This allows tacking on suplimental data to the capture with a build-id so
that our SysprofDocumentMmap can read it. The format is @build-id\0 after
the trailing \0 of the filename in a SysprofCaptureMap frame.
This is like sample but has an "enter/exit" flag with it. This can be
useful when you want to provide tracing instead of sampling. We use a
different frame type so that we can denote that this isn't traditional
sampling, and the flag can be used to find the next exit for the current
enter for calculating durations.
The entire stack trace is provided to make things easier on tools
which may want to deal with indirect functions that were not instrumented
but can be unwound. That may allow for tooling to give the user some
insight that it's not *just* this function entering, but some functions
before it were entered too.
This also adds a SysprofTracer instrument which will preload a
libsysprof-tracer-6.so into the process providing the
__cyg_profile_func_enter() and __cyg_profile_func_leave() hooks.
Really what we want to deal with here is tracking an overlay that we may
need to be able to decode after the fact (in case processes exit or we
need to do post-processing symbol resolution).
For the podman case, that is $some_path mapped to root (/), generally
speaking. For flatpak though, that would have two mappings, one for
/app and another for /usr (possibly more).
While I'm not thrilled to add new frame types for every sort of thing, I
think having this will be relatively useful so we can improve decoding
operations.
This adds SysprofCapturePidRoot which lets us specify a root directory
on the host system for which is the real root (/) of the PID. This can
be useful when reconstructing overlays for containers and you need to
direct access to alternate roots.
The layer gives us some ability to try to deal with overlayfs, albeit at
a very rudimentary level. In most cases I anticipate we just deal with
the main root and ignore overlays until necessary.
None of the code uses it any more. This means that `libsysprof-capture.a`
can now be used within `libglib-2.0.so` for collecting main loop
statistics.
Brought to you by Opeth’s Deliverance on repeat.
Signed-off-by: Philip Withnall <withnall@endlessm.com>
Fixes: #40
Use `errno` instead, which is icky, but given that all of the failure
modes are from POSIX I/O functions, it’s at least in keeping with them.
This is a major API break.
Signed-off-by: Philip Withnall <withnall@endlessm.com>
Helps: #40
It was unused anywhere within sysprof.git, and couldn’t be modified to
drop its GLib dependency while still retaining its functionality.
If it’s still needed, it’ll have to be reimplemented in libsysprof.
Signed-off-by: Philip Withnall <withnall@endlessm.com>
Helps: #40
This means we lose support for local timezones other than UTC, but is
otherwise equivalent.
Signed-off-by: Philip Withnall <withnall@endlessm.com>
Helps: #40
libsysprof-capture no longer calls any of the GLib logging functions
which make use of it.
Signed-off-by: Philip Withnall <withnall@endlessm.com>
Helps: #40
Calling this function without having registered the counter beforehand
seems reasonable to call a programmer error.
Signed-off-by: Philip Withnall <withnall@endlessm.com>
Helps: #40
Use `strcmp()` and `strdup()` rather than `g_strcmp0()` and
`g_strdup()`. In the latter case, this makes no difference. In the
former case it means we potentially need to do some additional `NULL`
checks before calling it; although most of the call sites use
fixed-length arrays, so no `NULL` check is needed.
Signed-off-by: Philip Withnall <withnall@endlessm.com>
Helps: #40
Another step away from GLib. This changes the OOM behaviour of the
library — previously it would immediately `abort()` on OOM. However, it
seems likely that given the small number of allocations
libsysprof-capture does, it should be able to recover from an OOM
situation more gracefully than larger libraries can — so the new
implementation tries to do that.
Signed-off-by: Philip Withnall <withnall@endlessm.com>
Helps: #40
If the system doesn’t provide `strlcpy()` (FreeBSD does, Linux doesn’t),
use an inbuilt copy instead.
Signed-off-by: Philip Withnall <withnall@endlessm.com>
Helps: #40
Use the intrinsic atomics provided by the compiler, instead of GLib’s
wrapper around them. This should work for all modern compilers.
Signed-off-by: Philip Withnall <withnall@endlessm.com>
Helps: #40
Another step towards dropping GLib as a dependency of
libsysprof-capture.
Unlike the previous commit which replaced GLib integer types with the
bitwise equivalent C standard types, `stdbool` is potentially a different
width from `gboolean`, so this is an ABI break.
It therefore involves some changes to callback functions in the tests
and tools, and in libsysprof.
Signed-off-by: Philip Withnall <withnall@endlessm.com>
Helps: #40
This is an almost entirely mechanical* conversion from (for example)
`gint` → `int`, `guint8` → `uint8_t`, etc.
It is not entirely complete, as many GLib functions are still used in
libsysprof-capture, which necessitate some use of GLib types.
It also avoids renaming `gboolean` → `bool` as that’s a slightly more
controversial change which will happen in the following commit.
*Code was manually realigned afterwards.
Signed-off-by: Philip Withnall <withnall@endlessm.com>
Helps: #40
128 is a bit much and can slow us down considerably with user-space stack
traces. This can mess up the tree a bit, but we can alter how we view
things later on if we need to so that it is easier to read.
This brings over some of the techniques from the old memprof design.
Sysprof and memprof shared a lot of code, so it is pretty natural to
bring back the same callgraph view based on memory allocations.
This reuses the StackStash just like it did in memprof. While it
would be nice to reuse some existing tools out there, the fit of
memprof with sysprof is so naturally aligned, it's not really a
big deal to bring back the LD_PRELOAD. The value really comes
from seeing all this stuff together instead of multiple apps.
There are plenty of things we can implement on top of this that
we are not doing yet such as temporary allocations, cross-thread
frees, graphing the heap, and graphing differences between the
heap at to points in time. I'd like all of these things, given
enough time to make them useful.
This is still a bit slow though due to the global lock we take
to access the writer. To improve the speed here we need to get
rid of that lock and head towards a design that allows a thread
to request a new writer from Sysprof and save it in TLS (to be
destroyed when the thread exits).
This allows embedding the static capture library without having to deal
with license incompatibility while retaining the patent clause. We want
as many consumers of the format as makes sense without the restriction
of what kind of app/library/tooling they are.
The application, sysprofd, libsysprof, and libsysprof-ui continue to be
GPLv3+.
