| Age | Commit message (Collapse) | Author |
|---|
|
We calculate dex2oatWallTimeMillis by (uptime - starttime). To get a
reasonable result, both times need to be from the same clock. However,
it's unclear which clock the kernel uses for the latter.
proc_pid_stat(5) does not have anything helpful on this matter:
(22) starttime %llu
The time the process started after system boot.
Before Linux 2.6, this value was expressed in
jiffies. Since Linux 2.6, the value is expressed
in clock ticks (divide by sysconf(_SC_CLK_TCK)).
The format for this field was %lu before Linux 2.6.
To make sure we use the same clock, we obtain the uptime on process
start as the start time.
Bug: 315061143
Test: Presubmit
Change-Id: I5f344e953a945656474b3f625368c1ccb5aa52d8
|
|
This change adds a new option `new_process_group` to ExecUtils, and this
option is only used by artd to manage its subprocesses. artd creates a
process group for each child process and its descendants, and it uses
`kill(-pid, SIGKILL)` to kill the whole process group upon job
cancellation.
Before this change, ExecUtils always creates a new process group for a
child process. This behavior can be dated back to
https://android.googlesource.com/platform/art/+/5643b786dd48df1f4b89a39e6024a22d032c79d4%5E%21/.
A comment on the `setpgid` call says: "change process groups, so we
don't get reaped by ProcessManager".
We cannot find out why changing the process group was needed or how
process reaping was related to process groups at that time because there
isn't such a thing called "ProcessManager" in Android anymore. However,
we don't need this behavior today because the usage of the code that
does fork and exec has completely changed.
At the time where the original CL was submitted, the code was for the
runtime to generate a persistent boot image on /data. Today, the runtime
never generates a persistent boot image because it's considered insecure.
Today, ExecUtils is mostly used in tests. Other than tests, it's only
used in four places: Zygote, odrefresh, artd, and dexopt_chroot_setup.
- Zygote uses it to call dex2oat to create an in-memory boot image. It
doesn't make sense to keep dex2oat running when Zygote (the memfd
owner) is killed.
- odrefresh uses it to call dex2oat to create persistent boot images. We
do want to kill dex2oat when killing odrefresh.
- artd uses it to call dex2oat, profman, derive_classpath, and
odrefresh. It sets `new_process_group` to true, so it's behavior
regarding process groups is unchanged. It's a service managed by
the service manager. The service manager uses cgroups, not process
grups, to manage and kill processes, so creating new process groups
for artd's children is fine.
- dexopt_chroot_setup uses it to call apexd and linkerconfig. We do want
to kill apexd and linkerconfig when dexopt_chroot_setup is killed.
Therefore, changing this behavior is fine.
After this change, ExecUtils no longer creates a new process group for a
child process unless `new_process_group` is set to true.
Bug: 345723405
Bug: 311377497
Test: m test-art-host-gtest-art_runtime_tests
Test: m test-art-host-gtest-art_artd_tests
Test: -
1. adb shell pm compile -m speed -f -v com.android.settings
2. adb shell pm art cancel <job-id>
3. No change to the existing behavior: dex2oat is still killed.
Test: -
1. adb shell pm art pr-dexopt-job --run
2. Press Ctrl+C.
3. Both odrefresh and dex2oat are killed.
Change-Id: I3058fc68a6bd5f98617422c1eb0861648b144a51
|
|
This reverts commit 7afcee1180986d8b87773d173af7ecd63859ffe3.
Reason for revert: It doesn't work. See details in b/345723405#comment2
We'll find an alternative.
Bug: 345723405
Bug: 311377497
Change-Id: I70152124bb2f996c5211b4f4bdf9519e80bdf6ca
Test: Presubmit
|
|
Before this change, we only trigger GC and sleep 5 seconds, hoping that
processes in chroot exit within that 5 seconds.
After this change, we use pidfd to wait for processes to exit, so
teardown can be performed as soon as all processes have exited, making
teardown considerably faster.
Also, if the processes don't exit within 5 seconds, we kill them, making
teardown more reliable.
Only one more SELinux permission is needed, which is the permission to
send `sigkill` to artd and its subprocesses. system_server already has
all the other permissions to perform the operations used in this CL.
1. system_server has supplementary group "readproc" (3009), to list all
pids in "/proc", bypassing "hidepid=2" (see proc(5)).
- https://cs.android.com/android/platform/superproject/main/+/main:frameworks/base/core/java/com/android/internal/os/ZygoteInit.java;l=658;drc=7d3ffbae618e9e728644a96647ed709bf39ae759
2. system_server can read /proc/<pid>/* for all domains.
- https://cs.android.com/android/platform/superproject/main/+/main:system/sepolicy/private/system_server.te;l=213;drc=ca2f3851afaee866d37caae16598b3d5c20844d4
3. system_server has CAP_SYS_PTRACE, to read the "/proc/<pid>/exe" link
of anyprocess (see proc(5)).
- https://cs.android.com/android/platform/superproject/main/+/main:frameworks/base/core/java/com/android/internal/os/ZygoteInit.java;l=635;drc=7d3ffbae618e9e728644a96647ed709bf39ae759
- https://cs.android.com/android/platform/superproject/main/+/main:system/sepolicy/private/system_server.te;l=145;drc=ca2f3851afaee866d37caae16598b3d5c20844d4
4. system_server has CAP_KILL, to kill processes that belong to other
users (see kill(2)).
- https://cs.android.com/android/platform/superproject/main/+/main:frameworks/base/core/java/com/android/internal/os/ZygoteInit.java;l=628;drc=7d3ffbae618e9e728644a96647ed709bf39ae759
- https://cs.android.com/android/platform/superproject/main/+/main:system/sepolicy/private/system_server.te;l=138;drc=ca2f3851afaee866d37caae16598b3d5c20844d4
Bug: 311377497
Test: m test-art-host-gtest-art_libarttools_tests
Test: atest art_standalone_libarttools_tests --iterations 10
Test: adb shell pm art pr-dexopt-job --test
Test: Run and cancel Pre-reboot Dexopt multiple times.
Change-Id: I1e41cd71402944e31b33e410ac1635766afe55c4
|
|
When we have a process hierarchy, such as artd->odrefresh->dex2oat, we
want to be able to kill a subtree. For example, if we kill odrefresh, we
want dex2oat to be killed as well; if we kill artd, we want both
odrefresh and dex2oat to be killed as well. This can be achieved by
PR_SET_PDEATHSIG.
Bug: 311377497
Test: Kill odrefresh and see dex2oat killed.
Change-Id: I28cc271f5c37b9ad50bee179eb9b211cb3545a2f
|
|
Bug: 260881207
Test: presubmit
Test: abtd app_compat_drm
Test: abtd app_compat_top_100
Test: abtd app_compat_banking
Change-Id: I0eca5c4fd64ec61388eded9b9e066091581e0c7e
|
|
- Handle the case where start time is 0.
- Handle the case where getting uptime failed.
- Use 64bit int. (32bit is enough to record a time less than 24 days, but
just in case.)
Bug: 315061143
Test: m test-art-host-gtest-art_runtime_tests
Change-Id: I3ac546f6921b8d6add2491170b5d8f50cc7169c1
|
|
Test: m tidy-art
Bug: 264654008
Merged-In: I6b0b495cbb934f9b8caedb389a643428d3cffc47
(cherry picked from commit d59d4d116f9d40113c32c8b1225388308830b099)
Change-Id: Ic0a52fb2c6842d18b292088135a39d6886b69d46
|
|
This is the first step to improve odrefresh metrics collection. More CLs
will follow.
After the change, the metrics should tell:
- whether dex2oat is invoked or not
- whether dex2oat crashed or exited with a code
Partially cherry-picked from
commit fa152ba922a6301a10671d4684b75fe8cb2a2330
Bug: 246534524
Test: m test-art-host-gtest-art_runtime_tests
Test: atest ArtGtestsTargetChroot:OdRefreshTest
Merged-In: Id962a6e6a765371d290a836c51059d725c9250f3
Change-Id: I69ddf64552d4f2e51e111a40871890d41c88721d
|
|
artd needs the callbacks to get the pid in order to kill the subprocess.
Partially cherry-picked from
commit 659f49bdb1263ceb26f666052f3ef7e4732b4eb2
Bug: 244412198
Test: m test-art-host-gtest-art_runtime_tests
Merged-In: I8c40949b5ed88ff85ddedad9d86f0b9bbfddb98d
Change-Id: Ia28410bdf6314178756ee5e3a94000dc6f3ecc97
|
|
Bug: 245380798
Test: m test-art-host-gtest-art_runtime_tests
Change-Id: I25d0d4c3dfaf24490c96ca650f076b2214597d2a
Merged-In: I25d0d4c3dfaf24490c96ca650f076b2214597d2a
(cherry picked from commit ebbd20c55adbdde7922c2ca402e1a04ed6ecee4c)
|
|
`WaitChildWithTimeout` didn't work with old Linux kernels because it
uses a syscall `pidfd_open`, which was added in Linux 5.4.
This change adds a fallback implementation of `WaitChildWithTimeout`
that creates a thread to wait instead of relying on `pidfd_open`.
Also:
- Use android::base::unique_fd to hold pidfd so that the fd is
guaranteed to be closed.
- Allow timeout_sec to be negative, in which case it blocks until the
subprocess exits.
Bug: 229268202
Test: m test-art-host-gtest-art_runtime_tests
Change-Id: Iec3f21db135d9a8a3a915372253f1ff3e285e5cf
Merged-In: Iec3f21db135d9a8a3a915372253f1ff3e285e5cf
(cherry picked from commit 99bd8dd2bf8eac1a60bf65e442462753ed7f2147)
|
|
This reverts commit ff95ad517297384d4a2ea589490019f90d637e17.
Reason for revert: Fix available (aosp/2148047)
Test: Presubmit
Change-Id: I952960949267a7b6e63cd7966a347cb1f3374b85
Merged-In: I375034162cded6fdf09ee4d4cd783a0d3987af49
|
|
This reverts commit f0061f62fbd5e4766220c154d914e7b9d07a86eb.
Reason for revert: Broke LUCI.
Change-Id: Iec793a667545a8469f2a712135353e816c623262
|
|
- Use pidfd_open/poll instaed of sigprocmask/sigtimedwait to implement
executing a command with a timeout because sigprocmask/sigtimedwait
does not support multithreaded program, while artd is multithreaded.
The same approach is used in installd in
https://r.android.com/1993670.
- Add "const" to the `std::vector<std::string>& arg_vector` parameter.
- Use "int" instead of "time_t" for the "timeout_sec" parameter because
"time_t" represents the time since epoch rather than a duration.
- Set an error message with the signal number when the subprocess is
terminated by a signal.
- Crash the subprocess instead of exiting it with an exit code when
execv fails.
The behavior does not change. ExecUtilsTest is still passing.
Bug: 229268202
Test: m test-art-host-gtest-art_runtime_tests
Test: atest ArtGtestsTargetChroot:ExecUtilsTest
Change-Id: I375034162cded6fdf09ee4d4cd783a0d3987af49
Merged-In: I375034162cded6fdf09ee4d4cd783a0d3987af49
(cherry picked from commit b48b732c535894bbae8f3c19eceeda67c40e9d61)
|
|
Enables callers to timeout waiting on a subprocess.
Bug: 177432913
Test: test-art-host-gtest-art_runtime_tests32
Change-Id: I7a27e0ca4679f45daf815d7e2563ca9b723bd701
|
|
Add a strlcpy shim for the host, so we can use strlcpy instead of
strcpy everywhere.
Fixed warnings include unused-decls, (some) unreachable code, use
after std::move, string char append, leaks, (some) excessive padding.
Disable some warnings we cannot or do not want to avoid.
Bug: 32619234
Test: m
Test: m test-art-host
Change-Id: Ie191985eebb160d94b988b41735d4f0a1fa1b54e
|
|
The rest of utils.cc does not depend on art::Runtime. This separates
the part dependent on that class, so that including utils.cc in the
build does not require the entire Runtime. Another preparatory cleanup
to getting tools to build on Windows.
Bug: 22322814
Test: test-art
Change-Id: I194ff363fc2ab87e5311ecea6973a2d0fad2621d
|