ThreadInfo provides a uniform interface to collect register sets or
the thread-local storage address across bitness for x86 and ARM family
architectures. Additionally, ThreadInfo.h defines context structs which
mirror those provided in sys/user.h. This allows tracing across bitness
as the structs in sys/user.h are only provided for a single target
architecture.
Bug: crashpad:30
Change-Id: I91d0d788927bdac5fb630a6ad3c6ea6d3645ef8a
Reviewed-on: https://chromium-review.googlesource.com/494075
Commit-Queue: Joshua Peraza <jperaza@chromium.org>
Reviewed-by: Mark Mentovai <mark@chromium.org>
Drop the text recommending the PolyGerrit UI, since it is now the
default Gerrit UI.
Bug: chromium:717982
Change-Id: I7041ee51670a7a18b510ed7a55045cc2eb09983e
Reviewed-on: https://chromium-review.googlesource.com/494726
Reviewed-by: Scott Graham <scottmg@chromium.org>
The lots-of-regions tests in the MemoryMap test case were very
time-consuming, particularly in debug mode. MemoryMap.MapRunningChild
took as long as 15 seconds on-device (Nexus 5X), and the best result was
in the neighborhood of 7 seconds.
The bulk of the time spent in these tests was in ExpectMappings(), which
calls MemoryMap::FindMapping() in a loop to verify each region. Each
call to FindMapping() traverses the MemoryMap (internally, currently
just a std::vector<>) from the beginning. With the need to verify 4,096
regions, a single call to ExpectMappings() had to perform over 8,000,000
checks to find the regions it needed. In turn, ExpectMappings() is
called once by the SelfLargeMapFile test, and eight times by
MapRunningChild. By reducing the number of regions to 1,024, each call
to ExpectMappings() needs to perform “only” fewer than 600,000 checks.
After this change, MemoryMap.MapRunningChild completes in about a half a
second on-device.
https://crashpad.chromium.org/bug/181 is concerned with implementing a
RangeMap to serve MemoryMap and other similar code. After that’s done,
it, it should be feasible to raise the number of regions used for these
tests again.
Bug: crashpad:30, crashpad:181
Test: crashpad_util_test MemoryMap.SelfLargeMapFile:MemoryMap.MapRunningChild
Change-Id: I8ff88dac72a63c97ac937304b578fbe3b4ebf316
Reviewed-on: https://chromium-review.googlesource.com/494128
Reviewed-by: Joshua Peraza <jperaza@chromium.org>
When the /proc/pid/maps file is not read atomically and the target
process is actively mapping memory, entries can be read multiple times
or missed entirely. This change makes MemoryMap read the whole contents
of the maps file before attempting to parse it as well as check for
duplication/overlap errors, retrying on failure. This change also adds
ptrace attachements to unit tests to reflect actual intended usage.
Bug: crashpad:30
Change-Id: Ie8549548e25c47baa418ee7439d82743f84ff41e
Reviewed-on: https://chromium-review.googlesource.com/491950
Reviewed-by: Joshua Peraza <jperaza@chromium.org>
Reviewed-by: Mark Mentovai <mark@chromium.org>
Commit-Queue: Joshua Peraza <jperaza@chromium.org>
syscall(0) results in SIGSYS on x86_64, but not 32-bit x86. Choose a
high number as a nonexistent syscall number. As of 10.12.4, the highest
known system call number is 521.
Test: crashpad_util_test Signals.Cause*
Change-Id: I82dbd210f0c90fe933898ea0d360b431b10d090e
Reviewed-on: https://chromium-review.googlesource.com/489826
Reviewed-by: Robert Sesek <rsesek@chromium.org>
I opted to leave casts to types that were definitely the same size
alone. reinterpret_cast<uintptr_t>(pointer) and
reinterpret_cast<intptr_t>(pointer) should always be safe, for example.
Casts to other integral types have been replaced with
FromPointerCast<>(), which does zero-extension or sign-extension based
on the target type.
To make it possible to use FromPointerCast<>() with some use sites that
were already using checked_cast<>(), FromPointerCast<>() now uses
check_cast<>() when converting to a narrower type.
Test: crashpad_util_test FromPointerCast*, others
Change-Id: I4a71b4aa2d87f545c75524290a702f5f3138d675
Reviewed-on: https://chromium-review.googlesource.com/489701
Reviewed-by: Scott Graham <scottmg@chromium.org>
Some of the new Linux/Android tests were failing in 32-bit code where
pointers were being casted via reinterpret_cast<>() to LinuxVMAddress,
an unsigned 64-bit type. The behavior of such casts is
implementation-defined, and in this case, sign-extension was being used
to convert the 32-bit pointers to 64 bits, resulting in very large
(unsigned) LinuxVMAddress values that could not possibly refer to proper
addresses in a 32-bit process’ address space.
The offending reinterpret_cast<>() conversions have been replaced with
the new FromPointerCast<>(), which is careful to do sign-extension when
converting to a signed type, and zero-extension when converting to an
unsigned type like LinuxVMAddress.
Bug: crashpad:30
Test: crashpad_util_test FromPointerCast*:MemoryMap.*:ProcessMemory.*
Change-Id: I6f1408dc63369a8740ecd6015d657e4407a7c271
Reviewed-on: https://chromium-review.googlesource.com/488264
Reviewed-by: Joshua Peraza <jperaza@chromium.org>
These tests:
- InitializationState.InitializationState
- InitializationStateDcheckDeathTest.Destroyed_NotUninitialized
- InitializationStateDcheckDeathTest.Destroyed_NotValid
rely on certain behavior from destroyed objects. This is undefined
behavior and we know it, but the whole point of the of
InitializationState and InitializationStateDcheck destructors is to try
to help catch other parts of the program making use of undefined
behavior.
To make it impossible for the memory that formerly hosted these objects
to be repurposed during tests after the objects are destroyed, these
tests that attempt to work with destroyed objects are changed to use
placement new, so that the lifetimes of the objects can be decoupled
from the lifetimes of the buffers.
Test: crashpad_util_test InitializationState*
Change-Id: Ie972a54116c8b90a21a502d3ba13623583dfac06
Reviewed-on: https://chromium-review.googlesource.com/486383
Reviewed-by: Joshua Peraza <jperaza@chromium.org>
aae1e3efb507 CQ config: add gerrit CQAbility verifier.
95da7665b1a3 [win-test] loosen win-driver-target-type test
eb296f67da07 [win] Add support for MS VS2017 (via Registry)
19495aa28282 Update test/no-cpp/gyptest-no-cpp.
a94b02ec68fb Disable a bunch of tests on Mac
ae76d9198630 Clean up gyptest.py
b62d04ff85e6 win,ninja: ninja generator better on windows
8dc77241251e Disable flaky test/copies/gyptest-all under msvs
e8850240a433 Fix MSVC++ 32-on-32 builds after b62d04ff85e6
ffd524cefaad win ninja/make: Always use a native compiler executable
with MSVS 2017
developing.md is updated to call out supported toolchain versions, and
to explain the CDB requirement for end_to_end_tests.py.
Change-Id: Iace68009aa22acec7303ea02a2ded755645ea96c
Reviewed-on: https://chromium-review.googlesource.com/486539
Reviewed-by: Scott Graham <scottmg@chromium.org>
Includes mini_chromium ef0ded8717340c9fe48e8e0f34f3e0e74d10a392.
1d2a024fdb1d android: Use _FILE_OFFSET_BITS after all (undo
dc3d480305b2)
ef0ded871734 win: MSVS 2017 (15)/C++ 14.1/C 19.10 compatibility
Change-Id: I5c814669a0ef8577872bddff9112ce28ec628ba3
Reviewed-on: https://chromium-review.googlesource.com/482639
Commit-Queue: Mark Mentovai <mark@chromium.org>
Reviewed-by: Scott Graham <scottmg@chromium.org>
TerminateProcess(), like most of the Windows API, is declared WINAPI,
which is __stdcall on 32-bit x86. That means that the callee,
TerminateProcess() itself, is responsible for cleaning up parameters on
the stack on return. In https://crashpad.chromium.org/bug/179, crashes
in ExceptionHandlerServer::OnNonCrashDumpEvent() were observed in ways
that make it evident that TerminateProcess() has been patched with a
__cdecl routine. The crucial difference between __stdcall and __cdecl is
that the caller is responsible for stack parameter cleanup in __cdecl.
The mismatch means that nobody cleans parameters from the stack, and the
stack pointer has an unexpected value, which in the case of the Crashpad
handler crash, results in TerminateProcess()’s second argument
erroneously being used as the lock address in the call to
ReleaseSRWLockExclusive() or LeaveCriticalSection().
As a workaround, on 32-bit x86, call through SafeTerminateProcess(), a
custom assembly routine that’s compatible with either __stdcall or
__cdecl implementations of TerminateProcess() by not trusting the value
of the stack pointer on return from that function. Instead, the stack
pointer is restored directly from the frame pointer.
Bug: crashpad:179
Test: crashpad_util_test SafeTerminateProcess.*, others
Change-Id: If9508f4eb7631020ea69ddbbe4a22eb335cdb325
Reviewed-on: https://chromium-review.googlesource.com/481180
Reviewed-by: Scott Graham <scottmg@chromium.org>
The references to RFC 2388 §3 and RFC 2047 are removed. RFC 7578 has
replaced RFC 2388, and RFC 7578 acknowledges that the area of RFC 2388
called into question by the previous comment in this code was not widely
adopted. The code does not violate RFC 7578, so the TODO is removed.
Change-Id: Ie68cba49f9fbc95a4ae3a156783a6db3b406950c
Reviewed-on: https://chromium-review.googlesource.com/481244
Reviewed-by: Robert Sesek <rsesek@chromium.org>
The "file-limit" annotation will be used to confirm the theory that
certain crashes are caused by systems at or near their file descriptor
table size limits.
The annotation records the system-wide kern.num_files and kern.maxfiles
values, and the process-specific current and maximum file descriptor
limits.
The annotation will be set on crashpad_handler startup, and will be
refreshed every time an exception is handled and every time the upload
thread processes a pending report.
It’s expected that this annotation will be removed after enough data has
been collected to confirm the theory. However, the principle is useful
enough that we may want to provide this feature more generally under
bugs 19 or 21.
Bug: crashpad:180
Change-Id: I3bb78fae60e0567bc4ac2625716e0abe0ddae08c
Reviewed-on: https://chromium-review.googlesource.com/479914
Reviewed-by: Robert Sesek <rsesek@chromium.org>
Self-monitoring revealed this CHECK was being hit in the wild:
base::debug::BreakDebugger() debugger_posix.cc:260
logging::LogMessage::~LogMessage() logging.cc:759
logging::MachLogMessage::~MachLogMessage() mach_logging.cc:45
crashpad::ExceptionHandlerServer::Run() exception_handler_server.cc:108
crashpad::HandlerMain() handler_main.cc:744
The MACH_CHECK() was:
108 MACH_CHECK(mr == MACH_MSG_SUCCESS, mr) << "MachMessageServer::Run";
Crash reports captured the full message, including the value of mr:
[0418/015158.777231:FATAL:exception_handler_server.cc(108)] Check failed: mr == MACH_MSG_SUCCESS. MachMessageServer::Run: (ipc/send) invalid destination port (0x10000003)
0x10000003 = MACH_SEND_INVALID_DEST.
This can happen when attempting to send a Mach message to a dead name.
Send (and send-once) rights become dead names when the corresponding
receive right dies. This would not normally happen for exception
requests originating in the kernel. It can happen for requests
originating from a user task: when the user task dies, the receive right
dies with it. All it takes to trigger this CHECK() in crashpad_handler
is for a Crashpad client to die (or be killed) while the handler is
processing a SimulateCrash() that the client originated.
Accept MACH_SEND_INVALID_DEST as a valid return value for
MachMessageServer::Run().
Note that MachMessageServer’s test coverage was already aware of this
behavior. MachMessageServer::Run()’s documentation is updated to reflect
it too.
Change-Id: I483c065d3c5f9a7da410ef3ad54db45ee53aa3c2
Reviewed-on: https://chromium-review.googlesource.com/479093
Commit-Queue: Mark Mentovai <mark@chromium.org>
Reviewed-by: Robert Sesek <rsesek@chromium.org>
76a67a37b1d0 adds crashpad_handler’s --monitor-self argument, which
results in a second crashpad_handler instance running out of the same
database as the initial crashpad_handler instance that it monitors. The
two handlers start at nearly the same time, and will initially be on
precisely the same schedule for periodic tasks such as scanning for new
reports to upload and pruning the database. This is an unnecessary
duplication of effort.
This adds a new --no-periodic-tasks argument to crashpad_handler. When
the first instance of crashpad_handler starts a second to monitor it, it
will use this argument, which prevents the second instance from
performing these tasks.
When --no-periodic-tasks is in effect, crashpad_handler will still be
able to upload crash reports that it knows about by virtue of having
written them itself, but it will not scan the database for other pending
reports to upload.
Bug: crashpad:143
Test: crashpad_util_test ThreadSafeVector.ThreadSafeVector
Change-Id: I7b249dd7b6d5782448d8071855818f986b98ab5a
Reviewed-on: https://chromium-review.googlesource.com/473827
Reviewed-by: Robert Sesek <rsesek@chromium.org>
The Groups: line unfortunately always had a trailing space, but Linux
4.10 takes this to a new level by including a trailing space even when
no groups are present. See commit f7a5f132b447,
linux-4.10.10/fs/proc/array.c task_state().
Bug: crashpad:30
Test: crashpad_util_test ProcessInfo.Pid1
Change-Id: If498abd929b27c7f28b69144e7c4928b1626acdb
Reviewed-on: https://chromium-review.googlesource.com/477070
Reviewed-by: Joshua Peraza <jperaza@chromium.org>
With multiple crashpad_handlers running out of the same database, it was
possible for more than one to attempt to upload the same report. Nothing
ensured that the reports remained pending between the calls to
CrashReportDatabaseMac::GetPendingReports() and
CrashReportDatabaseMac::GetReportForUploading().
The Windows equivalent did not share this bug, but it would return
kBusyError. kReportNotFound is a better code.
Test: crashpad_client_test CrashReportDatabaseTest.*
Change-Id: Ieaee7f94ca8e6f2606d000bd2ba508d3cfa2fe07
Reviewed-on: https://chromium-review.googlesource.com/473928
Reviewed-by: Robert Sesek <rsesek@chromium.org>
When GetProcessInformation() obtains SystemProcessInformation, it
resizes its buffer as directed by NtQuerySystemInformation(). Nothing of
value resides in the old buffer if a resize is attempted, so it can be
freed before attempting to allocate a resized one.
This may help crashes like go/crash/f385e94c80000000, which experience
out-of-memory while attempting to allocate a resized buffer. It also may
not help, because the required buffer size may just be too large to fit
in memory. See https://crashpad.chromium.org/bug/143#c19.
Change-Id: I63b9b8c1efda22d2fdbf05ef2b74975b92556bbd
Reviewed-on: https://chromium-review.googlesource.com/473792
Commit-Queue: Mark Mentovai <mark@chromium.org>
Reviewed-by: Scott Graham <scottmg@chromium.org>
This fixes ProcessMemory for 32-bit processes. All ProcessMemory tests
were failing on 32-bit ARM on Android like this:
[ RUN ] ProcessMemory.ReadSelf
[17345:17345:20170407,172222.579687:ERROR process_memory.cc:55] pread: Invalid argument (22)
../../../../util/linux/process_memory_test.cc:73: Failure
Value of: memory.Read(address, region_size_, result.get())
Actual: false
Expected: true
[ FAILED ] ProcessMemory.ReadSelf (5 ms)
Contemporary Linux doesn’t provide a pread() system call, it provides
pread64(), which operates on off64_t. pread() is a user-space wrapper
that accepts off_t. See Android 7.1.1
bionic/libc/bionic/legacy_32_bit_support.cpp pread().
Note that off_t is a signed type. With a 32-bit off_t, when the
“offset” parameter to pread() has its high bit set, it will be
sign-extended into the 64-bit off64_t, and when interpreted as a memory
address by virtue of being used as an offset into /proc/pid/mem, the
value will take on an incorrect meaning. In fact, the kernel will reject
it outright for its negativity. See linux-4.9.20/fs/read_write.c
[sys_]pread64().
Since ProcessMemory accepts its address parameter as a LinuxVMAddress,
which is wisely a uint64_t, it converts to off64_t properly, retaining
its original value.
Note, however, that the pread64() mechanism evidently cannot read memory
in the high half of a process’ address space even when pread64() is used
throughout. Most importantly, the (pos < 0) check in the kernel will be
tripped. Less importantly, the conversion of our unsigned LinuxVMAddress
to pread64’s signed off64_t, with the high bit set, is not defined. This
is not an immediate practical problem. With the exception of possible
shared pages mapped from kernel space (I only see this for the vsyscall
page on x86_64), Linux restricts 64-bit user process’ address space to
at least the lower half of the addressable range, with the high bit
clear. (The limit of the user address space is
linux-4.9.20/arch/x86/include/asm/processor.h TASK_SIZE_MAX =
0x7ffffffff000 for x86_64 and
linux-4.9.20/arch/arm64/include/asm/memory.h TASK_SIZE_64 =
0x1000000000000 at maximum for arm64.)
The 32-bit off_t may be a surprise, because
third_party/mini_chromium/mini_chromium/build/common.gypi sets
_FILE_OFFSET_BITS=64. Altough this macro is considered in the NDK’s
“unified headers”, in the classic NDK, this macro is never consulted.
Instead, off_t is always “long”, and pread() always gets the
compatibility shim in Bionic.
Bug: crashpad:30
Change-Id: Id00c882a3d521a46ef3fc0060d03ea0ab9493175
Reviewed-on: https://chromium-review.googlesource.com/472048
Reviewed-by: Joshua Peraza <jperaza@chromium.org>
When Yama is enabled and /proc/sys/kernel/yama/ptrace_scope is set to 1
(YAMA_SCOPE_RELATIONAL), for a child to ptrace() its parent, the parent
must first call prctl(PR_SET_PTRACER, child_pid, ...).
Bug: crashpad:30
Test: crashpad_util_test ScopedPtraceAttach.*
Change-Id: Ic85e8551259f17f372b2362887e7701b833b4cb4
Reviewed-on: https://chromium-review.googlesource.com/472006
Reviewed-by: Joshua Peraza <jperaza@chromium.org>
--monitor-self-annotations allows the Crashpad-using application to push
module-level annotations in to crashpad_handler. These annotations will
appear in any crash report written for that handler by --monitor-self.
Bug: crashpad:143
Change-Id: If47395da75a90be4f4bdce0630ce95ea93f9fcf3
Reviewed-on: https://chromium-review.googlesource.com/467746
Reviewed-by: Scott Graham <scottmg@chromium.org>
https://crbug.com/678959 added “fallback” crash reporting for
crashpad_handler on Windows, in a Chrome- and Windows-specific way. This
implements a more general self-monitor mechanism that will work on
multiple platforms and in the absence of Chrome.
When starting crashpad_handler (let’s call it the “first instance”) with
--monitor-self, it will start another crashpad_handler (the “second
instance”). The second instance monitors the first one for crashes. The
second instance will be started in mostly the same way as the first
instance, except --monitor-self will not be provided to the second
instance.
Bug: crashpad:143
Change-Id: I76f3f47d1762d8ecae1814357cb672c8b7bd5e95
Reviewed-on: https://chromium-review.googlesource.com/466267
Reviewed-by: Sigurður Ásgeirsson <siggi@chromium.org>
Reviewed-by: Scott Graham <scottmg@chromium.org>
gtest used to require (expected, actual) ordering for arguments to
EXPECT_EQ and ASSERT_EQ, and in failed test assertions would identify
each side as “expected” or “actual.” Tests in Crashpad adhered to this
traditional ordering. After a gtest change in February 2016, it is now
agnostic with respect to the order of these arguments.
This change mechanically updates all uses of these macros to (actual,
expected) by reversing them. This provides consistency with our use of
the logging CHECK_EQ and DCHECK_EQ macros, and makes for better
readability by ordinary native speakers. The rough (but working!)
conversion tool is
https://chromium-review.googlesource.com/c/466727/1/rewrite_expectassert_eq.py,
and “git cl format” cleaned up its output.
EXPECT_NE and ASSERT_NE never had a preferred ordering. gtest never made
a judgment that one side or the other needed to provide an “unexpected”
value. Consequently, some code used (unexpected, actual) while other
code used (actual, unexpected). For consistency with the new EXPECT_EQ
and ASSERT_EQ usage, as well as consistency with CHECK_NE and DCHECK_NE,
this change also updates these use sites to (actual, unexpected) where
one side can be called “unexpected” as, for example, std::string::npos
can be. Unfortunately, this portion was a manual conversion.
References:
https://github.com/google/googletest/blob/master/googletest/docs/Primer.md#binary-comparison77d6b17338https://github.com/google/googletest/pull/713
Change-Id: I978fef7c94183b8b1ef63f12f5ab4d6693626be3
Reviewed-on: https://chromium-review.googlesource.com/466727
Reviewed-by: Scott Graham <scottmg@chromium.org>
Provides Read, ReadCString, and ReadCStringSizeLimited. Does not provide
ReadMapped because Linux does not support mmap on /proc/pid/mem.
Bug: crashpad:30
Change-Id: Ia319c0107b1f138aeb8e5d0ee480c77310df7202
Reviewed-on: https://chromium-review.googlesource.com/459700
Reviewed-by: Mark Mentovai <mark@chromium.org>
Commit-Queue: Joshua Peraza <jperaza@chromium.org>
This supports the “double handler” or “double handler with low
probability” models from https://crashpad.chromium.org/bug/143.
For crashpad_handler to be become its own client, it needs access to its
own executable path to pass to CrashpadClient::StartHandler(). This was
formerly available in the test-only test::Paths::Executable(). Bring
that function’s implementation to the non-test Paths::Executable() in
util/misc, and rename test::Paths to test::TestPaths to avoid future
confusion.
test::TestPaths must still be used to access TestDataRoot(), which does
not make any sense to non-test code.
test::TestPaths::Executable() is retained for use by tests, which most
likely prefer the fatal semantics of that function. Paths::Executable()
is not fatal because for the purposes of implementing the double
handler, a failure to locate the executable path (which may happen on
some systems in deeply-nested directory hierarchies) shouldn’t cause the
initial crashpad_handler to abort, even if it does prevent a second
crashpad_handler from being started.
Bug: crashpad:143
Test: crashpad_util_test Paths.*, crashpad_test_test TestPaths.*
Change-Id: I9f75bf61839ce51e33c9f7c0d7031cebead6a156
Reviewed-on: https://chromium-review.googlesource.com/466346
Reviewed-by: Scott Graham <scottmg@chromium.org>
Commit-Queue: Mark Mentovai <mark@chromium.org>
mig-generated server dispatch routines used to not clear this field in
reply messages prepared from request messages. This oversight was
corrected in the migcom in bootstrap_cmds-96 (macOS 10.12 and Xcode
8.0). Maybe someone at Apple saw the admonishing comment that we had
left here. This comment can now be removed.
Change-Id: I73d965705a2ff5788afb59dd8ecdf4afe58ee47e
Reviewed-on: https://chromium-review.googlesource.com/465687
Reviewed-by: Robert Sesek <rsesek@chromium.org>
