Jérémie Galarneau [Wed, 21 Nov 2018 00:24:58 +0000 (19:24 -0500)]
Reference count ltt_session objects
The lifetime of ltt_session objects is mishandled in a number
of situations. For instance, if a rotation is launched on a
session and the session is destroyed during the rotation, it
is not possible for the session daemon to identify the completion
of the rotation. This then prevents the session daemon from
renaming the last chunk of a now-destroyed session and from
notifying clients that a new trace archive chunk is available.
This change only introduces reference counting of the ltt_session
objects; it does not change any behavior with regards to the
lifetime of the sessions themselved. Follow-up commits introduce
those changes.
Signed-off-by: Jérémie Galarneau <jeremie.galarneau@efficios.com>
Jérémie Galarneau [Fri, 16 Nov 2018 22:27:09 +0000 (17:27 -0500)]
Fix: session info reference is not put on unhandled condition type
This commit is only part of the stable-2.11 branch since the bug
results from a bad merge from master.
Signed-off-by: Jérémie Galarneau <jeremie.galarneau@efficios.com>
Michael Jeanson [Thu, 15 Nov 2018 17:24:48 +0000 (12:24 -0500)]
Fix: typo 'occured' -> 'occurred'
Signed-off-by: Michael Jeanson <mjeanson@efficios.com>
Signed-off-by: Jérémie Galarneau <jeremie.galarneau@efficios.com>
Michael Jeanson [Thu, 15 Nov 2018 17:24:47 +0000 (12:24 -0500)]
Fix: typo 'retrive' -> 'retrieve'
Signed-off-by: Michael Jeanson <mjeanson@efficios.com>
Signed-off-by: Jérémie Galarneau <jeremie.galarneau@efficios.com>
Michael Jeanson [Thu, 15 Nov 2018 17:16:37 +0000 (12:16 -0500)]
Fix: typo 'convet' -> 'convert'
Signed-off-by: Michael Jeanson <mjeanson@efficios.com>
Signed-off-by: Jérémie Galarneau <jeremie.galarneau@efficios.com>
Michael Jeanson [Thu, 15 Nov 2018 17:16:36 +0000 (12:16 -0500)]
Fix: typo 'leat' -> 'least'
Signed-off-by: Michael Jeanson <mjeanson@efficios.com>
Signed-off-by: Jérémie Galarneau <jeremie.galarneau@efficios.com>
Mathieu Desnoyers [Tue, 13 Nov 2018 17:12:21 +0000 (12:12 -0500)]
Fix: max_t/min_t macros are missing cast on input
The semantic expected from max_t and min_t is to perform the max/min
comparison in the type provided as first parameter.
Cast the input parameters to the proper type before comparing them,
rather than after. There is no more need to cast the result of the
expression now that both inputs are cast to the right type.
Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>
Signed-off-by: Jérémie Galarneau <jeremie.galarneau@efficios.com>
Mathieu Desnoyers [Tue, 13 Nov 2018 17:12:20 +0000 (12:12 -0500)]
Fix: Connect timeout arithmetic in inet/inet6 (v4)
The nanoseconds part of the timespec struct time_a is not always
bigger than time_b since it wraps around each second.
Use 64-bit arithmetic to compute the difference.
Merge/move duplicated code into utils.c.
This function is really doing two things. Split it into
timespec_to_ms() and timespec_abs_diff().
Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>
Signed-off-by: Jérémie Galarneau <jeremie.galarneau@efficios.com>
Jérémie Galarneau [Fri, 16 Nov 2018 21:46:36 +0000 (16:46 -0500)]
Fix: cmd_snapshot_record can return unexpected error codes
cmd_snapshot_record() does not correctly handle the various error
reporting conventions used by the functions it uses. In certain cases,
invalid error codes (not part of enum lttng_error_code) can bubble up
to the command handler which results in erroneous error messages being
reported by the client.
This change harmonizes the error handling and changes a number of
functions used to explicitly return 'enum lttng_error_code' instead
of 'int'.
Signed-off-by: Jérémie Galarneau <jeremie.galarneau@efficios.com>
Jérémie Galarneau [Fri, 16 Nov 2018 18:55:57 +0000 (13:55 -0500)]
Docs: document why a negative socket may be passed
The fix in
4821263c4 checks for a negative socket. While the
commit message explains the reason for the check, it is not
obvious why this check is needed when looking at the code
itself.
Signed-off-by: Jérémie Galarneau <jeremie.galarneau@efficios.com>
Philippe Proulx [Tue, 13 Nov 2018 17:52:04 +0000 (12:52 -0500)]
lttng-enable-event.1.txt: fix SDT acronym, add missing "are supported"
Signed-off-by: Philippe Proulx <eeppeliteloop@gmail.com>
Signed-off-by: Jérémie Galarneau <jeremie.galarneau@efficios.com>
Philippe Proulx [Tue, 13 Nov 2018 17:48:52 +0000 (12:48 -0500)]
lttng-enable-rotation.1.txt: --timer option should be fairly precise
Signed-off-by: Philippe Proulx <eeppeliteloop@gmail.com>
Signed-off-by: Jérémie Galarneau <jeremie.galarneau@efficios.com>
Philippe Proulx [Fri, 9 Nov 2018 17:01:37 +0000 (12:01 -0500)]
doc/man: remove AUTHORS section
Many people have contributed to LTTng-tools and the original authors are
not necessarily the top ones now. For instance, GitHub reports this top
5 as of this patch:
1. jgalar 952 commits 45,889 ++ 13,959 --
2. compudj 880 commits 56,579 ++ 28,095 --
3. PSRCode 303 commits 25,553 ++ 10,909 --
4. eepp 199 commits 12,119 ++ 6,461 --
5. jdesfossez 174 commits 32,546 ++ 3,381 --
Also, man-pages(7), the conventions for writing Linux man pages, says:
> Use of an AUTHORS section is strongly discouraged.
Let's just keep the project's copyrights and some special thanks without
selecting specific authors to list on each man page.
Signed-off-by: Philippe Proulx <eeppeliteloop@gmail.com>
Signed-off-by: Jérémie Galarneau <jeremie.galarneau@efficios.com>
Philippe Proulx [Fri, 9 Nov 2018 16:46:38 +0000 (11:46 -0500)]
lttng-enable-event.1.txt: uprobe: update style and content
* Better inline styles for option variables.
* Option descriptions moved to the OPTIONS section.
* Expand SDT acronym.
* Add note indicating that SDT probes surrounded by reference counts
are not supported.
* Use "application" and "library" instead of "executable".
Signed-off-by: Philippe Proulx <eeppeliteloop@gmail.com>
Reviewed-by: Francis Deslauriers <francis.deslauriers@efficios.com>
Signed-off-by: Jérémie Galarneau <jeremie.galarneau@efficios.com>
Philippe Proulx [Fri, 9 Nov 2018 15:28:49 +0000 (10:28 -0500)]
doc/man: add links between man pages about session rotation
Signed-off-by: Philippe Proulx <eeppeliteloop@gmail.com>
Signed-off-by: Jérémie Galarneau <jeremie.galarneau@efficios.com>
Philippe Proulx [Fri, 9 Nov 2018 00:09:26 +0000 (19:09 -0500)]
lttng-enable-event.1.txt: document dynamic user space probes
Signed-off-by: Philippe Proulx <eeppeliteloop@gmail.com>
Signed-off-by: Jérémie Galarneau <jeremie.galarneau@efficios.com>
Philippe Proulx [Fri, 9 Nov 2018 00:08:48 +0000 (19:08 -0500)]
lttng-{enable,disable}-rotation.1.txt: both types can be specified
Signed-off-by: Philippe Proulx <eeppeliteloop@gmail.com>
Signed-off-by: Jérémie Galarneau <jeremie.galarneau@efficios.com>
Philippe Proulx [Thu, 8 Nov 2018 21:37:27 +0000 (16:37 -0500)]
lttng-disable-rotation.1.txt: remove LIMITATIONS section
`lttng disable-rotation` only unsets what has been previously set by
`lttng enable-rotation`. The real limitations are for
`lttng enable-rotation` and `lttng rotate`.
Signed-off-by: Philippe Proulx <eeppeliteloop@gmail.com>
Signed-off-by: Jérémie Galarneau <jeremie.galarneau@efficios.com>
Philippe Proulx [Thu, 8 Nov 2018 21:36:57 +0000 (16:36 -0500)]
lttng-enable-rotation.1.txt: add that SIZE/PERIOD are approximations
Signed-off-by: Philippe Proulx <eeppeliteloop@gmail.com>
Signed-off-by: Jérémie Galarneau <jeremie.galarneau@efficios.com>
Philippe Proulx [Thu, 8 Nov 2018 21:36:29 +0000 (16:36 -0500)]
lttng-rotate.1.txt: lttng-relayd(8) also has --output option
Signed-off-by: Philippe Proulx <eeppeliteloop@gmail.com>
Signed-off-by: Jérémie Galarneau <jeremie.galarneau@efficios.com>
Mathieu Desnoyers [Wed, 14 Nov 2018 20:34:49 +0000 (15:34 -0500)]
Fix: sessiond: ust_app_rotate_session error handling
Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>
Signed-off-by: Jérémie Galarneau <jeremie.galarneau@efficios.com>
Mathieu Desnoyers [Wed, 14 Nov 2018 21:46:05 +0000 (16:46 -0500)]
Fix: sessiond: snapshot: handle consumer return codes
Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>
Signed-off-by: Jérémie Galarneau <jeremie.galarneau@efficios.com>
Mathieu Desnoyers [Wed, 14 Nov 2018 21:35:52 +0000 (16:35 -0500)]
Fix: consumer: snapshot error return code
Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>
Signed-off-by: Jérémie Galarneau <jeremie.galarneau@efficios.com>
Mathieu Desnoyers [Wed, 14 Nov 2018 21:28:24 +0000 (16:28 -0500)]
Fix: sessiond: consumer.c: rotation error handling
Return more specific error codes in the various error paths
of the session rotation commands.
Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>
Signed-off-by: Jérémie Galarneau <jeremie.galarneau@efficios.com>
Mathieu Desnoyers [Wed, 14 Nov 2018 21:11:48 +0000 (16:11 -0500)]
Fix: consumer: rotation error return codes
The return codes of rotation commands from consumerd are not
in sync with reality. Some are simply copy-pasted from old code.
Add new return codes to describe each error situation, and split the
"channel lookup" error from other errors so sessiond can distinguish
between an error caused by an exiting application (per-pid buffers)
(LTTCOMM_CONSUMERD_CHAN_NOT_FOUND) and an actual error while performing
the command.
Move the channel lookup into the rotation caller, because it is used by
two sub-functions: we don't want to lookup to succeed for the first and
then fail for the second.
Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>
Signed-off-by: Jérémie Galarneau <jeremie.galarneau@efficios.com>
Mathieu Desnoyers [Wed, 14 Nov 2018 22:49:21 +0000 (17:49 -0500)]
Fix: create_channel_per_pid: remove channel on error
The notification system of the session daemon tracks the lifetime
of per-pid buffers (in effect, channel keys) using two hooks:
- the successful completion of create_channel_per_pid()
- the deletion of the channel from an application's registry
Multiple error paths in create_channel_per_pid() can leave a channel
in an app's registry without notifying the notification system of its
existence.
When the channel is finally cleaned-up, the channel deletion hook will
inform the notification system that the channel should be deleted,
resulting in an error as that channel was previously unknown.
This fix ensures we remove the application's channel from its registry
on error (with the notify parameter set to 'false'), thus ensuring
that the notification system never tracks the incomplete channel.
Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>
Signed-off-by: Jérémie Galarneau <jeremie.galarneau@efficios.com>
Mathieu Desnoyers [Wed, 14 Nov 2018 22:46:26 +0000 (17:46 -0500)]
Fix: channel errors on local stress-tests
On local stress-tests with per-pid buffers, error paths triggered
when trying to create a channel on an application already going away
end up tiggering the channel destruction with "-1" socket parameter.
This function attempts to send a notification of channel destruction
(erroneously) and attempts to sum counters into a channel object which
may not have been populated yet.
Therefore, don't perform those side-effects when invoked from error path
caused by channel creation error.
Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>
Signed-off-by: Jérémie Galarneau <jeremie.galarneau@efficios.com>
Mathieu Desnoyers [Wed, 14 Nov 2018 22:45:38 +0000 (17:45 -0500)]
Fix: channel_ust_create: remove channel on agent error
Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>
Signed-off-by: Jérémie Galarneau <jeremie.galarneau@efficios.com>
Jérémie Galarneau [Wed, 14 Nov 2018 22:30:17 +0000 (17:30 -0500)]
Fix: cmd_rotate_session() returns unexpected error codes
Certain paths in cmd_rotate_session() result in its return value not
obeying the convention for commands: return LTTNG_OK on success, and a
negative LTTNG_ERR_* code on error.
This patch separates the use of 'int ret' from a separate 'enum
lttng_error_code' value to ensure 'ret' values never bubble-up to the
caller.
Note that this patch assumes that ust_app_rotate_session() returns an
lttng_error_code, which is not the case right now (upcoming patch).
Signed-off-by: Jérémie Galarneau <jeremie.galarneau@efficios.com>
Jérémie Galarneau [Wed, 14 Nov 2018 22:07:48 +0000 (17:07 -0500)]
Fix: session rotation logging statement references snapshot
Signed-off-by: Jérémie Galarneau <jeremie.galarneau@efficios.com>
Jérémie Galarneau [Wed, 14 Nov 2018 22:02:04 +0000 (17:02 -0500)]
Clean-up: typo in logging message Totation -> Rotation
Signed-off-by: Jérémie Galarneau <jeremie.galarneau@efficios.com>
Jérémie Galarneau [Wed, 14 Nov 2018 21:52:12 +0000 (16:52 -0500)]
Fix: rotation error may leave session in "ONGOING" state
The errors that can occur during the launch of a rotation may
leave a session's rotation state in the "ONGOING" state.
This means that clients polling for the rotation's state (or using the
notification channel) will never see the rotation enter the ERROR
or COMPLETED states, resulting in a hang.
This change introduces session_reset_rotation_state() which
implements the logic needed to reset a session's rotation state
and populate the result of the last rotation.
Signed-off-by: Jérémie Galarneau <jeremie.galarneau@efficios.com>
Jérémie Galarneau [Wed, 14 Nov 2018 21:00:15 +0000 (16:00 -0500)]
Fix: session_list lock must be held to launch a rotation
The timer.h functions are documented as needing the session_list
lock to be held. Since timers are setup during the launch of a
rotation, the session_list lock must be taken while a job is
executed by the rotation thread.
Signed-off-by: Jérémie Galarneau <jeremie.galarneau@efficios.com>
Jérémie Galarneau [Wed, 14 Nov 2018 20:43:49 +0000 (15:43 -0500)]
Add an ASSERT_LOCKED(lock) macro
This macro validates that a given lock is taken by using
pthread_mutex_trylock().
The use of this macro is disouraged in hot paths for performance
reasons. It is meant to catch errors in "cold" code path (e.g. error
handling) where the additional safety it provides has no material
effect on performance.
Signed-off-by: Jérémie Galarneau <jeremie.galarneau@efficios.com>
Jérémie Galarneau [Wed, 14 Nov 2018 20:32:24 +0000 (15:32 -0500)]
Clean-up: move global sessiond symbols out of main.o
The global symbols defined in main.c prevent the use of numerous
functions from various other files. For instance, using timer.h
utils from session.o forces the redefinition of symbols that
are defined in main.o in the session unit tests.
This is a problem that has often occured in the past, more or less
forcing the unit tests to link against all sessiond objects. Moving
the global symbols to their own files (thread-utils.c,
process-utils.c, globals.c, and ready.c) does not solve this problem,
but it allows the unit tests to link against their resulting
objects and forego the stub-ing of symbols.
In the future, we could consider moving most the session daemon
to an internal library which could be re-used by the unit tests.
Signed-off-by: Jérémie Galarneau <jeremie.galarneau@efficios.com>
Jérémie Galarneau [Mon, 12 Nov 2018 21:52:26 +0000 (16:52 -0500)]
Fix: split index and data file rotation logic
Issue
---
There is no guarantee that index and data positions (sequence numbers)
match at a given time as both are received on different sockets.
Currently, the relay determines whether or not it should rotate a
stream's data and index files based on the lowest sequence number
of both the data and index file.
When the data connection "lags" behind the control connection, index
entries received on the control connection that belong in the "next"
chunk are written to the "previous" chunk's index file.
While there is logic to ensure that trace data received after the
current rotation position is copied to the "previous" chunk when the
rotation is performed, the same is not true for the index file
rotation.
This is fine as it should not be needed. The consumerd <-> relayd
protocol ensures that a rotation command issued on a stream will
provide a rotation position that is either:
- the stream's current position,
- a position that will be reached in the future.
Then, the control connection will receive index entries that are
either <= to the rotation target sequence number.
Therefore, it is correct to check if an index file should be rotated
everytime an index entry is flushed and change the index output file
when the rotation position is reached.
Again, this is not true for the data connection as it would be
possible to receive a rotation command with a position that is
before the current data sequence number.
Solution
---
This change splits the logic to evaluate the moment at which a
rotation should be performed (changing the actual file to which data
is written) so that both index and data files are rotated
independently.
Hence, when an index is flushed, the relay will always evaluate
whether or not it should switch the index file to its new destination
(i.e. perform the rotation of the index).
The "data" reception logic remains mostly unchanged, except that
the trace sequence number is no longer used. Only the data stream's
position is considered to evaluate whether the data stream file
should be rotated.
Signed-off-by: Jérémie Galarneau <jeremie.galarneau@efficios.com>
Jérémie Galarneau [Mon, 12 Nov 2018 22:48:53 +0000 (17:48 -0500)]
Fix: trace_archive_id is not sent in add_stream command
The trace_archive_id argument of relayd_add_stream() is not
forwarded to the "2.11" implementation of the command. This means
that all streams on the relayd are created in the "chunk_id == 0"
EPOCH.
This may cause remote rotations to never complete when a stream
is created while a rotation is ongoing. In such a case, the
stream's chunk_id would be 0, but it would not have a rotation
sequence number. This would cause the rotation completion check
to always return that the rotation is "ongoing".
Signed-off-by: Jérémie Galarneau <jeremie.galarneau@efficios.com>
Jérémie Galarneau [Mon, 12 Nov 2018 22:26:46 +0000 (17:26 -0500)]
Fix: condition unsubscription error leaves session locked
Signed-off-by: Jérémie Galarneau <jeremie.galarneau@efficios.com>
Jérémie Galarneau [Mon, 12 Nov 2018 22:25:25 +0000 (17:25 -0500)]
Fix: rotation thread does not unregister from RCU on init error
Reported-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>
Signed-off-by: Jérémie Galarneau <jeremie.galarneau@efficios.com>
Jérémie Galarneau [Thu, 8 Nov 2018 17:09:04 +0000 (12:09 -0500)]
Remove unused nr_stream_rotate_pending from consumer channel
Signed-off-by: Jérémie Galarneau <jeremie.galarneau@efficios.com>
Jérémie Galarneau [Tue, 30 Oct 2018 12:47:52 +0000 (13:47 +0100)]
Fix: session destruction blocks indefinitely if rotation is ongoing
Issue
---
The destruction of an active session can hang indefinitely if it
occurs while a rotation is ongoing. This was observed when automatic
session rotations were scheduled on a time basis.
The destruction of the session causes it to be stopped. The 'stop'
command causes the session's timers to be stopped. These timers
include the rotation pending check timer.
Meanwhile, 'data pending' queries are performed against the session
until one of them returns that no data is pending.
The 'data pending' check returns that data is pending if a session
rotation is ongoing at the moment of the check.
Hence, stopping the rotation completion check timer causes the
session to remain in the 'session ongoing' state forever and
prevents the session destruction from completing.
Solution
---
The session's rotation schedule timer is correctly stopped when
a 'stop' is performed; we don't want new rotations to be issued
from this point. However, it is incorrect to stop the
'rotation pending check' timer at this stage if a rotation is
ongoing.
This commit leaves the 'rotation pending check' timer running,
allowing the rotation thread to update the session's rotation
state on completion of the rotation. The operations that were
performed as part of the stop command, namely renaming the
'current' chunk, are then performed from the context of the
rotation thread.
Signed-off-by: Jérémie Galarneau <jeremie.galarneau@efficios.com>
Jérémie Galarneau [Tue, 30 Oct 2018 13:15:27 +0000 (14:15 +0100)]
Clean-up: remove non-existent function's declaration
Signed-off-by: Jérémie Galarneau <jeremie.galarneau@efficios.com>
Mathieu Desnoyers [Wed, 17 Oct 2018 21:16:12 +0000 (17:16 -0400)]
Always choose large event header for UST channels
UST can receive the session start command before all probe provider
library constructors have completed running, therefore finding less
events than eventually enabled within the process. Moreover, with
per-uid buffers, many processes end up registering events into shared
buffers. Therefore, the guess based on number of events from the first
process to use the buffer is incorrect.
Considering that we typically have applications with more than 30
events, we will modify the session daemon so it selects the "large"
header type independently of the number of events.
Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>
Signed-off-by: Jérémie Galarneau <jeremie.galarneau@efficios.com>
Jérémie Galarneau [Fri, 12 Oct 2018 23:49:42 +0000 (19:49 -0400)]
Fix relayd: stream index file created in the wrong directory
This fix addresses an issue that can cause a stream's index
file to be created in the wrong trace archive chunk's
directory.
The data connection creates a stream's first index file.
This can happen _after_ a ROTATE_STREAM command. More specifically,
the data of the first packet of a stream can be received after a
ROTATE_STREAM command.
The ROTATE_STREAM command changes the streams path_name to point to
the "next" chunk. If a rotation is pending for a stream, as
indicated by "rotate_at_seq_num != -1ULL", it means that we are still
receiving data that belongs in the stream's former path.
In fact, we may have never received any data for this stream at this
point.
In this specific case, we must ensure that the index file is created
in the streams's former path, "prev_path_name", on reception of the
first packet's data on the data connection.
All other rotations beyond the first one are not affected by this
problem since the actual rotation operation creates the new chunk's
index file.
Signed-off-by: Jérémie Galarneau <jeremie.galarneau@efficios.com>
Jérémie Galarneau [Fri, 12 Oct 2018 23:49:18 +0000 (19:49 -0400)]
relayd: add payload logging to session rotation commands
Signed-off-by: Jérémie Galarneau <jeremie.galarneau@efficios.com>
Jérémie Galarneau [Fri, 12 Oct 2018 22:34:06 +0000 (18:34 -0400)]
relayd: rename stream prev_seq to prev_data_seq
Since there are now two "previous sequence numbers" that are
tracked, it makes sense to give them more descriptive names.
Signed-off-by: Jérémie Galarneau <jeremie.galarneau@efficios.com>
Jérémie Galarneau [Fri, 12 Oct 2018 22:31:51 +0000 (18:31 -0400)]
Fix: take index seq number into account for rotation pending check
The rotation pending check is only performed on the sequence number of
the received data. However, it is expected that the index of the
stream has been written to disk by the time this check returns that
no rotation is pending.
This patch ensures that the minimum between the data and index
sequence numbers are used to perform this check.
Signed-off-by: Jérémie Galarneau <jeremie.galarneau@efficios.com>
Jérémie Galarneau [Fri, 12 Oct 2018 22:22:35 +0000 (18:22 -0400)]
Fix: take index sequence number into account for data pending check
The data pending checks are only performed on the sequence number of
the received data. However, it is expected that the index of the
stream (when applicable) has been written to disk by the time this
check returns that no data is pending.
This patch ensures that the minimum between the data and index
sequence numbers are used to perform this check.
Signed-off-by: Jérémie Galarneau <jeremie.galarneau@efficios.com>
Jérémie Galarneau [Fri, 12 Oct 2018 22:05:10 +0000 (18:05 -0400)]
relayd: keep track of prev_index_seq in relayd_stream
The rotation and data pending checks are only performed on the
sequence number of the received data. However, it is expected
that the index of the stream (when applicable) has been written
to disk when those checks say that their respective operations
have completed.
This patch only introduces a new 'prev_index_seq' position that
is updated when an index is flushed to disk.
A follow-up fix addresses the issue mentioned above.
Signed-off-by: Jérémie Galarneau <jeremie.galarneau@efficios.com>
Jérémie Galarneau [Fri, 5 Oct 2018 15:55:39 +0000 (11:55 -0400)]
Fix: session conditions not evaluated at subscription/registration
Conditions bound to sessions (session rotation ongoing/completed)
are not automatically evaluated when a notification channel client
subscribes to them or when a client is subscribed _before_ the
trigger is created.
The problematic scenario is:
- Trigger is registered to notify on session rotation ongoing for
session 'foo',
- A rotation is launched on session foo (but not completed)
- A client subscribes to 'session rotation ongoing' notifications for
session 'foo'
In this scenario, the client would not be notified of the 'current'
state of the session.
Whether or not a client is notified of the 'current' state at the
time of subscription/registration is defined per-condition. In
the case of 'session rotation ongoing', it is desirable for clients
to be notified that the rotation is ongoing at the time of their
subscription/registration.
Signed-off-by: Jérémie Galarneau <jeremie.galarneau@efficios.com>
Jérémie Galarneau [Fri, 5 Oct 2018 16:06:37 +0000 (12:06 -0400)]
Remove unnecessary check of output parameter
It is not necessary to check for `_notification != NULL` as it
is done at the beginning of the function. Moreover, it confuses
Coverity which warns that `notification` will be leaked if the
output parameter is NULL.
Signed-off-by: Jérémie Galarneau <jeremie.galarneau@efficios.com>
Jérémie Galarneau [Thu, 4 Oct 2018 21:50:27 +0000 (17:50 -0400)]
Allow get_next_notification to return when interrupted
Applications (and scripts) which consume a given set of notifications
indefinitely may fail to exit if a SIGTERM handler is registered.
lttng_notification_channel_get_next_notification() blocks indefinitely
on recvmsg() until a new notification is available. The wrapper that
is used to do so automatically restarts the recvmsg() if it is
interrupted, thus not allowing clients a change to cleanly exit.
This change causes the notification channel to wait for a message to
be available using select() before starting the actual reception of
the data and return LTTNG_NOTIFICATION_CHANNEL_STATUS_INTERRUTPED if a
signal occurs during the wait.
If data is available, it is assumed that the message is well-formed
and can promptly be received in its entirety. The goal of this change
is to give a monitoring application a chance to leave the
get_next_notification() function and check if it should exit.
Signed-off-by: Jérémie Galarneau <jeremie.galarneau@efficios.com>
Jérémie Galarneau [Thu, 4 Oct 2018 02:05:32 +0000 (22:05 -0400)]
Fix: register rotation thread as RCU thread
Signed-off-by: Jérémie Galarneau <jeremie.galarneau@efficios.com>
Jérémie Galarneau [Thu, 4 Oct 2018 00:50:35 +0000 (20:50 -0400)]
Docs: comment typo fix (accomodates -> accommodates)
Signed-off-by: Jérémie Galarneau <jeremie.galarneau@efficios.com>
Jérémie Galarneau [Wed, 3 Oct 2018 15:28:11 +0000 (11:28 -0400)]
Fix: uninitialized variable may be used in local rotation check
** CID
1395985: Uninitialized variables (UNINIT)
'ret' may be left uninitialized if no consumer daemons are
iterated-upon to perform a local rotation pending check.
In practice this wont' happen as that would mean that the
ltt_session has no user space nor kernel session action, thus
no rotation would be launched.
Reported-by: Coverity Scan
Signed-off-by: Jérémie Galarneau <jeremie.galarneau@efficios.com>
Jérémie Galarneau [Tue, 2 Oct 2018 18:06:17 +0000 (14:06 -0400)]
Rename sessiond-timer.[hc] to timer.[hc]
There is no need to namespace the timer files as they are already
contained withing the lttng-sessiond directory.
Signed-off-by: Jérémie Galarneau <jeremie.galarneau@efficios.com>
Jérémie Galarneau [Fri, 21 Sep 2018 22:16:05 +0000 (18:16 -0400)]
Fix: rotation may never complete in per-PID buffering mode
Issue
-----
The current scheme to ensure that a rotation is completed
consists in the following, from the session daemon's perspective:
Iterate on all channels:
- Ask the consumerd to sample the current "write" positions
- Increment a count of channels being rotated
Wait for the consumer daemon to notify the session daemon every time
a channel's streams's "read" position have all reached the sampled
"write" position.
The idea behind this is making sure that all the data that was
produced before a rotation was triggered has been consumed (i.e.
been written to a local FS or streamed to the relay daemon) before
marking the rotation as completed.
However, this assumes that the session daemon is always aware of
all channels/streams that exist at the moment at which the rotation is
initiated. This is only true for the kernel domain.
In per-PID buffer mode, it is possible for an application, and its
buffers, to be torn down at any moment. Thus the following scenario
can happen:
- The application fills its buffers, causing the consumerd to fall
behind
- The application exits, leaving its full buffers behind to be
extracted by the consumer daemon
- The session daemon removes anything to do with the application from
its internal structures, including its channels
- A rotation is initiated
- The positions of the application's buffers are never sampled as the
session daemon does not see the channels when iterating on the
session's channels
Multiple bad things can happen from there.
First, the rotation can be marked as "completed" while the consumerd
is still exctracting the dead application's buffers, causing readers
to consume an incomplete/corrupted trace.
Second, if the session is being streamed to a relay daemon, it is
possible for the 'rename' command to be issued before the contents
of the buffers has been written causing indexes to fail to be
flushed (as the relay daemon attempts to write them to a now-defunct
location).
Solution
--------
Eliminate the pipe between the session daemon and consumer daemon that
is used to signify that a rotation is completed as the information is
unreliable.
The rotation thread now periodically asks the consumer daemon to check
for channels that have a pending rotation for a given session_id or
that belong to the ongoing rotation archive id.
Hence, for every stream:
- If the archive id during which it was created is '>' than that of
the ongoing rotation, we don't need to consider it
- If the current position is '>=' than the sampled rotation position,
we can consider its rotation 'done'
- If it belongs to the pending rotation archive id and doesn't have
a "target" position, it was unknown to the session daemon and the
application associated with it is dead. We must wait for the
stream to be flushed and torn down before assuming that the
rotation was completed.
Drawbacks
---------
This polling approach is somewhat inefficient and can cause rotations
to take longer to complete than necessary, especially in high-latency
networking conditions.
Signed-off-by: Jérémie Galarneau <jeremie.galarneau@efficios.com>
Jonathan Rajotte [Tue, 11 Sep 2018 00:09:15 +0000 (20:09 -0400)]
Fix: perform local data pending before checking data pending with relayd
Performing the data pending check in two phases, local and network,
reduces the total number network operations needed.
Doing the local check first enable early return in cases where data is
still pending locally.
Signed-off-by: Jonathan Rajotte <jonathan.rajotte-julien@efficios.com>
Signed-off-by: Jérémie Galarneau <jeremie.galarneau@efficios.com>
Jérémie Galarneau [Tue, 18 Sep 2018 01:18:33 +0000 (21:18 -0400)]
Fix: missing header breaks the cygwin build
stddef.h must be included to use ssize_t.
Signed-off-by: Jérémie Galarneau <jeremie.galarneau@efficios.com>
Jonathan Rajotte [Tue, 11 Sep 2018 00:09:11 +0000 (20:09 -0400)]
Fix: double put on error path
Let relay_index_try_flush be responsible for the self-reference put on
error path.
Code flow of relay_index_try_flush is a bit tricky but the only error
flow (via relay_index_file_write) will always mark the index as flushed
and perform the self-reference put.
Signed-off-by: Jonathan Rajotte <jonathan.rajotte-julien@efficios.com>
Signed-off-by: Jérémie Galarneau <jeremie.galarneau@efficios.com>
Jonathan Rajotte [Tue, 11 Sep 2018 00:09:14 +0000 (20:09 -0400)]
Fix: holding the stream lock does not equate to having data pending
The live timer can hold the stream lock while sending empty beacon. An
empty beacon does not mean that data is still pending for the stream.
Signed-off-by: Jonathan Rajotte <jonathan.rajotte-julien@efficios.com>
Signed-off-by: Jérémie Galarneau <jeremie.galarneau@efficios.com>
Jonathan Rajotte [Tue, 11 Sep 2018 00:09:13 +0000 (20:09 -0400)]
Fix: skip uid registry when metadata key is 0
A value of zero for the metadata key indicate that metadata was never
created/pushed to the consumer.
This can occur in scenario were a tracker is present since metadata
might never be created/pushed.
Signed-off-by: Jonathan Rajotte <jonathan.rajotte-julien@efficios.com>
Signed-off-by: Jérémie Galarneau <jeremie.galarneau@efficios.com>
Jérémie Galarneau [Mon, 17 Sep 2018 22:15:11 +0000 (18:15 -0400)]
Docs: document the meaning of a ust app channel key set to 0
Signed-off-by: Jérémie Galarneau <jeremie.galarneau@efficios.com>
Jérémie Galarneau [Tue, 11 Sep 2018 00:09:12 +0000 (20:09 -0400)]
Fix: acquire stream lock during kernel metadata snapshot
The stream lock is not taken when interacting with the kernel
metadata stream that is created at the time a snapshot is taken.
This was noticed while reviewing the code for an unrelated reason,
so there is no known problem caused by this. Nevertheless, this
is incorrect as the stream is globally visible in the consumer.
Moreover, the stream was not cleaned-up which can cause a leak
whenever a metadata snapshot fails.
Signed-off-by: Jérémie Galarneau <jeremie.galarneau@efficios.com>
Signed-off-by: Jonathan Rajotte <jonathan.rajotte-julien@efficios.com>
Jonathan Rajotte [Fri, 7 Sep 2018 19:18:38 +0000 (15:18 -0400)]
Fix: skip closed session on viewer listing
There is no value in listing a closed session. A viewer cannot hook
itself to a closed session in live mode and the session is about to be
removed from the sessions hash table.
Signed-off-by: Jonathan Rajotte <jonathan.rajotte-julien@efficios.com>
Signed-off-by: Jérémie Galarneau <jeremie.galarneau@efficios.com>
Jonathan Rajotte [Fri, 7 Sep 2018 19:18:37 +0000 (15:18 -0400)]
Fix: use LTTNG_VIEWER_ATTACH_UNK to report a closed session
LTTNG_VIEWER_NEW_STREAMS_HUP is not a valid error number for the
LTTNG_VIEWER_ATTACH_SESSION command. This result in erroneous error
reporting on the client side.
Signed-off-by: Jonathan Rajotte <jonathan.rajotte-julien@efficios.com>
Signed-off-by: Jérémie Galarneau <jeremie.galarneau@efficios.com>
Jonathan Rajotte [Mon, 17 Sep 2018 16:19:40 +0000 (12:19 -0400)]
Doc: withinin -> within
Signed-off-by: Jonathan Rajotte <jonathan.rajotte-julien@efficios.com>
Signed-off-by: Jérémie Galarneau <jeremie.galarneau@efficios.com>
Jonathan Rajotte [Wed, 6 Jun 2018 16:06:04 +0000 (12:06 -0400)]
Fix: cleanup relayd sockets on rotation command communication error
Signed-off-by: Jonathan Rajotte <jonathan.rajotte-julien@efficios.com>
Signed-off-by: Jérémie Galarneau <jeremie.galarneau@efficios.com>
Jonathan Rajotte [Wed, 6 Jun 2018 01:00:28 +0000 (21:00 -0400)]
Fix: perform relayd socket pair cleanup on control socket error
A reference to the local context for the socket pair is used to "force" an
evaluation of the data and metadata streams since we changed the endpoint
status. This imitates what is currently done for the data socket.
This prevents hitting network timeouts multiple times in a row when an
error occurs. For now, there is no mechanism for retry hence
"terminating" all communication make sense and prevent unwanted delays
on operation.
Signed-off-by: Jonathan Rajotte <jonathan.rajotte-julien@efficios.com>
Signed-off-by: Jérémie Galarneau <jeremie.galarneau@efficios.com>
Jérémie Galarneau [Thu, 13 Sep 2018 21:04:45 +0000 (17:04 -0400)]
Fix: relayd control socket mutex is not destroyed
Signed-off-by: Jérémie Galarneau <jeremie.galarneau@efficios.com>
Jonathan Rajotte [Tue, 3 Jul 2018 18:49:23 +0000 (14:49 -0400)]
Tests: do not bound test app iterations when in background mode
On systems with a high number of CPUs and slow disk, taking snapshots
can take a long time. When running a long regression test, the tests
sometimes outlive the test application.
The test application then exits since the required number of
iterations was completed
(NR_ITER=
2000000).
Set the iterations parameter to -1 to ensure the application keeps
producing events for the duration of the test.
Signed-off-by: Jonathan Rajotte <jonathan.rajotte-julien@efficios.com>
Signed-off-by: Jérémie Galarneau <jeremie.galarneau@efficios.com>
Jérémie Galarneau [Tue, 11 Sep 2018 19:11:39 +0000 (15:11 -0400)]
Tests: add missing rotation and autoload tests to check target
Signed-off-by: Jérémie Galarneau <jeremie.galarneau@efficios.com>
Jonathan Rajotte [Tue, 5 Jul 2016 19:23:42 +0000 (15:23 -0400)]
Tests: remove temporary folder
Signed-off-by: Jonathan Rajotte <jonathan.rajotte-julien@efficios.com>
Signed-off-by: Jérémie Galarneau <jeremie.galarneau@efficios.com>
Jonathan Rajotte [Tue, 5 Jul 2016 18:38:46 +0000 (14:38 -0400)]
Tests: remove mi result files when done
Signed-off-by: Jonathan Rajotte <jonathan.rajotte-julien@efficios.com>
Signed-off-by: Jérémie Galarneau <jeremie.galarneau@efficios.com>
Jonathan Rajotte [Wed, 14 Oct 2015 13:57:42 +0000 (09:57 -0400)]
Tests: Remove unused set +x
Signed-off-by: Jonathan Rajotte <jonathan.rajotte-julien@efficios.com>
Signed-off-by: Jérémie Galarneau <jeremie.galarneau@efficios.com>
Jonathan Rajotte [Tue, 6 Oct 2015 21:10:56 +0000 (17:10 -0400)]
Tests: Kill relayd after sessiond to ensure a clean tear down
Signed-off-by: Jonathan Rajotte <jonathan.rajotte-julien@efficios.com>
Signed-off-by: Jérémie Galarneau <jeremie.galarneau@efficios.com>
Jonathan Rajotte [Tue, 6 Oct 2015 16:07:41 +0000 (12:07 -0400)]
Tests: Remove unused variable
Signed-off-by: Jonathan Rajotte <jonathan.rajotte-julien@efficios.com>
Signed-off-by: Jérémie Galarneau <jeremie.galarneau@efficios.com>
Jonathan Rajotte [Wed, 30 Sep 2015 22:41:30 +0000 (18:41 -0400)]
Tests: Use stop relayd from utils.sh
Signed-off-by: Jonathan Rajotte <jonathan.rajotte-julien@efficios.com>
Signed-off-by: Jérémie Galarneau <jeremie.galarneau@efficios.com>
Jonathan Rajotte [Wed, 30 Sep 2015 22:38:13 +0000 (18:38 -0400)]
Tests: remove declaration already present in utils.sh
Signed-off-by: Jonathan Rajotte <jonathan.rajotte-julien@efficios.com>
Signed-off-by: Jérémie Galarneau <jeremie.galarneau@efficios.com>
Anders Wallin [Thu, 26 Jul 2018 07:46:28 +0000 (09:46 +0200)]
Tests: added test_autoload to noinst_SCRIPTS
Signed-off-by: Anders Wallin <wallinux@gmail.com>
Signed-off-by: Jérémie Galarneau <jeremie.galarneau@efficios.com>
Francis Deslauriers [Fri, 7 Sep 2018 14:40:04 +0000 (10:40 -0400)]
Fix: Memory leak on run_as worker restart error path
Reported-by: Coverity (1395614) Resource leak
Signed-off-by: Francis Deslauriers <francis.deslauriers@efficios.com>
Signed-off-by: Jérémie Galarneau <jeremie.galarneau@efficios.com>
Jérémie Galarneau [Fri, 7 Sep 2018 01:39:18 +0000 (21:39 -0400)]
Fix: non-zero return of open handled as error
The open() run_as wrapper marks any non-zero return value
of open() as an error, causing the transmission of the file
descriptor to be skipped.
Signed-off-by: Jérémie Galarneau <jeremie.galarneau@efficios.com>
Jérémie Galarneau [Fri, 7 Sep 2018 01:25:13 +0000 (21:25 -0400)]
Fix: global run_as worker lock released during restart
The global run_as should not be released during the restart of
the working as other threads could then start dispatching commands
while the worker is recovering from an error.
Signed-off-by: Jérémie Galarneau <jeremie.galarneau@efficios.com>
Jérémie Galarneau [Thu, 6 Sep 2018 22:11:25 +0000 (18:11 -0400)]
Fix: runas worker attempts to send invalid fd to master
Commands which return a file descriptor (i.e. RUN_AS_OPEN) attempt
to send the resulting file descriptor even on failure. However,
this is not permitted by the UNIX socket interface.
As a result, skip the reception of the file descriptor payload
when a command fails. The 'master' end is also adapted to skip
the reception of the file descriptor in the case of an error.
A check has also been added to ensure that the 'master' end does
not attempt to send invalid file descriptors to the worker process.
Signed-off-by: Jérémie Galarneau <jeremie.galarneau@efficios.com>
Jérémie Galarneau [Thu, 6 Sep 2018 21:40:06 +0000 (17:40 -0400)]
Fix runas: don't attempt close negative fd
Signed-off-by: Jérémie Galarneau <jeremie.galarneau@efficios.com>
Francis Deslauriers [Wed, 5 Sep 2018 03:24:19 +0000 (23:24 -0400)]
Fix: tests: missing frame pointer for callstack test on some compiler
The callstack testcase fails when the testapp is built with gcc 8. This
is because GCC8 may not emit frame pointers even when the
`-fno-omit-frame-pointer` is used.
To prevent that we manually mark these functions with optimization level
0.
On Clang we also need to include the `-mno-omit-leaf-frame-pointer` flag
along side with the existing `-fno-omit-frame-pointer` to ensure that
frame pointers are emitted. It's not clear if this incompatibility with
GCC is expected [1].
[1]: https://bugs.llvm.org/show_bug.cgi?id=9825
Signed-off-by: Francis Deslauriers <francis.deslauriers@efficios.com>
Signed-off-by: Jérémie Galarneau <jeremie.galarneau@efficios.com>
Jérémie Galarneau [Fri, 31 Aug 2018 21:04:45 +0000 (17:04 -0400)]
Add release name and description to configure.ac
Signed-off-by: Jérémie Galarneau <jeremie.galarneau@efficios.com>
Jérémie Galarneau [Fri, 31 Aug 2018 20:50:02 +0000 (16:50 -0400)]
Update version to v2.11.0-rc1
Signed-off-by: Jérémie Galarneau <jeremie.galarneau@efficios.com>
Jérémie Galarneau [Fri, 31 Aug 2018 18:49:24 +0000 (14:49 -0400)]
Missing kernel test files in dist target
Signed-off-by: Jérémie Galarneau <jeremie.galarneau@efficios.com>
Francis Deslauriers [Fri, 31 Aug 2018 15:59:56 +0000 (11:59 -0400)]
elf: support dynamic symbol table lookup
Background
==========
There may be two symbol tables in a shared object or executable. The
normal symbol table (.symtab) and the dynamic symbol table (.dynsym).
The normal symbol table contains lots of information, such as static
linking data, but none of it is used at runtime. This is why some
shared libraries are 'stripped', reducing the final size of the file.
Stripping an object file removes the entire .symtab section of the elf
file, amongst other things.
The dynamic symbol table contains symbols that are needed for dynamic
linking of the shared object. The symbols in that section form a subset
of the symbols contained in the normal symbol section (before
stripping). The .dynsym section is left untouched when stripping a file
as it is needed at runtime.
Current limitation
==================
The current elf parsing implementation looks for the normal symbol
section (.symtab) to find the target symbol. If the .symtab is not
found, the parsing stops and returns that the symbol was not found. As
explained in the section above, a shared library might be stripped from
its normal symbol table, but still have a dynamic symbol table (.dynsym)
containing the information of the target symbol. For example, on
distributions where libc is stripped, the malloc symbol can only be
found in the .dynsym section
Solution
========
Look for the normal symbol section first and, if it's found, use it to
find the symbol, as was previously done. If the .symtab is absent,
try to use the dynamic symbol section instead.
This commit also adds a testcase for this feature.
Signed-off-by: Francis Deslauriers <francis.deslauriers@efficios.com>
Signed-off-by: Jérémie Galarneau <jeremie.galarneau@efficios.com>
Jérémie Galarneau [Fri, 31 Aug 2018 17:56:04 +0000 (13:56 -0400)]
Fix: leak of event attributes on copy failure
Reported-by: Coverity Scan (1243042 Resource leak)
Signed-off-by: Jérémie Galarneau <jeremie.galarneau@efficios.com>
Jérémie Galarneau [Fri, 31 Aug 2018 17:47:36 +0000 (13:47 -0400)]
Test fix: check length of input string
Reported-by: Coverity Scan (395327 Unbounded source buffer)
Signed-off-by: Jérémie Galarneau <jeremie.galarneau@efficios.com>
Jérémie Galarneau [Fri, 31 Aug 2018 17:41:00 +0000 (13:41 -0400)]
Test cleanup: wrong indentation style in test_ust_data.c
Signed-off-by: Jérémie Galarneau <jeremie.galarneau@efficios.com>
Jérémie Galarneau [Fri, 31 Aug 2018 17:39:50 +0000 (13:39 -0400)]
Test fix: leak of exclusions on allocation error
Reported-by: Coverity Scan (1395328 Resource leak)
Signed-off-by: Jérémie Galarneau <jeremie.galarneau@efficios.com>
Jérémie Galarneau [Fri, 31 Aug 2018 17:33:51 +0000 (13:33 -0400)]
Fix: runas check fd value before calling close()
A bug could cause an 'open' command to return no FD in which
case the initial value of '-1' would be used in the call to
close().
Reported-by: Coverity Scan (1395329 Improper use of negative value)
Signed-off-by: Jérémie Galarneau <jeremie.galarneau@efficios.com>
Jérémie Galarneau [Thu, 30 Aug 2018 20:21:42 +0000 (16:21 -0400)]
Docs: multiple rotation schedules can be active
Signed-off-by: Jérémie Galarneau <jeremie.galarneau@efficios.com>
Jérémie Galarneau [Thu, 30 Aug 2018 20:19:53 +0000 (16:19 -0400)]
Docs: immadiate rotations can be performed with active schedules
Signed-off-by: Jérémie Galarneau <jeremie.galarneau@efficios.com>
Jérémie Galarneau [Thu, 30 Aug 2018 19:42:41 +0000 (15:42 -0400)]
Fix: ret variable is used instead of cmd_ret in disable-rotation
Signed-off-by: Jérémie Galarneau <jeremie.galarneau@efficios.com>
Jérémie Galarneau [Thu, 30 Aug 2018 19:39:30 +0000 (15:39 -0400)]
Cleanup: unused assignation on rotation error
Signed-off-by: Jérémie Galarneau <jeremie.galarneau@efficios.com>
Jérémie Galarneau [Thu, 30 Aug 2018 19:37:51 +0000 (15:37 -0400)]
Cleanup: unused assignation on rotation already pending
Signed-off-by: Jérémie Galarneau <jeremie.galarneau@efficios.com>
This page took 0.046272 seconds and 4 git commands to generate.