Btrfs maintains a fs_uuids cache, but there is no defined way to look
into it. This script reads the cache and prints details of the
filesystems present in it.
Signed-off-by: Srivathsa Dara <srivathsa.d.dara@oracle.com>
In the uncached case, the task variable is supposed to be the task that
matched the vm area. But, if we detect that CONFIG_VMAP_STACK is
disabled, we break without clearing the task variable. Fix that, and
also make sure to test both the cached and uncached cases.
Closes#414.
Fixes: 69bd07ddb2 ("drgn.helpers.common.memory: add optional cache to identify_address()")
Reported-by: Sourabh Jain <sourabhjain@linux.ibm.com>
Signed-off-by: Omar Sandoval <osandov@osandov.com>
At Meta, our Python executables are packaged in a way that doesn't leave
the original .py or .pyc files easily accessible or importable. This
makes the sudohelper fail with:
can't open file '/proc/self/fd/3/drgn/internal/sudohelper.pyc': [Errno 2] No such file or directory
We can work around this by passing the helper command as a string
literal. It's ugly, but it works with our Meta package and with
traditional installations.
Signed-off-by: Omar Sandoval <osandov@osandov.com>
The test for print_annotated_memory() shouldn't assert anything when
CONFIG_SLOB is enabled, because we can't find slab objects. However, the
test shouldn't be skipped entirely, because print_annotated_memory()
should run without error in this case. Skip the assertion only.
Fixes: c7717280 ("helpers.common.memory: add print_annotated_memory() helper")
Signed-off-by: Stephen Brennan <stephen.s.brennan@oracle.com>
In the Fixes commit, as part of the refactor, the
_identify_kernel_address() function started open-coding the
get_slab_object_info() function. Unfortunately, this resulted in the
function failing when CONFIG_SLOB was enabled, since the AttributeError
for accessing "slab.slab_cache" was not caught.
Open-coding was done to avoid the duplicate bounds check for
in_direct_map. In fact, in_direct_map seems like a useful helper itself,
and by factoring it out of _find_containing_slab() and into its callers,
we can simplify things a bit. Since _find_containing_slab() can handle
SLOB, this fixes the test failure too.
Fixes: 48e0c51b ("helper.common.memory: recognize vmap addresses in identify_address()")
Signed-off-by: Stephen Brennan <stephen.s.brennan@oracle.com>
The Github Actions disk space can get filled up if we test all kernel
flavors. So, delete kernels after testing when running in Github
Actions. However, if the download speed is far greater than the speed of
tests, then there is also the possibility that the disk space will be
filled up by the download thread, before the tests will run and delete
the older kernels. So we also need to ensure the download thread can't
get too far ahead of the tests, which we implement by setting a max size
on the download queue.
Signed-off-by: Stephen Brennan <stephen.s.brennan@oracle.com>
This makes it simple to run all tests for a specific flavor. That's
especially useful if submitting a pull request, where CI will test the
default flavor, but you want to double check some other flavor first.
Signed-off-by: Stephen Brennan <stephen.s.brennan@oracle.com>
Stephen noticed in #415 that if a new stable release is tagged after the
default vmtest build job runs but before another flavor does, we can end
up in a situation where the newest kernel for a given version isn't the
default flavor. This causes `setup.py test -K` to test a non-default
flavor. `-K` is specifically supposed to test the default, so specify it
explicitly.
Reported-by: Stephen Brennan <stephen.s.brennan@oracle.com>
Signed-off-by: Omar Sandoval <osandov@osandov.com>
There are a few common use cases that I find myself re-writing:
iterating over the list of modules, looking up a module by name, and
getting memory addresses related to modules (or vice versa). Add helpers
for each.
Signed-off-by: Stephen Brennan <stephen.s.brennan@oracle.com>
Python 3.12 introduces the ability to nest the quote character within
f-strings. Previously, this was not allowed. While contrib scripts don't
have any specific requirement for code quality or compatibility, it
would be nice to make this runnable on Python versions prior to 3.12.
Signed-off-by: Stephen Brennan <stephen.s.brennan@oracle.com>
Unlike `bpftool p` to show bpf map ids, `drgn contrib/bpf_inspect.py p
--show-details` is able to show more bpf map info, which is useful to
identify the used `.rodata` map.
Signed-off-by: Leon Hwang <hffilwlqm@gmail.com>
CentOS 7 is now EOL, so the manylinux2014 build is failing; see
pypa/manylinux#1641. Work around it by using the archive mirror.
Signed-off-by: Omar Sandoval <osandov@osandov.com>
libomp leaks memory on fork (see __kmp_atfork_child() in
llvm-project/openmp/src/z_Linux_util.cpp), which doesn't really matter
except that it causes our new C unit tests to fail under LeakSanitizer
because libdrgn has a constructor that calls omp_get_max_threads(). We
could work around it by using CK_NOFORK in the unit tests, but it's not
ideal for us to go digging in sysfs on process startup just because you
linked to libdrgn anyways. Initialize the number of threads when we
actually need it instead, which just requires some care around
concurrency.
Signed-off-by: Omar Sandoval <osandov@osandov.com>
Free-threading Python binaries aren't ABI compatible with the
traditional builds that still have a GIL. But currently the build_temp
directory, which is drgn's incremental build directory, is shared
between both kinds of interpreters. This can result in runtime issues
like segmentation faults if build artifacts are reused. To avoid this,
add a "t" suffix for free-threading build directories. We should only do
this if the "build-temp" option is not set by the user.
Signed-off-by: Stephen Brennan <stephen.s.brennan@oracle.com>
The latter fails on Python 3.13 since setuptools isn't installed by
default. If we need to install something anyways, we might as well use
the nicer pytest.
Signed-off-by: Omar Sandoval <osandov@osandov.com>
kdump_blob_new() was added in libkdumpfile 0.4.1.
Fixes: 6c7b271a53 ("libdrgn: kdump: pass architecture and vmcoreinfo to libkdumpfile")
Signed-off-by: Omar Sandoval <osandov@osandov.com>
When run drgn on kernel v5.5~5.9, there is no `dst_trampoline` in `prog->
aux`. At this time, `tr` must be inited as `None` in order to avoid
panic when get `trampoline` from `prog->aux`.
Signed-off-by: Leon Hwang <hffilwlqm@gmail.com>
In order to inspect more and more info of bpf objects, this commit
abstracts classes, such as `BpfProg`, `BpfMap`, `BpfLink` and so on.
Signed-off-by: Leon Hwang <hffilwlqm@gmail.com>
print_annotated_memory() and print_annotated_stack() are pretty slow
when called on large memory regions, largely because identify_address()
walks every task for every call on a non-direct map address in order to
resolve vmap stacks. We can cache the mapping from a vmap stack area to
a task, but we need callers to be able to indicate when it's okay to
cache in order to avoid stale results. To do so, add a cache parameter
to identify_address(), which is just an opaque dictionary that should be
passed between successive calls.
Signed-off-by: Omar Sandoval <osandov@osandov.com>
Drgn is used by multiple companies and stakeholders in a variety of
situations. I think it qualifies as a production / stable. Also, add in
the Linux kernel trove classifier.
Signed-off-by: Stephen Brennan <stephen@brennan.io>
Since 'prog->aux->linked_prog' has been moved to trampoline, it has to
inspect linked progs from tracing link's trampoline.
Therefore, add --show-details for link command to show the linked progs
for tracing link.
Signed-off-by: Leon Hwang <hffilwlqm@gmail.com>
We got a couple of reports about drgn failing to get a stack trace
(#391) or getting the wrong stack trace (#404) from a kernel core dump.
Both were caused because drgn assumes that there is an NT_PRSTATUS note
for each CPU in order by CPU number, and in these core dumps some
NT_PRSTATUS notes were missing. There are a least a couple of things
that can cause this: offline CPUs or CPUs that were in a bad state and
didn't respond to the kdump NMI. The former is expected and could be
special-cased, but the latter basically means that we can't trust the
order of the notes. Instead, look up the notes from the crash_notes
per-CPU variable that the kernel uses to populate the ELF notes. We
still need to use the actual NT_PRSTATUS notes for QEMU
dump-guest-memory dumps, but for those we need to use the PID field to
handle CPU hotplugging.
Closes#404.
Signed-off-by: Omar Sandoval <osandov@osandov.com>
We have a few places where we parse raw ELF notes independently of
libelf, which isn't complicated but also not trivial thanks to alignment
requirements. In preparation for adding another place, factor the
parsing out into a common helper. Also document the complexities around
figuring out the correct alignment.
Signed-off-by: Omar Sandoval <osandov@osandov.com>
The vmap_nodes global variable is optimized out on 32-bit kernels, so we
need to hard-code a workaround for the find_vmap_area() and
for_each_vmap_area() helpers.
Signed-off-by: Omar Sandoval <osandov@osandov.com>
The path of the Python interpreter in the Debian rootfs is
/usr/bin/python3, which may not be the same as sys.executable on the
host.
Signed-off-by: Omar Sandoval <osandov@osandov.com>
When running tests manually with vmtest.vm, it's tedious to set up the
test kernel module. Make vmtest.vm.run_in_vm() handle it and add command
line options.
Signed-off-by: Omar Sandoval <osandov@osandov.com>
