tequilaOS/platform_system_bpf
2
0
Fork 0
Code Issues Pull requests Projects Releases Packages Wiki Activity

bpfloader: add back support for 4.9-T kernels

Browse source 
This effectively reverts aosp/2268766

Test: TreeHugger
Signed-off-by: Maciej Żenczykowski <maze@google.com>
Change-Id: I42da5c56fe4b69635b9678ff57ade7788c17f7c8
This commit is contained in:
Maciej Żenczykowski 2024-03-11 12:40:06 +00:00 committed by zlewchan
parent 9bbad68230
commit d18d86dfe9
1 changed files with 18 additions and 0 deletions
Show stats Download patch file Download diff file Expand all files Collapse all files

18
loader/Loader.cpp
View file

@ -483,6 +483,14 @@ static int getSymNameByIdx(ifstream& elfFile, int index, string& name) {
static bool mapMatchesExpectations(const unique_fd& fd, const string& mapName, static bool mapMatchesExpectations(const unique_fd& fd, const string& mapName,
const struct bpf_map_def& mapDef, const enum bpf_map_type type) { const struct bpf_map_def& mapDef, const enum bpf_map_type type) {
// bpfGetFd... family of functions require at minimum a 4.14 kernel,
// so on 4.9-T kernels just pretend the map matches our expectations.
// Additionally we'll get almost equivalent test coverage on newer devices/kernels.
// This is because the primary failure mode we're trying to detect here
// is either a source code misconfiguration (which is likely kernel independent)
// or a newly introduced kernel feature/bug (which is unlikely to get backported to 4.9).
if (!isAtLeastKernelVersion(4, 14, 0)) return true;
// Assuming fd is a valid Bpf Map file descriptor then // Assuming fd is a valid Bpf Map file descriptor then
// all the following should always succeed on a 4.14+ kernel. // all the following should always succeed on a 4.14+ kernel.
// If they somehow do fail, they'll return -1 (and set errno), // If they somehow do fail, they'll return -1 (and set errno),
@ -572,6 +580,16 @@ static int createMaps(const char* elfPath, ifstream& elfFile, vector<unique_fd>&
} }
enum bpf_map_type type = md[i].type; enum bpf_map_type type = md[i].type;
if (type == BPF_MAP_TYPE_DEVMAP && !isAtLeastKernelVersion(4, 14, 0)) {
// On Linux Kernels older than 4.14 this map type doesn't exist, but it can kind
// of be approximated: ARRAY has the same userspace api, though it is not usable
// by the same ebpf programs. However, that's okay because the bpf_redirect_map()
// helper doesn't exist on 4.9-T anyway (so the bpf program would fail to load,
// and thus needs to be tagged as 4.14+ either way), so there's nothing useful you
// could do with a DEVMAP anyway (that isn't already provided by an ARRAY)...
// Hence using an ARRAY instead of a DEVMAP simply makes life easier for userspace.
type = BPF_MAP_TYPE_ARRAY;
}
if (type == BPF_MAP_TYPE_DEVMAP_HASH && !isAtLeastKernelVersion(5, 4, 0)) { if (type == BPF_MAP_TYPE_DEVMAP_HASH && !isAtLeastKernelVersion(5, 4, 0)) {
// On Linux Kernels older than 5.4 this map type doesn't exist, but it can kind // On Linux Kernels older than 5.4 this map type doesn't exist, but it can kind
// of be approximated: HASH has the same userspace visible api. // of be approximated: HASH has the same userspace visible api.