document that tracepoint names should ideally not be re-used

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

Remove LIBFORMAT config declaration, unused

Remove leftover from UST 0.x. Not needed anymore.

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

Add libc errno translation layer to UST error code

Also add four new possible code being EEXIST, EINVAL, ENOSYS and EPERM.

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

Revert "Fix: configure.ac should use "--target", not "--host""

This reverts commit 98c8a1d90d1052980365353eef201124863128b0.

No. --target is for building cross-compilers. --host was  appropriate.

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

Fix: configure.ac should use "--target", not "--host"

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

Fix: add const qualifier for filter local void *

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

Fix: re-allow non-lvalue string, sequence, array parameters

Issue introduced by upstream commit:

  commit 27f4b6094f399f2fe231b58801dce98cbd21baa9
  Author: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>
  Date: Tue Sep 4 12:17:07 2012 -0400

  Fix filter: pointer to string, not string, should be on stack

  Fixes #329

Reported-by: David Bryant <david.bryant@quantum.com>
Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>

Tear down handles associated with a closed sessiond socket

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

Distinguish UST return codes from transport return codes

One API change visible from lttng-tools: -ENOENT return values for
tracepoint list and tracepoint field list are now -LTTNG_UST_ERR_NOENT.

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

Fix: Conditionally disable tests requiring shared libs support

When building lttng-ust with shared library support explicitly
disabled (e.g.: ./configure --disable-shared), libtool fail with
a fatal error:

  CC       tp.lo
  CC       tp2.lo
  CCLD     liblttng-ust-provider-ust-tests-demo.la
libtool: link: can not build a shared library
libtool: link: See the libtool documentation for more information.
libtool: link: Fatal configuration error.

The build should not fail because some tests require explicit shared
library support, instead they should be skipped.

This patch detect that the --disable-shared flag was passed to the
configure script and toggle the "NO_SHARED" Automake variable.
Thus, the tests that require explicit shared library support can
be skipped when the NO_SHARED variable is true.

[ Edit by Mathieu Desnoyers: add "" in configure.ac to follow the local
  coding style. ]

Signed-off-by: Christian Babeux <christian.babeux@efficios.com>
Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>

Cleanup: don't spawn per-user thread if HOME is not set

Reported-by: David OShea <David.OShea@quantum.com>
Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>

Manpage: document supported UST contexts

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

Fix: procname context semantic

Cache the procname per-thread rather than per-process to take into
account that prctl() can be used to set thread names.

prctl() should be issued before tracing each thread's first event if we
care about the thread name.

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

Fix: Fix self-assign warning on struct ustfork_clone_info init

Use the proper field designator syntax (C99) to initialize the
ustfork_clone_info struct.

Signed-off-by: Christian Babeux <christian.babeux@efficios.com>
Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>

Fix: memcpy of string is larger than source

Hollis Blanchard <hollis_blanchard@mentor.com> wrote:
> I seem to have hit a little problem with a "hello world" test app and
> lttng-ust 2.0.3. lttng-ust.git seems to be affected as well. Basically,
> I created a single UST tracepoint, but as soon as I run "lttng
> enable-event -u -a", my app segfaults. The problem seems to be that when
> creating the event to pass to ltt_event_create(), we try to memcpy the
> full 256 bytes of name. However, the name might be shorter, and if we
> get unlucky it falls within 256 bytes of the segment boundary...

Fixing the 3 sites where this issue arise. Manually inspecting all
memcpy in the UST code returned by grep did the job.

Reported-by: Hollis Blanchard <hollis_blanchard@mentor.com>
Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>

Implement liblttng-ust-fork daemon() override test

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

liblttng-ust-fork: override daemon() call

Fixes #252

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

ustfork: set errno to ENOSYS if symbol lookup fails

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

Fix: be quiet on filter linker error

Could be reproduced by starting the application before enabling the
event with filter.

Fixes #362

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

Build out of src tree

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

Fix: filter bytecode specializer stack leak

Fixes #356

Signed-off-by: Christian Babeux <christian.babeux@efficios.com>
Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>

Fix: reloc offset validation error out on filters with no reloc table

The reloc table is currently appended at the end of the bytecode data.
With this scheme, the reloc table offset will be equal to the length
of the bytecode data.

<- length ->
+----------+-------------+
| BYTECODE | RELOC TABLE |
+----------+-------------+
           |
           +--> Reloc table offset

A special case arise with filters with no reloc table.

Example:

Filter: "myString" == "yourString"
./filter-grammar-test -p -B -i -b < bogus
<root>
        <op type="==">
                <expression>
                        <string value="myString"/>
                </expression>
                <expression>
                        <string value="yourString"/>
                </expression>
        </op>
</root>
Generating IR... done
Validating IR... done
Generating bytecode... done
Size of bytecode generated: 24 bytes.
Bytecode:

Val. Operator
---- --------
0x40 (FILTER_OP_LOAD_STRING)
0x6D m
0x79 y
0x53 S
0x74 t
0x72 r
0x69 i
0x6E n
0x67 g
0x00 \0
0x40 (FILTER_OP_LOAD_STRING)
0x79 y
0x6F o
0x75 u
0x72 r
0x53 S
0x74 t
0x72 r
0x69 i
0x6E n
0x67 g
0x00 \0
0x0C (FILTER_OP_EQ)
0x01 (FILTER_OP_RETURN)

Reloc table (offset: 24):
Empty

<-   24   ->
+----------+
| BYTECODE | <- No reloc table
+----------+
           |
           +--> Reloc table offset

In this case, we see that the reloc table offset (24) is indeed equal to
the length of the bytecode (24), but the reloc table is _empty_. Thus,
the reloc_offset received in handle_message() will be equal to the
data_size and will be wrongly flagged as not within the data even thought
the filter is entirely valid.

The fix is to simply allow a reloc_offset to be equal to the data_size.

Fixes #342

Signed-off-by: Christian Babeux <christian.babeux@efficios.com>
Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>

Perform calculation on bit size in 64-bit

allow 32-bit architectures to have sub-buffers larger than 256MB.

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

Use uint64_t for packet header content size and packet size

Allow packets larger than 256MB (4GB / 8 bits per byte).

Fixes #353

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

Fix: manpage typo "-lllttng-ust" -> "-llttng-ust"

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

Fix: BSD getprogname null pointer dereference

Signed-off-by: Christian Babeux <christian.babeux@efficios.com>
Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>

Add support for model.emf.uri event info

Declared with:

TRACEPOINT_MODEL_EMF_URI(provider, event, "string")

this info is optional, can be added in the tracepoint probe.

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

Filter error message cleanup

When encountering an invalid opcode, don't print that an overflow has
occurred.

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

Manpage update: document use in daemons

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

Fix: get_wait_shm() ust mutex deadlock (add 2 missing exit calls)

Reported-by: changz <zheng.chang@emc.com>
Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>

Fix: get_wait_shm() ust mutex deadlock

The main issue is that get_wait_shm() bypass the fork() wrapper (with
lttng_ust_nest_count), which is responsible for holding the UST mutex
across fork(). Therefore, when exiting the context of the child process,
we execute the destructor, which try to grab the UST mutex, which might
be in pretty much any state.

Given that we don't want this process to try to register to
lttng-sessiond (because this is internal to lttng-ust), we might want to
let it skip the destructor execution. This would actually be the easiest
way out.

Reported-by: changz <zheng.chang@emc.com>
Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>

Fix: add events with 0 field to field list

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

Update version to 2.1.0-rc1

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

Fix make dist: fix liblttng-ust-java dependencies

Acked-by: Alexandre Montplaisir <alexandre.montplaisir@polymtl.ca>
Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>

Fix make dist: add missing filter header

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

Fix: backward compatibility with UST 2.0 app probes

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

Fix: Filter ABI changes to support FILTER_BYTECODE_MAX_LEN (65536)

In order to support the filter bytecode maximum length (65536 bytes),
the lttng_ust_filter_bytecode len field type must be able to
hold more than a uint16_t. Change the field type to a uint32_t.

Also, since the relocation table is located at the end of the actual
bytecode, the reloc_table_offset (reloc_offset in ust-abi) field must
support offset values larger than 65535. Change the field type to a
uint32_t. This change will allow support of relocation table appended
to larger bytecode without breaking the ABI if the need arise in the
future.

Both changes currently breaks the filter ABI, but this should be a
reasonable compromise since the filtering feature has not been
released yet.

Signed-off-by: Christian Babeux <christian.babeux@efficios.com>
Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>

Export "written" information about fields

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

Fix filter: pointer to string, not string, should be on stack

Fixes #329

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

Fix: tracepoint float nowrite

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

Fix: accept 65536 bytes long bytecodes

We were limited to 65535.

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

Add hostname to env

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

ABI change: bump internal version to 3.0.0

We keep compatiblity with applications (so we're still in the 2.x
versions), but we are breaking compatibility with lttng-consumerd.
Therefore, push the internal version number to 3.0.0.

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

Fix: threads should be created in DETACHED state

Signed-off-by: Michael Greene <michael.greene@gmail.com>
Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>

Fix: 32-bit x86 strict-aliasing warnings

Compiling on x86-32 shows the following warnings for filter (with gcc
4.3 and 4.4):

././ust_tests_hello.h:28: note: initialized from here
././ust_tests_hello.h:28: error: dereferencing pointer ‘__stack_data.18’ does break strict-aliasing rules

Fix it by using memcpy when copying to the temporary "stack" array used
to send arguments to the filter.

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

Fix UST SIGPIPE handling

When the consumerd dies (from a SIGKILL), it may close all of its file
descriptors rather abruptly.

We ensured that the UST command threads have all signals blocked, and
they use MSG_NOSIGNAL when sending messages to the sessiond over
sockets.

However, the consumer scheme uses a pipe(2) to transport the "wakeup"
info from the application tracing site to the consumer daemon. It may
send a SIGPIPE to the application in that case, which could kill the
application, an unwanted side-effect.

Block thread SIGPIPE around write() and wait for the signal to fix this.

Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>
Acked-by: Christian Babeux <christian.babeux@efficios.com>
CC: David Goulet <dgoulet@efficios.com>

Fix: Libtool fails to find dependent libraries when cross-compiling lttng-ust

This problem arise when cross compiling and linking libraries with
indirect libraries dependencies (such as liblttng-ust). This "bug" is
caused by an upstream modification in the libtool package on Debian
system. The libtool "link_all_deplibs" flag is set to "no" by default
on linux targets (AFAIK, other distros set it to "unknown").

The chosen solution is to detect such cases via the configure script
and automagically patch the libtool.m4 by forcing the "link_all_deplibs"
to "unknown".

This fixup can be disabled with the appropriate configure flag:

./configure --disable-libtool-linkdep-fixup

Sample configure output on affected systems:

checking for occurence(s) of link_all_deplibs = no in
         ./config/libtool.m4... 3
configure: WARNING: the detected libtool will not link all
           dependencies, forcing link_all_deplibs = unknown

Fixes: #321
Signed-off-by: Christian Babeux <christian.babeux@efficios.com>
Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>

Cleanup: filter: turn bytecode linking error msg into debug

This can be triggered by enable event on a wildcard, with some of the
events that do not contain the requested fields.

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

Make lttng-ust robust against -finstrument-functions.

[ Edit by Mathieu Desnoyers:

We need to declare the no_instrument_function attribute on function
declarations (rather than definition) for g++. Moved the attribute prior
to the function declaration (rather than after) to follow the coding
style within LTTng-UST. ]

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

Filter: keep aliased ax and bx registers

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

Filter: remove interpreter dynamic typing

Now all performed by instruction selection.

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

Filter: add missing specialized op names

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

Filter: specialize double-s64 binary comparators

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

Fix filter: fix stack leak on taken branch

Also fix return op stack check in validation.

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

Filter: Implement stack-based interpreter

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

Filter: double comparator produces s64

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

Filter: use hash table to check merge points

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

Filter: split passes into separate components

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

Filter: cleanup macros

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

Filter: validate range overflow with end of insn

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

Filter: validate that field ref strings are non-NULL

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

Filter: ensure logical operator merge is always s64

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

Filter: we don't care if double/s64 are literals

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

Filter: specialize 'and' and 'or' ops.

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

Implement dispatch-table based interpretor

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

Filter: Specialize unary operators

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

Define switch use as macro in interpreter

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

Filter interpreter: mark float test as unlikely

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

Filter: fix bytecode validation typo

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

Filter: specialize comparators

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

Specialize load and unary ops

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

Validate registers, no need to initialize to 0

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

Filter: opcode for ref loads

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

Remove redundant validation from interpreter

With the new validation pass, we can remove redundant validation from
the interpreter.

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

filter: Add bytecode validation pass

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

TRACEPOINT_EVENT: add *_nowrite fields for filter

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

Only print filter errors if LTTNG_UST_DEBUG is set

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

Cleanup: remove debug define

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

Filter: add floating point support

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

Remove filter test printouts

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

Implement filter bytecode interpreter and linker

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

Filter: receive, attach and link empty filter

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

Filter: prepare filter stack data

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

Wrap dynamic len array into stackvar union

In preparation for filter array.

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

liblttng-ust-comm/lttng-ust-com.c: remove unnecessary goto in ustcomm_accept_unix_sock()

Signed-off-by: Henrik Hautakoski <henrik@fiktivkod.org>
Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>

liblttng-ust/lttng-ust-comm.c: fixing typo.

Signed-off-by: Henrik Hautakoski <henrik@fiktivkod.org>
Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>

Fix: remove unused texinfo dep from configure.ac

Fixes #274

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

Fix C99 strict compatibility: don't use void * for function pointers

compiling public headers with --std=x99 -pedantic shows:

warning: ISO C forbids conversion of object pointer to function pointer type

Use "void (*func)(void)" to represent a generic function pointer rather
than "void *".

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

Fix c99 compatibility: tp_rcu_dereference_bp() should not use braced-groups within expressions

Allow tp_rcu_dereference_bp() to be used within programs compiled with
--std=c99 -pedantic -Werror. Fixes the following:

In file included from hello.c:34:0:
ust_tests_hello.h: In function ‘__tracepoint_cb_ust_tests_hello___tptest’:
ust_tests_hello.h:28:1: warning: ISO C forbids braced-groups within expressions [-pedantic]
ust_tests_hello.h:28:1: warning: ISO C forbids conversion of object pointer to function pointer type [-pedantic]

We can easily fix this one since tp_rcu_dereference_bp() really
evaluates only a single expression.

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

Revert "Fix c99 compatibility: tp_rcu_dereference_bp() should not use braced-groups within expressions"

This reverts commit a4eaf8eabe829be8f7d7432ffaf83291a068b0ed.

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

Fix c99 compatibility: tp_rcu_dereference_bp() should not use braced-groups within expressions

Allow tp_rcu_dereference_bp() to be used within programs compiled with
--std=c99 -pedantic -Werror. Fixes the following:

In file included from hello.c:34:0:
ust_tests_hello.h: In function ‘__tracepoint_cb_ust_tests_hello___tptest’:
ust_tests_hello.h:28:1: warning: ISO C forbids braced-groups within expressions [-pedantic]
ust_tests_hello.h:28:1: warning: ISO C forbids conversion of object pointer to function pointer type [-pedantic]

We can easily fix this one since tp_rcu_dereference_bp() really
evaluates only a single expression.

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

Fix: perform TLS fixup of nest count outside of UST mutex

Could otherwise lead to a deadlock between the dynamic linker, TLS
fixup, and the UST mutex due to reversed lock acquisition order.

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

Fix: liblttng-ust-fork deadlock

* Mathieu Desnoyers (mathieu.desnoyers@efficios.com) wrote:
> * Burton, Michael (mburton@ciena.com) wrote:
> > Mathieu,
> >
> > I think there is a deadlock scenario in UST, which has been causing my problem.
>
> Good catch !
>
> >
> > sessiond is started as root:
> >     - creates global sockets ONLY
> >     - DOES NOT CREATE shm in $HOME/lttng-ust-wait-<uid>
> >
> > application linked against ust is run as root:
> >   - in lttng_ust_init constructor
> >   - ust_listener_thread  (local_apps)
> >     - fails to connect to local_apps in $HOME/.lttng (as expected)
> >     - prev_connect_failed=1
> >     - ust_unlock()
> >     - restart
> >     - wait_for_sessiond()
> >  -->  - ust_lock()
> >  |      - get_map_shm()
> >  |        - get_wait_shm()
> > DEADLOCK    - shm_open() FAILS (not created by sessiond when run by root)
> >  |            - fork() (trying to create shared memory itself)
> >  |              - ust_before_fork()
> >  ------------>    - ust_lock()
> >
> >
> > You should be able to create this with an empty main, with no
> > tracepoints.  As long as sessiond is started as root so
> > $HOME/lttng-ust-wait-<uid> is not created.  You can also make the
> > lttng-ust constructor (lttng_ust_init) wait forever and then you'll be
> > able to see the deadlock in gdb without even leaving the
> > lttng_ust_init constructor.
>
> Ah, I see. This deadlock is caused by the interaction between
> [ liblttng-ust-fork ] and liblttng-ust (the fork override is
> performed by [ liblttng-ust-fork ]).

This can be reproduced easily with the in-tree tests: by removing the
lttng-ust-apps-wait* files belonging to the user in /dev/shm, running
the "tests/fork" test (with ./run) hangs. If we run "hello" first, and
then the fork test, it works fine.

Fixing this by keeping a nesting counter around the fork() call, so we
return immediately from the pre/post fork handlers if they are
overridden by liblttng-ust-fork.

Reported-by: Michael Burton <mburton@ciena.com>
Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>

Fix: handle pthread errors

Show a descriptive error message if pthread calls fail.

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

Fix: local apps allowed should disable local (not global) tracing

Reported-by: Michael Burton <mburton@ciena.com>
Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>

Fix strict ISO-C compatibility for ust-tracepoint-event.h public header

../../include/lttng/ust-tracepoint-event.h:328:23: warning: ISO C does not permit named variadic macros [-Wvariadic-macros]

Use ... and __VA_ARGS__ instead of args... and args.

This enables ISO-C compability for the tracepoint headers for program
instrumentation. Note that the probes need to be built _without_ strict
C99 flags (they require gnu extensions).

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

Fix: support -std=c99 in tracepoint macros

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

Fix c99 compatibility: use __typeof__ instead of typeof in public headers

Reported-by: John Steele Scott <toojays@toojays.net>
Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>

hello test: fail on old style definition

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

Fix: tracepoint.h should not generate old-style definitions

I've defined a tracepoint without arguments:

   TRACEPOINT_EVENT(

        qemu_tb_hash,

        flushall,

        TP_ARGS(void),

        TP_FIELDS()

   )

When I build (with -Werror=old-style-definition), I get this error:

   In file included from /home/hollisb/work/qemu.git/exec.c:59:0:

   /home/hollisb/work/qemu.git/tracepoints.h:24:2: error: function declaration isn’t a prototype [-Werror=strict-prototypes]

   /home/hollisb/work/qemu.git/tracepoints.h: In function ‘__tracepoint_cb_qemu_tb_hash___flushall’:

   /home/hollisb/work/qemu.git/tracepoints.h:24:2: error: old-style function definition [-Werror=old-style-definition]

   cc1: all warnings being treated as errors

The preprocessed code looks like so:

   extern struct

   # 129 "/usr/local/include/lttng/tracepoint.h" 3

                  tracepoint

   # 24 "/home/hollisb/work/qemu.git/tracepoints.h"

     __tracepoint_qemu_tb_hash___flushall

   # 19 "/home/hollisb/work/qemu.git/tracepoints.h"

     ; static __attribute__ (( always_inline )) __inline__ void

     __tracepoint_cb_qemu_tb_hash___flushall

   # 19 "/home/hollisb/work/qemu.git/tracepoints.h"

     () { struct tracepoint_probe *__tp_probe; if (!tracepoint_dlopen.rcu_read_lock_sym_bp) return; tracepoint_dlopen.rcu_read_lock_sym_bp(); __tp_probe = ({ typeof(__tracepoint_qemu_tb_hash___flushall.probes) _________p1 = ((typeof(__tracepoint_qemu_tb_hash___flushall.probes)) (tracepoint_dlopen.rcu_dereference_sym_bp(((void *) (__tracepoint_qemu_tb_hash___flushall.probes))))); (_________p1); }); if (__builtin_expect(!!(!__tp_probe), 0)) goto end; do { void *__tp_cb = __tp_probe->func; void *__tp_data = __tp_probe->data; ((void (*)(void *__tp_data)) (__tp_cb)) (__tp_data); } while ((++__tp_probe)->func); end: tracepoint_dlopen.rcu_read_unlock_sym_bp(); } static __attribute__ (( always_inline )) __inline__ void

I believe the problem comes from -Werror=old-style-definition not liking
that empty "()", i.e. tracepoint_cb_qemu_tb_hash___flushall() { ... }.

Reported-by: Hollis Blanchard <hollis_blanchard@mentor.com>
Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>