1) All current users are better off ignoring SIGPIPE at the beginning
of their process instead of ignoring it just for SocketClient
2) This isn't thread safe if users did want it, since sigaction()
ignores SIGPIPE for the entire process
3) This costs 5-10% of logd CPU time when logcat is reading logs
Also clean up the error handling in SocketClient::sendDataLockedv().
Test: kill logcat and see that logd doesn't crash
Test: run simpleperf and see that no cycles are going to sigaction
Change-Id: I6532c8a0d71338e534411707b9a9bd785145c730
When parsing an RTM_NEWADDR or RTM_DELADDR, ifaddr is always
present (unless the message is invalid). So ifaddr->ifa_flags is
always known before any attributes are parsed.
Bug: 155005831
Test: atest NetworkStackNextIntegrationTests:IpClientIntegrationTest continues to apss
Change-Id: Id1998faccca7d81c1b7f3e85e4912aa22919e94a
In RTM_NEWADDR messages, the first 8 flags are reported in the
ifa_flags field in struct ifaddrmsg, but flags above 0x80 are
reported in the IFA_FLAGS attribute. NetlinkEvent currently only
looks at ifa_flags, so it cannot see higher-value flags such as
IFA_F_STABLE_PRIVACY. Fix this by parsing the IFA_FLAGS
attribute.
Bug: 155005831
Test: makes new test in aosp/1295495 pass
Change-Id: I723f1106cbcea2186fc6452305942a0f8301fd2a
The ipt_ULOG.h file has been deprecated, so it's being removed from
the bionic uapi kernel headers. Since this is the only code that
is using that file, copy the one structure needed to parse one type
of message into the file that needs it.
Test: Builds.
Change-Id: Ib879e655d598256c424792675d24e647eee33f89
Enable the same warnings used elsewhere in the tree, and fix the
two warnings it found (two safe uses of strcpy and one use of
atoi).
Test: builds, boots
Test: atest libsysutils_tests
Test: atest --test-mapping system/netd
Change-Id: I8e5d04da20cc4127439ad9fbda0e4e64f0e671d3
Children of processes that use SocketListener should not be able
to stop the SocketListener.
Test: builds, boots
Test: atest libsysutils_tests
Test: atest --test-mapping system/netd
Change-Id: I64898d9966f62004468b8e8a43b59be4a81a8cc4
Only Netd and Logd use FrameworkListener currently.
Run realted test to ensure anything are still fine.
Bug: 126307309
Test: built, flashed, booted
system/netd/tests/runtests.sh pass
atest CtsLogdTestCases
Change-Id: Iacd0cc13f23f457a0868a8a706080c3edc9a3cbd
Add FALLTHROUGH_INTENDED for clang compiler.
Bug: 112564944
Test: build with global -Wimplicit-fallthrough.
Change-Id: I40f8bbf94e207c9dd90921e9b762ba51abab5777
Convert the last user to a regular std::vector, and then put List.h out
of its misery.
Turns out there's another copy of this ancient header:
system/core/libutils/include/utils/List.h. This one is included in
various places, but most don't actually use it. There seems to
be a case for a tool to detect unused headers...
Test: m
Change-Id: Ie457bf5e06f6082537f5de2e1fefe7f05a1f0792
FD_SET is limited to 1024 file descriptors in Linux, which causes
processes with too many open files or connections to crash:
FORTIFY: FD_ISSET: file descriptor 1024 >= FD_SETSIZE 128
The fix we used elsewhere is replacing select() with poll(), but in the
case of SocketListener we additionally need to replace the SocketClient
list with a map indexed by fd in order to avoid quadratic behavior on
each poll() wakeup.
Bug: 79838856
Test: device boots and appears to work normally, tests pass
Change-Id: I4a8f1804fa990d3db3a2c96b9acd60b2c7135950
As documented at go/gerrit-owners, per-file directives cannot
contain directory paths. Therefore the current per-file
directives in system/core/OWNERS have no effect.
Test: build/make/tools/checkowners.py system/core/{,libsysutils/src/,include/sysutils/}OWNERS
Change-Id: Ia88b2be42dd50346578bb51dd17fb6136a4591f1
This reverts commit 0ad41cf3f4.
Reason for revert: Breaks many tests, see b/109745952#comment18 b/109745952#comment20 and b/79838856#comment19 amongst others
Change-Id: I92db38d47cd1dcd7841091e37e3b65dce9456cdf
FD_SET is limited to 1024 file descriptors in Linux, which causes
processes with too many open files or connections to crash:
FORTIFY: FD_ISSET: file descriptor 1024 >= FD_SETSIZE 128
The fix we used elsewhere is replacing select() with poll(), but in the
case of SocketListener we additionally need to replace the SocketClient
list with a map indexed by fd in order to avoid quadratic behavior on
each poll() wakeup.
Bug: 79838856
Test: device boots and appears to work normally
Change-Id: I19ca4be675e9638104c0e7acf4a4bc62085e8ecd
The TrafficController inside netd also get notified when the interface
on device added or removed, and it need both ifaceName and ifaceIndex
information from Netlink event. Add the ifaceIndex in the event can
avoid the TrafficController to run if_nametoindex again when receive
the event.
Bug: 30950746
Bug: 73137611
Test: new interface show up in bpf interface map.
Change-Id: Ia3b4aeb8eff345ce7cf03ccec88f0893cfe125c0
This option isn't unknown, we just don't support it.
Bug: 69778511
Test: walleye builds, boots, connects to IPv6 network
Change-Id: If3a37d855f404da708fec4a7ba242ed41a493895
We only use it for trivial functions. Replace them and drop the
dependency.
Bug: 67345547
Test: bullhead builds, boots
Test: CtsOsTestCases android.os.cts.StrictModeTest passes
Change-Id: I36254962284babdd1a55a32a76dd0dc92d85420c
As a side effect, this change fixes
libbsysutils => liblog dependencies. LOG_EVENT_*
macros are not defined for liblog.vendor.
Test: BOARD_VNDK_VERSION=current mma
Bug: 33241851
Bug: 63034227
Change-Id: I3df700d97481ec9aa59cf1796da1e991af28bdf5
Don't release the sender socket when FrameworkListener receives
a command that exceed the maximum buffer length
Bug: https://issuetracker.google.com/issues/62812018
Test: manual
Change-Id: I160f504f3fb902ba25fc79deb8ae7d78d811a807
This fixes warning: Null pointer argument in call to
memory comparison function [clang-analyzer-cplusplus.NewDeleteLeaks]
Bug: None
Test: The warning is gone.
Change-Id: I957365184966cc1435d7e37d64f2cc6a32846ebd
Point to log/log.h where necessary, define LOG_TAG where necessary.
Accept that private/android_logger.h is suitable replacement for
log/logger.h and android/log.h.
Correct liblog/README
Effectively a cleanup and controlled select revert of
'system/core: drop or replace log/logger.h' and
'system/core: Replace log/log.h with android/log.h'.
Test: compile
Bug: 30465923
Change-Id: Ic2ad157bad6f5efe2c6af293a73bb753300b17a2
Should use android/log.h instead of cutils/log.h as a good example
to all others. Adjust header order to comply with Android Coding
standards.
Test: Compile
Bug: 26552300
Bug: 31289077
Change-Id: I2c9cbbbd64d8dccf2d44356361d9742e4a9b9031
Modify FrameworkListener to ignore commands that exceed the maximum
buffer length and send an error message.
Bug: 29831647
Change-Id: I9e57d1648d55af2ca0191bb47868e375ecc26950
Signed-off-by: Connor O'Brien <connoro@google.com>
(cherry picked from commit baa126dc15)
When an interface is deleted (which can happen, for example, to
the tun interfaces used for VPN and 464xlat), the kernel sends
RTM_DELADDR messages for all its IP addresses.
Currently we ignore these because we cannot fetch the interface
name. Instead, pass them up with an empty interface name so that
NetlinkHandler can do something useful for them, such as destroy
connections on the IP addresses that are being deleted.
Bug: 26976388
Change-Id: I2ced7e389228f9d665d0a87d97bd3452c8c0c98b
Tom Cherry