Currently, the only information available from a stack frame is the
program counter. Eventually, we'd like to add support for getting
arguments and local variables, but that will require more work. In the
mean time, we can at least get the values of other registers. A
determined user can read the assembly for the code they're debugging and
derive the values of variables from the registers.
Rebase the existing patches and add the patches which extend the libdwfl
stack frame interface.
Based on:
47780c9e elflint, readelf: enhance error diagnostics
With the following patches:
configure: Add --disable-programs
configure: Add --disable-shared
configure: Fix -D_FORTIFY_SOURCE=2 check when CFLAGS contains -Wno-error
libcpu: compile i386_lex.c with -Wno-implicit-fallthrough
libdwfl: don't bother freeing frames outside of dwfl_thread_getframes
libdwfl: only use thread->unwound for initial frame
libdwfl: add interface for attaching to/detaching from threads
libdwfl: cache Dwfl_Module and Dwarf_Frame for Dwfl_Frame
libdwfl: add interface for evaluating DWARF expressions in a frame
In order to retrieve registers from stack traces, we need to know what
registers are defined for a platform. This adds a small DSL for defining
registers for an architecture. The DSL is parsed by an awk script that
generates the necessary tables, lookup functions, and enum definitions.
It's annoying to do obj.type_.size, and that doesn't even work for every
type. Add sizeof() that does the right thing whether it's given a Type
or Object.
For the following source code:
int arr[] = {};
GCC emits the following DWARF:
DWARF section [ 4] '.debug_info' at offset 0x40:
[Offset]
Compilation unit at offset 0:
Version: 4, Abbreviation section offset: 0, Address size: 8, Offset size: 4
[ b] compile_unit abbrev: 1
producer (strp) "GNU C17 9.2.0 -mtune=generic -march=x86-64 -g"
language (data1) C99 (12)
name (strp) "test.c"
comp_dir (strp) "/home/osandov"
stmt_list (sec_offset) 0
[ 1d] array_type abbrev: 2
type (ref4) [ 34]
sibling (ref4) [ 2d]
[ 26] subrange_type abbrev: 3
type (ref4) [ 2d]
upper_bound (sdata) -1
[ 2d] base_type abbrev: 4
byte_size (data1) 8
encoding (data1) signed (5)
name (strp) "ssizetype"
[ 34] base_type abbrev: 5
byte_size (data1) 4
encoding (data1) signed (5)
name (string) "int"
[ 3b] variable abbrev: 6
name (string) "arr"
decl_file (data1) test.c (1)
decl_line (data1) 1
decl_column (data1) 5
type (ref4) [ 1d]
external (flag_present) yes
location (exprloc)
[ 0] addr .bss+0 <arr>
Note the DW_AT_upper_bound of -1. We end up parsing this as UINT64_MAX
and returning a "DW_AT_upper_bound is too large" error. It appears that
GCC is simply emitting the array length minus one, so let's treat these
as having a length of zero.
Fixes#19.
There are a few places (e.g., Program.symbol(), Program.read()) where it
makes sense to accept, e.g., a drgn.Object with integer type. Replace
index_arg() with a converter function and use it everywhere that we use
the "K" format for PyArg_Parse*.
I got the error messages for DW_AT_upper_bound and DW_AT_count
backwards; fix it. Also fix the condition for word + 1 overflowing
dimension->length to be word >= UINT64_MAX. (Dwarf_Word is uint64_t so
this is kind of silly, but at least it documents the intent).
Sometimes, I'd like to see all of the missing debug info errors rather
than just the first 5. Allow setting this through the
DRGN_MAX_DEBUG_INFO_ERRORS environment variable.
Make the error message more concise, and reorder the sections so that we
check the most obviously-named section (.debug_info) first and least
important section (.debug_line) last.
Currently, the interface between the DWARF index, libdwfl, and the code
which finds and reports vmlinux/kernel modules is spaghetti. The DWARF
index tracks Dwfl_Modules via their userdata. However, despite
conceptually being owned by the DWARF index, the reporting code reports
the Dwfl_Modules and sets up the userdata. These Dwfl_Modules and
drgn_dwfl_module_userdatas are messy to track and pass between the
layers.
This reworks the architecture so that the DWARF index owns the Dwfl
instance and files are reported to the DWARF index; the DWARF index
takes care of reporting to libdwfl internally. In addition to making the
interface for the reporter much cleaner, this improves a few things as a
side-effect:
- We now deduplicate on build ID in addition to path.
- We now skip searching for vmlinux and/or kernel modules if they were
already indexed.
- We now support compressed ELF files via libdwelf.
- We can now load default debug info at the same time as additional
debug info.
It's undefined behavior to pass NULL to memcmp() even if the length is
zero. See also commit a17215e984 ("libdrgn: dwarf_index: fix memcpy()
undefined behavior").
vmcores don't include program headers for special memory regions like
vmalloc and percpu. Instead, we need to walk the kernel page table to
map those addresses. Luckily, libkdumpfile already does that. So, if
drgn was built with libkdumpfile support, use it for ELF vmcores. Also
add an environment variable to override this behavior.
Closes#15.
I didn't want to use BUILT_SOURCES before because that would break make
$TARGET. But, now that doesn't work anyways because we're using SUBDIRS,
so we might as well use BUILT_SOURCES.
Now that we have the bundled version of elfutils, build it from libdrgn
and link to it. We can also get rid of the elfutils version checks from
the libdrgn code.
Based on:
c950e8a9 config: Fix spec file, add manpages and new GFDL license.
With the following patches:
configure: Add --disable-programs
configure: Add --disable-shared
configure: Fix -D_FORTIFY_SOURCE=2 check when CFLAGS contains -Wno-error
libcpu: compile i386_lex.c with -Wno-implicit-fallthrough
The plan is to stop relying on the distribution's version of elfutils
and instead ship our own. This gives us freedom to assume that we're
using the latest version and even ship our own patches (starting with a
few build system improvements). More details are in
scripts/update-elfutils.sh, which was used to generate this commit.
Currently, we have a special Makefile target to output the files for a
libdrgn source tarball, and we use that for setuptools. However, the
next change is going to import elfutils, and it'd be a pain to add the
same thing for the elfutils sources. Instead, let's just use git
ls-files for everything. The only difference is that source
distributions won't have the autoconf/automake output.
I started with drgn_elf_relocator as a separate interface to parallelize
by relocation. However, the final result is parallelized by file, which
means that it can be done as part of the main read_cus() loop. Get rid
of the elf_relocator interface and do it in dwarf_index.c instead. This
means that if/when libdwfl gets faster at ELF relocations, we can rip
out the relocation code without any other changes.
We're too inconsistent with how we use these for them to be useful (and
it's impossible to distinguish between a format error and some other
error from libelf/libdw/libdwfl), so let's just get rid of them and make
it all DRGN_ERROR_OTHER/Exception.
For now, we only support stack traces for the Linux kernel (at least
v4.9) on x86-64, and we only support getting the program counter and
corresponding function symbol from each stack frame.
For stack trace support, we'll need to have some architecture-specific
functionality. drgn's current notion of an architecture doesn't actually
include the instruction set architecture. This change expands it to a
"platform", which includes the ISA as well as the existing flags.
Now that we're not overloading the name "symbol", we can define struct
drgn_symbol as a symbol table entry. For now, this is very minimal: it's
just a name, address, and size. We can then add a way to find the symbol
for a given address, drgn_program_find_symbol(). For now, this is only
supported through the actual ELF symbol tables. However, in the future,
we can probably support adding "symbol finders".
struct drgn_symbol doesn't really represent a symbol; it's just an
object which hasn't been fully initialized (see c2be52dff0 ("libdrgn:
rename object index to symbol index"), it used to be called a "partial
object"). For stack traces, we're going to have a notion of a symbol
that more closely represents an ELF symbol, so let's get rid of the
temporary struct drgn_symbol representation and just return an object
directly.
Currently, finders indicate a non-fatal lookup error by setting the type
member to NULL. This won't work when we replace the symbol finder with
an object finder (which shouldn't modify the object on failure).
Instead, use a static error for this purpose.
Currently, repr() of structure and union types goes arbitrarily deep
(except for cycles). However, for lots of real-world types, this is
easily deeper than Python's recursion limit, so we can't get a useful
repr() at all:
>>> repr(prog.type('struct task_struct'))
Traceback (most recent call last):
File "<console>", line 1, in <module>
RecursionError: maximum recursion depth exceeded while getting the repr of an object
Instead, only print one level of structure and union types.
We don't need to get the DWARF index at the time we get the Dwfl handle,
so get rid of drgn_program_get_dwarf(), add drgn_program_get_dwfl(), and
create the DWARF index right before we update in a new function,
drgn_program_update_dwarf_index().
After the libdwfl conversion, we apply ELF relocations with libdwfl
instead of our homegrown implementation. However, libdwfl is much slower
at it than the previous implementation. We can work around this by
(again) applying ELF relocations ourselves for architectures that we
care about (x86-64, to start). For other architectures, we can fall back
to libdwfl.
This new implementation of ELF relocation reworks the parallelization to
be per-file rather than per-relocation. The latter was done originally
because before commit 6f16ab09d6 ("libdrgn: only apply ELF relocations
to relocatable files"), we applied relocations to vmlinux, which is much
larger than most kernel modules. Now that we don't do that, it seems to
be slightly faster to parallelize by file.
libdwfl is the elfutils "DWARF frontend library". It has high-level
functionality for looking up symbols, walking stack traces, etc. In
order to use this functionality, we need to report our debugging
information through libdwfl. For userspace programs, libdwfl has a much
better implementation than drgn for automatically finding debug
information from a core dump or PID. However, for the kernel, libdwfl
has a few issues:
- It only supports finding debug information for the running kernel, not
vmcores.
- It determines the vmlinux address range by reading /proc/kallsyms,
which is slow (~70ms on my machine).
- If separate debug information isn't available for a kernel module, it
finds it by walking /lib/modules/$(uname -r)/kernel; this is repeated
for every module.
- It doesn't find kernel modules with names containing both dashes and
underscores (e.g., aes-x86_64).
Luckily, drgn already solved all of these problems, and with some
effort, we can keep doing it ourselves and report it to libdwfl.
The conversion replaces a bunch of code for dealing with userspace core
dump notes, /proc/$pid/maps, and relocations.
This converts several open-coded dynamic arrays to the new common vector
implementation:
- drgn_lexer stack
- Array dimension array for DWARF parsing
- drgn_program_read_c_string()
- DWARF index directory name hashes
- DWARF index file name hashes
- DWARF index abbreviation table
- DWARF index shard entries
drgn has enough open-coded dynamic arrays at this point to warrant a
common implementation. Add one inspired by hash_table.h. The API is
pretty minimal. I'll add more to it as the need arises.
There are some cases where we format a path (e.g., with asprintf()) and
keep it around only in case of errors. Add drgn_error_format_os() so we
can just reformat it if we hit the error, which simplifies cleanup.
Since we currently don't parse DWARF macro information, there's no easy
way to get the value PAGE_SIZE and friends in drgn. However, vmcoreinfo
contains the value of PAGE_SIZE, so let's add a special symbol finder
that returns that.
Currently, if we don't get vmcoreinfo from /proc/kcore, and we can't get
it from /sys/kernel/vmcoreinfo, then we manually determine the kernel
release and KASLR offset. This has a couple of issues:
1. We look for vmlinux to determine the KASLR offset, which may not be
in a standard location.
2. We might want to start using other information from vmcoreinfo which
can't be determined as easily.
Instead, we can get the virtual address of vmcoreinfo from
/proc/kallsyms and read it directly from there.
kernel_module_iterator_next() can also fail in
open_loaded_kernel_modules(), so handle it in the same way that we
currently handle kernel_module_iterator_init().
/proc/kcore contains segments which don't have a valid physical address,
which it indicates with a p_paddr of -1. Skip those segments, otherwise
we got an overflow error from the memory reader.
The current array-based memory reader has a bug in the following
scenario:
prog.add_memory_segment(0xffff0000, 128, ...)
# This should replace a subset of the first segment.
prog.add_memory_segment(0xffff0020, 32, ...)
# This moves the first segment back to the front of the array.
prog.read(0xffff0000, 32)
# This finds the first segment instead of the second segment.
prog.read(0xffff0032, 32)
Fix it by using the newly-added splay tree. This also splits up the
virtual and physical memory segments into separate trees.
This will be used to track memory segments instead of the array we
currently use. The API is based on the hash table API; it can support
alternative implementations in the future, like red-black trees.
This makes several improvements to the hash table API.
The first two changes make things more general in order to be consistent
with the upcoming binary search tree API:
- Items are renamed to entries.
- Positions are renamed to iterators.
- hash_table_empty() is added.
One change makes the definition API more convenient:
- It is no longer necessary to pass the types into
DEFINE_HASH_{MAP,SET}_FUNCTIONS().
A few changes take some good ideas from the C++ STL:
- hash_table_insert() now fails on duplicates instead of overwriting.
- hash_table_delete_iterator() returns the next iterator.
- hash_table_next() returns an iterator instead of modifying it.
One change reduces memory usage:
- The lower-level DEFINE_HASH_TABLE() is cleaned up and exposed as an
alternative to DEFINE_HASH_MAP() and DEFINE_HASH_SET(). This allows us
to get rid of the duplicated key where a hash map value already embeds
the key (the DWARF index file table) and gets rid of the need to make
a dummy hash set entry to do a search (the pointer and array type
caches).
Currently, we load debug information for every kernel module that we
find under /lib/modules/$(uname -r)/kernel. This has a few issues:
1. Distribution kernels have lots of modules (~3000 for Fedora and
Debian).
a) This can exceed the default soft limit on the number of open file
descriptors.
b) The mmap'd debug information can trip the overcommit heuristics
and cause OOM kills.
c) It can take a long time to parse all of the debug information.
2. Not all modules are under the "kernel" directory; some distros also
have an "extra" directory.
3. The user is not made aware of loaded kernel modules that don't have
debug information available.
So, instead of walking /lib/modules, walk the list of loaded kernel
modules and look up their debugging information.
const char * const * is not compatible with char * const *, so make
c_string_hash() and c_string_eq() macros so they can work with both
const char * and char * keys.
Currently, size_t and ptrdiff_t default to typedefs of the default
unsigned long and long, respectively, regardless of what the program
actually defines unsigned long or long as. Instead, make them refer the
whatever integer type (long, long long, or int) is the same size as the
word size.
Currently, programs can be created for three main use-cases: core dumps,
the running kernel, and a running process. However, internally, the
program memory, types, and symbols are pluggable. Expose that as a
callback API, which makes it possible to use drgn in much more creative
ways.
Similar to "libdrgn: make memory reader pluggable with callbacks", we
want to support custom type indexes (imagine, e.g., using drgn to parse
a binary format). For now, this disables the dwarf index tests; we'll
have a better way to test them later, so let's not bother adding more
test scaffolding.
I've been planning to make memory readers pluggable (in order to support
use cases like, e.g., reading a core file over the network), but the
C-style "inheritance" drgn uses internally is awkward as a library
interface; it's much easier to just register a callback. This change
effectively makes drgn_memory_reader a mapping from a memory range to an
arbitrary callback. As a bonus, this means that read callbacks can be
mixed and matched; a part of memory can be in a core file, another part
can be in the executable file, and another part could be filled from an
arbitrary buffer.
Currently, we deduplicate files for userspace mappings manually.
However, to prepare for adding symbol files at runtime, move the
deduplication to DWARF index. In the future, we probably want to
deduplicate based on build ID, as well.
Relocations are only supposed to be applied to ET_REL files, not ET_EXEC
files like vmlinux. This hasn't been an issue with the kernel builds
that I've tested on because the relocations match the contents of the
section. However, on Fedora, the relocation sections don't match,
probably because they post-process the binary in some way. This leads to
completely bogus debug information being parsed by drgn_dwarf_index. Fix
it by only relocating ET_REL files.
We need to set the value after we've reinitialized the object, otherwise
drgn_object_deinit() may try to free a buffer that we've already
overwritten. This also adds a test which triggers the crash.
There's a bug that we don't allow comparisons between void * and other
pointer types, so let's fix it by allowing all pointer comparisons
regardless of the referenced type. Although this isn't valid by the C
standard, GCC and Clang both allow it by default (with a warning).
drgn has pretty thorough in-program documentation, but it doesn't have a
nice overview or introduction to the basic concepts. This commit adds
that using Sphinx. In order to avoid documenting everything in two
places, the libdrgn bindings have their docstrings generated from the
API documentation. The alternative would be to use Sphinx's autodoc
extension, but that's not as flexible and would also require building
the extension to build the docs. The documentation for the helpers is
generated using autodoc and a small custom extension.
I went back and forth on using setuptools or autotools for the Python
extension, but I eventually settled on using only setuptools after
fighting to get the two to integrate well. However, setuptools is kind
of crappy; for one, it rebuilds every source file when rebuilding the
extension, which is really annoying for development. automake is a
better designed build system overall, so let's use that for the
extension. We override the build_ext command to build using autotools
and copy things where setuptools expects them.
The declaration file name of a DIE depends on the compilation directory,
which may not always be what the user expects. Instead, make the search
match as long as the full declaration file name ends with the given file
name. This is more convenient and more intuitive.
Running tests with Clang's AddressSanitizer fails with "runtime error:
index 1 out of bounds for type 'struct drgn_type_member [0]'". Zero
length arrays are a GCC extension and aren't buying us much anyways, so
just add a helper function that gets the array payload using pointer
arithmetic.
Older versions of Clang generate a call to __muloti4() for
__builtin_mul_overflow() with mixed signed and unsigned types. However,
Clang doesn't link to compiler-rt by default. Work around it by making
all of our calls to __builtin_mul_overflow() use unsigned types only.
1: https://bugs.llvm.org/show_bug.cgi?id=16404
The current mixed Python/C implementation works well, but it has a
couple of important limitations:
- It's too slow for some common use cases, like iterating over large
data structures.
- It can't be reused in utilities written in other languages.
This replaces the internals with a new library written in C, libdrgn. It
includes Python bindings with mostly the same public interface as
before, with some important improvements:
- Types are now represented by a single Type class rather than the messy
polymorphism in the Python implementation.
- Qualifiers are a bitmask instead of a set of strings.
- Bit fields are not considered a separate type.
- The lvalue/rvalue terminology is replaced with reference/value.
- Structure, union, and array values are better supported.
- Function objects are supported.
- Program distinguishes between lookups of variables, constants, and
functions.
The C rewrite is about 6x as fast as the original Python when using the
Python bindings, and about 8x when using the C API directly.
Currently, the exposed API in C is fairly conservative. In the future,
the memory reader, type index, and object index APIs will probably be
exposed for more flexibility.