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
);
138 * This is necessary since the creation time is present in the session
139 * name when it is generated.
141 if (session
->has_auto_generated_name
) {
142 ret
= kernctl_session_set_name(lks
->fd
, DEFAULT_SESSION_NAME
);
144 ret
= kernctl_session_set_name(lks
->fd
, session
->name
);
147 WARN("Could not set kernel session name for session %" PRIu64
" name: %s",
148 session
->id
, session
->name
);
151 ret
= kernctl_session_set_creation_time(lks
->fd
, session
->creation_time
);
153 WARN("Could not set kernel session creation time for session %" PRIu64
" name: %s",
154 session
->id
, session
->name
);
161 trace_kernel_destroy_session(lks
);
162 trace_kernel_free_session(lks
);
168 * Create a kernel channel, register it to the kernel tracer and add it to the
171 int kernel_create_channel(struct ltt_kernel_session
*session
,
172 struct lttng_channel
*chan
)
175 struct ltt_kernel_channel
*lkc
;
180 /* Allocate kernel channel */
181 lkc
= trace_kernel_create_channel(chan
);
186 DBG3("Kernel create channel %s with attr: %d, %" PRIu64
", %" PRIu64
", %u, %u, %d, %d",
187 chan
->name
, lkc
->channel
->attr
.overwrite
,
188 lkc
->channel
->attr
.subbuf_size
, lkc
->channel
->attr
.num_subbuf
,
189 lkc
->channel
->attr
.switch_timer_interval
, lkc
->channel
->attr
.read_timer_interval
,
190 lkc
->channel
->attr
.live_timer_interval
, lkc
->channel
->attr
.output
);
192 /* Kernel tracer channel creation */
193 ret
= kernctl_create_channel(session
->fd
, &lkc
->channel
->attr
);
195 PERROR("ioctl kernel create channel");
199 /* Setup the channel fd */
201 /* Prevent fd duplication after execlp() */
202 ret
= fcntl(lkc
->fd
, F_SETFD
, FD_CLOEXEC
);
204 PERROR("fcntl session fd");
207 /* Add channel to session */
208 cds_list_add(&lkc
->list
, &session
->channel_list
.head
);
209 session
->channel_count
++;
210 lkc
->session
= session
;
211 lkc
->key
= ++next_kernel_channel_key
;
213 DBG("Kernel channel %s created (fd: %d, key: %" PRIu64
")",
214 lkc
->channel
->name
, lkc
->fd
, lkc
->key
);
227 * Compute the offset of the instrumentation byte in the binary based on the
228 * function probe location using the ELF lookup method.
230 * Returns 0 on success and set the offset out parameter to the offset of the
232 * Returns -1 on error
235 int extract_userspace_probe_offset_function_elf(
236 const struct lttng_userspace_probe_location
*probe_location
,
237 struct ltt_kernel_session
*session
, uint64_t *offset
)
241 const char *symbol
= NULL
;
242 const struct lttng_userspace_probe_location_lookup_method
*lookup
= NULL
;
243 enum lttng_userspace_probe_location_lookup_method_type lookup_method_type
;
245 assert(lttng_userspace_probe_location_get_type(probe_location
) ==
246 LTTNG_USERSPACE_PROBE_LOCATION_TYPE_FUNCTION
);
248 lookup
= lttng_userspace_probe_location_get_lookup_method(
256 lttng_userspace_probe_location_lookup_method_get_type(lookup
);
258 assert(lookup_method_type
==
259 LTTNG_USERSPACE_PROBE_LOCATION_LOOKUP_METHOD_TYPE_FUNCTION_ELF
);
261 symbol
= lttng_userspace_probe_location_function_get_function_name(
268 fd
= lttng_userspace_probe_location_function_get_binary_fd(probe_location
);
274 ret
= run_as_extract_elf_symbol_offset(fd
, symbol
, session
->uid
,
275 session
->gid
, offset
);
277 DBG("userspace probe offset calculation failed for "
278 "function %s", symbol
);
282 DBG("userspace probe elf offset for %s is 0x%jd", symbol
, (intmax_t)(*offset
));
288 * Compute the offsets of the instrumentation bytes in the binary based on the
289 * tracepoint probe location using the SDT lookup method. This function
290 * allocates the offsets buffer, the caller must free it.
292 * Returns 0 on success and set the offset out parameter to the offsets of the
294 * Returns -1 on error.
297 int extract_userspace_probe_offset_tracepoint_sdt(
298 const struct lttng_userspace_probe_location
*probe_location
,
299 struct ltt_kernel_session
*session
, uint64_t **offsets
,
300 uint32_t *offsets_count
)
302 enum lttng_userspace_probe_location_lookup_method_type lookup_method_type
;
303 const struct lttng_userspace_probe_location_lookup_method
*lookup
= NULL
;
304 const char *probe_name
= NULL
, *provider_name
= NULL
;
308 assert(lttng_userspace_probe_location_get_type(probe_location
) ==
309 LTTNG_USERSPACE_PROBE_LOCATION_TYPE_TRACEPOINT
);
311 lookup
= lttng_userspace_probe_location_get_lookup_method(probe_location
);
318 lttng_userspace_probe_location_lookup_method_get_type(lookup
);
320 assert(lookup_method_type
==
321 LTTNG_USERSPACE_PROBE_LOCATION_LOOKUP_METHOD_TYPE_TRACEPOINT_SDT
);
324 probe_name
= lttng_userspace_probe_location_tracepoint_get_probe_name(
331 provider_name
= lttng_userspace_probe_location_tracepoint_get_provider_name(
333 if (!provider_name
) {
338 fd
= lttng_userspace_probe_location_tracepoint_get_binary_fd(probe_location
);
344 ret
= run_as_extract_sdt_probe_offsets(fd
, provider_name
, probe_name
,
345 session
->uid
, session
->gid
, offsets
, offsets_count
);
347 DBG("userspace probe offset calculation failed for sdt "
348 "probe %s:%s", provider_name
, probe_name
);
352 if (*offsets_count
== 0) {
353 DBG("no userspace probe offset found");
357 DBG("%u userspace probe SDT offsets found for %s:%s at:",
358 *offsets_count
, provider_name
, probe_name
);
359 for (i
= 0; i
< *offsets_count
; i
++) {
360 DBG("\t0x%jd", (intmax_t)((*offsets
)[i
]));
367 * Extract the offsets of the instrumentation point for the different lookup
371 int userspace_probe_add_callsites(struct lttng_event
*ev
,
372 struct ltt_kernel_session
*session
, int fd
)
374 const struct lttng_userspace_probe_location_lookup_method
*lookup_method
= NULL
;
375 enum lttng_userspace_probe_location_lookup_method_type type
;
376 const struct lttng_userspace_probe_location
*location
= NULL
;
380 assert(ev
->type
== LTTNG_EVENT_USERSPACE_PROBE
);
382 location
= lttng_event_get_userspace_probe_location(ev
);
388 lttng_userspace_probe_location_get_lookup_method(location
);
389 if (!lookup_method
) {
394 type
= lttng_userspace_probe_location_lookup_method_get_type(lookup_method
);
396 case LTTNG_USERSPACE_PROBE_LOCATION_LOOKUP_METHOD_TYPE_FUNCTION_ELF
:
398 struct lttng_kernel_event_callsite callsite
;
401 ret
= extract_userspace_probe_offset_function_elf(location
, session
, &offset
);
403 ret
= LTTNG_ERR_PROBE_LOCATION_INVAL
;
407 callsite
.u
.uprobe
.offset
= offset
;
408 ret
= kernctl_add_callsite(fd
, &callsite
);
410 WARN("Adding callsite to userspace probe "
411 "event %s failed.", ev
->name
);
412 ret
= LTTNG_ERR_KERN_ENABLE_FAIL
;
417 case LTTNG_USERSPACE_PROBE_LOCATION_LOOKUP_METHOD_TYPE_TRACEPOINT_SDT
:
420 uint64_t *offsets
= NULL
;
421 uint32_t offsets_count
;
422 struct lttng_kernel_event_callsite callsite
;
425 * This call allocates the offsets buffer. This buffer must be freed
428 ret
= extract_userspace_probe_offset_tracepoint_sdt(location
, session
,
429 &offsets
, &offsets_count
);
431 ret
= LTTNG_ERR_PROBE_LOCATION_INVAL
;
434 for (i
= 0; i
< offsets_count
; i
++) {
435 callsite
.u
.uprobe
.offset
= offsets
[i
];
436 ret
= kernctl_add_callsite(fd
, &callsite
);
438 WARN("Adding callsite to userspace probe "
439 "event %s failed.", ev
->name
);
440 ret
= LTTNG_ERR_KERN_ENABLE_FAIL
;
449 ret
= LTTNG_ERR_PROBE_LOCATION_INVAL
;
457 * Create a kernel event, enable it to the kernel tracer and add it to the
458 * channel event list of the kernel session.
459 * We own filter_expression and filter.
461 int kernel_create_event(struct lttng_event
*ev
,
462 struct ltt_kernel_channel
*channel
,
463 char *filter_expression
,
464 struct lttng_filter_bytecode
*filter
)
467 enum lttng_error_code ret
;
468 struct ltt_kernel_event
*event
;
473 /* We pass ownership of filter_expression and filter */
474 ret
= trace_kernel_create_event(ev
, filter_expression
,
476 if (ret
!= LTTNG_OK
) {
480 fd
= kernctl_create_event(channel
->fd
, event
->event
);
484 ret
= LTTNG_ERR_KERN_EVENT_EXIST
;
487 WARN("Event type not implemented");
488 ret
= LTTNG_ERR_KERN_EVENT_ENOSYS
;
491 WARN("Event %s not found!", ev
->name
);
492 ret
= LTTNG_ERR_KERN_ENABLE_FAIL
;
495 ret
= LTTNG_ERR_KERN_ENABLE_FAIL
;
496 PERROR("create event ioctl");
501 event
->type
= ev
->type
;
503 /* Prevent fd duplication after execlp() */
504 err
= fcntl(event
->fd
, F_SETFD
, FD_CLOEXEC
);
506 PERROR("fcntl session fd");
510 err
= kernctl_filter(event
->fd
, filter
);
514 ret
= LTTNG_ERR_FILTER_NOMEM
;
517 ret
= LTTNG_ERR_FILTER_INVAL
;
524 if (ev
->type
== LTTNG_EVENT_USERSPACE_PROBE
) {
525 ret
= userspace_probe_add_callsites(ev
, channel
->session
, event
->fd
);
527 goto add_callsite_error
;
531 err
= kernctl_enable(event
->fd
);
535 ret
= LTTNG_ERR_KERN_EVENT_EXIST
;
538 PERROR("enable kernel event");
539 ret
= LTTNG_ERR_KERN_ENABLE_FAIL
;
545 /* Add event to event list */
546 cds_list_add(&event
->list
, &channel
->events_list
.head
);
547 channel
->event_count
++;
549 DBG("Event %s created (fd: %d)", ev
->name
, event
->fd
);
559 closeret
= close(event
->fd
);
561 PERROR("close event fd");
571 * Disable a kernel channel.
573 int kernel_disable_channel(struct ltt_kernel_channel
*chan
)
579 ret
= kernctl_disable(chan
->fd
);
581 PERROR("disable chan ioctl");
586 DBG("Kernel channel %s disabled (fd: %d, key: %" PRIu64
")",
587 chan
->channel
->name
, chan
->fd
, chan
->key
);
596 * Enable a kernel channel.
598 int kernel_enable_channel(struct ltt_kernel_channel
*chan
)
604 ret
= kernctl_enable(chan
->fd
);
605 if (ret
< 0 && ret
!= -EEXIST
) {
606 PERROR("Enable kernel chan");
611 DBG("Kernel channel %s enabled (fd: %d, key: %" PRIu64
")",
612 chan
->channel
->name
, chan
->fd
, chan
->key
);
621 * Enable a kernel event.
623 int kernel_enable_event(struct ltt_kernel_event
*event
)
629 ret
= kernctl_enable(event
->fd
);
633 ret
= LTTNG_ERR_KERN_EVENT_EXIST
;
636 PERROR("enable kernel event");
643 DBG("Kernel event %s enabled (fd: %d)", event
->event
->name
, event
->fd
);
652 * Disable a kernel event.
654 int kernel_disable_event(struct ltt_kernel_event
*event
)
660 ret
= kernctl_disable(event
->fd
);
664 ret
= LTTNG_ERR_KERN_EVENT_EXIST
;
667 PERROR("disable kernel event");
674 DBG("Kernel event %s disabled (fd: %d)", event
->event
->name
, event
->fd
);
683 int kernel_track_pid(struct ltt_kernel_session
*session
, int pid
)
687 DBG("Kernel track PID %d for session id %" PRIu64
".",
689 ret
= kernctl_track_pid(session
->fd
, pid
);
695 return LTTNG_ERR_INVALID
;
697 return LTTNG_ERR_NOMEM
;
699 return LTTNG_ERR_PID_TRACKED
;
701 return LTTNG_ERR_UNK
;
705 int kernel_untrack_pid(struct ltt_kernel_session
*session
, int pid
)
709 DBG("Kernel untrack PID %d for session id %" PRIu64
".",
711 ret
= kernctl_untrack_pid(session
->fd
, pid
);
717 return LTTNG_ERR_INVALID
;
719 return LTTNG_ERR_NOMEM
;
721 return LTTNG_ERR_PID_NOT_TRACKED
;
723 return LTTNG_ERR_UNK
;
727 ssize_t
kernel_list_tracker_pids(struct ltt_kernel_session
*session
,
732 ssize_t nbmem
, count
= 0;
736 fd
= kernctl_list_tracker_pids(session
->fd
);
738 PERROR("kernel tracker pids list");
742 fp
= fdopen(fd
, "r");
744 PERROR("kernel tracker pids list fdopen");
748 nbmem
= KERNEL_TRACKER_PIDS_INIT_LIST_SIZE
;
749 pids
= zmalloc(sizeof(*pids
) * nbmem
);
751 PERROR("alloc list pids");
756 while (fscanf(fp
, "process { pid = %u; };\n", &pid
) == 1) {
757 if (count
>= nbmem
) {
761 new_nbmem
= nbmem
<< 1;
762 DBG("Reallocating pids list from %zu to %zu entries",
764 new_pids
= realloc(pids
, new_nbmem
* sizeof(*new_pids
));
765 if (new_pids
== NULL
) {
766 PERROR("realloc list events");
771 /* Zero the new memory */
772 memset(new_pids
+ nbmem
, 0,
773 (new_nbmem
- nbmem
) * sizeof(*new_pids
));
781 DBG("Kernel list tracker pids done (%zd pids)", count
);
783 ret
= fclose(fp
); /* closes both fp and fd */
799 * Create kernel metadata, open from the kernel tracer and add it to the
802 int kernel_open_metadata(struct ltt_kernel_session
*session
)
805 struct ltt_kernel_metadata
*lkm
= NULL
;
809 /* Allocate kernel metadata */
810 lkm
= trace_kernel_create_metadata();
815 /* Kernel tracer metadata creation */
816 ret
= kernctl_open_metadata(session
->fd
, &lkm
->conf
->attr
);
822 lkm
->key
= ++next_kernel_channel_key
;
823 /* Prevent fd duplication after execlp() */
824 ret
= fcntl(lkm
->fd
, F_SETFD
, FD_CLOEXEC
);
826 PERROR("fcntl session fd");
829 session
->metadata
= lkm
;
831 DBG("Kernel metadata opened (fd: %d)", lkm
->fd
);
836 trace_kernel_destroy_metadata(lkm
);
842 * Start tracing session.
844 int kernel_start_session(struct ltt_kernel_session
*session
)
850 ret
= kernctl_start_session(session
->fd
);
852 PERROR("ioctl start session");
856 DBG("Kernel session started");
865 * Make a kernel wait to make sure in-flight probe have completed.
867 void kernel_wait_quiescent(void)
870 int fd
= kernel_tracer_fd
;
872 DBG("Kernel quiescent wait on %d", fd
);
874 ret
= kernctl_wait_quiescent(fd
);
876 PERROR("wait quiescent ioctl");
877 ERR("Kernel quiescent wait failed");
882 * Force flush buffer of metadata.
884 int kernel_metadata_flush_buffer(int fd
)
888 DBG("Kernel flushing metadata buffer on fd %d", fd
);
890 ret
= kernctl_buffer_flush(fd
);
892 ERR("Fail to flush metadata buffers %d (ret: %d)", fd
, ret
);
899 * Force flush buffer for channel.
901 int kernel_flush_buffer(struct ltt_kernel_channel
*channel
)
904 struct ltt_kernel_stream
*stream
;
908 DBG("Flush buffer for channel %s", channel
->channel
->name
);
910 cds_list_for_each_entry(stream
, &channel
->stream_list
.head
, list
) {
911 DBG("Flushing channel stream %d", stream
->fd
);
912 ret
= kernctl_buffer_flush(stream
->fd
);
915 ERR("Fail to flush buffer for stream %d (ret: %d)",
924 * Stop tracing session.
926 int kernel_stop_session(struct ltt_kernel_session
*session
)
932 ret
= kernctl_stop_session(session
->fd
);
937 DBG("Kernel session stopped");
946 * Open stream of channel, register it to the kernel tracer and add it
947 * to the stream list of the channel.
949 * Note: given that the streams may appear in random order wrt CPU
950 * number (e.g. cpu hotplug), the index value of the stream number in
951 * the stream name is not necessarily linked to the CPU number.
953 * Return the number of created stream. Else, a negative value.
955 int kernel_open_channel_stream(struct ltt_kernel_channel
*channel
)
958 struct ltt_kernel_stream
*lks
;
962 while ((ret
= kernctl_create_stream(channel
->fd
)) >= 0) {
963 lks
= trace_kernel_create_stream(channel
->channel
->name
,
964 channel
->stream_count
);
974 /* Prevent fd duplication after execlp() */
975 ret
= fcntl(lks
->fd
, F_SETFD
, FD_CLOEXEC
);
977 PERROR("fcntl session fd");
980 lks
->tracefile_size
= channel
->channel
->attr
.tracefile_size
;
981 lks
->tracefile_count
= channel
->channel
->attr
.tracefile_count
;
983 /* Add stream to channel stream list */
984 cds_list_add(&lks
->list
, &channel
->stream_list
.head
);
985 channel
->stream_count
++;
987 DBG("Kernel stream %s created (fd: %d, state: %d)", lks
->name
, lks
->fd
,
991 return channel
->stream_count
;
998 * Open the metadata stream and set it to the kernel session.
1000 int kernel_open_metadata_stream(struct ltt_kernel_session
*session
)
1006 ret
= kernctl_create_stream(session
->metadata
->fd
);
1008 PERROR("kernel create metadata stream");
1012 DBG("Kernel metadata stream created (fd: %d)", ret
);
1013 session
->metadata_stream_fd
= ret
;
1014 /* Prevent fd duplication after execlp() */
1015 ret
= fcntl(session
->metadata_stream_fd
, F_SETFD
, FD_CLOEXEC
);
1017 PERROR("fcntl session fd");
1027 * Get the event list from the kernel tracer and return the number of elements.
1029 ssize_t
kernel_list_events(struct lttng_event
**events
)
1033 size_t nbmem
, count
= 0;
1035 struct lttng_event
*elist
;
1039 fd
= kernctl_tracepoint_list(kernel_tracer_fd
);
1041 PERROR("kernel tracepoint list");
1045 fp
= fdopen(fd
, "r");
1047 PERROR("kernel tracepoint list fdopen");
1052 * Init memory size counter
1053 * See kernel-ctl.h for explanation of this value
1055 nbmem
= KERNEL_EVENT_INIT_LIST_SIZE
;
1056 elist
= zmalloc(sizeof(struct lttng_event
) * nbmem
);
1057 if (elist
== NULL
) {
1058 PERROR("alloc list events");
1063 while (fscanf(fp
, "event { name = %m[^;]; };\n", &event
) == 1) {
1064 if (count
>= nbmem
) {
1065 struct lttng_event
*new_elist
;
1068 new_nbmem
= nbmem
<< 1;
1069 DBG("Reallocating event list from %zu to %zu bytes",
1071 new_elist
= realloc(elist
, new_nbmem
* sizeof(struct lttng_event
));
1072 if (new_elist
== NULL
) {
1073 PERROR("realloc list events");
1079 /* Zero the new memory */
1080 memset(new_elist
+ nbmem
, 0,
1081 (new_nbmem
- nbmem
) * sizeof(struct lttng_event
));
1085 strncpy(elist
[count
].name
, event
, LTTNG_SYMBOL_NAME_LEN
);
1086 elist
[count
].name
[LTTNG_SYMBOL_NAME_LEN
- 1] = '\0';
1087 elist
[count
].enabled
= -1;
1093 DBG("Kernel list events done (%zu events)", count
);
1095 ret
= fclose(fp
); /* closes both fp and fd */
1111 * Get kernel version and validate it.
1113 int kernel_validate_version(struct lttng_kernel_tracer_version
*version
,
1114 struct lttng_kernel_tracer_abi_version
*abi_version
)
1118 ret
= kernctl_tracer_version(kernel_tracer_fd
, version
);
1120 ERR("Failed to retrieve the lttng-modules version");
1124 /* Validate version */
1125 if (version
->major
!= VERSION_MAJOR
) {
1126 ERR("Kernel tracer major version (%d) is not compatible with lttng-tools major version (%d)",
1127 version
->major
, VERSION_MAJOR
);
1130 ret
= kernctl_tracer_abi_version(kernel_tracer_fd
, abi_version
);
1132 ERR("Failed to retrieve lttng-modules ABI version");
1135 if (abi_version
->major
!= LTTNG_MODULES_ABI_MAJOR_VERSION
) {
1136 ERR("Kernel tracer ABI version (%d.%d) does not match the expected ABI major version (%d.*)",
1137 abi_version
->major
, abi_version
->minor
,
1138 LTTNG_MODULES_ABI_MAJOR_VERSION
);
1141 DBG2("Kernel tracer version validated (%d.%d, ABI %d.%d)",
1142 version
->major
, version
->minor
,
1143 abi_version
->major
, abi_version
->minor
);
1150 ERR("Kernel tracer version check failed; kernel tracing will not be available");
1155 * Kernel work-arounds called at the start of sessiond main().
1157 int init_kernel_workarounds(void)
1163 * boot_id needs to be read once before being used concurrently
1164 * to deal with a Linux kernel race. A fix is proposed for
1165 * upstream, but the work-around is needed for older kernels.
1167 fp
= fopen("/proc/sys/kernel/random/boot_id", "r");
1174 ret
= fread(buf
, 1, sizeof(buf
), fp
);
1176 /* Ignore error, we don't really care */
1188 * Teardown of a kernel session, keeping data required by destroy notifiers.
1190 void kernel_destroy_session(struct ltt_kernel_session
*ksess
)
1192 struct lttng_trace_chunk
*trace_chunk
;
1194 if (ksess
== NULL
) {
1195 DBG3("No kernel session when tearing down session");
1199 DBG("Tearing down kernel session");
1200 trace_chunk
= ksess
->current_trace_chunk
;
1203 * Destroy channels on the consumer if at least one FD has been sent and we
1204 * are in no output mode because the streams are in *no* monitor mode so we
1205 * have to send a command to clean them up or else they leaked.
1207 if (!ksess
->output_traces
&& ksess
->consumer_fds_sent
) {
1209 struct consumer_socket
*socket
;
1210 struct lttng_ht_iter iter
;
1212 /* For each consumer socket. */
1214 cds_lfht_for_each_entry(ksess
->consumer
->socks
->ht
, &iter
.iter
,
1215 socket
, node
.node
) {
1216 struct ltt_kernel_channel
*chan
;
1218 /* For each channel, ask the consumer to destroy it. */
1219 cds_list_for_each_entry(chan
, &ksess
->channel_list
.head
, list
) {
1220 ret
= kernel_consumer_destroy_channel(socket
, chan
);
1222 /* Consumer is probably dead. Use next socket. */
1230 /* Close any relayd session */
1231 consumer_output_send_destroy_relayd(ksess
->consumer
);
1233 trace_kernel_destroy_session(ksess
);
1234 lttng_trace_chunk_put(trace_chunk
);
1237 /* Teardown of data required by destroy notifiers. */
1238 void kernel_free_session(struct ltt_kernel_session
*ksess
)
1240 if (ksess
== NULL
) {
1243 trace_kernel_free_session(ksess
);
1247 * Destroy a kernel channel object. It does not do anything on the tracer side.
1249 void kernel_destroy_channel(struct ltt_kernel_channel
*kchan
)
1251 struct ltt_kernel_session
*ksess
= NULL
;
1254 assert(kchan
->channel
);
1256 DBG3("Kernel destroy channel %s", kchan
->channel
->name
);
1258 /* Update channel count of associated session. */
1259 if (kchan
->session
) {
1260 /* Keep pointer reference so we can update it after the destroy. */
1261 ksess
= kchan
->session
;
1264 trace_kernel_destroy_channel(kchan
);
1267 * At this point the kernel channel is not visible anymore. This is safe
1268 * since in order to work on a visible kernel session, the tracing session
1269 * lock (ltt_session.lock) MUST be acquired.
1272 ksess
->channel_count
--;
1277 * Take a snapshot for a given kernel session.
1279 * Return LTTNG_OK on success or else return a LTTNG_ERR code.
1281 enum lttng_error_code
kernel_snapshot_record(
1282 struct ltt_kernel_session
*ksess
,
1283 const struct consumer_output
*output
, int wait
,
1284 uint64_t nb_packets_per_stream
)
1286 int err
, ret
, saved_metadata_fd
;
1287 enum lttng_error_code status
= LTTNG_OK
;
1288 struct consumer_socket
*socket
;
1289 struct lttng_ht_iter iter
;
1290 struct ltt_kernel_metadata
*saved_metadata
;
1291 char *trace_path
= NULL
;
1294 assert(ksess
->consumer
);
1297 DBG("Kernel snapshot record started");
1299 /* Save current metadata since the following calls will change it. */
1300 saved_metadata
= ksess
->metadata
;
1301 saved_metadata_fd
= ksess
->metadata_stream_fd
;
1305 ret
= kernel_open_metadata(ksess
);
1307 status
= LTTNG_ERR_KERN_META_FAIL
;
1311 ret
= kernel_open_metadata_stream(ksess
);
1313 status
= LTTNG_ERR_KERN_META_FAIL
;
1314 goto error_open_stream
;
1317 trace_path
= setup_channel_trace_path(ksess
->consumer
,
1318 DEFAULT_KERNEL_TRACE_DIR
);
1320 status
= LTTNG_ERR_INVALID
;
1323 /* Send metadata to consumer and snapshot everything. */
1324 cds_lfht_for_each_entry(output
->socks
->ht
, &iter
.iter
,
1325 socket
, node
.node
) {
1326 struct ltt_kernel_channel
*chan
;
1328 pthread_mutex_lock(socket
->lock
);
1329 /* This stream must not be monitored by the consumer. */
1330 ret
= kernel_consumer_add_metadata(socket
, ksess
, 0);
1331 pthread_mutex_unlock(socket
->lock
);
1333 status
= LTTNG_ERR_KERN_META_FAIL
;
1334 goto error_consumer
;
1337 /* For each channel, ask the consumer to snapshot it. */
1338 cds_list_for_each_entry(chan
, &ksess
->channel_list
.head
, list
) {
1339 status
= consumer_snapshot_channel(socket
, chan
->key
, output
, 0,
1340 ksess
->uid
, ksess
->gid
,
1342 nb_packets_per_stream
);
1343 if (status
!= LTTNG_OK
) {
1344 (void) kernel_consumer_destroy_metadata(socket
,
1346 goto error_consumer
;
1350 /* Snapshot metadata, */
1351 status
= consumer_snapshot_channel(socket
, ksess
->metadata
->key
, output
,
1352 1, ksess
->uid
, ksess
->gid
, trace_path
, wait
, 0);
1353 if (status
!= LTTNG_OK
) {
1354 goto error_consumer
;
1358 * The metadata snapshot is done, ask the consumer to destroy it since
1359 * it's not monitored on the consumer side.
1361 (void) kernel_consumer_destroy_metadata(socket
, ksess
->metadata
);
1365 /* Close newly opened metadata stream. It's now on the consumer side. */
1366 err
= close(ksess
->metadata_stream_fd
);
1368 PERROR("close snapshot kernel");
1372 trace_kernel_destroy_metadata(ksess
->metadata
);
1374 /* Restore metadata state.*/
1375 ksess
->metadata
= saved_metadata
;
1376 ksess
->metadata_stream_fd
= saved_metadata_fd
;
1383 * Get the syscall mask array from the kernel tracer.
1385 * Return 0 on success else a negative value. In both case, syscall_mask should
1388 int kernel_syscall_mask(int chan_fd
, char **syscall_mask
, uint32_t *nr_bits
)
1390 assert(syscall_mask
);
1393 return kernctl_syscall_mask(chan_fd
, syscall_mask
, nr_bits
);
1397 * Check for the support of the RING_BUFFER_SNAPSHOT_SAMPLE_POSITIONS via abi
1400 * Return 1 on success, 0 when feature is not supported, negative value in case
1403 int kernel_supports_ring_buffer_snapshot_sample_positions(void)
1405 int ret
= 0; // Not supported by default
1406 struct lttng_kernel_tracer_abi_version abi
;
1408 ret
= kernctl_tracer_abi_version(kernel_tracer_fd
, &abi
);
1410 ERR("Failed to retrieve lttng-modules ABI version");
1415 * RING_BUFFER_SNAPSHOT_SAMPLE_POSITIONS was introduced in 2.3
1417 if (abi
.major
>= 2 && abi
.minor
>= 3) {
1429 * Rotate a kernel session.
1431 * Return LTTNG_OK on success or else an LTTng error code.
1433 enum lttng_error_code
kernel_rotate_session(struct ltt_session
*session
)
1436 enum lttng_error_code status
= LTTNG_OK
;
1437 struct consumer_socket
*socket
;
1438 struct lttng_ht_iter iter
;
1439 struct ltt_kernel_session
*ksess
= session
->kernel_session
;
1442 assert(ksess
->consumer
);
1444 DBG("Rotate kernel session %s started (session %" PRIu64
")",
1445 session
->name
, session
->id
);
1450 * Note that this loop will end after one iteration given that there is
1451 * only one kernel consumer.
1453 cds_lfht_for_each_entry(ksess
->consumer
->socks
->ht
, &iter
.iter
,
1454 socket
, node
.node
) {
1455 struct ltt_kernel_channel
*chan
;
1457 /* For each channel, ask the consumer to rotate it. */
1458 cds_list_for_each_entry(chan
, &ksess
->channel_list
.head
, list
) {
1459 DBG("Rotate kernel channel %" PRIu64
", session %s",
1460 chan
->key
, session
->name
);
1461 ret
= consumer_rotate_channel(socket
, chan
->key
,
1462 ksess
->uid
, ksess
->gid
, ksess
->consumer
,
1463 /* is_metadata_channel */ false);
1465 status
= LTTNG_ERR_KERN_CONSUMER_FAIL
;
1471 * Rotate the metadata channel.
1473 ret
= consumer_rotate_channel(socket
, ksess
->metadata
->key
,
1474 ksess
->uid
, ksess
->gid
, ksess
->consumer
,
1475 /* is_metadata_channel */ true);
1477 status
= LTTNG_ERR_KERN_CONSUMER_FAIL
;
1487 enum lttng_error_code
kernel_create_channel_subdirectories(
1488 const struct ltt_kernel_session
*ksess
)
1490 enum lttng_error_code ret
= LTTNG_OK
;
1491 enum lttng_trace_chunk_status chunk_status
;
1494 assert(ksess
->current_trace_chunk
);
1497 * Create the index subdirectory which will take care
1498 * of implicitly creating the channel's path.
1500 chunk_status
= lttng_trace_chunk_create_subdirectory(
1501 ksess
->current_trace_chunk
,
1502 DEFAULT_KERNEL_TRACE_DIR
"/" DEFAULT_INDEX_DIR
);
1503 if (chunk_status
!= LTTNG_TRACE_CHUNK_STATUS_OK
) {
1504 ret
= LTTNG_ERR_CREATE_DIR_FAIL
;
1513 * Setup necessary data for kernel tracer action.
1516 int init_kernel_tracer(void)
1519 bool is_root
= !getuid();
1521 /* Modprobe lttng kernel modules */
1522 ret
= modprobe_lttng_control();
1527 /* Open debugfs lttng */
1528 kernel_tracer_fd
= open(module_proc_lttng
, O_RDWR
);
1529 if (kernel_tracer_fd
< 0) {
1530 DBG("Failed to open %s", module_proc_lttng
);
1534 /* Validate kernel version */
1535 ret
= kernel_validate_version(&kernel_tracer_version
,
1536 &kernel_tracer_abi_version
);
1541 ret
= modprobe_lttng_data();
1546 ret
= kernel_supports_ring_buffer_snapshot_sample_positions();
1552 WARN("Kernel tracer does not support buffer monitoring. "
1553 "The monitoring timer of channels in the kernel domain "
1554 "will be set to 0 (disabled).");
1557 DBG("Kernel tracer fd %d", kernel_tracer_fd
);
1559 ret
= syscall_init_table(kernel_tracer_fd
);
1561 ERR("Unable to populate syscall table. Syscall tracing won't "
1562 "work for this session daemon.");
1567 modprobe_remove_lttng_control();
1568 ret
= close(kernel_tracer_fd
);
1572 kernel_tracer_fd
= -1;
1573 return LTTNG_ERR_KERN_VERSION
;
1576 ret
= close(kernel_tracer_fd
);
1582 modprobe_remove_lttng_control();
1585 WARN("No kernel tracer available");
1586 kernel_tracer_fd
= -1;
1588 return LTTNG_ERR_NEED_ROOT_SESSIOND
;
1590 return LTTNG_ERR_KERN_NA
;
1595 void cleanup_kernel_tracer(void)
1599 DBG2("Closing kernel fd");
1600 if (kernel_tracer_fd
>= 0) {
1601 ret
= close(kernel_tracer_fd
);
1605 kernel_tracer_fd
= -1;
1607 DBG("Unloading kernel modules");
1608 modprobe_remove_lttng_all();
1609 free(syscall_table
);
1613 bool kernel_tracer_is_initialized(void)
1615 return kernel_tracer_fd
>= 0;