• Home
  • Line#
  • Scopes#
  • Navigate#
  • Raw
  • Download
1# zygote-start is what officially starts netd (see //system/core/rootdir/init.rc)
2# However, on some hardware it's started from post-fs-data as well, which is just
3# a tad earlier.  There's no benefit to that though, since on 4.9+ P+ devices netd
4# will just block until bpfloader finishes and sets the bpf.progs_loaded property.
5#
6# It is important that we start bpfloader after:
7#   - /sys/fs/bpf is already mounted,
8#   - apex (incl. rollback) is initialized (so that in the future we can load bpf
9#     programs shipped as part of apex mainline modules)
10#   - logd is ready for us to log stuff
11#
12# At the same time we want to be as early as possible to reduce races and thus
13# failures (before memory is fragmented, and cpu is busy running tons of other
14# stuff) and we absolutely want to be before netd and the system boot slot is
15# considered to have booted successfully.
16#
17on load_bpf_programs
18    # Linux 5.16-rc1 has changed the default to 2 (disabled but changeable),
19    # but we need 0
20    write /proc/sys/kernel/unprivileged_bpf_disabled 0
21    # Enable the eBPF JIT -- but do note that on 64-bit kernels it is likely
22    # already force enabled by the kernel config option BPF_JIT_ALWAYS_ON
23    write /proc/sys/net/core/bpf_jit_enable 1
24    # Enable JIT kallsyms export for privileged users only
25    write /proc/sys/net/core/bpf_jit_kallsyms 1
26    exec_start bpfloader
27
28service bpfloader /system/bin/bpfloader
29    capabilities CHOWN SYS_ADMIN NET_ADMIN
30    #
31    # Set RLIMIT_MEMLOCK to 1GiB for bpfloader
32    #
33    # Actually only 8MiB would be needed if bpfloader ran as its own uid.
34    #
35    # However, while the rlimit is per-thread, the accounting is system wide.
36    # So, for example, if the graphics stack has already allocated 10MiB of
37    # memlock data before bpfloader even gets a chance to run, it would fail
38    # if its memlock rlimit is only 8MiB - since there would be none left for it.
39    #
40    # bpfloader succeeding is critical to system health, since a failure will
41    # cause netd crashloop and thus system server crashloop... and the only
42    # recovery is a full kernel reboot.
43    #
44    # We've had issues where devices would sometimes (rarely) boot into
45    # a crashloop because bpfloader would occasionally lose a boot time
46    # race against the graphics stack's boot time locked memory allocation.
47    #
48    # Thus bpfloader's memlock has to be 8MB higher then the locked memory
49    # consumption of the root uid anywhere else in the system...
50    # But we don't know what that is for all possible devices...
51    #
52    # Ideally, we'd simply grant bpfloader the IPC_LOCK capability and it
53    # would simply ignore it's memlock rlimit... but it turns that this
54    # capability is not even checked by the kernel's bpf system call.
55    #
56    # As such we simply use 1GiB as a reasonable approximation of infinity.
57    #
58    rlimit memlock 1073741824 1073741824
59    oneshot
60    #
61    # How to debug bootloops caused by 'bpfloader-failed'.
62    #
63    # 1. On some lower RAM devices (like wembley) you may need to first enable developer mode
64    #    (from the Settings app UI), and change the developer option "Logger buffer sizes"
65    #    from the default (wembley: 64kB) to the maximum (1M) per log buffer.
66    #    Otherwise buffer will overflow before you manage to dump it and you'll get useless logs.
67    #
68    # 2. comment out 'reboot_on_failure reboot,bpfloader-failed' below
69    # 3. rebuild/reflash/reboot
70    # 4. as the device is booting up capture bpfloader logs via:
71    #    adb logcat -s 'bpfloader:*' 'LibBpfLoader:*'
72    #
73    # something like:
74    #   $ adb reboot; sleep 1; adb wait-for-device; adb root; sleep 1; adb wait-for-device; adb logcat -s 'bpfloader:*' 'LibBpfLoader:*'
75    # will take care of capturing logs as early as possible
76    #
77    # 5. look through the logs from the kernel's bpf verifier that bpfloader dumps out,
78    #    it usually makes sense to search back from the end and find the particular
79    #    bpf verifier failure that caused bpfloader to terminate early with an error code.
80    #    This will probably be something along the lines of 'too many jumps' or
81    #    'cannot prove return value is 0 or 1' or 'unsupported / unknown operation / helper',
82    #    'invalid bpf_context access', etc.
83    #
84    reboot_on_failure reboot,bpfloader-failed
85    # we're not really updatable, but want to be able to load bpf programs shipped in apexes
86    updatable
87