2 * Copyright (C) 2011 - David Goulet <david.goulet@polymtl.ca>
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 only,
6 * as 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 along
14 * with this program; if not, write to the Free Software Foundation, Inc.,
15 * 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA.
27 #include <common/common.h>
28 #include <common/kernel-ctl/kernel-ctl.h>
29 #include <common/kernel-ctl/kernel-ioctl.h>
30 #include <common/sessiond-comm/sessiond-comm.h>
34 #include "kernel-consumer.h"
35 #include "kern-modules.h"
39 * Add context on a kernel channel.
41 * Assumes the ownership of ctx.
43 int kernel_add_channel_context(struct ltt_kernel_channel
*chan
,
44 struct ltt_kernel_context
*ctx
)
51 DBG("Adding context to channel %s", chan
->channel
->name
);
52 ret
= kernctl_add_context(chan
->fd
, &ctx
->ctx
);
56 /* Exists but not available for this kernel */
57 ret
= LTTNG_ERR_KERN_CONTEXT_UNAVAILABLE
;
60 /* If EEXIST, we just ignore the error */
64 PERROR("add context ioctl");
65 ret
= LTTNG_ERR_KERN_CONTEXT_FAIL
;
71 cds_list_add_tail(&ctx
->list
, &chan
->ctx_list
);
75 trace_kernel_destroy_context(ctx
);
81 * Create a new kernel session, register it to the kernel tracer and add it to
82 * the session daemon session.
84 int kernel_create_session(struct ltt_session
*session
, int tracer_fd
)
87 struct ltt_kernel_session
*lks
;
91 /* Allocate data structure */
92 lks
= trace_kernel_create_session();
98 /* Kernel tracer session creation */
99 ret
= kernctl_create_session(tracer_fd
);
101 PERROR("ioctl kernel create session");
106 /* Prevent fd duplication after execlp() */
107 ret
= fcntl(lks
->fd
, F_SETFD
, FD_CLOEXEC
);
109 PERROR("fcntl session fd");
112 lks
->id
= session
->id
;
113 lks
->consumer_fds_sent
= 0;
114 session
->kernel_session
= lks
;
116 DBG("Kernel session created (fd: %d)", lks
->fd
);
122 trace_kernel_destroy_session(lks
);
128 * Create a kernel channel, register it to the kernel tracer and add it to the
131 int kernel_create_channel(struct ltt_kernel_session
*session
,
132 struct lttng_channel
*chan
)
135 struct ltt_kernel_channel
*lkc
;
140 /* Allocate kernel channel */
141 lkc
= trace_kernel_create_channel(chan
);
146 DBG3("Kernel create channel %s with attr: %d, %" PRIu64
", %" PRIu64
", %u, %u, %d, %d",
147 chan
->name
, lkc
->channel
->attr
.overwrite
,
148 lkc
->channel
->attr
.subbuf_size
, lkc
->channel
->attr
.num_subbuf
,
149 lkc
->channel
->attr
.switch_timer_interval
, lkc
->channel
->attr
.read_timer_interval
,
150 lkc
->channel
->attr
.live_timer_interval
, lkc
->channel
->attr
.output
);
152 /* Kernel tracer channel creation */
153 ret
= kernctl_create_channel(session
->fd
, &lkc
->channel
->attr
);
155 PERROR("ioctl kernel create channel");
159 /* Setup the channel fd */
161 /* Prevent fd duplication after execlp() */
162 ret
= fcntl(lkc
->fd
, F_SETFD
, FD_CLOEXEC
);
164 PERROR("fcntl session fd");
167 /* Add channel to session */
168 cds_list_add(&lkc
->list
, &session
->channel_list
.head
);
169 session
->channel_count
++;
170 lkc
->session
= session
;
172 DBG("Kernel channel %s created (fd: %d)", lkc
->channel
->name
, lkc
->fd
);
185 * Create a kernel event, enable it to the kernel tracer and add it to the
186 * channel event list of the kernel session.
187 * We own filter_expression and filter.
189 int kernel_create_event(struct lttng_event
*ev
,
190 struct ltt_kernel_channel
*channel
,
191 char *filter_expression
,
192 struct lttng_filter_bytecode
*filter
)
195 struct ltt_kernel_event
*event
;
200 /* We pass ownership of filter_expression and filter */
201 event
= trace_kernel_create_event(ev
, filter_expression
,
208 ret
= kernctl_create_event(channel
->fd
, event
->event
);
214 WARN("Event type not implemented");
217 WARN("Event %s not found!", ev
->name
);
220 PERROR("create event ioctl");
226 event
->type
= ev
->type
;
228 /* Prevent fd duplication after execlp() */
229 ret
= fcntl(event
->fd
, F_SETFD
, FD_CLOEXEC
);
231 PERROR("fcntl session fd");
235 ret
= kernctl_filter(event
->fd
, filter
);
241 ret
= kernctl_enable(event
->fd
);
245 ret
= LTTNG_ERR_KERN_EVENT_EXIST
;
248 PERROR("enable kernel event");
254 /* Add event to event list */
255 cds_list_add(&event
->list
, &channel
->events_list
.head
);
256 channel
->event_count
++;
258 DBG("Event %s created (fd: %d)", ev
->name
, event
->fd
);
267 closeret
= close(event
->fd
);
269 PERROR("close event fd");
279 * Disable a kernel channel.
281 int kernel_disable_channel(struct ltt_kernel_channel
*chan
)
287 ret
= kernctl_disable(chan
->fd
);
289 PERROR("disable chan ioctl");
295 DBG("Kernel channel %s disabled (fd: %d)", chan
->channel
->name
, chan
->fd
);
304 * Enable a kernel channel.
306 int kernel_enable_channel(struct ltt_kernel_channel
*chan
)
312 ret
= kernctl_enable(chan
->fd
);
313 if (ret
< 0 && errno
!= EEXIST
) {
314 PERROR("Enable kernel chan");
319 DBG("Kernel channel %s enabled (fd: %d)", chan
->channel
->name
, chan
->fd
);
328 * Enable a kernel event.
330 int kernel_enable_event(struct ltt_kernel_event
*event
)
336 ret
= kernctl_enable(event
->fd
);
340 ret
= LTTNG_ERR_KERN_EVENT_EXIST
;
343 PERROR("enable kernel event");
350 DBG("Kernel event %s enabled (fd: %d)", event
->event
->name
, event
->fd
);
359 * Disable a kernel event.
361 int kernel_disable_event(struct ltt_kernel_event
*event
)
367 ret
= kernctl_disable(event
->fd
);
371 ret
= LTTNG_ERR_KERN_EVENT_EXIST
;
374 PERROR("disable kernel event");
381 DBG("Kernel event %s disabled (fd: %d)", event
->event
->name
, event
->fd
);
390 int kernel_track_pid(struct ltt_kernel_session
*session
, int pid
)
394 DBG("Kernel track PID %d for session id %" PRIu64
".",
396 ret
= kernctl_track_pid(session
->fd
, pid
);
402 return LTTNG_ERR_INVALID
;
404 return LTTNG_ERR_NOMEM
;
406 return LTTNG_ERR_PID_TRACKED
;
408 return LTTNG_ERR_UNK
;
412 int kernel_untrack_pid(struct ltt_kernel_session
*session
, int pid
)
416 DBG("Kernel untrack PID %d for session id %" PRIu64
".",
418 ret
= kernctl_untrack_pid(session
->fd
, pid
);
424 return LTTNG_ERR_INVALID
;
426 return LTTNG_ERR_NOMEM
;
428 return LTTNG_ERR_PID_NOT_TRACKED
;
430 return LTTNG_ERR_UNK
;
434 ssize_t
kernel_list_tracker_pids(struct ltt_kernel_session
*session
,
439 ssize_t nbmem
, count
= 0;
443 fd
= kernctl_list_tracker_pids(session
->fd
);
445 PERROR("kernel tracker pids list");
449 fp
= fdopen(fd
, "r");
451 PERROR("kernel tracker pids list fdopen");
455 nbmem
= KERNEL_TRACKER_PIDS_INIT_LIST_SIZE
;
456 pids
= zmalloc(sizeof(*pids
) * nbmem
);
458 PERROR("alloc list pids");
463 while (fscanf(fp
, "process { pid = %u; };\n", &pid
) == 1) {
464 if (count
>= nbmem
) {
468 new_nbmem
= nbmem
<< 1;
469 DBG("Reallocating pids list from %zu to %zu entries",
471 new_pids
= realloc(pids
, new_nbmem
* sizeof(*new_pids
));
472 if (new_pids
== NULL
) {
473 PERROR("realloc list events");
478 /* Zero the new memory */
479 memset(new_pids
+ nbmem
, 0,
480 (new_nbmem
- nbmem
) * sizeof(*new_pids
));
488 DBG("Kernel list tracker pids done (%zd pids)", count
);
490 ret
= fclose(fp
); /* closes both fp and fd */
506 * Create kernel metadata, open from the kernel tracer and add it to the
509 int kernel_open_metadata(struct ltt_kernel_session
*session
)
512 struct ltt_kernel_metadata
*lkm
= NULL
;
516 /* Allocate kernel metadata */
517 lkm
= trace_kernel_create_metadata();
522 /* Kernel tracer metadata creation */
523 ret
= kernctl_open_metadata(session
->fd
, &lkm
->conf
->attr
);
529 /* Prevent fd duplication after execlp() */
530 ret
= fcntl(lkm
->fd
, F_SETFD
, FD_CLOEXEC
);
532 PERROR("fcntl session fd");
535 session
->metadata
= lkm
;
537 DBG("Kernel metadata opened (fd: %d)", lkm
->fd
);
542 trace_kernel_destroy_metadata(lkm
);
548 * Start tracing session.
550 int kernel_start_session(struct ltt_kernel_session
*session
)
556 ret
= kernctl_start_session(session
->fd
);
558 PERROR("ioctl start session");
562 DBG("Kernel session started");
571 * Make a kernel wait to make sure in-flight probe have completed.
573 void kernel_wait_quiescent(int fd
)
577 DBG("Kernel quiescent wait on %d", fd
);
579 ret
= kernctl_wait_quiescent(fd
);
581 PERROR("wait quiescent ioctl");
582 ERR("Kernel quiescent wait failed");
589 int kernel_calibrate(int fd
, struct lttng_kernel_calibrate
*calibrate
)
595 ret
= kernctl_calibrate(fd
, calibrate
);
597 PERROR("calibrate ioctl");
606 * Force flush buffer of metadata.
608 int kernel_metadata_flush_buffer(int fd
)
612 DBG("Kernel flushing metadata buffer on fd %d", fd
);
614 ret
= kernctl_buffer_flush(fd
);
616 ERR("Fail to flush metadata buffers %d (ret: %d)", fd
, ret
);
623 * Force flush buffer for channel.
625 int kernel_flush_buffer(struct ltt_kernel_channel
*channel
)
628 struct ltt_kernel_stream
*stream
;
632 DBG("Flush buffer for channel %s", channel
->channel
->name
);
634 cds_list_for_each_entry(stream
, &channel
->stream_list
.head
, list
) {
635 DBG("Flushing channel stream %d", stream
->fd
);
636 ret
= kernctl_buffer_flush(stream
->fd
);
639 ERR("Fail to flush buffer for stream %d (ret: %d)",
648 * Stop tracing session.
650 int kernel_stop_session(struct ltt_kernel_session
*session
)
656 ret
= kernctl_stop_session(session
->fd
);
661 DBG("Kernel session stopped");
670 * Open stream of channel, register it to the kernel tracer and add it
671 * to the stream list of the channel.
673 * Note: given that the streams may appear in random order wrt CPU
674 * number (e.g. cpu hotplug), the index value of the stream number in
675 * the stream name is not necessarily linked to the CPU number.
677 * Return the number of created stream. Else, a negative value.
679 int kernel_open_channel_stream(struct ltt_kernel_channel
*channel
)
682 struct ltt_kernel_stream
*lks
;
686 while ((ret
= kernctl_create_stream(channel
->fd
)) >= 0) {
687 lks
= trace_kernel_create_stream(channel
->channel
->name
,
688 channel
->stream_count
);
698 /* Prevent fd duplication after execlp() */
699 ret
= fcntl(lks
->fd
, F_SETFD
, FD_CLOEXEC
);
701 PERROR("fcntl session fd");
704 lks
->tracefile_size
= channel
->channel
->attr
.tracefile_size
;
705 lks
->tracefile_count
= channel
->channel
->attr
.tracefile_count
;
707 /* Add stream to channel stream list */
708 cds_list_add(&lks
->list
, &channel
->stream_list
.head
);
709 channel
->stream_count
++;
711 DBG("Kernel stream %s created (fd: %d, state: %d)", lks
->name
, lks
->fd
,
715 return channel
->stream_count
;
722 * Open the metadata stream and set it to the kernel session.
724 int kernel_open_metadata_stream(struct ltt_kernel_session
*session
)
730 ret
= kernctl_create_stream(session
->metadata
->fd
);
732 PERROR("kernel create metadata stream");
736 DBG("Kernel metadata stream created (fd: %d)", ret
);
737 session
->metadata_stream_fd
= ret
;
738 /* Prevent fd duplication after execlp() */
739 ret
= fcntl(session
->metadata_stream_fd
, F_SETFD
, FD_CLOEXEC
);
741 PERROR("fcntl session fd");
751 * Get the event list from the kernel tracer and return the number of elements.
753 ssize_t
kernel_list_events(int tracer_fd
, struct lttng_event
**events
)
757 size_t nbmem
, count
= 0;
759 struct lttng_event
*elist
;
763 fd
= kernctl_tracepoint_list(tracer_fd
);
765 PERROR("kernel tracepoint list");
769 fp
= fdopen(fd
, "r");
771 PERROR("kernel tracepoint list fdopen");
776 * Init memory size counter
777 * See kernel-ctl.h for explanation of this value
779 nbmem
= KERNEL_EVENT_INIT_LIST_SIZE
;
780 elist
= zmalloc(sizeof(struct lttng_event
) * nbmem
);
782 PERROR("alloc list events");
787 while (fscanf(fp
, "event { name = %m[^;]; };\n", &event
) == 1) {
788 if (count
>= nbmem
) {
789 struct lttng_event
*new_elist
;
792 new_nbmem
= nbmem
<< 1;
793 DBG("Reallocating event list from %zu to %zu bytes",
795 new_elist
= realloc(elist
, new_nbmem
* sizeof(struct lttng_event
));
796 if (new_elist
== NULL
) {
797 PERROR("realloc list events");
803 /* Zero the new memory */
804 memset(new_elist
+ nbmem
, 0,
805 (new_nbmem
- nbmem
) * sizeof(struct lttng_event
));
809 strncpy(elist
[count
].name
, event
, LTTNG_SYMBOL_NAME_LEN
);
810 elist
[count
].name
[LTTNG_SYMBOL_NAME_LEN
- 1] = '\0';
811 elist
[count
].enabled
= -1;
817 DBG("Kernel list events done (%zu events)", count
);
819 ret
= fclose(fp
); /* closes both fp and fd */
835 * Get kernel version and validate it.
837 int kernel_validate_version(int tracer_fd
)
840 struct lttng_kernel_tracer_version version
;
841 struct lttng_kernel_tracer_abi_version abi_version
;
843 ret
= kernctl_tracer_version(tracer_fd
, &version
);
845 ERR("Failed to retrieve the lttng-modules version");
849 /* Validate version */
850 if (version
.major
!= VERSION_MAJOR
) {
851 ERR("Kernel tracer major version (%d) is not compatible with lttng-tools major version (%d)",
852 version
.major
, VERSION_MAJOR
);
855 ret
= kernctl_tracer_abi_version(tracer_fd
, &abi_version
);
857 ERR("Failed to retrieve lttng-modules ABI version");
860 if (abi_version
.major
!= LTTNG_MODULES_ABI_MAJOR_VERSION
) {
861 ERR("Kernel tracer ABI version (%d.%d) does not match the expected ABI major version (%d.*)",
862 abi_version
.major
, abi_version
.minor
,
863 LTTNG_MODULES_ABI_MAJOR_VERSION
);
866 DBG2("Kernel tracer version validated (%d.%d, ABI %d.%d)",
867 version
.major
, version
.minor
,
868 abi_version
.major
, abi_version
.minor
);
875 ERR("Kernel tracer version check failed; kernel tracing will not be available");
880 * Kernel work-arounds called at the start of sessiond main().
882 int init_kernel_workarounds(void)
888 * boot_id needs to be read once before being used concurrently
889 * to deal with a Linux kernel race. A fix is proposed for
890 * upstream, but the work-around is needed for older kernels.
892 fp
= fopen("/proc/sys/kernel/random/boot_id", "r");
899 ret
= fread(buf
, 1, sizeof(buf
), fp
);
901 /* Ignore error, we don't really care */
913 * Complete teardown of a kernel session.
915 void kernel_destroy_session(struct ltt_kernel_session
*ksess
)
918 DBG3("No kernel session when tearing down session");
922 DBG("Tearing down kernel session");
925 * Destroy channels on the consumer if at least one FD has been sent and we
926 * are in no output mode because the streams are in *no* monitor mode so we
927 * have to send a command to clean them up or else they leaked.
929 if (!ksess
->output_traces
&& ksess
->consumer_fds_sent
) {
931 struct consumer_socket
*socket
;
932 struct lttng_ht_iter iter
;
934 /* For each consumer socket. */
936 cds_lfht_for_each_entry(ksess
->consumer
->socks
->ht
, &iter
.iter
,
938 struct ltt_kernel_channel
*chan
;
940 /* For each channel, ask the consumer to destroy it. */
941 cds_list_for_each_entry(chan
, &ksess
->channel_list
.head
, list
) {
942 ret
= kernel_consumer_destroy_channel(socket
, chan
);
944 /* Consumer is probably dead. Use next socket. */
952 /* Close any relayd session */
953 consumer_output_send_destroy_relayd(ksess
->consumer
);
955 trace_kernel_destroy_session(ksess
);
959 * Destroy a kernel channel object. It does not do anything on the tracer side.
961 void kernel_destroy_channel(struct ltt_kernel_channel
*kchan
)
963 struct ltt_kernel_session
*ksess
= NULL
;
966 assert(kchan
->channel
);
968 DBG3("Kernel destroy channel %s", kchan
->channel
->name
);
970 /* Update channel count of associated session. */
971 if (kchan
->session
) {
972 /* Keep pointer reference so we can update it after the destroy. */
973 ksess
= kchan
->session
;
976 trace_kernel_destroy_channel(kchan
);
979 * At this point the kernel channel is not visible anymore. This is safe
980 * since in order to work on a visible kernel session, the tracing session
981 * lock (ltt_session.lock) MUST be acquired.
984 ksess
->channel_count
--;
989 * Take a snapshot for a given kernel session.
991 * Return 0 on success or else return a LTTNG_ERR code.
993 int kernel_snapshot_record(struct ltt_kernel_session
*ksess
,
994 struct snapshot_output
*output
, int wait
,
995 uint64_t nb_packets_per_stream
)
997 int err
, ret
, saved_metadata_fd
;
998 struct consumer_socket
*socket
;
999 struct lttng_ht_iter iter
;
1000 struct ltt_kernel_metadata
*saved_metadata
;
1003 assert(ksess
->consumer
);
1006 DBG("Kernel snapshot record started");
1008 /* Save current metadata since the following calls will change it. */
1009 saved_metadata
= ksess
->metadata
;
1010 saved_metadata_fd
= ksess
->metadata_stream_fd
;
1014 ret
= kernel_open_metadata(ksess
);
1016 ret
= LTTNG_ERR_KERN_META_FAIL
;
1020 ret
= kernel_open_metadata_stream(ksess
);
1022 ret
= LTTNG_ERR_KERN_META_FAIL
;
1023 goto error_open_stream
;
1026 /* Send metadata to consumer and snapshot everything. */
1027 cds_lfht_for_each_entry(ksess
->consumer
->socks
->ht
, &iter
.iter
,
1028 socket
, node
.node
) {
1029 struct consumer_output
*saved_output
;
1030 struct ltt_kernel_channel
*chan
;
1033 * Temporarly switch consumer output for our snapshot output. As long
1034 * as the session lock is taken, this is safe.
1036 saved_output
= ksess
->consumer
;
1037 ksess
->consumer
= output
->consumer
;
1039 pthread_mutex_lock(socket
->lock
);
1040 /* This stream must not be monitored by the consumer. */
1041 ret
= kernel_consumer_add_metadata(socket
, ksess
, 0);
1042 pthread_mutex_unlock(socket
->lock
);
1043 /* Put back the saved consumer output into the session. */
1044 ksess
->consumer
= saved_output
;
1046 ret
= LTTNG_ERR_KERN_CONSUMER_FAIL
;
1047 goto error_consumer
;
1050 /* For each channel, ask the consumer to snapshot it. */
1051 cds_list_for_each_entry(chan
, &ksess
->channel_list
.head
, list
) {
1052 pthread_mutex_lock(socket
->lock
);
1053 ret
= consumer_snapshot_channel(socket
, chan
->fd
, output
, 0,
1054 ksess
->uid
, ksess
->gid
,
1055 DEFAULT_KERNEL_TRACE_DIR
, wait
,
1056 nb_packets_per_stream
);
1057 pthread_mutex_unlock(socket
->lock
);
1059 ret
= LTTNG_ERR_KERN_CONSUMER_FAIL
;
1060 (void) kernel_consumer_destroy_metadata(socket
,
1062 goto error_consumer
;
1066 /* Snapshot metadata, */
1067 pthread_mutex_lock(socket
->lock
);
1068 ret
= consumer_snapshot_channel(socket
, ksess
->metadata
->fd
, output
,
1069 1, ksess
->uid
, ksess
->gid
,
1070 DEFAULT_KERNEL_TRACE_DIR
, wait
, 0);
1071 pthread_mutex_unlock(socket
->lock
);
1073 ret
= LTTNG_ERR_KERN_CONSUMER_FAIL
;
1074 goto error_consumer
;
1078 * The metadata snapshot is done, ask the consumer to destroy it since
1079 * it's not monitored on the consumer side.
1081 (void) kernel_consumer_destroy_metadata(socket
, ksess
->metadata
);
1087 /* Close newly opened metadata stream. It's now on the consumer side. */
1088 err
= close(ksess
->metadata_stream_fd
);
1090 PERROR("close snapshot kernel");
1094 trace_kernel_destroy_metadata(ksess
->metadata
);
1096 /* Restore metadata state.*/
1097 ksess
->metadata
= saved_metadata
;
1098 ksess
->metadata_stream_fd
= saved_metadata_fd
;
1105 * Get the syscall mask array from the kernel tracer.
1107 * Return 0 on success else a negative value. In both case, syscall_mask should
1110 int kernel_syscall_mask(int chan_fd
, char **syscall_mask
, uint32_t *nr_bits
)
1112 assert(syscall_mask
);
1115 return kernctl_syscall_mask(chan_fd
, syscall_mask
, nr_bits
);