ELF modules are loaded in memory in several, possibly discontiguous,
segments. If the holes between segments are large enough, other things,
possibly other ELF modules may be mapped in that space. Crashpad
records the range of modules as the base address of the lowest mapped
segment to the high address of the highest mapped segment. This means
that when one module is mapped into a hole in another, it appears to
the Breakpad processor as overlapping modules. Module ranges are
relevant to the Breakpad processor during stackwalking for identifying
which module a particular program counter belongs to (i.e. mapping the
address to a module's text segment). This patch addresses this issue of
overlapping modules by truncating the range of the module with the
lower base address. A typical module's text segment is the first loaded
segment which would leave the text segment range unaffected. Module
producers can restrict the size of holes in their ELF modules with the
flag "-Wl,-z,max-page-size=4096", preventing other modules from being
mapped in their address range.
Properly contemplating ELF module address ranges would require
extensions to the minidump format to encode any holes.
crbug.com/crashpad/298
This patch also renames the concept of "shrinking down" (which
truncated the upper of two overlapping ranges) to "truncate upper".
Change-Id: I4599201f1e43918db036c390961f8b39e3af1849
Reviewed-on: https://chromium-review.googlesource.com/c/breakpad/breakpad/+/1646932
Reviewed-by: Mark Mentovai <mark@chromium.org>
The path NSCachesDirectory may change across app updates and sometimes
even across app launches. As a result, the Config-XXX files may end up
with an outdated path to the associated minidump file.
Change-Id: I0befde26b2ac406c154ce7c7e9be0063ee99892d
Bug:850379
Reviewed-on: https://chromium-review.googlesource.com/c/breakpad/breakpad/+/1592561
Reviewed-by: Mark Mentovai <mark@chromium.org>
This CL adds a result callback on report upload completion.
On failure, Breakpad deletes the configuration file and does retry to
upload a report.
Using this callback, the client will be able to log some metrics and to
act on upload failure.
Bug: 954175
Change-Id: I95a3264b65d4c06ba5d8dde8377440d23f1e2081
Reviewed-on: https://chromium-review.googlesource.com/c/breakpad/breakpad/+/1572661
Reviewed-by: Mark Mentovai <mark@chromium.org>
Chrome's test runner on Linux installs its own StackDumpSignalHandler
which swallows signals and doesn't re-raise them. This is sloppy, but
apparently there are reasons (https://crbug.com/551681). For
breakpad_unittests, it causes problems where a test process expects (via
waitpid()) to observe a child crash. Deal with those cases by
explicitly restoring the default signal handler.
In another case, Chrome's test runner seems to have been arriving at the
conclusion that it was to expect output from a child. Transitioning from
exit() to _exit() fixes this problem, and it's not necessarily a bad
idea to do this in post-fork() children without an execve() anyway.
Bug: chromium:949098
Change-Id: I5a6af0c2a09cd8eac9998358f6d5ea665288236f
Reviewed-on: https://chromium-review.googlesource.com/c/breakpad/breakpad/+/1575670
Reviewed-by: Mark Mentovai <mark@chromium.org>
Running this test on android-kitkat-arm-rel fails for some reason:
[ RUN ] LinuxCoreDumperTest.VerifyExceptionDetails
linux_core_dumper_unittest.cc:170: Failure
Expected: (0U) != (dumper.crash_address()), actual: 0 vs 0
linux_core_dumper_unittest.cc:178: Failure
Expected equality of these values:
2U
Which is: 2
info.size()
Which is: 0
[ FAILED ] LinuxCoreDumperTest.VerifyExceptionDetails (7 ms)
Disable it for now on Android until someone can look into it.
Bug: google-breakpad:791
Change-Id: I40a5e3dbeeb44e5eb0df187e61d55e07d8ad3613
Reviewed-on: https://chromium-review.googlesource.com/c/breakpad/breakpad/+/1546778
Reviewed-by: Mark Mentovai <mark@chromium.org>
Some of the fields we save might have signed types depending on the
system (e.g. `typedef int pid_t`). Depending on the toolchain, we
can trip -Werror=narrowing failures like:
src/client/linux/minidump_writer/linux_core_dumper.cc:248:66: error:
narrowing conversion of ‘(__pid_t)info->siginfo_t::_sifields.siginfo_t::<anonymous union>::_kill.siginfo_t::<anonymous union>::<anonymous struct>::si_pid’
from ‘__pid_t {aka int}’ to ‘long unsigned int’ inside { } [-Werror=narrowing]
set_crash_exception_info({info->si_pid, info->si_uid});
^^^^^^
src/client/linux/minidump_writer/linux_core_dumper.cc:252:71: error:
narrowing conversion of ‘(int)info->siginfo_t::_sifields.siginfo_t::<anonymous union>::_sigsys.siginfo_t::<anonymous union>::<anonymous struct>::_syscall’
from ‘int’ to ‘long unsigned int’ inside { } [-Werror=narrowing]
set_crash_exception_info({info->si_syscall, info->si_arch});
^^^^^^^^^^
Since the exception info fields are all uint64_t which should be large
enough to handle all the fields in the siginfo_t structure, add casts
for all the assignments to avoid these errors. We have implicit casts
even without them, so we aren't changing behavior.
Bug: google-breakpad:791
Bug: chromium:945653
Change-Id: Ib04e015998f08b857159ac13e9a065a66d228d49
Reviewed-on: https://chromium-review.googlesource.com/c/breakpad/breakpad/+/1544862
Reviewed-by: Mark Mentovai <mark@chromium.org>
Even 64-bit Mach-O (MH_MAGIC_64 = 0xfeedfacf) is not a fully 64-bit file
format. File offsets in sections are stored in 32-bit fields, with
Mach-O writers typically truncating offsets too large to fit to just
their low 32 bits. When a section begins at a file offset >= 4GB,
dump_syms would produce an error such as:
Google Chrome Framework.dSYM/Contents/Resources/DWARF/Google Chrome Framework: the section '__apple_names' in segment '__DWARF' claims its contents lie outside the segment's contents
As a workaround, this implements the strategy I first described in
https://crbug.com/940823#c22.
Segment file offsets are stored in 64-bit fields. Because segments
contain sections and must load contiguously, it’s possible to infer a
section’s actual offset by computing its load address relative to its
containing segment’s load address, and treating this as an offset into
the containing segment’s file offset. For safety, this is only done for
64-bit segments (LC_SEGMENT_64) where the 32-bit section offset stored
in the Mach-O file is equal to the low (truncated) 32 bits of the
section offset recomputed per the above strategy.
Beware that this does not provide full “large file” support for 64-bit
Mach-O files. There are other file offsets within Mach-O files aside
from section file offsets that are stored in 32-bit fields even in the
64-bit format, including offsets to symbol table data (LC_SYMTAB and
LC_DYSYMTAB). No attempt is made to recover correct file offsets for
such data because, at present, such data is always stored by dsymutil
near the beginning of .dSYM files, within the first 4GB. If it becomes
necessary to address these other offsets, it should be possible to
recover these offsets by reference to the __LINKEDIT segment that
normally contains them, provided that __LINKEDIT doesn’t span more than
4GB, according to the strategy discussed at the bottom of
https://crbug.com/940823#c22.
Although this is sufficient to allow dump_syms to interpret Chromium
.dSYM files that exceed 4GB, be warned that these Mach-O files are still
technically malformed, and most other tools that consume Mach-O files
will continue to have difficulties interpreting these large files.
As further warning, note that should any individual DWARF section exceed
4GB, internal section offsets will be truncated irrecoverably, unless
and until the toolchain implements support for DWARF64.
https://bugs.llvm.org/show_bug.cgi?id=14969
With this change, dump_syms is able to correctly recover file offsets
from and continue processing a .dSYM file with length 4530593528
(4321MB), whose largest section (__DWARF,__debug_info = .debug_info) has
size 0x8d64c0b8 (2262MB), and which contains four sections (starting
with __DWARF,__apple_names) beginning at file offsets >= 4GB.
Bug: chromium:940823, chromium:946404
Change-Id: I23f5f3b07773fa2f010204d5bb53b6fb1d4926f7
Reviewed-on: https://chromium-review.googlesource.com/c/breakpad/breakpad/+/1541830
Reviewed-by: Robert Sesek <rsesek@chromium.org>
Reviewed-by: Mike Frysinger <vapier@chromium.org>
- Baselines appeared to be stale. dump_syms now prints FUNC entries
with the full function signature, whereas the baselines only
contained the function name. The current state of the symbol file
docs
(https://chromium.googlesource.com/breakpad/breakpad/+/refs/heads/master/docs/symbol_files.md)
seem to agree with the new FUNC entries rather than the old ones.
Example of a name given in current docs:
"nsQueryInterfaceWithError::operator()(nsID const&, void**) const".
Change-Id: I9e01354cd82b7184b7cba31d132603e949a657ac
Reviewed-on: https://chromium-review.googlesource.com/c/breakpad/breakpad/+/1529133
Reviewed-by: Ivan Penkov <ivanpe@chromium.org>
If the kernel/C library headers are old, they might not have the
fields needed for SIGSYS decoding. Add ifdef checks for that and
skip the logic entirely. Easier than adding arch-specific siginfo
structs to the codebase.
Bug: google-breakpad:791
Change-Id: Ia473e3ffa61fce4c42cf4c1e73a9df044599bc5c
Reviewed-on: https://chromium-review.googlesource.com/c/breakpad/breakpad/+/1524447
Reviewed-by: Mark Mentovai <mark@chromium.org>
Many signals in Linux support additional metadata on a per-signal
basis. We can extract that from NT_SIGINFO and pass it through
in the exception_information fields.
The current core dumper logic doesn't set exception_information
at all, so this is an improvement.
Bug: google-breakpad:791
Change-Id: I38b78d6494e9bc682441750d98ac9be5b0656f5a
Reviewed-on: https://chromium-review.googlesource.com/c/breakpad/breakpad/+/1497662
Reviewed-by: Mark Mentovai <mark@chromium.org>
When building on an old system with outdated headers, this define
might not be available. Add a fallback to our existing elf header.
Bug: google-breakpad:790
Change-Id: I4dfe7a5cebd414cca3582a1a9cfc983503d5a779
Reviewed-on: https://chromium-review.googlesource.com/c/breakpad/breakpad/+/1507073
Reviewed-by: Mark Mentovai <mark@chromium.org>
The current failure message omits the underlying errno. This can
make diagnosing failures a bit difficult unless you run everything
through strace. For example:
$ core2md core /proc/self md
$ core2md core /proc/self md
Unable to generate minidump
Now we get the errno details:
Unable to generate minidump: File exists
Change-Id: I67f30879868ce4a726d5d888ee8c0a4a316b5186
Reviewed-on: https://chromium-review.googlesource.com/c/breakpad/breakpad/+/1497660
Reviewed-by: Ted Mielczarek <ted.mielczarek@gmail.com>
The current core dumper only parses NT_PRSTATUS notes. With signal
details, this note only includes three fields: signo, code, and errno.
We set exception_code to signo and exception_flag to code. The errno
value isn't set by the kernel, so there's no need to save it.
However, we never fill in exception_address which means all converted
crashes look like they happen at address 0. This implies a NULL jump
which is usually not the case, so it's just confusing. The prstatus
structure doesn't offer anything directly that tracks this.
Starting with linux-3.7, the kernel writes out the full siginfo
structure in the NT_SIGINFO note. So lets support that to pull out
si_addr which, for a bunch of common signals, is the value we want in
exception_address.
The size of the siginfo_t structure should be locked to 128 bytes at
build time for all architectures, so this should hopefully be stable.
Bug: google-breakpad:790
Change-Id: I458bad4787b1a8b73fad8fe068e9f23bec957599
Reviewed-on: https://chromium-review.googlesource.com/c/1497661
Reviewed-by: Mark Mentovai <mark@chromium.org>
TYPED_TEST_CASE is deprecated in modern googletest.
BUG=chromium:936654
Change-Id: I08004ffbb26089ebe17302934ed6d3268220d151
Reviewed-on: https://chromium-review.googlesource.com/c/1493423
Reviewed-by: Mark Mentovai <mark@chromium.org>
Certain minidumps for 32-bit crashes have the upper 32-bit of the crash
address (which is a 64-bit value) set to non-zero values. This caused a
crash address with more than 32-bits to be printed out for minidumps of
32-bit architectures. This patch masks out those bits when reading the
raw minidump data to ensure this doesn't happen anymore.
Bug: google-breakpad:783
Change-Id: Ieef6dff759fd0ee2efc47c4c4a3cf863a48f0659
Reviewed-on: https://chromium-review.googlesource.com/c/1427819
Reviewed-by: Ted Mielczarek <ted.mielczarek@gmail.com>
This affects the output of tools like minidump_stackwalk which currently
print out the hexadecimal representation of the architecture instead of
the "arm64" string.
BUG=780
Change-Id: Id1d9d65fa5f3509c8c6580e2e3042f7d682b52be
Reviewed-on: https://chromium-review.googlesource.com/c/1412004
Reviewed-by: Mike Frysinger <vapier@chromium.org>
Eventually, I want to remove the current version of
SetFirstChanceHandler. That is why I changed the name of the current
callback type to FirstChanceHandlerDeprecated.
I also made sure that it is not possible to have two different
FirstChanceHandlers set at the same time.
This is the first of a set of CLs to clean up the API between Chrome,
BreakPad, and V8. See more information in the tracking bug.
R=mark@chromium.org
Bug: chromium:921971
Change-Id: Ia8c2fd9bd875c36dd7ae8bb4a02e538556bc67a1
Reviewed-on: https://chromium-review.googlesource.com/c/1411776
Reviewed-by: Mark Mentovai <mark@chromium.org>
This allows BPLOG_LAZY_STREAM to be overridden by BP_LOGGING_INCLUDE
Change-Id: I5c9ec19b619ad5db9e97f3a1813b0f965a357b38
Reviewed-on: https://chromium-review.googlesource.com/c/1351361
Reviewed-by: Ivan Penkov <ivanpe@chromium.org>
dwarf_range_list_handler.{cc,h} were added in 16e08520.
Default to building with c++11.
Change-Id: Iceb29ab665260a9e71a30920fdfb5623d10a9cfa
Reviewed-on: https://chromium-review.googlesource.com/c/1351351
Reviewed-by: Mark Mentovai <mark@chromium.org>
This allows BPLOG_IF to be overriden by defines in BP_LOGGING_INCLUDE.
Change-Id: Ic6e8373476cc4d1f73d55e13a23686a2c8309fdc
Reviewed-on: https://chromium-review.googlesource.com/c/1278104
Reviewed-by: Ivan Penkov <ivanpe@chromium.org>
Also adds method to determine if breakpad is started.
Change-Id: I272765e7ac6bbc07d77ca2d8dcc34d51c205116e
Reviewed-on: https://chromium-review.googlesource.com/c/1260625
Reviewed-by: Joshua Peraza <jperaza@chromium.org>
Also remove ranges_handler_ which is unused.
Change-Id: I771bf4b5fc4410f0406bf26e1e405905b55389ab
Reviewed-on: https://chromium-review.googlesource.com/1180587
Reviewed-by: Mark Mentovai <mark@chromium.org>
The high_pc is an address and has already been read from .debug_addr
before being passed into FuncHandler::ProcessAttributeUnsigned.
Bug:870908
Change-Id: I950098e360b5193f26bf767b8fa0a5f9d59e66ce
Reviewed-on: https://chromium-review.googlesource.com/1178760
Reviewed-by: Mark Mentovai <mark@chromium.org>
This enables the DWARF reader to properly parse DW_AT_ranges attributes
in compilation units and functions. Code covered by a function is now
represented by a vector of ranges instead of a single contiguous range
and DW_AT_ranges entries are used to populate it. All the code and tests
that assumed functions to be contiguous entities has been updated to
reflect the change. DW_AT_ranges attributes found in compilation units
are parsed but no data is generated for them as it is not currently needed.
BUG=754
Change-Id: I310391b525aaba0dd329f1e3187486f2e0c6d442
Reviewed-on: https://chromium-review.googlesource.com/1124721
Reviewed-by: Ted Mielczarek <ted.mielczarek@gmail.com>
This struct matches the layout defined by Microsoft and replaces
Breakpad's MDRawContextARM64_Old. This CL updates the processor to
understand either the old or new structs, but clients continue to write
the old structs.
Change-Id: I8dedd9ddb2ec083b802723b9ac87beb18d98edbd
Reviewed-on: https://chromium-review.googlesource.com/1155938
Reviewed-by: Mark Mentovai <mark@chromium.org>
This makes way for the addition of a struct matching Microsoft's layout
for ARM64.
Change-Id: I115f25290863e7438852691d1ec3c9324a42f7a5
Reviewed-on: https://chromium-review.googlesource.com/1152158
Reviewed-by: Mark Mentovai <mark@chromium.org>
Reviewed-by: Mike Frysinger <vapier@chromium.org>
Fixes a build break of dump_syms with Xcode.
Patch provided by Hiroyuki Komatsu.
Change-Id: I3bd3772060afee9f78dc99c75cd94f96a56c7617
Reviewed-on: https://chromium-review.googlesource.com/1144604
Reviewed-by: Joshua Peraza <jperaza@chromium.org>
Reviewed-by: Mark Mentovai <mark@chromium.org>
dump_syms produces incomplete CFI info on iOS because it doesn't support
converting compact unwind to Breakpad symbols. Attempting to use
incomplete CFI can result in infinte stack traces.
Bug: google-breakpad:764
Change-Id: Id042aa515d17928cb5503a79038607d95c56238d
Reviewed-on: https://chromium-review.googlesource.com/1128252
Reviewed-by: Ivan Penkov <ivanpe@chromium.org>
When DW_AT_MIPS_linkage_name doesn't demangle, breakpad currently throws
the symbol completely, but in some cases, there is no DW_AT_name or
DW_AT_abstract_origin to figure out a name, and the raw value from
DW_AT_MIPS_linkage_name is still better than nothing. Fall back to that
in when there is nothing else.
R=ted@mielczarek.org
Change-Id: I5cc7580244f2b99f5f1f279d09b904031cae1a37
Reviewed-on: https://chromium-review.googlesource.com/1082176
Reviewed-by: Ted Mielczarek <ted.mielczarek@gmail.com>
endl flushes output after each line. Using "\n" instead significantly improves
I/O efficiency.
Change-Id: If6a5549fc3613ca3a7c9a71838ec36c5b7a20580
Reviewed-on: https://chromium-review.googlesource.com/1077626
Reviewed-by: Ivan Penkov <ivanpe@chromium.org>
Reviewed-by: Lei Zhang <thestig@chromium.org>
For common signals: SIGILL, SIGFPE, SIGSEGV, and SIGBUS.
Change-Id: I80048f70445c3fa6accd548704c5700b3bed12a4
Reviewed-on: https://chromium-review.googlesource.com/1012589
Reviewed-by: Robert Sesek <rsesek@chromium.org>
The variables in the CL are not initialized. Even if it's safe not to
initialize them here, MSAN doesn't know that.
Bug: 394028
Change-Id: I597a7d76aa19d5789decd0f85150fa31c9655269
Reviewed-on: https://chromium-review.googlesource.com/1001573
Reviewed-by: Lei Zhang <thestig@chromium.org>
Store the information in the exception record's exception_information
field.
Change-Id: Ie215cae2f070fdab63c3d05cc1bc4fb4b7b095fa
Reviewed-on: https://chromium-review.googlesource.com/990799
Reviewed-by: Mark Mentovai <mark@chromium.org>
for being too long.
We've seen some minidumps that fail to process because they contain
a ridiculous number of modules (usually due to something leaking shm
mappings, it looks like). They're annoying to investigate because even
minidump_dump fails to load and print the module list. This patch makes
minidump_dump effectively remove the limit on the number of modules it
will load, so inspecting the dump by hand is possible.
R=vapier@chromium.org
Change-Id: I7a55387ca4aaad8664cd4d2651052da989366027
Reviewed-on: https://chromium-review.googlesource.com/957130
Reviewed-by: Mike Frysinger <vapier@chromium.org>
Noticed while adding an include guard presubmit check in
Chromium.
Change-Id: I9e677412d881b32a58d695208045b575bb8f8be6
Reviewed-on: https://chromium-review.googlesource.com/934448
Reviewed-by: Mark Mentovai <mark@chromium.org>
old_handlers is zeroish whenever an NSException is thrown. This caused PROT_WRITE to never be set and resulted in an EXC_BAD_ACCESS when trying to set the handler to NULL.
Change-Id: Ibb7da448204431c7602b1001f3a5216303c4c9d1
Reviewed-on: https://chromium-review.googlesource.com/899907
Reviewed-by: Mark Mentovai <mark@chromium.org>
The previous change to fix compiling on Android < N forgot to include a
helper in the ifdef, thus not fixing the problem. This change extends
the ifdef to include all helpers used by the test.
Change-Id: Ibb3030f54a81b5609a0b55ccef387a3cba22d088
Reviewed-on: https://chromium-review.googlesource.com/895240
Reviewed-by: Mike Frysinger <vapier@chromium.org>
The unittest for #752 made use of pthread_barrier_t, which is not
supported on Android. This change replaces the barrier code with a
simple sleep, which proved sufficient to trigger the race. It only
affects the test and does not affect the original fix for #752.
Change-Id: I82c32cf00899176fa09089e716ed85850b8711e6
Reviewed-on: https://chromium-review.googlesource.com/895168
Reviewed-by: Mike Frysinger <vapier@chromium.org>
Fixes a bug where MD_EXCEPTION_CODE_MAC_PPC_ALTIVEC_ASSIST
would unintentionally get two reason strings appended.
Bug: 177475
Change-Id: I4957268328a242c7c75bbff8add98e9a48ba83ad
Reviewed-on: https://chromium-review.googlesource.com/895705
Reviewed-by: Mike Frysinger <vapier@chromium.org>
Reviewed-by: Mark Mentovai <mark@chromium.org>
Restoring the signal handler in ExceptionHandler::SignalHandler() can
lead to a race in scenarios where multiple threads crash within a short
time. This can cause threads to alternately try to write a minidump
without ever terminating the process.
The first thread to write a minidump will reset the signal handler to
the SIG_DFL using signal() in InstallDefaultHandler(). The next thread
to execute SignalHandler() will detect this and will reset the signal
handler to SignalHandler(). If the first thread takes too long to write
its minidump (e.g. when there are many threads), the chances increase
that the second thread will enter SignalHandler() before the first one
leaves the critical section.
After resetting the signal handler, the second thread will fail to write
a minidump (since the file already exists) and will try to reset the
signal handler to the default by calling RestoreHandlersLocked().
However, in the meantime the first thread will have entered
SignalHandler() again and will overwrite it one more time.
After that, no further attempts will be made to restore the default
signal handler and both threads will continue to re-raise the signal and
attempt to write minidump files.
This change adds a check to make sure that cur_handler.sa_sigaction is
still pointing to SignalHandler() before re-installing the handler.
To test this we start a large number of sleeping threads and two threads
that will crash simultaneously. Without the fix, this would reproducibly
lead to a loop between the two crashing threads.
Bug: 752
Change-Id: I784328cfff17ddc7476d6668354570ab867ba405
Reviewed-on: https://chromium-review.googlesource.com/855137
Reviewed-by: Mike Frysinger <vapier@chromium.org>
Bug:
Change-Id: I02ceca2ff7cb87bb2b8f0cf02d31f9ab6d46a8da
src/tools/linux/tools_linux.gypi was using 'symupload/minidump_upload.m' whereas it should have been 'symupload/minidump_upload.cc' for linux. '.m' is for mac.
Reviewed-on: https://chromium-review.googlesource.com/840622
Reviewed-by: Mike Frysinger <vapier@chromium.org>
Adds asm/ and machine/ directories to src/common/android/include. This
is necessary because some required files for MIPS were removed in newer
Android NDK versions, which broke Breakpad compilation.
Bug: 771171
Change-Id: Ie6a079b6b8130b549ebc6d0bc4aef0e47e7bd6c2
Reviewed-on: https://chromium-review.googlesource.com/835282
Reviewed-by: Mark Mentovai <mark@chromium.org>
Breakpad shouldn't be hacking up headers for LSS. This was eventually
fixed in LSS directly in https://codereview.chromium.org/1248033002, so
we can drop this hack on our side.
Change-Id: Iff29efe7f6af40835e0aab1f6ac3fd8d167045ef
Reviewed-on: https://chromium-review.googlesource.com/843124
Reviewed-by: Mark Seaborn <mseaborn@chromium.org>
Reviewed-by: Mark Mentovai <mark@chromium.org>
Chrome somehow changed the memory mapping with hugepage enabled.
This makes the hack in CrOSPostProcessMappings more general.
BUG=chromium:793452
TEST=with this patch on Chromium,
minidump_dump *dmp shows the right information on chrome
Change-Id: Iff58bf1a712a6e66cbd2d813422db7549a3080a5
Reviewed-on: https://chromium-review.googlesource.com/837963
Reviewed-by: Mark Mentovai <mark@chromium.org>
The header states that if the controller is not -start:'ed that it will call
the block with a NULL BreakpadRef. As previously implemented, it asserted if
it was not started.
Change-Id: I3a329a773c0484dc1b74013717b68426758ea2cd
Reviewed-on: https://chromium-review.googlesource.com/829834
Reviewed-by: Mark Mentovai <mark@chromium.org>
Updates dump_syms to write the optional 'm' first field in FUNCTION and
PUBLIC records to indicate that the address corresponds to more than one
symbol.
Bug: google-breakpad:751
Change-Id: I850b0122324ed5f9ec747aa92ba354a3126a7ef9
Reviewed-on: https://chromium-review.googlesource.com/820711
Reviewed-by: Mark Mentovai <mark@chromium.org>
The mac exception_handler is included in a conditional below.
Change-Id: I505fad7ef6731706a39b7aaacc9a948800fc3069
Reviewed-on: https://chromium-review.googlesource.com/809306
Reviewed-by: Mark Mentovai <mark@chromium.org>
Adds an optional 'm' as the first field in FUNCTION and PUBLIC records
to indicate that the address corresponds to more than one symbol.
Controls this by a command line flag for now to give symbol file users
a chance to update.
Also reduces the number of IDiaSymbols retained in memory to one per
address. This reduces memory consumption by 8% when processing
chrome.dll.pdb.
Updates the processor to parse the new optional field.
Bug: google-breakpad:751
Change-Id: I6503edaf057312d21a1d63d9c84e5a4fa019dc46
Reviewed-on: https://chromium-review.googlesource.com/773418
Reviewed-by: Mark Mentovai <mark@chromium.org>
Consistently output the "least" symbol by decorated name when
multiple symbols share an address.
Testing with chrome.dll.pdb the diffs between the new and old output
look sensible, and this is actually ~20% faster than the existing
implementation.
Bug: 749
Change-Id: Ie638559b63f0eb2dcb80b1ebb579228d62c63bb2
Reviewed-on: https://chromium-review.googlesource.com/758885
Reviewed-by: Mark Mentovai <mark@chromium.org>
This enables repeatedly setting a value based on index, which avoids a
linear scan of the entry table after the first SetKeyValue().
Bug: chromium:598854
Change-Id: I9964670a09dcd8ff76180d031a373f20990bf4d8
Reviewed-on: https://chromium-review.googlesource.com/757579
Reviewed-by: Mark Mentovai <mark@chromium.org>
The lld linker has native support for creating packed relocation
sections, and as a result we can expect files with these sections to
have symbols.
Bug: chromium:742655
Change-Id: I48a50bff041146f51b3a8b730d7a778f832787f6
Reviewed-on: https://chromium-review.googlesource.com/754239
Reviewed-by: Mike Frysinger <vapier@chromium.org>
Reviewed-by: Mark Mentovai <mark@chromium.org>
This relands fd0a0d2b7a which was reverted
in 5dad29423e, with a fix for guarding
kMaxSuffixLength which only used in assert()s with macros which breaks
chromium.mac/ios-device.
Change-Id: I5ee21b7f290517d6e7a0ef90b693b97f92392549
Reviewed-on: https://chromium-review.googlesource.com/751922
Reviewed-by: Mark Mentovai <mark@chromium.org>
On tvOS, the app fails to shutdown after write.
Allow exit_after_write to be false for tvOS in order to force an exit() after write.
Change-Id: Ib2e1e1d03264a2972f5607b3070f4a6287aa0a98
Reviewed-on: https://chromium-review.googlesource.com/752071
Reviewed-by: Mark Mentovai <mark@chromium.org>
This silences a warning in newer versions of clang that complains
about "register" being a deprecated keyword.
Bug: chromium:780692
Change-Id: If354b9b18421e3e910849b385c44207e0ce02590
Reviewed-on: https://chromium-review.googlesource.com/750362
Reviewed-by: Mark Mentovai <mark@chromium.org>
As of Android API level 16 tgkill is declared in the NDK version of
signal.h, which conflicts with the static definition found in
src/client/linux/handler/exception_handler.cc. This change removes
the static tgkill definition and replaces its use with sys_tgkill
from the linux syscall support library.
Bug:
Change-Id: Ic70addd8a064cfa36345d86b7e36409e2089e909
Reviewed-on: https://chromium-review.googlesource.com/738912
Reviewed-by: Mike Frysinger <vapier@chromium.org>
C++ doesn't allow skipping initialization with a goto. This means that
this code is illegal:
void func(bool b) {
if(b) goto END;
int value = 0; //error C2362 with /permissive-
//... value used here
END:
return;
}
Adding an extra scope makes the code legal. This problem is only
detected with /permissive- but now that compiling with this
switch is practical we might as well stay /permissive- clean:
https://blogs.msdn.microsoft.com/vcblog/2016/11/16/permissive-switch/
Note that compiling /permissive- clean only works with the 10.0.16299.0
SDK which currently has other issues...
Bug: 773476
Change-Id: I54e64aaef46d70a817cf7da272f76d9ae5f6a6f7
Reviewed-on: https://chromium-review.googlesource.com/740287
Reviewed-by: Mark Mentovai <mark@chromium.org>
This hides the need to provide mutable C strings, and unifies
existing basename calls and variations in a single location.
Change-Id: Idfb449c47b1421f1a751efc3d7404f15f8b369ca
Reviewed-on: https://chromium-review.googlesource.com/725731
Reviewed-by: Mark Mentovai <mark@chromium.org>
Reviewed-by: Mike Frysinger <vapier@chromium.org>
If the mapping for the main executable needed to be merged (for
example, if it was linked with lld and therefore contains an r mapping
followed by an r/x mapping), we would never reach the code that makes
it the first module. Handle that situation by moving that code into
a separate loop.
This fixes an issue where breakpad_unittests fails on Android devices
when linked with lld. It appears that the glibc dynamic loader
happens to always load executables (or at least the executables that
we create) at a lower address than DSOs, so we never hit this bug on
desktop Linux.
Testing: "make check" with both gold and lld as linker. Also
breakpad_unittests when patched into Chromium on Linux (lld) and
Android (gold and lld).
Bug: chromium:469376
Change-Id: I6329e4afd2f1bf44c25a6c3e684495e21dba83a6
Reviewed-on: https://chromium-review.googlesource.com/722286
Reviewed-by: Mike Frysinger <vapier@chromium.org>
Reviewed-by: Mark Mentovai <mark@chromium.org>
memory.h shadows a system header which normally isn't a problem
because of the include paths in Breakpad, but the Firefox build
system winds up with src/common in the include path so we've had
a workaround for this for years. Renaming the file lets us get
rid of that workaround and shouldn't hurt anything.
Change-Id: I3b7c4239dc77f3b2b7cf2b572a0cad88cd7e8522
Reviewed-on: https://chromium-review.googlesource.com/723261
Reviewed-by: Mark Mentovai <mark@chromium.org>
Note that the current MicrodumpProcessor::Process implementation has a
bug due to the fact that it creates a local Microdump instance, and then
holds onto a pointer to the object returned by microdump.GetMemory()
which is destroyed when microdump goes out of scope. This CL fixes the
crash by making Microdump outlive MicrodumpProcessor, which is the same
pattern that Minidump/MinidumpProcessor uses.
Bug: google-breakpad:748
Change-Id: I554b46d309649cf404523722bd9ee39e17a10139
Reviewed-on: https://chromium-review.googlesource.com/720809
Reviewed-by: Primiano Tucci <primiano@chromium.org>
Reviewed-by: Ivan Penkov <ivanpe@chromium.org>
This exception_handler_no_mach does not use Mach for exception handling
so that clients such as tvOS and watchOS that do not support mach
messages can handle POSIX signals.
Change-Id: I4a4574e58834bc590e110e6ecd1825f8af1437a2
Reviewed-on: https://chromium-review.googlesource.com/714276
Reviewed-by: Mark Mentovai <mark@chromium.org>
When using traditional headers, sys/types.h is needed to define __u64
for sys/user.h. Previously, we thought this would be provided by
stdint.h, but it is not.
Change-Id: I0e648712f4ef1e303104a5264d3d2d0b218f5d45
Reviewed-on: https://chromium-review.googlesource.com/705267
Reviewed-by: Mark Mentovai <mark@chromium.org>
Mostly int<->size_t implicit conversions.
Warning 4366 (The result of the unary '&' operator may be unaligned)
appears in minidump.cc:907, but I don't know why. It looks aligned to me.
Change-Id: I641942adc324f8f9832b20662083dc83498688a8
Reviewed-on: https://chromium-review.googlesource.com/637390
Reviewed-by: Mike Frysinger <vapier@chromium.org>
This was lost in afa9c52715, but it turns out that it’s still
necessary.
Bug: google-breakpad:733
Change-Id: I4e0e4e4d2e80c22df1ff6b82e471905773c940a3
Reviewed-on: https://chromium-review.googlesource.com/675732
Reviewed-by: Joshua Peraza <jperaza@chromium.org>
1. testing.gyp is a gyp file, not a gypi file. It is only referenced in
“dependencies” sections. The gypi extension is used for files that are
included by an “includes” section.
2. Update paths in testing.gyp to reflect the real locations of
googletest and googlemock following their merge into a single
repository.
Change-Id: If9c356d93aa5ffda54af46fbed648baa2274dac6
Reviewed-on: https://chromium-review.googlesource.com/673404
Reviewed-by: Mike Frysinger <vapier@chromium.org>
Chrome uses API 16 for 32-bit builds and API 21 for 64-bit builds. The
NDK’s <link.h> provides r_debug and link_map structure definitions only
at API 21 and above. Breakpad used a custom <link.h> to define these
structures only during 64-bit builds, which worked for Chrome’s
purposes. However, other consumers may wish to build Breakpad at
arbitrary API levels without regard to bitness. This alters Breakpad’s
custom <link.h> to correctly check the NDK API level rather than target
CPU bitness.
Likewise for <sys/user.h> on 32-bit x86, which provided a typedef for
user_fpxregs_struct to user_fxsr_struct. API 21 and above, as well as
the unified headers at any API level, always name the structure
user_fpxregs_struct.
Definitions for 64-bit ARM’s user_regs_struct and user_fpsimd_struct
have been removed from Breakpad’s copy of <sys/user.h>. The header
claims that these fallback definitions are only necessary with NDK r10,
which should no longer be in use even by Chromium, which now uses NDK
r12b. This removes the Chromium-specific ANDROID_NDK_MAJOR_VERSION macro
from use entirely.
Fixes https://stackoverflow.com/questions/44141159/ and b/65630828.
Bug: google-breakpad:733
Change-Id: I5841906297cd15b15ce48b73fd8332fd40afc9a0
Reviewed-on: https://chromium-review.googlesource.com/665740
Reviewed-by: Primiano Tucci <primiano@chromium.org>
Reviewed-by: Joshua Peraza <jperaza@chromium.org>
The only code using gflags is google_crash_report_sender, and nothing
builds or tests that code currently. Switch it over to using system
versions of gflags so we can drop the local prebuilts. Tested local
builds by hand of the tool.
Bug: google-breakpad:360
Change-Id: I75d79b176468c948773079a54d87e70709feaf87
Reviewed-on: https://chromium-review.googlesource.com/665799
Reviewed-by: Mark Mentovai <mark@chromium.org>
Nothing appears to be using this anymore, so stop bundling it.
Bug: google-breakpad:360
Change-Id: Id95b36994379da92f8ef2a81754b3da5f1f79cae
Reviewed-on: https://chromium-review.googlesource.com/665503
Reviewed-by: Mark Mentovai <mark@chromium.org>
Breakpad’s DWARF line table reader only understood line tables at the
level of DWARF 2. This wasn’t a problem because LLVM only produced line
tables at this level, even when generating DWARF 4. But LLVM would like
to output DWARF 4 line tables when generating DWARF 4, and Breakpad
needs to understand this format. (Meanwhile, it seems that GCC has used
DWARF 4 line tables with DWARF 4 output since 4.5.0, 2010-04-14.)
DWARF 3 line tables are fully compatible with DWARF 2 (assuming that
nothing needs “prologue end,” “epilogue begin,” or “isa”, and opcodes
related to these fields are properly skipped). DWARF 4 changes the line
number program header slightly to include a “maximum operations per
instruction” field. This field must be recognized, but can safely be
ignored (and assumed to be always 1) if VLIW architectures are not
supported (they aren’t). DWARF 4 also introduces a “discriminator”,
whose opcode can also be skipped if these values are not needed (they
shouldn’t be).
This recognizes the “maximum operations per instruction” field when
processing DWARF 4 line tables, but asserts that its value is 1 and
otherwise ignores it.
This is not compatible with VLIW architectures that set this field to a
value other than 1. Such architectures are irrelevant to Breakpad, and
mainline GCC and the proposed LLVM patch always set this field to 1.
There are other things that could be extracted from DWARF 3 and 4 line
tables that aren’t currently extracted (although these are currently
irrelevant to Breakpad too).
Bug: google-breakpad:745
Change-Id: I5bf9c0b1aa654849c9cce64e60682447d10be8ba
Reviewed-on: https://chromium-review.googlesource.com/663441
Reviewed-by: Mike Frysinger <vapier@chromium.org>
This will allow us to provide the right information for webview renderer
crashes. At the moment the crash information for the browser process is
captured (from the debuggerd output) instead.
BUG=754715
Change-Id: I409546311b6e38fe1cf804097c18d7bb2a015d83
Reviewed-on: https://chromium-review.googlesource.com/612381
Reviewed-by: Robert Sesek <rsesek@chromium.org>
Change I361d8812df7b2977fe2630289059d31c3c9a4cc3 increased the maximum
number of threads for minidump_stackwalk. This change also increases the
maximum number of regions.
Change-Id: I61efd4453df8809bd9cd657546d1d6727cd10281
Reviewed-on: https://chromium-review.googlesource.com/588384
Reviewed-by: Mike Frysinger <vapier@chromium.org>
The main motivation for this change is to handle very large stack
traces, normally the result of infinite recursion. This part is
actually fairly simple, relaxing a few self-imposed limits on how
many frames we can unwind and the max size for stack memory.
Relaxing these limits requires stricter and more consistent checks for
stack unwinding. There are a number of unwinding invariants that apply
to all the platforms:
1. stack pointer (and frame pointer) must be within the stack memory
(frame pointer, if preset, must point to the right frame too)
2. unwinding must monotonically increase SP
(except for the first frame unwind, this must be a strict increase)
3. Instruction pointer (return address) must point to a valid location
4. stack pointer (and frame pointer) must be appropriately aligned
This change is focused on 2), which is enough to guarantee that the
unwinding doesn't get stuck in an infinite loop.
1) is implicitly validated part of accessing the stack memory
(explicit checks might be nice though).
4) is ABI specific and while it may be valuable in catching suspicious
frames is not in the scope of this change.
3) is also an interesting check but thanks to just-in-time compilation
it's more complex than just calling
StackWalker::InstructionAddressSeemsValid()
and we don't want to drop parts of the callstack due to an overly
conservative check.
Bug: chromium:735989
Change-Id: I9aaba77c7fd028942d77c87d51b5e6f94e136ddd
Reviewed-on: https://chromium-review.googlesource.com/563771
Reviewed-by: Mark Mentovai <mark@chromium.org>
Reviewed-by: Ivan Penkov <ivanpe@chromium.org>
1. Fixing ExceptionHandlerTest.FirstChanceHandlerRuns:
exit() is not an async-signal-safe function (http://man7.org/linux/man-pages/man7/signal-safety.7.html)
2. Fixing entry point signature in minidump_dump
Changed "const char* argv[]" to "char* argv[]" to match the standard entry point signature
3. Updating .gitignore to exclude unit test artifacts
Change-Id: I9662898d0bd97769621fb6476a720105821c60f0
Reviewed-on: https://chromium-review.googlesource.com/562356
Reviewed-by: Mark Mentovai <mark@chromium.org>
Reviewed-by: Ivan Penkov <ivanpe@chromium.org>
Reviewed-by: Joshua Peraza <jperaza@chromium.org>
When rolling this into Chrome, we got compile failures due to
DoNullPointerDereference being undefined but the new FirstChanceHandlerRuns
tests depends on this and was still defined.
The fix is to only enable the FirstChanceHandlerRuns test on non-asan builds.
Bug:
Change-Id: I5a3da0a21e2d0dd663ffc01137496d16905293a6
Reviewed-on: https://chromium-review.googlesource.com/544186
Reviewed-by: Mark Mentovai <mark@chromium.org>
This change adds the option for Breakpad hosts to register a callback
that gets the first chance to handle an exception. The handler will
return true if it handled the exception and false otherwise.
The primary use case is V8's trap-based bounds checking support for
WebAssembly.
Bug:
Change-Id: I5aa5b87d1229f1cef905a00404fa2027ee86be56
Reviewed-on: https://chromium-review.googlesource.com/509994
Reviewed-by: Mark Mentovai <mark@chromium.org>
The bfd and gold linkers create segments like this: r/x, r/w where
the r/x segment covers the start of the ELF file.
lld's segments look like this: r, r/x, r/w where the r segment covers
the start of the ELF file.
So we cannot rely on the location of the r/x to tell where the start
of the ELF is. But we can still rely on the r and r/x mappings being
adjacent. So what we do is when we see an r segment followed by an r/x,
merge the r into the r/x and claim that it is executable. This way,
the minidump writer will continue to see a single executable segment
covering the entire executable.
Testing: "make check" passes when breakpad is compiled with
lld compiled from trunk (requires bug fix from LLVM r303689).
Also patched change into chromium and tested these builds:
$ cat args.gn
is_chrome_branded = true
is_debug = false
is_official_build = true
use_lld = true
allow_posix_link_time_opt = false
is_cfi = false
$ cat args.gn
target_os = "android"
target_cpu = "arm"
is_debug = false
is_official_build = true
is_chrome_branded = true
With both builds breakpad_unittests passes and
chrome/chrome_modern_public_apk create good minidumps after navigating
to chrome://inducebrowsercrashforrealz (checked that minidump contains
stack trace entry for content::HandleDebugURL).
Bug: chromium:716484
Change-Id: Ib6ed3a8420b83acf4a5962843930fb006734cb95
Reviewed-on: https://chromium-review.googlesource.com/513610
Reviewed-by: Primiano Tucci <primiano@chromium.org>
It is legal for an ELF to contain multiple PT_NOTEs, and that is in
fact what lld's output looks like.
Testing: "make check" and breakpad_unittests when patched into
chromium.
Bug: chromium:716484
Change-Id: I01d3f8679961e2cb7e789d4007de8914c6af357d
Reviewed-on: https://chromium-review.googlesource.com/513512
Reviewed-by: Primiano Tucci <primiano@chromium.org>
Reviewed-by: Ted Mielczarek <ted@mielczarek.org>
Reviewed-by: Mark Mentovai <mark@chromium.org>
x86_64h has a different cpusubtype from x86_64. The h is for Haswell.
BUG=
Change-Id: Icf884e5699fe120c12d13aa57cd62db5b69a2ce6
Reviewed-on: https://chromium-review.googlesource.com/457171
Reviewed-by: Ted Mielczarek <ted@mielczarek.org>
The layout of Elf32_Nhdr and Elf64_Nhdr is the same, so remove
templating and code that extracts the elfclass from the ELF file.
Testing: "make check" and breakpad_unittests when patched into
chromium.
Bug: chromium:716484
Change-Id: I41442cfff48afc6ae1a5b604d22b67550a910376
Reviewed-on: https://chromium-review.googlesource.com/514450
Reviewed-by: Mike Frysinger <vapier@chromium.org>
Also adds waits for all child processes spawned in MinidumpWriterTest.
Bug: 725754
Change-Id: I3248925993dede2c113ab1989b322a9d9c8f24bd
Reviewed-on: https://chromium-review.googlesource.com/513480
Reviewed-by: Mark Mentovai <mark@chromium.org>
Change a9fca58 made use of the O_CLOEXEC flag, which is not supported on
older Linux kernels. This change makes the use contingent on kernel
support.
Testing: I manually compiled breakpad on CentOS 5.8 running kernel
2.6.18-308.8.2.el5.centos.plusxen.
Bug: 730
Change-Id: I21dff928cfba3c156a56708913f65a0c7b5396a6
Reviewed-on: https://chromium-review.googlesource.com/498528
Reviewed-by: Mike Frysinger <vapier@chromium.org>
When writing a minidump on Linux, we called clone() in
linux/handler/exception_handler.cc with the CLONE_FILES flag. If the
parent process died while the child waited for the continuation signal,
the write side of the pipe 'fdes' stayed open in the child. The child
would not receive a SIGPIPE and would wait forever.
To fix this, we clone without CLONE_FILES and then close the
read-side of fdes in the master before the ptrace call. That way, if the
master dies, the child will receive a SIGPIPE and will die, too.
To test this I added a sleep() call before SendContinueSignalToChild()
and then killed the master, manually observing that the child would die,
too.
Bug: 728
Change-Id: Ifd72de835a34e7d9852ae1a362e707fdc6c96c7e
Reviewed-on: https://chromium-review.googlesource.com/464708
Reviewed-by: Mike Frysinger <vapier@chromium.org>
Try to read the trace's registers by PTRACE_GETREGS if kernel doesn't support PTRACE_GETREGSET.
Bug:
Change-Id: I881f3a868789747ca217f22a93370c6914881f9a
Reviewed-on: https://chromium-review.googlesource.com/484479
Reviewed-by: Mike Frysinger <vapier@chromium.org>
This patch ensures that two crashes taken within the same second have
different minidump names. The random characters used in the minidump
filename are now read from /dev/urandom where possible or generated via
arc4random(). If neither is available we fall back to regular rand() but
mixing the address of an object to the current time when generating the
random seed to make it slightly less predictable.
BUG=681
Change-Id: I2e97454859ed386e199b2628d6b7e87e16481b75
Reviewed-on: https://chromium-review.googlesource.com/445784
Reviewed-by: Mike Frysinger <vapier@chromium.org>
Because we can't determine the top of userspace mappable memory
directly, we rely on the fact that the process stack is allocated at the
top of the address space (minus some randomization). Anything after that
should not count as free space.
BUG=695382
Change-Id: I68453aac9732c2bd4b87236b234518068dec6640
Reviewed-on: https://chromium-review.googlesource.com/446100
Reviewed-by: Primiano Tucci <primiano@chromium.org>
Fix some build & test failures in the previous minidump_dump code.
BUG=chromium:598947
Change-Id: Ia8fce453265167368de96747a8a92af930e78245
Reviewed-on: https://chromium-review.googlesource.com/458881
Reviewed-by: Mike Frysinger <vapier@chromium.org>
The current stack output is one line byte string which is not easy for
humans to parse. Extend the print mode to support a hexdump-like view
and switch to that by default. Now we get something like:
Stack
00000000 20 67 7b 53 94 7f 00 00 01 00 00 00 00 00 00 00 | g{S...........|
00000010 00 70 c4 44 9a 25 00 00 08 65 7a 53 94 7f 00 00 |.p.D.%...ezS...|
BUG=chromium:598947
Change-Id: I868e1cf4faa435a14c5f1c35f94a5db4a49b6a6d
Reviewed-on: https://chromium-review.googlesource.com/404008
Reviewed-by: Mark Mentovai <mark@chromium.org>
In preparation for adding more flexibility to this tool, add a
proper parser for the command line flags. This uses the style
as seen in other breakpad tools.
BUG=chromium:598947
Change-Id: I95495e6ca7093be34d0d426f98a6c22880ff24a3
Reviewed-on: https://chromium-review.googlesource.com/457019
Reviewed-by: Mark Mentovai <mark@chromium.org>
If the crashing thread doesn't reference the principal mapping we can
assume that not only is that thread uninteresting from a debugging
perspective, the whole crash is uninteresting. In that case we should
not generate a minidump at all.
BUG=703599
Change-Id: Ia25bbb8adb79d04dcaf3992c3d2474f3b9b1f796
Reviewed-on: https://chromium-review.googlesource.com/457338
Reviewed-by: Robert Sesek <rsesek@chromium.org>
This change is fixing LinuxPtraceDumperTest.SanitizeStackCopy
test case.
Change-Id: I1eb3becfd4b3660bc5529b5d2a5e35db0b6eb6e0
Reviewed-on: https://chromium-review.googlesource.com/458277
Reviewed-by: Mark Mentovai <mark@chromium.org>
If another memory region of interest (e.g. a thread stack) randomly happens
to lie immediately before the page allocated by this test, the memory
regions can be coalesced in the minidump generated. Relax this test so it
correctly handles the case where the expected 256 bytes around the IP aren't
at the start of the minidump memory region.
Alternatively, that could be avoided by reserving the page before the page
used for this test, in which case this test is degenerate with
InstructionPointerMemoryMinBound and can be removed.
BUG=
Change-Id: Ib1bfb242b2c0acaa090df68334a02ac434ad880c
Reviewed-on: https://chromium-review.googlesource.com/456702
Reviewed-by: Mike Frysinger <vapier@chromium.org>
