rprichard@ pointed out a bug where LogBufferElement::setDropped()
reallocates mMsg to the size of mMsgLen in the case where getTag() !=
0. However, mMsgLen is in a union with mDroppedCount and
mDroppedCount is the value used when a message is already dropped.
Therefore, it's possible that logd uses the wrong value and allocates
much more memory than intended.
We do call setDropped() on elements that have already been dropped in
LogBufferElementLast::coalesce(), so this is not a superfluous issue.
To simplify this even more, this code puts an mTag in a union with
mMsg, such that if mDropped is true, there will never be an allocated
message; mTag will be directly referred to instead. This also reduces
the number of allocations needed very slightly.
Test: logd/liblog/logcat unit tests.
Change-Id: Ia1bfba076439fe31c745a243283d41902bca45ac
logd isn't meant to be modularized. The previous user was using a
small subset of LogListener.cpp, which is now copied into their
project.
Test: liblog, logd unit tests
This reverts commit fafea32468.
Change-Id: I05ec764db2d9395f2d5b69a1a610c9c55240ab3a
We don't want to fake socket credentials if they were not provided by
the kernel. If there is a bug preventing us from reading the
credentials then it must be solved directly.
Test: logd, liblog unit tests
Test: boot and ensure overflow uid doesn't show up
This reverts commit c4e4823b00.
Change-Id: I683129a8a214637635f163ae25c39bb8a47cd50f
for arm64, userspace get time by vDSO, no syscall overhead.
multiple_test_2 sometimes get 2 same timestamps now, then testcase
fails.
Bug: b/135555666
Test: ./CtsLogdTestCases64 --gtest_filter=*multiple_test*
uidToName() originally used a separate worker thread with additional
group permissions. Threads are not security boundaries however, so
these group permissions are removed in a previous change.
This change handles the lookup for uidToName() directly without using
a separate thread.
Test: boot CF, logd unit tests
Change-Id: If245388bc221bc77102a0bbcee82c8f42b140760
On Android, unlike POSIX, groups and capabilities are able to be set
per thread. This is useless however, since threads are not a security
boundary. This change drops the logic to set groups and capabilities
per thread and instead leaves all threads running with the initial
user and groups.
This does still drop some capabilities if they're unneeded due to
features being disabled.
This also moves the setpriority() call from code into the init script.
Test: logd runs with the expected user/groups and with the expected
capabilities and priority without any errors
Change-Id: Ibb0e529ea1574a2b8ec391a2678504ca9fbe19be
Logcatd has capability to output logs to filesystem with certain size
and certain file count, however file size is not configurable, fixed
as 1024 kbytes, file count is configurable, but original property
name didn't match the logcat parameter well.
This patch add interface rotate_kbytes and count into logcatd.rc.
rotate_kbytes used to control each logcat file size.
count is another alias for logd.logpersistd.size to control file number.
Bug: 133362078
Test: Can use logcat -r -n with configures
Change-Id: I9954c9c125a4ab4e49310986f81c734bf8ee96b3
Signed-off-by: Tian, Baofeng <baofeng.tian@intel.com>
Signed-off-by: Duan, YayongX <yayongx.duan@intel.com>
CTS tests are not supposed to link against platform-private
libraries. As it is this test put an implicit requirement on
having libselinux.so in system image. Which is not intended.
Link statically against listed libraries to avoid such unintended
dependencies.
Test: make
Change-Id: I47a5b032000ee2c1ff764d6cc57683343851a07f
Remove the code which attempts to collapse multiple similar SELinux
denials into one denial. SELinux logs are already rate throttled by the
kernel
(https://android-review.googlesource.com/c/platform/system/core/+/941357
for example), so additional filtering here only adds complexity and may
result in SELinux denials being inappropriately suppressed or dropped.
Bug: 118815957
Test: atest logd-unit-tests
Change-Id: I611fa43dc56c99aa992aad002c16c742e6034e5c
In an effort to ensure that our development community does not
introduce new code without corresponding SELinux changes, Android
closely monitors the number of SELinux denials which occur during
boot. This monitoring occurs both in treehugger, as well as various
dashboards. If SELinux denials are dropped during early boot, this
could result in non-determinism for the various SELinux treehugger
tests.
Introduce /system/bin/auditctl. This tool, model after
https://linux.die.net/man/8/auditctl , allows for configuring the
throttling rate for the kernel auditing system.
Remove any throttling from early boot. This will hopefully reduce
treehugger flakiness by making denial generation more predictible
during early boot.
Reapply the throttling at boot complete, to avoid denial of service
attacks against the auditing subsystem.
Delete pre-existing unittests for logd / SELinux integration. It's
intended that all throttling decisions be made in the kernel, and
shouldn't be a concern of logd.
Bug: 118815957
Test: Perform an operation which generates lots of SELinux denials,
and count how many occur before and after the time period.
Change-Id: I6c787dbdd4a28208dc854b543e1727ae92e5eeed
This reverts commit 1bef8c550c.
Reason for revert: AOSP is fixed with new vendor image
Change-Id: Ib341ac80e2f88c13a7815a490ea2d9422ebdf55f
Signed-off-by: Suren Baghdasaryan <surenb@google.com>
selinux_denial_metadata contains device-specific information, so it is
being moved to /vendor.
Bug: 5159394
Test: bug information is still preserved in avc logs, e.g.
audit(0.0:248): avc: denied { read } for
name="u:object_r:vendor_default_prop:s0" dev="tmpfs" ino=18012
scontext=u:r:platform_app:s0:c512,c768
tcontext=u:object_r:vendor_default_prop:s0 tclass=file permissive=0
b/79617173 app=com.android.systemui
Change-Id: I23293e715b5d793a54466a128ccd57f44370cf5a
It's replaced with entries in PRODUCT_PACKAGES_DEBUG in
build/make/target/product/base_system.mk
Test: treehugger
Change-Id: I4dc69c34ddc2c494fc74bc4afee6efa240c9b0d3
logd/logcat use more symbols than the rest of the platform, so let's
statically link them for now while we understand the API that the
platform overall uses.
Test: build
Change-Id: I21ae1bd3ff9b310f099612fbe790f769712661df
- Zero initialize log_time instances by default
- Disable implicit conversions by making constructors explicit
- Explicitly initialize to EPOCH in most places
- Change sniffTime() to avoid in-place modification of a log_time
I stopped here, but we could consider following up with a more invasive
change to make log_time instances immutable and perhaps also remove the
default constructor to force explicit initialization to EPOCH.
Test: atest libbase_test netd_unit_test
Change-Id: I67e716ef74adaaf40ab2c6e2e0dddb8d309bc7ca
upon memory usage high(>log_buffer_size), logd will try to prune(erase) all those old log elements which have been read by all readers for reclaiming the memory. As such, a too slow reader will be a hinder to the success of the prune. Logd has to try to kick-out the slow-est reader when memory usage is really too high(>2 * log_buffer_size). But the kick-out operation is just a request to the reader and at what time the reader will exit is always uncertain, beyond control. Furthermore, if you kick-out reader-A, waiting for A to exit; then, another reader-B may come in; then A exit; and then you kick-out-B, waiting for B to exit; and then, ...loop for ever: yes, logd may find that there seems to be always a slow reader hinder its pruning. As we all know, that, logd will probably kick-out a slow reader(logcat), hence, indeed, almost all log capturing tools will try to re-connect logd immediately after it being kick-out-ed. Such retry makes the issue easy to happen. And, we observed that the reader thread may often be blocked by socket write operation, which hindering its exiting and hereby hindering the prune progress. We need gracefully shutdown socket to relieve it from blocking and eventually stop such disaster from happening.
Test: monkey test for one day and one night
Change-Id: I5496ff74168b71e261914b91c145aa44814a5def
LogBufferElement copy constructor supported only partial function.
Solution: handle all cases.
Test: unit test for calling the copy constructor with all possible states.
Change-Id: I55091569d98eb35a09b4c3fc068836ecd256558c
There is an issue in LogAudit::auditParse where
android::uidToName(uid) crashes with a null pointer dereference.
Include a null check on the value before passing it on.
Bug: 120043607
Test: End-to-end test with syzkaller as per instructions in bug.
Change-Id: Ic0ac5c3003fcd289ec156ce63fbd668413763429
Kernel panic - not syncing: Out of memory occurred once after 8 hours
automatic test which cases are about Summary:logd memory leak UTResult
: 1/1
IssueID: 93440
Rootcause: the method "LogTimeEntry::FilterSecondPass"(called inside
"LogBuffer::flushTo") does not check the condition flag
"LogTimeEntry::mRelease" which will be set to true when "logd" begins
to kick off a client when it finding the client is slow-est.
Therefore the client cannot be kicked off and hence the log-element
list becomes longer and longer if the slowest client reading more
slower than the writer writing.
Change-Id: I91a44a393e63b3122c15ce269ee195e0eb4339e8
Solution: add the check.
Test: logd kills slow reader
This code was accidentally deleted in the refactoring of
4f22786cc9 cause the
logd.timeout_no_start test to start failing.
Bug: 117942520
Test: logd unit tests
Change-Id: Id7994230a89d7f8c2cb1f36a8ca1175a6b398de0
LogTimeEntry's lifecycle is spread out in various locations. It
further seems incomplete as there is logic that assumes that its
associated thread can exit while the underlying LogTimeEntry remains
valid, however it doesn't appear that that is actually a supported
situation.
This change simplifies this logic to have only one valid state for a
LogTimeEntry: it must have its thread running and be present in
LastLogTimes. A LogTimeEntry will never be placed into LastLogTimes
unless its thread is running and its thread will remove its associated
LogTimeEntry from LastLogTimes before it has exited.
This admittedly breaks situations where a blocking socket gets issued
multiple commands with different pid filters, tail lines, etc,
however, I'm reasonably sure that these situations were already
broken. A check is added to close the socket in this case.
Test: multiple logcat instances work, logd.reader.per's are cleaned up
Change-Id: Ibe8651e7d530c5e9a8d6ce3150cd247982887cbe
Add FALLTHROUGH_INTENDED for clang compiler.
Bug: 112564944
Test: build with global -Wimplicit-fallthrough.
Change-Id: I40f8bbf94e207c9dd90921e9b762ba51abab5777
This change removes the call to prctl(PR_SET_KEEPCAPS, 1), because now
that logd is being run with ambient capabilities, init will set the
KEEP_CAPS / KEEP_CAPS_LOCKED securebits.
Bug: 110127913
Test: liblog-unit-tests, logd-unit-tests, logcat-unit-tests
(on an aosp_sailfish-userdebug)
Change-Id: I72b6a49c5cd1552085f9ad44cb19a7333a13eeea
This change moves logd's capabilities from being file based to being set
by init through ambient capabilities.
Bug: 62845925
Test: sailfish:/ # grep Cap /proc/`pidof logd`/status
CapInh: 0000000000000000
CapPrm: 0000000440000000
CapEff: 0000000440000000
CapBnd: 0000000440000040
CapAmb: 0000000000000000
$ capsh --decode=0x440000040
0x440000040=cap_setgid,cap_audit_control,cap_syslog
Test: liblog-unit-tests, logd-unit-tests, logcat-unit-tests
Change-Id: I28c7c4ad37ad5eafd399aef78c303ce31298a9ef
