Philippe Proulx [Mon, 17 May 2021 18:02:45 +0000 (14:02 -0400)]
lttng-add-trigger(1): DESCRIPTION: move up the link to "EXAMPLES"
In lttng(1) manual pages, put
See the <<examples,EXAMPLES>> section below for usage examples.
above references to other manual pages at the bottom of the
"DESCRIPTION" section.
Signed-off-by: Philippe Proulx <eeppeliteloop@gmail.com>
Signed-off-by: Jérémie Galarneau <jeremie.galarneau@efficios.com>
Change-Id: I498a49536d6a876af9a0ed009378d257e483c975
Philippe Proulx [Mon, 17 May 2021 18:02:15 +0000 (14:02 -0400)]
lttng-add-context(1): EXAMPLES: add internal links to relevant options
Signed-off-by: Philippe Proulx <eeppeliteloop@gmail.com>
Signed-off-by: Jérémie Galarneau <jeremie.galarneau@efficios.com>
Change-Id: I242445993ae85c66074b90078dcd1f8e18cb86db
Philippe Proulx [Mon, 17 May 2021 17:57:48 +0000 (13:57 -0400)]
lttng-disable-event(1): `--tracepoint` option is not the default
I previously thought that not specifying any instrumentation point type
option was equivalent to specifying `--tracepoint`, like with the
`enable-event` command.
It's not the case: `--tracepoint` is just a filter, like `--syscall` and
the others. With `--kernel`, if you don't specify any instrumentation
point type condition option, the `disable-event` command disables
recording event rules regardless of their instrumentation point type.
This patch updates lttng-disable-event(1) to explain this.
Signed-off-by: Philippe Proulx <eeppeliteloop@gmail.com>
Signed-off-by: Jérémie Galarneau <jeremie.galarneau@efficios.com>
Change-Id: I489c5d16dae954f4b5116edce556fdd942c98c4d
Francis Deslauriers [Fri, 14 May 2021 20:04:28 +0000 (16:04 -0400)]
Cleanup: tests: name all temporary files to better identify leakage
When using a template, we need to add `--tmpdir` to the `mktemp`
arguments to place the tmp files in `/tmp` or `$TMPDIR`.
Signed-off-by: Francis Deslauriers <francis.deslauriers@efficios.com>
Signed-off-by: Jérémie Galarneau <jeremie.galarneau@efficios.com>
Change-Id: Id107530578d91700b726ceec016a8cef772e94b0
Francis Deslauriers [Mon, 17 May 2021 19:01:33 +0000 (15:01 -0400)]
Cleanup: tests: use find's `-name` option instead of grep
This prevents grep of finding matches in the leading directories listed
by find.
It caused problem when I used the "metadata" string in one of the
directory names:
/tmp/tmp.test_regen_metadata_ust_trace_path.6FbwPF/
Signed-off-by: Francis Deslauriers <francis.deslauriers@efficios.com>
Signed-off-by: Jérémie Galarneau <jeremie.galarneau@efficios.com>
Change-Id: I7165a3b4e4a16c793fe51c40484978be3a99dbfc
Jérémie Galarneau [Sat, 15 May 2021 03:32:57 +0000 (23:32 -0400)]
Update version to v2.13.0-rc2
Signed-off-by: Jérémie Galarneau <jeremie.galarneau@efficios.com>
Mathieu Desnoyers [Sat, 15 May 2021 00:57:33 +0000 (20:57 -0400)]
Fix: expected procname should not have -ust suffix
commit ("Fix: ustcomm: application name uses the '-ust'-suffixed thread
name") in LTTng-UST removes "-ust" suffix from procname, which appears
in the output trace directory hierarchy.
Adapt tests to not expect this quirk.
Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>
Signed-off-by: Jérémie Galarneau <jeremie.galarneau@efficios.com>
Change-Id: I96c3944124bbe154875085a37a51ba2bd458f83b
Francis Deslauriers [Fri, 14 May 2021 19:45:15 +0000 (15:45 -0400)]
Fix: Tests: leftover temporary files after tests
Signed-off-by: Francis Deslauriers <francis.deslauriers@efficios.com>
Signed-off-by: Jérémie Galarneau <jeremie.galarneau@efficios.com>
Change-Id: Ie0c66d45a00e890c1f1e8dc8db7b1baa2cca50f1
Francis Deslauriers [Wed, 12 May 2021 15:28:26 +0000 (11:28 -0400)]
Fix: trigger: abort() when adding `--notify` action with python event rule matches
Issue
=====
Adding the following trigger makes the sessiond abort:
lttng add-trigger --condition=event-rule-matches --domain=python --action=notify
With the following stacktrace:
(gdb) bt
#0 __GI_raise (sig=sig@entry=6) at ../sysdeps/unix/sysv/linux/raise.c:50
#1 0x00007ffff7a71859 in __GI_abort () at abort.c:79
#2 0x00000000004a1eeb in event_notifier_error_accounting_register_event_notifier (trigger=0x7fffe0000f50, error_counter_index=0x7fffefffde38) at event-notifier-error-accounting.c:1075
#3 0x00000000004743e6 in setup_tracer_notifier (state=0x7fffefffe3e8, trigger=0x7fffe0000f50) at notification-thread-events.c:2606
#4 0x000000000046dacb in handle_notification_thread_command_register_trigger (state=0x7fffefffe3e8, trigger=0x7fffe0000f50, is_trigger_anonymous=false, cmd_result=0x7fffedfdd6e8) at notification-thread-events.c:2751
#5 0x000000000046d083 in handle_notification_thread_command (handle=0x601460, state=0x7fffefffe3e8) at notification-thread-events.c:3112
#6 0x00000000004687bd in thread_notification (data=0x601460) at notification-thread.c:710
#7 0x0000000000486703 in launch_thread (data=0x601550) at thread.c:66
#8 0x00007ffff7c47609 in start_thread (arg=<optimized out>) at pthread_create.c:477
#9 0x00007ffff7b6e293 in clone () at ../sysdeps/unix/sysv/linux/x86_64/clone.S:95
This happens because the LTTNG_DOMAIN_PYTHON domain is not handled by
the switch-case.
Fix
===
Add LTTNG_DOMAIN_PYTHON (all other agent domains) as a fallthrough
LTTNG_DOMAIN_UST.
Note
====
Add a basic test case for python agent.
Signed-off-by: Francis Deslauriers <francis.deslauriers@efficios.com>
Signed-off-by: Jérémie Galarneau <jeremie.galarneau@efficios.com>
Change-Id: I9971cdad8bbc6adca8f6ba49964483c42a25be7d
Francis Deslauriers [Wed, 12 May 2021 15:09:37 +0000 (11:09 -0400)]
Tests: move `list_triggers_matches_ok` to utils.sh
Signed-off-by: Francis Deslauriers <francis.deslauriers@efficios.com>
Signed-off-by: Jérémie Galarneau <jeremie.galarneau@efficios.com>
Change-Id: I4cd3816fa7cbc1032582b2a99f67a3b14222f972
Francis Deslauriers [Thu, 13 May 2021 01:21:28 +0000 (21:21 -0400)]
Tests: Move python test app to `tests/utils/testapp` directory
Signed-off-by: Francis Deslauriers <francis.deslauriers@efficios.com>
Signed-off-by: Jérémie Galarneau <jeremie.galarneau@efficios.com>
Change-Id: I9bc55c0fc0b8de42157d64884c7e1ee08b800525
Mathieu Desnoyers [Wed, 12 May 2021 20:05:29 +0000 (16:05 -0400)]
Fix: validate whether event notifiers are supported by UST application
Considering that lttng-sessiond interacts with older (2.12) applications
as well as 2.13+, it needs not to send event notifiers related commands
to an older UST, because the unknown command will cause the protocol to
become out of sync on the communication socket, because the older
lttng-ust does not know how many bytes must be read when receiving the
unknown command. So even though it can return that the command is
unknown, the communication socket becomes out of sync.
Depends-on: lttng-ust: If048c739dd37147ffb2a54715c2101177d2df4f7
Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>
Signed-off-by: Jérémie Galarneau <jeremie.galarneau@efficios.com>
Change-Id: If9cbc7a7cd6cb957134bf3a2e284ab99ef20e93b
Mathieu Desnoyers [Wed, 12 May 2021 18:09:00 +0000 (14:09 -0400)]
Fix: ust: keep using lttng-ust-sock-8 and lttng-ust-wait-8 filenames
Replicate a change from a lttng-ust header for which there is an
internal copy within lttng-tools for building without lttng-ust.
lttng-ust commit
6a359b8a4006 ("Bump LTTNG_UST_ABI version from 8.1 to
9.0") attempted to ensure that a session daemon linked against an old
lttng-ust-ctl (2.12) would not attempt to interact with newer
applications.
This was done by increasing the major ABI version number, but
introducing LTTNG_UST_ABI_MAJOR_VERSION_OLDEST_COMPATIBLE (set to 8) as
a mean to allow session daemon to accept both major versions 8 and 9.
Unfortunately, changing LTTNG_UST_ABI_MAJOR_VERSION means the filenames
used for communication changed as well, meaning that applications linked
against older lttng-ust 2.12 cannot interact with lttng sessiond linked
against lttng-ust-ctl 2.13, which is unintended.
Therefore, go back to using the previous filenames: lttng-ust-sock-8,
and lttng-ust-wait-8 for communication.
Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>
Signed-off-by: Jérémie Galarneau <jeremie.galarneau@efficios.com>
Change-Id: I121c6e5aa13637f13b996dd9e31e06d060b52ca6
Mathieu Desnoyers [Wed, 7 Oct 2020 20:42:05 +0000 (16:42 -0400)]
Fix: Handle SIGBUS in sessiond and consumerd
There is an issue with the security model of lib ring buffer (lttng-ust)
vs SIGBUS handling by consumer daemon. We do not handle SIGBUS in the
consumer daemon. An application using ftruncate on a ring buffer shm
could cause the consumer to be killed with SIGBUS.
Wire up SIGBUS handling in the session daemon as well given that it also
uses liblttng-ust-ctl.
This depends on "liblttng-ust-ctl: Implement SIGBUS handling" in
lttng-ust, which extends the API of liblttng-ust-ctl, which requires
the user application to define the TLS sigbus state with
DEFINE_LTTNG_UST_SIGBUS_STATE(). It therefore needs to be introduced in
locked-step between lttng-ust and lttng-tools.
Considering that this change in liblttng-ust-ctl modifies the ABI, it is
done with a major soname version bump of the library, so it is allowed
to break the API.
Depends-on: lttng-ust: I7ade988e3e68a87930fbcee3e14e59c3fb66e755
Fixes: #1284
Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>
Signed-off-by: Jérémie Galarneau <jeremie.galarneau@efficios.com>
Change-Id: I76a91616bd35b21c5e891c8306c73ed5e8725ebb
Jérémie Galarneau [Thu, 13 May 2021 19:45:43 +0000 (15:45 -0400)]
Clean-up: rename double 'lttng' prefixed functions
Rename `lttng_lttng_ust_ctl_flush_buffer` to `lttng_ust_flush_buffer`
since the double prefix is odd, but needed to work-around a clash
with symbols introduced in liblttng-ust-ctl as part of 2.13.
Two functions stubs are not used are removed:
- lttng_lttng_ust_ctl_get_mmap_read_offset,
- lttng_lttng_ust_ctl_get_mmap_base.
Signed-off-by: Jérémie Galarneau <jeremie.galarneau@efficios.com>
Change-Id: I9a924f503c7d409ed1a28768d07cbbad9d23f503
Philippe Proulx [Thu, 13 May 2021 02:59:59 +0000 (22:59 -0400)]
lttng-{enable-event(1),event-rule(7)}: `--filter` is not always avail.
The `--filter` option of both lttng-enable-event(1) and a common event
rule specification is not available with the Linux kprobe, kretprobe,
and user space probe instrumentation point types.
Signed-off-by: Philippe Proulx <eeppeliteloop@gmail.com>
Signed-off-by: Jérémie Galarneau <jeremie.galarneau@efficios.com>
Change-Id: I2cea488359bbe8ac8494d785fe3282419ee66d46
Philippe Proulx [Thu, 13 May 2021 02:52:25 +0000 (22:52 -0400)]
lttng(1)/lttng-create(1): document `--relayd-path`
Signed-off-by: Philippe Proulx <eeppeliteloop@gmail.com>
Signed-off-by: Jérémie Galarneau <jeremie.galarneau@efficios.com>
Change-Id: I83d65e13834593c97e296db2505dc97f6a8e4568
Philippe Proulx [Thu, 13 May 2021 02:51:41 +0000 (22:51 -0400)]
doc/man: document `LTTNG_SESSIOND_PATH` env. var. like `--sessiond-path`
Signed-off-by: Philippe Proulx <eeppeliteloop@gmail.com>
Signed-off-by: Jérémie Galarneau <jeremie.galarneau@efficios.com>
Change-Id: Ie6ff3f2f1003d506e5d4a351af0f811b4ca8b556
Philippe Proulx [Thu, 13 May 2021 02:24:09 +0000 (22:24 -0400)]
lttng-save(1): document `--all` option in "SYNOPSIS" and "DESCRIPTION"
Signed-off-by: Philippe Proulx <eeppeliteloop@gmail.com>
Signed-off-by: Jérémie Galarneau <jeremie.galarneau@efficios.com>
Change-Id: I895e2c8277cb8c1b7644f6d8607cd77230706515
Jonathan Rajotte [Tue, 1 Dec 2020 17:19:53 +0000 (12:19 -0500)]
Fix: snapshot path have domain subdir duplicate "ust/ust" or "kernel/kernel"
Observed issue
==============
lttng-ivc observed unexpected path generated for streamed snapshot:
joraj-alpa/
test-
20190319-120000-
20210113-110101/
snapshot-1-
20210113-110102-0/
* ust/
* ust/
pid/
app-ust-
2362198-
20210113-110101/
channel_0
channel_1
metadata
channel_2
channel_3
"ust" the domain subdir is present two-time instead off only one time.
The same problem is seen for kernel snapshots.
Cause
=====
Based on dissection the problem was introduced by commit
5da88b0f58d7f838068037ea449ddfb25d3e85ad [1]
For snapshots, the consumer output object of the snapshot output and
information is fetched from the *original* consumer output.
(code block around src/bin/lttng-sessiond/cmd.c:4748)
The snapshot consumer output does not contains the necessary information
to populate domain_subdir on copy (domain_subdir is '\0').
This would lead to a len evaluation of 1 for the consumer_path_offset in
setup_channel_trace_path:70. This would end up not "skipping" the "ust"
and "kernel" part of the path.
Solution
========
Part of the solution is to copy the domain_subdir from the original
session consumer output to the snapshot output.
Still, the problem was still present since that now that the
domain_subdir was not "\0", the value was suffixed to the passed
session_path. In the snapshot code path, "ust/" and "kernel/" were
already present in the session_path passed to setup_channel_trace_path.
A quick modification at the caller level in the snapshot code path fixes
the issues once and for all.
Tests
=======
The test suit is augmented for certain key tests to validate the
complete path of a trace.
Path validation is based on crude bash pattern matching.
Known drawbacks
=========
None
References
==========
[1] https://github.com/lttng/lttng-tools/commit/
5da88b0f58d7f838068037ea449ddfb25d3e85ad
Signed-off-by: Jonathan Rajotte <jonathan.rajotte-julien@efficios.com>
Signed-off-by: Jérémie Galarneau <jeremie.galarneau@efficios.com>
Change-Id: Ie7cd2d5471ee3a942fa511e2f4cab09e3aa499e4
Jérémie Galarneau [Tue, 11 May 2021 18:37:02 +0000 (14:37 -0400)]
Fix: error: remove trailing period from error descriptions
Trailing periods in some error descriptions cause the client to print
two dots at the end of some error messages:
Error: Failed to register trigger: Tracing the kernel requires a root
lttng-sessiond daemon, as well as "tracing" group membership or root
user ID for the lttng client..
Signed-off-by: Jérémie Galarneau <jeremie.galarneau@efficios.com>
Change-Id: Ib6522252fb716b1baad3c7416e2db150c614702c
Jérémie Galarneau [Mon, 10 May 2021 19:02:51 +0000 (15:02 -0400)]
Make the subitems array a part of action_work_item
Make the subitems array a part of action_work_item instead of allocating
it dynamically when a trigger is enqueued for a given action executor.
Signed-off-by: Jérémie Galarneau <jeremie.galarneau@efficios.com>
Change-Id: I474e43ba3ba9267d629d3ce561c8677c90bddf27
Jérémie Galarneau [Tue, 11 May 2021 19:52:50 +0000 (15:52 -0400)]
Test: fix: firing policy name has changed
The firing policy name from "after every" to "every" in
4edabba83. The
tests are adapted to reflect this change.
Signed-off-by: Jérémie Galarneau <jeremie.galarneau@efficios.com>
Change-Id: Ic563c8ce959a6876e1300079ed236c83072efc58
Philippe Proulx [Thu, 6 May 2021 20:50:28 +0000 (16:50 -0400)]
lttng-add-trigger(1): add usage examples
Signed-off-by: Philippe Proulx <eeppeliteloop@gmail.com>
Signed-off-by: Jérémie Galarneau <jeremie.galarneau@efficios.com>
Change-Id: Iac386141968106e648da67498696ecba9524491e
Philippe Proulx [Thu, 6 May 2021 20:50:17 +0000 (16:50 -0400)]
lttng-add-context(1): add usage examples
Signed-off-by: Philippe Proulx <eeppeliteloop@gmail.com>
Signed-off-by: Jérémie Galarneau <jeremie.galarneau@efficios.com>
Change-Id: I8ce928325a659676210ea9d2647ce7223f9527aa
Philippe Proulx [Thu, 6 May 2021 20:19:39 +0000 (16:19 -0400)]
list_triggers.c: replace "after every" -> "every"
Signed-off-by: Philippe Proulx <eeppeliteloop@gmail.com>
Signed-off-by: Jérémie Galarneau <jeremie.galarneau@efficios.com>
Change-Id: If6121fda5bb6d7cf5c941c1ca134191c72518422
Philippe Proulx [Thu, 6 May 2021 20:10:01 +0000 (16:10 -0400)]
Fix: add_trigger.c: `goto error` with a wrong UID for `--owner-uid`
Signed-off-by: Philippe Proulx <eeppeliteloop@gmail.com>
Signed-off-by: Jérémie Galarneau <jeremie.galarneau@efficios.com>
Change-Id: I8650af1f43e8ed8bad58a928cfcdf78d382212f1
Philippe Proulx [Thu, 6 May 2021 19:15:29 +0000 (15:15 -0400)]
lttng-add-context(1): use a description list to explain `--channel`
Signed-off-by: Philippe Proulx <eeppeliteloop@gmail.com>
Signed-off-by: Jérémie Galarneau <jeremie.galarneau@efficios.com>
Change-Id: Ie2d701820bf1e086205254c0712fe088c3295b2d
Philippe Proulx [Thu, 6 May 2021 19:09:58 +0000 (15:09 -0400)]
doc/man/manpage.xsl: indent the body of formal (example) blocks
Signed-off-by: Philippe Proulx <eeppeliteloop@gmail.com>
Signed-off-by: Jérémie Galarneau <jeremie.galarneau@efficios.com>
Change-Id: I184b1bc54728cf0d770bc44f9f0923b9f8724396
Philippe Proulx [Thu, 6 May 2021 18:45:26 +0000 (14:45 -0400)]
doc/man: lttng(1) command pages: always include `common-footer.txt`
The end of an lttng(1) command manual page source now looks like this:
include::common-lttng-cmd-help-options.txt[]
include::common-lttng-cmd-after-options.txt[]
include::common-footer.txt[]
`common-lttng-cmd-after-options.txt` contains the "EXIT STATUS",
"ENVIRONMENT", and "FILES" sections.
`common-footer.txt` begins with the "RESOURCES" section.
This will make it possible to insert an "EXAMPLES" section between
`common-lttng-cmd-after-options.txt` and `common-footer.txt`.
Signed-off-by: Philippe Proulx <eeppeliteloop@gmail.com>
Signed-off-by: Jérémie Galarneau <jeremie.galarneau@efficios.com>
Change-Id: I1eee42d7386f4671d9825e9d3e131e54f868ee39
Philippe Proulx [Thu, 6 May 2021 18:14:08 +0000 (14:14 -0400)]
doc/man: follow man-pages(7) for section names/order and for "SEE ALSO"
Changes, following the "Sections within a manual page" section of
man-pages(7):
* Rename the "ENVIRONMENT VARIABLES" section to "ENVIRONMENT".
* Rename the "COPYRIGHTS" section to "COPYRIGHT".
* Move the "EXIT STATUS" section immediately after the "OPTIONS"
section.
* Remove the "BUGS" section.
Such a section indicates known bugs (why would you ever have such
a section anyway?).
The "RESOURCES" section now contains the URL of our bug tracker.
* Sort the manual pages by section, and then by name in the "SEE ALSO"
section.
Signed-off-by: Philippe Proulx <eeppeliteloop@gmail.com>
Signed-off-by: Jérémie Galarneau <jeremie.galarneau@efficios.com>
Change-Id: I3d91073ec876efd98dcc723ddf40272c814663dd
Francis Deslauriers [Fri, 7 May 2021 21:44:52 +0000 (17:44 -0400)]
Fix: action-executor: leak of `work_item::subitems` field
This leak can be reproduced by adding a trigger and looking at the used
memory of the lttng-sessiond process.
Valgrind reports the following leak:
==2472== 7,087,968 bytes in 147,666 blocks are definitely lost in loss record 24 of 24
==2472== at 0x4C31B25: calloc (in /usr/lib/valgrind/vgpreload_memcheck-amd64-linux.so)
==2472== by 0x191C37: zmalloc (macros.h:45)
==2472== by 0x195234: action_executor_enqueue_trigger (action-executor.c:913)
==2472== by 0x171AD7: dispatch_one_event_notifier_notification (notification-thread-events.c:4626)
==2472== by 0x171D1E: handle_one_event_notifier_notification (notification-thread-events.c:4715)
==2472== by 0x171DA4: handle_notification_thread_event_notification (notification-thread-events.c:4730)
==2472== by 0x164EA6: handle_event_notification_pipe (notification-thread.c:591)
==2472== by 0x1656C3: thread_notification (notification-thread.c:727)
==2472== by 0x17BD16: launch_thread (thread.c:66)
==2472== by 0x5F086DA: start_thread (pthread_create.c:463)
==2472== by 0x624188E: clone (clone.S:95)
Signed-off-by: Francis Deslauriers <francis.deslauriers@efficios.com>
Signed-off-by: Jérémie Galarneau <jeremie.galarneau@efficios.com>
Change-Id: I3d98e3ae18c5b2d7f400407842a7de17716f680c
Simon Marchi [Tue, 27 Apr 2021 22:00:28 +0000 (18:00 -0400)]
lttng: change a few instances of trigger "id" to "name"
The term trigger "id" has been changed to trigger "name", these are
leftovers. Fix them.
Change-Id: Ic05925af1cd25917de75a0a4d061324604e5a599
Signed-off-by: Simon Marchi <simon.marchi@efficios.com>
Signed-off-by: Jérémie Galarneau <jeremie.galarneau@efficios.com>
Simon Marchi [Tue, 27 Apr 2021 18:51:08 +0000 (14:51 -0400)]
Change "event rule hit" to "event-rule matches" in user interface strings
The condition name was changed from "event rule hit" to "event rule
matches", but there are some leftover user-visible strings with the old
name. Fix them.
Change-Id: I619c9265d68ca6bf4a3ac11ef38cabc715b01110
Signed-off-by: Simon Marchi <simon.marchi@efficios.com>
Signed-off-by: Jérémie Galarneau <jeremie.galarneau@efficios.com>
Jonathan Rajotte [Thu, 6 May 2021 15:14:52 +0000 (11:14 -0400)]
Fix: relayd: live: data is missing between viewer attach and retry
Observed issue
==============
Data produced between the time a live viewer attach and a viewer
subsequent "polling" is missing from the point of view of the viewer.
Using the following reproducer:
# A lttng-sessiond is already running
lttng-relayd -b -vvv > relayd.log 2>&1
lttng create live --live
lttng enable-event -u "hello_world:*"
lttng start
date
# Set the "polling" interval at ~30 seconds
babeltrace2 --retry-duration=
30000000 -i lttng-live net://localhost/host/$HOSTNAME/live &
pid=$!
# Make sure we are in the retry phase of bt2
date
sleep 2
# Produce events
# Note here that hello perform a sleep of 1 seconds before generating
# 10 events to make sure lttng-ust had time to register. A total of
# 100 events is generated.
date
for i in `seq 1 10`; do
./hello
done
# Here we expect to see 100 events outputted at some point (~18
# seconds (30 - 2 - 10) from that time).
# No events are received by babeltrace2.
date
sleep 40
date
# We retry the same exercise a second time.
echo "Moving to phase 2"
date
for i in `seq 1 10`; do
./hello
done
# Here data is received correctly.
date
sleep 40
date
kill $pid
wait
lttng destroy live
pkill lttng-relayd
Cause
=====
At the moment the viewer attach is done, no streams exist. On the
following viewer_get_new_streams (~30 seconds later), streams now exist
since they were allocated lazily following the first application
registration, but LTTNG_VIEWER_SEEK_LAST is used. This essentially
"discards" any events in the stream that happened between the attach
time and the viewer_get_new_stream execution.
Note that the same problem would happen if a new UID (in per-uid mode)
would happen to begin tracing in between viewer retry.
Solution
========
Always use `LTTNG_VIEWER_SEEK_BEGINNING` during
`viewer_get_new_streams`.
Known drawbacks
=========
This does not fix the per-pid problem where an app run and die in between
the viewer retries.
References
==========
https://lists.lttng.org/pipermail/lttng-dev/2021-May/029953.html
Signed-off-by: Jonathan Rajotte <jonathan.rajotte-julien@efficios.com>
Signed-off-by: Jérémie Galarneau <jeremie.galarneau@efficios.com>
Change-Id: I92fbe6511fb93e31d20922b4eef5c9a221e07716
Jérémie Galarneau [Thu, 6 May 2021 22:43:44 +0000 (18:43 -0400)]
Build fix: filter-grammar-test: incorrect inclusion flag
Including src/common causes LTTng files that clash with system headers
to be included before the system headers and breaks the build on
multiple platforms and architectures.
Signed-off-by: Jérémie Galarneau <jeremie.galarneau@efficios.com>
Change-Id: Iff7b6e7000d907dc076a467d223c91569b783861
Jérémie Galarneau [Thu, 6 May 2021 19:48:02 +0000 (15:48 -0400)]
Tests: fix: test_tp_fail: bail out on non-existing relay daemon
Fatal thread errors simulated by the tp_fail test cause the relay
daemon to shutdown. This is unexepected by stop_lttng_relayd_notap
which bails out, causing the test to fail.
We bail-out when the daemon is already dead to catch crashes
during the test suite since
a0f8e310.
Use the clean-up variant so that we don't fail the tests for this
expected outcome.
Signed-off-by: Jérémie Galarneau <jeremie.galarneau@efficios.com>
Change-Id: I02e7f35451b3c81e7e808e9ff96b6c824fa8f904
Jonathan Rajotte [Wed, 28 Apr 2021 00:13:13 +0000 (20:13 -0400)]
build system: support rotation example
Signed-off-by: Jonathan Rajotte <jonathan.rajotte-julien@efficios.com>
Signed-off-by: Jérémie Galarneau <jeremie.galarneau@efficios.com>
Change-Id: Icf5f88dcc725967b7fda66db6e047cdf18a89f01
Jonathan Rajotte [Fri, 14 Feb 2020 23:26:39 +0000 (18:26 -0500)]
Add trigger on event rule matches application example
Signed-off-by: Jonathan Rajotte <jonathan.rajotte-julien@efficios.com>
Signed-off-by: Jérémie Galarneau <jeremie.galarneau@efficios.com>
Change-Id: Ice78e15aa1f4380925a272def8bcc7ca331c0aa0
Philippe Proulx [Thu, 29 Apr 2021 21:00:07 +0000 (17:00 -0400)]
Update the remaining manual pages for LTTng-tools 2.13
This patch updates the remaining manual pages for LTTng-tools 2.13.
This patch:
* Improves the consistency of some command descriptions in
`configure.ac`.
* Adds `common-daemon-cfg.txt` which is a section explaining how to
configure a daemon (session or relay).
lttng-sessiond(8) and lttng-relayd(8) include this file.
* Adds `lttng-concepts.7.txt` which is an adapted copy of the
"Core concepts" section of the online LTTng Documentation.
This centralizes all the LTTng theory into a single manual page
instead of having this information split into multiple lttng(1)
command manual pages.
Many manual pages now refer to lttng-concepts(7), making it possible
to cut a lot of text in those.
* Updates existing manual pages to:
* Have a style and voice which is more consistent with the LTTng
Documentation (website) for 2.13.
* Fix various terminology ambiguities.
* Use more textual variables and lists to explain more complex logic
and processes.
* Always use the same pattern to specify the behaviour of an lttng(1)
command depending on the `SESSION` argument or the `--session`
option.
* For the commands which can perform more than one task, list their
available tasks at the beginning of the "DESCRIPTION" section.
* For some lttng(1) commands which can operate on all tracing sessions
(for example, lttng-clear(1) and lttng-destroy(1)), always indicate
that they target all your Unix user's tracing sessions or, if your
Unix user is `root`, the tracing sessions of all the Unix users
within the root session daemon.
* Clean the "SEE ALSO" sections.
* Always have "LTTng" in the "NAME" section of a manual page.
More specifically:
lttng-create(1):
* Clarify the tracing session modes.
* Clarify how the command adds (or not) a snapshot output for a
snapshot mode tracing session.
* Specify that `--output=DIR` is equivalent to
`--set-url=file://DIR`.
lttng-enable-channel(1):
Include the `--discard`, `--buffers-uid`, and `--buffers-global`
options in the "SYNOPSIS" section even if they are the current
defaults.
lttng-list(1):
Explain what this command does exactly using a tree of options
and arguments.
lttng-load(1):
Clarify how LTTng finds tracing session configurations.
lttng-relayd(8):
* Document the missing `--group` option.
* Rework the text in general.
* Add a daemon configuration section with an INI file example.
* Add more cross-references between options and equivalent
environment variables.
lttng-rotate(1):
Specify that the `rotate-session` trigger action can also rotate
a tracing session.
lttng-save(1):
Clarify the output path.
lttng-sessiond(8):
Add more cross-references between options and equivalent
environment variables.
lttng-shapshot(1):
* Clarify everything related to the snapshot output of a tracing
session, including when and how the lttng-create(1) command adds
an initial snapshot output.
* Specify that the `snapshot-session` trigger action can also take
a snapshot of a tracing session.
lttng-track(1):
lttng-untrack(1):
* Simply refer to allowing processes to record events and to
process attribute inclusion sets instead of using the vague
"tracker" terminology.
* Restate that those commands control an implicit condition of
a recording event rule, as per lttng-concepts(7).
* Improve the documentation of each inclusion set selection
option.
Signed-off-by: Philippe Proulx <eeppeliteloop@gmail.com>
Signed-off-by: Jérémie Galarneau <jeremie.galarneau@efficios.com>
Change-Id: Iac7498ee979fe077f0927a9b8335f6c07f203989
Philippe Proulx [Mon, 3 May 2021 19:46:51 +0000 (15:46 -0400)]
doc/man/manpage.xsl: do not indent the title of an admonition block
See the new comment in `manpage.xsl` for more details.
I copied and modified some upstream DocBook XSL (as found in
`xsl/manpages/block.xsl`).
Signed-off-by: Philippe Proulx <eeppeliteloop@gmail.com>
Signed-off-by: Jérémie Galarneau <jeremie.galarneau@efficios.com>
Change-Id: I765ce37957907edb4088a4212c1b12af8b485e60
Jérémie Galarneau [Thu, 29 Apr 2021 23:01:44 +0000 (19:01 -0400)]
Fix: lttng: add-trigger: invalid access past end of exclusions buffer
The following command causes the `lttng` client to SEGFAULT on 32-bit
machines:
lttng add-trigger \
--condition event-rule-matches --domain=user \
'--name=jean-*' \
--exclude-name jean-chretien -x jean-charest \
--action notify
Running with valgrind results in the following report:
Invalid read of size 8
at 0x12EEA4: validate_exclusion_list (enable_events.c:354)
by 0x149018: parse_event_rule (add_trigger.c:964)
by 0x148356: handle_condition_event (add_trigger.c:1222)
by 0x147EC3: parse_condition (add_trigger.c:1300)
by 0x147702: cmd_add_trigger (add_trigger.c:2107)
by 0x154067: handle_command (lttng.c:237)
by 0x1534D1: parse_args (lttng.c:426)
by 0x152C54: main (lttng.c:475)
validate_exclusion_list expects a NULL terminated array of exclusions
while handle_condition_event prepares an array of exclusions using
the dynamic pointer array facilities (which doesn't need to
null-terminate arrays).
The code that deals with exclusions "lists" mixes various conventions
(null-terminated vs explicit size) in different places leading to this
kind of errors.
All the code that references exclusion lists, along with the string
utils, are adapted to make use of the common dynamic pointer array
facilities.
Signed-off-by: Jérémie Galarneau <jeremie.galarneau@efficios.com>
Change-Id: Icbb7f0e8601c7ecc887dc9ae64d0ec6390e6aba3
Jonathan Rajotte [Wed, 5 May 2021 17:34:00 +0000 (13:34 -0400)]
Fix: tests: extra arguments are not passed to the remove trigger command
Observed issue
==============
CI root jobs fails on the remove trigger operation for the user_trigger
trigger during the test_ust_notifier_discarded_count_multi_uid test.
Cause
=====
The `--owner uid` option and its argument are not passed during the
command execution.
Solution
========
Pass all extra parameters for the command execution.
Known drawbacks
=========
None.
Signed-off-by: Jonathan Rajotte <jonathan.rajotte-julien@efficios.com>
Signed-off-by: Jérémie Galarneau <jeremie.galarneau@efficios.com>
Change-Id: Ic7b5d46b6793cda16a474c5459f9bf533812fe61
Jérémie Galarneau [Mon, 3 May 2021 16:50:25 +0000 (12:50 -0400)]
Fix: lttng-ctl: erroneous check if user is part of the tracing group
in_tgroup is set to `-1` whenever the current user is not part of the
tracing group _or_ if an error occurred while looking up if the user
is part of the tracing group. In other words, the value '0' is unused.
in_tgroup must be explicitly checked against '1' and can't be assumed
to behave as a boolean value.
This is _not_ a security issue: if the user is not part of the tracing
group, she will fail to open the root session damon's socket because
of the kernel-side permission checking. However, the behaviour of the
lttng client (and error reporting) will be confusing.
Signed-off-by: Jérémie Galarneau <jeremie.galarneau@efficios.com>
Change-Id: I614da0123d0546c5f54f121e8ed9716d6e292400
Jérémie Galarneau [Mon, 3 May 2021 15:52:51 +0000 (11:52 -0400)]
Build fix: kernel-consumerd: merge conflict left in file
Signed-off-by: Jérémie Galarneau <jeremie.galarneau@efficios.com>
Change-Id: I71defaee6724a08661567af879b91a57d9dae74c
Jérémie Galarneau [Fri, 30 Apr 2021 16:12:44 +0000 (12:12 -0400)]
Clean-up: consumerd: use a specific status code for get_next_subbuffer
The implementation of "get next subbuffer" of the user space and kernel
tracers return different error codes (-ENODATA and -EAGAIN) which are
are confusing to handle in the generic code.
Since the difference between -ENODATA and -EAGAIN makes no material
difference in the current consumerd implementation, those conditions
are abstracted by a common GET_NEXT_SUBBEFFER_STATUS_NO_DATA.
Otherwise, the callers handle 'OK' and the generic 'ERROR' condition
which makes the transport of more specific "errno" values useless for
the moment.
Signed-off-by: Jérémie Galarneau <jeremie.galarneau@efficios.com>
Change-Id: Ibdb2837396e4b8cd291ffd80f6ca59b39ce3f707
Mathieu Desnoyers [Fri, 30 Apr 2021 16:02:47 +0000 (12:02 -0400)]
Fix: kernel consumer: get next subbuffer EAGAIN handling
The caller of get next subbuffer (data and metadata) callbacks only
expects -ENODATA when there is no data to read. However, the kernel
tracer distinguishes between no data for a finalized stream (-ENODATA)
and no data for a non-finalized stream (-EAGAIN).
Given that the consumer daemon uses the POLLHUP returned by epoll to
detect stream end of life, it does not care about the distinction
between -EAGAIN and -ENODATA when streaming.
However, taking a snapshot of a metadata stream uses the distinction
between nodata and again. Change this so it considers a return value of
0 from lttng_consumer_read_subbuffer to mean there is no more data to
read, so we can combine -EAGAIN and -ENODATA within get next subbuffer
callbacks and return -ENODATA for both.
Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>
Signed-off-by: Jérémie Galarneau <jeremie.galarneau@efficios.com>
Change-Id: I30786985c2389570dd342e55d4faf7c552532f8b
Mathieu Desnoyers [Fri, 30 Apr 2021 14:44:44 +0000 (10:44 -0400)]
Fix: kernel consumer: signal metadata ready condition variable
The behaviour of the kernel consumer does not match that of the user
space consumer. When a live metadata stream goes back to "sleep", it
must broadcast on the metadata_rdv to wake-up any thread that is
performing a metadata sync.
This mismatch causes a hang in the kernel consumer during the kernel
clear tests.
Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>
Signed-off-by: Jérémie Galarneau <jeremie.galarneau@efficios.com>
Change-Id: Ief15fc2b7c92582ea36bf64d185720e48d2c94d5
Jérémie Galarneau [Thu, 29 Apr 2021 20:17:45 +0000 (16:17 -0400)]
Fix: consumerd: unbalanced subbuffer 'get' when checking operation availability
A WARN_ON in lttng-modules is hit in the get_subbuf ioctl, which
pointed to an unbalanced get/put pair when accessing the subbuffers
of a channel.
517716.168856] ------------[ cut here ]------------
[517716.171559] WARNING: CPU: 1 PID: 19313 at /home/efficios/git/lttng-modules/src/lib/ringbuffer/ring_buffer_frontend.c:1263 lib_ring_buffer_get_subbuf+0x24f/0x260 [lttng_lib_ring_buffer]
[517716.180096] Modules linked in: lttng_test(O) lttng_probe_x86_exceptions(O) lttng_probe_x86_irq_vectors(O) lttng_probe_writeback(O) lttng_probe_workqueue(O) lttng_probe_vmscan(O) lttng_probe_udp(O) lttng_probe_timer(O) lttng_probe_sunrpc(O) lttng_probe_statedump(O) lttng_probe_sock(O) lttng_probe_skb(O) lttng_probe_signal(O) lttng_probe_scsi(O) lttng_probe_sched(O) lttng_probe_regulator(O) lttng_probe_regmap(O) lttng_probe_rcu(O) lttng_probe_random(O) lttng_probe_printk(O) lttng_probe_power(O) lttng_probe_net(O) lttng_probe_napi(O) lttng_probe_module(O) lttng_probe_kmem(O) lttng_probe_jbd2(O) lttng_probe_irq(O) lttng_probe_i2c(O) lttng_probe_gpio(O) lttng_probe_ext4(O) lttng_probe_compaction(O) lttng_probe_btrfs(O) lttng_probe_block(O) lttng_counter_client_percpu_32_modular(O) lttng_counter_client_percpu_64_modular(O) lttng_counter(O) lttng_ring_buffer_event_notifier_client(O) lttng_ring_buffer_metadata_mmap_client(O) lttng_ring_buffer_client_mmap_overwrite(O)
[517716.180815] lttng_ring_buffer_client_mmap_discard(O) lttng_ring_buffer_metadata_client(O) lttng_ring_buffer_client_overwrite(O) lttng_ring_buffer_client_discard(O) lttng_tracer(O) lttng_statedump(O) lttng_wrapper(O) lttng_uprobes(O) lttng_clock(O) lttng_kprobes(O) lttng_lib_ring_buffer(O) lttng_kretprobes(O) [last unloaded: lttng_wrapper]
[517716.213228] CPU: 1 PID: 19313 Comm: lttng-consumerd Tainted: G O 5.11.2 #80
[517716.215573] Hardware name: QEMU Standard PC (i440FX + PIIX, 1996), BIOS Bochs 01/01/2011
[517716.220341] RIP: 0010:lib_ring_buffer_get_subbuf+0x24f/0x260 [lttng_lib_ring_buffer]
[517716.222579] Code: 50 f0 ff 00 0f 0b 49 03 5f 28 44 8b 85 80 00 00 00 49 8b 77 30 45 85 c0 48 89 d9 0f 85 4f ff ff ff e9 25 ff ff ff f0 ff 45 00 <0f> 0b b8 f0 ff ff ff e9 a6 fe ff ff 0f 1f 44 00 00 0f 1f 44 00 00
[517716.227610] RSP: 0018:
ffffbafd09023e88 EFLAGS:
00010202
[517716.229147] RAX:
0000000000000000 RBX:
ffff9906ed069a00 RCX:
ffff9905c4e4a400
[517716.231186] RDX:
ffffdafcffc4ec90 RSI:
0000000000200000 RDI:
ffffdafcffc4e9f0
[517716.233265] RBP:
ffff9905c4e4a400 R08:
0000000000300000 R09:
0000000000200000
[517716.235284] R10:
0000000000200000 R11:
0000000000000000 R12:
0000000000000000
[517716.237333] R13:
0000000000000000 R14:
000000000000005e R15:
0000000000000000
[517716.239360] FS:
00007ff327fff700(0000) GS:
ffff9905a7a40000(0000) knlGS:
0000000000000000
[517716.241634] CS: 0010 DS: 0000 ES: 0000 CR0:
0000000080050033
[517716.243262] CR2:
000055b91cf07730 CR3:
000000076f45e003 CR4:
00000000001706e0
[517716.245316] Call Trace:
[517716.246281] lib_ring_buffer_ioctl+0x181/0x300 [lttng_lib_ring_buffer]
[517716.248301] lttng_stream_ring_buffer_ioctl+0x1a3/0x200 [lttng_tracer]
[517716.252621] __x64_sys_ioctl+0x8e/0xd0
[517716.253931] do_syscall_64+0x33/0x80
[517716.255016] entry_SYSCALL_64_after_hwframe+0x44/0xa9
[517716.256448] RIP: 0033:0x7ff3372f46d7
[517716.257586] Code: b3 66 90 48 8b 05 b1 47 2d 00 64 c7 00 26 00 00 00 48 c7 c0 ff ff ff ff c3 66 2e 0f 1f 84 00 00 00 00 00 b8 10 00 00 00 0f 05 <48> 3d 01 f0 ff ff 73 01 c3 48 8b 0d 81 47 2d 00 f7 d8 64 89 01 48
[517716.262472] RSP: 002b:
00007ff327ffe2c8 EFLAGS:
00000246 ORIG_RAX:
0000000000000010
[517716.264634] RAX:
ffffffffffffffda RBX:
00007ff310002740 RCX:
00007ff3372f46d7
[517716.266674] RDX:
0000000000000000 RSI:
000000000000f605 RDI:
000000000000005e
[517716.268713] RBP:
00007ff327ffe310 R08:
00007ff310002870 R09:
a002000000000000
[517716.270732] R10:
000055992b0a6530 R11:
0000000000000246 R12:
000055992c47da70
[517716.272768] R13:
00007ff318005e80 R14:
00007ff310002740 R15:
000055992b0a6528
[517716.274800] irq event stamp:
4526705
[517716.275897] hardirqs last enabled at (
4526713): [<
ffffffff9016a474>] console_unlock+0x4b4/0x5b0
[517716.278320] hardirqs last disabled at (
4526722): [<
ffffffff9016a3d0>] console_unlock+0x410/0x5b0
[517716.280759] softirqs last enabled at (
4526658): [<
ffffffff9120030f>] __do_softirq+0x30f/0x432
[517716.285125] softirqs last disabled at (
4526653): [<
ffffffff91001052>] asm_call_irq_on_stack+0x12/0x20
[517716.287648] ---[ end trace
506e55b312b731bf ]---
The check for the availability of the 'get_next_check_metadata'
operation attempts to use the operation at the creation of the metadata
stream. Most of the time this occurs before any metadata could be
generated.
However, the check will sometimes (very rarely) occur after the
generation of some metadata causing the 'get' to succeed and,
consequently, a subbuffer to be acquired. In those cases, the subbuffer
must be released immediately.
Fixes #1313
Signed-off-by: Jérémie Galarneau <jeremie.galarneau@efficios.com>
Change-Id: I906dca7d83ce103c009aa477ae5881145e4aa081
Jérémie Galarneau [Wed, 28 Apr 2021 21:50:41 +0000 (17:50 -0400)]
sessiond: remove commented-out logging
Signed-off-by: Jérémie Galarneau <jeremie.galarneau@efficios.com>
Change-Id: I7b8991ba994585fd8f2b5c6e02b9be7cf8e7123b
Jérémie Galarneau [Wed, 28 Apr 2021 21:44:52 +0000 (17:44 -0400)]
sessiond: Remove [.*-thread] logging prefix
The logger now logs the thread name making this prefix useless.
Signed-off-by: Jérémie Galarneau <jeremie.galarneau@efficios.com>
Change-Id: Ie9093e4032d2cf0d3c0e9fa0d7eda28467033931
Francis Deslauriers [Mon, 26 Apr 2021 22:18:11 +0000 (18:18 -0400)]
run-as: reduce verbosity of fd sending error paths
Issue
=====
A testcase in `tests/regression/tools/save-load/test_save` tests that
saving a session on an already existing configuration file fails.
The test case fails as expected but it is a bit noisy in terms of error
reporting:
ok 9 - Enable channel chan-save for session save-42
ok 10 - Enable ust event tp:tptest for session save-42
Error: Attempt to send invalid file descriptor to master (fd = -1)
PERROR - 09:57:10.
893683118 [Client management]: Could not create configuration file: File exists (in save_session() at save.c:2706)
PERROR - 09:57:10.
893714862 [Main]: Failed to close result file descriptor: Bad file descriptor (in send_fds_to_master() at runas.c:824)
ok 11 - Session failed to be saved. Expected!
We see that 3 error statements are printed by the sessiond but only the
second is really relevant.
Fix
===
This commit:
- changes the first `ERR()` statement to a `DBG()` statement, and
- only call `close()` on seemingly valid FDs.
Notes
=====
This commit also removes the mention of "master" in the first `DBG()`
statement as this function is used by both the master and the runas
process.
Signed-off-by: Francis Deslauriers <francis.deslauriers@efficios.com>
Signed-off-by: Jérémie Galarneau <jeremie.galarneau@efficios.com>
Change-Id: Ie77d44233a770610f8a3f4412b84c0fd70c0812e
Francis Deslauriers [Mon, 26 Apr 2021 22:09:54 +0000 (18:09 -0400)]
Set thread name of run-as worker process
Signed-off-by: Francis Deslauriers <francis.deslauriers@efficios.com>
Signed-off-by: Jérémie Galarneau <jeremie.galarneau@efficios.com>
Change-Id: Ifd89ad19a17d96975f712bcd298588f2fc28d04f
Francis Deslauriers [Tue, 27 Apr 2021 18:26:09 +0000 (14:26 -0400)]
Use new ioctl macros with fixed ioctl direction
Update the ioctl commands to follow the changes in this commit:
commit
8c71721f7a868b575b05e24bc3a3dcc967e6d5d6
Author: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>
Date: Tue Apr 20 11:05:19 2021 -0400
Fix: LTTng-modules ABI ioctl wrong direction
This commit uses the new command first, and if -ENOSYS is returned by
the LTTng kernel tracer it falls back to the old command.
Signed-off-by: Francis Deslauriers <francis.deslauriers@efficios.com>
Signed-off-by: Jérémie Galarneau <jeremie.galarneau@efficios.com>
Change-Id: I2604ea5902fec3bf574c7a249cb65784685a0ae8
Francis Deslauriers [Tue, 27 Apr 2021 16:13:57 +0000 (12:13 -0400)]
Sync lttng-modules ABI in internal kernel-ioctl.h
This commit bumps the modules minor version number in `kernel-ioctl.h`
and applies the following renames:
define renames:
`LTTNG_KERNEL_x` -> `LTTNG_KERNEL_ABI_x`
`RING_BUFFER_x` -> `LTTNG_KERNEL_ABI_RING_BUFFER_x`
struct renames:
`struct lttng_kernel_x` -> `struct lttng_kernel_abi_x`
No behavior change intended.
Signed-off-by: Francis Deslauriers <francis.deslauriers@efficios.com>
Signed-off-by: Jérémie Galarneau <jeremie.galarneau@efficios.com>
Change-Id: I0bb0ac6bca56b6740ea2d543a6d1bd618d39b36c
Jonathan Rajotte [Tue, 27 Apr 2021 01:09:35 +0000 (21:09 -0400)]
Fix: data reception is called event when only FDS are expected for reception
In non blocking mode, the FDs reception code can hit EAGAIN. In such
cases, skip the data reception completely when performing the reception.
The same behavior could have been implemented with a noop at the
`lttcomm_recv_unix_sock_non_block` level when the passed `len` is equal
to 0. The handling of it at the notification thread reception level was
chosen to be explicit with regards to the expected behaviour and not
rely on implicit functionality of `lttcomm_recv_unix_sock_non_block`.
While there, augment assertions for code in unix.c
Change-Id: I0fa6df1c9630caa6e877a39372bdfd3d3a01b648
Signed-off-by: Jonathan Rajotte <jonathan.rajotte-julien@efficios.com>
Signed-off-by: Jérémie Galarneau <jeremie.galarneau@efficios.com>
Jonathan Rajotte [Tue, 27 Apr 2021 01:08:18 +0000 (21:08 -0400)]
Fix: FDs are added to payload despite not being received
`lttcomm_recv_fds_unix_sock_non_block` can return zero on EAGAIN.
It is important not to add the raw_fds array to the payload in such case
since concretely no FDs have been received.
Signed-off-by: Jonathan Rajotte <jonathan.rajotte-julien@efficios.com>
Signed-off-by: Jérémie Galarneau <jeremie.galarneau@efficios.com>
Change-Id: I551457856c885098bf0de2388936e97456051f93
Jonathan Rajotte [Tue, 27 Apr 2021 01:07:08 +0000 (21:07 -0400)]
Fix: initialize temporary fd array element to -1
These are FDs, an initialization value of zero is a valid FD number
hence we must initialize the element to -1.
This allows us to be more strict in `add_fds_to_payload` and assert that
the passed FDs are actual FDs and not only bare zero-initialized
element.
Change-Id: Icc05d98d51906636362a67588d7c42b58cf33ded
Signed-off-by: Jonathan Rajotte <jonathan.rajotte-julien@efficios.com>
Signed-off-by: Jérémie Galarneau <jeremie.galarneau@efficios.com>
Jérémie Galarneau [Mon, 26 Apr 2021 23:21:53 +0000 (19:21 -0400)]
Fix: common: hide shm* util symbols
Internal symbols that are part of libcommon must be marked as "hidden".
Otherwise, the symbols are exported as part of liblttng-ctl.so.
Signed-off-by: Jérémie Galarneau <jeremie.galarneau@efficios.com>
Change-Id: I41e7746a8044e85a7d422f138654de1579bf22f1
Francis Deslauriers [Thu, 25 Feb 2021 23:19:37 +0000 (18:19 -0500)]
Fix: mark channel as disabled even if the session is inactive
Observed issue
==============
When the session is stopped, disable-channel commands are no-op.
The following commands reproduce the issue:
lttng create
lttng enable-event -u -a
lttng start
sleep 4
lttng stop
lttng disable-channel -u channel0
sleep 10
lttng start
sleep 4
lttng stop
lttng view
Note that the sleep command there are to give the UST application time
to produce events.
Even after disabling the channel, we can see that events are still
traced.
This is due to the fact that the `channel_ust_disable()` function
returns early if the session is inactive and omits to set the channel as
disabled.
Proposed fix
============
Move this following line before the check:
uchan->enabled = 0;
Test
====
Add a test case to exercise this exact scenario.
Signed-off-by: Francis Deslauriers <francis.deslauriers@efficios.com>
Signed-off-by: Jérémie Galarneau <jeremie.galarneau@efficios.com>
Change-Id: I9660064ac6eb99f2aea8851dc98a94cfc1b810e6
Francis Deslauriers [Fri, 23 Apr 2021 18:45:31 +0000 (14:45 -0400)]
Fix: condition: buffer-usage: use double instead of fixed point
Issue observed
==============
When running the test_notification_ust_buffer_usage test on x86
(32 bit), the session daemon and test client both crash. The session
daemon dies while attempting to lock a NULL client list during the
execution of an enqueued action in the action executor.
See the following backtrace:
#0 0xf7c6c756 in __GI___pthread_mutex_lock (mutex=0x0) at ../nptl/pthread_mutex_lock.c:67
#1 0x565afe96 in notification_client_list_send_evaluation (client_list=0x0, trigger=0xf0f225e0, evaluation=0xf330c830, source_object_creds=0xf330e5cc, client_report=0x565cf81b <client_handle_transmission_status>, user_data=0xf330c320) at notification-thread-events.c:4372
#2 0x565cfb41 in action_executor_notify_handler (executor=0xf330c320, work_item=0xf330e5b0, item=0xf330c7b0) at action-executor.c:269
#3 0x565d1a58 in action_executor_generic_handler (executor=0xf330c320, work_item=0xf330e5b0, item=0xf330c7b0) at action-executor.c:696
#4 0x565d1b7f in action_work_item_execute (executor=0xf330c320, work_item=0xf330e5b0) at action-executor.c:715
#5 0x565d212f in action_executor_thread (_data=0xf330c320) at action-executor.c:797
#6 0x565b9d0e in launch_thread (data=0xf330c390) at thread.c:66
#7 0xf7c69fd2 in start_thread (arg=<optimized out>) at pthread_create.c:486
#8 0xf7b7f6d6 in clone () at ../sysdeps/unix/sysv/linux/i386/clone.S:108
This crash causes an assertion to fail in the test client; checking for
data pending was not expected to return a negative value. In this case,
the negative return value is justified as it is -LTTNG_ERR_NO_SESSIOND.
Cause
=====
Equipped with coffee, a debugger, and a healthy dose of print
statements, it appeared that the following was taking place:
- Register a trigger (T1): high buffer usage (0.99) -> notify (succeeds)
- Subscribe to high buffer usage (0.99) notifications (succeeds)
- Subscribe to high buffer usage (0.99) notifications
(fails duplicate, expected)
- Unregister trigger (fails unexpectedly)
- Notification client destroys its channel, causing the condition to be
unsubscribed-from
- Another test registers a trigger (T2): high buffer usage (0.90) ->
notify (succeeds)
- Session daemon evaluates a channel sample against T1's condition,
which evaluates to true and produces an "evaluation" to send to
clients
- The client list associated to T1's condition is not found (but this
isn't checked)
- An action executor work item is queued to run T1's actions (notify),
but without a client list, resulting in the crash when it is executed.
We could confirm that the client list associated to T1's condition was
created and never destroyed making the failure to find it rather
puzzling.
It turns out that the hash of T1's condition did not match the hash of
the client list's condition. This is unexpected as both conditions are
copies of one another.
It turns out that, on x86, the scheme being used to transmit the
condition's buffer usage threshold floating point value is not compiled
to numerically stable code. Serializing such a buffer condition and
creating it from the resulting payload in a loop showed that the
threshold value gradually drifted. This isn't the case on the other
architectures we support.
On x86-64, gcc makes use of SSE instructions to perform the conversion
to an integral value (with double precision). However, on x86, it makes
use of the x87 fpu stack instructions which carry 80-bit of precision
internally, resulting in a loss of precision as the value is
transformed, back and forth, between 80-bit to double precision
representations.
Solution
========
Since conditions are not carried between hosts (only between clients
and the session daemon), a fixed-point conversion scheme is unnecessary.
The 'double' value provided by the client is carried directly which
bypasses the problem completely.
Drawbacks
=========
None.
Signed-off-by: Francis Deslauriers <francis.deslauriers@efficios.com>
Signed-off-by: Jérémie Galarneau <jeremie.galarneau@efficios.com>
Change-Id: Ie524e7362626406327f4f56e1dba5c8cf469df31
Jérémie Galarneau [Wed, 28 Apr 2021 17:16:12 +0000 (13:16 -0400)]
Tests: notification: fail on trigger unregistration failure
Signed-off-by: Jérémie Galarneau <jeremie.galarneau@efficios.com>
Change-Id: I474c92f34604100081676dda484513cc8a950c73
Jérémie Galarneau [Sat, 24 Apr 2021 01:28:58 +0000 (21:28 -0400)]
Update version to v2.13.0-rc1
Signed-off-by: Jérémie Galarneau <jeremie.galarneau@efficios.com>
Jérémie Galarneau [Fri, 23 Apr 2021 15:21:13 +0000 (11:21 -0400)]
Set the 2.13 release codename and description
Signed-off-by: Jérémie Galarneau <jeremie.galarneau@efficios.com>
Change-Id: I6d388b879bc10f5fd5d869352a90701faa20a378
Jérémie Galarneau [Fri, 23 Apr 2021 22:30:22 +0000 (18:30 -0400)]
Tests: fix: test_notification_notifier_discarded_count missing renames
Some options of lttng-add-trigger were renamed without adapting the
test since that part is only ran when destructive tests are enabled.
Signed-off-by: Jérémie Galarneau <jeremie.galarneau@efficios.com>
Change-Id: Ia5b01e64a3766adced292be7f7436d742b3308c7
Jérémie Galarneau [Fri, 23 Apr 2021 22:29:29 +0000 (18:29 -0400)]
Docs: lttng-add-trigger and lttng-remove-trigger refer to owner-id
The proper name of the option is --owner-uid.
Signed-off-by: Jérémie Galarneau <jeremie.galarneau@efficios.com>
Change-Id: I8db096acc5ddd06107762a44ebfda8eaf96ba0d4
Francis Deslauriers [Fri, 23 Apr 2021 21:15:28 +0000 (17:15 -0400)]
Fix: tests: test case using old condition syntax
This commit also changes "on-event" -> "event-rule-matches"
Signed-off-by: Francis Deslauriers <francis.deslauriers@efficios.com>
Signed-off-by: Jérémie Galarneau <jeremie.galarneau@efficios.com>
Change-Id: I0b9ff1302ca5a3e8bbdddc8a626890c527dcbd59
Philippe Proulx [Fri, 23 Apr 2021 21:12:12 +0000 (17:12 -0400)]
lttng-add-trigger(1): clarify the `--rate-policy` documentation
Signed-off-by: Philippe Proulx <eeppeliteloop@gmail.com>
Signed-off-by: Jérémie Galarneau <jeremie.galarneau@efficios.com>
Change-Id: I6144521f3186e33c52d7ac20b64cf3e8f7663a55
Philippe Proulx [Wed, 3 Mar 2021 20:18:30 +0000 (15:18 -0500)]
Update some manual pages for LTTng-tools 2.13
This patch updates some manual pages for LTTng-tools 2.13. A few pages
still remain to be updated, a task which is reserved for a subsequent
patch.
This patch:
* Improves the consistency of the the command descriptions in
`configure.ac`.
* Adds `common-help-option.txt` which is the description of any
`--help` option.
* Adds `common-intro.txt` which is a common description introduction for
the top-level manual pages.
* Adds `lttng-event-rule.7.txt` which describes the common way to
specify an event rule on the command-line.
lttng-event-rule(7) has a "Migration from a recording event rule
specification" section with a table which shows the relationship
between lttng-enable-event(1) command-line arguments and
lttng-event-rule(7) options.
As of this patch, only `lttng-add-trigger.1.txt` references it,
for the `event-rule-matches` trigger condition.
`Makefile.am` is also updated to build and include manual pages of
section 7.
* Updates existing manual pages to:
* Have a style and voice which is more consistent with the
LTTng Documentation (website) for 2.13.
* Fix various terminology ambiguities.
* Use more textual variables and lists to explain more complex logic
and processes.
More specifically:
lttng-add-context(1):
Specify that this command adds context fields to be recorded
to the event records of one or more channels.
In other words, this is a recording-related command. You don't
need to use it to access context fields with the filter expression
of an event rule, for example.
lttng-add-trigger(1):
* Update the "NAME" section.
* Add internal option links where missing.
* Improve the description.
Add links to the lttng-remove-trigger(1) and
lttng-list-triggers(1) manual pages and explain what those
commands are used for.
Use "condition specifier" and "action specifier" terms to
describe those groups of options.
For condition and action specifiers, use localized synopses.
For action specifiers, add links to the corresponding LTTng
command manual pages.
* Document the `--owner-id` option.
* Group option descriptions.
* Use "name" instead of "ID".
* Refer to the new lttng-event-rule(7) manual page.
* Remove the "no context field" limitation for `ERSPEC`.
* Fix verse blocks nested in lists.
lttng-create(1):
* Add more documentation about tracing sessions.
* Specify that the `create` command can spawn a session daemon.
* Add the "Current tracing session" section to explain this
concept and where it applies.
* Clarify the "URL format" section.
lttng-disable-event(1):
Explain how this command can only find recording event rules
to disable by instrumentation point type and event name condition.
lttng-enable-event(1):
I more or less completely rewrote this page.
The document now clearly explains the related core concepts, shows
the explicit and implicit conditions of a recording event rule,
has one section for each condition explaining how an event can
satisfy it, and more.
The synopsis is more accurate.
I added an "Event record name" section to indicate what's the name
of a matched event depending on the instrumentation point type and
some command-line arguments.
I also added an "Enable a disabled recording event rule" section
to explain how the `enable-event` command enables existing,
disabled events.
This manual page now documents all the options, even if they're
the default, as defaults may change in the future.
The new lttng-event-rule(7) manual page is based on this one, but
with its own ways to specify event rule conditions.
lttng-remove-trigger(1):
* Use "name" instead of "ID".
* Use `--owner-id` instead of `--user-id`.
lttng-sessiond(8):
* Explain what an LTTng session daemon does.
* Clarify everything related to the tracing group and root
session daemon.
* Add a "Daemon configuration" section which explains the
INI configuration files and the `--config` option.
* Make the "Tracing session configuration loading" section
(renamed) much more straightforward, with less text.
* Specify that the `--daemonize` and `--background` options
make `lttng-sessiond` only exit when the daemon is ready to
receive client commands.
lttng-set-session(1):
List which commands rely on the current tracing session
concept.
lttng(1):
* Add a "Session daemon connection" section which shows how the
`lttng` tool (or any LTTng tracing control application) connects
to a session daemon (user-specific vs. root session daemon).
* Use tables to list the available commands.
Signed-off-by: Philippe Proulx <eeppeliteloop@gmail.com>
Signed-off-by: Jérémie Galarneau <jeremie.galarneau@efficios.com>
Change-Id: I6b98f4907d94763f3bfcb6576e4add9cfc59a2e3
Jérémie Galarneau [Fri, 23 Apr 2021 20:21:08 +0000 (16:21 -0400)]
.gitignore: add register-some-triggers trigger test utility
Signed-off-by: Jérémie Galarneau <jeremie.galarneau@efficios.com>
Change-Id: Iba2ccd5d23e90a421386d1ff81cde71fcbbe72fd
Francis Deslauriers [Thu, 22 Apr 2021 22:50:57 +0000 (18:50 -0400)]
Fix: event-expr.c: use-after-free and NULL ptr deref in error path
use-after-free
==============
The following functions are affected:
`lttng_event_expr_app_specific_context_field_create()`, and
`lttng_event_expr_array_field_element_create()`.
In one error path we call `lttng_event_expr_destroy()` with the
`&expr->parent` pointer (which is dynamically allocated) and this
function then calls free() on that pointer. Right after that function
call we return the pointer that was just freed.
Fix that by adding a `ret_parent_expr` pointer that is set to NULL on
the error path;
Null pointer dereference
========================
The following functions are affected:
`lttng_event_expr_app_specific_context_field_create()`,
`lttng_event_expr_array_field_element_create()`, and
`create_field_event_expr()`.
We dereference a NULL pointer if the argument sanity check fails.
Fix that by checking if `expr` is non-null before dereferencing it.
Found with scan-build.
Signed-off-by: Francis Deslauriers <francis.deslauriers@efficios.com>
Signed-off-by: Jérémie Galarneau <jeremie.galarneau@efficios.com>
Change-Id: I5fdde462d7345d0dae7ecd2e4f46473a92cd11a9
Michael Jeanson [Thu, 22 Apr 2021 22:46:06 +0000 (18:46 -0400)]
Adapt to namespaced liblttng-ust-ctl symbols
The liblttng-ust ABI bump gave us the opportunity to namespace all
public symbols under the 'lttng_ust_' prefix. All of liblttng-ust-ctl
public symbols were renamed from 'ustctl_' to 'lttng_ust_ctl_'. Adapt
the tools codebase for this API change.
Depends-on: lttng-ust: Ic2b9da019e7776b24b98824ae63d16b86c5a5742
Change-Id: I64b61a7fad86a1c0b4734a4a181f6ea7a9de7e24
Signed-off-by: Michael Jeanson <mjeanson@efficios.com>
Signed-off-by: Jérémie Galarneau <jeremie.galarneau@efficios.com>
Francis Deslauriers [Thu, 22 Apr 2021 21:39:33 +0000 (17:39 -0400)]
Cleanup: ust-app.c: Dead assignment
Found using scan-build.
Signed-off-by: Francis Deslauriers <francis.deslauriers@efficios.com>
Signed-off-by: Jérémie Galarneau <jeremie.galarneau@efficios.com>
Change-Id: I8920ee21d8202e9b94b95f517c8e1fbe8c9224a0
Francis Deslauriers [Thu, 22 Apr 2021 21:24:57 +0000 (17:24 -0400)]
Fix: tests: gen-ust-events-ns: Uninitialized argument value
If both `if (snprintf(...` of the `get_ns_inum()` function fail, the
function will not uninitialize the `ns_inum` output parameter and still
return 0. Leading to the argument `ns1` of debug_printf() being used
uninitialized.
Reported-by: scan-build.
Signed-off-by: Francis Deslauriers <francis.deslauriers@efficios.com>
Signed-off-by: Jérémie Galarneau <jeremie.galarneau@efficios.com>
Change-Id: I47286312095ca0f6a889eb0faa93661f92156ec1
Jérémie Galarneau [Thu, 22 Apr 2021 23:54:47 +0000 (19:54 -0400)]
Fix: consumerd: strlen called on uninitialized path
closed_trace_chunk_path is uninitialized when relayd_id != 0.
Signed-off-by: Jérémie Galarneau <jeremie.galarneau@efficios.com>
Change-Id: Ica3763e5e73a9f2796b96c8030c551e730fe63d8
Jérémie Galarneau [Thu, 22 Apr 2021 23:33:02 +0000 (19:33 -0400)]
Fix: sessiond: notification: leak of condition on duplicate subscription
An lttng_condition is leaked when a notification channel client attempts
to subscribe to a condition she is already subscribed to.
Signed-off-by: Jérémie Galarneau <jeremie.galarneau@efficios.com>
Change-Id: I3474ba07046c75bc3cfc2c32501d77b3ea880d33
Jonathan Rajotte [Thu, 22 Apr 2021 02:01:02 +0000 (22:01 -0400)]
Fix: action executor: deadlock on session_list_lock
Observed issue
==============
A dead lock is observed during the start-stop test suite for triggers.
Cause
=====
A start session action is executed by the action executor, the
`cmd_start_trace` function is called and effectively holds the
`session_list_lock.`. During `cmd_start_trace` a call to
`notification_thread_command_add_channel` is performed to inform the
notification thread of the new channel presence.
At the same time, a tracer event notification is received by the
notification thread. The actions are queued up and the sample of the
session id take place and a call to `session_lock_list` is performed and
blocks on the lock operation.
The notification thread wait on the `session_list_lock` and the
`session_list_lock` holder, the action executor, waits on the completion
of a command the be run by the notification thread: deadlock.
The backtrace:
Thread 6 (Thread 0x7f831c8a6700 (LWP
3046458)):
#0 syscall () at ../sysdeps/unix/sysv/linux/x86_64/syscall.S:38
#1 0x000000000053b852 in futex (uaddr=0x7f831c8a45e0, op=0, val=0, timeout=0x0, uaddr2=0x0, val3=0) at /home/joraj/lttng/master/install/include/urcu/futex.h:72
#2 0x000000000053b4f9 in futex_noasync (uaddr=0x7f831c8a45e0, op=0, val=0, timeout=0x0, uaddr2=0x0, val3=0) at /home/joraj/lttng/master/install/include/urcu/futex.h:81
#3 0x000000000053af10 in lttng_waiter_wait (waiter=0x7f831c8a45d8) at waiter.c:55
#4 0x000000000046b0f2 in run_command_wait (handle=0xe60520, cmd=0x7f831c8a4588) at notification-thread-commands.c:49
#5 0x000000000046b270 in notification_thread_command_add_channel (handle=0xe60520, session_name=0x7f8300006c30 "my_triggered_session", uid=1000, gid=1000, channel_name=0x7f82dc00be04 "channel0", key=1, domain=LTTNG_DOMAIN_UST, capacity=
2097152) at notification-thread-commands.c:184
#6 0x00000000004c7f65 in create_channel_per_uid (app=0x7f82d8000bf0, usess=0x7f8300000bb0, ua_sess=0x7f82dc002600, ua_chan=0x7f82dc00bde0) at ust-app.c:3360
#7 0x00000000004c6f98 in ust_app_channel_send (app=0x7f82d8000bf0, usess=0x7f8300000bb0, ua_sess=0x7f82dc002600, ua_chan=0x7f82dc00bde0) at ust-app.c:3514
#8 0x00000000004c6bde in ust_app_channel_create (usess=0x7f8300000bb0, ua_sess=0x7f82dc002600, uchan=0x7f8300005a90, app=0x7f82d8000bf0, _ua_chan=0x7f831c8a48b0) at ust-app.c:4771
#9 0x00000000004c6968 in find_or_create_ust_app_channel (usess=0x7f8300000bb0, ua_sess=0x7f82dc002600, app=0x7f82d8000bf0, uchan=0x7f8300005a90, ua_chan=0x7f831c8a48b0) at ust-app.c:5610
#10 0x00000000004c4f09 in ust_app_synchronize_all_channels (usess=0x7f8300000bb0, ua_sess=0x7f82dc002600, app=0x7f82d8000bf0) at ust-app.c:5820
#11 0x00000000004b958c in ust_app_synchronize (usess=0x7f8300000bb0, app=0x7f82d8000bf0) at ust-app.c:5886
#12 0x00000000004b8500 in ust_app_global_update (usess=0x7f8300000bb0, app=0x7f82d8000bf0) at ust-app.c:5960
#13 0x00000000004b7ec2 in ust_app_start_trace_all (usess=0x7f8300000bb0) at ust-app.c:5520
#14 0x0000000000444e86 in cmd_start_trace (session=0x7f8300006c30) at cmd.c:2707
#15 0x00000000004a5af9 in action_executor_start_session_handler (executor=0x7f8314004410, work_item=0x7f8314005100, item=0x7f83140050b0) at action-executor.c:342
#16 0x00000000004a537f in action_executor_generic_handler (executor=0x7f8314004410, work_item=0x7f8314005100, item=0x7f83140050b0) at action-executor.c:696
#17 0x00000000004a4dbc in action_work_item_execute (executor=0x7f8314004410, work_item=0x7f8314005100) at action-executor.c:715
#18 0x00000000004a37e6 in action_executor_thread (_data=0x7f8314004410) at action-executor.c:797
#19 0x0000000000486193 in launch_thread (data=0x7f83140044b0) at thread.c:66
#20 0x00007f8320b60609 in start_thread (arg=<optimized out>) at pthread_create.c:477
#21 0x00007f8320a87293 in clone () at ../sysdeps/unix/sysv/linux/x86_64/clone.S:95
Thread 5 (Thread 0x7f831d0a7700 (LWP
3046457)):
#0 __lll_lock_wait (futex=futex@entry=0x5e1c10 <ltt_session_list>, private=0) at lowlevellock.c:52
#1 0x00007f8320b630a3 in __GI___pthread_mutex_lock (mutex=0x5e1c10 <ltt_session_list>) at ../nptl/pthread_mutex_lock.c:80
#2 0x00000000004378c3 in session_lock_list () at session.c:156
#3 0x00000000004a871c in add_action_to_subitem_array (action=0x7f830001a730, subitems=0x7f83140051d0) at action-executor.c:1081
#4 0x00000000004a8578 in add_action_to_subitem_array (action=0x7f830001a620, subitems=0x7f83140051d0) at action-executor.c:1025
#5 0x00000000004a4922 in populate_subitem_array_from_trigger (trigger=0x7f830001a950, subitems=0x7f83140051d0) at action-executor.c:1116
#6 0x00000000004a416e in action_executor_enqueue_trigger (executor=0x7f8314004410, trigger=0x7f830001a950, evaluation=0x7f8314005190, object_creds=0x0, client_list=0x7f8314004980) at action-executor.c:924
#7 0x0000000000479481 in dispatch_one_event_notifier_notification (state=0x7f831d0a63e8, notification=0x7f8314005160) at notification-thread-events.c:4613
#8 0x0000000000472324 in handle_one_event_notifier_notification (state=0x7f831d0a63e8, pipe=65, domain=LTTNG_DOMAIN_UST) at notification-thread-events.c:4702
#9 0x0000000000472271 in handle_notification_thread_event_notification (state=0x7f831d0a63e8, pipe=65, domain=LTTNG_DOMAIN_UST) at notification-thread-events.c:4717
#10 0x00000000004695a3 in handle_event_notification_pipe (event_source_fd=65, domain=LTTNG_DOMAIN_UST, revents=1, state=0x7f831d0a63e8) at notification-thread.c:591
#11 0x000000000046849b in thread_notification (data=0xe60520) at notification-thread.c:727
#12 0x0000000000486193 in launch_thread (data=0xe60610) at thread.c:66
#13 0x00007f8320b60609 in start_thread (arg=<optimized out>) at pthread_create.c:477
#14 0x00007f8320a87293 in clone () at ../sysdeps/unix/sysv/linux/x86_64/clone.S:95
Solution
========
Instead of using session_find_by_name() which requires the
`session_list_lock`, we introduce `sample_session_id_by_name` that uses
a urcu backed data structure. This allows the sampling of the session
id without holding the session list lock. We accept the small window
where a session object is still accessible but concretely not valid
since the actual execution context will be validated at the moment of
execution. The execution side already handles the possibility that the
session is removed at that point or is not the same session. The
execution side acquires the session_list_lock for validation.
Known drawbacks
=========
None
Signed-off-by: Jonathan Rajotte <jonathan.rajotte-julien@efficios.com>
Signed-off-by: Jérémie Galarneau <jeremie.galarneau@efficios.com>
Change-Id: I5ad2c57acc0d03d2814dda59f8ecf2d831fd961e
Simon Marchi [Tue, 20 Apr 2021 02:13:28 +0000 (22:13 -0400)]
lttng add-trigger: replace --exclude-names with --exclude-name
Implement the proposal to make the --exclude-names option singular. The
user can specify it multiple times to specify multiple exclusions,
rather than specifying a comma-separated list.
Change-Id: Id95313bb5576a8d6cecb73b61b537ae190554cd0
Signed-off-by: Simon Marchi <simon.marchi@efficios.com>
Signed-off-by: Jérémie Galarneau <jeremie.galarneau@efficios.com>
Francis Deslauriers [Fri, 16 Apr 2021 15:14:21 +0000 (11:14 -0400)]
sessiond: refcount `ust_error_accounting_entry`
Keep a reference count of the `ust_error_accounting_entry` to reclaim
memory and cleanup handles as we go.
Triggers on a sessiond target all applications regardless of the UID of
both the trigger and the application. This means that whenever a new UST
counter is created for a UID, we need to keep it around as long as there
is 1. an application from that UID, or 2. an event notifier registered.
This commit achieves that by keeping a count of the registered event
notifiers. If the count is above zero, we keep a reference on every UID
entries. When the count reaches zero, we put that reference on every
entries.
The event notifier count is guarded by a mutex since it's accessed by
both the notification thread and UST registration thread.
Here is the pseudo code for the relevant error accounting events:
App registration:
lookup entry for application UID in hashtable
if found:
get a reference on that existing entry
else:
create a new entry for that UID
get a reference on that new entry
if event_notifier_count > 0:
get an extra reference on the behalf of the event notifiers
Event notifier registration:
increment event_notifier_count
if event_notifier_count == 1:
get a reference on all existing UID entries
App unregistration:
put reference on application UID entry
Event notifier unregistration:
decrement event_notifier_count
if event_notifier_count == 0:
put ref on all existing UID entries
Signed-off-by: Francis Deslauriers <francis.deslauriers@efficios.com>
Signed-off-by: Jérémie Galarneau <jeremie.galarneau@efficios.com>
Change-Id: I74667d30a5b6975bad7975b82f6819e263199cd0
Jérémie Galarneau [Thu, 22 Apr 2021 18:56:59 +0000 (14:56 -0400)]
action list: missing renames from previous name "group"
Signed-off-by: Jérémie Galarneau <jeremie.galarneau@efficios.com>
Change-Id: I9b7379d5f2ed66677666365cf5cf0b98c774f219
Jérémie Galarneau [Thu, 22 Apr 2021 18:50:55 +0000 (14:50 -0400)]
Fix: error-query: uninitialized action_index value
action_index will be used uninitialized if the user passes an action
that is not part of the trigger (an API use error).
CID
1452662 (#2 of 2): Uninitialized scalar variable (UNINIT)
14. uninit_use: Using uninitialized value action_index.is_set.
Reported-by: Coverity Scan
Signed-off-by: Jérémie Galarneau <jeremie.galarneau@efficios.com>
Change-Id: I9afc6730b5462f4bf55e0dba3454efb5b4ca22c9
Jérémie Galarneau [Thu, 22 Apr 2021 18:47:14 +0000 (14:47 -0400)]
Fix: sessiond: leak of config_path on duplicate --config option
1452207 Resource leak
The system resource will not be reclaimed and reused, reducing the
future availability of the resource.
CID
1452373 (#3 of 3): Resource leak (RESOURCE_LEAK)
26. overwrite_var: Overwriting config_path in config_path =
utils_expand_path(optarg) leaks the storage that config_path points to.
Reported-by: Coverity Scan
Signed-off-by: Jérémie Galarneau <jeremie.galarneau@efficios.com>
Change-Id: I70eff0f4de5ac9b84b699e88232669259eab54db
Jérémie Galarneau [Thu, 22 Apr 2021 18:43:02 +0000 (14:43 -0400)]
Fix: tests: base notification client: unchecked sscanf return value
CID
1407934 (#2 of 2): Unchecked return value (CHECKED_RETURN)
8. check_return: Calling sscanf without checking return value (as is
done elsewhere 17 out of 19 times).
Reported-by: Coverity Scan
Signed-off-by: Jérémie Galarneau <jeremie.galarneau@efficios.com>
Change-Id: Ifcef490aa15aea8ae4846553fbc63c596583d274
Simon Marchi [Fri, 9 Apr 2021 16:48:30 +0000 (12:48 -0400)]
Fix: argpar: Error out when passing an argument to long option that takes no argument
If the user defines a long option `--foo` and passes `--foo=arg`, the
current behavior is that the option foo is recognized and the argument
is ignored. Change that to be an error.
It is currently not possible to pass arguments to short options using an
equal sign, so this doesn't apply to short options.
Equivalent accepted change in argpar: https://review.lttng.org/c/argpar/+/5361
Change-Id: I6fd6271abb7e271704db1e60f680729931a1fbe1
Signed-off-by: Simon Marchi <simon.marchi@efficios.com>
Signed-off-by: Jérémie Galarneau <jeremie.galarneau@efficios.com>
Jérémie Galarneau [Thu, 22 Apr 2021 16:37:37 +0000 (12:37 -0400)]
Clean-up: sessiond: identation fix
Signed-off-by: Jérémie Galarneau <jeremie.galarneau@efficios.com>
Change-Id: I10e56e226c7ea3eb6eae6a2ea17f5eae7cff3b6c
Jérémie Galarneau [Thu, 22 Apr 2021 16:32:26 +0000 (12:32 -0400)]
Fix: tests: health thread stall: only stop consumerd when required
Since
a0f8e3109, stop_lttng_consumerd will report a failure when
there is no consumer daemon to kill. This fix ensures it is only
invoked for tests that launch a consumer daemon.
Signed-off-by: Jérémie Galarneau <jeremie.galarneau@efficios.com>
Change-Id: I6831fbe7563d2e1804f10588494d126fbb4202ff
Francis Deslauriers [Thu, 22 Apr 2021 15:07:26 +0000 (11:07 -0400)]
Fix: tests: notification: update kernel testcase count
Signed-off-by: Francis Deslauriers <francis.deslauriers@efficios.com>
Signed-off-by: Jérémie Galarneau <jeremie.galarneau@efficios.com>
Change-Id: Ib65e4ea8ced397bdd3b30f1401121adbfeeff23a
Jérémie Galarneau [Thu, 22 Apr 2021 14:17:04 +0000 (10:17 -0400)]
Fix: tests: integer truncation warning on 32-bit platforms
large integer implicitly truncated to unsigned type [-Woverflow]
Signed-off-by: Jérémie Galarneau <jeremie.galarneau@efficios.com>
Change-Id: I3e479d240c25550f44bbf40944484870da85a401
Jérémie Galarneau [Wed, 21 Apr 2021 04:41:09 +0000 (00:41 -0400)]
Fix: sessiond: triggers: failure to insert trigger is fatal
Inserting a trigger by name/owner should succeed if it could be added
to the top-level triggers_ht.
Signed-off-by: Jérémie Galarneau <jeremie.galarneau@efficios.com>
Change-Id: I46518b86b34cb13074723b4f2bef38026d898700
Jérémie Galarneau [Wed, 21 Apr 2021 03:57:11 +0000 (23:57 -0400)]
sessiond: notification-thread: switch logging to debug on GET_TRIGGER
Not finding a trigger is not necessarily an error (espeacially during
the execution of tests).
Signed-off-by: Jérémie Galarneau <jeremie.galarneau@efficios.com>
Change-Id: I238e6ca42f65e773e7e88773bc2e4f54b0a78ad5
Jérémie Galarneau [Mon, 19 Apr 2021 21:44:55 +0000 (17:44 -0400)]
Tests: add trigger name/unnamed backwards compatibility test
Signed-off-by: Jérémie Galarneau <jeremie.galarneau@efficios.com>
Change-Id: Ib19de93e316640fb9ba37d2df621c3c97cd712f6
Jérémie Galarneau [Tue, 20 Apr 2021 02:38:49 +0000 (22:38 -0400)]
lttng-ctl: separate support of named/unnamed trigger registration
Signed-off-by: Jérémie Galarneau <jeremie.galarneau@efficios.com>
Change-Id: I37e78344dd14d00c617cd462914dee287e3b24bb
Jérémie Galarneau [Tue, 20 Apr 2021 04:43:22 +0000 (00:43 -0400)]
sessiond: add support for anonymous triggers
Signed-off-by: Jérémie Galarneau <jeremie.galarneau@efficios.com>
Change-Id: I5b7fb29700af7ac7b633e5d73fb29f99f55ebfe8
Jérémie Galarneau [Tue, 20 Apr 2021 16:48:05 +0000 (12:48 -0400)]
Fix: lttng-ctl: assertion failure during unregistration of trigger
Issue observed
==============
lt-trigger_name: trigger.c:302: int lttng_trigger_serialize(const struct lttng_trigger *, struct lttng_payload *): Assertion `(creds->uid).is_set' failed.
Program terminated with signal SIGABRT, Aborted.
#0 0x00007fb74129eef5 in raise () from /usr/lib/libc.so.6
#1 0x00007fb741288862 in abort () from /usr/lib/libc.so.6
#2 0x00007fb741288747 in __assert_fail_base.cold () from /usr/lib/libc.so.6
#3 0x00007fb741297646 in __assert_fail () from /usr/lib/libc.so.6
#4 0x00007fb74169bab7 in lttng_trigger_serialize (trigger=0x5616f6f70060, payload=0x7ffe5819d140) at trigger.c:302
#5 0x00007fb74169cef0 in lttng_trigger_copy (trigger=0x5616f6f70060) at trigger.c:859
#6 0x00007fb74164302e in lttng_unregister_trigger (trigger=0x5616f6f70060) at lttng-ctl.c:3350
#7 0x00005616f50c675f in register_named_trigger () at trigger_name.c:295
#8 0x00005616f50c6879 in main (argc=1, argv=0x7ffe581a07d8) at trigger_name.c:343
Cause
=====
When creating a trigger instance and using it to unregister an existing
trigger, its credentials are unset (meaning 'default'). Expecting this,
lttng_unregister_trigger() copies the source trigger to change its
credentials to those of the caller.
Unfortunately, the trigger copy operation expects credentials to be set.
We don't run into this situation typically since the trigger instance
used to perform the unregistration is sourced from a listing or is the
same instance that was used to perform the registration (which sets the
credentials before serializing).
Solution
========
A proper implementation of "copy" is provided for the trigger object
itself. For its condition and action, we still use the same "trick"
of leveraging the serdes code to perform a deep-copy, keeping the change
small
Drawbacks
=========
None really, except that we lose some of the code sharing between
copy and serdes.
Signed-off-by: Jérémie Galarneau <jeremie.galarneau@efficios.com>
Change-Id: I71b7b075c959bc4935621543c4d379f62b7dabdf
Jonathan Rajotte [Tue, 20 Apr 2021 02:06:33 +0000 (22:06 -0400)]
Rename group action files to list files
Signed-off-by: Jonathan Rajotte <jonathan.rajotte-julien@efficios.com>
Signed-off-by: Jérémie Galarneau <jeremie.galarneau@efficios.com>
Change-Id: Iea4c2fe0a2d7001e8c05b4bb33c24e5641b193b0
Jonathan Rajotte [Tue, 20 Apr 2021 01:57:13 +0000 (21:57 -0400)]
Rename action group to action list
This is necessary to convey that the `group` an ordered sequence of
actions. We use the term `list` to put an emphasis on the ordered nature
of the group.
Files rename follows.
Signed-off-by: Jonathan Rajotte <jonathan.rajotte-julien@efficios.com>
Signed-off-by: Jérémie Galarneau <jeremie.galarneau@efficios.com>
Change-Id: Idb85d8ca5b60947d8dc3d43d98c8d915bf34c3f2
Jonathan Rajotte [Tue, 20 Apr 2021 01:22:09 +0000 (21:22 -0400)]
Rename on-event to event-rule-matches
Signed-off-by: Jonathan Rajotte <jonathan.rajotte-julien@efficios.com>
Signed-off-by: Jérémie Galarneau <jeremie.galarneau@efficios.com>
Change-Id: Icb596abdd0c291643f144ce274919b4094828e95
Jonathan Rajotte [Tue, 20 Apr 2021 01:03:40 +0000 (21:03 -0400)]
Move on-event*.* to event-rule-matches*.*
Code renaming follows.
Signed-off-by: Jonathan Rajotte <jonathan.rajotte-julien@efficios.com>
Signed-off-by: Jérémie Galarneau <jeremie.galarneau@efficios.com>
Change-Id: Ifb2de118546d39877e77ba30ba8720f75240eb39
Jonathan Rajotte [Fri, 16 Apr 2021 21:24:20 +0000 (17:24 -0400)]
Introduce emission site for syscall event rule
The tracer support multiple emission site configuration: entry, exit,
both.
On the CLI, the emission site is exposed directly on the event rule type
with the following syntax:
--type=syscall:entry
--type=syscall:exit
--type=syscall:entry+exit
`--type=syscall` is equivalent to `--type=syscall:entry+exit`.
Signed-off-by: Jonathan Rajotte <jonathan.rajotte-julien@efficios.com>
Signed-off-by: Jérémie Galarneau <jeremie.galarneau@efficios.com>
Change-Id: I65faea6872f121b98b781d931b6cc7301f20a0b6
Simon Marchi [Thu, 15 Apr 2021 13:07:31 +0000 (09:07 -0400)]
lttng list-triggers: handle all condition types
At the moment, only the event-rule-matches condition is handled by the
list-triggers commands. Change it to handle all existing condition
types.
Because these other condition types can't be created using the
command-line interface, add a util program that creates and registers
triggers for the purpose of the test. The test triggers with the
event-rule-matches condition are kept as is, created using the
command-line interface.
Change-Id: I6a2b0442800c93f7e8815dce96d1cf1928d5725a
Signed-off-by: Simon Marchi <simon.marchi@efficios.com>
Signed-off-by: Jérémie Galarneau <jeremie.galarneau@efficios.com>
This page took 0.057463 seconds and 5 git commands to generate.