Previously, crashpad_handler made its own receive right, and transferred
a corresponding send right to its client. There are two advantages to
making the receive right in the client:
- It is possible to monitor the receive right for a port-destroyed
notificaiton in the client, allowing the handler to be restarted if
it dies.
- For the future run-from-launchd mode (bug crashpad:25), the handler
will obtain its receive right from the bootstrap server instead of
making its own. Having the handler get its receive right from
different sources allows more code to be shared than if it were to
sometimes get a receive right and sometimes make a receive right and
transfer a send right.
This includes a restructuring in crashpad_client_mac.cc that will make
it easier to give it an option to restart crashpad_handler if it dies.
The handler starting logic should all behave the same as before.
BUG=crashpad:68
R=rsesek@chromium.org
Review URL: https://codereview.chromium.org/1409073013 .
The intended use is to flip the client-server relationship in
CrashpadClient so that the initial client (parent process) furnishes the
handler process with a receive right. The parent can optionally receive
a port-destroyed notification allowing it to restart the handler if it
exits prematurely.
R=rsesek@chromium.org
Review URL: https://codereview.chromium.org/1408473002 .
In https://codereview.chromium.org/1411523006, the Mach port scopers are
becoming better ScopedGenerics and are losing the type conversion
operators in the process. This is needed to adapt to that change. get()
is ugly, but being explicit about conversion isn’t a bad thing, and
these scopers will gain functionality such as Pass() as part of the
switch.
As a bonus, some would-be uses of get() to check for valid port rights
are becoming a more descriptive is_valid().
R=rsesek@chromium.org
Review URL: https://codereview.chromium.org/1405273002 .
ExceptionPorts::GetExceptionPorts() returned a
std::vector<ExceptionPorts::ExceptionHandler>, which contained send
rights to Mach ports. The interface required callers to assume ownership
of each send right contained within the vector. This was cumbersome and
error-prone, and despite the care taken in Crashpad, port right leaks
did occur:
- SimulateCrash() didn’t make any attempt to release these resources at
all.
- Neither did crashpad_util_test ExceptionPorts.HostExceptionPorts,
which also reused a vector.
This replaces the vector with the interface-compatible (as far as
necessary) ExceptionPorts::ExceptionHandlerVector, which deallocates
collected port rights on destruction or clear().
R=rsesek@chromium.org
Review URL: https://codereview.chromium.org/1381023007 .
This wraps bootstrap_check_in() in BootstrapCheckIn(), and
bootstrap_look_up() in BootstrapLookUp(). The wrappers make it more
difficult to accidentally leak a returned right. They’re easier to use,
encapsulating common error checking and logging, simplifying all call
sites.
TEST=crashpad_util_test MachExtensions.BootstrapCheckInAndLookUp
R=rsesek@chromium.org
Review URL: https://codereview.chromium.org/1383283003 .
Chrome’s relauncher process needs a way to sever ties with the
crashpad_handler instance running from the disk image in order to cause
that instance to exit so that the disk image may be unmounted. This new
function is otherwise not thought to be interesting, and its use is not
recommended.
This comes with a small refactoring to create a
SystemCrashReporterHandler() function, and a fix for a minor port leak
in CrashReportExceptionHandler::CatchMachException().
BUG=chromium:538373
R=rsesek@chromium.org
Review URL: https://codereview.chromium.org/1375573005 .
CrashReportExceptionHandler::CatchMachException() must always set a
valid new_state. Failing to do so appears to trigger corpse generation
on OS X 10.11. This is addressed by calling ExcServerCopyState().
Previously, this was not done for exceptions forwarded to the user
ReportCrash, under the apparent mistaken assumption that ReportCrash
would do it. However, ReportCrash is given copies of out-parameters like
new_state to explicitly prevent it from influencing Crashpad’s returned
state.
ExcServerSuccessfulReturnValue() must not return MACH_RCV_PORT_DIED for
an EXC_CRASH handler on OS X 10.11. This appears to trigger corpse
generation. This is addressed by always returning KERN_SUCCESS from
EXC_CRASH handlers on OS X 10.11.
This also adds generic EXC_CORPSE_NOTIFY support throughout Crashpad.
The crashpad_handler does not listen for this exception type, but it is
now possible to work with this exception type using tools like
exception_port_tool and catch_exception_tool.
BUG=crashpad:48
TEST=Crashes handled by crashpad_handler do not result in the generation
of reports in the root /Library/Logs/DiagnosticReports.
R=kerrnel@chromium.org, rsesek@chromium.org
Review URL: https://codereview.chromium.org/1305893010 .
Calling std::vector<>::operator[]() with an out-of-range index argument
is undefined behavior. In two cases, Crashpad used &v[0] in situations
where it was known that the address would not be used. These calls were
wrapped in conditions guarding against vector emptiness.
While s[0] is valid on an empty string, in two cases, Crashpad used
&s[0] as an argument to a system call that would be a no-op. These calls
were wrapped in similar conditions to avoid the system call.
The two uses of vector with undefined behavior were caught by the
following tests in crashpad_snapshot_test with
UndefinedBehaviorSanitizer:
[ RUN ] CrashpadInfoClientOptions.OneModule
/Users/mark/compilatorium/llvm.build/bin/../include/c++/v1/vector:1493:12:
runtime error: reference binding to null pointer of type
'crashpad::process_types::section'
[ OK ] CrashpadInfoClientOptions.OneModule (72 ms)
[ RUN ] ProcessSnapshotMinidump.Empty
/Users/mark/compilatorium/llvm.build/bin/../include/c++/v1/vector:1493:12:
runtime error: reference binding to null pointer of type
'MINIDUMP_DIRECTORY'
[ OK ] ProcessSnapshotMinidump.Empty (1 ms)
The Crashpad codebase was audited by searching for resize() calls and
analyzing how resized strings and vectors are used.
TEST=*
BUG=
R=rsesek@chromium.org
Review URL: https://codereview.chromium.org/1283243004 .
While not strictly asan-related, this bug was found while running tests
under asan. Evidently, strings are pooled differently in that build
configuration.
TEST=crashpad_util_test ExceptionPorts.*
R=rsesek@chromium.org
Review URL: https://codereview.chromium.org/1291573004 .
The warnings are emitted when a translation unit attempts to reference
a function whose availability is newer than the deployment target.
BUG=471823
R=mark@chromium.org
Review URL: https://codereview.chromium.org/1109273002
Patch from Erik Chen <erikchen@chromium.org>.
This adds IsExceptionNonfatalResource() and its test, and uses it in
crashpad_handler. When non-fatal resource exceptions are encountered, no
crash report is generated. crashpad_handler swallows these exceptions.
Alternatively, it could allow them to be sent to the system’s host-level
resource exception handler, normally com.apple.ReportCrash.root, which
would allow them to be processed in the same way as when Crashpad is not
in use. I’m not sure which option is better. I chose to swallow them
because there doesn’t appear to be much value in letting
com.apple.ReportCrash.root and spindump look at them.
This also moves ExcCrashRecoverOriginalException() to the new file as a
sibling of IsExceptionNonfatalResource(). This provides better
organization.
BUG=crashpad:35, chromium:474163, chromium:474326
TEST=crashpad_util_test ExceptionTypes.IsExceptionNonfatalResource
R=rsesek@chromium.org
Review URL: https://codereview.chromium.org/1066243002
ExcServerCopyState() properly sets the new_state and new_state_count
out-parameters for exception handler routines that may deal with
state-carrying exceptions.
This used to exist inline in catch_exception_tool, but that
implementation had a bug caught by the new test.
TEST=crashpad_util_test ExcServerVariants.ExcServerCopyState and others
R=rsesek@chromium.org
Review URL: https://codereview.chromium.org/1049023003
After 9e79ea1da719, it no longer makes sense for crashpad_util_test_lib
to “hide” in util/util_test.gyp. All of util/test is moved to its own
top-level directory, test, which all other test code is allowed to
depend on. test, too, is allowed to depend on all other non-test code.
In a future change, when crashpad_util_test_lib gains a dependency on
crashpad_client, it won’t look so weird for something in util (even
though it’s in util/test) to depend on something in client, because the
thing that needs to depend on client will live in test, not util.
BUG=crashpad:33
R=scottmg@chromium.org
Review URL: https://codereview.chromium.org/1051533002
origin.
This adds AuditPIDFromMachMessageTrailer() to get the process ID of a
Mach message’s sender. Exception messages are considered suspicious when
not sent by the kernel or the exception process.
TEST=crashpad_util_test
R=rsesek@chromium.org
Review URL: https://codereview.chromium.org/1001943002
optimization.
Newer versions of clang (in this case, trunk r231191) can see through
the pointless division by zero and optimize it away. This caused the
test to hang in release mode.
A 50ms timeout is added to each test to transform the hang into a
failure. The test was split into 12 tests to provide better feedback and
control.
To fix the bug, the division by zero is replaced by __builtin_trap().
TEST=crashpad_util_test ExceptionPorts.*
R=rsesek@chromium.org
Review URL: https://codereview.chromium.org/993613004
Likewise for EXPECT_DEATH_CHECK() and EXPECT_DEATH().
In the in-Chromium build configured for official builds in Release mode,
CHECK() throws away its condition string and stream parameters without
ever printing them, although it still evaluates the condition and
triggers death appropriately. {ASSERT,EXPECT}_DEATH(statement, regex)
will not work correctly for any regex that attempts to match what
CHECK() prints. In these build configurations,
{ASSERT,EXPECT}_DEATH_CHECK() use a match-all regex (""). In other build
configurations, they transparently wrap {ASSERT,EXPECT}_DEATH().
BUG=crashpad:12
R=rsesek@chromium.org, scottmg@chromium.org
Review URL: https://codereview.chromium.org/992693003
- Various "FD" to "Handle"
- Existing Multiprocess implementation moves to _posix.
- Stub implementation for _win.
At the moment, multiprocess_exec_win.cc contains implementations of both
Multiprocess methods and MultiprocessExec functions. This will need more
work in the future, but reflects the idea that all tests should be in
terms of MultiprocessExec eventually.
Currently, this works sufficiently to have util_test succeed (including
multiprocess_exec_test, and the recently ported HTTPTransport tests.)
R=mark@chromium.org
BUG=crashpad:1, crashpad:7
Review URL: https://codereview.chromium.org/880763002
Rename fd_io to file_io, and ReadFD to ReadFile, etc.
file_io.cc is the higher level versions that call the basic ReadFile/WriteFile
and then file_io_posix.cc and file_io_win.cc are the implementations of
those functions.
The Windows path is as yet untested, lacking the ability to link the test binary.
R=cpu@chromium.org, mark@chromium.org
BUG=crashpad:1
Review URL: https://codereview.chromium.org/811823003
There were many call sites that wasted a few lines on
mach_port_allocate() and sticking the result into a scoper. I was about
to add three more, so I took the opportunity to simplify things.
TEST=util_test
R=rsesek@chromium.org
Review URL: https://codereview.chromium.org/809103002
The implementations for the exc and mach_exc subsystems were nearly
identical, and were a good target for templatization. The existing
split between exc and mach_exc was a good candidate for unification
based on CompositeMachMessageServer instead of the custom unification
previously done in UniversalMachMessageServer.
TEST=util_test ExcServerVariants.*
R=rsesek@chromium.org
Review URL: https://codereview.chromium.org/766193006
MachMessageServer::Run()’s distinct |nonblocking| parameter is removed.
The information it formerly conveyed is now implied by the |timeout_ms|
parameter, which can accept two special values,
kMachMessageTimeoutNonblocking and kMachMessageTimeoutWaitIndefinitely.
TEST=client_test, snapshot_test, util_test
R=rsesek@chromium.org
Review URL: https://codereview.chromium.org/777993002
UniversalMachExcServer provided both an interface and an implementation,
contrary to the other classes in the exc_server_variants family. This
was mostly done for reasons of economy in an already-large class family.
Unfortunately, this decision meant that it was impossible for other code
to use UniversalMachExcServer, which required that CatchMachException()
be implemented, and also extend another class without violating the
style guide’s prohibition of multiple implementation inheritance. This
became a problem in a lot of test code, which extended MachMultiprocess
and UniversalMachExcServer.
UniversalMachExcServer is now given its own nested Interface class,
which is a pure interface. All users of UniversalMachExcServer are
changed from “is-a” UniversalMachExcServer to “has-a”
UniversalMachExcServer and “is-a” UniversalMachExcServer::Interface.
TEST=client_test, snapshot_test, util_test
R=rsesek@chromium.org
Review URL: https://codereview.chromium.org/775943005
This exposed a bug in the ExcClientVariants test, which was expecting
the memory used for new_state to be initialized with zeroes. In reality,
no guarantee of initialization is made. MIG “out” parameters are
strictly “out” and may contain garbage at function entry.
TEST=util_test
R=rsesek@chromium.org
Review URL: https://codereview.chromium.org/779633004
The F_SETNOSIGPIPE fcntl() command is not available on 10.6. Use
socketpair() instead of pipe(), so that the SO_NOSIGPIPE socket option
can be used.
TEST=util_test ChildPortHandshake.*
R=rsesek@chromium.org
Review URL: https://codereview.chromium.org/777573002
MachMessageWithDeadline() is a mach_msg() wrapper that deals with
deadlines instead of timeouts. It is a slight simplification of the
mach_msg() interface because the deadline parameter implies the timeout
option bits, and because the caller does not need to specify send_size
during sends as the message itself already carries this information.
TEST=util_test MachMessage.MachMessageDeadlineFromTimeout
R=rsesek@chromium.org
Review URL: https://codereview.chromium.org/773943002
A subsequent change will add MachMessageWithDeadline(), a mach_msg()
wrapper. Conceptually, it makes sense to include that function in this
file family. Since this file family now contains a mach_msg() wrapper,
it makes sense to rename it mach_message and lose the _util suffix.
R=rsesek@chromium.org
Review URL: https://codereview.chromium.org/772133004
MachMessageServer was wasteful with allocations for request and reply
messages. It allocated new memory for each request receive and for each
reply send, and if it needed to resize an allocation for a request, it
would maintain two request allocations simultaneously. The new behavior
allocates memory for a new request only if it needs a different size
than for the previous request, and it never maintains two request
allocations simultaneously. Memory for a reply is allocated once per
method invocation and maintained, since this never needs to be resized.
One pass of the loop is now guaranteed, even if a caller specifies a
very small timeout that expires before attempting to receive a message.
An infinite looping bug that could occur when ignoring large messages
has also been fixed.
TEST=util_test MachMessageServer.*
R=rsesek@chromium.org
Review URL: https://codereview.chromium.org/759023004
The buffer sizing logic was correct to start with. I don’t know why I
misread it. It should say “if this would resize to receive a large
message, use the entire allocation rounded up to full page size,
otherwise, only use the space expected for a message.”
TEST=util_test
R=rsesek@chromium.org
Review URL: https://codereview.chromium.org/760573003
These port dispositions were naïvely taken from excUser.c and
mach_excUser.c, but the local and remote portions were not swapped as
they would be upon receipt in a server. This swaps them to match how
they’d be visible in a server, and uses the port disposition name
aliases expected to be used in servers: MACH_MSG_TYPE_PORT_* instead of
MACH_MSG_TYPE_{MAKE,COPY,MOVE}_*.
TEST=util_test ExcServerVariants.*
R=rsesek@chromium.org
Review URL: https://codereview.chromium.org/755323002
The existing implementation used the same logic as is found in
mach_msg_server(), but that logic seems incorrect. When the caller wants
to retry a mach_msg() receive of a too-large message that returns
MACH_RCV_TOO_LARGE, there’s no harm in attempting the receive with a
larger buffer initially. On the other hand, if the caller does not want
to retry such mach_msg() receive attempts, it’s an indication that the
caller is expecting to be intolerant of too-large messages, and there’s
no need to attempt the receive with a buffer any larger than requested.
TEST=util_test
R=rsesek@chromium.org
Review URL: https://codereview.chromium.org/753363003
As documented, MACH_SEND_TRAILER would allow a sender to provide its own
message trailer instead of having the kernel append its own
kernel-generated trailer. This is a Mach feature that supports a network
of multiple Mach hosts, but even in that environment, the option is
restricted to use by privileged callers. In reality, MACH_SEND_TRAILER
has never been implemented in OS X.
The system’s mach_msg_server() family does consider the value of
MACH_SEND_TRAILER, but this is pointless. Any purported trailer set by a
server function would be ignored.
Maintaining this code gives the illusion that it’s functional, so it’s
being removed.
TEST=util_test MachMessageServer.*
R=rsesek@chromium.org
Review URL: https://codereview.chromium.org/736493007
ChildPortHandshake is the most generic system yet to allow child
processes to provide their parents with Mach rights. These are
ordinarily expected to be send rights to the children’s own task ports,
or send rights to servers that the children hold receive rights to.
This updates DEPS to pull mini_chromium 1d3523dbda93, which includes
base::mac::ScopedMachPortSet.
TEST=util_test ChildPortHandshake.*
R=rsesek@chromium.org
Review URL: https://codereview.chromium.org/756603003
Previously, MachMessageServer::Run() only provided two strategies for
dealing with large messages, indicated by mach_msg() returning
MACH_RCV_TOO_LARGE: the receive buffer could be reallocated and the
message received, or the entire function could return MACH_RCV_TOO_LARGE
to the caller. There are situations where an intermediate behavior might
be desirable. This intermediate behavior would allow the function to
continue waiting for another message without returning an error to the
caller or attempting to receive the large message. This is desirable
when dealing with fixed-sized messages and a receiver that might be sent
messages by unknown, possibly-malicious callers. This can happen when
the corresponding send right is published with the bootstrap server, for
example.
Existing users continue to request their existing behavior, typically
receiving an error when encountering a large message.
catch_exception_tool will use the new “ignore” behavior when running in
persistent mode.
TEST=util_test MachMessageServer.*
R=rsesek@chromium.org
Review URL: https://codereview.chromium.org/756803002
the child_port subsystem.
Common routines shared with the ExcServer family of classes have been
moved to a new file, where they can be shared between different
MachMessageServer::Interface implementations.
TEST=util_test ChildPortServer.*:MachMessageUtil.*
R=rsesek@chromium.org
Review URL: https://codereview.chromium.org/754123002
