Decouple `struct lttng_session` from filter code

Use `struct lttng_ctx` directly so that the future event notifier
infrastructure, which will be outside any session, can use it.

Signed-off-by: Francis Deslauriers <francis.deslauriers@efficios.com>
Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>
Change-Id: I84109a0c01636bc5206f43e7abdceee27124c520

port: FreeBSD 12.2 added pthread_setname_np

With a minor difference that it doesn't return an error when the name is
too long, add this check in the wrapper to have the same behavior on
all platforms.

Signed-off-by: Michael Jeanson <mjeanson@efficios.com>
Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>
Change-Id: Id6fa0970b011867424a215e5d7ba69e9fe617389

port: tests: Add a simple unit test for shm operations

Signed-off-by: Michael Jeanson <mjeanson@efficios.com>
Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>
Change-Id: Idf347eb2985cac2216da0bb13aee5b09884c8b48

port: fsync(2) on a POSIX shm fd returns EINVAL on FreeBSD

The POSIX shared memory object implementation on FreeBSD is not file
based and doesn't support fsync(2).

Signed-off-by: Michael Jeanson <mjeanson@efficios.com>
Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>
Change-Id: Ie170254122aeab858a12c0522dc4e78075f243f9

port: set shm size with ftruncate before writing

According to POSIX, the size of shared memory object must be set with
ftruncate before writing to it. This happenen to work on Linux because
the implementation simply creates files on a tmpfs but on FreeBSD it is
implemented using actual shared memory objects and as such requires the
size to be set first.

Signed-off-by: Michael Jeanson <mjeanson@efficios.com>
Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>
Change-Id: I048f65104d7c34142025337ec1f6777650bfd62f

fix: pass the detected CFLAGS to the C++ compiler

Signed-off-by: Michael Jeanson <mjeanson@efficios.com>
Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>
Change-Id: I8f76c87c1acb772d93292ca36bd9b7d56a147b5e

Cleanup: silent rules are always available in automake >= 1.12

Signed-off-by: Michael Jeanson <mjeanson@efficios.com>
Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>
Change-Id: Ib0acb195594e1b788c8c6d1b2cd879c3c271aef3

Cleanup: Use pkg-config to detect liburcu

Signed-off-by: Michael Jeanson <mjeanson@efficios.com>
Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>
Change-Id: Iddfc721318e3ac06f2403876c1aa196da797d113

tests: Move tap-driver.sh out of the autotools aux directory

We have made local modifications to this script, move it to the test
suite directory so it doesn't get overwritten by an updated version from
autotools.

Signed-off-by: Michael Jeanson <mjeanson@efficios.com>
Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>
Change-Id: Ie46f0c39fe4d196b2bab3978ac02c2b1fed883c7

Enable autotools warnings as errors

Enable all autotools warnings and error out to help catch all the
wonderful, shall we say, pecularities of M4sh.

And fix some associated errors in configure.ac.

Signed-off-by: Michael Jeanson <mjeanson@efficios.com>
Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>
Change-Id: Icc693a1913a003f2c5b299e82084252c5735603f

Cleanup: separate tests between 'unit' and 'compile'

Change-Id: Ib4326540609dc464b9a89ba9b891cf9c410d902b
Signed-off-by: Michael Jeanson <mjeanson@efficios.com>
Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>

Cleanup: Bump autoconf required version to 2.69

We already depend on automake 1.12 released in 2012, set the minimum
autoconf version to 2.69 which was released on the same year.

Signed-off-by: Michael Jeanson <mjeanson@efficios.com>
Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>
Change-Id: Ib3c8bb5fccca3eacb80e78f361bc1b3fddbe00e1

Cleanup: Remove obsolete and unused tests

These tests have not been built in years and are not compatible with the
current API.

Signed-off-by: Michael Jeanson <mjeanson@efficios.com>
Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>
Change-Id: I6e18ba701d5deec72102e9b652ec9dfc9f08a9c2

fix: pthread_setname_np tests to match compat behavior

Adjust the test to expect an error when setting a thread name of more
than 16 bytes. Also don't override global AM_CPPFLAGS in
pthread_name/Makefile.am so that we get the proper include
configuration.

Signed-off-by: Michael Jeanson <mjeanson@efficios.com>
Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>
Change-Id: Iee41d8ea845836d273f01e890ea4196ae2ed43ef

fix: pthread_setname_np fails on longer tread names

When supplied with a thread name of more than 16 bytes including the
null terminating byte pthread_setname_np will fail with ERANGE,
replicate this behavior in the compat wrappers.

Change-Id: I91ac35a400a39c297e49fcab83b4f345b7ad92d0
Signed-off-by: Michael Jeanson <mjeanson@efficios.com>
Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>

Namespace lttng/align.h macros

These macros are part of the API and should have been namespaced
initialy. Namespace them and add backwards compatibility defines, make
sure to include the relevant system headers first for platforms that
have conflicting defines.

Change-Id: If2929ec2202dbde896614a25440ab200514ef4ad
Signed-off-by: Michael Jeanson <mjeanson@efficios.com>
Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>

Fix: remove redefinition of _GNU_SOURCE

Now that _GNU_SOURCE is defined by the compiler command line, no need to
redefine it.

Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>
Change-Id: I5a02dd945f0dfec778779a8c5806cc9f45960520

Add pthread_setname_np tests

Signed-off-by: Michael Jeanson <mjeanson@efficios.com>
Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>
Change-Id: I5999ecddf62be9ad25cd67a61ff7315b48f3306a

port: fix pthread_setname_np integration

The detection was broken and initialy went undetected because of the
failover code. Replace it with a hard fail and fix the detection for
each platform.

Also, replace the private LTTNG_UST_PROCNAME_LEN with
LTTNG_UST_ABI_PROCNAME_LEN from the public headers. The limit of 16
bytes does in fact include the terminating null byte contrary to what
was defined in LTTNG_UST_PROCNAME_LEN. This was not catched before since
the string is silently truncated if it exceeds 16 bytes.

Change-Id: I1ea95dfd882dfeb80fdc3d71ebec6618b1324a79
Signed-off-by: Michael Jeanson <mjeanson@efficios.com>
Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>

Cleanup: standardise include path

Use the same include setup as our other projects, set the default
includes globally in configure.ac by exporting AM_CPPFLAGS. Move the
local config.h to the include directory and include it by default on the
command line.

Signed-off-by: Michael Jeanson <mjeanson@efficios.com>
Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>
Change-Id: I28b48dc47cfc923d588a0033b97c2917a9c893a5

port: no libnuma on FreeBSD

Signed-off-by: Michael Jeanson <mjeanson@efficios.com>
Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>
Change-Id: I1c5b376de188a8f23ae90c2a6ab72fb1ecc42ace

port: use /bin/sh in test_ust_elf

The script doesn't use any bash extensions, use the default shell.

Signed-off-by: Michael Jeanson <mjeanson@efficios.com>
Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>
Change-Id: I7f7c843087400938f1e4d0de140d5deaff4efcd9

port: fix typo 'lttng_ust_safe_closefrom' -> 'lttng_ust_safe_closefrom_fd'

Signed-off-by: Michael Jeanson <mjeanson@efficios.com>
Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>
Change-Id: I6f9ddd4b46e83c379ed76dd435ce44c49393f7cf

port: include limits.h first for CHAR_BIT

On FreeBSD, CHAR_BIT is defined in limits.h, include it first so we
don't get redifinition errors.

Signed-off-by: Michael Jeanson <mjeanson@efficios.com>
Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>
Change-Id: I694cc906d801bcd9b1e82dcbcf1163cdb3c41100

port: fix endian and byteswap compat on FreeBSD

Signed-off-by: Michael Jeanson <mjeanson@efficios.com>
Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>
Change-Id: I8edd89df2c99406b10b846eb1a8bcf959e8bbde4

port: Silence macro redefinition warnings on FreeBSD

Signed-off-by: Michael Jeanson <mjeanson@efficios.com>
Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>
Change-Id: I32473a65116fd9e4120a852a8cbc5c145e9609f7

port: fix pthread autoconf detection to support FreeBSD

Use the same autoconf pthread code as lttng-tools to handle pthread
compat on FreeBSD.

Signed-off-by: Michael Jeanson <mjeanson@efficios.com>
Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>
Change-Id: I713446b1f5ec2eaab96cdfc7e9888ab543e8a67f

Set version to 2.13-pre

Signed-off-by: Michael Jeanson <mjeanson@efficios.com>
Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>
Change-Id: Id17c197aa437f0cf7a5d489f62f5f6b5cffb22e0

Fix: ustctl_release_object: eliminate double-close/free on error

When ustctl_release_object returns an error, it is unclear to the caller
what close/free side effects were effectively performed and which were
not. So the only courses of action are to either leak file descriptors
or memory, or call ustctl_release_object again which can trigger double
close or double free.

Fix this by setting the file descriptors to -1 after successful close,
and pointers to NULL after successful free.

Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>

tests: return the proper TAP exit code

The C TAP library provides the 'exit_status()' function that will return
the proper exit code according to the number of tests that succeeded or
failed.

Signed-off-by: Michael Jeanson <mjeanson@efficios.com>
Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>
Change-Id: I786527dfa9cfe2d1a7c8bc80086d54186f60b4d9

Add userspace time namespace context

This was introduced in kernel v5.6.0, on a system running an older kernel,
zero will be returned.

Signed-off-by: Michael Jeanson <mjeanson@efficios.com>
Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>
Change-Id: I1ea66a98c96a4de084c02521f6982e8c7d1b53bb

Use libtool syntax in LDADD

Use .la files to add link dependencies between libraries, linking
directly to the content of '.libs' is relying on an implementation
detail of libtool.

Signed-off-by: Michael Jeanson <mjeanson@efficios.com>
Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>
Change-Id: I6fe0157e23f13cd69c436111a2e743fa1c763d64

Add missing files to distribution

Signed-off-by: Michael Jeanson <mjeanson@efficios.com>
Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>
Change-Id: I5b0be9c7c87dc7fef1c6f46f15091d4e3c2a09a8

Fix: python agent: 'time' has no attribute 'clock'

The time.clock() function was removed in python 3.8 and is marked as
deprecated since python 3.3. See PEP 418 for more details [1].

Solution
=====

When the python version is greater than 3.2 use the
`time.perf_counter()` function [2]. Otherwise, fall back to
`time.clock()`.

Introduce a compat module to the lttngust agent package providing the
`_clock` function.

[1] https://www.python.org/dev/peps/pep-0418/
[2] https://docs.python.org/3/library/time.html#time.perf_counter

Signed-off-by: Jonathan Rajotte <jonathan.rajotte-julien@efficios.com>
Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>
Change-Id: I3d6d8b24309d45d43b634dc2a6b4d5dbc12da3aa

Fix: libc-wrapper: undef temporary token rather than value

The lttng-ust malloc wrapper defines pthread_mutex_lock/unlock
preprocessor tokens to ust_malloc_spin_lock/unlock around the definition
of a TLS variable, which uses pthread mutexes when relying on the
pthread key fallback. Undefining those tokens should be done on the
preprocessor token, rather than its value.

Signed-off-by: Christophe Bedard <bedard.christophe@gmail.com>
Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>
Change-Id: I2053b79c88000e272c29b25ca105b1352ecfabd7

Fix: support compile units including 'sys/sdt.h' without defining SDT_USE_VARIADIC

Instead of using SDT_USE_VARIADIC from 'sys/sdt.h', use our own namespaced
macros since the instrumented application might already have included
'sys/sdt.h' without variadic support.

Signed-off-by: Michael Jeanson <mjeanson@efficios.com>
Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>
Change-Id: Idfece1a65a5296a90d33370bc8d73ea554c14b0f

Cleanup: have interpreter functions return _DISCARD instead of 0

It's easier to understand the meaning of the zero return value of these
function using the enum. It makes it obvious.

Signed-off-by: Francis Deslauriers <francis.deslauriers@efficios.com>
Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>
Change-Id: Id8f199d82b17228fdc882e3ccbab8b3535840984

Cleanup: bytecode: typo: "s16" -> "u16"

Signed-off-by: Francis Deslauriers <fdeslaur@gmail.com>
Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>
Change-Id: If3a0e0422ce91d26c33b2e040a1463879e8d8505

doc: Explain `struct lttng_bytecode_runtime` layout restrictions

Signed-off-by: Francis Deslauriers <francis.deslauriers@efficios.com>
Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>
Change-Id: I875b11c644917a3b9f15bf88ab44098813e59db0

Fix: event probes attached before event enabled

Background
==========
When userspace events with exclusions are enabled by the CLI user, the
session daemon enables the events in a 3-steps process using the
following ustctl commands:
1. `LTTNG_UST_EVENT` to create an event and get an handle on it,
2. `LTTNG_UST_EXCLUSION` to attach exclusions, and
3. `LTTNG_UST_ENABLE` to activate the tracing of this event.

Also, the session daemon uses the `LTTNG_UST_SESSION_START` to start the
tracing of events. In various use cases, this can happen before OR after
the 3-steps process above.

Scenario
========
If the`LTTNG_UST_SESSION_START` is done before the 3-steps process the
tracer will end up not considering the exclusions and trace all events
matching the event name glob pattern provided at step #1.

This is due to the fact that when we do step #1, we end up calling
`lttng_session_lazy_sync_enablers()`. This function will sync the event
enablers if the session is active (which it is in our scenario because
of the _SESSION_START).
Syncing the enablers will then match event name glob patterns provided
by the user to the event descriptors of the tracepoints and attach
probes in their corresponding callsite on matches.

All of this is done before we even received the exclusions in step #2.

Problem
=======
The problem is that we attach probes to tracepoints before being sure we
received the entire description of the events.
Step #2 may reduced the set of tracepoints to trace so we must wait
until exclusions are all received to attached probes to tracepoints.

Solution
========
Only match event names and exclusions (and ultimately, register probes)
on enabled enablers to ensure that no other modifications will be
applied to the event.

Event enablers are enabled when at step #3.

Note
====
Filters are not causing problems here because adding a filter won't
change what tracepoints are to be traced.

Signed-off-by: Francis Deslauriers <francis.deslauriers@efficios.com>
Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>
Change-Id: Id984f266d976f346b001db81cd8a2b74965b5ef2

Fix: use underlying types for array and sequence assertion

Based on https://gcc.gnu.org/onlinedocs/gcc/Other-Builtins.html:

"Built-in Function: int __builtin_types_compatible_p (type1, type2)

[...] two types that are typedefed are considered compatible if
their underlying types are compatible."

This definition explains why listing all the {u,}intX_t types misses
the underlying type "long" on 32-bit architectures which rely on
"long long" for 64-bit integers. Therefore, list all the underlying
integer types.

Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>

Add compile time assertion that array and sequence have integer elements

Currently, LTTng-UST only supports ctf_arrays(and ctf_sequences) of
integers and chars but there is nothing preventing the user of compiling
an array of double. This is problematic because the resulting metadata
for that array will end up being erroneous.

This commit adds a new stage in the probe building process that asserts
that the element type of arrays or sequences are supported. This stage
adds a compile time static assertion, so the probes won't build if the
type is not supported.

Here is the error message the user will get when compiling with a C99
compiler:
  gcc -c -sdt=c99 -I. tp.c
    In file included from /usr/local/include/lttng/tracepoint-event.h:58:0,
       from tp.h:77,
       from tp.c:4:
    ././tp.h:71:9: error: size of array ‘lttng_static_assert_Non_integer_type__my_array__not_supported_as_element_of_CTF_ARRAY_or_CTF_SEQUENCE’ is negative
    ctf_array(double, my_array_for_double, my_array_for_double, 3)

Here is the error message the user will get when compiling with a C++ or
C11 compiler or above:
  g++ -c -std=c11 -I. tp.c
    In file included from /usr/local/include/lttng/tracepoint-event.h:58:0,
         from tp.h:77,
         from tp.c:4:
    ././tp.h:71:9: error: static assertion failed: Non-integer type `my_array` not supported as element of CTF_ARRAY or CTF_SEQUENCE
    ctf_array(double, my_array, my_array_for_double, 3)

Signed-off-by: Francis Deslauriers <francis.deslauriers@efficios.com>
Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>
Change-Id: Ie958635a8dc6262793d45c5f7458d50a0fecf7ab

Implement compile time assertion macro wrapper

Starting from C++11[1] and C11[2] both C++ and C standards implement the
`static_assert()` macro that allows the compile time evaluation of an
expression and emitting of a compiler error when the expression
evaluates to false.

This commit implements a wrapper for compile time assertions on C
compilers implementing C standards prior C11.

On such compilers, we emulate a static assert by typedef'ing an array of
negative size in case of predicate failure.

The downside of this method is that error messages are a bit cryptic as
it mentions the negative-sized array. We overcome this issue by using a
user-provided message as part of variable name that gets printed on
error. For this reason, we decide to require 2 different messages in
addition of the predicate.

Here is the signature of the macro:
  #define lttng_static_assert(predicate, msg, c_identifier_msg)

The first message, `msg`, is the used with C++/C11 compilers can be any
string. The second message, `c_identifier_msg`, is used with older
standards and MUST be a valid C identifier as it's will be concatenated
to a typedef name.

For example:
If the user uses the macro such as:
  lttng_static_assert(false, "My assert message", MY_ASSERT_MSG);

The C99 user will get an error looking like this:
 error: size of array ‘lttng_static_assert_MY_ASSERT_MSG’ is negative

The C11 or C++ user will get an error looking like this:
  error: static assertion failed: "My assert message"

[1] https://en.cppreference.com/w/cpp/language/static_assert
[2] https://en.cppreference.com/w/c/language/_Static_assert

Signed-off-by: Francis Deslauriers <francis.deslauriers@efficios.com>
Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>
Change-Id: I725f6e77f1858b8d88ffae781b648ac5b5c64b28

Fix: agent: read: end of loop condition should exclude 0

Causes the agent to hang endlessly waiting for completion of the
loop.

Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>

fix: Java examples CLASSPATH override

Variables provided to make as arguments, called 'command variables' can't
be reassigned a new value in the Makefile. Since the CLASSPATH is now
passed this way in the glue between automake and the example makefiles,
use different names for the internal variables.

Change-Id: Id6289273f211f544a66d933a96f06df75243415f
Signed-off-by: Michael Jeanson <mjeanson@efficios.com>
Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>

Add missing files to .gitignore

Signed-off-by: Michael Jeanson <mjeanson@efficios.com>
Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>
Change-Id: I5050772c58342d3eb956e0cbd56983fbb1ed3b70

fix: Add CLASSPATH to autoconf precious variables

This will record the value of CLASSPATH in the autoconf cache and ensure
it's used when calling make even if it's not exported to the
environment.

Signed-off-by: Michael Jeanson <mjeanson@efficios.com>
Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>
Change-Id: Ia06070dd352f9d3b6956a1ec0f4eb1d024f73179

Fix: Java agent: close session daemon socket on error

When catching an error, close the socket used to communicate with the
session daemon rather than leaking it, before retrying to connect.

Also, when sleep is interrupted, there is no point in printing
out the stack trace. Just retry connection immediately.

Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>

Java agent: Use inequality for read bound check

Don't continue looping if bytesLeft becomes negative for unforeseen
reasons.

Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>

Fix: Java agent: handle partial payload read

When reading from a TCP socket, there is no guarantee that the
read will return all the requested data. We need to loop and continue
reading until we gather all the expected data.

This fixes flakiness of the lttng-ust-java-tests.

Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>

Fix: incorrect field accounting for dynamic type

Two field accounting issues were introduced by
commit 218deb69baab "tracepoint: Refactor representation of nested
types" which affect the dynamic types used by Java application contexts.

The structure types should account 1 for the structure per se, in
addition to the nested fields.

The atype_dynamic includes an implicit enumeration, which now has two
fields instead of 1.

This was causing memory corruption issues in the lttng-ust-java-tests.

Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>

ust-compiler.h: Allocate compound literals on heap with gcc <= 4.8

g++ 4.8 and prior do not support C99 compound literals. Therefore,
force allocating those on the heap with these C++ compilers.

Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>

Require C++11 for building probe providers with C++ compiler

The __LTTNG_COMPOUND_LITERAL(), in C++, requires either that the
compiler supports C99 compound literals (which is not the case for
g++ 4.8), or that LTTNG_ALLOCATE_COMPOUND_LITERAL_ON_HEAP is
defined so the compiler allocates the compound literal on the
heap with "new", which is available since c++11.

Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>

filter: bytecode already in the list should go before

Background
==========
This `seqnum` (sequence number) feature is currently unused. It was
designed so that the session daemon could tell the tracer the order in
which the bytecode should be run.

Issue
=====
The current implementation of the session daemon doesn't use this
feature so there is only ever a single bytecode to execute per callsite.

During work on an upcoming feature uses this `seqnum` became useful and
it was realized that the current bytecode linking code would reverse the
order in which the bytecode were executed when all bytecodes have the
same `seqnum` value.

This is due to the fact that the `cds_list_for_each_entry_reverse` loops
until it finds a `seqnum` smaller than the new one.

So if all bytecodes have the same `seqnum`, the new bytecode will be
added at the beginning of the list.

This is not technically a problem since it's the session daemon's job to
set the sequence number if it wants a particular ordering. Even
considering that, we found it counterintuitive that new bytecodes are
added at the beginning of the list in those cases.

Solution
========
This commit makes it so that on equality, the insertion is done after
the existing bytecodes.

Signed-off-by: Francis Deslauriers <francis.deslauriers@efficios.com>
Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>
Change-Id: Ie323b3ff5df346f222e22abfb9bbad6c5b7e4a27

Bump LTTNG_UST_ABI version from 8.1 to 9.0

Ensure an older liblttng-ust communicating with a newer liblttng-ust
will refuse the handshake, so older session daemon do not get to
interact with more recent lttng-ust which implements nested arrays and
sequences.

Introduce LTTNG_UST_ABI_MAJOR_VERSION_OLDEST_COMPATIBLE to allow
liblttng-ust to be backward compatible with a range of major versions.
In this case, liblttng-ust ABI 9.0 is fully backward compatible with
8.x, given that none of the existing ABI in 8.x has changed in 9.0.

Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>

tracepoint: Refactor representation of nested types

Refactor enumeration, sequence, array, structure, and variant types.

Implement internal data structures to support nested types.

All probe providers using ctf_enum(), ctf_array*() and ctf_sequence*()
are switched to this new binary layout, and the prior enum, array and
sequence abstract types are kept only for backward compatibility with
probe providers generated by older lttng-ust headers.

Each of sequence, array, struct and variant gain a "alignment" property,
which is a feature which was needed in lttng-modules to express
alignment for an array or sequence of bits.

The ust-ctl protocol with session daemon is extended to support those
new types.

Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>

Remove has_strcpy check following probe provider version bump

Bumping the probe provider major version from 1 to 2 allows us to
remove a dynamic check within ctf_string() in the probe providers.

Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>

Bump LTTNG_UST_PROVIDER_MAJOR from 1 to 2

The introduction of nested arrays and sequences requires to bump the
provider major version from 1 to 2. Newer lttng-ust are backward
compatible with older probes, but newer probes cannot be used by
older lttng-ust versions.

Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>

ust-compiler.h: Implement __LTTNG_COMPOUND_LITERAL

Implement __LTTNG_COMPOUND_LITERAL to allow defining a compound literal
with static storage.

Compound literals with static storage are needed by LTTng to implement
nested arrays and sequences in TRACEPOINT_EVENT.

Compound literals are part of the C99 and C11 standards, but not
part of the C++ standards. However, those are supported by both g++ and
clang as C99 extensions when compiling C++ code.

Strict C++11 compliance can be obtained by defining
LTTNG_ALLOCATE_COMPOUND_LITERAL_ON_HEAP before including this header
to allocate those on the heap in C++.

Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>

Document dependency on liburcu >= 0.12

Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>

configure: add check for DEFINE_URCU_TLS_INIT

Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>

Cleanup: use DEFINE_URCU_TLS_INIT for all CONFIG_RCU_TLS configurations

Introduce a dependency on DEFINE_URCU_TLS_INIT for all CONFIG_RCU_TLS
configurations.

Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>

Fix: namespace contexts !CONFIG_RCU_TLS variable initialization

The namespace contexts introduced in lttng-ust 2.12 require to
initialize TLS variable to nonzero values. However, in !CONFIG_RCU_TLS
(compatibility mode using pthread setspecific), this initialization
does not build.

Use the new DEFINE_URCU_TLS_INIT from liburcu when building
!CONFIG_RCU_TLS to fix this issue. Since this requires a dependency on
a new liburcu version, only !CONFIG_RCU_TLS adds this dependency in the
fix. A followup cleanup patch will use DEFINE_URCU_TLS_INIT as we add
a strict version dependency.

Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>

Record event as soon as one filters evaluates to TRUE

This is a tiny optimization but it can reduce tracing cost in cases
where multiple filters are attached to a probe.

Since we want to trace the event if any of its filter expressions
evaluates to TRUE, there is no need to keep on evaluating expression
once one is found to be TRUE.

Signed-off-by: Francis Deslauriers <francis.deslauriers@efficios.com>
Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>
Change-Id: I2bb8fafb3052fce02b10f541e025a9a74ffa1895

Cleanup: ust-tracepoint-event.h: extract `__num_fields` in a variable

This makes it easier to understand what the `_TP_ARRAY_SIZE()` macro
expands to in this context.

Signed-off-by: Francis Deslauriers <francis.deslauriers@efficios.com>
Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>
Change-Id: I6b44e1e444df1dab0496282af4cea260ae895ca0

Fix: v.u.d might be uninitialized

lttng-filter-interpreter.c:301:17: warning: ‘v.u.d’ may be used uninitialized in this function [-Wmaybe-uninitialized]
   ptr->u.d = v.u.d;

Indeed it seems that the value is never fetched if we compare to other
call sites and the dynamic call site further below.

Signed-off-by: Jonathan Rajotte <jonathan.rajotte-julien@efficios.com>
Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>
Change-Id: I542d6db7514a2c39afe3613228577f34f194d672

Fix: Add missing vtracelog demo to make dist

Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>

Fix: Add missing vtracef demo to make dist

Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>

Introduce vtracelog

vtracelog works the same as vtracef, but takes a log level
as a parameter and has the same limitations.

Signed-off-by: Maxime Roussin-Belanger <maxime.roussinbelanger@gmail.com>
Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>

Introduce vtracef

vtracef accepts a va_list argument to simplify
tracing functions which use a va_list

Here's an example from wpa_supplicant that I wanted to
trace:

void wpa_debug(int level, const char* fmt, ...)
{
va_list ap;
va_start(ap, fmt);

...
// The call I want to easily trace with vtracef
vprintf(fmt, ap);

...
va_end(ap);
}

wpa_debug is used a fair amount and it would be annoying to
replace all the wpa_debug calls with tracef.

With vtracef, it simplifies the find and replace effort by
only changing it at one place.

Signed-off-by: Maxime Roussin-Belanger <maxime.roussinbelanger@gmail.com>
Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>

Fix: set FD_CLOEXEC on incoming FDs.

The stream shm FDs are allocated by the consumer process, and then
passed to the applications over unix sockets. When opening those
file descriptors on reception, the FD_CLOEXEC flag is not set.

In a fork + exec scenario, parent process streams shm FDs and channel
wake FDs are present in the resulting child process.

Set FD_CLOEXEC on reception (ustcomm_recv_fds_unix_sock) to
prevent such scenario.

Change-Id: Id58077b272be9c1ab239846639ffd8103b3d50f1
Signed-off-by: Jonathan Rajotte <jonathan.rajotte-julien@efficios.com>
Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>

Fix: tracepoint.h: Disable address sanitizer on pointer array section variables

The tracepoint header declares pointer global variables meant to be
placed contiguously within the __tracepoints_ptrs section, and then used
as an array of pointers when loading an executable or shared object.

Clang Address Sanitizer adds redzones around each variable, thus leading to
detection of a global buffer overflow.

Those redzones should not be placed within this section, because it
defeats its purpose. Therefore, teach asan not to add redzones
around those variables with an attribute.

Note that there does not appear to be any issue with gcc (tested with
gcc-8 with address sanitization enabled), and gcc ignores the
no_sanitize_address attribute when applied to a global variable.

Fixes: #1238
Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>

Fix: jhash.h: remove out-of-bound reads

jhash.h implements "special" code for valgrind because it reads memory
out-of-bound (and then applies a mask) when reading strings.

Considering that lttng-ust does not use jhash.h in a fast-path, remove
this "optimization" and use the verifiable VALGRIND code instead. This
fixes an ASan splat.

Fixes: #1238
Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>

Fix: generation of man pages with multiple refnames

In this proposed patch [1], we're trying to make a man page with two
command names in the NAME section, like this:

  tracef, vtracef - LTTng-UST printf(3)-like interface

Doing so causes the intermediary XML (what asciidoc outputs) to contain
this:

    <refnamediv>
      <refname>tracef, vtracef</refname>
      <refpurpose>LTTng-UST printf(3)-like interface</refpurpose>
    </refnamediv>

The refname is used by the docbook XSL to determine the output
filename.  This therefore results in an output file named
`tracef,_vtracef.3`, which is not desirable.

The problem is in the asciidoc pass, more specifically the part of
asciidoc.conf removed by this patch.  Is is there to override the
`source`, `version` and `manual` fields, but it also inadvertently
affects the refname generation.

Instead of playing with the asciidoc output template, this patch sets
the same `source`, `version` and `manual` attributes on the asciidoc
command line.  With the present patch applied, the XML output contains:

    <refnamediv>
        <refname>tracef</refname>
        <refname>vtracef</refname>
        <refpurpose>LTTng-UST printf(3)-like interface</refpurpose>
    </refnamediv>

The xmlto pass (docbook XML -> man page) then generates two files,
`tracef.3` and `vtracef.3`, where `vtracef.3` simply sources `tracef.3`
(which is what we want).

[1] https://lists.lttng.org/pipermail/lttng-dev/2020-February/029524.html

Signed-off-by: Simon Marchi <simon.marchi@efficios.com>
Signed-off-by: Philippe Proulx <eeppeliteloop@gmail.com>
Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>

Cleanup: remove trailing white spaces across project

Signed-off-by: Francis Deslauriers <francis.deslauriers@efficios.com>
Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>
Change-Id: I15202338465ee56d33316cbc632d9e3bf44ee31e

Fix: lttng-ust-comm.c: return number of fd rather size of array

There are two conflicting comments for this function. One says it
returns the size of the received data and the other says it returns the
number of fd received.

It's more useful to receive the number of fd.

Signed-off-by: Francis Deslauriers <francis.deslauriers@efficios.com>
Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>
Change-Id: I74084b461d396c3e623fa55100e6dd7e59dbea83

liblttng-ust-comm: move `_unlock_fd_tracker()` after `close()` on error paths

Right now, receiving an error from `lttng_ust_add_fd_to_tracker()` means that
the fd was _not_ added to the fd set. So the `lttng_ust_safe_close_fd()`
(overriding `close()`) call will indeed close the fd as expected. So, it
doesn't matter if the `close()` is before or after the `_unlock_`.

Even considering that, I believe that it's clearer and more common to
have all related operations within the `_lock_` and  `_unlock_`
functions. Also, `lttng_ust_add_fd_to_tracker()` might be modified in
the future and fail for some other reason.

Signed-off-by: Francis Deslauriers <francis.deslauriers@efficios.com>
Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>
Change-Id: Id29a6ab004cfd5ca601615e1a70c74cf754b12e2

Cleanup: liblttng-ust: change `int` flag to `bool`

Signed-off-by: Francis Deslauriers <francis.deslauriers@efficios.com>
Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>
Change-Id: I78b35b6b99afe8a84aba6c836df3bce8d2532760

liblttng-ust: exit loop early on event enabler match

Signed-off-by: Francis Deslauriers <francis.deslauriers@efficios.com>
Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>
Change-Id: I0fa3215f7cd6a2d32ac00d66cf5fc184abd14612

Cleanup: remove redundant memory barrier

This memory barrier is redundant with the one already issued in
`rcu_assign_pointer()` function during the `set_tracepoint()` call.

Signed-off-by: Francis Deslauriers <francis.deslauriers@efficios.com>
Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>
Change-Id: I78decd9ae87a63d7663928ad99e03831155548f7

Cleanup: remove unused `lttng_bytecode_runtime::event` field

Signed-off-by: Francis Deslauriers <francis.deslauriers@efficios.com>
Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>
Change-Id: Id946b69e8cb4e79415d221f0be0889cfe707b439

Docs: explain why unused `lttng_enabler::ctx` is kept around

Signed-off-by: Francis Deslauriers <francis.deslauriers@efficios.com>
Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>
Change-Id: I93cc035ced57e29f3d675c8c611d5319194e6cfb

Cleanup: remove unused `lttng_free_enabler_filter_bytecode()` func

Signed-off-by: Francis Deslauriers <francis.deslauriers@efficios.com>
Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>
Change-Id: Id9326aec2e6758ffffeb3eb9805bb0fa85d3076c

Cleanup: move unused function to deprecated symbol list

Signed-off-by: Francis Deslauriers <francis.deslauriers@efficios.com>
Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>
Change-Id: I9c75feee029364ab17ef3783a6c8f0d45ff2948b

Cleanup: remove unused `__check_ust_safe_fmt()` function

Usages of this function were removed by this commit:
  commit 5e96a46756a5dcd6e348afd84078b9e26438245d
  Author: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>
  Date:   Tue Aug 9 09:46:51 2011 -0400

      Add UST_DEBUG env. var. support

Signed-off-by: Francis Deslauriers <francis.deslauriers@efficios.com>
Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>
Change-Id: I535028fc9338c0c2cef51a993bcab8f0e6c062c6

Cleanup: silence unused parameter `ps` warning

This function was copied here from one of the BSDs by this commit:
e5bc3b0f4d6c0407492ebdea863483925393e1bc with the /*ARGSUSED*/
lint-style comment to prevent any `unused` warning.

It seems clang doesn't support those lint-style comments, so it prints
this warning:
  CC       libustsnprintf_la-core.lo
  mbrtowc_sb.c:39:68: error: unused parameter 'ps' [-Werror,-Wunused-parameter]
  ust_safe_mbrtowc(wchar_t *pwc, const char *s, size_t n, mbstate_t *ps)

So mark that parameter as unused to prevent any warning.

We could also simply remove the unused parameter and change the
signature of the function.

Signed-off-by: Francis Deslauriers <francis.deslauriers@efficios.com>
Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>
Change-Id: I09f4fd3938058ed985c1525d8b9b06a27529691e

Cleanup: apply `include-what-you-use` guideline for `close()`

I saw that some files use the `close()` function but there are not
including a header for it. The `close()` function is defined by the
following header: <unistd.h>

So, to follow the best practice of including what is used in a file, I
added `#include <unistd.h>` in files using `close()`.

Signed-off-by: Francis Deslauriers <francis.deslauriers@efficios.com>
Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>
Change-Id: I07e104e957857c869576af7c2704e98584ecc763

Cleanup: apply `include-what-you-use` guideline for `uint*_t`

I saw that some files use `uint*_t` types but there are not including a
header for it. The `uint*_t` types is defined by the following header:
  <stdint.h>

So, to follow the best practice of including what is used in a file, I
added `#include <stdint.h>` in files using `uint*_t`.

Signed-off-by: Francis Deslauriers <francis.deslauriers@efficios.com>
Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>
Change-Id: I5da13744858a57fc8c9bf6a6cf1b29299c2211cc

Cleanup: apply `include-what-you-use` guideline for `mbstate_t`

I saw that some files use the `mbstate_t` type but there are not including
a header for it. The `mbstate_t` type is defined by the following header:
  <wchar.h>

So, to follow the best practice of including what is used in a file, I
added `#include <wchar.h>` in files using `mbstate_t`.

Signed-off-by: Francis Deslauriers <francis.deslauriers@efficios.com>
Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>
Change-Id: I103d144ede1a8488d2662d3b8ba5337accafda99

Cleanup: apply `include-what-you-use` guideline for `fpos_t`

I saw that some files use the `fpos_t` type but there are not including
a header for it. The `fpos_t` type is defined by the following header:
  <stdio.h>

So, to follow the best practice of including what is used in a file, I
added `#include <stdio.h>` in files using `fpos_t`.

Signed-off-by: Francis Deslauriers <francis.deslauriers@efficios.com>
Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>
Change-Id: I808ad0b23f04389fd56f89aa001095b771a327d6

Cleanup: apply `include-what-you-use` guideline for `size_t`

I saw that some files use the `size_t` type but there are not including
a header for it. The `size_t` type is defined by the following headers:
  <stddef.h>
  <stdio.h>
  <stdlib.h>
  <string.h>
  <time.h>
  <wchar.h>

So, to follow the best practice of including what is used in a file, I
added `#include <stddef.h>` in files using `size_t` but did not include
any of the previously listed header.

Signed-off-by: Francis Deslauriers <francis.deslauriers@efficios.com>
Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>
Change-Id: Ica1d82691335294decf13ffcdf4257e35d6a44c2

Cleanup: typo: column -> colon

Signed-off-by: Francis Deslauriers <francis.deslauriers@efficios.com>
Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>
Change-Id: I6873c94805ad02e4bba1431f7202ae0185d9698f

Add git-review config

Add .gitreview for contributors wishing to use gerrit for patch
reviews.

Signed-off-by: Francis Deslauriers <francis.deslauriers@efficios.com>
Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>
Change-Id: I3ba853d0d68a660ac5d2d24c757e0f28b634977c

Version 2.12.0-rc1

Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>

Fix: build with -fno-common

GCC 10 will default to building with -fno-common, this inhibits the
linker from merging multiple tentative definitions of a symbol in an
archive. Keep only the declaration in the libustsnprintf.la convenience
library.

Signed-off-by: Michael Jeanson <mjeanson@efficios.com>
Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>
Change-Id: I8fb7c72811ce7e62f10342f55fcabeeabfdd4c67

Bump LTTNG_UST_ABI_MINOR_VERSION to 1

Increment the minor version of lttng-ust ABI to 1, to take into
account that the "clear" command was added in this release cycle.

This will allow future LTTng-tools versions to check for this
capability.

Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>

lttng-clear: stop tracing required

Require that tracing is stopped when buffers are cleared. Update
comments and warning checks to that effect.

Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>

doc: fix build failure due to wrong whitespace character

The previous, commit:

    54435b75df4c ("doc: reformat long lines in doc/examples/Makefile.am")

introduced the following build failure, when the support for JUL is
enabled:

    make[1]: Entering directory '/home/smarchi/build/lttng-ust/doc/examples/java-jul'
    javac -classpath "../../../liblttng-ust-java-agent/java/lttng-ust-agent-jul/lttng-ust-agent-jul.jar:../../../liblttng-ust-java-agent/java/lttng-ust-agent-common/lttng-ust-agent-common.jar:." -g Hello.java
    javac -classpath "../../../liblttng-ust-java-agent/java/lttng-ust-agent-jul/lttng-ust-agent-jul.jar:../../../liblttng-ust-java-agent/java/lttng-ust-agent-common/lttng-ust-agent-common.jar:." -g FilterChangeListenerExample.java
    javac -classpath "../../../liblttng-ust-java-agent/java/lttng-ust-agent-jul/lttng-ust-agent-jul.jar:../../../liblttng-ust-java-agent/java/lttng-ust-agent-common/lttng-ust-agent-common.jar:." -g ApplicationContextExample.java
    make[1]: *** No rule to make target ' '.  Stop.
    make[1]: Leaving directory '/home/smarchi/build/lttng-ust/doc/examples/java-jul'
    Makefile:979: recipe for target 'all-local' failed
    make: *** [all-local] Error 1

I inadvertently inserted a character that looks like a space, but that
is not a space.  make tries to interpret it as a target name, which
obviously fails.

Replace it with a proper space.

Signed-off-by: Simon Marchi <simon.marchi@efficios.com>
Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>

doc: reformat long lines in doc/examples/Makefile.am

Format the long lines in the all-local target a bit like the "cmake"
target is formatted already.  I think it helps readability to have one
argument per line instead of very long lines.

At the same time, I removed the "cd .." at the end of parentheses.  The
parentheses start a new subshell, so it's unnecessary to do "cd .."
before the subshell exits.

Signed-off-by: Simon Marchi <simon.marchi@efficios.com>
Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>

doc: pass AR when building examples

As reported here [1], when cross-compiling lttng-ust, the
"hello-static-lib" example uses the ar tool made for the --build machine
instead of the prefixed one, for the --host machine.

The Makefiles in the subdirectories of doc/examples are written by hand,
so that they can be easily copied and modified by users.  They are
therefore not integrated in the automake build system, and any value
detected by configure must be passed explicitly when invoking it.
For example, the CC value is already explicitly passed, so that the
compiler value found by configure is passed down.  We just need to do
the same for AR.

This patch adds AM_PROG_AR in configure.ac, so that configure finds the
prefixed version of ar, if cross-compiling.

It then sets the AR variable in doc/examples/Makefile.am, when invoking
sub-Makefiles.  I don't think we really need it in the cmake case, but
it doesn't hurt to have it there.

[1] https://lists.lttng.org/pipermail/lttng-dev/2019-November/029388.html

Reported-by: Rolf Eike Beer <eb@emlix.com>
Signed-off-by: Simon Marchi <simon.marchi@efficios.com>
Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>