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 int kernel_add_channel_context(struct ltt_kernel_channel
*chan
,
41 struct ltt_kernel_context
*ctx
)
48 DBG("Adding context to channel %s", chan
->channel
->name
);
49 ret
= kernctl_add_context(chan
->fd
, &ctx
->ctx
);
53 /* Exists but not available for this kernel */
54 ret
= LTTNG_ERR_KERN_CONTEXT_UNAVAILABLE
;
57 /* If EEXIST, we just ignore the error */
61 PERROR("add context ioctl");
62 ret
= LTTNG_ERR_KERN_CONTEXT_FAIL
;
69 cds_list_add_tail(&ctx
->list
, &chan
->ctx_list
);
75 * Create a new kernel session, register it to the kernel tracer and add it to
76 * the session daemon session.
78 int kernel_create_session(struct ltt_session
*session
, int tracer_fd
)
81 struct ltt_kernel_session
*lks
;
85 /* Allocate data structure */
86 lks
= trace_kernel_create_session();
92 /* Kernel tracer session creation */
93 ret
= kernctl_create_session(tracer_fd
);
95 PERROR("ioctl kernel create session");
100 /* Prevent fd duplication after execlp() */
101 ret
= fcntl(lks
->fd
, F_SETFD
, FD_CLOEXEC
);
103 PERROR("fcntl session fd");
106 lks
->id
= session
->id
;
107 lks
->consumer_fds_sent
= 0;
108 session
->kernel_session
= lks
;
110 DBG("Kernel session created (fd: %d)", lks
->fd
);
116 trace_kernel_destroy_session(lks
);
122 * Create a kernel channel, register it to the kernel tracer and add it to the
125 int kernel_create_channel(struct ltt_kernel_session
*session
,
126 struct lttng_channel
*chan
)
129 struct ltt_kernel_channel
*lkc
;
134 /* Allocate kernel channel */
135 lkc
= trace_kernel_create_channel(chan
);
140 DBG3("Kernel create channel %s with attr: %d, %" PRIu64
", %" PRIu64
", %u, %u, %d, %d",
141 chan
->name
, lkc
->channel
->attr
.overwrite
,
142 lkc
->channel
->attr
.subbuf_size
, lkc
->channel
->attr
.num_subbuf
,
143 lkc
->channel
->attr
.switch_timer_interval
, lkc
->channel
->attr
.read_timer_interval
,
144 lkc
->channel
->attr
.live_timer_interval
, lkc
->channel
->attr
.output
);
146 /* Kernel tracer channel creation */
147 ret
= kernctl_create_channel(session
->fd
, &lkc
->channel
->attr
);
149 PERROR("ioctl kernel create channel");
153 /* Setup the channel fd */
155 /* Prevent fd duplication after execlp() */
156 ret
= fcntl(lkc
->fd
, F_SETFD
, FD_CLOEXEC
);
158 PERROR("fcntl session fd");
161 /* Add channel to session */
162 cds_list_add(&lkc
->list
, &session
->channel_list
.head
);
163 session
->channel_count
++;
164 lkc
->session
= session
;
166 DBG("Kernel channel %s created (fd: %d)", lkc
->channel
->name
, lkc
->fd
);
179 * Create a kernel event, enable it to the kernel tracer and add it to the
180 * channel event list of the kernel session.
181 * We own filter_expression and filter.
183 int kernel_create_event(struct lttng_event
*ev
,
184 struct ltt_kernel_channel
*channel
,
185 char *filter_expression
,
186 struct lttng_filter_bytecode
*filter
)
189 struct ltt_kernel_event
*event
;
194 /* We pass ownership of filter_expression and filter */
195 event
= trace_kernel_create_event(ev
, filter_expression
,
202 ret
= kernctl_create_event(channel
->fd
, event
->event
);
208 WARN("Event type not implemented");
211 WARN("Event %s not found!", ev
->name
);
214 PERROR("create event ioctl");
219 event
->type
= ev
->type
;
221 /* Prevent fd duplication after execlp() */
222 ret
= fcntl(event
->fd
, F_SETFD
, FD_CLOEXEC
);
224 PERROR("fcntl session fd");
228 ret
= kernctl_filter(event
->fd
, filter
);
234 ret
= kernctl_enable(event
->fd
);
238 ret
= LTTNG_ERR_KERN_EVENT_EXIST
;
241 PERROR("enable kernel event");
247 /* Add event to event list */
248 cds_list_add(&event
->list
, &channel
->events_list
.head
);
249 channel
->event_count
++;
251 DBG("Event %s created (fd: %d)", ev
->name
, event
->fd
);
260 closeret
= close(event
->fd
);
262 PERROR("close event fd");
272 * Disable a kernel channel.
274 int kernel_disable_channel(struct ltt_kernel_channel
*chan
)
280 ret
= kernctl_disable(chan
->fd
);
282 PERROR("disable chan ioctl");
287 DBG("Kernel channel %s disabled (fd: %d)", chan
->channel
->name
, chan
->fd
);
296 * Enable a kernel channel.
298 int kernel_enable_channel(struct ltt_kernel_channel
*chan
)
304 ret
= kernctl_enable(chan
->fd
);
305 if (ret
< 0 && ret
!= -EEXIST
) {
306 PERROR("Enable kernel chan");
311 DBG("Kernel channel %s enabled (fd: %d)", chan
->channel
->name
, chan
->fd
);
320 * Enable a kernel event.
322 int kernel_enable_event(struct ltt_kernel_event
*event
)
328 ret
= kernctl_enable(event
->fd
);
332 ret
= LTTNG_ERR_KERN_EVENT_EXIST
;
335 PERROR("enable kernel event");
342 DBG("Kernel event %s enabled (fd: %d)", event
->event
->name
, event
->fd
);
351 * Disable a kernel event.
353 int kernel_disable_event(struct ltt_kernel_event
*event
)
359 ret
= kernctl_disable(event
->fd
);
363 ret
= LTTNG_ERR_KERN_EVENT_EXIST
;
366 PERROR("disable kernel event");
373 DBG("Kernel event %s disabled (fd: %d)", event
->event
->name
, event
->fd
);
382 int kernel_track_pid(struct ltt_kernel_session
*session
, int pid
)
386 DBG("Kernel track PID %d for session id %" PRIu64
".",
388 ret
= kernctl_track_pid(session
->fd
, pid
);
394 return LTTNG_ERR_INVALID
;
396 return LTTNG_ERR_NOMEM
;
398 return LTTNG_ERR_PID_TRACKED
;
400 return LTTNG_ERR_UNK
;
404 int kernel_untrack_pid(struct ltt_kernel_session
*session
, int pid
)
408 DBG("Kernel untrack PID %d for session id %" PRIu64
".",
410 ret
= kernctl_untrack_pid(session
->fd
, pid
);
416 return LTTNG_ERR_INVALID
;
418 return LTTNG_ERR_NOMEM
;
420 return LTTNG_ERR_PID_NOT_TRACKED
;
422 return LTTNG_ERR_UNK
;
426 ssize_t
kernel_list_tracker_pids(struct ltt_kernel_session
*session
,
431 ssize_t nbmem
, count
= 0;
435 fd
= kernctl_list_tracker_pids(session
->fd
);
437 PERROR("kernel tracker pids list");
441 fp
= fdopen(fd
, "r");
443 PERROR("kernel tracker pids list fdopen");
447 nbmem
= KERNEL_TRACKER_PIDS_INIT_LIST_SIZE
;
448 pids
= zmalloc(sizeof(*pids
) * nbmem
);
450 PERROR("alloc list pids");
455 while (fscanf(fp
, "process { pid = %u; };\n", &pid
) == 1) {
456 if (count
>= nbmem
) {
460 new_nbmem
= nbmem
<< 1;
461 DBG("Reallocating pids list from %zu to %zu entries",
463 new_pids
= realloc(pids
, new_nbmem
* sizeof(*new_pids
));
464 if (new_pids
== NULL
) {
465 PERROR("realloc list events");
470 /* Zero the new memory */
471 memset(new_pids
+ nbmem
, 0,
472 (new_nbmem
- nbmem
) * sizeof(*new_pids
));
480 DBG("Kernel list tracker pids done (%zd pids)", count
);
482 ret
= fclose(fp
); /* closes both fp and fd */
498 * Create kernel metadata, open from the kernel tracer and add it to the
501 int kernel_open_metadata(struct ltt_kernel_session
*session
)
504 struct ltt_kernel_metadata
*lkm
= NULL
;
508 /* Allocate kernel metadata */
509 lkm
= trace_kernel_create_metadata();
514 /* Kernel tracer metadata creation */
515 ret
= kernctl_open_metadata(session
->fd
, &lkm
->conf
->attr
);
521 /* Prevent fd duplication after execlp() */
522 ret
= fcntl(lkm
->fd
, F_SETFD
, FD_CLOEXEC
);
524 PERROR("fcntl session fd");
527 session
->metadata
= lkm
;
529 DBG("Kernel metadata opened (fd: %d)", lkm
->fd
);
534 trace_kernel_destroy_metadata(lkm
);
540 * Start tracing session.
542 int kernel_start_session(struct ltt_kernel_session
*session
)
548 ret
= kernctl_start_session(session
->fd
);
550 PERROR("ioctl start session");
554 DBG("Kernel session started");
563 * Make a kernel wait to make sure in-flight probe have completed.
565 void kernel_wait_quiescent(int fd
)
569 DBG("Kernel quiescent wait on %d", fd
);
571 ret
= kernctl_wait_quiescent(fd
);
573 PERROR("wait quiescent ioctl");
574 ERR("Kernel quiescent wait failed");
579 * Force flush buffer of metadata.
581 int kernel_metadata_flush_buffer(int fd
)
585 DBG("Kernel flushing metadata buffer on fd %d", fd
);
587 ret
= kernctl_buffer_flush(fd
);
589 ERR("Fail to flush metadata buffers %d (ret: %d)", fd
, ret
);
596 * Force flush buffer for channel.
598 int kernel_flush_buffer(struct ltt_kernel_channel
*channel
)
601 struct ltt_kernel_stream
*stream
;
605 DBG("Flush buffer for channel %s", channel
->channel
->name
);
607 cds_list_for_each_entry(stream
, &channel
->stream_list
.head
, list
) {
608 DBG("Flushing channel stream %d", stream
->fd
);
609 ret
= kernctl_buffer_flush(stream
->fd
);
612 ERR("Fail to flush buffer for stream %d (ret: %d)",
621 * Stop tracing session.
623 int kernel_stop_session(struct ltt_kernel_session
*session
)
629 ret
= kernctl_stop_session(session
->fd
);
634 DBG("Kernel session stopped");
643 * Open stream of channel, register it to the kernel tracer and add it
644 * to the stream list of the channel.
646 * Note: given that the streams may appear in random order wrt CPU
647 * number (e.g. cpu hotplug), the index value of the stream number in
648 * the stream name is not necessarily linked to the CPU number.
650 * Return the number of created stream. Else, a negative value.
652 int kernel_open_channel_stream(struct ltt_kernel_channel
*channel
)
655 struct ltt_kernel_stream
*lks
;
659 while ((ret
= kernctl_create_stream(channel
->fd
)) >= 0) {
660 lks
= trace_kernel_create_stream(channel
->channel
->name
,
661 channel
->stream_count
);
671 /* Prevent fd duplication after execlp() */
672 ret
= fcntl(lks
->fd
, F_SETFD
, FD_CLOEXEC
);
674 PERROR("fcntl session fd");
677 lks
->tracefile_size
= channel
->channel
->attr
.tracefile_size
;
678 lks
->tracefile_count
= channel
->channel
->attr
.tracefile_count
;
680 /* Add stream to channel stream list */
681 cds_list_add(&lks
->list
, &channel
->stream_list
.head
);
682 channel
->stream_count
++;
684 DBG("Kernel stream %s created (fd: %d, state: %d)", lks
->name
, lks
->fd
,
688 return channel
->stream_count
;
695 * Open the metadata stream and set it to the kernel session.
697 int kernel_open_metadata_stream(struct ltt_kernel_session
*session
)
703 ret
= kernctl_create_stream(session
->metadata
->fd
);
705 PERROR("kernel create metadata stream");
709 DBG("Kernel metadata stream created (fd: %d)", ret
);
710 session
->metadata_stream_fd
= ret
;
711 /* Prevent fd duplication after execlp() */
712 ret
= fcntl(session
->metadata_stream_fd
, F_SETFD
, FD_CLOEXEC
);
714 PERROR("fcntl session fd");
724 * Get the event list from the kernel tracer and return the number of elements.
726 ssize_t
kernel_list_events(int tracer_fd
, struct lttng_event
**events
)
730 size_t nbmem
, count
= 0;
732 struct lttng_event
*elist
;
736 fd
= kernctl_tracepoint_list(tracer_fd
);
738 PERROR("kernel tracepoint list");
742 fp
= fdopen(fd
, "r");
744 PERROR("kernel tracepoint list fdopen");
749 * Init memory size counter
750 * See kernel-ctl.h for explanation of this value
752 nbmem
= KERNEL_EVENT_INIT_LIST_SIZE
;
753 elist
= zmalloc(sizeof(struct lttng_event
) * nbmem
);
755 PERROR("alloc list events");
760 while (fscanf(fp
, "event { name = %m[^;]; };\n", &event
) == 1) {
761 if (count
>= nbmem
) {
762 struct lttng_event
*new_elist
;
765 new_nbmem
= nbmem
<< 1;
766 DBG("Reallocating event list from %zu to %zu bytes",
768 new_elist
= realloc(elist
, new_nbmem
* sizeof(struct lttng_event
));
769 if (new_elist
== NULL
) {
770 PERROR("realloc list events");
776 /* Zero the new memory */
777 memset(new_elist
+ nbmem
, 0,
778 (new_nbmem
- nbmem
) * sizeof(struct lttng_event
));
782 strncpy(elist
[count
].name
, event
, LTTNG_SYMBOL_NAME_LEN
);
783 elist
[count
].name
[LTTNG_SYMBOL_NAME_LEN
- 1] = '\0';
784 elist
[count
].enabled
= -1;
790 DBG("Kernel list events done (%zu events)", count
);
792 ret
= fclose(fp
); /* closes both fp and fd */
808 * Get kernel version and validate it.
810 int kernel_validate_version(int tracer_fd
)
813 struct lttng_kernel_tracer_version version
;
814 struct lttng_kernel_tracer_abi_version abi_version
;
816 ret
= kernctl_tracer_version(tracer_fd
, &version
);
818 ERR("Failed to retrieve the lttng-modules version");
822 /* Validate version */
823 if (version
.major
!= VERSION_MAJOR
) {
824 ERR("Kernel tracer major version (%d) is not compatible with lttng-tools major version (%d)",
825 version
.major
, VERSION_MAJOR
);
828 ret
= kernctl_tracer_abi_version(tracer_fd
, &abi_version
);
830 ERR("Failed to retrieve lttng-modules ABI version");
833 if (abi_version
.major
!= LTTNG_MODULES_ABI_MAJOR_VERSION
) {
834 ERR("Kernel tracer ABI version (%d.%d) does not match the expected ABI major version (%d.*)",
835 abi_version
.major
, abi_version
.minor
,
836 LTTNG_MODULES_ABI_MAJOR_VERSION
);
839 DBG2("Kernel tracer version validated (%d.%d, ABI %d.%d)",
840 version
.major
, version
.minor
,
841 abi_version
.major
, abi_version
.minor
);
848 ERR("Kernel tracer version check failed; kernel tracing will not be available");
853 * Kernel work-arounds called at the start of sessiond main().
855 int init_kernel_workarounds(void)
861 * boot_id needs to be read once before being used concurrently
862 * to deal with a Linux kernel race. A fix is proposed for
863 * upstream, but the work-around is needed for older kernels.
865 fp
= fopen("/proc/sys/kernel/random/boot_id", "r");
872 ret
= fread(buf
, 1, sizeof(buf
), fp
);
874 /* Ignore error, we don't really care */
886 * Complete teardown of a kernel session.
888 void kernel_destroy_session(struct ltt_kernel_session
*ksess
)
891 DBG3("No kernel session when tearing down session");
895 DBG("Tearing down kernel session");
898 * Destroy channels on the consumer if at least one FD has been sent and we
899 * are in no output mode because the streams are in *no* monitor mode so we
900 * have to send a command to clean them up or else they leaked.
902 if (!ksess
->output_traces
&& ksess
->consumer_fds_sent
) {
904 struct consumer_socket
*socket
;
905 struct lttng_ht_iter iter
;
907 /* For each consumer socket. */
909 cds_lfht_for_each_entry(ksess
->consumer
->socks
->ht
, &iter
.iter
,
911 struct ltt_kernel_channel
*chan
;
913 /* For each channel, ask the consumer to destroy it. */
914 cds_list_for_each_entry(chan
, &ksess
->channel_list
.head
, list
) {
915 ret
= kernel_consumer_destroy_channel(socket
, chan
);
917 /* Consumer is probably dead. Use next socket. */
925 /* Close any relayd session */
926 consumer_output_send_destroy_relayd(ksess
->consumer
);
928 trace_kernel_destroy_session(ksess
);
932 * Destroy a kernel channel object. It does not do anything on the tracer side.
934 void kernel_destroy_channel(struct ltt_kernel_channel
*kchan
)
936 struct ltt_kernel_session
*ksess
= NULL
;
939 assert(kchan
->channel
);
941 DBG3("Kernel destroy channel %s", kchan
->channel
->name
);
943 /* Update channel count of associated session. */
944 if (kchan
->session
) {
945 /* Keep pointer reference so we can update it after the destroy. */
946 ksess
= kchan
->session
;
949 trace_kernel_destroy_channel(kchan
);
952 * At this point the kernel channel is not visible anymore. This is safe
953 * since in order to work on a visible kernel session, the tracing session
954 * lock (ltt_session.lock) MUST be acquired.
957 ksess
->channel_count
--;
962 * Take a snapshot for a given kernel session.
964 * Return 0 on success or else return a LTTNG_ERR code.
966 int kernel_snapshot_record(struct ltt_kernel_session
*ksess
,
967 struct snapshot_output
*output
, int wait
,
968 uint64_t nb_packets_per_stream
)
970 int err
, ret
, saved_metadata_fd
;
971 struct consumer_socket
*socket
;
972 struct lttng_ht_iter iter
;
973 struct ltt_kernel_metadata
*saved_metadata
;
976 assert(ksess
->consumer
);
979 DBG("Kernel snapshot record started");
981 /* Save current metadata since the following calls will change it. */
982 saved_metadata
= ksess
->metadata
;
983 saved_metadata_fd
= ksess
->metadata_stream_fd
;
987 ret
= kernel_open_metadata(ksess
);
989 ret
= LTTNG_ERR_KERN_META_FAIL
;
993 ret
= kernel_open_metadata_stream(ksess
);
995 ret
= LTTNG_ERR_KERN_META_FAIL
;
996 goto error_open_stream
;
999 /* Send metadata to consumer and snapshot everything. */
1000 cds_lfht_for_each_entry(ksess
->consumer
->socks
->ht
, &iter
.iter
,
1001 socket
, node
.node
) {
1002 struct consumer_output
*saved_output
;
1003 struct ltt_kernel_channel
*chan
;
1006 * Temporarly switch consumer output for our snapshot output. As long
1007 * as the session lock is taken, this is safe.
1009 saved_output
= ksess
->consumer
;
1010 ksess
->consumer
= output
->consumer
;
1012 pthread_mutex_lock(socket
->lock
);
1013 /* This stream must not be monitored by the consumer. */
1014 ret
= kernel_consumer_add_metadata(socket
, ksess
, 0);
1015 pthread_mutex_unlock(socket
->lock
);
1016 /* Put back the saved consumer output into the session. */
1017 ksess
->consumer
= saved_output
;
1019 ret
= LTTNG_ERR_KERN_CONSUMER_FAIL
;
1020 goto error_consumer
;
1023 /* For each channel, ask the consumer to snapshot it. */
1024 cds_list_for_each_entry(chan
, &ksess
->channel_list
.head
, list
) {
1025 pthread_mutex_lock(socket
->lock
);
1026 ret
= consumer_snapshot_channel(socket
, chan
->fd
, output
, 0,
1027 ksess
->uid
, ksess
->gid
,
1028 DEFAULT_KERNEL_TRACE_DIR
, wait
,
1029 nb_packets_per_stream
);
1030 pthread_mutex_unlock(socket
->lock
);
1032 ret
= LTTNG_ERR_KERN_CONSUMER_FAIL
;
1033 (void) kernel_consumer_destroy_metadata(socket
,
1035 goto error_consumer
;
1039 /* Snapshot metadata, */
1040 pthread_mutex_lock(socket
->lock
);
1041 ret
= consumer_snapshot_channel(socket
, ksess
->metadata
->fd
, output
,
1042 1, ksess
->uid
, ksess
->gid
,
1043 DEFAULT_KERNEL_TRACE_DIR
, wait
, 0);
1044 pthread_mutex_unlock(socket
->lock
);
1046 ret
= LTTNG_ERR_KERN_CONSUMER_FAIL
;
1047 goto error_consumer
;
1051 * The metadata snapshot is done, ask the consumer to destroy it since
1052 * it's not monitored on the consumer side.
1054 (void) kernel_consumer_destroy_metadata(socket
, ksess
->metadata
);
1060 /* Close newly opened metadata stream. It's now on the consumer side. */
1061 err
= close(ksess
->metadata_stream_fd
);
1063 PERROR("close snapshot kernel");
1067 trace_kernel_destroy_metadata(ksess
->metadata
);
1069 /* Restore metadata state.*/
1070 ksess
->metadata
= saved_metadata
;
1071 ksess
->metadata_stream_fd
= saved_metadata_fd
;
1078 * Get the syscall mask array from the kernel tracer.
1080 * Return 0 on success else a negative value. In both case, syscall_mask should
1083 int kernel_syscall_mask(int chan_fd
, char **syscall_mask
, uint32_t *nr_bits
)
1085 assert(syscall_mask
);
1088 return kernctl_syscall_mask(chan_fd
, syscall_mask
, nr_bits
);
1092 * Check for the support of the RING_BUFFER_SNAPSHOT_SAMPLE_POSITIONS via abi
1095 * Return 1 on success, 0 when feature is not supported, negative value in case
1098 int kernel_supports_ring_buffer_snapshot_sample_positions(int tracer_fd
)
1100 int ret
= 0; // Not supported by default
1101 struct lttng_kernel_tracer_abi_version abi
;
1103 ret
= kernctl_tracer_abi_version(tracer_fd
, &abi
);
1105 ERR("Failed to retrieve lttng-modules ABI version");
1110 * RING_BUFFER_SNAPSHOT_SAMPLE_POSITIONS was introduced in 2.3
1112 if (abi
.major
>= 2 && abi
.minor
>= 3) {