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.
28 #include <common/common.h>
29 #include <common/kernel-ctl/kernel-ctl.h>
30 #include <common/kernel-ctl/kernel-ioctl.h>
31 #include <common/sessiond-comm/sessiond-comm.h>
35 #include "kernel-consumer.h"
36 #include "kern-modules.h"
40 * Add context on a kernel channel.
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
);
53 if (errno
!= EEXIST
) {
54 PERROR("add context ioctl");
56 /* If EEXIST, we just ignore the error */
62 cds_list_add_tail(&ctx
->list
, &chan
->ctx_list
);
71 * Create a new kernel session, register it to the kernel tracer and add it to
72 * the session daemon session.
74 int kernel_create_session(struct ltt_session
*session
, int tracer_fd
)
77 struct ltt_kernel_session
*lks
;
81 /* Allocate data structure */
82 lks
= trace_kernel_create_session();
88 /* Kernel tracer session creation */
89 ret
= kernctl_create_session(tracer_fd
);
91 PERROR("ioctl kernel create session");
96 /* Prevent fd duplication after execlp() */
97 ret
= fcntl(lks
->fd
, F_SETFD
, FD_CLOEXEC
);
99 PERROR("fcntl session fd");
102 lks
->id
= session
->id
;
103 lks
->consumer_fds_sent
= 0;
104 session
->kernel_session
= lks
;
106 DBG("Kernel session created (fd: %d)", lks
->fd
);
112 trace_kernel_destroy_session(lks
);
118 * Create a kernel channel, register it to the kernel tracer and add it to the
121 int kernel_create_channel(struct ltt_kernel_session
*session
,
122 struct lttng_channel
*chan
)
125 struct ltt_kernel_channel
*lkc
;
130 /* Allocate kernel channel */
131 lkc
= trace_kernel_create_channel(chan
);
136 DBG3("Kernel create channel %s with attr: %d, %" PRIu64
", %" PRIu64
", %u, %u, %d, %d",
137 chan
->name
, lkc
->channel
->attr
.overwrite
,
138 lkc
->channel
->attr
.subbuf_size
, lkc
->channel
->attr
.num_subbuf
,
139 lkc
->channel
->attr
.switch_timer_interval
, lkc
->channel
->attr
.read_timer_interval
,
140 lkc
->channel
->attr
.live_timer_interval
, lkc
->channel
->attr
.output
);
142 /* Kernel tracer channel creation */
143 ret
= kernctl_create_channel(session
->fd
, &lkc
->channel
->attr
);
145 PERROR("ioctl kernel create channel");
149 /* Setup the channel fd */
151 /* Prevent fd duplication after execlp() */
152 ret
= fcntl(lkc
->fd
, F_SETFD
, FD_CLOEXEC
);
154 PERROR("fcntl session fd");
157 /* Add channel to session */
158 cds_list_add(&lkc
->list
, &session
->channel_list
.head
);
159 session
->channel_count
++;
160 lkc
->session
= session
;
162 DBG("Kernel channel %s created (fd: %d)", lkc
->channel
->name
, lkc
->fd
);
175 * Create a kernel event, enable it to the kernel tracer and add it to the
176 * channel event list of the kernel session.
177 * We own filter_expression and filter.
179 int kernel_create_event(struct lttng_event
*ev
,
180 struct ltt_kernel_channel
*channel
)
183 struct ltt_kernel_event
*event
;
188 event
= trace_kernel_create_event(ev
);
194 ret
= kernctl_create_event(channel
->fd
, event
->event
);
200 WARN("Event type not implemented");
203 WARN("Event %s not found!", ev
->name
);
206 PERROR("create event ioctl");
213 * LTTNG_KERNEL_SYSCALL event creation will return 0 on success.
215 if (ret
== 0 && event
->event
->instrumentation
== LTTNG_KERNEL_SYSCALL
) {
216 DBG2("Kernel event syscall creation success");
218 * We use fd == -1 to ensure that we never trigger a close of fd
226 /* Prevent fd duplication after execlp() */
227 ret
= fcntl(event
->fd
, F_SETFD
, FD_CLOEXEC
);
229 PERROR("fcntl session fd");
233 /* Add event to event list */
234 cds_list_add(&event
->list
, &channel
->events_list
.head
);
235 channel
->event_count
++;
237 DBG("Event %s created (fd: %d)", ev
->name
, event
->fd
);
248 * Disable a kernel channel.
250 int kernel_disable_channel(struct ltt_kernel_channel
*chan
)
256 ret
= kernctl_disable(chan
->fd
);
258 PERROR("disable chan ioctl");
264 DBG("Kernel channel %s disabled (fd: %d)", chan
->channel
->name
, chan
->fd
);
273 * Enable a kernel channel.
275 int kernel_enable_channel(struct ltt_kernel_channel
*chan
)
281 ret
= kernctl_enable(chan
->fd
);
282 if (ret
< 0 && errno
!= EEXIST
) {
283 PERROR("Enable kernel chan");
288 DBG("Kernel channel %s enabled (fd: %d)", chan
->channel
->name
, chan
->fd
);
297 * Enable a kernel event.
299 int kernel_enable_event(struct ltt_kernel_event
*event
)
305 ret
= kernctl_enable(event
->fd
);
309 ret
= LTTNG_ERR_KERN_EVENT_EXIST
;
312 PERROR("enable kernel event");
319 DBG("Kernel event %s enabled (fd: %d)", event
->event
->name
, event
->fd
);
328 * Disable a kernel event.
330 int kernel_disable_event(struct ltt_kernel_event
*event
)
336 ret
= kernctl_disable(event
->fd
);
340 ret
= LTTNG_ERR_KERN_EVENT_EXIST
;
343 PERROR("disable kernel event");
350 DBG("Kernel event %s disabled (fd: %d)", event
->event
->name
, event
->fd
);
358 int kernel_enable_syscall(const char *syscall_name
,
359 struct ltt_kernel_channel
*channel
)
361 return kernctl_enable_syscall(channel
->fd
, syscall_name
);
364 int kernel_disable_syscall(const char *syscall_name
,
365 struct ltt_kernel_channel
*channel
)
367 return kernctl_disable_syscall(channel
->fd
, syscall_name
);
371 * Create kernel metadata, open from the kernel tracer and add it to the
374 int kernel_open_metadata(struct ltt_kernel_session
*session
)
377 struct ltt_kernel_metadata
*lkm
= NULL
;
381 /* Allocate kernel metadata */
382 lkm
= trace_kernel_create_metadata();
387 /* Kernel tracer metadata creation */
388 ret
= kernctl_open_metadata(session
->fd
, &lkm
->conf
->attr
);
394 /* Prevent fd duplication after execlp() */
395 ret
= fcntl(lkm
->fd
, F_SETFD
, FD_CLOEXEC
);
397 PERROR("fcntl session fd");
400 session
->metadata
= lkm
;
402 DBG("Kernel metadata opened (fd: %d)", lkm
->fd
);
407 trace_kernel_destroy_metadata(lkm
);
413 * Start tracing session.
415 int kernel_start_session(struct ltt_kernel_session
*session
)
421 ret
= kernctl_start_session(session
->fd
);
423 PERROR("ioctl start session");
427 DBG("Kernel session started");
436 * Make a kernel wait to make sure in-flight probe have completed.
438 void kernel_wait_quiescent(int fd
)
442 DBG("Kernel quiescent wait on %d", fd
);
444 ret
= kernctl_wait_quiescent(fd
);
446 PERROR("wait quiescent ioctl");
447 ERR("Kernel quiescent wait failed");
454 int kernel_calibrate(int fd
, struct lttng_kernel_calibrate
*calibrate
)
460 ret
= kernctl_calibrate(fd
, calibrate
);
462 PERROR("calibrate ioctl");
471 * Force flush buffer of metadata.
473 int kernel_metadata_flush_buffer(int fd
)
477 DBG("Kernel flushing metadata buffer on fd %d", fd
);
479 ret
= kernctl_buffer_flush(fd
);
481 ERR("Fail to flush metadata buffers %d (ret: %d)", fd
, ret
);
488 * Force flush buffer for channel.
490 int kernel_flush_buffer(struct ltt_kernel_channel
*channel
)
493 struct ltt_kernel_stream
*stream
;
497 DBG("Flush buffer for channel %s", channel
->channel
->name
);
499 cds_list_for_each_entry(stream
, &channel
->stream_list
.head
, list
) {
500 DBG("Flushing channel stream %d", stream
->fd
);
501 ret
= kernctl_buffer_flush(stream
->fd
);
504 ERR("Fail to flush buffer for stream %d (ret: %d)",
513 * Stop tracing session.
515 int kernel_stop_session(struct ltt_kernel_session
*session
)
521 ret
= kernctl_stop_session(session
->fd
);
526 DBG("Kernel session stopped");
535 * Open stream of channel, register it to the kernel tracer and add it
536 * to the stream list of the channel.
538 * Return the number of created stream. Else, a negative value.
540 int kernel_open_channel_stream(struct ltt_kernel_channel
*channel
)
543 struct ltt_kernel_stream
*lks
;
547 while ((ret
= kernctl_create_stream(channel
->fd
)) >= 0) {
548 lks
= trace_kernel_create_stream(channel
->channel
->name
, count
);
558 /* Prevent fd duplication after execlp() */
559 ret
= fcntl(lks
->fd
, F_SETFD
, FD_CLOEXEC
);
561 PERROR("fcntl session fd");
564 lks
->tracefile_size
= channel
->channel
->attr
.tracefile_size
;
565 lks
->tracefile_count
= channel
->channel
->attr
.tracefile_count
;
567 /* Add stream to channe stream list */
568 cds_list_add(&lks
->list
, &channel
->stream_list
.head
);
569 channel
->stream_count
++;
571 /* Increment counter which represent CPU number. */
574 DBG("Kernel stream %s created (fd: %d, state: %d)", lks
->name
, lks
->fd
,
578 return channel
->stream_count
;
585 * Open the metadata stream and set it to the kernel session.
587 int kernel_open_metadata_stream(struct ltt_kernel_session
*session
)
593 ret
= kernctl_create_stream(session
->metadata
->fd
);
595 PERROR("kernel create metadata stream");
599 DBG("Kernel metadata stream created (fd: %d)", ret
);
600 session
->metadata_stream_fd
= ret
;
601 /* Prevent fd duplication after execlp() */
602 ret
= fcntl(session
->metadata_stream_fd
, F_SETFD
, FD_CLOEXEC
);
604 PERROR("fcntl session fd");
614 * Get the event list from the kernel tracer and return the number of elements.
616 ssize_t
kernel_list_events(int tracer_fd
, struct lttng_event
**events
)
620 size_t nbmem
, count
= 0;
622 struct lttng_event
*elist
;
626 fd
= kernctl_tracepoint_list(tracer_fd
);
628 PERROR("kernel tracepoint list");
632 fp
= fdopen(fd
, "r");
634 PERROR("kernel tracepoint list fdopen");
639 * Init memory size counter
640 * See kernel-ctl.h for explanation of this value
642 nbmem
= KERNEL_EVENT_INIT_LIST_SIZE
;
643 elist
= zmalloc(sizeof(struct lttng_event
) * nbmem
);
645 PERROR("alloc list events");
650 while (fscanf(fp
, "event { name = %m[^;]; };\n", &event
) == 1) {
651 if (count
>= nbmem
) {
652 struct lttng_event
*new_elist
;
655 new_nbmem
= nbmem
<< 1;
656 DBG("Reallocating event list from %zu to %zu bytes",
658 new_elist
= realloc(elist
, new_nbmem
* sizeof(struct lttng_event
));
659 if (new_elist
== NULL
) {
660 PERROR("realloc list events");
666 /* Zero the new memory */
667 memset(new_elist
+ nbmem
, 0,
668 (new_nbmem
- nbmem
) * sizeof(struct lttng_event
));
672 strncpy(elist
[count
].name
, event
, LTTNG_SYMBOL_NAME_LEN
);
673 elist
[count
].name
[LTTNG_SYMBOL_NAME_LEN
- 1] = '\0';
674 elist
[count
].enabled
= -1;
680 DBG("Kernel list events done (%zu events)", count
);
682 ret
= fclose(fp
); /* closes both fp and fd */
698 * Get kernel version and validate it.
700 int kernel_validate_version(int tracer_fd
)
703 struct lttng_kernel_tracer_version version
;
704 struct lttng_kernel_tracer_abi_version abi_version
;
706 ret
= kernctl_tracer_version(tracer_fd
, &version
);
708 ERR("Failed at getting the lttng-modules version");
712 /* Validate version */
713 if (version
.major
!= VERSION_MAJOR
) {
714 ERR("Kernel tracer major version (%d) is not compatible with lttng-tools major version (%d)",
715 version
.major
, VERSION_MAJOR
);
718 ret
= kernctl_tracer_abi_version(tracer_fd
, &abi_version
);
720 ERR("Failed at getting lttng-modules ABI version");
723 if (abi_version
.major
!= LTTNG_MODULES_ABI_MAJOR_VERSION
) {
724 ERR("Kernel tracer ABI version (%d.%d) is not compatible with expected ABI major version (%d.*)",
725 abi_version
.major
, abi_version
.minor
,
726 LTTNG_MODULES_ABI_MAJOR_VERSION
);
729 DBG2("Kernel tracer version validated (%d.%d, ABI %d.%d)",
730 version
.major
, version
.minor
,
731 abi_version
.major
, abi_version
.minor
);
742 * Kernel work-arounds called at the start of sessiond main().
744 int init_kernel_workarounds(void)
750 * boot_id needs to be read once before being used concurrently
751 * to deal with a Linux kernel race. A fix is proposed for
752 * upstream, but the work-around is needed for older kernels.
754 fp
= fopen("/proc/sys/kernel/random/boot_id", "r");
761 ret
= fread(buf
, 1, sizeof(buf
), fp
);
763 /* Ignore error, we don't really care */
775 * Complete teardown of a kernel session.
777 void kernel_destroy_session(struct ltt_kernel_session
*ksess
)
780 DBG3("No kernel session when tearing down session");
784 DBG("Tearing down kernel session");
787 * Destroy channels on the consumer if at least one FD has been sent and we
788 * are in no output mode because the streams are in *no* monitor mode so we
789 * have to send a command to clean them up or else they leaked.
791 if (!ksess
->output_traces
&& ksess
->consumer_fds_sent
) {
793 struct consumer_socket
*socket
;
794 struct lttng_ht_iter iter
;
796 /* For each consumer socket. */
798 cds_lfht_for_each_entry(ksess
->consumer
->socks
->ht
, &iter
.iter
,
800 struct ltt_kernel_channel
*chan
;
802 /* For each channel, ask the consumer to destroy it. */
803 cds_list_for_each_entry(chan
, &ksess
->channel_list
.head
, list
) {
804 ret
= kernel_consumer_destroy_channel(socket
, chan
);
806 /* Consumer is probably dead. Use next socket. */
814 /* Close any relayd session */
815 consumer_output_send_destroy_relayd(ksess
->consumer
);
817 trace_kernel_destroy_session(ksess
);
821 * Destroy a kernel channel object. It does not do anything on the tracer side.
823 void kernel_destroy_channel(struct ltt_kernel_channel
*kchan
)
825 struct ltt_kernel_session
*ksess
= NULL
;
828 assert(kchan
->channel
);
830 DBG3("Kernel destroy channel %s", kchan
->channel
->name
);
832 /* Update channel count of associated session. */
833 if (kchan
->session
) {
834 /* Keep pointer reference so we can update it after the destroy. */
835 ksess
= kchan
->session
;
838 trace_kernel_destroy_channel(kchan
);
841 * At this point the kernel channel is not visible anymore. This is safe
842 * since in order to work on a visible kernel session, the tracing session
843 * lock (ltt_session.lock) MUST be acquired.
846 ksess
->channel_count
--;
851 * Take a snapshot for a given kernel session.
853 * Return 0 on success or else return a LTTNG_ERR code.
855 int kernel_snapshot_record(struct ltt_kernel_session
*ksess
,
856 struct snapshot_output
*output
, int wait
,
857 uint64_t nb_packets_per_stream
)
859 int err
, ret
, saved_metadata_fd
;
860 struct consumer_socket
*socket
;
861 struct lttng_ht_iter iter
;
862 struct ltt_kernel_metadata
*saved_metadata
;
865 assert(ksess
->consumer
);
868 DBG("Kernel snapshot record started");
870 /* Save current metadata since the following calls will change it. */
871 saved_metadata
= ksess
->metadata
;
872 saved_metadata_fd
= ksess
->metadata_stream_fd
;
876 ret
= kernel_open_metadata(ksess
);
878 ret
= LTTNG_ERR_KERN_META_FAIL
;
882 ret
= kernel_open_metadata_stream(ksess
);
884 ret
= LTTNG_ERR_KERN_META_FAIL
;
885 goto error_open_stream
;
888 /* Send metadata to consumer and snapshot everything. */
889 cds_lfht_for_each_entry(ksess
->consumer
->socks
->ht
, &iter
.iter
,
891 struct consumer_output
*saved_output
;
892 struct ltt_kernel_channel
*chan
;
895 * Temporarly switch consumer output for our snapshot output. As long
896 * as the session lock is taken, this is safe.
898 saved_output
= ksess
->consumer
;
899 ksess
->consumer
= output
->consumer
;
901 pthread_mutex_lock(socket
->lock
);
902 /* This stream must not be monitored by the consumer. */
903 ret
= kernel_consumer_add_metadata(socket
, ksess
, 0);
904 pthread_mutex_unlock(socket
->lock
);
905 /* Put back the saved consumer output into the session. */
906 ksess
->consumer
= saved_output
;
908 ret
= LTTNG_ERR_KERN_CONSUMER_FAIL
;
912 /* For each channel, ask the consumer to snapshot it. */
913 cds_list_for_each_entry(chan
, &ksess
->channel_list
.head
, list
) {
914 pthread_mutex_lock(socket
->lock
);
915 ret
= consumer_snapshot_channel(socket
, chan
->fd
, output
, 0,
916 ksess
->uid
, ksess
->gid
,
917 DEFAULT_KERNEL_TRACE_DIR
, wait
,
918 nb_packets_per_stream
);
919 pthread_mutex_unlock(socket
->lock
);
921 ret
= LTTNG_ERR_KERN_CONSUMER_FAIL
;
922 (void) kernel_consumer_destroy_metadata(socket
,
928 /* Snapshot metadata, */
929 pthread_mutex_lock(socket
->lock
);
930 ret
= consumer_snapshot_channel(socket
, ksess
->metadata
->fd
, output
,
931 1, ksess
->uid
, ksess
->gid
,
932 DEFAULT_KERNEL_TRACE_DIR
, wait
, 0);
933 pthread_mutex_unlock(socket
->lock
);
935 ret
= LTTNG_ERR_KERN_CONSUMER_FAIL
;
940 * The metadata snapshot is done, ask the consumer to destroy it since
941 * it's not monitored on the consumer side.
943 (void) kernel_consumer_destroy_metadata(socket
, ksess
->metadata
);
949 /* Close newly opened metadata stream. It's now on the consumer side. */
950 err
= close(ksess
->metadata_stream_fd
);
952 PERROR("close snapshot kernel");
956 trace_kernel_destroy_metadata(ksess
->metadata
);
958 /* Restore metadata state.*/
959 ksess
->metadata
= saved_metadata
;
960 ksess
->metadata_stream_fd
= saved_metadata_fd
;
967 * Get the syscall mask array from the kernel tracer.
969 * Return 0 on success else a negative value. In both case, syscall_mask should
972 int kernel_syscall_mask(int chan_fd
, char **syscall_mask
, uint32_t *nr_bits
)
974 assert(syscall_mask
);
977 return kernctl_syscall_mask(chan_fd
, syscall_mask
, nr_bits
);