* changes:
NetBpfLoad: Add method to pin programs loaded by libbpf
NetBpfLoad: Add method to pin maps loaded by libbpf
NetBpfLoad: Update helper methods to use borrowed_fd
Treehugger Robotauthored and Android (Google) Code Reviewcommited 2025-07-24 21:28:55 -07:00
Reduces ~900 lines of log spam by suppressing non-warning messages.
Verbose logging can be re-enabled for debugging by defining the
NETBPFLOAD_VERBOSE_LOG macro.
Bug: 428792764
Flag: EXEMPT logging only change
Test: TH
Change-Id: Ieac3e7ee843b889fb404a89cfe2319ec34ab365c
* changes:
NetBpfLoad: Add method to prepare map load by libbpf
NetBpfLoad: Extract building prog pin loc to a helper function
NetBpfLoad: Extract building map pin loc to a helper function
Treehugger Robotauthored and Android (Google) Code Reviewcommited 2025-07-24 00:45:20 -07:00
The function doesn't enforce anything specific to /proc/sys.
The name should reflect that.
(prep work for reusing this for other purposes - incl. selinux booleans)
Test: TreeHugger
Flag: EXEMPT mainline/no-op
Signed-off-by: Maciej Żenczykowski <maze@google.com>
Change-Id: Ia206c19043e63e274c7370739750e1a8236e4127
Maciej Żenczykowskicommited 2025-07-22 12:32:08 -07:00
Kernel uapi documents this field as unused.
libbpf sets this to 8+8+8 bits, while we do 8+8+16 bits
(to support kernels x.x.>255)
This causes crazy strace output...
Let's just remove it entirely.
Test: TreeHugger
Flag: EXEMPT mainline
Signed-off-by: Maciej Żenczykowski <maze@google.com>
Change-Id: Ie67f5a4b479d4555a593a4e6d8d14ff82efce98f
Maciej Żenczykowskicommited 2025-07-22 12:28:08 -07:00
This is to prepare for splitting out the method for pinning a prog.
Bug: 428792764
Flag: EXEMPT logging only change
Test: TH
Change-Id: Iedfc8f3941f8e8b0e70e4dfbc47520b29d332c3e
This is to prepare for splitting out the method for pinning a map.
Bug: 428792764
Flag: EXEMPT logging only change
Test: TH
Change-Id: Iebaeea3150a99e85c25bb106e74ce3fc7f9ddb50
This updates the BPF map macros so programs can be loaded by both the
existing code and the libbpf.
The macros now generate ".maps" section in addition to the
".android_maps" section.
To prevent name collisions, the legacy map definitions are suffixed with
_def. This CL also adds the readMapNames method to get map names from
the ".android_maps" section, allowing the existing code path to continue
working.
Bug: 428792764
Flag: EXEMPT no-op
Test: TH
Change-Id: Ibe6f100b5684bc85d2cf9c100f2431740fa516ca
The reuse boolean was set to true only in a code path that
immediately called abort().
Bug: 428792764
Test: TH
Flag: EXEMPT no-op
Change-Id: I157e9f09c87eedc82c24d4ec821dcb132a389e31
This is a no-op because the data in the ".android_maps"/"progs" section
is an array of struct bpf_map_def/bpf_prog_def.
Test: TH
Flag: EXEMPT no-op
Bug: 428792764
Change-Id: I80bd8c72c68b5359cf1b16ed5d4739a854169d7f
These tend to be huge (because they can contain the full bpf disassembly)
but not very worthwhile, because they often result in logs being rate
limitted, and thus we miss out on the useful information.
This is particularly true for user builds on non-Google devices
installing the production signed tethering apex mainline module.
We're better off seeing more of the other log lines...
Test: TreeHugger
Flag: EXEMPT mainline
Signed-off-by: Maciej Żenczykowski <maze@google.com>
Change-Id: Ie25a60c2e602772ab0de2050e7964f0202ff2685
Maciej Żenczykowskicommited 2025-07-15 15:00:25 -07:00
Not needed any more, since no more loader vs .o version skew.
Test: TreeHugger
Flag: EXEMPT mainline
Signed-off-by: Maciej Żenczykowski <maze@google.com>
Change-Id: I6c0b9ff2683809f3e732ec950bc91cf87040642f
Maciej Żenczykowskicommited 2025-07-15 14:46:10 -07:00
For libbpf compatibility we will want to have a 'normal' looking
.maps section & an android-only .android_maps section.
Test: TreeHugger
Flag: EXEMPT mainline
Signed-off-by: Maciej Żenczykowski <maze@google.com>
Change-Id: I9114d27c408cb084626ec1151826ae722797296f
Maciej Żenczykowskicommited 2025-07-15 14:43:49 -07:00
This is not needed, as this is now simply hard coded (along with the
.o names themselves) inside of NetBpfLoad.cpp itself - see loadAllObjects().
Even if it wasn't, we're better off using the per-map & per-program
limits instead. These will already cause all maps/programs to be skipped...
(The only reason to have a global per .o one was to solve
.o vs loader version skew issues, which we can no longer have)
Test: TreeHugger
Flag: EXEMPT mainline
Signed-off-by: Maciej Żenczykowski <maze@google.com>
Change-Id: I8e8f62ed3534d54dd3382df53637f76bb72dab5e
Maciej Żenczykowskicommited 2025-07-15 14:40:11 -07:00
Flag: EXEMPT mainline
Test: Boot on 4.14 and BTF is disabled
Test: Boot on 4.19 and BTF is enabled
Bug: 429278733
Change-Id: Icdb85891c3d70f1f1949c5b6f5f82ed6bb862873
This is an array of two ints:
- index 0 should have value 0
(a value of 1 implies a kernel uapi bug)
- index 1 should have value 123
(a value of 0 implies bpf array writes at non-zero index fail)
Note that these bugs can only be present on Android S & Sv2,
as these kernel bugs are considered fatal on Android T+.
(btw. we need to do this write manually as we cannot write into
maps we pin and then reopen - due to sepolicy)
Test: TreeHugger
Flag: EXEMPT mainline
Signed-off-by: Maciej Żenczykowski <maze@google.com>
Change-Id: I9b2ada7f3e299014d39461a8d0718396daf0c91a
Maciej Żenczykowskicommited 2025-06-17 01:49:23 -07:00
See https://b.corp.google.com/issues/414570528#comment96
which can be summarized as
https://github.com/MiCode/Xiaomi_Kernel_OpenSource/blob/vayu-r-oss/include/uapi/linux/bpf.h has:
enum bpf_cmd {
BPF_MAP_CREATE, <-- UAPI 0
BPF_MAP_LOOKUP_ELEM, <-- UAPI 1
BPF_MAP_UPDATE_ELEM, <-- UAPI 2
BPF_MAP_DELETE_ELEM, <-- UAPI 3
BPF_MAP_GET_NEXT_KEY, <-- UAPI 4
BPF_PROG_LOAD, <-- UAPI 5
BPF_OBJ_PIN, <-- UAPI 6
BPF_OBJ_GET, <-- UAPI 7
BPF_PROG_ATTACH, <-- UAPI 8
BPF_PROG_DETACH, <-- UAPI 9
BPF_GET_COMM_HASH, <-- breaks Linux uapi (uses 10)
BPF_PROG_TEST_RUN, <-- UAPI 10 becomes 11 (aka. BPF_PROG_RUN)
BPF_PROG_GET_NEXT_ID, <-- UAPI 11 becomes 12
BPF_MAP_GET_NEXT_ID, <-- UAPI 12 becomes 13
BPF_PROG_GET_FD_BY_ID, <-- UAPI 13 becomes 14
BPF_MAP_GET_FD_BY_ID, <-- UAPI 14 becomes 15
BPF_OBJ_GET_INFO_BY_FD, <-- UAPI 15 becomes 16
};
this addition in the middle of the userspace api enum,
causes all following enum values to be greater by one.
In particular this increases the value of
BPF_OBJ_GET_INFO_BY_FD from 15 to 16, thus break it.
It also causes UAPI 12 (BPF_MAP_GET_NEXT_ID)
to actually execute BPF_PROG_GET_NEXT_ID (normally UAPI 13)
As such this can be detected via finding the index of
the last map (on buggy kernels this will actually find
the index of the last program), then creating a map,
then checking if the last map index has changed
(on buggy kernels this will not be the case because
we did not create another program).
Bug: 414570528
Test: TreeHugger
Flag: EXEMPT mainline
Change-Id: Iecad121ac624df2d123a13e90a46792bd567a4c6
Signed-off-by: Maciej Żenczykowski <maze@google.com>
Maciej Żenczykowskicommited 2025-06-10 02:43:28 -07:00
We sometimes see issues when trying to load BPF maps, it's helpful to know
what errno was set.
Bug: 414570528
Test: TreeHugger
Flag: EXEMPT mainline
Signed-off-by: Maciej Żenczykowski <maze@google.com>
Change-Id: I8b78ee17ebad076d073d50bfd0f01440fe7187a2
Maciej Żenczykowskicommited 2025-05-29 01:25:26 -07:00
`ignore_on_*` directives haven't been used in a while and require a special version of bpfloader. Remove them.
While doing so, pad the `map_def` and `prog_def` structs, to maintain existing size.
Note: The mainline shipped bpf .o files are now in a location
that no platform bpfloader ever looked/looks at.
As such maintaining .o format compatibility with older release
is not needed.
Test: TreeHugger
Flag: EXEMPT dead code removal / mainline
Signed-off-by: Maciej Żenczykowski <maze@google.com>
Change-Id: I485c9645bdece2ad27add08347688d0f1c93e5d2
Maciej Żenczykowskicommited 2025-05-20 15:02:15 -07:00
should be unused
Test: TreeHugger
Flag: EXEMPT dead code
Signed-off-by: Maciej Żenczykowski <maze@google.com>
Change-Id: Ice72d940dd3492d2b9e14dd32390a49f87bd54d1
Maciej Żenczykowskicommited 2025-05-20 08:45:15 -07:00
This is to identify if it actually happens.
It is expected not to happen, because we now pin every map and are run only once on boot.
Test: TreeHugger
Flag: EXEMPT dead code
Signed-off-by: Maciej Żenczykowski <maze@google.com>
Change-Id: Ie46ace40f20fb4b061a91cd53f8256caebdfa71f
Maciej Żenczykowskicommited 2025-05-20 06:41:18 -07:00
This code should be dead. Abort to see whether it is.
Test: TreeHugger
Flag: EXEMPT dead code
Signed-off-by: Maciej Żenczykowski <maze@google.com>
Change-Id: Ib5309ff683339e70bc44e135578ede75548b022d
Maciej Żenczykowskicommited 2025-05-20 06:38:57 -07:00
This is basically a no-op as 32-bit netd already fails on android-mainline @ 6.13 (we're now at 6.14, 6.15 is coming in a few weeks).
I know this because we had to fix ACK 6.12, and didn't port it forward.
Note: AFAIK TV is planning to release on 25Q2 and then 27Q2,
so this 26Q2+ doesn't actually affect them till a year later in 27Q2.
Modern arm cores are dropping 32-bit cpu instruction decode support.
This already happened on Pixel 8+ (Tensor G3+) and is soon coming
to the GCE cloud machines underpinning our cuttlefish VMs
(this is very much an industry wide trend for modern aarch64 cores,
I believe all new application cores since 2023 have been 64-bit only).
Pixel 6a is AFAIK the last pixel with 32-bit application support,
and that is reaching EOL approximately in July 2027.
(while most test coverage is on Pixel 6/6pro which EOLs in Oct 2026)
Note though that this isn't a great example: AFAIK all Pixels
have had 64-bit system userspace (which is what we care about here)
The transition happened in the final Nexus, the year before Pixel 1.
This means we're going to start losing *any* remaining vestiges
of test coverage that we still (might) have.
Test: TreeHugger
Flag: EXEMPT mainline
Signed-off-by: Maciej Żenczykowski <maze@google.com>
Change-Id: Ie8a5a5dafd2f0e20ef4651c3a69957fde374ae4b
Maciej Żenczykowskicommited 2025-05-17 06:00:03 -07:00
The full set of 5 .o's is known at build time, and basically almost
never changes (and indeed is likely to shrink, with clatd.o merged
into offload.o && dscpPolicy.c merged into netd.c).
We might as well just code this knowledge straight into the mainline
loader.
This also allows us to move all the files to an apex subdirectory
that old platform bpfloaders don't even look at, which eliminates
the need for any backward compatibility with them.
Test: TreeHugger
Flag: EXEMPT mainline
Signed-off-by: Maciej Żenczykowski <maze@google.com>
Change-Id: Ie3d27178c31a68ed7cf5173b5983ad3009a3cf94
Maciej Żenczykowskicommited 2025-05-15 01:38:16 -07:00
Treehugger Robot
authored and
Android (Google) Code Review
commited 
Maciej Żenczykowski
authored and
Motomu Utsumi
commited 
Elliott Hughes
commited