();
912 ucb
= per_cpu_ptr(uprobe_cpu_buffer
, cpu
);
915 * Use per-cpu buffers for fastest access, but we might migrate
916 * so the mutex makes sure we have sole access to it.
918 mutex_lock(&ucb
->mutex
);
923 static void uprobe_buffer_put(struct uprobe_cpu_buffer
*ucb
)
925 mutex_unlock(&ucb
->mutex
);
928 static void __uprobe_trace_func(struct trace_uprobe
*tu
,
929 unsigned long func
, struct pt_regs
*regs
,
930 struct uprobe_cpu_buffer
*ucb
, int dsize
,
931 struct trace_event_file
*trace_file
)
933 struct uprobe_trace_entry_head
*entry
;
934 struct trace_buffer
*buffer
;
935 struct ring_buffer_event
*event
;
938 struct trace_event_call
*call
= trace_probe_event_call(&tu
->tp
);
940 WARN_ON(call
!= trace_file
->event_call
);
942 if (WARN_ON_ONCE(tu
->tp
.size
+ dsize
> PAGE_SIZE
))
945 if (trace_trigger_soft_disabled(trace_file
))
948 esize
= SIZEOF_TRACE_ENTRY(is_ret_probe(tu
));
949 size
= esize
+ tu
->tp
.size
+ dsize
;
950 event
= trace_event_buffer_lock_reserve(&buffer
, trace_file
,
951 call
->event
.type
, size
, 0, 0);
955 entry
= ring_buffer_event_data(event
);
956 if (is_ret_probe(tu
)) {
957 entry
->vaddr
[0] = func
;
958 entry
->vaddr
[1] = instruction_pointer(regs
);
959 data
= DATAOF_TRACE_ENTRY(entry
, true);
961 entry
->vaddr
[0] = instruction_pointer(regs
);
962 data
= DATAOF_TRACE_ENTRY(entry
, false);
965 memcpy(data
, ucb
->buf
, tu
->tp
.size
+ dsize
);
967 event_trigger_unlock_commit(trace_file
, buffer
, event
, entry
, 0, 0);
971 static int uprobe_trace_func(struct trace_uprobe
*tu
, struct pt_regs
*regs
,
972 struct uprobe_cpu_buffer
*ucb
, int dsize
)
974 struct event_file_link
*link
;
976 if (is_ret_probe(tu
))
980 trace_probe_for_each_link_rcu(link
, &tu
->tp
)
981 __uprobe_trace_func(tu
, 0, regs
, ucb
, dsize
, link
->file
);
987 static void uretprobe_trace_func(struct trace_uprobe
*tu
, unsigned long func
,
988 struct pt_regs
*regs
,
989 struct uprobe_cpu_buffer
*ucb
, int dsize
)
991 struct event_file_link
*link
;
994 trace_probe_for_each_link_rcu(link
, &tu
->tp
)
995 __uprobe_trace_func(tu
, func
, regs
, ucb
, dsize
, link
->file
);
999 /* Event entry printers */
1000 static enum print_line_t
1001 print_uprobe_event(struct trace_iterator
*iter
, int flags
, struct trace_event
*event
)
1003 struct uprobe_trace_entry_head
*entry
;
1004 struct trace_seq
*s
= &iter
->seq
;
1005 struct trace_uprobe
*tu
;
1008 entry
= (struct uprobe_trace_entry_head
*)iter
->ent
;
1009 tu
= trace_uprobe_primary_from_call(
1010 container_of(event
, struct trace_event_call
, event
));
1014 if (is_ret_probe(tu
)) {
1015 trace_seq_printf(s
, "%s: (0x%lx <- 0x%lx)",
1016 trace_probe_name(&tu
->tp
),
1017 entry
->vaddr
[1], entry
->vaddr
[0]);
1018 data
= DATAOF_TRACE_ENTRY(entry
, true);
1020 trace_seq_printf(s
, "%s: (0x%lx)",
1021 trace_probe_name(&tu
->tp
),
1023 data
= DATAOF_TRACE_ENTRY(entry
, false);
1026 if (print_probe_args(s
, tu
->tp
.args
, tu
->tp
.nr_args
, data
, entry
) < 0)
1029 trace_seq_putc(s
, '\n');
1032 return trace_handle_return(s
);
1035 typedef bool (*filter_func_t
)(struct uprobe_consumer
*self
,
1036 enum uprobe_filter_ctx ctx
,
1037 struct mm_struct
*mm
);
1039 static int trace_uprobe_enable(struct trace_uprobe
*tu
, filter_func_t filter
)
1043 tu
->consumer
.filter
= filter
;
1044 tu
->inode
= d_real_inode(tu
->path
.dentry
);
1046 if (tu
->ref_ctr_offset
)
1047 ret
= uprobe_register_refctr(tu
->inode
, tu
->offset
,
1048 tu
->ref_ctr_offset
, &tu
->consumer
);
1050 ret
= uprobe_register(tu
->inode
, tu
->offset
, &tu
->consumer
);
1058 static void __probe_event_disable(struct trace_probe
*tp
)
1060 struct trace_probe
*pos
;
1061 struct trace_uprobe
*tu
;
1063 tu
= container_of(tp
, struct trace_uprobe
, tp
);
1064 WARN_ON(!uprobe_filter_is_empty(tu
->tp
.event
->filter
));
1066 list_for_each_entry(pos
, trace_probe_probe_list(tp
), list
) {
1067 tu
= container_of(pos
, struct trace_uprobe
, tp
);
1071 uprobe_unregister(tu
->inode
, tu
->offset
, &tu
->consumer
);
1076 static int probe_event_enable(struct trace_event_call
*call
,
1077 struct trace_event_file
*file
, filter_func_t filter
)
1079 struct trace_probe
*pos
, *tp
;
1080 struct trace_uprobe
*tu
;
1084 tp
= trace_probe_primary_from_call(call
);
1085 if (WARN_ON_ONCE(!tp
))
1087 enabled
= trace_probe_is_enabled(tp
);
1089 /* This may also change "enabled" state */
1091 if (trace_probe_test_flag(tp
, TP_FLAG_PROFILE
))
1094 ret
= trace_probe_add_file(tp
, file
);
1098 if (trace_probe_test_flag(tp
, TP_FLAG_TRACE
))
1101 trace_probe_set_flag(tp
, TP_FLAG_PROFILE
);
1104 tu
= container_of(tp
, struct trace_uprobe
, tp
);
1105 WARN_ON(!uprobe_filter_is_empty(tu
->tp
.event
->filter
));
1110 ret
= uprobe_buffer_enable();
1114 list_for_each_entry(pos
, trace_probe_probe_list(tp
), list
) {
1115 tu
= container_of(pos
, struct trace_uprobe
, tp
);
1116 ret
= trace_uprobe_enable(tu
, filter
);
1118 __probe_event_disable(tp
);
1126 uprobe_buffer_disable();
1130 trace_probe_remove_file(tp
, file
);
1132 trace_probe_clear_flag(tp
, TP_FLAG_PROFILE
);
1137 static void probe_event_disable(struct trace_event_call
*call
,
1138 struct trace_event_file
*file
)
1140 struct trace_probe
*tp
;
1142 tp
= trace_probe_primary_from_call(call
);
1143 if (WARN_ON_ONCE(!tp
))
1146 if (!trace_probe_is_enabled(tp
))
1150 if (trace_probe_remove_file(tp
, file
) < 0)
1153 if (trace_probe_is_enabled(tp
))
1156 trace_probe_clear_flag(tp
, TP_FLAG_PROFILE
);
1158 __probe_event_disable(tp
);
1159 uprobe_buffer_disable();
1162 static int uprobe_event_define_fields(struct trace_event_call
*event_call
)
1165 struct uprobe_trace_entry_head field
;
1166 struct trace_uprobe
*tu
;
1168 tu
= trace_uprobe_primary_from_call(event_call
);
1172 if (is_ret_probe(tu
)) {
1173 DEFINE_FIELD(unsigned long, vaddr
[0], FIELD_STRING_FUNC
, 0);
1174 DEFINE_FIELD(unsigned long, vaddr
[1], FIELD_STRING_RETIP
, 0);
1175 size
= SIZEOF_TRACE_ENTRY(true);
1177 DEFINE_FIELD(unsigned long, vaddr
[0], FIELD_STRING_IP
, 0);
1178 size
= SIZEOF_TRACE_ENTRY(false);
1181 return traceprobe_define_arg_fields(event_call
, size
, &tu
->tp
);
1184 #ifdef CONFIG_PERF_EVENTS
1186 __uprobe_perf_filter(struct trace_uprobe_filter
*filter
, struct mm_struct
*mm
)
1188 struct perf_event
*event
;
1190 if (filter
->nr_systemwide
)
1193 list_for_each_entry(event
, &filter
->perf_events
, hw
.tp_list
) {
1194 if (event
->hw
.target
->mm
== mm
)
1202 trace_uprobe_filter_event(struct trace_uprobe_filter
*filter
,
1203 struct perf_event
*event
)
1205 return __uprobe_perf_filter(filter
, event
->hw
.target
->mm
);
1208 static bool trace_uprobe_filter_remove(struct trace_uprobe_filter
*filter
,
1209 struct perf_event
*event
)
1213 write_lock(&filter
->rwlock
);
1214 if (event
->hw
.target
) {
1215 list_del(&event
->hw
.tp_list
);
1216 done
= filter
->nr_systemwide
||
1217 (event
->hw
.target
->flags
& PF_EXITING
) ||
1218 trace_uprobe_filter_event(filter
, event
);
1220 filter
->nr_systemwide
--;
1221 done
= filter
->nr_systemwide
;
1223 write_unlock(&filter
->rwlock
);
1228 /* This returns true if the filter always covers target mm */
1229 static bool trace_uprobe_filter_add(struct trace_uprobe_filter
*filter
,
1230 struct perf_event
*event
)
1234 write_lock(&filter
->rwlock
);
1235 if (event
->hw
.target
) {
1237 * event->parent != NULL means copy_process(), we can avoid
1238 * uprobe_apply(). current->mm must be probed and we can rely
1239 * on dup_mmap() which preserves the already installed bp's.
1241 * attr.enable_on_exec means that exec/mmap will install the
1242 * breakpoints we need.
1244 done
= filter
->nr_systemwide
||
1245 event
->parent
|| event
->attr
.enable_on_exec
||
1246 trace_uprobe_filter_event(filter
, event
);
1247 list_add(&event
->hw
.tp_list
, &filter
->perf_events
);
1249 done
= filter
->nr_systemwide
;
1250 filter
->nr_systemwide
++;
1252 write_unlock(&filter
->rwlock
);
1257 static int uprobe_perf_close(struct trace_event_call
*call
,
1258 struct perf_event
*event
)
1260 struct trace_probe
*pos
, *tp
;
1261 struct trace_uprobe
*tu
;
1264 tp
= trace_probe_primary_from_call(call
);
1265 if (WARN_ON_ONCE(!tp
))
1268 tu
= container_of(tp
, struct trace_uprobe
, tp
);
1269 if (trace_uprobe_filter_remove(tu
->tp
.event
->filter
, event
))
1272 list_for_each_entry(pos
, trace_probe_probe_list(tp
), list
) {
1273 tu
= container_of(pos
, struct trace_uprobe
, tp
);
1274 ret
= uprobe_apply(tu
->inode
, tu
->offset
, &tu
->consumer
, false);
1282 static int uprobe_perf_open(struct trace_event_call
*call
,
1283 struct perf_event
*event
)
1285 struct trace_probe
*pos
, *tp
;
1286 struct trace_uprobe
*tu
;
1289 tp
= trace_probe_primary_from_call(call
);
1290 if (WARN_ON_ONCE(!tp
))
1293 tu
= container_of(tp
, struct trace_uprobe
, tp
);
1294 if (trace_uprobe_filter_add(tu
->tp
.event
->filter
, event
))
1297 list_for_each_entry(pos
, trace_probe_probe_list(tp
), list
) {
1298 err
= uprobe_apply(tu
->inode
, tu
->offset
, &tu
->consumer
, true);
1300 uprobe_perf_close(call
, event
);
1308 static bool uprobe_perf_filter(struct uprobe_consumer
*uc
,
1309 enum uprobe_filter_ctx ctx
, struct mm_struct
*mm
)
1311 struct trace_uprobe_filter
*filter
;
1312 struct trace_uprobe
*tu
;
1315 tu
= container_of(uc
, struct trace_uprobe
, consumer
);
1316 filter
= tu
->tp
.event
->filter
;
1318 read_lock(&filter
->rwlock
);
1319 ret
= __uprobe_perf_filter(filter
, mm
);
1320 read_unlock(&filter
->rwlock
);
1325 static void __uprobe_perf_func(struct trace_uprobe
*tu
,
1326 unsigned long func
, struct pt_regs
*regs
,
1327 struct uprobe_cpu_buffer
*ucb
, int dsize
)
1329 struct trace_event_call
*call
= trace_probe_event_call(&tu
->tp
);
1330 struct uprobe_trace_entry_head
*entry
;
1331 struct hlist_head
*head
;
1336 if (bpf_prog_array_valid(call
) && !trace_call_bpf(call
, regs
))
1339 esize
= SIZEOF_TRACE_ENTRY(is_ret_probe(tu
));
1341 size
= esize
+ tu
->tp
.size
+ dsize
;
1342 size
= ALIGN(size
+ sizeof(u32
), sizeof(u64
)) - sizeof(u32
);
1343 if (WARN_ONCE(size
> PERF_MAX_TRACE_SIZE
, "profile buffer not large enough"))
1347 head
= this_cpu_ptr(call
->perf_events
);
1348 if (hlist_empty(head
))
1351 entry
= perf_trace_buf_alloc(size
, NULL
, &rctx
);
1355 if (is_ret_probe(tu
)) {
1356 entry
->vaddr
[0] = func
;
1357 entry
->vaddr
[1] = instruction_pointer(regs
);
1358 data
= DATAOF_TRACE_ENTRY(entry
, true);
1360 entry
->vaddr
[0] = instruction_pointer(regs
);
1361 data
= DATAOF_TRACE_ENTRY(entry
, false);
1364 memcpy(data
, ucb
->buf
, tu
->tp
.size
+ dsize
);
1366 if (size
- esize
> tu
->tp
.size
+ dsize
) {
1367 int len
= tu
->tp
.size
+ dsize
;
1369 memset(data
+ len
, 0, size
- esize
- len
);
1372 perf_trace_buf_submit(entry
, size
, rctx
, call
->event
.type
, 1, regs
,
1378 /* uprobe profile handler */
1379 static int uprobe_perf_func(struct trace_uprobe
*tu
, struct pt_regs
*regs
,
1380 struct uprobe_cpu_buffer
*ucb
, int dsize
)
1382 if (!uprobe_perf_filter(&tu
->consumer
, 0, current
->mm
))
1383 return UPROBE_HANDLER_REMOVE
;
1385 if (!is_ret_probe(tu
))
1386 __uprobe_perf_func(tu
, 0, regs
, ucb
, dsize
);
1390 static void uretprobe_perf_func(struct trace_uprobe
*tu
, unsigned long func
,
1391 struct pt_regs
*regs
,
1392 struct uprobe_cpu_buffer
*ucb
, int dsize
)
1394 __uprobe_perf_func(tu
, func
, regs
, ucb
, dsize
);
1397 int bpf_get_uprobe_info(const struct perf_event
*event
, u32
*fd_type
,
1398 const char **filename
, u64
*probe_offset
,
1399 bool perf_type_tracepoint
)
1401 const char *pevent
= trace_event_name(event
->tp_event
);
1402 const char *group
= event
->tp_event
->class->system
;
1403 struct trace_uprobe
*tu
;
1405 if (perf_type_tracepoint
)
1406 tu
= find_probe_event(pevent
, group
);
1408 tu
= event
->tp_event
->data
;
1412 *fd_type
= is_ret_probe(tu
) ? BPF_FD_TYPE_URETPROBE
1413 : BPF_FD_TYPE_UPROBE
;
1414 *filename
= tu
->filename
;
1415 *probe_offset
= tu
->offset
;
1418 #endif /* CONFIG_PERF_EVENTS */
1421 trace_uprobe_register(struct trace_event_call
*event
, enum trace_reg type
,
1424 struct trace_event_file
*file
= data
;
1427 case TRACE_REG_REGISTER
:
1428 return probe_event_enable(event
, file
, NULL
);
1430 case TRACE_REG_UNREGISTER
:
1431 probe_event_disable(event
, file
);
1434 #ifdef CONFIG_PERF_EVENTS
1435 case TRACE_REG_PERF_REGISTER
:
1436 return probe_event_enable(event
, NULL
, uprobe_perf_filter
);
1438 case TRACE_REG_PERF_UNREGISTER
:
1439 probe_event_disable(event
, NULL
);
1442 case TRACE_REG_PERF_OPEN
:
1443 return uprobe_perf_open(event
, data
);
1445 case TRACE_REG_PERF_CLOSE
:
1446 return uprobe_perf_close(event
, data
);
1455 static int uprobe_dispatcher(struct uprobe_consumer
*con
, struct pt_regs
*regs
)
1457 struct trace_uprobe
*tu
;
1458 struct uprobe_dispatch_data udd
;
1459 struct uprobe_cpu_buffer
*ucb
;
1464 tu
= container_of(con
, struct trace_uprobe
, consumer
);
1468 udd
.bp_addr
= instruction_pointer(regs
);
1470 current
->utask
->vaddr
= (unsigned long) &udd
;
1472 if (WARN_ON_ONCE(!uprobe_cpu_buffer
))
1475 dsize
= __get_data_size(&tu
->tp
, regs
);
1476 esize
= SIZEOF_TRACE_ENTRY(is_ret_probe(tu
));
1478 ucb
= uprobe_buffer_get();
1479 store_trace_args(ucb
->buf
, &tu
->tp
, regs
, esize
, dsize
);
1481 if (trace_probe_test_flag(&tu
->tp
, TP_FLAG_TRACE
))
1482 ret
|= uprobe_trace_func(tu
, regs
, ucb
, dsize
);
1484 #ifdef CONFIG_PERF_EVENTS
1485 if (trace_probe_test_flag(&tu
->tp
, TP_FLAG_PROFILE
))
1486 ret
|= uprobe_perf_func(tu
, regs
, ucb
, dsize
);
1488 uprobe_buffer_put(ucb
);
1492 static int uretprobe_dispatcher(struct uprobe_consumer
*con
,
1493 unsigned long func
, struct pt_regs
*regs
)
1495 struct trace_uprobe
*tu
;
1496 struct uprobe_dispatch_data udd
;
1497 struct uprobe_cpu_buffer
*ucb
;
1500 tu
= container_of(con
, struct trace_uprobe
, consumer
);
1505 current
->utask
->vaddr
= (unsigned long) &udd
;
1507 if (WARN_ON_ONCE(!uprobe_cpu_buffer
))
1510 dsize
= __get_data_size(&tu
->tp
, regs
);
1511 esize
= SIZEOF_TRACE_ENTRY(is_ret_probe(tu
));
1513 ucb
= uprobe_buffer_get();
1514 store_trace_args(ucb
->buf
, &tu
->tp
, regs
, esize
, dsize
);
1516 if (trace_probe_test_flag(&tu
->tp
, TP_FLAG_TRACE
))
1517 uretprobe_trace_func(tu
, func
, regs
, ucb
, dsize
);
1519 #ifdef CONFIG_PERF_EVENTS
1520 if (trace_probe_test_flag(&tu
->tp
, TP_FLAG_PROFILE
))
1521 uretprobe_perf_func(tu
, func
, regs
, ucb
, dsize
);
1523 uprobe_buffer_put(ucb
);
1527 static struct trace_event_functions uprobe_funcs
= {
1528 .trace
= print_uprobe_event
1531 static struct trace_event_fields uprobe_fields_array
[] = {
1532 { .type
= TRACE_FUNCTION_TYPE
,
1533 .define_fields
= uprobe_event_define_fields
},
1537 static inline void init_trace_event_call(struct trace_uprobe
*tu
)
1539 struct trace_event_call
*call
= trace_probe_event_call(&tu
->tp
);
1540 call
->event
.funcs
= &uprobe_funcs
;
1541 call
->class->fields_array
= uprobe_fields_array
;
1543 call
->flags
= TRACE_EVENT_FL_UPROBE
| TRACE_EVENT_FL_CAP_ANY
;
1544 call
->class->reg
= trace_uprobe_register
;
1547 static int register_uprobe_event(struct trace_uprobe
*tu
)
1549 init_trace_event_call(tu
);
1551 return trace_probe_register_event_call(&tu
->tp
);
1554 static int unregister_uprobe_event(struct trace_uprobe
*tu
)
1556 return trace_probe_unregister_event_call(&tu
->tp
);
1559 #ifdef CONFIG_PERF_EVENTS
1560 struct trace_event_call
*
1561 create_local_trace_uprobe(char *name
, unsigned long offs
,
1562 unsigned long ref_ctr_offset
, bool is_return
)
1564 struct trace_uprobe
*tu
;
1568 ret
= kern_path(name
, LOOKUP_FOLLOW
, &path
);
1570 return ERR_PTR(ret
);
1572 if (!d_is_reg(path
.dentry
)) {
1574 return ERR_PTR(-EINVAL
);
1578 * local trace_kprobes are not added to dyn_event, so they are never
1579 * searched in find_trace_kprobe(). Therefore, there is no concern of
1580 * duplicated name "DUMMY_EVENT" here.
1582 tu
= alloc_trace_uprobe(UPROBE_EVENT_SYSTEM
, "DUMMY_EVENT", 0,
1586 pr_info("Failed to allocate trace_uprobe.(%d)\n",
1589 return ERR_CAST(tu
);
1594 tu
->ref_ctr_offset
= ref_ctr_offset
;
1595 tu
->filename
= kstrdup(name
, GFP_KERNEL
);
1596 init_trace_event_call(tu
);
1598 if (traceprobe_set_print_fmt(&tu
->tp
, is_ret_probe(tu
)) < 0) {
1603 return trace_probe_event_call(&tu
->tp
);
1605 free_trace_uprobe(tu
);
1606 return ERR_PTR(ret
);
1609 void destroy_local_trace_uprobe(struct trace_event_call
*event_call
)
1611 struct trace_uprobe
*tu
;
1613 tu
= trace_uprobe_primary_from_call(event_call
);
1615 free_trace_uprobe(tu
);
1617 #endif /* CONFIG_PERF_EVENTS */
1619 /* Make a trace interface for controling probe points */
1620 static __init
int init_uprobe_trace(void)
1622 struct dentry
*d_tracer
;
1625 ret
= dyn_event_register(&trace_uprobe_ops
);
1629 d_tracer
= tracing_init_dentry();
1630 if (IS_ERR(d_tracer
))
1633 trace_create_file("uprobe_events", 0644, d_tracer
,
1634 NULL
, &uprobe_events_ops
);
1635 /* Profile interface */
1636 trace_create_file("uprobe_profile", 0444, d_tracer
,
1637 NULL
, &uprobe_profile_ops
);
1641 fs_initcall(init_uprobe_trace
);