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.
25 #include <sys/types.h>
27 #include <common/common.h>
28 #include <common/trace-chunk.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>
33 #include "lttng-sessiond.h"
34 #include "lttng-syscall.h"
37 #include "kernel-consumer.h"
38 #include "kern-modules.h"
44 * Key used to reference a channel between the sessiond and the consumer. This
45 * is only read and updated with the session_list lock held.
47 static uint64_t next_kernel_channel_key
;
49 static const char *module_proc_lttng
= "/proc/lttng";
51 static int kernel_tracer_fd
= -1;
53 #include <lttng/userspace-probe.h>
54 #include <lttng/userspace-probe-internal.h>
56 * Add context on a kernel channel.
58 * Assumes the ownership of ctx.
60 int kernel_add_channel_context(struct ltt_kernel_channel
*chan
,
61 struct ltt_kernel_context
*ctx
)
68 DBG("Adding context to channel %s", chan
->channel
->name
);
69 ret
= kernctl_add_context(chan
->fd
, &ctx
->ctx
);
73 /* Exists but not available for this kernel */
74 ret
= LTTNG_ERR_KERN_CONTEXT_UNAVAILABLE
;
77 /* If EEXIST, we just ignore the error */
81 PERROR("add context ioctl");
82 ret
= LTTNG_ERR_KERN_CONTEXT_FAIL
;
89 cds_list_add_tail(&ctx
->list
, &chan
->ctx_list
);
94 trace_kernel_destroy_context(ctx
);
100 * Create a new kernel session, register it to the kernel tracer and add it to
101 * the session daemon session.
103 int kernel_create_session(struct ltt_session
*session
)
106 struct ltt_kernel_session
*lks
;
110 /* Allocate data structure */
111 lks
= trace_kernel_create_session();
117 /* Kernel tracer session creation */
118 ret
= kernctl_create_session(kernel_tracer_fd
);
120 PERROR("ioctl kernel create session");
125 /* Prevent fd duplication after execlp() */
126 ret
= fcntl(lks
->fd
, F_SETFD
, FD_CLOEXEC
);
128 PERROR("fcntl session fd");
131 lks
->id
= session
->id
;
132 lks
->consumer_fds_sent
= 0;
133 session
->kernel_session
= lks
;
135 DBG("Kernel session created (fd: %d)", lks
->fd
);
141 trace_kernel_destroy_session(lks
);
142 trace_kernel_free_session(lks
);
148 * Create a kernel channel, register it to the kernel tracer and add it to the
151 int kernel_create_channel(struct ltt_kernel_session
*session
,
152 struct lttng_channel
*chan
)
155 struct ltt_kernel_channel
*lkc
;
160 /* Allocate kernel channel */
161 lkc
= trace_kernel_create_channel(chan
);
166 DBG3("Kernel create channel %s with attr: %d, %" PRIu64
", %" PRIu64
", %u, %u, %d, %d",
167 chan
->name
, lkc
->channel
->attr
.overwrite
,
168 lkc
->channel
->attr
.subbuf_size
, lkc
->channel
->attr
.num_subbuf
,
169 lkc
->channel
->attr
.switch_timer_interval
, lkc
->channel
->attr
.read_timer_interval
,
170 lkc
->channel
->attr
.live_timer_interval
, lkc
->channel
->attr
.output
);
172 /* Kernel tracer channel creation */
173 ret
= kernctl_create_channel(session
->fd
, &lkc
->channel
->attr
);
175 PERROR("ioctl kernel create channel");
179 /* Setup the channel fd */
181 /* Prevent fd duplication after execlp() */
182 ret
= fcntl(lkc
->fd
, F_SETFD
, FD_CLOEXEC
);
184 PERROR("fcntl session fd");
187 /* Add channel to session */
188 cds_list_add(&lkc
->list
, &session
->channel_list
.head
);
189 session
->channel_count
++;
190 lkc
->session
= session
;
191 lkc
->key
= ++next_kernel_channel_key
;
193 DBG("Kernel channel %s created (fd: %d, key: %" PRIu64
")",
194 lkc
->channel
->name
, lkc
->fd
, lkc
->key
);
207 * Compute the offset of the instrumentation byte in the binary based on the
208 * function probe location using the ELF lookup method.
210 * Returns 0 on success and set the offset out parameter to the offset of the
212 * Returns -1 on error
215 int extract_userspace_probe_offset_function_elf(
216 const struct lttng_userspace_probe_location
*probe_location
,
217 struct ltt_kernel_session
*session
, uint64_t *offset
)
221 const char *symbol
= NULL
;
222 const struct lttng_userspace_probe_location_lookup_method
*lookup
= NULL
;
223 enum lttng_userspace_probe_location_lookup_method_type lookup_method_type
;
225 assert(lttng_userspace_probe_location_get_type(probe_location
) ==
226 LTTNG_USERSPACE_PROBE_LOCATION_TYPE_FUNCTION
);
228 lookup
= lttng_userspace_probe_location_get_lookup_method(
236 lttng_userspace_probe_location_lookup_method_get_type(lookup
);
238 assert(lookup_method_type
==
239 LTTNG_USERSPACE_PROBE_LOCATION_LOOKUP_METHOD_TYPE_FUNCTION_ELF
);
241 symbol
= lttng_userspace_probe_location_function_get_function_name(
248 fd
= lttng_userspace_probe_location_function_get_binary_fd(probe_location
);
254 ret
= run_as_extract_elf_symbol_offset(fd
, symbol
, session
->uid
,
255 session
->gid
, offset
);
257 DBG("userspace probe offset calculation failed for "
258 "function %s", symbol
);
262 DBG("userspace probe elf offset for %s is 0x%jd", symbol
, (intmax_t)(*offset
));
268 * Compute the offsets of the instrumentation bytes in the binary based on the
269 * tracepoint probe location using the SDT lookup method. This function
270 * allocates the offsets buffer, the caller must free it.
272 * Returns 0 on success and set the offset out parameter to the offsets of the
274 * Returns -1 on error.
277 int extract_userspace_probe_offset_tracepoint_sdt(
278 const struct lttng_userspace_probe_location
*probe_location
,
279 struct ltt_kernel_session
*session
, uint64_t **offsets
,
280 uint32_t *offsets_count
)
282 enum lttng_userspace_probe_location_lookup_method_type lookup_method_type
;
283 const struct lttng_userspace_probe_location_lookup_method
*lookup
= NULL
;
284 const char *probe_name
= NULL
, *provider_name
= NULL
;
288 assert(lttng_userspace_probe_location_get_type(probe_location
) ==
289 LTTNG_USERSPACE_PROBE_LOCATION_TYPE_TRACEPOINT
);
291 lookup
= lttng_userspace_probe_location_get_lookup_method(probe_location
);
298 lttng_userspace_probe_location_lookup_method_get_type(lookup
);
300 assert(lookup_method_type
==
301 LTTNG_USERSPACE_PROBE_LOCATION_LOOKUP_METHOD_TYPE_TRACEPOINT_SDT
);
304 probe_name
= lttng_userspace_probe_location_tracepoint_get_probe_name(
311 provider_name
= lttng_userspace_probe_location_tracepoint_get_provider_name(
313 if (!provider_name
) {
318 fd
= lttng_userspace_probe_location_tracepoint_get_binary_fd(probe_location
);
324 ret
= run_as_extract_sdt_probe_offsets(fd
, provider_name
, probe_name
,
325 session
->uid
, session
->gid
, offsets
, offsets_count
);
327 DBG("userspace probe offset calculation failed for sdt "
328 "probe %s:%s", provider_name
, probe_name
);
332 if (*offsets_count
== 0) {
333 DBG("no userspace probe offset found");
337 DBG("%u userspace probe SDT offsets found for %s:%s at:",
338 *offsets_count
, provider_name
, probe_name
);
339 for (i
= 0; i
< *offsets_count
; i
++) {
340 DBG("\t0x%jd", (intmax_t)((*offsets
)[i
]));
347 * Extract the offsets of the instrumentation point for the different lookup
351 int userspace_probe_add_callsites(struct lttng_event
*ev
,
352 struct ltt_kernel_session
*session
, int fd
)
354 const struct lttng_userspace_probe_location_lookup_method
*lookup_method
= NULL
;
355 enum lttng_userspace_probe_location_lookup_method_type type
;
356 const struct lttng_userspace_probe_location
*location
= NULL
;
360 assert(ev
->type
== LTTNG_EVENT_USERSPACE_PROBE
);
362 location
= lttng_event_get_userspace_probe_location(ev
);
368 lttng_userspace_probe_location_get_lookup_method(location
);
369 if (!lookup_method
) {
374 type
= lttng_userspace_probe_location_lookup_method_get_type(lookup_method
);
376 case LTTNG_USERSPACE_PROBE_LOCATION_LOOKUP_METHOD_TYPE_FUNCTION_ELF
:
378 struct lttng_kernel_event_callsite callsite
;
381 ret
= extract_userspace_probe_offset_function_elf(location
, session
, &offset
);
383 ret
= LTTNG_ERR_PROBE_LOCATION_INVAL
;
387 callsite
.u
.uprobe
.offset
= offset
;
388 ret
= kernctl_add_callsite(fd
, &callsite
);
390 WARN("Adding callsite to userspace probe "
391 "event %s failed.", ev
->name
);
392 ret
= LTTNG_ERR_KERN_ENABLE_FAIL
;
397 case LTTNG_USERSPACE_PROBE_LOCATION_LOOKUP_METHOD_TYPE_TRACEPOINT_SDT
:
400 uint64_t *offsets
= NULL
;
401 uint32_t offsets_count
;
402 struct lttng_kernel_event_callsite callsite
;
405 * This call allocates the offsets buffer. This buffer must be freed
408 ret
= extract_userspace_probe_offset_tracepoint_sdt(location
, session
,
409 &offsets
, &offsets_count
);
411 ret
= LTTNG_ERR_PROBE_LOCATION_INVAL
;
414 for (i
= 0; i
< offsets_count
; i
++) {
415 callsite
.u
.uprobe
.offset
= offsets
[i
];
416 ret
= kernctl_add_callsite(fd
, &callsite
);
418 WARN("Adding callsite to userspace probe "
419 "event %s failed.", ev
->name
);
420 ret
= LTTNG_ERR_KERN_ENABLE_FAIL
;
429 ret
= LTTNG_ERR_PROBE_LOCATION_INVAL
;
437 * Create a kernel event, enable it to the kernel tracer and add it to the
438 * channel event list of the kernel session.
439 * We own filter_expression and filter.
441 int kernel_create_event(struct lttng_event
*ev
,
442 struct ltt_kernel_channel
*channel
,
443 char *filter_expression
,
444 struct lttng_filter_bytecode
*filter
)
447 enum lttng_error_code ret
;
448 struct ltt_kernel_event
*event
;
453 /* We pass ownership of filter_expression and filter */
454 ret
= trace_kernel_create_event(ev
, filter_expression
,
456 if (ret
!= LTTNG_OK
) {
460 fd
= kernctl_create_event(channel
->fd
, event
->event
);
464 ret
= LTTNG_ERR_KERN_EVENT_EXIST
;
467 WARN("Event type not implemented");
468 ret
= LTTNG_ERR_KERN_EVENT_ENOSYS
;
471 WARN("Event %s not found!", ev
->name
);
472 ret
= LTTNG_ERR_KERN_ENABLE_FAIL
;
475 ret
= LTTNG_ERR_KERN_ENABLE_FAIL
;
476 PERROR("create event ioctl");
481 event
->type
= ev
->type
;
483 /* Prevent fd duplication after execlp() */
484 err
= fcntl(event
->fd
, F_SETFD
, FD_CLOEXEC
);
486 PERROR("fcntl session fd");
490 err
= kernctl_filter(event
->fd
, filter
);
494 ret
= LTTNG_ERR_FILTER_NOMEM
;
497 ret
= LTTNG_ERR_FILTER_INVAL
;
504 if (ev
->type
== LTTNG_EVENT_USERSPACE_PROBE
) {
505 ret
= userspace_probe_add_callsites(ev
, channel
->session
, event
->fd
);
507 goto add_callsite_error
;
511 err
= kernctl_enable(event
->fd
);
515 ret
= LTTNG_ERR_KERN_EVENT_EXIST
;
518 PERROR("enable kernel event");
519 ret
= LTTNG_ERR_KERN_ENABLE_FAIL
;
525 /* Add event to event list */
526 cds_list_add(&event
->list
, &channel
->events_list
.head
);
527 channel
->event_count
++;
529 DBG("Event %s created (fd: %d)", ev
->name
, event
->fd
);
539 closeret
= close(event
->fd
);
541 PERROR("close event fd");
551 * Disable a kernel channel.
553 int kernel_disable_channel(struct ltt_kernel_channel
*chan
)
559 ret
= kernctl_disable(chan
->fd
);
561 PERROR("disable chan ioctl");
566 DBG("Kernel channel %s disabled (fd: %d, key: %" PRIu64
")",
567 chan
->channel
->name
, chan
->fd
, chan
->key
);
576 * Enable a kernel channel.
578 int kernel_enable_channel(struct ltt_kernel_channel
*chan
)
584 ret
= kernctl_enable(chan
->fd
);
585 if (ret
< 0 && ret
!= -EEXIST
) {
586 PERROR("Enable kernel chan");
591 DBG("Kernel channel %s enabled (fd: %d, key: %" PRIu64
")",
592 chan
->channel
->name
, chan
->fd
, chan
->key
);
601 * Enable a kernel event.
603 int kernel_enable_event(struct ltt_kernel_event
*event
)
609 ret
= kernctl_enable(event
->fd
);
613 ret
= LTTNG_ERR_KERN_EVENT_EXIST
;
616 PERROR("enable kernel event");
623 DBG("Kernel event %s enabled (fd: %d)", event
->event
->name
, event
->fd
);
632 * Disable a kernel event.
634 int kernel_disable_event(struct ltt_kernel_event
*event
)
640 ret
= kernctl_disable(event
->fd
);
644 ret
= LTTNG_ERR_KERN_EVENT_EXIST
;
647 PERROR("disable kernel event");
654 DBG("Kernel event %s disabled (fd: %d)", event
->event
->name
, event
->fd
);
663 int kernel_track_pid(struct ltt_kernel_session
*session
, int pid
)
667 DBG("Kernel track PID %d for session id %" PRIu64
".",
669 ret
= kernctl_track_pid(session
->fd
, pid
);
675 return LTTNG_ERR_INVALID
;
677 return LTTNG_ERR_NOMEM
;
679 return LTTNG_ERR_PID_TRACKED
;
681 return LTTNG_ERR_UNK
;
685 int kernel_untrack_pid(struct ltt_kernel_session
*session
, int pid
)
689 DBG("Kernel untrack PID %d for session id %" PRIu64
".",
691 ret
= kernctl_untrack_pid(session
->fd
, pid
);
697 return LTTNG_ERR_INVALID
;
699 return LTTNG_ERR_NOMEM
;
701 return LTTNG_ERR_PID_NOT_TRACKED
;
703 return LTTNG_ERR_UNK
;
707 ssize_t
kernel_list_tracker_pids(struct ltt_kernel_session
*session
,
712 ssize_t nbmem
, count
= 0;
716 fd
= kernctl_list_tracker_pids(session
->fd
);
718 PERROR("kernel tracker pids list");
722 fp
= fdopen(fd
, "r");
724 PERROR("kernel tracker pids list fdopen");
728 nbmem
= KERNEL_TRACKER_PIDS_INIT_LIST_SIZE
;
729 pids
= zmalloc(sizeof(*pids
) * nbmem
);
731 PERROR("alloc list pids");
736 while (fscanf(fp
, "process { pid = %u; };\n", &pid
) == 1) {
737 if (count
>= nbmem
) {
741 new_nbmem
= nbmem
<< 1;
742 DBG("Reallocating pids list from %zu to %zu entries",
744 new_pids
= realloc(pids
, new_nbmem
* sizeof(*new_pids
));
745 if (new_pids
== NULL
) {
746 PERROR("realloc list events");
751 /* Zero the new memory */
752 memset(new_pids
+ nbmem
, 0,
753 (new_nbmem
- nbmem
) * sizeof(*new_pids
));
761 DBG("Kernel list tracker pids done (%zd pids)", count
);
763 ret
= fclose(fp
); /* closes both fp and fd */
779 * Create kernel metadata, open from the kernel tracer and add it to the
782 int kernel_open_metadata(struct ltt_kernel_session
*session
)
785 struct ltt_kernel_metadata
*lkm
= NULL
;
789 /* Allocate kernel metadata */
790 lkm
= trace_kernel_create_metadata();
795 /* Kernel tracer metadata creation */
796 ret
= kernctl_open_metadata(session
->fd
, &lkm
->conf
->attr
);
802 lkm
->key
= ++next_kernel_channel_key
;
803 /* Prevent fd duplication after execlp() */
804 ret
= fcntl(lkm
->fd
, F_SETFD
, FD_CLOEXEC
);
806 PERROR("fcntl session fd");
809 session
->metadata
= lkm
;
811 DBG("Kernel metadata opened (fd: %d)", lkm
->fd
);
816 trace_kernel_destroy_metadata(lkm
);
822 * Start tracing session.
824 int kernel_start_session(struct ltt_kernel_session
*session
)
830 ret
= kernctl_start_session(session
->fd
);
832 PERROR("ioctl start session");
836 DBG("Kernel session started");
845 * Make a kernel wait to make sure in-flight probe have completed.
847 void kernel_wait_quiescent(void)
850 int fd
= kernel_tracer_fd
;
852 DBG("Kernel quiescent wait on %d", fd
);
854 ret
= kernctl_wait_quiescent(fd
);
856 PERROR("wait quiescent ioctl");
857 ERR("Kernel quiescent wait failed");
862 * Force flush buffer of metadata.
864 int kernel_metadata_flush_buffer(int fd
)
868 DBG("Kernel flushing metadata buffer on fd %d", fd
);
870 ret
= kernctl_buffer_flush(fd
);
872 ERR("Fail to flush metadata buffers %d (ret: %d)", fd
, ret
);
879 * Force flush buffer for channel.
881 int kernel_flush_buffer(struct ltt_kernel_channel
*channel
)
884 struct ltt_kernel_stream
*stream
;
888 DBG("Flush buffer for channel %s", channel
->channel
->name
);
890 cds_list_for_each_entry(stream
, &channel
->stream_list
.head
, list
) {
891 DBG("Flushing channel stream %d", stream
->fd
);
892 ret
= kernctl_buffer_flush(stream
->fd
);
895 ERR("Fail to flush buffer for stream %d (ret: %d)",
904 * Stop tracing session.
906 int kernel_stop_session(struct ltt_kernel_session
*session
)
912 ret
= kernctl_stop_session(session
->fd
);
917 DBG("Kernel session stopped");
926 * Open stream of channel, register it to the kernel tracer and add it
927 * to the stream list of the channel.
929 * Note: given that the streams may appear in random order wrt CPU
930 * number (e.g. cpu hotplug), the index value of the stream number in
931 * the stream name is not necessarily linked to the CPU number.
933 * Return the number of created stream. Else, a negative value.
935 int kernel_open_channel_stream(struct ltt_kernel_channel
*channel
)
938 struct ltt_kernel_stream
*lks
;
942 while ((ret
= kernctl_create_stream(channel
->fd
)) >= 0) {
943 lks
= trace_kernel_create_stream(channel
->channel
->name
,
944 channel
->stream_count
);
954 /* Prevent fd duplication after execlp() */
955 ret
= fcntl(lks
->fd
, F_SETFD
, FD_CLOEXEC
);
957 PERROR("fcntl session fd");
960 lks
->tracefile_size
= channel
->channel
->attr
.tracefile_size
;
961 lks
->tracefile_count
= channel
->channel
->attr
.tracefile_count
;
963 /* Add stream to channel stream list */
964 cds_list_add(&lks
->list
, &channel
->stream_list
.head
);
965 channel
->stream_count
++;
967 DBG("Kernel stream %s created (fd: %d, state: %d)", lks
->name
, lks
->fd
,
971 return channel
->stream_count
;
978 * Open the metadata stream and set it to the kernel session.
980 int kernel_open_metadata_stream(struct ltt_kernel_session
*session
)
986 ret
= kernctl_create_stream(session
->metadata
->fd
);
988 PERROR("kernel create metadata stream");
992 DBG("Kernel metadata stream created (fd: %d)", ret
);
993 session
->metadata_stream_fd
= ret
;
994 /* Prevent fd duplication after execlp() */
995 ret
= fcntl(session
->metadata_stream_fd
, F_SETFD
, FD_CLOEXEC
);
997 PERROR("fcntl session fd");
1007 * Get the event list from the kernel tracer and return the number of elements.
1009 ssize_t
kernel_list_events(struct lttng_event
**events
)
1013 size_t nbmem
, count
= 0;
1015 struct lttng_event
*elist
;
1019 fd
= kernctl_tracepoint_list(kernel_tracer_fd
);
1021 PERROR("kernel tracepoint list");
1025 fp
= fdopen(fd
, "r");
1027 PERROR("kernel tracepoint list fdopen");
1032 * Init memory size counter
1033 * See kernel-ctl.h for explanation of this value
1035 nbmem
= KERNEL_EVENT_INIT_LIST_SIZE
;
1036 elist
= zmalloc(sizeof(struct lttng_event
) * nbmem
);
1037 if (elist
== NULL
) {
1038 PERROR("alloc list events");
1043 while (fscanf(fp
, "event { name = %m[^;]; };\n", &event
) == 1) {
1044 if (count
>= nbmem
) {
1045 struct lttng_event
*new_elist
;
1048 new_nbmem
= nbmem
<< 1;
1049 DBG("Reallocating event list from %zu to %zu bytes",
1051 new_elist
= realloc(elist
, new_nbmem
* sizeof(struct lttng_event
));
1052 if (new_elist
== NULL
) {
1053 PERROR("realloc list events");
1059 /* Zero the new memory */
1060 memset(new_elist
+ nbmem
, 0,
1061 (new_nbmem
- nbmem
) * sizeof(struct lttng_event
));
1065 strncpy(elist
[count
].name
, event
, LTTNG_SYMBOL_NAME_LEN
);
1066 elist
[count
].name
[LTTNG_SYMBOL_NAME_LEN
- 1] = '\0';
1067 elist
[count
].enabled
= -1;
1073 DBG("Kernel list events done (%zu events)", count
);
1075 ret
= fclose(fp
); /* closes both fp and fd */
1091 * Get kernel version and validate it.
1093 int kernel_validate_version(struct lttng_kernel_tracer_version
*version
,
1094 struct lttng_kernel_tracer_abi_version
*abi_version
)
1098 ret
= kernctl_tracer_version(kernel_tracer_fd
, version
);
1100 ERR("Failed to retrieve the lttng-modules version");
1104 /* Validate version */
1105 if (version
->major
!= VERSION_MAJOR
) {
1106 ERR("Kernel tracer major version (%d) is not compatible with lttng-tools major version (%d)",
1107 version
->major
, VERSION_MAJOR
);
1110 ret
= kernctl_tracer_abi_version(kernel_tracer_fd
, abi_version
);
1112 ERR("Failed to retrieve lttng-modules ABI version");
1115 if (abi_version
->major
!= LTTNG_MODULES_ABI_MAJOR_VERSION
) {
1116 ERR("Kernel tracer ABI version (%d.%d) does not match the expected ABI major version (%d.*)",
1117 abi_version
->major
, abi_version
->minor
,
1118 LTTNG_MODULES_ABI_MAJOR_VERSION
);
1121 DBG2("Kernel tracer version validated (%d.%d, ABI %d.%d)",
1122 version
->major
, version
->minor
,
1123 abi_version
->major
, abi_version
->minor
);
1130 ERR("Kernel tracer version check failed; kernel tracing will not be available");
1135 * Kernel work-arounds called at the start of sessiond main().
1137 int init_kernel_workarounds(void)
1143 * boot_id needs to be read once before being used concurrently
1144 * to deal with a Linux kernel race. A fix is proposed for
1145 * upstream, but the work-around is needed for older kernels.
1147 fp
= fopen("/proc/sys/kernel/random/boot_id", "r");
1154 ret
= fread(buf
, 1, sizeof(buf
), fp
);
1156 /* Ignore error, we don't really care */
1168 * Teardown of a kernel session, keeping data required by destroy notifiers.
1170 void kernel_destroy_session(struct ltt_kernel_session
*ksess
)
1172 struct lttng_trace_chunk
*trace_chunk
;
1174 if (ksess
== NULL
) {
1175 DBG3("No kernel session when tearing down session");
1179 DBG("Tearing down kernel session");
1180 trace_chunk
= ksess
->current_trace_chunk
;
1183 * Destroy channels on the consumer if at least one FD has been sent and we
1184 * are in no output mode because the streams are in *no* monitor mode so we
1185 * have to send a command to clean them up or else they leaked.
1187 if (!ksess
->output_traces
&& ksess
->consumer_fds_sent
) {
1189 struct consumer_socket
*socket
;
1190 struct lttng_ht_iter iter
;
1192 /* For each consumer socket. */
1194 cds_lfht_for_each_entry(ksess
->consumer
->socks
->ht
, &iter
.iter
,
1195 socket
, node
.node
) {
1196 struct ltt_kernel_channel
*chan
;
1198 /* For each channel, ask the consumer to destroy it. */
1199 cds_list_for_each_entry(chan
, &ksess
->channel_list
.head
, list
) {
1200 ret
= kernel_consumer_destroy_channel(socket
, chan
);
1202 /* Consumer is probably dead. Use next socket. */
1210 /* Close any relayd session */
1211 consumer_output_send_destroy_relayd(ksess
->consumer
);
1213 trace_kernel_destroy_session(ksess
);
1214 lttng_trace_chunk_put(trace_chunk
);
1217 /* Teardown of data required by destroy notifiers. */
1218 void kernel_free_session(struct ltt_kernel_session
*ksess
)
1220 if (ksess
== NULL
) {
1223 trace_kernel_free_session(ksess
);
1227 * Destroy a kernel channel object. It does not do anything on the tracer side.
1229 void kernel_destroy_channel(struct ltt_kernel_channel
*kchan
)
1231 struct ltt_kernel_session
*ksess
= NULL
;
1234 assert(kchan
->channel
);
1236 DBG3("Kernel destroy channel %s", kchan
->channel
->name
);
1238 /* Update channel count of associated session. */
1239 if (kchan
->session
) {
1240 /* Keep pointer reference so we can update it after the destroy. */
1241 ksess
= kchan
->session
;
1244 trace_kernel_destroy_channel(kchan
);
1247 * At this point the kernel channel is not visible anymore. This is safe
1248 * since in order to work on a visible kernel session, the tracing session
1249 * lock (ltt_session.lock) MUST be acquired.
1252 ksess
->channel_count
--;
1257 * Take a snapshot for a given kernel session.
1259 * Return LTTNG_OK on success or else return a LTTNG_ERR code.
1261 enum lttng_error_code
kernel_snapshot_record(
1262 struct ltt_kernel_session
*ksess
,
1263 const struct consumer_output
*output
, int wait
,
1264 uint64_t nb_packets_per_stream
)
1266 int err
, ret
, saved_metadata_fd
;
1267 enum lttng_error_code status
= LTTNG_OK
;
1268 struct consumer_socket
*socket
;
1269 struct lttng_ht_iter iter
;
1270 struct ltt_kernel_metadata
*saved_metadata
;
1271 char *trace_path
= NULL
;
1274 assert(ksess
->consumer
);
1277 DBG("Kernel snapshot record started");
1279 /* Save current metadata since the following calls will change it. */
1280 saved_metadata
= ksess
->metadata
;
1281 saved_metadata_fd
= ksess
->metadata_stream_fd
;
1285 ret
= kernel_open_metadata(ksess
);
1287 status
= LTTNG_ERR_KERN_META_FAIL
;
1291 ret
= kernel_open_metadata_stream(ksess
);
1293 status
= LTTNG_ERR_KERN_META_FAIL
;
1294 goto error_open_stream
;
1297 trace_path
= setup_channel_trace_path(ksess
->consumer
,
1298 DEFAULT_KERNEL_TRACE_DIR
);
1300 status
= LTTNG_ERR_INVALID
;
1303 /* Send metadata to consumer and snapshot everything. */
1304 cds_lfht_for_each_entry(output
->socks
->ht
, &iter
.iter
,
1305 socket
, node
.node
) {
1306 struct ltt_kernel_channel
*chan
;
1308 pthread_mutex_lock(socket
->lock
);
1309 /* This stream must not be monitored by the consumer. */
1310 ret
= kernel_consumer_add_metadata(socket
, ksess
, 0);
1311 pthread_mutex_unlock(socket
->lock
);
1313 status
= LTTNG_ERR_KERN_META_FAIL
;
1314 goto error_consumer
;
1317 /* For each channel, ask the consumer to snapshot it. */
1318 cds_list_for_each_entry(chan
, &ksess
->channel_list
.head
, list
) {
1319 status
= consumer_snapshot_channel(socket
, chan
->key
, output
, 0,
1320 ksess
->uid
, ksess
->gid
,
1322 nb_packets_per_stream
);
1323 if (status
!= LTTNG_OK
) {
1324 (void) kernel_consumer_destroy_metadata(socket
,
1326 goto error_consumer
;
1330 /* Snapshot metadata, */
1331 status
= consumer_snapshot_channel(socket
, ksess
->metadata
->key
, output
,
1332 1, ksess
->uid
, ksess
->gid
, trace_path
, wait
, 0);
1333 if (status
!= LTTNG_OK
) {
1334 goto error_consumer
;
1338 * The metadata snapshot is done, ask the consumer to destroy it since
1339 * it's not monitored on the consumer side.
1341 (void) kernel_consumer_destroy_metadata(socket
, ksess
->metadata
);
1345 /* Close newly opened metadata stream. It's now on the consumer side. */
1346 err
= close(ksess
->metadata_stream_fd
);
1348 PERROR("close snapshot kernel");
1352 trace_kernel_destroy_metadata(ksess
->metadata
);
1354 /* Restore metadata state.*/
1355 ksess
->metadata
= saved_metadata
;
1356 ksess
->metadata_stream_fd
= saved_metadata_fd
;
1363 * Get the syscall mask array from the kernel tracer.
1365 * Return 0 on success else a negative value. In both case, syscall_mask should
1368 int kernel_syscall_mask(int chan_fd
, char **syscall_mask
, uint32_t *nr_bits
)
1370 assert(syscall_mask
);
1373 return kernctl_syscall_mask(chan_fd
, syscall_mask
, nr_bits
);
1377 * Check for the support of the RING_BUFFER_SNAPSHOT_SAMPLE_POSITIONS via abi
1380 * Return 1 on success, 0 when feature is not supported, negative value in case
1383 int kernel_supports_ring_buffer_snapshot_sample_positions(void)
1385 int ret
= 0; // Not supported by default
1386 struct lttng_kernel_tracer_abi_version abi
;
1388 ret
= kernctl_tracer_abi_version(kernel_tracer_fd
, &abi
);
1390 ERR("Failed to retrieve lttng-modules ABI version");
1395 * RING_BUFFER_SNAPSHOT_SAMPLE_POSITIONS was introduced in 2.3
1397 if (abi
.major
>= 2 && abi
.minor
>= 3) {
1409 * Rotate a kernel session.
1411 * Return LTTNG_OK on success or else an LTTng error code.
1413 enum lttng_error_code
kernel_rotate_session(struct ltt_session
*session
)
1416 enum lttng_error_code status
= LTTNG_OK
;
1417 struct consumer_socket
*socket
;
1418 struct lttng_ht_iter iter
;
1419 struct ltt_kernel_session
*ksess
= session
->kernel_session
;
1422 assert(ksess
->consumer
);
1424 DBG("Rotate kernel session %s started (session %" PRIu64
")",
1425 session
->name
, session
->id
);
1430 * Note that this loop will end after one iteration given that there is
1431 * only one kernel consumer.
1433 cds_lfht_for_each_entry(ksess
->consumer
->socks
->ht
, &iter
.iter
,
1434 socket
, node
.node
) {
1435 struct ltt_kernel_channel
*chan
;
1437 /* For each channel, ask the consumer to rotate it. */
1438 cds_list_for_each_entry(chan
, &ksess
->channel_list
.head
, list
) {
1439 DBG("Rotate kernel channel %" PRIu64
", session %s",
1440 chan
->key
, session
->name
);
1441 ret
= consumer_rotate_channel(socket
, chan
->key
,
1442 ksess
->uid
, ksess
->gid
, ksess
->consumer
,
1443 /* is_metadata_channel */ false);
1445 status
= LTTNG_ERR_KERN_CONSUMER_FAIL
;
1451 * Rotate the metadata channel.
1453 ret
= consumer_rotate_channel(socket
, ksess
->metadata
->key
,
1454 ksess
->uid
, ksess
->gid
, ksess
->consumer
,
1455 /* is_metadata_channel */ true);
1457 status
= LTTNG_ERR_KERN_CONSUMER_FAIL
;
1467 enum lttng_error_code
kernel_create_channel_subdirectories(
1468 const struct ltt_kernel_session
*ksess
)
1470 enum lttng_error_code ret
= LTTNG_OK
;
1471 enum lttng_trace_chunk_status chunk_status
;
1474 assert(ksess
->current_trace_chunk
);
1477 * Create the index subdirectory which will take care
1478 * of implicitly creating the channel's path.
1480 chunk_status
= lttng_trace_chunk_create_subdirectory(
1481 ksess
->current_trace_chunk
,
1482 DEFAULT_KERNEL_TRACE_DIR
"/" DEFAULT_INDEX_DIR
);
1483 if (chunk_status
!= LTTNG_TRACE_CHUNK_STATUS_OK
) {
1484 ret
= LTTNG_ERR_CREATE_DIR_FAIL
;
1493 * Setup necessary data for kernel tracer action.
1496 int init_kernel_tracer(void)
1499 bool is_root
= !getuid();
1501 /* Modprobe lttng kernel modules */
1502 ret
= modprobe_lttng_control();
1507 /* Open debugfs lttng */
1508 kernel_tracer_fd
= open(module_proc_lttng
, O_RDWR
);
1509 if (kernel_tracer_fd
< 0) {
1510 DBG("Failed to open %s", module_proc_lttng
);
1514 /* Validate kernel version */
1515 ret
= kernel_validate_version(&kernel_tracer_version
,
1516 &kernel_tracer_abi_version
);
1521 ret
= modprobe_lttng_data();
1526 ret
= kernel_supports_ring_buffer_snapshot_sample_positions();
1532 WARN("Kernel tracer does not support buffer monitoring. "
1533 "The monitoring timer of channels in the kernel domain "
1534 "will be set to 0 (disabled).");
1537 DBG("Kernel tracer fd %d", kernel_tracer_fd
);
1539 ret
= syscall_init_table(kernel_tracer_fd
);
1541 ERR("Unable to populate syscall table. Syscall tracing won't "
1542 "work for this session daemon.");
1547 modprobe_remove_lttng_control();
1548 ret
= close(kernel_tracer_fd
);
1552 kernel_tracer_fd
= -1;
1553 return LTTNG_ERR_KERN_VERSION
;
1556 ret
= close(kernel_tracer_fd
);
1562 modprobe_remove_lttng_control();
1565 WARN("No kernel tracer available");
1566 kernel_tracer_fd
= -1;
1568 return LTTNG_ERR_NEED_ROOT_SESSIOND
;
1570 return LTTNG_ERR_KERN_NA
;
1575 void cleanup_kernel_tracer(void)
1579 DBG2("Closing kernel fd");
1580 if (kernel_tracer_fd
>= 0) {
1581 ret
= close(kernel_tracer_fd
);
1585 kernel_tracer_fd
= -1;
1587 DBG("Unloading kernel modules");
1588 modprobe_remove_lttng_all();
1589 free(syscall_table
);
1593 bool kernel_tracer_is_initialized(void)
1595 return kernel_tracer_fd
>= 0;