2 * Copyright (C) 2011 David Goulet <david.goulet@polymtl.ca>
4 * SPDX-License-Identifier: GPL-2.0-only
15 #include <sys/types.h>
17 #include <common/common.h>
18 #include <common/trace-chunk.h>
19 #include <common/kernel-ctl/kernel-ctl.h>
20 #include <common/kernel-ctl/kernel-ioctl.h>
21 #include <common/sessiond-comm/sessiond-comm.h>
23 #include "lttng-sessiond.h"
24 #include "lttng-syscall.h"
27 #include "kernel-consumer.h"
28 #include "kern-modules.h"
34 * Key used to reference a channel between the sessiond and the consumer. This
35 * is only read and updated with the session_list lock held.
37 static uint64_t next_kernel_channel_key
;
39 static const char *module_proc_lttng
= "/proc/lttng";
41 static int kernel_tracer_fd
= -1;
43 #include <lttng/userspace-probe.h>
44 #include <lttng/userspace-probe-internal.h>
46 * Add context on a kernel channel.
48 * Assumes the ownership of ctx.
50 int kernel_add_channel_context(struct ltt_kernel_channel
*chan
,
51 struct ltt_kernel_context
*ctx
)
58 DBG("Adding context to channel %s", chan
->channel
->name
);
59 ret
= kernctl_add_context(chan
->fd
, &ctx
->ctx
);
63 /* Exists but not available for this kernel */
64 ret
= LTTNG_ERR_KERN_CONTEXT_UNAVAILABLE
;
67 /* If EEXIST, we just ignore the error */
71 PERROR("add context ioctl");
72 ret
= LTTNG_ERR_KERN_CONTEXT_FAIL
;
79 cds_list_add_tail(&ctx
->list
, &chan
->ctx_list
);
84 trace_kernel_destroy_context(ctx
);
90 * Create a new kernel session, register it to the kernel tracer and add it to
91 * the session daemon session.
93 int kernel_create_session(struct ltt_session
*session
)
96 struct ltt_kernel_session
*lks
;
100 /* Allocate data structure */
101 lks
= trace_kernel_create_session();
107 /* Kernel tracer session creation */
108 ret
= kernctl_create_session(kernel_tracer_fd
);
110 PERROR("ioctl kernel create session");
115 /* Prevent fd duplication after execlp() */
116 ret
= fcntl(lks
->fd
, F_SETFD
, FD_CLOEXEC
);
118 PERROR("fcntl session fd");
121 lks
->id
= session
->id
;
122 lks
->consumer_fds_sent
= 0;
123 session
->kernel_session
= lks
;
125 DBG("Kernel session created (fd: %d)", lks
->fd
);
128 * This is necessary since the creation time is present in the session
129 * name when it is generated.
131 if (session
->has_auto_generated_name
) {
132 ret
= kernctl_session_set_name(lks
->fd
, DEFAULT_SESSION_NAME
);
134 ret
= kernctl_session_set_name(lks
->fd
, session
->name
);
137 WARN("Could not set kernel session name for session %" PRIu64
" name: %s",
138 session
->id
, session
->name
);
141 ret
= kernctl_session_set_creation_time(lks
->fd
, session
->creation_time
);
143 WARN("Could not set kernel session creation time for session %" PRIu64
" name: %s",
144 session
->id
, session
->name
);
151 trace_kernel_destroy_session(lks
);
152 trace_kernel_free_session(lks
);
158 * Create a kernel channel, register it to the kernel tracer and add it to the
161 int kernel_create_channel(struct ltt_kernel_session
*session
,
162 struct lttng_channel
*chan
)
165 struct ltt_kernel_channel
*lkc
;
170 /* Allocate kernel channel */
171 lkc
= trace_kernel_create_channel(chan
);
176 DBG3("Kernel create channel %s with attr: %d, %" PRIu64
", %" PRIu64
", %u, %u, %d, %d",
177 chan
->name
, lkc
->channel
->attr
.overwrite
,
178 lkc
->channel
->attr
.subbuf_size
, lkc
->channel
->attr
.num_subbuf
,
179 lkc
->channel
->attr
.switch_timer_interval
, lkc
->channel
->attr
.read_timer_interval
,
180 lkc
->channel
->attr
.live_timer_interval
, lkc
->channel
->attr
.output
);
182 /* Kernel tracer channel creation */
183 ret
= kernctl_create_channel(session
->fd
, &lkc
->channel
->attr
);
185 PERROR("ioctl kernel create channel");
189 /* Setup the channel fd */
191 /* Prevent fd duplication after execlp() */
192 ret
= fcntl(lkc
->fd
, F_SETFD
, FD_CLOEXEC
);
194 PERROR("fcntl session fd");
197 /* Add channel to session */
198 cds_list_add(&lkc
->list
, &session
->channel_list
.head
);
199 session
->channel_count
++;
200 lkc
->session
= session
;
201 lkc
->key
= ++next_kernel_channel_key
;
203 DBG("Kernel channel %s created (fd: %d, key: %" PRIu64
")",
204 lkc
->channel
->name
, lkc
->fd
, lkc
->key
);
217 * Compute the offset of the instrumentation byte in the binary based on the
218 * function probe location using the ELF lookup method.
220 * Returns 0 on success and set the offset out parameter to the offset of the
222 * Returns -1 on error
225 int extract_userspace_probe_offset_function_elf(
226 const struct lttng_userspace_probe_location
*probe_location
,
227 struct ltt_kernel_session
*session
, uint64_t *offset
)
231 const char *symbol
= NULL
;
232 const struct lttng_userspace_probe_location_lookup_method
*lookup
= NULL
;
233 enum lttng_userspace_probe_location_lookup_method_type lookup_method_type
;
235 assert(lttng_userspace_probe_location_get_type(probe_location
) ==
236 LTTNG_USERSPACE_PROBE_LOCATION_TYPE_FUNCTION
);
238 lookup
= lttng_userspace_probe_location_get_lookup_method(
246 lttng_userspace_probe_location_lookup_method_get_type(lookup
);
248 assert(lookup_method_type
==
249 LTTNG_USERSPACE_PROBE_LOCATION_LOOKUP_METHOD_TYPE_FUNCTION_ELF
);
251 symbol
= lttng_userspace_probe_location_function_get_function_name(
258 fd
= lttng_userspace_probe_location_function_get_binary_fd(probe_location
);
264 ret
= run_as_extract_elf_symbol_offset(fd
, symbol
, session
->uid
,
265 session
->gid
, offset
);
267 DBG("userspace probe offset calculation failed for "
268 "function %s", symbol
);
272 DBG("userspace probe elf offset for %s is 0x%jd", symbol
, (intmax_t)(*offset
));
278 * Compute the offsets of the instrumentation bytes in the binary based on the
279 * tracepoint probe location using the SDT lookup method. This function
280 * allocates the offsets buffer, the caller must free it.
282 * Returns 0 on success and set the offset out parameter to the offsets of the
284 * Returns -1 on error.
287 int extract_userspace_probe_offset_tracepoint_sdt(
288 const struct lttng_userspace_probe_location
*probe_location
,
289 struct ltt_kernel_session
*session
, uint64_t **offsets
,
290 uint32_t *offsets_count
)
292 enum lttng_userspace_probe_location_lookup_method_type lookup_method_type
;
293 const struct lttng_userspace_probe_location_lookup_method
*lookup
= NULL
;
294 const char *probe_name
= NULL
, *provider_name
= NULL
;
298 assert(lttng_userspace_probe_location_get_type(probe_location
) ==
299 LTTNG_USERSPACE_PROBE_LOCATION_TYPE_TRACEPOINT
);
301 lookup
= lttng_userspace_probe_location_get_lookup_method(probe_location
);
308 lttng_userspace_probe_location_lookup_method_get_type(lookup
);
310 assert(lookup_method_type
==
311 LTTNG_USERSPACE_PROBE_LOCATION_LOOKUP_METHOD_TYPE_TRACEPOINT_SDT
);
314 probe_name
= lttng_userspace_probe_location_tracepoint_get_probe_name(
321 provider_name
= lttng_userspace_probe_location_tracepoint_get_provider_name(
323 if (!provider_name
) {
328 fd
= lttng_userspace_probe_location_tracepoint_get_binary_fd(probe_location
);
334 ret
= run_as_extract_sdt_probe_offsets(fd
, provider_name
, probe_name
,
335 session
->uid
, session
->gid
, offsets
, offsets_count
);
337 DBG("userspace probe offset calculation failed for sdt "
338 "probe %s:%s", provider_name
, probe_name
);
342 if (*offsets_count
== 0) {
343 DBG("no userspace probe offset found");
347 DBG("%u userspace probe SDT offsets found for %s:%s at:",
348 *offsets_count
, provider_name
, probe_name
);
349 for (i
= 0; i
< *offsets_count
; i
++) {
350 DBG("\t0x%jd", (intmax_t)((*offsets
)[i
]));
357 * Extract the offsets of the instrumentation point for the different lookup
361 int userspace_probe_add_callsites(struct lttng_event
*ev
,
362 struct ltt_kernel_session
*session
, int fd
)
364 const struct lttng_userspace_probe_location_lookup_method
*lookup_method
= NULL
;
365 enum lttng_userspace_probe_location_lookup_method_type type
;
366 const struct lttng_userspace_probe_location
*location
= NULL
;
370 assert(ev
->type
== LTTNG_EVENT_USERSPACE_PROBE
);
372 location
= lttng_event_get_userspace_probe_location(ev
);
378 lttng_userspace_probe_location_get_lookup_method(location
);
379 if (!lookup_method
) {
384 type
= lttng_userspace_probe_location_lookup_method_get_type(lookup_method
);
386 case LTTNG_USERSPACE_PROBE_LOCATION_LOOKUP_METHOD_TYPE_FUNCTION_ELF
:
388 struct lttng_kernel_event_callsite callsite
;
391 ret
= extract_userspace_probe_offset_function_elf(location
, session
, &offset
);
393 ret
= LTTNG_ERR_PROBE_LOCATION_INVAL
;
397 callsite
.u
.uprobe
.offset
= offset
;
398 ret
= kernctl_add_callsite(fd
, &callsite
);
400 WARN("Adding callsite to userspace probe "
401 "event %s failed.", ev
->name
);
402 ret
= LTTNG_ERR_KERN_ENABLE_FAIL
;
407 case LTTNG_USERSPACE_PROBE_LOCATION_LOOKUP_METHOD_TYPE_TRACEPOINT_SDT
:
410 uint64_t *offsets
= NULL
;
411 uint32_t offsets_count
;
412 struct lttng_kernel_event_callsite callsite
;
415 * This call allocates the offsets buffer. This buffer must be freed
418 ret
= extract_userspace_probe_offset_tracepoint_sdt(location
, session
,
419 &offsets
, &offsets_count
);
421 ret
= LTTNG_ERR_PROBE_LOCATION_INVAL
;
424 for (i
= 0; i
< offsets_count
; i
++) {
425 callsite
.u
.uprobe
.offset
= offsets
[i
];
426 ret
= kernctl_add_callsite(fd
, &callsite
);
428 WARN("Adding callsite to userspace probe "
429 "event %s failed.", ev
->name
);
430 ret
= LTTNG_ERR_KERN_ENABLE_FAIL
;
439 ret
= LTTNG_ERR_PROBE_LOCATION_INVAL
;
447 * Create a kernel event, enable it to the kernel tracer and add it to the
448 * channel event list of the kernel session.
449 * We own filter_expression and filter.
451 int kernel_create_event(struct lttng_event
*ev
,
452 struct ltt_kernel_channel
*channel
,
453 char *filter_expression
,
454 struct lttng_filter_bytecode
*filter
)
457 enum lttng_error_code ret
;
458 struct ltt_kernel_event
*event
;
463 /* We pass ownership of filter_expression and filter */
464 ret
= trace_kernel_create_event(ev
, filter_expression
,
466 if (ret
!= LTTNG_OK
) {
470 fd
= kernctl_create_event(channel
->fd
, event
->event
);
474 ret
= LTTNG_ERR_KERN_EVENT_EXIST
;
477 WARN("Event type not implemented");
478 ret
= LTTNG_ERR_KERN_EVENT_ENOSYS
;
481 WARN("Event %s not found!", ev
->name
);
482 ret
= LTTNG_ERR_KERN_ENABLE_FAIL
;
485 ret
= LTTNG_ERR_KERN_ENABLE_FAIL
;
486 PERROR("create event ioctl");
491 event
->type
= ev
->type
;
493 /* Prevent fd duplication after execlp() */
494 err
= fcntl(event
->fd
, F_SETFD
, FD_CLOEXEC
);
496 PERROR("fcntl session fd");
500 err
= kernctl_filter(event
->fd
, filter
);
504 ret
= LTTNG_ERR_FILTER_NOMEM
;
507 ret
= LTTNG_ERR_FILTER_INVAL
;
514 if (ev
->type
== LTTNG_EVENT_USERSPACE_PROBE
) {
515 ret
= userspace_probe_add_callsites(ev
, channel
->session
, event
->fd
);
517 goto add_callsite_error
;
521 err
= kernctl_enable(event
->fd
);
525 ret
= LTTNG_ERR_KERN_EVENT_EXIST
;
528 PERROR("enable kernel event");
529 ret
= LTTNG_ERR_KERN_ENABLE_FAIL
;
535 /* Add event to event list */
536 cds_list_add(&event
->list
, &channel
->events_list
.head
);
537 channel
->event_count
++;
539 DBG("Event %s created (fd: %d)", ev
->name
, event
->fd
);
549 closeret
= close(event
->fd
);
551 PERROR("close event fd");
561 * Disable a kernel channel.
563 int kernel_disable_channel(struct ltt_kernel_channel
*chan
)
569 ret
= kernctl_disable(chan
->fd
);
571 PERROR("disable chan ioctl");
576 DBG("Kernel channel %s disabled (fd: %d, key: %" PRIu64
")",
577 chan
->channel
->name
, chan
->fd
, chan
->key
);
586 * Enable a kernel channel.
588 int kernel_enable_channel(struct ltt_kernel_channel
*chan
)
594 ret
= kernctl_enable(chan
->fd
);
595 if (ret
< 0 && ret
!= -EEXIST
) {
596 PERROR("Enable kernel chan");
601 DBG("Kernel channel %s enabled (fd: %d, key: %" PRIu64
")",
602 chan
->channel
->name
, chan
->fd
, chan
->key
);
611 * Enable a kernel event.
613 int kernel_enable_event(struct ltt_kernel_event
*event
)
619 ret
= kernctl_enable(event
->fd
);
623 ret
= LTTNG_ERR_KERN_EVENT_EXIST
;
626 PERROR("enable kernel event");
633 DBG("Kernel event %s enabled (fd: %d)", event
->event
->name
, event
->fd
);
642 * Disable a kernel event.
644 int kernel_disable_event(struct ltt_kernel_event
*event
)
650 ret
= kernctl_disable(event
->fd
);
654 ret
= LTTNG_ERR_KERN_EVENT_EXIST
;
657 PERROR("disable kernel event");
664 DBG("Kernel event %s disabled (fd: %d)", event
->event
->name
, event
->fd
);
672 static struct lttng_tracker_list
*get_id_tracker_list(
673 struct ltt_kernel_session
*session
,
674 enum lttng_tracker_type tracker_type
)
676 switch (tracker_type
) {
677 case LTTNG_TRACKER_PID
:
678 return session
->tracker_list_pid
;
679 case LTTNG_TRACKER_VPID
:
680 return session
->tracker_list_vpid
;
681 case LTTNG_TRACKER_UID
:
682 return session
->tracker_list_uid
;
683 case LTTNG_TRACKER_VUID
:
684 return session
->tracker_list_vuid
;
685 case LTTNG_TRACKER_GID
:
686 return session
->tracker_list_gid
;
687 case LTTNG_TRACKER_VGID
:
688 return session
->tracker_list_vgid
;
694 int kernel_track_id(enum lttng_tracker_type tracker_type
,
695 struct ltt_kernel_session
*session
,
696 const struct lttng_tracker_id
*id
)
699 struct lttng_tracker_list
*tracker_list
;
700 struct lttng_tracker_ids
*saved_ids
;
702 ret
= lttng_tracker_id_lookup_string(tracker_type
, id
, &value
);
703 if (ret
!= LTTNG_OK
) {
707 tracker_list
= get_id_tracker_list(session
, tracker_type
);
709 return LTTNG_ERR_INVALID
;
712 /* Save list for restore on error. */
713 ret
= lttng_tracker_id_get_list(tracker_list
, &saved_ids
);
714 if (ret
!= LTTNG_OK
) {
715 return LTTNG_ERR_INVALID
;
719 ret
= lttng_tracker_list_add(tracker_list
, id
);
720 if (ret
!= LTTNG_OK
) {
724 switch (tracker_type
) {
725 case LTTNG_TRACKER_PID
:
726 DBG("Kernel track PID %d for session id %" PRIu64
".", value
,
728 ret
= kernctl_track_pid(session
->fd
, value
);
734 case LTTNG_TRACKER_VPID
:
735 DBG("Kernel track VPID %d for session id %" PRIu64
".", value
,
737 ret
= kernctl_track_id(session
->fd
, LTTNG_TRACKER_VPID
, value
);
743 case LTTNG_TRACKER_UID
:
744 DBG("Kernel track UID %d for session id %" PRIu64
".", value
,
746 ret
= kernctl_track_id(session
->fd
, LTTNG_TRACKER_UID
, value
);
752 case LTTNG_TRACKER_GID
:
753 DBG("Kernel track GID %d for session id %" PRIu64
".", value
,
755 ret
= kernctl_track_id(session
->fd
, LTTNG_TRACKER_GID
, value
);
761 case LTTNG_TRACKER_VUID
:
762 DBG("Kernel track VUID %d for session id %" PRIu64
".", value
,
764 ret
= kernctl_track_id(session
->fd
, LTTNG_TRACKER_VUID
, value
);
770 case LTTNG_TRACKER_VGID
:
771 DBG("Kernel track VGID %d for session id %" PRIu64
".", value
,
773 ret
= kernctl_track_id(session
->fd
, LTTNG_TRACKER_VGID
, value
);
784 /* Error handling. */
787 ret
= LTTNG_ERR_INVALID
;
790 ret
= LTTNG_ERR_NOMEM
;
793 ret
= LTTNG_ERR_ID_TRACKED
;
800 if (lttng_tracker_id_set_list(tracker_list
, saved_ids
) != LTTNG_OK
) {
801 ERR("Error on tracker add error handling.\n");
804 lttng_tracker_ids_destroy(saved_ids
);
808 int kernel_untrack_id(enum lttng_tracker_type tracker_type
,
809 struct ltt_kernel_session
*session
,
810 const struct lttng_tracker_id
*id
)
813 struct lttng_tracker_list
*tracker_list
;
814 struct lttng_tracker_ids
*saved_ids
;
816 ret
= lttng_tracker_id_lookup_string(tracker_type
, id
, &value
);
817 if (ret
!= LTTNG_OK
) {
821 tracker_list
= get_id_tracker_list(session
, tracker_type
);
823 return LTTNG_ERR_INVALID
;
825 /* Save list for restore on error. */
826 ret
= lttng_tracker_id_get_list(tracker_list
, &saved_ids
);
827 if (ret
!= LTTNG_OK
) {
828 return LTTNG_ERR_INVALID
;
830 /* Remove from list. */
831 ret
= lttng_tracker_list_remove(tracker_list
, id
);
832 if (ret
!= LTTNG_OK
) {
836 switch (tracker_type
) {
837 case LTTNG_TRACKER_PID
:
838 DBG("Kernel untrack PID %d for session id %" PRIu64
".", value
,
840 ret
= kernctl_untrack_pid(session
->fd
, value
);
846 case LTTNG_TRACKER_VPID
:
847 DBG("Kernel untrack VPID %d for session id %" PRIu64
".", value
,
849 ret
= kernctl_untrack_id(
850 session
->fd
, LTTNG_TRACKER_VPID
, value
);
856 case LTTNG_TRACKER_UID
:
857 DBG("Kernel untrack UID %d for session id %" PRIu64
".", value
,
859 ret
= kernctl_untrack_id(session
->fd
, LTTNG_TRACKER_UID
, value
);
865 case LTTNG_TRACKER_GID
:
866 DBG("Kernel untrack GID %d for session id %" PRIu64
".", value
,
868 ret
= kernctl_untrack_id(session
->fd
, LTTNG_TRACKER_GID
, value
);
874 case LTTNG_TRACKER_VUID
:
875 DBG("Kernel untrack VUID %d for session id %" PRIu64
".", value
,
877 ret
= kernctl_untrack_id(
878 session
->fd
, LTTNG_TRACKER_VUID
, value
);
884 case LTTNG_TRACKER_VGID
:
885 DBG("Kernel untrack VGID %d for session id %" PRIu64
".", value
,
887 ret
= kernctl_untrack_id(
888 session
->fd
, LTTNG_TRACKER_VGID
, value
);
899 /* Error handling. */
902 ret
= LTTNG_ERR_INVALID
;
905 ret
= LTTNG_ERR_NOMEM
;
908 ret
= LTTNG_ERR_ID_TRACKED
;
915 if (lttng_tracker_id_set_list(tracker_list
, saved_ids
) != LTTNG_OK
) {
916 ERR("Error on tracker remove error handling.\n");
919 lttng_tracker_ids_destroy(saved_ids
);
924 * Called with session lock held.
926 int kernel_list_tracker_ids(enum lttng_tracker_type tracker_type
,
927 struct ltt_kernel_session
*session
,
928 struct lttng_tracker_ids
**_ids
)
931 struct lttng_tracker_list
*tracker_list
;
933 tracker_list
= get_id_tracker_list(session
, tracker_type
);
935 ret
= -LTTNG_ERR_INVALID
;
939 ret
= lttng_tracker_id_get_list(tracker_list
, _ids
);
940 if (ret
!= LTTNG_OK
) {
941 ret
= -LTTNG_ERR_INVALID
;
950 * Create kernel metadata, open from the kernel tracer and add it to the
953 int kernel_open_metadata(struct ltt_kernel_session
*session
)
956 struct ltt_kernel_metadata
*lkm
= NULL
;
960 /* Allocate kernel metadata */
961 lkm
= trace_kernel_create_metadata();
966 /* Kernel tracer metadata creation */
967 ret
= kernctl_open_metadata(session
->fd
, &lkm
->conf
->attr
);
973 lkm
->key
= ++next_kernel_channel_key
;
974 /* Prevent fd duplication after execlp() */
975 ret
= fcntl(lkm
->fd
, F_SETFD
, FD_CLOEXEC
);
977 PERROR("fcntl session fd");
980 session
->metadata
= lkm
;
982 DBG("Kernel metadata opened (fd: %d)", lkm
->fd
);
987 trace_kernel_destroy_metadata(lkm
);
993 * Start tracing session.
995 int kernel_start_session(struct ltt_kernel_session
*session
)
1001 ret
= kernctl_start_session(session
->fd
);
1003 PERROR("ioctl start session");
1007 DBG("Kernel session started");
1016 * Make a kernel wait to make sure in-flight probe have completed.
1018 void kernel_wait_quiescent(void)
1021 int fd
= kernel_tracer_fd
;
1023 DBG("Kernel quiescent wait on %d", fd
);
1025 ret
= kernctl_wait_quiescent(fd
);
1027 PERROR("wait quiescent ioctl");
1028 ERR("Kernel quiescent wait failed");
1033 * Force flush buffer of metadata.
1035 int kernel_metadata_flush_buffer(int fd
)
1039 DBG("Kernel flushing metadata buffer on fd %d", fd
);
1041 ret
= kernctl_buffer_flush(fd
);
1043 ERR("Fail to flush metadata buffers %d (ret: %d)", fd
, ret
);
1050 * Force flush buffer for channel.
1052 int kernel_flush_buffer(struct ltt_kernel_channel
*channel
)
1055 struct ltt_kernel_stream
*stream
;
1059 DBG("Flush buffer for channel %s", channel
->channel
->name
);
1061 cds_list_for_each_entry(stream
, &channel
->stream_list
.head
, list
) {
1062 DBG("Flushing channel stream %d", stream
->fd
);
1063 ret
= kernctl_buffer_flush(stream
->fd
);
1066 ERR("Fail to flush buffer for stream %d (ret: %d)",
1075 * Stop tracing session.
1077 int kernel_stop_session(struct ltt_kernel_session
*session
)
1083 ret
= kernctl_stop_session(session
->fd
);
1088 DBG("Kernel session stopped");
1097 * Open stream of channel, register it to the kernel tracer and add it
1098 * to the stream list of the channel.
1100 * Note: given that the streams may appear in random order wrt CPU
1101 * number (e.g. cpu hotplug), the index value of the stream number in
1102 * the stream name is not necessarily linked to the CPU number.
1104 * Return the number of created stream. Else, a negative value.
1106 int kernel_open_channel_stream(struct ltt_kernel_channel
*channel
)
1109 struct ltt_kernel_stream
*lks
;
1113 while ((ret
= kernctl_create_stream(channel
->fd
)) >= 0) {
1114 lks
= trace_kernel_create_stream(channel
->channel
->name
,
1115 channel
->stream_count
);
1125 /* Prevent fd duplication after execlp() */
1126 ret
= fcntl(lks
->fd
, F_SETFD
, FD_CLOEXEC
);
1128 PERROR("fcntl session fd");
1131 lks
->tracefile_size
= channel
->channel
->attr
.tracefile_size
;
1132 lks
->tracefile_count
= channel
->channel
->attr
.tracefile_count
;
1134 /* Add stream to channel stream list */
1135 cds_list_add(&lks
->list
, &channel
->stream_list
.head
);
1136 channel
->stream_count
++;
1138 DBG("Kernel stream %s created (fd: %d, state: %d)", lks
->name
, lks
->fd
,
1142 return channel
->stream_count
;
1149 * Open the metadata stream and set it to the kernel session.
1151 int kernel_open_metadata_stream(struct ltt_kernel_session
*session
)
1157 ret
= kernctl_create_stream(session
->metadata
->fd
);
1159 PERROR("kernel create metadata stream");
1163 DBG("Kernel metadata stream created (fd: %d)", ret
);
1164 session
->metadata_stream_fd
= ret
;
1165 /* Prevent fd duplication after execlp() */
1166 ret
= fcntl(session
->metadata_stream_fd
, F_SETFD
, FD_CLOEXEC
);
1168 PERROR("fcntl session fd");
1178 * Get the event list from the kernel tracer and return the number of elements.
1180 ssize_t
kernel_list_events(struct lttng_event
**events
)
1184 size_t nbmem
, count
= 0;
1186 struct lttng_event
*elist
;
1190 fd
= kernctl_tracepoint_list(kernel_tracer_fd
);
1192 PERROR("kernel tracepoint list");
1196 fp
= fdopen(fd
, "r");
1198 PERROR("kernel tracepoint list fdopen");
1203 * Init memory size counter
1204 * See kernel-ctl.h for explanation of this value
1206 nbmem
= KERNEL_EVENT_INIT_LIST_SIZE
;
1207 elist
= zmalloc(sizeof(struct lttng_event
) * nbmem
);
1208 if (elist
== NULL
) {
1209 PERROR("alloc list events");
1214 while (fscanf(fp
, "event { name = %m[^;]; };\n", &event
) == 1) {
1215 if (count
>= nbmem
) {
1216 struct lttng_event
*new_elist
;
1219 new_nbmem
= nbmem
<< 1;
1220 DBG("Reallocating event list from %zu to %zu bytes",
1222 new_elist
= realloc(elist
, new_nbmem
* sizeof(struct lttng_event
));
1223 if (new_elist
== NULL
) {
1224 PERROR("realloc list events");
1230 /* Zero the new memory */
1231 memset(new_elist
+ nbmem
, 0,
1232 (new_nbmem
- nbmem
) * sizeof(struct lttng_event
));
1236 strncpy(elist
[count
].name
, event
, LTTNG_SYMBOL_NAME_LEN
);
1237 elist
[count
].name
[LTTNG_SYMBOL_NAME_LEN
- 1] = '\0';
1238 elist
[count
].enabled
= -1;
1244 DBG("Kernel list events done (%zu events)", count
);
1246 ret
= fclose(fp
); /* closes both fp and fd */
1262 * Get kernel version and validate it.
1264 int kernel_validate_version(struct lttng_kernel_tracer_version
*version
,
1265 struct lttng_kernel_tracer_abi_version
*abi_version
)
1269 ret
= kernctl_tracer_version(kernel_tracer_fd
, version
);
1271 ERR("Failed to retrieve the lttng-modules version");
1275 /* Validate version */
1276 if (version
->major
!= VERSION_MAJOR
) {
1277 ERR("Kernel tracer major version (%d) is not compatible with lttng-tools major version (%d)",
1278 version
->major
, VERSION_MAJOR
);
1281 ret
= kernctl_tracer_abi_version(kernel_tracer_fd
, abi_version
);
1283 ERR("Failed to retrieve lttng-modules ABI version");
1286 if (abi_version
->major
!= LTTNG_MODULES_ABI_MAJOR_VERSION
) {
1287 ERR("Kernel tracer ABI version (%d.%d) does not match the expected ABI major version (%d.*)",
1288 abi_version
->major
, abi_version
->minor
,
1289 LTTNG_MODULES_ABI_MAJOR_VERSION
);
1292 DBG2("Kernel tracer version validated (%d.%d, ABI %d.%d)",
1293 version
->major
, version
->minor
,
1294 abi_version
->major
, abi_version
->minor
);
1301 ERR("Kernel tracer version check failed; kernel tracing will not be available");
1306 * Kernel work-arounds called at the start of sessiond main().
1308 int init_kernel_workarounds(void)
1314 * boot_id needs to be read once before being used concurrently
1315 * to deal with a Linux kernel race. A fix is proposed for
1316 * upstream, but the work-around is needed for older kernels.
1318 fp
= fopen("/proc/sys/kernel/random/boot_id", "r");
1325 ret
= fread(buf
, 1, sizeof(buf
), fp
);
1327 /* Ignore error, we don't really care */
1339 * Teardown of a kernel session, keeping data required by destroy notifiers.
1341 void kernel_destroy_session(struct ltt_kernel_session
*ksess
)
1343 struct lttng_trace_chunk
*trace_chunk
;
1345 if (ksess
== NULL
) {
1346 DBG3("No kernel session when tearing down session");
1350 DBG("Tearing down kernel session");
1351 trace_chunk
= ksess
->current_trace_chunk
;
1354 * Destroy channels on the consumer if at least one FD has been sent and we
1355 * are in no output mode because the streams are in *no* monitor mode so we
1356 * have to send a command to clean them up or else they leaked.
1358 if (!ksess
->output_traces
&& ksess
->consumer_fds_sent
) {
1360 struct consumer_socket
*socket
;
1361 struct lttng_ht_iter iter
;
1363 /* For each consumer socket. */
1365 cds_lfht_for_each_entry(ksess
->consumer
->socks
->ht
, &iter
.iter
,
1366 socket
, node
.node
) {
1367 struct ltt_kernel_channel
*chan
;
1369 /* For each channel, ask the consumer to destroy it. */
1370 cds_list_for_each_entry(chan
, &ksess
->channel_list
.head
, list
) {
1371 ret
= kernel_consumer_destroy_channel(socket
, chan
);
1373 /* Consumer is probably dead. Use next socket. */
1381 /* Close any relayd session */
1382 consumer_output_send_destroy_relayd(ksess
->consumer
);
1384 trace_kernel_destroy_session(ksess
);
1385 lttng_trace_chunk_put(trace_chunk
);
1388 /* Teardown of data required by destroy notifiers. */
1389 void kernel_free_session(struct ltt_kernel_session
*ksess
)
1391 if (ksess
== NULL
) {
1394 trace_kernel_free_session(ksess
);
1398 * Destroy a kernel channel object. It does not do anything on the tracer side.
1400 void kernel_destroy_channel(struct ltt_kernel_channel
*kchan
)
1402 struct ltt_kernel_session
*ksess
= NULL
;
1405 assert(kchan
->channel
);
1407 DBG3("Kernel destroy channel %s", kchan
->channel
->name
);
1409 /* Update channel count of associated session. */
1410 if (kchan
->session
) {
1411 /* Keep pointer reference so we can update it after the destroy. */
1412 ksess
= kchan
->session
;
1415 trace_kernel_destroy_channel(kchan
);
1418 * At this point the kernel channel is not visible anymore. This is safe
1419 * since in order to work on a visible kernel session, the tracing session
1420 * lock (ltt_session.lock) MUST be acquired.
1423 ksess
->channel_count
--;
1428 * Take a snapshot for a given kernel session.
1430 * Return LTTNG_OK on success or else return a LTTNG_ERR code.
1432 enum lttng_error_code
kernel_snapshot_record(
1433 struct ltt_kernel_session
*ksess
,
1434 const struct consumer_output
*output
, int wait
,
1435 uint64_t nb_packets_per_stream
)
1437 int err
, ret
, saved_metadata_fd
;
1438 enum lttng_error_code status
= LTTNG_OK
;
1439 struct consumer_socket
*socket
;
1440 struct lttng_ht_iter iter
;
1441 struct ltt_kernel_metadata
*saved_metadata
;
1442 char *trace_path
= NULL
;
1443 size_t consumer_path_offset
= 0;
1446 assert(ksess
->consumer
);
1449 DBG("Kernel snapshot record started");
1451 /* Save current metadata since the following calls will change it. */
1452 saved_metadata
= ksess
->metadata
;
1453 saved_metadata_fd
= ksess
->metadata_stream_fd
;
1457 ret
= kernel_open_metadata(ksess
);
1459 status
= LTTNG_ERR_KERN_META_FAIL
;
1463 ret
= kernel_open_metadata_stream(ksess
);
1465 status
= LTTNG_ERR_KERN_META_FAIL
;
1466 goto error_open_stream
;
1469 trace_path
= setup_channel_trace_path(ksess
->consumer
,
1470 DEFAULT_KERNEL_TRACE_DIR
, &consumer_path_offset
);
1472 status
= LTTNG_ERR_INVALID
;
1475 /* Send metadata to consumer and snapshot everything. */
1476 cds_lfht_for_each_entry(output
->socks
->ht
, &iter
.iter
,
1477 socket
, node
.node
) {
1478 struct ltt_kernel_channel
*chan
;
1480 pthread_mutex_lock(socket
->lock
);
1481 /* This stream must not be monitored by the consumer. */
1482 ret
= kernel_consumer_add_metadata(socket
, ksess
, 0);
1483 pthread_mutex_unlock(socket
->lock
);
1485 status
= LTTNG_ERR_KERN_META_FAIL
;
1486 goto error_consumer
;
1489 /* For each channel, ask the consumer to snapshot it. */
1490 cds_list_for_each_entry(chan
, &ksess
->channel_list
.head
, list
) {
1491 status
= consumer_snapshot_channel(socket
, chan
->key
, output
, 0,
1492 ksess
->uid
, ksess
->gid
,
1493 &trace_path
[consumer_path_offset
], wait
,
1494 nb_packets_per_stream
);
1495 if (status
!= LTTNG_OK
) {
1496 (void) kernel_consumer_destroy_metadata(socket
,
1498 goto error_consumer
;
1502 /* Snapshot metadata, */
1503 status
= consumer_snapshot_channel(socket
, ksess
->metadata
->key
, output
,
1504 1, ksess
->uid
, ksess
->gid
, &trace_path
[consumer_path_offset
],
1506 if (status
!= LTTNG_OK
) {
1507 goto error_consumer
;
1511 * The metadata snapshot is done, ask the consumer to destroy it since
1512 * it's not monitored on the consumer side.
1514 (void) kernel_consumer_destroy_metadata(socket
, ksess
->metadata
);
1518 /* Close newly opened metadata stream. It's now on the consumer side. */
1519 err
= close(ksess
->metadata_stream_fd
);
1521 PERROR("close snapshot kernel");
1525 trace_kernel_destroy_metadata(ksess
->metadata
);
1527 /* Restore metadata state.*/
1528 ksess
->metadata
= saved_metadata
;
1529 ksess
->metadata_stream_fd
= saved_metadata_fd
;
1536 * Get the syscall mask array from the kernel tracer.
1538 * Return 0 on success else a negative value. In both case, syscall_mask should
1541 int kernel_syscall_mask(int chan_fd
, char **syscall_mask
, uint32_t *nr_bits
)
1543 assert(syscall_mask
);
1546 return kernctl_syscall_mask(chan_fd
, syscall_mask
, nr_bits
);
1550 * Check for the support of the RING_BUFFER_SNAPSHOT_SAMPLE_POSITIONS via abi
1553 * Return 1 on success, 0 when feature is not supported, negative value in case
1556 int kernel_supports_ring_buffer_snapshot_sample_positions(void)
1558 int ret
= 0; // Not supported by default
1559 struct lttng_kernel_tracer_abi_version abi
;
1561 ret
= kernctl_tracer_abi_version(kernel_tracer_fd
, &abi
);
1563 ERR("Failed to retrieve lttng-modules ABI version");
1568 * RING_BUFFER_SNAPSHOT_SAMPLE_POSITIONS was introduced in 2.3
1570 if (abi
.major
>= 2 && abi
.minor
>= 3) {
1582 * Check for the support of the packet sequence number via abi version number.
1584 * Return 1 on success, 0 when feature is not supported, negative value in case
1587 int kernel_supports_ring_buffer_packet_sequence_number(void)
1589 int ret
= 0; // Not supported by default
1590 struct lttng_kernel_tracer_abi_version abi
;
1592 ret
= kernctl_tracer_abi_version(kernel_tracer_fd
, &abi
);
1594 ERR("Failed to retrieve lttng-modules ABI version");
1599 * Packet sequence number was introduced in LTTng 2.8,
1600 * lttng-modules ABI 2.1.
1602 if (abi
.major
>= 2 && abi
.minor
>= 1) {
1614 * Rotate a kernel session.
1616 * Return LTTNG_OK on success or else an LTTng error code.
1618 enum lttng_error_code
kernel_rotate_session(struct ltt_session
*session
)
1621 enum lttng_error_code status
= LTTNG_OK
;
1622 struct consumer_socket
*socket
;
1623 struct lttng_ht_iter iter
;
1624 struct ltt_kernel_session
*ksess
= session
->kernel_session
;
1627 assert(ksess
->consumer
);
1629 DBG("Rotate kernel session %s started (session %" PRIu64
")",
1630 session
->name
, session
->id
);
1635 * Note that this loop will end after one iteration given that there is
1636 * only one kernel consumer.
1638 cds_lfht_for_each_entry(ksess
->consumer
->socks
->ht
, &iter
.iter
,
1639 socket
, node
.node
) {
1640 struct ltt_kernel_channel
*chan
;
1642 /* For each channel, ask the consumer to rotate it. */
1643 cds_list_for_each_entry(chan
, &ksess
->channel_list
.head
, list
) {
1644 DBG("Rotate kernel channel %" PRIu64
", session %s",
1645 chan
->key
, session
->name
);
1646 ret
= consumer_rotate_channel(socket
, chan
->key
,
1647 ksess
->uid
, ksess
->gid
, ksess
->consumer
,
1648 /* is_metadata_channel */ false);
1650 status
= LTTNG_ERR_KERN_CONSUMER_FAIL
;
1656 * Rotate the metadata channel.
1658 ret
= consumer_rotate_channel(socket
, ksess
->metadata
->key
,
1659 ksess
->uid
, ksess
->gid
, ksess
->consumer
,
1660 /* is_metadata_channel */ true);
1662 status
= LTTNG_ERR_KERN_CONSUMER_FAIL
;
1672 enum lttng_error_code
kernel_create_channel_subdirectories(
1673 const struct ltt_kernel_session
*ksess
)
1675 enum lttng_error_code ret
= LTTNG_OK
;
1676 enum lttng_trace_chunk_status chunk_status
;
1679 assert(ksess
->current_trace_chunk
);
1682 * Create the index subdirectory which will take care
1683 * of implicitly creating the channel's path.
1685 chunk_status
= lttng_trace_chunk_create_subdirectory(
1686 ksess
->current_trace_chunk
,
1687 DEFAULT_KERNEL_TRACE_DIR
"/" DEFAULT_INDEX_DIR
);
1688 if (chunk_status
!= LTTNG_TRACE_CHUNK_STATUS_OK
) {
1689 ret
= LTTNG_ERR_CREATE_DIR_FAIL
;
1698 * Setup necessary data for kernel tracer action.
1701 int init_kernel_tracer(void)
1704 bool is_root
= !getuid();
1706 /* Modprobe lttng kernel modules */
1707 ret
= modprobe_lttng_control();
1712 /* Open debugfs lttng */
1713 kernel_tracer_fd
= open(module_proc_lttng
, O_RDWR
);
1714 if (kernel_tracer_fd
< 0) {
1715 DBG("Failed to open %s", module_proc_lttng
);
1719 /* Validate kernel version */
1720 ret
= kernel_validate_version(&kernel_tracer_version
,
1721 &kernel_tracer_abi_version
);
1726 ret
= modprobe_lttng_data();
1731 ret
= kernel_supports_ring_buffer_snapshot_sample_positions();
1737 WARN("Kernel tracer does not support buffer monitoring. "
1738 "The monitoring timer of channels in the kernel domain "
1739 "will be set to 0 (disabled).");
1742 DBG("Kernel tracer fd %d", kernel_tracer_fd
);
1744 ret
= syscall_init_table(kernel_tracer_fd
);
1746 ERR("Unable to populate syscall table. Syscall tracing won't "
1747 "work for this session daemon.");
1752 modprobe_remove_lttng_control();
1753 ret
= close(kernel_tracer_fd
);
1757 kernel_tracer_fd
= -1;
1758 return LTTNG_ERR_KERN_VERSION
;
1761 ret
= close(kernel_tracer_fd
);
1767 modprobe_remove_lttng_control();
1770 WARN("No kernel tracer available");
1771 kernel_tracer_fd
= -1;
1773 return LTTNG_ERR_NEED_ROOT_SESSIOND
;
1775 return LTTNG_ERR_KERN_NA
;
1780 void cleanup_kernel_tracer(void)
1784 DBG2("Closing kernel fd");
1785 if (kernel_tracer_fd
>= 0) {
1786 ret
= close(kernel_tracer_fd
);
1790 kernel_tracer_fd
= -1;
1792 DBG("Unloading kernel modules");
1793 modprobe_remove_lttng_all();
1794 free(syscall_table
);
1798 bool kernel_tracer_is_initialized(void)
1800 return kernel_tracer_fd
>= 0;
1804 * Clear a kernel session.
1806 * Return LTTNG_OK on success or else an LTTng error code.
1808 enum lttng_error_code
kernel_clear_session(struct ltt_session
*session
)
1811 enum lttng_error_code status
= LTTNG_OK
;
1812 struct consumer_socket
*socket
;
1813 struct lttng_ht_iter iter
;
1814 struct ltt_kernel_session
*ksess
= session
->kernel_session
;
1817 assert(ksess
->consumer
);
1819 DBG("Clear kernel session %s (session %" PRIu64
")",
1820 session
->name
, session
->id
);
1824 if (ksess
->active
) {
1825 ERR("Expecting inactive session %s (%" PRIu64
")", session
->name
, session
->id
);
1826 status
= LTTNG_ERR_FATAL
;
1831 * Note that this loop will end after one iteration given that there is
1832 * only one kernel consumer.
1834 cds_lfht_for_each_entry(ksess
->consumer
->socks
->ht
, &iter
.iter
,
1835 socket
, node
.node
) {
1836 struct ltt_kernel_channel
*chan
;
1838 /* For each channel, ask the consumer to clear it. */
1839 cds_list_for_each_entry(chan
, &ksess
->channel_list
.head
, list
) {
1840 DBG("Clear kernel channel %" PRIu64
", session %s",
1841 chan
->key
, session
->name
);
1842 ret
= consumer_clear_channel(socket
, chan
->key
);
1848 if (!ksess
->metadata
) {
1850 * Nothing to do for the metadata.
1851 * This is a snapshot session.
1852 * The metadata is genererated on the fly.
1858 * Clear the metadata channel.
1859 * Metadata channel is not cleared per se but we still need to
1860 * perform a rotation operation on it behind the scene.
1862 ret
= consumer_clear_channel(socket
, ksess
->metadata
->key
);
1871 case LTTCOMM_CONSUMERD_RELAYD_CLEAR_DISALLOWED
:
1872 status
= LTTNG_ERR_CLEAR_RELAY_DISALLOWED
;
1875 status
= LTTNG_ERR_CLEAR_FAIL_CONSUMER
;