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.
26 #include <common/common.h>
27 #include <common/kernel-ctl/kernel-ctl.h>
28 #include <common/kernel-ctl/kernel-ioctl.h>
29 #include <common/sessiond-comm/sessiond-comm.h>
33 #include "kernel-consumer.h"
34 #include "kern-modules.h"
38 * Add context on a kernel channel.
40 * Assumes the ownership of ctx.
42 int kernel_add_channel_context(struct ltt_kernel_channel
*chan
,
43 struct ltt_kernel_context
*ctx
)
50 DBG("Adding context to channel %s", chan
->channel
->name
);
51 ret
= kernctl_add_context(chan
->fd
, &ctx
->ctx
);
55 /* Exists but not available for this kernel */
56 ret
= LTTNG_ERR_KERN_CONTEXT_UNAVAILABLE
;
59 /* If EEXIST, we just ignore the error */
63 PERROR("add context ioctl");
64 ret
= LTTNG_ERR_KERN_CONTEXT_FAIL
;
70 cds_list_add_tail(&ctx
->list
, &chan
->ctx_list
);
74 trace_kernel_destroy_context(ctx
);
80 * Create a new kernel session, register it to the kernel tracer and add it to
81 * the session daemon session.
83 int kernel_create_session(struct ltt_session
*session
, int tracer_fd
)
86 struct ltt_kernel_session
*lks
;
90 /* Allocate data structure */
91 lks
= trace_kernel_create_session();
97 /* Kernel tracer session creation */
98 ret
= kernctl_create_session(tracer_fd
);
100 PERROR("ioctl kernel create session");
105 /* Prevent fd duplication after execlp() */
106 ret
= fcntl(lks
->fd
, F_SETFD
, FD_CLOEXEC
);
108 PERROR("fcntl session fd");
111 lks
->id
= session
->id
;
112 lks
->consumer_fds_sent
= 0;
113 session
->kernel_session
= lks
;
115 DBG("Kernel session created (fd: %d)", lks
->fd
);
121 trace_kernel_destroy_session(lks
);
127 * Create a kernel channel, register it to the kernel tracer and add it to the
130 int kernel_create_channel(struct ltt_kernel_session
*session
,
131 struct lttng_channel
*chan
)
134 struct ltt_kernel_channel
*lkc
;
139 /* Allocate kernel channel */
140 lkc
= trace_kernel_create_channel(chan
);
145 DBG3("Kernel create channel %s with attr: %d, %" PRIu64
", %" PRIu64
", %u, %u, %d, %d",
146 chan
->name
, lkc
->channel
->attr
.overwrite
,
147 lkc
->channel
->attr
.subbuf_size
, lkc
->channel
->attr
.num_subbuf
,
148 lkc
->channel
->attr
.switch_timer_interval
, lkc
->channel
->attr
.read_timer_interval
,
149 lkc
->channel
->attr
.live_timer_interval
, lkc
->channel
->attr
.output
);
151 /* Kernel tracer channel creation */
152 ret
= kernctl_create_channel(session
->fd
, &lkc
->channel
->attr
);
154 PERROR("ioctl kernel create channel");
158 /* Setup the channel fd */
160 /* Prevent fd duplication after execlp() */
161 ret
= fcntl(lkc
->fd
, F_SETFD
, FD_CLOEXEC
);
163 PERROR("fcntl session fd");
166 /* Add channel to session */
167 cds_list_add(&lkc
->list
, &session
->channel_list
.head
);
168 session
->channel_count
++;
169 lkc
->session
= session
;
171 DBG("Kernel channel %s created (fd: %d)", lkc
->channel
->name
, lkc
->fd
);
184 * Create a kernel event, enable it to the kernel tracer and add it to the
185 * channel event list of the kernel session.
186 * We own filter_expression and filter.
188 int kernel_create_event(struct lttng_event
*ev
,
189 struct ltt_kernel_channel
*channel
,
190 char *filter_expression
,
191 struct lttng_filter_bytecode
*filter
)
194 struct ltt_kernel_event
*event
;
199 /* We pass ownership of filter_expression and filter */
200 event
= trace_kernel_create_event(ev
, filter_expression
,
207 ret
= kernctl_create_event(channel
->fd
, event
->event
);
213 WARN("Event type not implemented");
216 WARN("Event %s not found!", ev
->name
);
219 PERROR("create event ioctl");
224 event
->type
= ev
->type
;
226 /* Prevent fd duplication after execlp() */
227 ret
= fcntl(event
->fd
, F_SETFD
, FD_CLOEXEC
);
229 PERROR("fcntl session fd");
233 ret
= kernctl_filter(event
->fd
, filter
);
239 ret
= kernctl_enable(event
->fd
);
243 ret
= LTTNG_ERR_KERN_EVENT_EXIST
;
246 PERROR("enable kernel event");
252 /* Add event to event list */
253 cds_list_add(&event
->list
, &channel
->events_list
.head
);
254 channel
->event_count
++;
256 DBG("Event %s created (fd: %d)", ev
->name
, event
->fd
);
265 closeret
= close(event
->fd
);
267 PERROR("close event fd");
277 * Disable a kernel channel.
279 int kernel_disable_channel(struct ltt_kernel_channel
*chan
)
285 ret
= kernctl_disable(chan
->fd
);
287 PERROR("disable chan ioctl");
292 DBG("Kernel channel %s disabled (fd: %d)", chan
->channel
->name
, chan
->fd
);
301 * Enable a kernel channel.
303 int kernel_enable_channel(struct ltt_kernel_channel
*chan
)
309 ret
= kernctl_enable(chan
->fd
);
310 if (ret
< 0 && ret
!= -EEXIST
) {
311 PERROR("Enable kernel chan");
316 DBG("Kernel channel %s enabled (fd: %d)", chan
->channel
->name
, chan
->fd
);
325 * Enable a kernel event.
327 int kernel_enable_event(struct ltt_kernel_event
*event
)
333 ret
= kernctl_enable(event
->fd
);
337 ret
= LTTNG_ERR_KERN_EVENT_EXIST
;
340 PERROR("enable kernel event");
347 DBG("Kernel event %s enabled (fd: %d)", event
->event
->name
, event
->fd
);
356 * Disable a kernel event.
358 int kernel_disable_event(struct ltt_kernel_event
*event
)
364 ret
= kernctl_disable(event
->fd
);
368 ret
= LTTNG_ERR_KERN_EVENT_EXIST
;
371 PERROR("disable kernel event");
378 DBG("Kernel event %s disabled (fd: %d)", event
->event
->name
, event
->fd
);
387 int kernel_track_pid(struct ltt_kernel_session
*session
, int pid
)
391 DBG("Kernel track PID %d for session id %" PRIu64
".",
393 ret
= kernctl_track_pid(session
->fd
, pid
);
399 return LTTNG_ERR_INVALID
;
401 return LTTNG_ERR_NOMEM
;
403 return LTTNG_ERR_PID_TRACKED
;
405 return LTTNG_ERR_UNK
;
409 int kernel_untrack_pid(struct ltt_kernel_session
*session
, int pid
)
413 DBG("Kernel untrack PID %d for session id %" PRIu64
".",
415 ret
= kernctl_untrack_pid(session
->fd
, pid
);
421 return LTTNG_ERR_INVALID
;
423 return LTTNG_ERR_NOMEM
;
425 return LTTNG_ERR_PID_NOT_TRACKED
;
427 return LTTNG_ERR_UNK
;
431 ssize_t
kernel_list_tracker_pids(struct ltt_kernel_session
*session
,
436 ssize_t nbmem
, count
= 0;
440 fd
= kernctl_list_tracker_pids(session
->fd
);
442 PERROR("kernel tracker pids list");
446 fp
= fdopen(fd
, "r");
448 PERROR("kernel tracker pids list fdopen");
452 nbmem
= KERNEL_TRACKER_PIDS_INIT_LIST_SIZE
;
453 pids
= zmalloc(sizeof(*pids
) * nbmem
);
455 PERROR("alloc list pids");
460 while (fscanf(fp
, "process { pid = %u; };\n", &pid
) == 1) {
461 if (count
>= nbmem
) {
465 new_nbmem
= nbmem
<< 1;
466 DBG("Reallocating pids list from %zu to %zu entries",
468 new_pids
= realloc(pids
, new_nbmem
* sizeof(*new_pids
));
469 if (new_pids
== NULL
) {
470 PERROR("realloc list events");
475 /* Zero the new memory */
476 memset(new_pids
+ nbmem
, 0,
477 (new_nbmem
- nbmem
) * sizeof(*new_pids
));
485 DBG("Kernel list tracker pids done (%zd pids)", count
);
487 ret
= fclose(fp
); /* closes both fp and fd */
503 * Create kernel metadata, open from the kernel tracer and add it to the
506 int kernel_open_metadata(struct ltt_kernel_session
*session
)
509 struct ltt_kernel_metadata
*lkm
= NULL
;
513 /* Allocate kernel metadata */
514 lkm
= trace_kernel_create_metadata();
519 /* Kernel tracer metadata creation */
520 ret
= kernctl_open_metadata(session
->fd
, &lkm
->conf
->attr
);
526 /* Prevent fd duplication after execlp() */
527 ret
= fcntl(lkm
->fd
, F_SETFD
, FD_CLOEXEC
);
529 PERROR("fcntl session fd");
532 session
->metadata
= lkm
;
534 DBG("Kernel metadata opened (fd: %d)", lkm
->fd
);
539 trace_kernel_destroy_metadata(lkm
);
545 * Start tracing session.
547 int kernel_start_session(struct ltt_kernel_session
*session
)
553 ret
= kernctl_start_session(session
->fd
);
555 PERROR("ioctl start session");
559 DBG("Kernel session started");
568 * Make a kernel wait to make sure in-flight probe have completed.
570 void kernel_wait_quiescent(int fd
)
574 DBG("Kernel quiescent wait on %d", fd
);
576 ret
= kernctl_wait_quiescent(fd
);
578 PERROR("wait quiescent ioctl");
579 ERR("Kernel quiescent wait failed");
584 * Force flush buffer of metadata.
586 int kernel_metadata_flush_buffer(int fd
)
590 DBG("Kernel flushing metadata buffer on fd %d", fd
);
592 ret
= kernctl_buffer_flush(fd
);
594 ERR("Fail to flush metadata buffers %d (ret: %d)", fd
, ret
);
601 * Force flush buffer for channel.
603 int kernel_flush_buffer(struct ltt_kernel_channel
*channel
)
606 struct ltt_kernel_stream
*stream
;
610 DBG("Flush buffer for channel %s", channel
->channel
->name
);
612 cds_list_for_each_entry(stream
, &channel
->stream_list
.head
, list
) {
613 DBG("Flushing channel stream %d", stream
->fd
);
614 ret
= kernctl_buffer_flush(stream
->fd
);
617 ERR("Fail to flush buffer for stream %d (ret: %d)",
626 * Stop tracing session.
628 int kernel_stop_session(struct ltt_kernel_session
*session
)
634 ret
= kernctl_stop_session(session
->fd
);
639 DBG("Kernel session stopped");
648 * Open stream of channel, register it to the kernel tracer and add it
649 * to the stream list of the channel.
651 * Note: given that the streams may appear in random order wrt CPU
652 * number (e.g. cpu hotplug), the index value of the stream number in
653 * the stream name is not necessarily linked to the CPU number.
655 * Return the number of created stream. Else, a negative value.
657 int kernel_open_channel_stream(struct ltt_kernel_channel
*channel
)
660 struct ltt_kernel_stream
*lks
;
664 while ((ret
= kernctl_create_stream(channel
->fd
)) >= 0) {
665 lks
= trace_kernel_create_stream(channel
->channel
->name
,
666 channel
->stream_count
);
676 /* Prevent fd duplication after execlp() */
677 ret
= fcntl(lks
->fd
, F_SETFD
, FD_CLOEXEC
);
679 PERROR("fcntl session fd");
682 lks
->tracefile_size
= channel
->channel
->attr
.tracefile_size
;
683 lks
->tracefile_count
= channel
->channel
->attr
.tracefile_count
;
685 /* Add stream to channel stream list */
686 cds_list_add(&lks
->list
, &channel
->stream_list
.head
);
687 channel
->stream_count
++;
689 DBG("Kernel stream %s created (fd: %d, state: %d)", lks
->name
, lks
->fd
,
693 return channel
->stream_count
;
700 * Open the metadata stream and set it to the kernel session.
702 int kernel_open_metadata_stream(struct ltt_kernel_session
*session
)
708 ret
= kernctl_create_stream(session
->metadata
->fd
);
710 PERROR("kernel create metadata stream");
714 DBG("Kernel metadata stream created (fd: %d)", ret
);
715 session
->metadata_stream_fd
= ret
;
716 /* Prevent fd duplication after execlp() */
717 ret
= fcntl(session
->metadata_stream_fd
, F_SETFD
, FD_CLOEXEC
);
719 PERROR("fcntl session fd");
729 * Get the event list from the kernel tracer and return the number of elements.
731 ssize_t
kernel_list_events(int tracer_fd
, struct lttng_event
**events
)
735 size_t nbmem
, count
= 0;
737 struct lttng_event
*elist
;
741 fd
= kernctl_tracepoint_list(tracer_fd
);
743 PERROR("kernel tracepoint list");
747 fp
= fdopen(fd
, "r");
749 PERROR("kernel tracepoint list fdopen");
754 * Init memory size counter
755 * See kernel-ctl.h for explanation of this value
757 nbmem
= KERNEL_EVENT_INIT_LIST_SIZE
;
758 elist
= zmalloc(sizeof(struct lttng_event
) * nbmem
);
760 PERROR("alloc list events");
765 while (fscanf(fp
, "event { name = %m[^;]; };\n", &event
) == 1) {
766 if (count
>= nbmem
) {
767 struct lttng_event
*new_elist
;
770 new_nbmem
= nbmem
<< 1;
771 DBG("Reallocating event list from %zu to %zu bytes",
773 new_elist
= realloc(elist
, new_nbmem
* sizeof(struct lttng_event
));
774 if (new_elist
== NULL
) {
775 PERROR("realloc list events");
781 /* Zero the new memory */
782 memset(new_elist
+ nbmem
, 0,
783 (new_nbmem
- nbmem
) * sizeof(struct lttng_event
));
787 strncpy(elist
[count
].name
, event
, LTTNG_SYMBOL_NAME_LEN
);
788 elist
[count
].name
[LTTNG_SYMBOL_NAME_LEN
- 1] = '\0';
789 elist
[count
].enabled
= -1;
795 DBG("Kernel list events done (%zu events)", count
);
797 ret
= fclose(fp
); /* closes both fp and fd */
813 * Get kernel version and validate it.
815 int kernel_validate_version(int tracer_fd
)
818 struct lttng_kernel_tracer_version version
;
819 struct lttng_kernel_tracer_abi_version abi_version
;
821 ret
= kernctl_tracer_version(tracer_fd
, &version
);
823 ERR("Failed to retrieve the lttng-modules version");
827 /* Validate version */
828 if (version
.major
!= VERSION_MAJOR
) {
829 ERR("Kernel tracer major version (%d) is not compatible with lttng-tools major version (%d)",
830 version
.major
, VERSION_MAJOR
);
833 ret
= kernctl_tracer_abi_version(tracer_fd
, &abi_version
);
835 ERR("Failed to retrieve lttng-modules ABI version");
838 if (abi_version
.major
!= LTTNG_MODULES_ABI_MAJOR_VERSION
) {
839 ERR("Kernel tracer ABI version (%d.%d) does not match the expected ABI major version (%d.*)",
840 abi_version
.major
, abi_version
.minor
,
841 LTTNG_MODULES_ABI_MAJOR_VERSION
);
844 DBG2("Kernel tracer version validated (%d.%d, ABI %d.%d)",
845 version
.major
, version
.minor
,
846 abi_version
.major
, abi_version
.minor
);
853 ERR("Kernel tracer version check failed; kernel tracing will not be available");
858 * Kernel work-arounds called at the start of sessiond main().
860 int init_kernel_workarounds(void)
866 * boot_id needs to be read once before being used concurrently
867 * to deal with a Linux kernel race. A fix is proposed for
868 * upstream, but the work-around is needed for older kernels.
870 fp
= fopen("/proc/sys/kernel/random/boot_id", "r");
877 ret
= fread(buf
, 1, sizeof(buf
), fp
);
879 /* Ignore error, we don't really care */
891 * Complete teardown of a kernel session.
893 void kernel_destroy_session(struct ltt_kernel_session
*ksess
)
896 DBG3("No kernel session when tearing down session");
900 DBG("Tearing down kernel session");
903 * Destroy channels on the consumer if at least one FD has been sent and we
904 * are in no output mode because the streams are in *no* monitor mode so we
905 * have to send a command to clean them up or else they leaked.
907 if (!ksess
->output_traces
&& ksess
->consumer_fds_sent
) {
909 struct consumer_socket
*socket
;
910 struct lttng_ht_iter iter
;
912 /* For each consumer socket. */
914 cds_lfht_for_each_entry(ksess
->consumer
->socks
->ht
, &iter
.iter
,
916 struct ltt_kernel_channel
*chan
;
918 /* For each channel, ask the consumer to destroy it. */
919 cds_list_for_each_entry(chan
, &ksess
->channel_list
.head
, list
) {
920 ret
= kernel_consumer_destroy_channel(socket
, chan
);
922 /* Consumer is probably dead. Use next socket. */
930 /* Close any relayd session */
931 consumer_output_send_destroy_relayd(ksess
->consumer
);
933 trace_kernel_destroy_session(ksess
);
937 * Destroy a kernel channel object. It does not do anything on the tracer side.
939 void kernel_destroy_channel(struct ltt_kernel_channel
*kchan
)
941 struct ltt_kernel_session
*ksess
= NULL
;
944 assert(kchan
->channel
);
946 DBG3("Kernel destroy channel %s", kchan
->channel
->name
);
948 /* Update channel count of associated session. */
949 if (kchan
->session
) {
950 /* Keep pointer reference so we can update it after the destroy. */
951 ksess
= kchan
->session
;
954 trace_kernel_destroy_channel(kchan
);
957 * At this point the kernel channel is not visible anymore. This is safe
958 * since in order to work on a visible kernel session, the tracing session
959 * lock (ltt_session.lock) MUST be acquired.
962 ksess
->channel_count
--;
967 * Take a snapshot for a given kernel session.
969 * Return 0 on success or else return a LTTNG_ERR code.
971 int kernel_snapshot_record(struct ltt_kernel_session
*ksess
,
972 struct snapshot_output
*output
, int wait
,
973 uint64_t nb_packets_per_stream
)
975 int err
, ret
, saved_metadata_fd
;
976 struct consumer_socket
*socket
;
977 struct lttng_ht_iter iter
;
978 struct ltt_kernel_metadata
*saved_metadata
;
981 assert(ksess
->consumer
);
984 DBG("Kernel snapshot record started");
986 /* Save current metadata since the following calls will change it. */
987 saved_metadata
= ksess
->metadata
;
988 saved_metadata_fd
= ksess
->metadata_stream_fd
;
992 ret
= kernel_open_metadata(ksess
);
994 ret
= LTTNG_ERR_KERN_META_FAIL
;
998 ret
= kernel_open_metadata_stream(ksess
);
1000 ret
= LTTNG_ERR_KERN_META_FAIL
;
1001 goto error_open_stream
;
1004 /* Send metadata to consumer and snapshot everything. */
1005 cds_lfht_for_each_entry(ksess
->consumer
->socks
->ht
, &iter
.iter
,
1006 socket
, node
.node
) {
1007 struct consumer_output
*saved_output
;
1008 struct ltt_kernel_channel
*chan
;
1011 * Temporarly switch consumer output for our snapshot output. As long
1012 * as the session lock is taken, this is safe.
1014 saved_output
= ksess
->consumer
;
1015 ksess
->consumer
= output
->consumer
;
1017 pthread_mutex_lock(socket
->lock
);
1018 /* This stream must not be monitored by the consumer. */
1019 ret
= kernel_consumer_add_metadata(socket
, ksess
, 0);
1020 pthread_mutex_unlock(socket
->lock
);
1021 /* Put back the saved consumer output into the session. */
1022 ksess
->consumer
= saved_output
;
1024 ret
= LTTNG_ERR_KERN_CONSUMER_FAIL
;
1025 goto error_consumer
;
1028 /* For each channel, ask the consumer to snapshot it. */
1029 cds_list_for_each_entry(chan
, &ksess
->channel_list
.head
, list
) {
1030 pthread_mutex_lock(socket
->lock
);
1031 ret
= consumer_snapshot_channel(socket
, chan
->fd
, output
, 0,
1032 ksess
->uid
, ksess
->gid
,
1033 DEFAULT_KERNEL_TRACE_DIR
, wait
,
1034 nb_packets_per_stream
);
1035 pthread_mutex_unlock(socket
->lock
);
1037 ret
= LTTNG_ERR_KERN_CONSUMER_FAIL
;
1038 (void) kernel_consumer_destroy_metadata(socket
,
1040 goto error_consumer
;
1044 /* Snapshot metadata, */
1045 pthread_mutex_lock(socket
->lock
);
1046 ret
= consumer_snapshot_channel(socket
, ksess
->metadata
->fd
, output
,
1047 1, ksess
->uid
, ksess
->gid
,
1048 DEFAULT_KERNEL_TRACE_DIR
, wait
, 0);
1049 pthread_mutex_unlock(socket
->lock
);
1051 ret
= LTTNG_ERR_KERN_CONSUMER_FAIL
;
1052 goto error_consumer
;
1056 * The metadata snapshot is done, ask the consumer to destroy it since
1057 * it's not monitored on the consumer side.
1059 (void) kernel_consumer_destroy_metadata(socket
, ksess
->metadata
);
1065 /* Close newly opened metadata stream. It's now on the consumer side. */
1066 err
= close(ksess
->metadata_stream_fd
);
1068 PERROR("close snapshot kernel");
1072 trace_kernel_destroy_metadata(ksess
->metadata
);
1074 /* Restore metadata state.*/
1075 ksess
->metadata
= saved_metadata
;
1076 ksess
->metadata_stream_fd
= saved_metadata_fd
;
1083 * Get the syscall mask array from the kernel tracer.
1085 * Return 0 on success else a negative value. In both case, syscall_mask should
1088 int kernel_syscall_mask(int chan_fd
, char **syscall_mask
, uint32_t *nr_bits
)
1090 assert(syscall_mask
);
1093 return kernctl_syscall_mask(chan_fd
, syscall_mask
, nr_bits
);