* relocation: handle ARM binaries

* relocation: handle R_ARM_ABS32 for ARM machines

* testfiles: add reloc_arm_gcc.o.elf

Generated on Ubuntu 14.04 using: arm-linux-gnueabi-gcc-4.7 -c -g -o reloc_armhf_gcc.o.elf hello.c

* testfiles: add reloc_armhf_gcc.o.elf

Generated on Ubuntu 14.04 using: arm-linux-gnueabihf-gcc-4.7 -c -g -o reloc_armhf_gcc.o.elf hello.c

* readelf: print soft-float abi for ARM if EF_ARM_ABI_FLOAT_SOFT in flags

* readelf: print hard-float abi for ARM if EF_ARM_ABI_FLOAT_HARD in flags

* readelf: print BE8 info for armeb binaries

* testfiles: add simple_armhf_gcc.o.elf

    Generated on Ubuntu 14.04 using: arm-linux-gnueabihf-gcc-4.7  -g -o simple_armhf_gcc.o.elf hello.c

* elf: remove unwind from dicts and set ARM_EXIDX description

* testfiles: add  reloc_armsf_gcc.o.elf as soft float testcase taken from binutils 2.30

* testfiles: add reloc_armeb_gcc.o.elf as arm big endian testcase taken from binutils 2.30 testcase arm-be8

* readelf: print endian info LE8 if flag was set in header flags

* Add support for 'R_ARM_CALL' relocation type

* Add test script and test files to verify support for 'R_ARM_CALL'

Signed-off-by: Koltunov Dmitry <[email protected]>

… tests

* Provide enums for DT_FLAGS and DT_FLAGS_1

This change adds two enums with the name to value mappings
for the two flags fields in the dynamic section. The values
and corresponding names are taken from the elf/elf.h file
in the most recent glibc version.

The enums are also used to print the names instead of the
raw hex values for DT_FLAGS and DT_FLAGS_1 in
scripts/readelf.py.

Fixes: eliben#189

* Add test file for DT_FLAGS/DT_FLAGS_1 parsing

The test file has the DF_BIND_NOW and DF_ORIGIN flags set
in DT_FLAGS as well as DF_1_NOW, DF_1_GLOBAL, DF_1_NOOPEN
and DF_1_ORIGIN flags in DF_FLAGS_1.

This is the source code for the dt_flags.elf file:

  #include <stdio.h>

  int function(const char *arg){
      printf("Hello, %s!", arg);
      return 0;
  }

and was compiled using the following command line:

$ gcc -shared -m32 \
  -Wl,-rpath,'$ORIGIN/lib',-z,global,-z,origin,-z,nodlopen,-z,now \
  -o testfiles_for_readelf/dt_flags.elf dt_flags.c

The __init__ function of ARMAttribute has two parameters
structs and stream through which the caller can pass in the
relevant objects (ARMAttributesSubsubsection does that after
seeking to the right position in stream).

The accesses for TAG_SECTION and TAG_SYMBOL, however, were
referring to non-existing members instead of the parameters.

Additionally, one assertion tries to access an undefined
'null_byte' variable which should be 'nul' instead.

The stream position in the .debug_info stream can't change when
reading from the .debug_abbrev stream.

…eliben#206)

* Implemented ELFFile.get_machine_arch for the remaining architectures.

Added all architectures according to the ENUM_E_MACHINE.

* Refactored if statement into dict.get.

The code that is intended to coalesce null DIEs into the DIE that
precedes them does not do that and is actually not needed as the
'unflattening' procedure takes care of any unexpected null DIEs.

Also added a unit test for verifying the DIE size calculation.

…ns (eliben#208)

* Added support for decoding .debug_pubtypes and .debug_pubnames sections

* Added reference output to dwarf_pubnames_types.py example.

* Added readelf support, fixed review comments and documentation updates

* Avoid printing the entire die in pubnames example to workaround Python2 vs 3 imcompatibilites

Create all the AbbrevDecl objects during parsing and later return
references to them - this gives a small performance gain.

…#214)

In DWARFv4 the location lists are referenced with the 'sec_offset'
attribute form instead of 'data4' or 'data8'.

* tox: explicitly set locale

Locale affects GNU binutils output translation which cause
run_readelf_tests.py to fail if system language is not English.

Signed-off-by: Efimov Vasily <[email protected]>

* test: unittest reproducing error with empty ".debug_pubtypes" section

Signed-off-by: Efimov Vasily <[email protected]>

* NameLUT: use `construct.If` to declare "name" field

This patch also fixes problem with empty first entry.

Signed-off-by: Efimov Vasily <[email protected]>

* NameLUT._get_entries: remove unused `bytes_read`

Signed-off-by: Efimov Vasily <[email protected]>

StringTableSection.get_string() returns an UTF-8 decoded
string (or '' if fetching the string failed) since eliben#182
but the code in _DynamicStringTable was never updated to
decode anything at all so it just returns a bytes sequence
in Python 3.

Let's convert the string there as well to be able to use
both string tables the same way without having to worry
about decoding. Adapt the test cases accordingly.

On macOS I'm getting the following error when testing with tox on py27:

```
ERROR: invocation failed (exit code 1), logfile: /devel/pyelftools/.tox/py27/log/py27-33.log
ERROR: actionid: py27
msg: installpkg
cmdargs: ['/devel/pyelftools/.tox/py27/bin/pip', 'install', '-U', '--no-deps', '/devel/pyelftools/.tox/dist/pyelftools-0.25.zip']

DEPRECATION: Python 2.7 will reach the end of its life on January 1st, 2020. Please upgrade your Python as Python 2.7 won't be maintained after that date. A future version of pip will drop support for Python 2.7.
Processing ./.tox/dist/pyelftools-0.25.zip
    Complete output from command python setup.py egg_info:
    Traceback (most recent call last):
      File "<string>", line 1, in <module>
      File "/private/var/folders/qz/XXX/T/pip-req-build-890d2p/setup.py", line 47, in <module>
        scripts=['scripts/readelf.py']
      File "/devel/pyelftools/.tox/py27/lib/python2.7/site-packages/setuptools/__init__.py", line 144, in setup
        _install_setup_requires(attrs)
      File "/devel/pyelftools/.tox/py27/lib/python2.7/site-packages/setuptools/__init__.py", line 137, in _install_setup_requires
        dist.parse_config_files(ignore_option_errors=True)
      File "/devel/pyelftools/.tox/py27/lib/python2.7/site-packages/setuptools/dist.py", line 704, in parse_config_files
        self._parse_config_files(filenames=filenames)
      File "/devel/pyelftools/.tox/py27/lib/python2.7/site-packages/setuptools/dist.py", line 600, in _parse_config_files
        reader = io.TextIOWrapper(fp, encoding=encoding)
    LookupError: unknown encoding:
```

This is due to the specification of LC_ALL as simply `en_US` without an encoding. Python 3.x seems to be fine with this, but Python 2.7 barfs. As a fix, setting `LC_ALL` to `en_US.utf-8` (including an explicit encoding spec) works.

dynamic: parse DT_{GNU_}HASH for number of symbols

In ultra-stripped binaries we can find the symbol table by
parsing the dynamic segment and using the pointer in the
DT_SYMTAB tag as the base address. However, we don't know
anything about the number of symbols in the symbol table.

Earlier, this code relied on finding the closest pointer
value bigger than the base address of the symbol table. In
PIE executables and shared libraries however this method
could break as the pointer value for DT_SYMTAB is in the same
range as things like DT_RELASZ or DT_STRSZ, leading to a too
small number of symbols returned by iter_symbols().

The crashpad project has implemented a different strategy to
find the number of symbols: parsing the symbol lookup hash
tables (see [0]) as every symbol must have a corresponding
entry in the hash table. This commit implements this
behaviour for DynamicSegment, leaving the old code as a
backup if neither DT_HASH or DT_GNU_HASH tags have been
found.

For DT_HASH type tables, it is quite easy as the header
already contains the number of entries. For DT_GNU_HASH
things are a bit more complicated as we need to work forward
from the highest symbol referenced in the header (a good
explanation of the format can be found at [1]).

[0]: chromium/crashpad@1f1657d
[1]: https://flapenguin.me/2017/05/10/elf-lookup-dt-gnu-hash/

* dynamic: provide more functions for symbol access

So far, the DynamicSegment only provided a method to iterate
over all symbols but for some use cases it might be useful to
use the recovered symbol table more like a normal
SymbolTableSection.

To this end, provide get_symbol(index) to fetch a symbol by
its index, num_symbols() to get the total number of symbols
and get_symbol_by_name(name) to look for a list of symbols
with a given name.

$SITE_PYTHON/lib/python3.7/site-packages/elftools/construct/lib/container.py:5
 Using or importing the ABCs from 'collections' instead of from 'collections.abc' is deprecated, and in 3.8 it will stop working

This change is compatible with Python 3.3 and up, when the ABCs were
moved to collections.abc. Backward compatibility is retained through
the try/except block.

… 3.7

The standard defines only DW_LANG_UPC, and this
value also contained a typo.

* test: test `CFIEntry.get_decoded()`

This test detects an error in `CFIEntry.get_decoded()`, that occurs when
decodes the `DW_CFA_def_cfa_register` instruction without some CFA
definition previously.

Signed-off-by: Koltunov Dmitry <[email protected]>

* add empty `cfa` for fixup decode of the `DW_CFA_def_cfa_register`

Signed-off-by: Koltunov Dmitry <[email protected]>

This commit moves some of the location-handling code from the examples
to a new class (LocationParser) in order to make it more reusable.

Also adds two test files containing location information.

Tested with Python 3.8

Based on eliben#237 by @Plailect. Closes eliben#237

Most of these were added in DWARFv5.

* dwarf/die: Handle DW_FORM_flag_present in value translation

When an attribute has form DW_FORM_flag_present it is implicitly
indicated as present, with no actual value.

Ref. DWARFv4, section 7.

* test: Add DW_FORM_flag_present value test

* test: Fix iteration

* test: Remove old assert

`setup.py bdist_wheel` warns about not finding README.

Supersedes/closes eliben#216.

Closes eliben#258

While the comment in section_in_segment() suggests that the
logic follows the logic inside ELF_SECTION_IN_SEGMENT_1 with
the strict parameter set, all of the checks in the binutils
macro are written so that they must succeed for the section
to be contained in the current segment. In our implementation,
however, the checks were not properly negated.

This showed in the case of .tdata and .tbss which did not
appear in the section to segment mapping (these sections are
found in glibc, for example).

Fix it up by aligning the logic more closely to the binutils
macro by implementing the same logic and returning False only
if the checks fail. Additionally, introduce the third check
from the upstream binutils which checks the combination of
SHT_ALLOC sections and PT_LOAD-like segments.

Furthermore, in the original check, the PT_GNU_RELRO type was
misspelled with a 0 (zero) instead of an O so this check
could never have worked.

Fixes: eliben#263

Fixes eliben#260

Add a forgotten import of SizeofError.

Fixes: eliben#278

This adds an example of the operation discussed in eliben#283.

Usage:

   python3 ./dwarf_lineprogram_filenames.py --test x.elf y.elf z.elf

In DWARF 2, the DW_FORM_ref_addr format matches the target address size, while in DWARF3+ it matches the bitness of the CU record. Here are the relevant fragments from the spec, part 7:

v2:

    The second type of reference is the address of any debugging information entry within the same executable or shared object; it may refer to an entry in a different compilation unit from the unit containing the reference. This type of reference (DW_FORM_ref_addr) is the size of an address on the target architecture; it is relocatable in a relocatable object file and relocated in an executable file or shared object.

v3:

    The second type of reference can identify any debugging information entry within a program; in particular, it may refer to an entry in a different compilation unit from the unit containing the reference, and may refer to an entry in a different shared object. This type of reference (DW_FORM_ref_addr) is an offset from the beginning of the .debug_info section of the target executable or shared object; it is relocatable in a relocatable object file and frequently relocated in an executable file or shared object. For references from one shared object or static executable file to another, the relocation and identification of the target object must be performed by the consumer. In the 32-bit DWARF format, this offset is a 4-byte unsigned value; in the 64-bit DWARF format, it is an 8-byte unsigned value (see Section 7.4).

If elftools encounters 32-bit DWARF v2 targeting a 64-bit architecture, it will misparse DW_FORM_ref_addr and crash downstream.

I encountered this in an iOS binary from 2017, built with Xcode several versions ago for ARM64. This probably never came up before because by the time 64 bit code became relevant, most toolchains would generate DWARF 3 or newer.


Co-authored-by: Seva Alekseyev <[email protected]>

…information (eliben#287)

Some ELF files which contain only debug symbols have important sections present in the section table but marked as NOBITS instead of PROGBITS. Attempting to extract the segments can lead to crashes through parsing invalid data.

The first patch modifies the dynamic segment/section specifically to add a flag for this case, since it seems to assume that there will always be at least one entry, DT_NULL.

The second patch modifies the segment code more generally to return a dummy answer for what data it holds. The actual way that this change prevents a crash is while trying to parse .eh_frame when it is in fact NOBITS - originally I had a more targeted patch, but decided that it was important enough to do more generally

Fixes eliben#274

* readelf.py: adapt section mapping output for .tbss sections

GNU readelf does not show the .tbss section as part of the
loaded data segment when listing the section to segment
mappings, using the ELF_TBSS_SPECIAL macro in
include/elf/internal.h to skip printing the section name.

Implement the same logic in readelf.py.

* test: add test file for .tbss output in readelf.py

This test file includes a .tbss section which is not output
by GNU readelf as part of the loaded data segment when
listing the section to segment mappings.

The source code for tls.elf is simply:

  __thread int i;

  int main(){}

The file was compiled using the following command line:

$ gcc -o tls.elf -m32 tls.c

* readelf: print addend for RELA relocations without symbol

When processing relocations from a SHT_RELA type section, GNU
readelf displays the value of the 'r_addend' field if no
symbol index is given (that is, 'r_info_sym' is 0).

By also implementing this we can better test the output for
64-bit binaries which commonly use SHT_RELA relocations.

The included test file is the same as tls.elf but compiled
for x86_64. Its code is the following:

  __thread int i;

  int main(){}

and it is compiled using the following command line:

$ gcc -m64 -o tls64.elf tls.c

* test: add source file for tls{,64}.elf

The comments at the top describe how to compile the source
file into tls.elf and tls64.elf.

* Add handling for SHT_MIPS_ABIFLAGS section types

* Add handling for SHF_MASKPROC section flags

* Add handling for DT_MIPS_FLAGS dynamic table entries

* Display DT_MIPS_SYMTABNO and DT_MIPS_LOCAL_GOTNO entries as decimal ints

* Adjust display of NT_GNU_GOLD_VERSION notes

In super-stripped binaries, symbol tables can not be accessed
directly as we do not have section headers to find them. In
this case, we can already use the mandatory DynamicSegment
which provides methods for individual access and iteration
over symbols via a minimal implementation of symbol hash
sections which only provided the number of symbols so far.

As we can also directly look up symbols via the hash table,
let's implement this functionality as well.

The code is based on @rhelmot's implementation as discussed
in eliben#219, with some changes around reading the hash parameters.

For supporting individual symbol lookup, we also need the
corresponding symbol table to get the Symbol objects if the
matching hash was found in the hash section. In regular ELF
files, the symbol table is denoted by the section index
provided in the sh_link field of the hash section and
automatically created when building the hash section, for
super-stripped binaries we can use the DynamicSegment (which
needs to be present in any case) as the symbol table as it
also provides a get_symbol() method relying on other ways to
determine the list of symbols. Both of these variants can be
seen in the improved test_hash.py file.

The hash tables are implemented in a base class which does not
derive from the Section class in order to allow instantiation
even if no section headers are present in the underlying file.

* readelf.py: add an option to show traceback on error

* readelf.py: flush stdout before printing to sys.stderr

This is necessary to make error messages appear after any display that
was emitted before the error actually happened.

* Handle type2/type3 relocation fields for ELF64 MIPS binaries

* dwarf/callframe.py: fix field read using the DW_EH_PE_absptr encoding

This encoding represents target addresses, so it is the virtual address
space determines its size, not the DWARF format.

Fixes eliben#288

Basic unit tests pass, but old code is still in place and descriptions is not
yet converted.

)

In descriptions, ExprDumper invokes parse_expr many times on small
expressions. Initializing the dispatch table for every parse is
wasteful.

Wrap parse_expr with a simple object that generates and caches the
dispatch table during initialization. parse_expr remains stateless.

Updates eliben#298

We no longer need the part-by-part dumping and separate process/get_str.

Also simplify tests.

Fixes eliben#298

…locations (eliben#300)

* Add handling for SHF_MASKOS section flags

* Add readelf testcases for MIPS64 specificities

* Simplify the decoding of MIPS64 relocations

Instead of using "fake" fields to parse the relocation structure and
then use complex shift/masks to recover the conveyed information (once
for big endian binaries, twice for little endian ones), use fields
actually described in the spec and use straightforward shifts to
synthetize the "fake" fields.

Previous commit broke them because lambdas can't be picked by multiprocessing

* dwarf/compileunit: Lookup DIE from a reference

Accept a resolved reference address for a DIE in a compile unit and
parse the DIE at that location.  Insert into the _diemap / _dielist
cache shared with iter_DIE_children() for fast repeated lookups.

This can be used to follow attribute references to a DIE that be
referenced several times (eg for a DW_AT_type reference) or find
a DIE referenced in a lookup table.

* dwarf/dwarfinfo: Cache CUs, direct parse or search from known units

Maintain a cache of compile units parsed and a map of their offsets
similar to the one mainained of DIEs by compile units.

Add the ability to parse a random compile unit when the offset of
the compile unit header is known.

Add the ability to search for a compile unit containing (spanning)
a given refaddr, such as that obtained from a DIE reference class
attribute, starting from the closest previous cached compile unit.

* dwarf/die: search for parents on demand

Add a function to set the _parent link of known chldren, iterating
down each parent of a target DIE.  Walk all children of a given
parent and set each child's ._parent to avoid O(n^2) walking.

A future commit will add other methods to instatiate a DIE that will
not set the _parent link as the DIE is instantiated.

This walk uses the knowledge that in a flattened tree a parents offset
will always be less than the childs.

The call to die.set_parent in compile_unit iter_DIE_children could be
removed to make the method private,, but it is free to set starting
from the top DIE.  Alternativly make it an optional argument to
DIE creation.

* dwarf/dwarfinfo: APIs to lookup DIEs

Add APIs to lookup a DIE from: (a) a DIE reference class attribute
taking into account the attribute form, (b) from a lookup table entry
(NameLUTEntry) from a .pub_types or .pub_names section, or (c) directly
from a reference addresss (.debug_info offset) regardless of how it
was obtained.

Add a test that will lookup dies from pubnames and follow die by ref.

	This is a simple test that exercises the new cache lookup
	methods and provides a starting point on how to determine a
	variables type.

For now raise NotImplemented exception for type signature lookup
and supplemental dwarf object files.

Jython has neither `resource` nor `mmap`, therefore just use a
reasonable default.

* fix issue in aranges cu_offset_at_addr

if there are aranges for some parts of the binary but not others, incorrect aranges may be returned

* add test suite for absent/partial/complete aranges