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
);
76 trace_kernel_destroy_context(ctx
);
82 * Create a new kernel session, register it to the kernel tracer and add it to
83 * the session daemon session.
85 int kernel_create_session(struct ltt_session
*session
, int tracer_fd
)
88 struct ltt_kernel_session
*lks
;
92 /* Allocate data structure */
93 lks
= trace_kernel_create_session();
99 /* Kernel tracer session creation */
100 ret
= kernctl_create_session(tracer_fd
);
102 PERROR("ioctl kernel create session");
107 /* Prevent fd duplication after execlp() */
108 ret
= fcntl(lks
->fd
, F_SETFD
, FD_CLOEXEC
);
110 PERROR("fcntl session fd");
113 lks
->id
= session
->id
;
114 lks
->consumer_fds_sent
= 0;
115 session
->kernel_session
= lks
;
117 DBG("Kernel session created (fd: %d)", lks
->fd
);
123 trace_kernel_destroy_session(lks
);
129 * Create a kernel channel, register it to the kernel tracer and add it to the
132 int kernel_create_channel(struct ltt_kernel_session
*session
,
133 struct lttng_channel
*chan
)
136 struct ltt_kernel_channel
*lkc
;
141 /* Allocate kernel channel */
142 lkc
= trace_kernel_create_channel(chan
);
147 DBG3("Kernel create channel %s with attr: %d, %" PRIu64
", %" PRIu64
", %u, %u, %d, %d",
148 chan
->name
, lkc
->channel
->attr
.overwrite
,
149 lkc
->channel
->attr
.subbuf_size
, lkc
->channel
->attr
.num_subbuf
,
150 lkc
->channel
->attr
.switch_timer_interval
, lkc
->channel
->attr
.read_timer_interval
,
151 lkc
->channel
->attr
.live_timer_interval
, lkc
->channel
->attr
.output
);
153 /* Kernel tracer channel creation */
154 ret
= kernctl_create_channel(session
->fd
, &lkc
->channel
->attr
);
156 PERROR("ioctl kernel create channel");
160 /* Setup the channel fd */
162 /* Prevent fd duplication after execlp() */
163 ret
= fcntl(lkc
->fd
, F_SETFD
, FD_CLOEXEC
);
165 PERROR("fcntl session fd");
168 /* Add channel to session */
169 cds_list_add(&lkc
->list
, &session
->channel_list
.head
);
170 session
->channel_count
++;
171 lkc
->session
= session
;
173 DBG("Kernel channel %s created (fd: %d)", lkc
->channel
->name
, lkc
->fd
);
186 * Create a kernel event, enable it to the kernel tracer and add it to the
187 * channel event list of the kernel session.
188 * We own filter_expression and filter.
190 int kernel_create_event(struct lttng_event
*ev
,
191 struct ltt_kernel_channel
*channel
,
192 char *filter_expression
,
193 struct lttng_filter_bytecode
*filter
)
196 struct ltt_kernel_event
*event
;
201 /* We pass ownership of filter_expression and filter */
202 event
= trace_kernel_create_event(ev
, filter_expression
,
209 ret
= kernctl_create_event(channel
->fd
, event
->event
);
215 WARN("Event type not implemented");
218 WARN("Event %s not found!", ev
->name
);
221 PERROR("create event ioctl");
227 event
->type
= ev
->type
;
229 /* Prevent fd duplication after execlp() */
230 ret
= fcntl(event
->fd
, F_SETFD
, FD_CLOEXEC
);
232 PERROR("fcntl session fd");
236 ret
= kernctl_filter(event
->fd
, filter
);
242 ret
= kernctl_enable(event
->fd
);
246 ret
= LTTNG_ERR_KERN_EVENT_EXIST
;
249 PERROR("enable kernel event");
255 /* Add event to event list */
256 cds_list_add(&event
->list
, &channel
->events_list
.head
);
257 channel
->event_count
++;
259 DBG("Event %s created (fd: %d)", ev
->name
, event
->fd
);
268 closeret
= close(event
->fd
);
270 PERROR("close event fd");
280 * Disable a kernel channel.
282 int kernel_disable_channel(struct ltt_kernel_channel
*chan
)
288 ret
= kernctl_disable(chan
->fd
);
290 PERROR("disable chan ioctl");
296 DBG("Kernel channel %s disabled (fd: %d)", chan
->channel
->name
, chan
->fd
);
305 * Enable a kernel channel.
307 int kernel_enable_channel(struct ltt_kernel_channel
*chan
)
313 ret
= kernctl_enable(chan
->fd
);
314 if (ret
< 0 && errno
!= EEXIST
) {
315 PERROR("Enable kernel chan");
320 DBG("Kernel channel %s enabled (fd: %d)", chan
->channel
->name
, chan
->fd
);
329 * Enable a kernel event.
331 int kernel_enable_event(struct ltt_kernel_event
*event
)
337 ret
= kernctl_enable(event
->fd
);
341 ret
= LTTNG_ERR_KERN_EVENT_EXIST
;
344 PERROR("enable kernel event");
351 DBG("Kernel event %s enabled (fd: %d)", event
->event
->name
, event
->fd
);
360 * Disable a kernel event.
362 int kernel_disable_event(struct ltt_kernel_event
*event
)
368 ret
= kernctl_disable(event
->fd
);
372 ret
= LTTNG_ERR_KERN_EVENT_EXIST
;
375 PERROR("disable kernel event");
382 DBG("Kernel event %s disabled (fd: %d)", event
->event
->name
, event
->fd
);
391 int kernel_track_pid(struct ltt_kernel_session
*session
, int pid
)
395 DBG("Kernel track PID %d for session id %" PRIu64
".",
397 ret
= kernctl_track_pid(session
->fd
, pid
);
403 return LTTNG_ERR_INVALID
;
405 return LTTNG_ERR_NOMEM
;
407 return LTTNG_ERR_PID_TRACKED
;
409 return LTTNG_ERR_UNK
;
413 int kernel_untrack_pid(struct ltt_kernel_session
*session
, int pid
)
417 DBG("Kernel untrack PID %d for session id %" PRIu64
".",
419 ret
= kernctl_untrack_pid(session
->fd
, pid
);
425 return LTTNG_ERR_INVALID
;
427 return LTTNG_ERR_NOMEM
;
429 return LTTNG_ERR_PID_NOT_TRACKED
;
431 return LTTNG_ERR_UNK
;
435 ssize_t
kernel_list_tracker_pids(struct ltt_kernel_session
*session
,
440 ssize_t nbmem
, count
= 0;
444 fd
= kernctl_list_tracker_pids(session
->fd
);
446 PERROR("kernel tracker pids list");
450 fp
= fdopen(fd
, "r");
452 PERROR("kernel tracker pids list fdopen");
456 nbmem
= KERNEL_TRACKER_PIDS_INIT_LIST_SIZE
;
457 pids
= zmalloc(sizeof(*pids
) * nbmem
);
459 PERROR("alloc list pids");
464 while (fscanf(fp
, "process { pid = %u; };\n", &pid
) == 1) {
465 if (count
>= nbmem
) {
469 new_nbmem
= nbmem
<< 1;
470 DBG("Reallocating pids list from %zu to %zu entries",
472 new_pids
= realloc(pids
, new_nbmem
* sizeof(*new_pids
));
473 if (new_pids
== NULL
) {
474 PERROR("realloc list events");
479 /* Zero the new memory */
480 memset(new_pids
+ nbmem
, 0,
481 (new_nbmem
- nbmem
) * sizeof(*new_pids
));
489 DBG("Kernel list tracker pids done (%zd pids)", count
);
491 ret
= fclose(fp
); /* closes both fp and fd */
507 * Create kernel metadata, open from the kernel tracer and add it to the
510 int kernel_open_metadata(struct ltt_kernel_session
*session
)
513 struct ltt_kernel_metadata
*lkm
= NULL
;
517 /* Allocate kernel metadata */
518 lkm
= trace_kernel_create_metadata();
523 /* Kernel tracer metadata creation */
524 ret
= kernctl_open_metadata(session
->fd
, &lkm
->conf
->attr
);
530 /* Prevent fd duplication after execlp() */
531 ret
= fcntl(lkm
->fd
, F_SETFD
, FD_CLOEXEC
);
533 PERROR("fcntl session fd");
536 session
->metadata
= lkm
;
538 DBG("Kernel metadata opened (fd: %d)", lkm
->fd
);
543 trace_kernel_destroy_metadata(lkm
);
549 * Start tracing session.
551 int kernel_start_session(struct ltt_kernel_session
*session
)
557 ret
= kernctl_start_session(session
->fd
);
559 PERROR("ioctl start session");
563 DBG("Kernel session started");
572 * Make a kernel wait to make sure in-flight probe have completed.
574 void kernel_wait_quiescent(int fd
)
578 DBG("Kernel quiescent wait on %d", fd
);
580 ret
= kernctl_wait_quiescent(fd
);
582 PERROR("wait quiescent ioctl");
583 ERR("Kernel quiescent wait failed");
590 int kernel_calibrate(int fd
, struct lttng_kernel_calibrate
*calibrate
)
596 ret
= kernctl_calibrate(fd
, calibrate
);
598 PERROR("calibrate ioctl");
607 * Force flush buffer of metadata.
609 int kernel_metadata_flush_buffer(int fd
)
613 DBG("Kernel flushing metadata buffer on fd %d", fd
);
615 ret
= kernctl_buffer_flush(fd
);
617 ERR("Fail to flush metadata buffers %d (ret: %d)", fd
, ret
);
624 * Force flush buffer for channel.
626 int kernel_flush_buffer(struct ltt_kernel_channel
*channel
)
629 struct ltt_kernel_stream
*stream
;
633 DBG("Flush buffer for channel %s", channel
->channel
->name
);
635 cds_list_for_each_entry(stream
, &channel
->stream_list
.head
, list
) {
636 DBG("Flushing channel stream %d", stream
->fd
);
637 ret
= kernctl_buffer_flush(stream
->fd
);
640 ERR("Fail to flush buffer for stream %d (ret: %d)",
649 * Stop tracing session.
651 int kernel_stop_session(struct ltt_kernel_session
*session
)
657 ret
= kernctl_stop_session(session
->fd
);
662 DBG("Kernel session stopped");
671 * Open stream of channel, register it to the kernel tracer and add it
672 * to the stream list of the channel.
674 * Note: given that the streams may appear in random order wrt CPU
675 * number (e.g. cpu hotplug), the index value of the stream number in
676 * the stream name is not necessarily linked to the CPU number.
678 * Return the number of created stream. Else, a negative value.
680 int kernel_open_channel_stream(struct ltt_kernel_channel
*channel
)
683 struct ltt_kernel_stream
*lks
;
687 while ((ret
= kernctl_create_stream(channel
->fd
)) >= 0) {
688 lks
= trace_kernel_create_stream(channel
->channel
->name
,
689 channel
->stream_count
);
699 /* Prevent fd duplication after execlp() */
700 ret
= fcntl(lks
->fd
, F_SETFD
, FD_CLOEXEC
);
702 PERROR("fcntl session fd");
705 lks
->tracefile_size
= channel
->channel
->attr
.tracefile_size
;
706 lks
->tracefile_count
= channel
->channel
->attr
.tracefile_count
;
708 /* Add stream to channel stream list */
709 cds_list_add(&lks
->list
, &channel
->stream_list
.head
);
710 channel
->stream_count
++;
712 DBG("Kernel stream %s created (fd: %d, state: %d)", lks
->name
, lks
->fd
,
716 return channel
->stream_count
;
723 * Open the metadata stream and set it to the kernel session.
725 int kernel_open_metadata_stream(struct ltt_kernel_session
*session
)
731 ret
= kernctl_create_stream(session
->metadata
->fd
);
733 PERROR("kernel create metadata stream");
737 DBG("Kernel metadata stream created (fd: %d)", ret
);
738 session
->metadata_stream_fd
= ret
;
739 /* Prevent fd duplication after execlp() */
740 ret
= fcntl(session
->metadata_stream_fd
, F_SETFD
, FD_CLOEXEC
);
742 PERROR("fcntl session fd");
752 * Get the event list from the kernel tracer and return the number of elements.
754 ssize_t
kernel_list_events(int tracer_fd
, struct lttng_event
**events
)
758 size_t nbmem
, count
= 0;
760 struct lttng_event
*elist
;
764 fd
= kernctl_tracepoint_list(tracer_fd
);
766 PERROR("kernel tracepoint list");
770 fp
= fdopen(fd
, "r");
772 PERROR("kernel tracepoint list fdopen");
777 * Init memory size counter
778 * See kernel-ctl.h for explanation of this value
780 nbmem
= KERNEL_EVENT_INIT_LIST_SIZE
;
781 elist
= zmalloc(sizeof(struct lttng_event
) * nbmem
);
783 PERROR("alloc list events");
788 while (fscanf(fp
, "event { name = %m[^;]; };\n", &event
) == 1) {
789 if (count
>= nbmem
) {
790 struct lttng_event
*new_elist
;
793 new_nbmem
= nbmem
<< 1;
794 DBG("Reallocating event list from %zu to %zu bytes",
796 new_elist
= realloc(elist
, new_nbmem
* sizeof(struct lttng_event
));
797 if (new_elist
== NULL
) {
798 PERROR("realloc list events");
804 /* Zero the new memory */
805 memset(new_elist
+ nbmem
, 0,
806 (new_nbmem
- nbmem
) * sizeof(struct lttng_event
));
810 strncpy(elist
[count
].name
, event
, LTTNG_SYMBOL_NAME_LEN
);
811 elist
[count
].name
[LTTNG_SYMBOL_NAME_LEN
- 1] = '\0';
812 elist
[count
].enabled
= -1;
818 DBG("Kernel list events done (%zu events)", count
);
820 ret
= fclose(fp
); /* closes both fp and fd */
836 * Get kernel version and validate it.
838 int kernel_validate_version(int tracer_fd
)
841 struct lttng_kernel_tracer_version version
;
842 struct lttng_kernel_tracer_abi_version abi_version
;
844 ret
= kernctl_tracer_version(tracer_fd
, &version
);
846 ERR("Failed to retrieve the lttng-modules version");
850 /* Validate version */
851 if (version
.major
!= VERSION_MAJOR
) {
852 ERR("Kernel tracer major version (%d) is not compatible with lttng-tools major version (%d)",
853 version
.major
, VERSION_MAJOR
);
856 ret
= kernctl_tracer_abi_version(tracer_fd
, &abi_version
);
858 ERR("Failed to retrieve lttng-modules ABI version");
861 if (abi_version
.major
!= LTTNG_MODULES_ABI_MAJOR_VERSION
) {
862 ERR("Kernel tracer ABI version (%d.%d) does not match the expected ABI major version (%d.*)",
863 abi_version
.major
, abi_version
.minor
,
864 LTTNG_MODULES_ABI_MAJOR_VERSION
);
867 DBG2("Kernel tracer version validated (%d.%d, ABI %d.%d)",
868 version
.major
, version
.minor
,
869 abi_version
.major
, abi_version
.minor
);
876 ERR("Kernel tracer version check failed; kernel tracing will not be available");
881 * Kernel work-arounds called at the start of sessiond main().
883 int init_kernel_workarounds(void)
889 * boot_id needs to be read once before being used concurrently
890 * to deal with a Linux kernel race. A fix is proposed for
891 * upstream, but the work-around is needed for older kernels.
893 fp
= fopen("/proc/sys/kernel/random/boot_id", "r");
900 ret
= fread(buf
, 1, sizeof(buf
), fp
);
902 /* Ignore error, we don't really care */
914 * Complete teardown of a kernel session.
916 void kernel_destroy_session(struct ltt_kernel_session
*ksess
)
919 DBG3("No kernel session when tearing down session");
923 DBG("Tearing down kernel session");
926 * Destroy channels on the consumer if at least one FD has been sent and we
927 * are in no output mode because the streams are in *no* monitor mode so we
928 * have to send a command to clean them up or else they leaked.
930 if (!ksess
->output_traces
&& ksess
->consumer_fds_sent
) {
932 struct consumer_socket
*socket
;
933 struct lttng_ht_iter iter
;
935 /* For each consumer socket. */
937 cds_lfht_for_each_entry(ksess
->consumer
->socks
->ht
, &iter
.iter
,
939 struct ltt_kernel_channel
*chan
;
941 /* For each channel, ask the consumer to destroy it. */
942 cds_list_for_each_entry(chan
, &ksess
->channel_list
.head
, list
) {
943 ret
= kernel_consumer_destroy_channel(socket
, chan
);
945 /* Consumer is probably dead. Use next socket. */
953 /* Close any relayd session */
954 consumer_output_send_destroy_relayd(ksess
->consumer
);
956 trace_kernel_destroy_session(ksess
);
960 * Destroy a kernel channel object. It does not do anything on the tracer side.
962 void kernel_destroy_channel(struct ltt_kernel_channel
*kchan
)
964 struct ltt_kernel_session
*ksess
= NULL
;
967 assert(kchan
->channel
);
969 DBG3("Kernel destroy channel %s", kchan
->channel
->name
);
971 /* Update channel count of associated session. */
972 if (kchan
->session
) {
973 /* Keep pointer reference so we can update it after the destroy. */
974 ksess
= kchan
->session
;
977 trace_kernel_destroy_channel(kchan
);
980 * At this point the kernel channel is not visible anymore. This is safe
981 * since in order to work on a visible kernel session, the tracing session
982 * lock (ltt_session.lock) MUST be acquired.
985 ksess
->channel_count
--;
990 * Take a snapshot for a given kernel session.
992 * Return 0 on success or else return a LTTNG_ERR code.
994 int kernel_snapshot_record(struct ltt_kernel_session
*ksess
,
995 struct snapshot_output
*output
, int wait
,
996 uint64_t nb_packets_per_stream
)
998 int err
, ret
, saved_metadata_fd
;
999 struct consumer_socket
*socket
;
1000 struct lttng_ht_iter iter
;
1001 struct ltt_kernel_metadata
*saved_metadata
;
1004 assert(ksess
->consumer
);
1007 DBG("Kernel snapshot record started");
1009 /* Save current metadata since the following calls will change it. */
1010 saved_metadata
= ksess
->metadata
;
1011 saved_metadata_fd
= ksess
->metadata_stream_fd
;
1015 ret
= kernel_open_metadata(ksess
);
1017 ret
= LTTNG_ERR_KERN_META_FAIL
;
1021 ret
= kernel_open_metadata_stream(ksess
);
1023 ret
= LTTNG_ERR_KERN_META_FAIL
;
1024 goto error_open_stream
;
1027 /* Send metadata to consumer and snapshot everything. */
1028 cds_lfht_for_each_entry(ksess
->consumer
->socks
->ht
, &iter
.iter
,
1029 socket
, node
.node
) {
1030 struct consumer_output
*saved_output
;
1031 struct ltt_kernel_channel
*chan
;
1034 * Temporarly switch consumer output for our snapshot output. As long
1035 * as the session lock is taken, this is safe.
1037 saved_output
= ksess
->consumer
;
1038 ksess
->consumer
= output
->consumer
;
1040 pthread_mutex_lock(socket
->lock
);
1041 /* This stream must not be monitored by the consumer. */
1042 ret
= kernel_consumer_add_metadata(socket
, ksess
, 0);
1043 pthread_mutex_unlock(socket
->lock
);
1044 /* Put back the saved consumer output into the session. */
1045 ksess
->consumer
= saved_output
;
1047 ret
= LTTNG_ERR_KERN_CONSUMER_FAIL
;
1048 goto error_consumer
;
1051 /* For each channel, ask the consumer to snapshot it. */
1052 cds_list_for_each_entry(chan
, &ksess
->channel_list
.head
, list
) {
1053 pthread_mutex_lock(socket
->lock
);
1054 ret
= consumer_snapshot_channel(socket
, chan
->fd
, output
, 0,
1055 ksess
->uid
, ksess
->gid
,
1056 DEFAULT_KERNEL_TRACE_DIR
, wait
,
1057 nb_packets_per_stream
);
1058 pthread_mutex_unlock(socket
->lock
);
1060 ret
= LTTNG_ERR_KERN_CONSUMER_FAIL
;
1061 (void) kernel_consumer_destroy_metadata(socket
,
1063 goto error_consumer
;
1067 /* Snapshot metadata, */
1068 pthread_mutex_lock(socket
->lock
);
1069 ret
= consumer_snapshot_channel(socket
, ksess
->metadata
->fd
, output
,
1070 1, ksess
->uid
, ksess
->gid
,
1071 DEFAULT_KERNEL_TRACE_DIR
, wait
, 0);
1072 pthread_mutex_unlock(socket
->lock
);
1074 ret
= LTTNG_ERR_KERN_CONSUMER_FAIL
;
1075 goto error_consumer
;
1079 * The metadata snapshot is done, ask the consumer to destroy it since
1080 * it's not monitored on the consumer side.
1082 (void) kernel_consumer_destroy_metadata(socket
, ksess
->metadata
);
1088 /* Close newly opened metadata stream. It's now on the consumer side. */
1089 err
= close(ksess
->metadata_stream_fd
);
1091 PERROR("close snapshot kernel");
1095 trace_kernel_destroy_metadata(ksess
->metadata
);
1097 /* Restore metadata state.*/
1098 ksess
->metadata
= saved_metadata
;
1099 ksess
->metadata_stream_fd
= saved_metadata_fd
;
1106 * Get the syscall mask array from the kernel tracer.
1108 * Return 0 on success else a negative value. In both case, syscall_mask should
1111 int kernel_syscall_mask(int chan_fd
, char **syscall_mask
, uint32_t *nr_bits
)
1113 assert(syscall_mask
);
1116 return kernctl_syscall_mask(chan_fd
, syscall_mask
, nr_bits
);