On Mon, Jun 27, 2022 at 09:40:52PM -0300, Jason Gunthorpe wrote: > On Mon, Jun 27, 2022 at 08:27:37PM +0200, Daniel Borkmann wrote: > > [...] > > Fyi, this breaks BPF CI: > > > > https://github.com/kernel-patches/bpf/runs/7078719372?check_suite_focus=true > > > > [...] > > progs/map_ptr_kern.c:314:26: error: field 'trie_key' with variable sized type 'struct bpf_lpm_trie_key' not at the end of a struct or class is a GNU extension [-Werror,-Wgnu-variable-sized-type-not-at-end] > > struct bpf_lpm_trie_key trie_key; > > ^ The issue here seems to be a collision between "unknown array size" and known sizes: struct bpf_lpm_trie_key { __u32 prefixlen; /* up to 32 for AF_INET, 128 for AF_INET6 */ __u8 data[0]; /* Arbitrary size */ }; struct lpm_key { struct bpf_lpm_trie_key trie_key; __u32 data; }; This is treating trie_key as a header, which it's not: it's a complete structure. :) Perhaps: struct lpm_key { __u32 prefixlen; __u32 data; }; I don't see anything else trying to include bpf_lpm_trie_key. > > This will break the rdma-core userspace as well, with a similar > error: > > /usr/bin/clang-13 -DVERBS_DEBUG -Dibverbs_EXPORTS -Iinclude -I/usr/include/libnl3 -I/usr/include/drm -g -O2 -fdebug-prefix-map=/__w/1/s=. -fstack-protector-strong -Wformat -Werror=format-security -Wdate-time -D_FORTIFY_SOURCE=2 -Wmissing-prototypes -Wmissing-declarations -Wwrite-strings -Wformat=2 -Wcast-function-type -Wformat-nonliteral -Wdate-time -Wnested-externs -Wshadow -Wstrict-prototypes -Wold-style-definition -Werror -Wredundant-decls -g -fPIC -std=gnu11 -MD -MT libibverbs/CMakeFiles/ibverbs.dir/cmd_flow.c.o -MF libibverbs/CMakeFiles/ibverbs.dir/cmd_flow.c.o.d -o libibverbs/CMakeFiles/ibverbs.dir/cmd_flow.c.o -c ../libibverbs/cmd_flow.c > In file included from ../libibverbs/cmd_flow.c:33: > In file included from include/infiniband/cmd_write.h:36: > In file included from include/infiniband/cmd_ioctl.h:41: > In file included from include/infiniband/verbs.h:48: > In file included from include/infiniband/verbs_api.h:66: > In file included from include/infiniband/ib_user_ioctl_verbs.h:38: > include/rdma/ib_user_verbs.h:436:34: error: field 'base' with variable sized type 'struct ib_uverbs_create_cq_resp' not at the end of a struct or class is a GNU extension [-Werror,-Wgnu-variable-sized-type-not-at-end] > struct ib_uverbs_create_cq_resp base; > ^ > include/rdma/ib_user_verbs.h:644:34: error: field 'base' with variable sized type 'struct ib_uverbs_create_qp_resp' not at the end of a struct or class is a GNU extension [-Werror,-Wgnu-variable-sized-type-not-at-end] > struct ib_uverbs_create_qp_resp base; This looks very similar, a struct of unknown size is being treated as a header struct: struct ib_uverbs_create_cq_resp { __u32 cq_handle; __u32 cqe; __aligned_u64 driver_data[0]; }; struct ib_uverbs_ex_create_cq_resp { struct ib_uverbs_create_cq_resp base; __u32 comp_mask; __u32 response_length; }; And it only gets used here: DECLARE_UVERBS_WRITE(IB_USER_VERBS_CMD_CREATE_CQ, ib_uverbs_create_cq, UAPI_DEF_WRITE_UDATA_IO( struct ib_uverbs_create_cq, struct ib_uverbs_create_cq_resp), ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ UAPI_DEF_METHOD_NEEDS_FN(create_cq)), which must also be assuming it's a header. So probably better to just drop the driver_data field? I don't see anything using it (that I can find) besides as a sanity-check that the field exists and is at the end of the struct. -- Kees Cook
