If we create a pending CU for a namespace, then add more CUs to the
index, the CU might get reallocated, resulting in a use after free. Fix
it by storing the index of the CU instead of the pointer.
Signed-off-by: Omar Sandoval <osandov@osandov.com>
Debugging information tracking is currently in two places: drgn_program
finds debugging information, and drgn_dwarf_index stores it. Both of
these responsibilities make more sense as part of drgn_debug_info, so
let's move them there. This prepares us to track extra debugging
information that isn't pertinent to indexing.
This also reworks a couple of details of loading debugging information:
- drgn_dwarf_module and drgn_dwfl_module_userdata are consolidated into
a single structure, drgn_debug_info_module.
- The first pass of DWARF indexing now happens in parallel with reading
compilation units (by using OpenMP tasks).
Signed-off-by: Omar Sandoval <osandov@osandov.com>
DWARF represents namespaces with DW_TAG_namespace DIEs. Add these to the
DWARF index, with each namespace being its own sub-index. We only index
the namespace itself when it is first accessed, which should help with
startup time and simplifies tracking.
Signed-off-by: Jay Kamat <jaygkamat@gmail.com>
In order to index namespaces lazily, we need the CU structures. Rename
struct compilation_unit to the less generic struct drgn_dwarf_index_cu
and keep the CUs in a vector in the dindex.
Signed-off-by: Jay Kamat <jaygkamat@gmail.com>
For namespace support, we will want to access the struct
drgn_dwarf_index_die for namespaces instead of the Dwarf_Die. Split
drgn_dwarf_index_get_die() out of drgn_dwarf_index_iterator_next().
Signed-off-by: Omar Sandoval <osandov@osandov.com>
There are a couple of related ways that we can cause undefined behavior
when parsing a malformed DWARF or depmod index file:
1. There are several places where we increment the cursor to skip past
some data. It is undefined behavior if the result points out of
bounds of the data, even if we don't attempt to dereference it.
2. read_in_bounds() checks that ptr <= end. This pointer comparison is
only defined if ptr and end both point to elements of the same array
object or one past the last element. If ptr has gone past end, then
this comparison is likely undefined anyways.
Fix it by adding a helper to skip past data with bounds checking. Then,
all of the helpers can assume that ptr <= end and maintain that
invariant. while we're here and auditing all of the call sites, let's
clean up the API and rename it from read_foo() to the less generic
mread_foo().
Signed-off-by: Omar Sandoval <osandov@osandov.com>
Now that we can handle a DW_AT_specification that references another
compilation unit, add support for DW_FORM_ref_addr.
Signed-off-by: Omar Sandoval <osandov@osandov.com>
We currently handle DIEs with a DW_AT_specification attribute by parsing
the corresponding declaration to get the name and inserting the DIE as
usual. This has a couple of problems:
1. It only works if DW_AT_specification refers to the same compilation
unit, which is true for DW_FORM_ref{1,2,4,8,_udata}, but not
DW_FORM_ref_addr. As a result, drgn doesn't support the latter.
2. It assumes that the DIE with DW_AT_specification is in the correct
"scope". Unfortunately, this is not true for g++: for a variable
definition in a C++ namespace, it generates a DIE with
DW_AT_declaration as a child of the DW_TAG_namespace DIE and a DIE
which refers to the declaration with DW_AT_specification _outside_ of
the DW_TAG_namespace as a child of the DW_TAG_compilation_unit DIE.
Supporting both of these cases requires reworking how we handle
DW_AT_specification. This commit takes an approach of parsing the DWARF
data in two passes: the first pass reads the abbrevation and file name
tables and builds a map of instances of DW_AT_specification; the second
pass indexes DIEs as before, but ignores DIEs with DW_AT_specification
and handles DIEs with DW_AT_declaration by looking them up in the map
built by the first pass.
This approach is a 10-20% regression in indexing time in the benchmarks
I ran. Thankfully, it is not 100% slower for a couple of reasons. The
first is that the two passes are simpler than the original combined
pass. The second is that a decent part of the indexing time is spent
faulting in the mapped debugging information, which only needs to happen
once (even if the file is cached, minor page faults add non-negligible
overhead).
This doesn't handle DW_AT_specification "chains" yet, but neither did
the original code. If it is necessary, it shouldn't be too difficult to
add.
Signed-off-by: Omar Sandoval <osandov@osandov.com>
It's very unlikely that we'll ever index more than 4 billion DIEs in a
single shard, so we can shrink the index a bit by using uint32_t
indices (and uint8_t tag).
Signed-off-by: Omar Sandoval <osandov@osandov.com>
I originally copied the sections into each compilation unit to avoid a
pointer indirection, but performance-wise it's a wash, so we might as
well save the memory. This will be more important when we keep the CUs
after indexing.
Signed-off-by: Omar Sandoval <osandov@osandov.com>
In read_cus(), the master thread can use the final CUs vector directly
and the rest of the threads can merge their private vectors in. This
consistently shaves a few milliseconds off of startup.
Signed-off-by: Omar Sandoval <osandov@osandov.com>
We currently assume that if DW_AT_declaration is present, it is true.
This seems to be true in practice, and I see no reason to ever use
DW_FORM_flag with a value of zero. There's no performance hit to handle
it, though, so we might as well.
Signed-off-by: Omar Sandoval <osandov@osandov.com>
As a small simplification, we can take commit 9bb2ccecb7 ("Enable
DWARF indexing to work with partial units") further and not look at the
tag of the top-level DIE at all.
Signed-off-by: Omar Sandoval <osandov@osandov.com>
The CU unit length and DIE offset are both limited by the size of the
mapped debugging information, i.e., size_t.
Signed-off-by: Omar Sandoval <osandov@osandov.com>
This is more clear: although these flags happen to be encoded with the
DWARF tag, they are flags regarding the DIE.
Signed-off-by: Omar Sandoval <osandov@osandov.com>
If we fail to read an include directory in read_file_name_table(), we
need to free the directory hashes.
Signed-off-by: Omar Sandoval <osandov@osandov.com>
Commit 326107f054 ("libdrgn: add task_state_to_char() helper")
implemented task_state_to_char() in libdrgn so that it could be used in
commit 4780c7a266 ("libdrgn: stack_trace: prohibit unwinding stack of
running tasks"). As of commit eea5422546 ("libdrgn: make Linux kernel
stack unwinding more robust"), it is no longer used in libdrgn, so we
can translate it to Python. This removes a bunch of code and is more
useful as an example.
Signed-off-by: Omar Sandoval <osandov@osandov.com>
I originally did it this way because pydoc doesn't handle non-trivial
defaults in signature very well (see commit 67a16a09b8 ("tests: test
that Python documentation renders")). drgndoc doesn't generate signature
for pydoc anymore, though, so we don't need to worry about it and can
clean up the typing.
Signed-off-by: Omar Sandoval <osandov@osandov.com>
I originally envisioned types as dumb descriptors. This mostly works for
C because in C, types are fairly simple. However, even then the
drgn_program_member_info() API is awkward. You should be able to look up
a member directly from a type, but we need the program for caching
purposes. This has also held me back from adding offsetof() or
has_member() APIs.
Things get even messier with C++. C++ template parameters can be objects
(e.g., template <int N>). Such parameters would best be represented by a
drgn object, which we need a drgn program for. Static members are a
similar case.
So, let's reimagine types as being owned by a program. This has a few
parts:
1. In libdrgn, simple types are now created by factory functions,
drgn_foo_type_create().
2. To handle their variable length fields, compound types, enum types,
and function types are constructed with a "builder" API.
3. Simple types are deduplicated.
4. The Python type factory functions are replaced by methods of the
Program class.
5. While we're changing the API, the parameters to pointer_type() and
array_type() are reordered to be more logical (and to allow
pointer_type() to take a default size of None for the program's
default pointer size).
6. Likewise, the type factory methods take qualifiers as a keyword
argument only.
A big part of this change is updating the tests and splitting up large
test cases into smaller ones in a few places.
Signed-off-by: Omar Sandoval <osandov@osandov.com>
struct drgn_program has a bunch of state scattered around. Group it
together more logically, even if it means sacrificing some padding here
and there.
Signed-off-by: Omar Sandoval <osandov@osandov.com>
Most places that call these check has_platform and return an error, and
those that don't can live with the extra check.
Signed-off-by: Omar Sandoval <osandov@osandov.com>
Program::objects is used to store references to objects that must stay
alive while the Program is alive. It is currently a PyDict where the
keys are the object addresses as PyLong and the values are the objects
themselves. This has two problems:
1. Allocating the key as a full object is obviously wasteful.
2. PyDict doesn't have an API for reserving capacity ahead of time,
which we want for an upcoming change.
Both of these are easily fixed by using our own hash table.
Signed-off-by: Omar Sandoval <osandov@osandov.com>
drgn_object_init() is available in drgh.h file and seems to a required
call before calling drgn_program_find_object().
Without this, trying to call drgn_object_init() from an external C
application results in undefined reference.
Signed-off-by: Aditya Sarwade <asarwade@fb.com>
Now that drgndoc can handle overloads and we have the IntegerLike and
Path aliases, we can add type annotations to all helpers. There are also
a couple of functional changes that snuck in here to make annotating
easier.
Signed-off-by: Omar Sandoval <osandov@osandov.com>
Rather than duplicating Union[str, bytes, os.PathLike] everywhere, add
an alias. Also make it explicitly os.PathLike[str] or os.PathLike[bytes]
to get rid of some mypy --strict errors.
Signed-off-by: Omar Sandoval <osandov@osandov.com>
Lots if interfaces in drgn transparently turn an integer Object into an
int by using __index__(), so add an IntegerLike protocol for this and
use it everywhere applicable.
Signed-off-by: Omar Sandoval <osandov@osandov.com>
Make the KASLR offset available to Python in a new
drgn.helpers.linux.boot module, and move pgtable_l5_enabled() there,
too.
Signed-off-by: Omar Sandoval <osandov@osandov.com>
Clean up the coding style of the remaining few places that the last
couple of changes didn't rewrite.
Signed-off-by: Omar Sandoval <osandov@osandov.com>
The folly F14 implementation provides 3 storage policies: value, node,
and vector. The default F14FastMap/F14FastSet chooses between the value
and vector policies based on the value size.
We currently only implement the value policy, as the node policy is easy
to emulate and the vector policy would've added more complexity. This
adds support for the vector policy (adding even more C abuse :) and
automatically chooses the policy the same way as folly. It'd be easy to
add a way to choose the policy if needed.
Signed-off-by: Omar Sandoval <osandov@osandov.com>
The only major change to the folly F14 implementation since I originally
ported it is commit 3d169f4365cf ("memory savings for F14 tables with
explicit reserve()"). That is a small improvement for small tables and a
large improvement for vector tables, which are about to be added.
Signed-off-by: Omar Sandoval <osandov@osandov.com>
posix_memalign() doesn't have the restriction that the size must be a
multiple of the alignment like aligned_alloc() does in C11.
Signed-off-by: Omar Sandoval <osandov@osandov.com>
Compile-time constants have DW_AT_const_value instead of DW_AT_location.
We can translate those to a value object.
Signed-off-by: Omar Sandoval <osandov@osandov.com>
Really it's more of a test program than an example program. It's useful
for benchmarking, testing with valgrind, etc. It's not built by default,
but it can be built manually with:
$ make -C build/temp.* examples/load_debug_info
And run with:
$ ./build/temp.*/examples/load_debug_info
Signed-off-by: Omar Sandoval <osandov@osandov.com>
Commit eea5422546 ("libdrgn: make Linux kernel stack unwinding more
robust") overlooked that if the task is running in userspace, the stack
pointer in PRSTATUS obviously won't match the kernel stack pointer.
Let's bite the bullet and use the PID. If the race shows up in practice,
we can try to come up with another workaround.
I once tried to implement a generic arithmetic right shift macro without
relying on any implementation-defined behavior, but this turned out to
be really hard. drgn is fairly tied to GCC and GCC-compatible compilers
(like Clang), so let's just assume GCC's model [1]: modular conversion
to signed types, two's complement signed bitwise operators, and sign
extension for signed right shift.
1: https://gcc.gnu.org/onlinedocs/gcc/Integers-implementation.html
Declaring a local vector or hash table and separately initializing it
with vector_init()/hash_table_init() is annoying. Add macros that can be
used as initializers.
This exposes several places where the C89 style of placing all
declarations at the beginning of a block is awkward. I adopted this
style from the Linux kernel, which uses C89 and thus requires this
style. I'm now convinced that it's usually nicer to declare variables
where they're used. So let's officially adopt the style of mixing
declarations and code (and ditch the blank line after declarations) and
update the functions touched by this change.
We were forgetting to mask away the extra bits. There are two places
that we use the tag without converting it to a uint8_t:
hash_table_probe_delta(), which is mostly benign since we mask it by the
chunk mask anyways; and table_chunk_match() without SSE 2, which
completely breaks.
While we're here, let's align the comments better.
After thinking about it some more, I realized that "libdwfl: simplify
activation frame logic" breaks the case where during unwinding someone
queries isactivation for reasons other than knowing whether to decrement
program counter. Revert the patch and refactor "libdwfl: add interface
for getting Dwfl_Module and Dwarf_Frame for Dwfl_Frame" to handle it
differently.
Based on:
c95081596 size: Also obey radix printing for bsd format.
With the following patches:
configure: Add --disable-programs
configure: Add --disable-shared
libdwfl: add interface for attaching to/detaching from threads
libdwfl: export __libdwfl_frame_reg_get as dwfl_frame_register
libdwfl: add interface for getting Dwfl_Module and Dwarf_Frame for Dwfl_Frame
libdwfl: add interface for evaluating DWARF expressions in a frame
drgn has a couple of issues unwinding stack traces for kernel core
dumps:
1. It can't unwind the stack for the idle task (PID 0), which commonly
appears in core dumps.
2. It uses the PID in PRSTATUS, which is racy and can't actually be
trusted.
The solution for both of these is to look up the PRSTATUS note by CPU
instead of PID.
For the live kernel, drgn refuses to unwind the stack of tasks in the
"R" state. However, the "R" state is running *or runnable*, so in the
latter case, we can still unwind the stack. The solution for this is to
look at on_cpu for the task instead of the state.
We currently unwind from pt_regs and NT_PRSTATUS using an array of
register definitions. It's more flexible and more efficient to do this
with an architecture-specific callback. For x86-64, this change also
makes us depend on the binary layout rather than member names of struct
pt_regs, but that shouldn't matter unless people are defining their own,
weird struct pt_regs.
The model has always been that drgn Objects are immutable, but for some
reason I went through the trouble of allowing __init__() to reinitialize
an already initialized Object. Instead, let's fully initialize the
Object in __new__() and get rid of __init__().
It's annoying to have to do value= when creating objects, especially in
interactive mode. Let's allow passing in the value positionally so that
`Object(prog, "int", value=0)` becomes `Object(prog, "int", 0)`. It's
clear enough that this is creating an int with value 0.
drgn was originally my side project, but for awhile now it's also been
my work project. Update the copyright headers to reflect this, and add a
copyright header to various files that were missing it.
For functions that call a noreturn function, the compiler may omit code
after the call instruction. This means that the return address may not
lie in the caller's symbol. dwfl_frame_pc() returns whether a frame is
an "activation", i.e., its program counter is guaranteed to lie within
the caller. This is only the case for the initial frame, frames
interrupted by a signal, and the signal trampoline frame. For everything
else, we need to decrement the program counter before doing any lookups.
Rebase on master and fix dwfl_frame_module/dwfl_frame_dwarf_frame to
decrement the program counter when necessary.
Based on:
a8493c12a libdw: Skip imported compiler_units in libdw_visit_scopes walking DIE tree
With the following patches:
configure: Add --disable-programs
configure: Add --disable-shared
libdwfl: simplify activation frame logic
libdwfl: add interface for attaching to/detaching from threads
libdwfl: add interface for getting Dwfl_Module and Dwarf_Frame for Dwfl_Frame
libdwfl: export __libdwfl_frame_reg_get as dwfl_frame_register
libdwfl: add interface for evaluating DWARF expressions in a frame
Now that we can walk page tables, we can use it in a memory reader that
reads kernel memory via the kernel page table. This means that we don't
need libkdumpfile for ELF vmcores anymore (although I'll keep the
functionality around until this code has been validated more).
I originally wanted to avoid depending on another vmcoreinfo field, but
an the next change is going to depend on swapper_pg_dir in vmcoreinfo
anyways, and it ends up being simpler to use it.
There are a few big use cases for this in drgn:
* Helpers for accessing memory in the virtual address space of userspace
tasks.
* Removing the libkdumpfile dependency for vmcores.
* Handling gaps in the virtual address space of /proc/kcore (cf. #27).
I dragged my feet on implementing this because I thought it would be
more complicated, but the page table layout on x86-64 isn't too bad.
This commit implements page table walking using a page table iterator
abstraction. The first thing we'll add on top of this will be a helper
for reading memory from a virtual address space, but in the future it'd
also be possible to export the page table iterator directly.
UNARY_OP_SIGNED_2C() uses a union of int64_t and uint64_t to avoid
signed integer overflow... except that there's a typo and the uint64_t
is actually an int64_t. Fix it and add a test that would catch it with
-fsanitize=undefined.
internal.h includes both drgn-specific helpers and generic utility
functions. Split the latter into their own util.h header and use it
instead of internal.h in the generic data structure code. This makes it
easier to copy the data structures into other projects/test programs.
Program_load_debug_info() is the last user of the
resize_array()/realloc_array() utility functions. We can clean it up by
using a vector and finally get rid of those functions.
This also happens to fix three bugs in Program_load_debug_info(): we
weren't setting a Python exception if we couldn't allocate the path_args
array, we weren't zeroing path_args after resizing the array, and we
weren't freeing the path_args array. Shame on whoever wrote this.
DRGN_UNREACHABLE() currently expands to abort(), but assert() provides
more information. If NDEBUG is defined, we can use
__builtin_unreachable() instead.
DRGN_UNREACHABLE() isn't drgn-specific, so this renames it to
UNREACHABLE(). It's also not really related to errors, so this moves it
to internal.h.
Before Linux v4.11, /proc/kcore didn't have valid physical addresses, so
it's currently not possible to read from physical memory on old kernels.
However, if we can figure out the address of the direct mapping, then we
can determine the corresponding physical addresses for the segments and
add them.
We treat core dumps with all zero p_paddrs as not having valid physical
addresses. However, it is theoretically possible for a kernel core dump
to only have one segment which legitimately has a p_paddr of 0 (e.g., if
it only has a segment for the direct mapping, although note that this
isn't currently possible on x86, as Linux on x86 reserves PFN 0 for the
BIOS [1]).
If the core dump has a VMCOREINFO note, then it is either a vmcore,
which has valid physical addresses, or it is /proc/kcore with Linux
kernel commit 23c85094fe18 ("proc/kcore: add vmcoreinfo note to
/proc/kcore") (in v4.19), so it must also have Linux kernel commit
464920104bf7 ("/proc/kcore: update physical address for kcore ram and
text") (in v4.11) (ignoring the possibility of a franken-kernel which
backported the former but not the latter). Therefore, treat core dumps
with a VMCOREINFO note as having valid physical addresses.
1: https://git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git/tree/arch/x86/kernel/setup.c?h=v5.6#n678
Even on 32-bit architectures, physical addresses can be 64 bits (e.g.,
x86 with PAE). It's unlikely that the vmcoreinfo note would be in such
an address, but let's always parse it as a uint64_t just to be safe.
I've found that I do this manually a lot (e.g., when digging through a
task's stack). Add shortcuts for reading unsigned integers and a note
for how to manually read other formats.
DrgnObject_getattro() uses PyObject_GenericGetAttr() and catches the
AttributeError raised if the name is not an attribute of the Object
class. If the member is found, we then destroy the AttributeError.
Raising an exception only to destroy it is obviously wasteful. Luckily,
as of Python 3.7, the lower-level _PyObject_GenericGetAttrWithDict() can
suppress the AttributeError; we can raise it ourselves if we need it. In
my microbenchmarks, this makes Object.__getattribute__() at least twice
as fast when the member exists.
This also fixes a drgn_error leak.
Loading debuginfo from a kernel with separate or re-combined debuginfo
results in the following failure:
Traceback (most recent call last):
File "/home/jeffm/.local/bin/drgn", line 11, in <module>
load_entry_point('drgn==0.0.3+39.gbf05d9bf', 'console_scripts', 'drgn')()
File "/home/jeffm/.local/lib/python3.6/site-packages/drgn-0.0.3+39.gbf05d9bf-py3.6-linux-x86_64.egg/drgn/internal/cli.py", line 121, in main
prog.load_debug_info(args.symbols, **args.default_symbols)
Exception: invalid DW_AT_decl_file 10
A typical kernel build results in DWARF information with
DW_TAG_compile_unit tags but the objcopy --only-keep-debug and eu-unstrip
commands use DW_TAG_partial_unit tags. The DWARF indexer only handles
the former, so the file table isn't populated and we get the exception.
Fortunately, the two are more or less interchangeable. (See
http://dwarfstd.org/doc/Dwarf3.pdf, section E.2.3). Accepting both for
indexing compile units results in a working session.
compound_initializer_init_next() saves a pointer to the compound
initializer stack and uses it after appending to the stack, which may
have reallocated the stack.
For C++ support, we need to add an array of template parameters to
struct drgn_type. struct drgn_type already has arrays for members,
enumerators, and parameters embedded at the end of the structure,
because no type needs more than one of those. However, struct, union,
and class types may need members and template parameters. We could add a
separate array of templates, but then it gets confusing having two
methods of storing arrays in struct drgn_type. Let's make these arrays
separate instead of embedding them.
Similarly to PAGE_OFFSET, vmemmap makes more sense as part of the Linux
kernel object finder than an internal helper.
While we're here, let's fix the definition for 5-level page tables. This
only matters for kernels with commit 77ef56e4f0fb ("x86: Enable 5-level
paging support via CONFIG_X86_5LEVEL=y") but without eedb92abb9bb
("x86/mm: Make virtual memory layout dynamic for CONFIG_X86_5LEVEL=y")
(namely, v4.14, v4.15, and v4.16); since v4.17, 5-level page table
support enables KASLR.
The internal _page_offset() helper gets the value of PAGE_OFFSET, but
the fallback when KASLR is disabled has been out of date since Linux
v4.20 and never handled 5-level page tables. Additionally, it makes more
sense as part of the Linux kernel (formerly vmcoreinfo) object finder so
that it's cleanly accessible outside of drgn internals.
The configure script allows the user to not use any openmp
implementation but dwarf_index.c uses the locking APIs unconditionally.
This compiles but fails at runtime.
Adding simple stubs for the locking API. This is useful when debugging
crashes in dwarf indexing during development.
elfutils 0.179 was just released, and it includes my fix for vmcores
with >= 2^16 phdrs.
Based on:
3a7728808 Prepare for 0.179
With the following patches:
configure: Add --disable-programs
configure: Add --disable-shared
libdwfl: add interface for attaching to/detaching from threads
libdwfl: add interface for getting Dwfl_Module and Dwarf_Frame for Dwfl_Frame
libdwfl: export __libdwfl_frame_reg_get as dwfl_frame_register
libdwfl: add interface for evaluating DWARF expressions in a frame
Jay reported that the default language detection was happening too early
and not finding "main". We need to make sure to do it after the DWARF
index is actually populated. The problem with that is that it makes
error reporting much harder, as we don't want to return a fatal error
from drgn_program_set_language_from_main() if we actually succeeded in
loading debug info. That means we probably need to ignore errors in
drgn_program_set_language_from_main(). To reduce the surface area where
we'd be failing, let's get the language directly from the DWARF index.
This also allows us to avoid setting the language if it's actually
unknown (information which is lost by the time we convert it to a
drgn_object in the current code).
Currently, drgn_language_from_die() returns the default language when it
encounters an unknown DW_LANG because the dwarf_info_cache always wants
a language. The next change will want to detect the unknown language
case, so make drgn_language_from_die() return NULL if the language is
unknown, move it to language.c, and fold drgn_language_from_dw_lang()
into it.
We need to keep the Program alive for its types to stay valid, not just
the objects the Program has pinned. (I have no idea why I changed this
in commit 565e0343ef ("libdrgn: make symbol index pluggable with
callbacks").)
For operations where we don't have a type available, we currently fall
back to C. Instead, we should guess the language of the program and use
that as the default. The heurisitic implemented here gets the language
of the CU containing "main" (except for the Linux kernel, which is
always C). In the future, we should allow manually overriding the
automatically determined language.
For types obtained from DWARF, we determine it from the language of the
CU. For other types, it can be specified manually or fall back to the
default (C). Then, we can use the language for operations where the type
is available.
This way, languages can be identified by an index, which will be useful
for adding Python bindings for drgn_language and for adding a language
field to drgn_type.
