The current code is relying on info->si_pid to figure out whether
the exception handler was triggered by a signal coming from the kernel
(that will re-trigger until the cause that triggered the signal has
been cleared) or from user-space e.g., kill -SIGNAL pid, which will NOT
automatically re-trigger in the next signal handler in the chain.
While the intentions are good (manually re-triggering user-space
signals), the current implementation mistakenly looks at the si_pid
field in siginfo_t, assuming that it is coming from the kernel if
si_pid == 0.
This is wrong. siginfo_t, in fact, is a union and si_pid is meaningful
only for userspace signals. For signals originated by the kernel,
instead, si_pid overlaps with si_addr (the faulting address).
As a matter of facts, the current implementation is mistakenly
re-triggering the signal using tgkill for most of the kernel-space
signals (unless the fault address is exactly 0x0).
This is not completelly correct for the case of SIGSEGV/SIGBUS. The
next handler in the chain will stil see the signal, but the |siginfo|
and the |context| arguments of the handler will be meaningless
(retriggering a signal with tgkill doesn't preserve them).
Therefore, if the next handler in the chain expects those arguments
to be set, it will fail.
Concretelly, this is causing problems to WebView. In some rare
circumstances, the next handler in the chain is a user-space runtime
which does SIGSEGV handling to implement speculative null pointer
managed exceptions (see as an example
http://www.mono-project.com/docs/advanced/runtime/docs/exception-handling/)
The fix herein proposed consists in using the si_code (see SI_FROMUSER
macros) to determine whether a signal is coming form the kernel
(and therefore just re-establish the next signal handler) or from
userspace (and use the tgkill logic).
Repro case:
This issue is visible in Chrome for Android with this simple repro case:
- Add a non-null pointer dereference in the codebase:
*((volatile int*)0xbeef) = 42
Without this change: the next handler (the libc trap) prints:
F/libc ( 595): Fatal signal 11 (SIGSEGV), code 1, fault addr 0x487
where 0x487 is actually the PID of the process (which is wrong).
With this change: the next handler prints:
F/libc ( 595): Fatal signal 11 (SIGSEGV), code 1, fault addr 0xbeef
which is the correct answer.
BUG=chromium:481937
R=mark@chromium.org
Review URL: https://breakpad.appspot.com/6844002.
git-svn-id: http://google-breakpad.googlecode.com/svn/trunk@1461 4c0a9323-5329-0410-9bdc-e9ce6186880e
Shared libraries containing Android packed relocations have a load
bias that differs from the start address in /proc/$$/maps. Current
breakpad assumes that the load bias and mapping start address are
the same.
Fixed by changing the client to detect the presence of Android packed
relocations in the address space of a loaded library, and adjusting the
stored mapping start address of any that are packed so that it contains
the linker's load bias.
For this to work properly, it is important that the non-packed library
is symbolized for breakpad. Either packed or non-packed libraries may
be run on the device; the client detects which has been loaded by the
linker.
BUG=499747
R=primiano@chromium.org, rmcilroy@chromium.org
Review URL: https://codereview.chromium.org/1189823002.
Patch from Simon Baldwin <simonb@chromium.org>.
git-svn-id: http://google-breakpad.googlecode.com/svn/trunk@1459 4c0a9323-5329-0410-9bdc-e9ce6186880e
This is to add build fingerprint and product name/version to
microdumps. Conversely to what happens in the case of minidumps
with MIME fields, due to the nature of minidumps, extra metadata
cannot be reliably injected after the dump is completed.
This CL adds the plumbing to inject two optional fields plus the
corresponding tests.
BUG=chromium:410294
R=thestig@chromium.org
Review URL: https://codereview.chromium.org/1125153008
git-svn-id: http://google-breakpad.googlecode.com/svn/trunk@1456 4c0a9323-5329-0410-9bdc-e9ce6186880e
The current code is relying on info->si_pid to figure out whether
the exception handler was triggered by a signal coming from the kernel
(that will re-trigger until the cause that triggered the signal has
been cleared) or from user-space e.g., kill -SIGNAL pid, which will NOT
automatically re-trigger in the next signal handler in the chain.
While the intentions are good (manually re-triggering user-space
signals), the current implementation mistakenly looks at the si_pid
field in siginfo_t, assuming that it is coming from the kernel if
si_pid == 0.
This is wrong. siginfo_t, in fact, is a union and si_pid is meaningful
only for userspace signals. For signals originated by the kernel,
instead, si_pid overlaps with si_addr (the faulting address).
As a matter of facts, the current implementation is mistakenly
re-triggering the signal using tgkill for most of the kernel-space
signals (unless the fault address is exactly 0x0).
This is not completelly correct for the case of SIGSEGV/SIGBUS. The
next handler in the chain will stil see the signal, but the |siginfo|
and the |context| arguments of the handler will be meaningless
(retriggering a signal with tgkill doesn't preserve them).
Therefore, if the next handler in the chain expects those arguments
to be set, it will fail.
Concretelly, this is causing problems to WebView. In some rare
circumstances, the next handler in the chain is a user-space runtime
which does SIGSEGV handling to implement speculative null pointer
managed exceptions (see as an example
http://www.mono-project.com/docs/advanced/runtime/docs/exception-handling/)
The fix herein proposed consists in using the si_code (see SI_FROMUSER
macros) to determine whether a signal is coming form the kernel
(and therefore just re-establish the next signal handler) or from
userspace (and use the tgkill logic).
Repro case:
This issue is visible in Chrome for Android with this simple repro case:
- Add a non-null pointer dereference in the codebase:
*((volatile int*)0xbeef) = 42
Without this change: the next handler (the libc trap) prints:
F/libc ( 595): Fatal signal 11 (SIGSEGV), code 1, fault addr 0x487
where 0x487 is actually the PID of the process (which is wrong).
With this change: the next handler prints:
F/libc ( 595): Fatal signal 11 (SIGSEGV), code 1, fault addr 0xbeef
which is the correct answer.
BUG=chromium:481937
R=mark@chromium.org
Review URL: https://breakpad.appspot.com/6844002
git-svn-id: http://google-breakpad.googlecode.com/svn/trunk@1454 4c0a9323-5329-0410-9bdc-e9ce6186880e
Despite the fact that many places imply that sigaction and rt_sigaction
are essentially the same, rt_sigaction's signature is actually
different-- it takes the size of the kernel's sigset_t as an extra argument.
BUG=473973
git-svn-id: http://google-breakpad.googlecode.com/svn/trunk@1447 4c0a9323-5329-0410-9bdc-e9ce6186880e
The current processor implementation is grepping for /google-breakpad(
in the logcat lines, to filter out microdump lines, which by default
look like this:
W/google-breakpad( 3728): -----BEGIN BREAKPAD MICRODUMP-----
Turns out that logcat format can vary, when passing optional arguments,
and produce something like the following:
04-13 12:30:35.563 6531 6531 W google-breakpad: -----BEGIN ...
In the latter case, the "/google-breakpad(" filter is too aggressive.
This change is relaxing it, so it is compatible also with non-default
logcat arguments.
BUG=640
R=mmandlis@chromium.org
Review URL: https://breakpad.appspot.com/2864002
git-svn-id: http://google-breakpad.googlecode.com/svn/trunk@1442 4c0a9323-5329-0410-9bdc-e9ce6186880e
ACCESS_VIOLATION and IN_PAGE_ERROR both specify
read/write/dep flags and address. ACCESS_VIOLATION currently
reports these, but IN_PAGE_ERROR does not. This change makes
IN_PAGE_ERROR report this information as well, and also the
additional NTSTATUS value for the underlying cause.
Patch by bungeman@chromium.org
Review URL: https://breakpad.appspot.com/1794002/
git-svn-id: http://google-breakpad.googlecode.com/svn/trunk@1441 4c0a9323-5329-0410-9bdc-e9ce6186880e
On Android L+, signal and sigaction symbols are provided by libsigchain
that override the system's versions. There is a bug in these functions
where they essentially ignore requests to install SIG_DFL.
Workaround this issue by explicitly performing a syscall to
__NR_rt_sigaction to install SIG_DFL on Android.
BUG=473973
Patch by Chris Hopman <cjhopman@chromium.org>
Review URL: https://breakpad.appspot.com/1804002/
git-svn-id: http://google-breakpad.googlecode.com/svn/trunk@1438 4c0a9323-5329-0410-9bdc-e9ce6186880e
This is a reland of the previous CL (r1433). r1433 did not achieve what
intended and failed the x86_64 build of Chrome with NDK r10c.
The workaround logic in this CL is identical to r1433, but the #define
magic is applied in a more appropriate proper place this time. Turns
out Breakpad already has an Android compatibility layer, which is
common/android/include. Piggybacking the fix there.
BUG=breakpad:642
R=fdegans@chromium.org, rmcilroy@chromium.org
Review URL: https://breakpad.appspot.com/3794002
git-svn-id: http://google-breakpad.googlecode.com/svn/trunk@1434 4c0a9323-5329-0410-9bdc-e9ce6186880e
r1397 did introduce a workaround to deal with a typo in sys/user.h
in the Android NDK. The typo has been fixed in [1]. However, breakpad
cannot just switch to the fixed version as this would require atomic
rolls of Breakpad and NDK in chromium, which would make reverts hard
to handle.
This change introduces an inelegant yet functional hack which makes
breakpad compatible with both versions of the NDK, with and without
the typo. It can be reverted once Chrome has stably rolled to NDK
r10d.
[1] https://android.googlesource.com/platform/bionic/+/f485547b
BUG=breakpad:642
R=fdegans@chromium.org, rmcilroy@chromium.org
Review URL: https://breakpad.appspot.com/7814002
git-svn-id: http://google-breakpad.googlecode.com/svn/trunk@1433 4c0a9323-5329-0410-9bdc-e9ce6186880e
A recent change in the client-side microdump write (r1404) ended up
introducing a call to new() to instantiate the line buffer that
microdump uses to dump its lines. new/malloc is a luxury we cannot
afford in a compromised context.
This change switches the line buffer to be backed by the dumper
page allocator, which on Linux/Android ends up requesting pages
via mmap.
Also, the microdump write bails out without crashing if the page
allocator failed (crash during severe OOM).
BUG=640
git-svn-id: http://google-breakpad.googlecode.com/svn/trunk@1432 4c0a9323-5329-0410-9bdc-e9ce6186880e
This feature is enabled only when "-s" is provided as a commandline option.
minidump_stackwalk.cc:
- Add a new commandline option "-s" to output stack contents.
- Instantiate Minidump object in PrintMinidumpProcess() to keep it alive longer so that accessing process_state.thread_memory_regions() in stackwalk_common.cc doesn't result in use-after-free.
stackwalk_common.cc:
- Add a new function PrintStackContents() to output stack contents.
R=mark@chromium.org
Review URL: https://breakpad.appspot.com/9774002
git-svn-id: http://google-breakpad.googlecode.com/svn/trunk@1429 4c0a9323-5329-0410-9bdc-e9ce6186880e
A computation in the stack unwind algorithm could cause an overflow if a base
pointer read from crashed process is sufficiently close to top of address space.
This causes a memory read that causes the dump thread to crash, resulting in a
failure to generate crash dump. Check fixed to properly detect that this pointer
is greater than actual memory range of current stack.
Patch by Kyle Joswiak <kjoswiak@chromium.org>
Review URL: https://breakpad.appspot.com/3754003/
git-svn-id: http://google-breakpad.googlecode.com/svn/trunk@1425 4c0a9323-5329-0410-9bdc-e9ce6186880e
When LinuxCoreDumper fails to copy process data from a core file, it
fills the return buffer with a repeated sequence of a special marker.
However, MinidumpWriter doesn't know about that and may incorrectly
interpret the data. In many cases, MinidumpWriter simply copies the
gibberish data to the minidump, which isn't too bad. However, the
gibberish data may cause MinidumpWriter to behave badly in some other
cases. For example, when MinidumpWriter tries to iterate through the
linked list of all loaded DSOs via the r_map field of a r_debug struct,
if the linked list is filed with the special marker, the code keeps
iterating through the same address.
This CL addresses the issue by having LinuxCoreDumper::CopyFromProcess()
returns a Boolean value to indicate if the expected data is found from
the core file. MinidumpWriter can then decide how to handle that.
BUG=chromium:453484
TEST=Run core2md with the test data attached to chromium:453484.
R=mark@chromium.org
Review URL: https://breakpad.appspot.com/4724002
git-svn-id: http://google-breakpad.googlecode.com/svn/trunk@1420 4c0a9323-5329-0410-9bdc-e9ce6186880e
problems.
- The ordering of keys in stl containers cannot change. Make the relevant
members const to guarantee this assumption.
- Add handling and logging for demangle errors.
- Fix a potential double-delete bug if a function passed to AddFunction() is
already present.
BUG=chromium:449214
R=mark@chromium.org
Review URL: https://breakpad.appspot.com/10704002/
git-svn-id: http://google-breakpad.googlecode.com/svn/trunk@1415 4c0a9323-5329-0410-9bdc-e9ce6186880e
gcc has a single exception setting for all languages. Saying -fno-exceptions
in gcc disables exceptions and cleanups for cc files, but has no effect for mm
files.
In clang, -fno-exceptions only disables c++ exceptions, but keeps objective-c
exceptions and cleanups enabled.
http://llvm.org/viewvc/llvm-project?view=revision&revision=220714 changed
__EXCEPTIONS to be defined for clang when cleanups are enabled, independent of
if c++ exceptions are enabled. (This was necessary to have clang work with
glibc which looks at __EXCEPTIONS to decide if cleanups are enabled.)
Breakpad tried to use __EXCEPTIONS to figure out if c++ exceptions are enabled.
In cc files, this works: -fno-exceptions will disable c++ exceptions and
cleanups. But in mm files, -fno-exceptions will disable c++ exceptions and
objective-c exceptions will still be enabled, and so cleanups must run and hence
__EXCEPTIONS is defined.
To make things work with both old and new compilers, do the try/catch hack in
mm files either if __EXCEPTIONS is not defined (for old compilers) or if the
compiler is clang and __has_feature(cxx_exceptions) isn't set (which will work
for new clangs too, and which cleanly maps to if c++ exceptions are enabled).
Patch by Nico Weber <thakis@chromium.org>
Review URL: https://breakpad.appspot.com/1774002/
git-svn-id: http://google-breakpad.googlecode.com/svn/trunk@1409 4c0a9323-5329-0410-9bdc-e9ce6186880e
Previously, dump_syms did not handle DW_FORM_ref_addr if it appeared
in DWARF 4 debugging info.
Also fix a DW_FORM_ref_addr case so that it doesn't fall through to
the next switch case when assertions are disabled and the DWARF
version isn't recognised.
The following steps will reproduce the problem when using LLVM 3.4:
cat <<END >example1.c
int main() { return 0; }
END
cat <<END >example2.c
void foo(int x) {}
END
clang -emit-llvm -g -c example1.c -o example1.bc
clang -emit-llvm -g -c example2.c -o example2.bc
llvm-link-3.4 example1.bc example2.bc -o combined.bc
clang combined.bc -o executable
./google-breakpad/build/src/tools/linux/dump_syms/dump_syms executable
When using LLVM bitcode linking in this way, LLVM's backend generates
partially-merged DWARF debugging info in which some of the references
to the "int" type go via "DW_FORM_ref_addr". Since PNaCl uses LLVM
bitcode linking, this dump_syms failure occurs with nexes produced by
the PNaCl toolchain.
BUG= https://code.google.com/p/chromium/issues/detail?id=416368
TEST= see above
R=mark@chromium.org, mcgrathr@chromium.org
Review URL: https://breakpad.appspot.com/5744002
git-svn-id: http://google-breakpad.googlecode.com/svn/trunk@1408 4c0a9323-5329-0410-9bdc-e9ce6186880e
- Filter modules by prot flags (only +x) not extensions. It wouldn't
otherwise catch the case of Chrome mapping the library from the
apk (which is mapped r-x but doesn't end in .so).
- Use compile-time detection of target arch, in order to cope with
multilib OSes, where uname() doesn't reflect the run-time arch.
- Add OS information and CPU arch / count.
- Add support for aarch64.
- Add tests and stackwalk expectations for aarch64.
- Fix a potential overflow bug in the processor.
- Rebaseline the tests using smaller symbols.
- Fix microdump_writer_unittest.cc on 32-bit host.
BUG=chromium:410294
git-svn-id: http://google-breakpad.googlecode.com/svn/trunk@1407 4c0a9323-5329-0410-9bdc-e9ce6186880e
and updating minidump_stackwalk to show process uptime.
I tested this with a minidump from Chrome and I got a result that
is inline with what the Windows debugger is showing for that dump:
minidump_stackwalk output:
--------------------------
Process uptime: 601 seconds
WinDBG output:
--------------
Process Uptime: 0 days 0:10:01.000
I didn't update the machine readable output of minidump_stackwalk
on purpose in order to avoid breaking someone that uses it.
It can be added later to the machine output if needed.
R=mark@chromium.org
Review URL: https://breakpad.appspot.com/7754002
git-svn-id: http://google-breakpad.googlecode.com/svn/trunk@1406 4c0a9323-5329-0410-9bdc-e9ce6186880e
This introduces the microdump_stackwalk binary which takes advantage
of the MicrodumpProcessor to symbolize microdumps.
Its operation is identical to the one of minidump_stackwalk.
This CL, in fact, is also refactoring most of the common bits into
stackwalk_common.
BUG=chromium:410294
R=mmandlis@chromium.org
Review URL: https://breakpad.appspot.com/4704002
git-svn-id: http://google-breakpad.googlecode.com/svn/trunk@1405 4c0a9323-5329-0410-9bdc-e9ce6186880e
