This adds the missing assembler for riscv64, even though I don't have a
working tombstoned yet to test it with. There's a distinct possibility
we'll be back to fix the test (because although "register 1" is harmless
for the other architectures, it's the ra register on riscv64; the default
link register), but at least this lets us build the test.
I've also simplified all the assembly to be the simplest sequence I
know that writes 0 to address 0 (because if there was a reason to use
so many instructions before, I want to know what it is so I can write
the missing comment!).
Test: treehugger
Change-Id: I10d117eaedf361d9759a450e0973d07c4f97090e
r29 is the stack pointer on mips, but it's x2 on riscv64 (and the git
history shows that this was indeed copy & pasted from the mips code)
and since bionic always sets up a signal stack with sigaltstack() I
doubt the comment was relevant even on mips (but no-one ever used it,
so who'd know?).
While I'm here, stop using decimal arithmetic --- the whole point was to
have each register contain the value that was obviously appropriate for
that register. (riscv64's mips-like mess of registers all over the place
means that's not going to be super readable, but there's no reason to
make it worse.)
Also, even though I personally prefer the 0xdead from the old mips code,
everyone else is using 0xa5a5, so let's make riscv64 match the others.
Test: treehugger
Change-Id: Ibbae821bc0a02e07164147d621e342224528c2c9
This changes the crash export mechanism in Microdroid. For this, we
create module tombstone_handler which exports methods very similar to
tombstoned.h
For Microdroid (detected using prop: ro.hardware): It calls newly
introduces microdroid specific methods to connect/notify completion of
crash.
Individual methods:
connect -> For Android, it would connect to
tombstoned which would send it the fd corresponding to newly created
file on /data/tombstone_ . For Microdroid, we connect to tombstone
server on host via vsock & populate these sockets as the output fd.
crash_dump, in the later case, would directly write on the socket(s).
notify_completion: For Microdroid, it would simply shutdown the vsock
connections.
Note when OS is not Microdroid: It calls corresponding methods of
tombstoned_client, essentially serving as a proxy.
Detailed design: go/vm-tombstone
Test: atest MicrodroidHostTests#testTombstonesAreGeneratedUponUserspaceCrash
Bug: 243494912
Change-Id: I68537b967f2ee48c1647f0f923aa79e8bcc66942
When this was translated to riscv64, someone "fixed" the crashing bugs
that were the whole point of these two functions. Fix them back so they
actually crash, and add the CFI directives.
Test: treehugger
Change-Id: I312c51fa4c893d27b0f4e39383521657a5870a0d
See aosp/2485756 for more details, but this patch introduces a new
protocol between debuggerd and ActivityManager. This new protocol allows
ActivityManager to correctly handle recoverable crashes.
Bug: 270720163
Test: atest CtsGwpAsanTestCases
Change-Id: Icac6262d608dd57a5daf51699064ab28b0c4703f
They're the same script right now, but gdbclient.py is a bit misleading,
even if we're not likely to ever actually remove it.
Test: treehugger
Change-Id: Ic514f98bf13b3e699be4dbad2bafef22d41d9ffd
1. Fixes this test under clang coverage, which is run under presubmit
for TEST_MAPPING files. When we spawn under a minijail, and the
process exited normally (which is the case for recoverable), clang
coverage would use atexit handlers to dump some stuff using banned
prctl's and other syscalls. Instead of allow-listing them all which
sounds like a huge pain, call _exit() which skips those handlers.
2. Extends the invariant testing to make sure that recoverable GWP-ASan
recovers both the first time, and a second time in a different slot.
Bug: N/A
Test: CLANG_COVERAGE=true NATIVE_COVERAGE_PATHS="*" atest debuggerd_test
Change-Id: I6059e21db4c2898b1c9777a00d2a54497d80ef79
Currently, debuggerd tells the teacher that an app that received a fatal
signal. On the playground, dobbing on a process that doesn't actually
need to be killed is considered a friendship-ending move.
Because recoverable GWP-ASan is *supposed* to not crash your app,
suppress this behaviour and don't let ActivityManager know about the
crash.
Bug: N/A
Test: Run a use-after-free in an app that's using recoverable GWP-ASan,
through the 'libc.debug.gwp_asan.recoverable.<app_name>=1' and
'libc.debug.gwp_asan.process_sampling.<app_name>=1' sysprops.
Change-Id: I033ea67d577573df10936e37db7302d4f4bc0069
Recoverable GWP-ASan is a mode landed upstream in
https://reviews.llvm.org/D140173. For more information about why/what it
is, see
https://android-review.git.corp.google.com/c/platform/bionic/+/2394588.
This patch makes debuggerd call the required libc callbacks for GWP-ASan
to recover from the memory corruption. It also adds the functionality
that libart/sigchain eventually ends up calling, which dumps a GWP-ASan
report for the first error encountered.
Test: Build the platform, run sanitizer-status in recoverable mode,
asserting that it doesn't crash but we get a debuggerd report.
Bug: 247012630
Change-Id: I27212f7250844c20a8fd1e961417cdb4e5bd3626
When moving to a proto tombstone, backtraces no longer contain
an offset when a frame is in a shared library from an apk.
Add the offset display again if needed, and add a test to
verify this behavior.
Bug: 267341682
Test: All unit tests pass.
Test: Dumped a process running through an apk to verify the offset
Test: is present.
Change-Id: Ib720ccb5bfcc8531d1e407f3d01817e8a0b9128c
An early return out of this function makes it harder to add new prints
after the memory maps.
Test: m, flash, look at tombstone
Change-Id: Id06e432918d69ac3307761b244473b6b7ab769e8
GWP-ASan changed one of the APIs upstream to now take the fault address
as well. This is to support the recoverable mode.
Add the fault address as well.
Test: gwp_asan_unittest
Bug: N/A
Change-Id: I8a4edd3fad159d91cc036050d330bbb8f9c8d435
This is a no-op but will be used in upcoming scudo changes that allow to
change the buffer size at process startup time, and as such we will no
longer be able to call __scudo_get_ring_buffer_size in debuggerd.
Bug: 263287052
Change-Id: I350421d1fcdf22ce3b8b73780b88c1e10fa8a074
The current logging...
```
F libc : Fatal signal 31 (SIGSYS), code 1 (SYS_SECCOMP) in tid 6640 (logcat), pid 6640 (logcat)
```
...isn't super useful if crash_dump then fails, because you have no idea
what syscall caused the problem.
We already include the fault address in this line for relevant cases,
so include the syscall number in this case.
Bug: http://b/262391724
Test: treehugger
Change-Id: I45ad7d99c9904bab32b65efeb19be232e59ab3a4
* Intentional crash test code with null/free/escape warnings.
Test: make tidy-system-core-debuggerd_subset
Change-Id: Ib1255c17a374729c82aa246c6a59156dbc4e1b77
The long term plan is to completely remove tombstoned from microdroid (b/243494912), however it might take time some time to implement it.
In the meantime, we've recently removed cgroups support from the microdroid kernel. This means that starting a tombstoned results in a bunch of non-fatal errors in the logs that are related to the fact that tombstoned service specifies task_profiles.
To get rid of these error messages we temporary add a microdroid variant of the tombstoned (tombstoned.microdroid) that doesn't specify task_profiles.
Bug: 239367015
Test: microdroid presubmit
Change-Id: Ia7d37ede2276790008702e48fdfaf37f4c1fd251
Fixes
```
out/soong/installs-aosp_riscv64.mk:56833: error: overriding commands for target `out/target/product/generic_riscv64/system/etc/seccomp_policy/crash_dump.riscv64.policy', previously defined at out/soong/installs-aosp_riscv64.mk:56829
```
Test: m
Change-Id: I78a1c6b10dac2da704515f33b492ff37cc086dd6
The use of __builtin_abort in CrasherTest::Trap breaks with
-ftrap-function=abort, because then the argument of Trap is no longer in
the first argument register at the time of crash.
This flag is added when *any* sanitizer is enabled on the target, even harmless
ones like memtag-heap. See sanitize.go:769.
Fix CrasherTest::Trap to be a little more reliable.
Test: debuggerd_test with SANITIZE_TARGET=memtag_heap
Change-Id: I150f1c0355bd6f2bfabfa5a7bba125acdde1120e
Signed-off-by: Liu Cunyuan <liucunyuan.lcy@linux.alibaba.com>
Signed-off-by: Mao Han <han_mao@linux.alibaba.com>
Change-Id: Ie22c2895fc30fab68eddc18713c80e403f44b203
Signed-off-by: Chen Guoyin <chenguoyin.cgy@linux.alibaba.com>
Signed-off-by: Mao Han <han_mao@linux.alibaba.com>
Change-Id: Ie58bd7cf5dde792d8fba78602b5f53471752ab24
Signed-off-by: Xia Lifang <lifang_xia@linux.alibaba.com>
Signed-off-by: Mao Han <han_mao@linux.alibaba.com>
Change-Id: I521c6da61cf2f6f67a73febf368068c430d94cdb
This uses an std::string, which causes a heap allocation, which is not
async-safe.
Test: atest --no-bazel-mode permissive_mte_test
Change-Id: I4bd53d42d9a6a659abe62a964f14c81d9ec059d0
Unify all our "noinline" variants to the current most common one, not
least because the new [[noinline]] syntax is fussier about where it goes.
Test: treehugger
Change-Id: Icfcb75c9d687f0f05c19f66ee778fd8962519436
There's a link here from the javadoc, but a link to the javadoc from
here seems like a good idea.
Test: N/A
Change-Id: I89a29f72d086d08174e72f7d0aa0421fe417f733
liblog_for_runtime_apex is a static variant of liblog which is
explicitly marked as available to the runtime APEX. Any static
dependency to liblog from inside the runtime APEX is changed from liblog
to liblog_for_runtime_apex.
Previously, to support the need for using liblog inside the runtime
APEX, the entire (i.e. both static and shared variants) liblog module
was marked as available to the runtime APEX, although in reality only
the static variant of the library was needed there. This was not only
looking dirty, but also has caused a problem like b/241259844.
To fix this, liblog is separated into two parts. (1) liblog and (2)
liblog_for_runtime_apex. (1) no longer is available to the runtime APEX
and is intended to be depended on in most cases: either from the
non-updatable platform, or from other APEXes. (2) is a static library
which is explicitly marked as available to the runtime APEX and also
visible to certain modules that are included in the runtime APEX.
Bug: 241259844
Test: m and check that liblog depends on stub library of libc
Change-Id: I10edd4487a6f090ef026acffe1ffbd067387a0d3
r.android.com/2108505 was intended to fix a crash in Scudo in
the case where the stack depot, region info or ring buffer were
unreadable. However, it also ended up introducing a number of bugs into
the code. It failed to call __scudo_get_error_info if the page at the
fault address was unreadable. This can happen in legitimate crash cases
if a primary allocation was close to the boundary of a mapped region,
or if the allocation was a secondary allocation with guard pages. It
also used long as the type for tags, whereas Scudo expects it to be
char. In combination this ended up causing most of the MTE tests to
fail. Therefore, mostly revert that change.
Fix the original crash by null checking the pointers returned by
AllocAndReadFully before proceeding with the rest of the function.
Bug: 233720136
Change-Id: I04d70d2abffaa35fe315d15d9224f9b412a9825d
Added SPDX-license-identifier-Apache-2.0 to:
debuggerd/test_permissive_mte/Android.bp
Bug: 68860345
Bug: 151177513
Bug: 151953481
Test: m all
Change-Id: Ic48cf8a972147eba8a955136be74204c013ca436
In the fallback path, if the non-main thread is the target
to be dumped, then no other threads are dumped when creating
a tombstone. Fix this and add unit tests to verify that
this all threads, including the main thread are dumped.
Bug: 234058038
Test: All unit tests pass.
Test: debuggerd -b media.swcodec process
Test: debuggerd media.swcodec process
Change-Id: Ibb75264f7b3847acdbab939a66902d986c0d0e5c
The code doesn't properly check if data is not read properly, so
make it fail if reads fail. Also, change the algorithm so that
first try and read the faulting page then 16 pages before and 16
pages after. Rather than trying to read every one of these pages,
stop as soon as one is unreadable. This means that the total memory
passed to the scudo error function is all valid data, rather than
potentially being some uninitialized memory.
Added new unit tests to cover scudo address processing.
Bug: 233720136
Test: All unit tests pass.
Test: atest CtsIncidentHostTestCases
Change-Id: I18a97bdee9a0c44075c1c31ccd1b546d10895be9
This is not meant to be enabled long-term, but can be used to assess system
stability with MTE before enabling it.
Bug: 202037138
Change-Id: I9fb9b63ff94da2de0a814fd7150f51559d3af079
If a process requires executing fallback unwinder and the thread
crashing is not the main thread, the wrong unwinder is used.
Fix this case, and add a new unit test that causes an abort in
the non main thread.
Bug: 233721755
Test: New unit test passes with fix and fails without.
Test: Ran debuggerd on swcodec process and it still dumps all threads.
Change-Id: I70fffc5d680256ce867e7a1d427593b584259160
Merged-In: I70fffc5d680256ce867e7a1d427593b584259160
(cherry picked from commit 2d5d46ca85)
This simplifies most of the calls to avoid doing any Android
specific code.
Bug: 120606663
Test: All unit tests pass.
Change-Id: I511e637b9459a1f052a01e501b134e31d65b5fbe
With the addition of runtime-configurable GWP-ASan, there might be many,
many more than 1,000 allocations. Have support for them, but keep a
hopefully-won't-crash-the-device limit.
Bug: 219651032
Test: atest bionic-unit-tests
Change-Id: I7b8e2bf5ab7c723ab6c61365f0dc610e400dbbce
One is intentionally seeting the abort message. The other is to set
the abort message to null.
Also, make the libseccomp_policy static so that the crasher
executable can be copied to the system afterwards without
requiring libseccomp_policy.so.
Test: Ran both new crash commands on device.
Test: Ran the seccomp crash command to verify seccomp still works.
Change-Id: I255b5f37e6eb188719e5b72302ca3f5911c8d821
Inline definitions of a few constants that don't appear on Q/R devices,
so that this works for us in mainline modules that are loaded on those
older devices.
Bug: 225406881
Test: boot on Q, watch logcat
Test: boot on R, watch logcat
Change-Id: Ic5781976d4c1e2d16e230c015fc49d9fde74e289
The functionality moved from the Unwinder object to the MapInfo
object and means that the individual unreadable files can be
displayed now.
Included adding the unreadable elfs per thread in the protobuf.
Updated the unwinder test.
Test: All unit tests pass.
Change-Id: I7140bde16938736da005f926e10bbdb3dbc0f6f5
When dumping a tombstone using the fallback path, only the main
thread was showing up. Modify the code to dump the threads using
a slightly different path for the tombstone generation code.
In addition, while looking at this code, two MTE variables were
not set in the tombstone fallback code. Added those variables
so MTE devices will work properly in this fallback path.
Modified the tombstone unit tests for seccomp to have
multiple threads and verify those threads show up in the tombstone.
Bug: 208933016
Test: Ran unit tests.
Test: Ran debuggerd <PID> on a privileged process and verified
Test: all threads dumped. Also verified that the tagged_addr_ctrl
Test: variable is present on the raven device.
Change-Id: I16eadb0cc2c37a7dbc5cac16af9b5051008b5127
debuggerd_test depends on it, and the easiest way to
ensure that the file is available when running the tests
is to make it a dependency of crash_dump.
Change-Id: Iebea8e0c49d8d49d52a434e4194e870793758988
Change use of new_ to old_ to save the old sigaction data. This hasn't
caused any issues, but it's obviously wrong.
Test: Ran unit tests on coral.
Change-Id: I96be5b0980c323c3aeafb422fbc06202577604a2
Hard to get otherwise if you're trying to debug PAC issues.
Bug: http://b/214314197
Test: treehugger
Change-Id: I2e5502809f84579bf287364e59d6e7ff67770919
The frame data no longer contains map_XXX fields which represent
the map data. Now there is only a shared pointer to the MapInfo
object with which this frame is associated.
Bug: 120606663
Test: Unit tests pass.
Change-Id: I89282963f742f6fcc07e48533da4108dc16bdce9
It is expensive to keep the non-protobuf path around and it hasn't
been used for an entire release without anyone noticing, so remove it.
Create new end-to-end unit tests that cover tests of the non-proto
code paths that are being deleted.
Bug: 197981919
Test: Unit tests pass.
Change-Id: Ia1c45572300bd63e5f196ad61e5e5386830c8ece
- Use "likelihood" instead of "probability" since that has connotations
of being less precise, and our probability ordering isn't very precise
anyway.
- Hide the fault address with SEGV_MTEAERR because it is not available.
- Pad the fault address with leading zeroes to make it clearer which
bits of the top byte (and any following bytes such as PAC signature
bits) are set.
Bug: 206015287
Change-Id: I5e1e99b7f3e967c44781d8550bbd7158eb421b64
On the main thread, the siginfo pointer will never be nullptr.
Add a CHECK to make sure this is true.
Test: Unit tests pass both 32 bit and 64 bit.
Test: Ran with debug.debuggerd.translate_proto_to_text set to 0
Test: to exercise old path.
Change-Id: I9d5ed0de5d652de8a4f9cd85eb57cbb1ec676404
This code was added, but a svelte config still tries to use scudo
related code that doesn't exist.
Bug: 201007100
Test: Ran unit tests on normal config.
Test: Ran unit tests on svelte config.
Change-Id: Ic84bae37717d213121aef182bac2f82dbee25213
strerror is nice, but usually I don't care about the text, I care about
the uppercase enum
Bug: N/A
Test: N/A
Change-Id: I8ea86220cb04cbded701379c47b8aba8ea8864b8
I was here because we have a case where timeout(1) kills logcat, but
debuggerd alleges that the process that was killed had started less than
a second ago. I'm not sure this is the problem there, but I did notice
that far too many tombstones were claiming improbably short process
uptimes. It turns out that the code was measuring the *thread* uptime,
not the *process* uptime.
Also simplify the code a bit by switching to sysinfo(2) rather than
reading a file.
Test: manual, plus the existing unit test
Change-Id: Ie2810b1d5777ad9182be92bfb3f60795dc978b24
