diff options
Diffstat (limited to 'test/syscalls')
55 files changed, 5789 insertions, 1531 deletions
diff --git a/test/syscalls/BUILD b/test/syscalls/BUILD index 9800a0cdf..c19b30b4a 100644 --- a/test/syscalls/BUILD +++ b/test/syscalls/BUILD @@ -2,47 +2,78 @@ load("//test/runner:defs.bzl", "syscall_test") package(licenses = ["notice"]) -syscall_test(test = "//test/syscalls/linux:32bit_test") +syscall_test( + test = "//test/syscalls/linux:32bit_test", + vfs2 = "True", +) -syscall_test(test = "//test/syscalls/linux:accept_bind_stream_test") +syscall_test( + test = "//test/syscalls/linux:accept_bind_stream_test", + vfs2 = "True", +) syscall_test( size = "large", shard_count = 50, test = "//test/syscalls/linux:accept_bind_test", + vfs2 = "True", ) syscall_test( add_overlay = True, test = "//test/syscalls/linux:access_test", + vfs2 = "True", ) -syscall_test(test = "//test/syscalls/linux:affinity_test") +syscall_test( + test = "//test/syscalls/linux:affinity_test", + vfs2 = "True", +) syscall_test( add_overlay = True, test = "//test/syscalls/linux:aio_test", + vfs2 = "True", ) syscall_test( size = "medium", shard_count = 5, test = "//test/syscalls/linux:alarm_test", + vfs2 = "True", ) -syscall_test(test = "//test/syscalls/linux:arch_prctl_test") +syscall_test( + test = "//test/syscalls/linux:arch_prctl_test", + vfs2 = "True", +) -syscall_test(test = "//test/syscalls/linux:bad_test") +syscall_test( + test = "//test/syscalls/linux:bad_test", + vfs2 = "True", +) syscall_test( size = "large", add_overlay = True, test = "//test/syscalls/linux:bind_test", + vfs2 = "True", +) + +syscall_test( + test = "//test/syscalls/linux:brk_test", + vfs2 = "True", ) -syscall_test(test = "//test/syscalls/linux:brk_test") +syscall_test( + test = "//test/syscalls/linux:socket_test", + vfs2 = "True", +) -syscall_test(test = "//test/syscalls/linux:socket_test") +syscall_test( + test = "//test/syscalls/linux:socket_capability_test", + vfs2 = "True", +) syscall_test( size = "large", @@ -51,16 +82,19 @@ syscall_test( # involve much concurrency, TSAN's usefulness here is limited anyway. tags = ["nogotsan"], test = "//test/syscalls/linux:socket_stress_test", + vfs2 = "True", ) syscall_test( add_overlay = True, test = "//test/syscalls/linux:chdir_test", + vfs2 = "True", ) syscall_test( add_overlay = True, test = "//test/syscalls/linux:chmod_test", + vfs2 = "True", ) syscall_test( @@ -68,77 +102,116 @@ syscall_test( add_overlay = True, test = "//test/syscalls/linux:chown_test", use_tmpfs = True, # chwon tests require gofer to be running as root. + vfs2 = "True", ) syscall_test( add_overlay = True, test = "//test/syscalls/linux:chroot_test", + vfs2 = "True", ) -syscall_test(test = "//test/syscalls/linux:clock_getres_test") +syscall_test( + test = "//test/syscalls/linux:clock_getres_test", + vfs2 = "True", +) syscall_test( size = "medium", test = "//test/syscalls/linux:clock_gettime_test", + vfs2 = "True", ) -syscall_test(test = "//test/syscalls/linux:clock_nanosleep_test") +syscall_test( + test = "//test/syscalls/linux:clock_nanosleep_test", + vfs2 = "True", +) -syscall_test(test = "//test/syscalls/linux:concurrency_test") +syscall_test( + test = "//test/syscalls/linux:concurrency_test", + vfs2 = "True", +) syscall_test( add_uds_tree = True, test = "//test/syscalls/linux:connect_external_test", use_tmpfs = True, + vfs2 = "True", ) syscall_test( add_overlay = True, test = "//test/syscalls/linux:creat_test", + vfs2 = "True", ) -syscall_test(test = "//test/syscalls/linux:dev_test") +syscall_test( + fuse = "True", + test = "//test/syscalls/linux:dev_test", + vfs2 = "True", +) syscall_test( add_overlay = True, test = "//test/syscalls/linux:dup_test", + vfs2 = "True", ) -syscall_test(test = "//test/syscalls/linux:epoll_test") +syscall_test( + test = "//test/syscalls/linux:epoll_test", + vfs2 = "True", +) -syscall_test(test = "//test/syscalls/linux:eventfd_test") +syscall_test( + test = "//test/syscalls/linux:eventfd_test", + vfs2 = "True", +) -syscall_test(test = "//test/syscalls/linux:exceptions_test") +syscall_test( + test = "//test/syscalls/linux:exceptions_test", + vfs2 = "True", +) syscall_test( size = "medium", add_overlay = True, test = "//test/syscalls/linux:exec_test", + vfs2 = "True", ) syscall_test( size = "medium", add_overlay = True, test = "//test/syscalls/linux:exec_binary_test", + vfs2 = "True", ) -syscall_test(test = "//test/syscalls/linux:exit_test") +syscall_test( + test = "//test/syscalls/linux:exit_test", + vfs2 = "True", +) syscall_test( add_overlay = True, test = "//test/syscalls/linux:fadvise64_test", + vfs2 = "True", ) syscall_test( add_overlay = True, test = "//test/syscalls/linux:fallocate_test", + vfs2 = "True", ) -syscall_test(test = "//test/syscalls/linux:fault_test") +syscall_test( + test = "//test/syscalls/linux:fault_test", + vfs2 = "True", +) syscall_test( add_overlay = True, test = "//test/syscalls/linux:fchdir_test", + vfs2 = "True", ) syscall_test( @@ -150,326 +223,505 @@ syscall_test( size = "medium", add_overlay = True, test = "//test/syscalls/linux:flock_test", + vfs2 = "True", ) -syscall_test(test = "//test/syscalls/linux:fork_test") +syscall_test( + test = "//test/syscalls/linux:fork_test", + vfs2 = "True", +) -syscall_test(test = "//test/syscalls/linux:fpsig_fork_test") +syscall_test( + test = "//test/syscalls/linux:fpsig_fork_test", + vfs2 = "True", +) -syscall_test(test = "//test/syscalls/linux:fpsig_nested_test") +syscall_test( + test = "//test/syscalls/linux:fpsig_nested_test", + vfs2 = "True", +) syscall_test( add_overlay = True, test = "//test/syscalls/linux:fsync_test", + vfs2 = "True", ) syscall_test( size = "medium", shard_count = 5, test = "//test/syscalls/linux:futex_test", + vfs2 = "True", ) -syscall_test(test = "//test/syscalls/linux:getcpu_host_test") +syscall_test( + test = "//test/syscalls/linux:getcpu_host_test", + vfs2 = "True", +) -syscall_test(test = "//test/syscalls/linux:getcpu_test") +syscall_test( + test = "//test/syscalls/linux:getcpu_test", + vfs2 = "True", +) syscall_test( add_overlay = True, test = "//test/syscalls/linux:getdents_test", + vfs2 = "True", ) -syscall_test(test = "//test/syscalls/linux:getrandom_test") +syscall_test( + test = "//test/syscalls/linux:getrandom_test", + vfs2 = "True", +) -syscall_test(test = "//test/syscalls/linux:getrusage_test") +syscall_test( + test = "//test/syscalls/linux:getrusage_test", + vfs2 = "True", +) syscall_test( size = "medium", add_overlay = False, # TODO(gvisor.dev/issue/317): enable when fixed. test = "//test/syscalls/linux:inotify_test", + vfs2 = "True", ) syscall_test( size = "medium", add_overlay = True, test = "//test/syscalls/linux:ioctl_test", + vfs2 = "True", ) syscall_test( test = "//test/syscalls/linux:iptables_test", + vfs2 = "True", ) syscall_test( size = "large", shard_count = 5, test = "//test/syscalls/linux:itimer_test", + vfs2 = "True", ) -syscall_test(test = "//test/syscalls/linux:kill_test") +syscall_test( + test = "//test/syscalls/linux:kill_test", + vfs2 = "True", +) syscall_test( add_overlay = True, test = "//test/syscalls/linux:link_test", use_tmpfs = True, # gofer needs CAP_DAC_READ_SEARCH to use AT_EMPTY_PATH with linkat(2) + vfs2 = "True", ) syscall_test( add_overlay = True, test = "//test/syscalls/linux:lseek_test", + vfs2 = "True", ) -syscall_test(test = "//test/syscalls/linux:madvise_test") +syscall_test( + test = "//test/syscalls/linux:madvise_test", + vfs2 = "True", +) -syscall_test(test = "//test/syscalls/linux:memory_accounting_test") +syscall_test( + test = "//test/syscalls/linux:memory_accounting_test", + vfs2 = "True", +) -syscall_test(test = "//test/syscalls/linux:mempolicy_test") +syscall_test( + test = "//test/syscalls/linux:mempolicy_test", + vfs2 = "True", +) -syscall_test(test = "//test/syscalls/linux:mincore_test") +syscall_test( + test = "//test/syscalls/linux:mincore_test", + vfs2 = "True", +) syscall_test( add_overlay = True, test = "//test/syscalls/linux:mkdir_test", + vfs2 = "True", ) syscall_test( add_overlay = True, test = "//test/syscalls/linux:mknod_test", + vfs2 = "True", ) syscall_test( size = "medium", shard_count = 5, test = "//test/syscalls/linux:mmap_test", + vfs2 = "True", ) syscall_test( add_overlay = True, test = "//test/syscalls/linux:mount_test", + vfs2 = "True", ) syscall_test( size = "medium", test = "//test/syscalls/linux:mremap_test", + vfs2 = "True", ) syscall_test( size = "medium", test = "//test/syscalls/linux:msync_test", + vfs2 = "True", ) -syscall_test(test = "//test/syscalls/linux:munmap_test") +syscall_test( + test = "//test/syscalls/linux:munmap_test", + vfs2 = "True", +) -syscall_test(test = "//test/syscalls/linux:network_namespace_test") +syscall_test( + test = "//test/syscalls/linux:network_namespace_test", + vfs2 = "True", +) syscall_test( add_overlay = True, test = "//test/syscalls/linux:open_create_test", + vfs2 = "True", ) syscall_test( add_overlay = True, test = "//test/syscalls/linux:open_test", + vfs2 = "True", ) -syscall_test(test = "//test/syscalls/linux:packet_socket_raw_test") +syscall_test( + test = "//test/syscalls/linux:packet_socket_raw_test", + vfs2 = "True", +) -syscall_test(test = "//test/syscalls/linux:packet_socket_test") +syscall_test( + test = "//test/syscalls/linux:packet_socket_test", + vfs2 = "True", +) -syscall_test(test = "//test/syscalls/linux:partial_bad_buffer_test") +syscall_test( + test = "//test/syscalls/linux:partial_bad_buffer_test", + vfs2 = "True", +) -syscall_test(test = "//test/syscalls/linux:pause_test") +syscall_test( + test = "//test/syscalls/linux:pause_test", + vfs2 = "True", +) + +syscall_test( + size = "medium", + # Takes too long under gotsan to run. + tags = ["nogotsan"], + test = "//test/syscalls/linux:ping_socket_test", + vfs2 = "True", +) syscall_test( size = "large", add_overlay = True, shard_count = 5, test = "//test/syscalls/linux:pipe_test", + vfs2 = "True", ) -syscall_test(test = "//test/syscalls/linux:poll_test") +syscall_test( + test = "//test/syscalls/linux:poll_test", + vfs2 = "True", +) syscall_test( size = "medium", test = "//test/syscalls/linux:ppoll_test", + vfs2 = "True", ) -syscall_test(test = "//test/syscalls/linux:prctl_setuid_test") +syscall_test( + test = "//test/syscalls/linux:prctl_setuid_test", + vfs2 = "True", +) -syscall_test(test = "//test/syscalls/linux:prctl_test") +syscall_test( + test = "//test/syscalls/linux:prctl_test", + vfs2 = "True", +) syscall_test( add_overlay = True, test = "//test/syscalls/linux:pread64_test", + vfs2 = "True", ) syscall_test( add_overlay = True, test = "//test/syscalls/linux:preadv_test", + vfs2 = "True", ) syscall_test( add_overlay = True, test = "//test/syscalls/linux:preadv2_test", + vfs2 = "True", ) -syscall_test(test = "//test/syscalls/linux:priority_test") +syscall_test( + test = "//test/syscalls/linux:priority_test", + vfs2 = "True", +) syscall_test( size = "medium", test = "//test/syscalls/linux:proc_test", + vfs2 = "True", ) -syscall_test(test = "//test/syscalls/linux:proc_net_test") +syscall_test( + test = "//test/syscalls/linux:proc_net_test", + vfs2 = "True", +) -syscall_test(test = "//test/syscalls/linux:proc_pid_oomscore_test") +syscall_test( + test = "//test/syscalls/linux:proc_pid_oomscore_test", + vfs2 = "True", +) -syscall_test(test = "//test/syscalls/linux:proc_pid_smaps_test") +syscall_test( + test = "//test/syscalls/linux:proc_pid_smaps_test", + vfs2 = "True", +) -syscall_test(test = "//test/syscalls/linux:proc_pid_uid_gid_map_test") +syscall_test( + test = "//test/syscalls/linux:proc_pid_uid_gid_map_test", + vfs2 = "True", +) syscall_test( size = "medium", test = "//test/syscalls/linux:pselect_test", + vfs2 = "True", ) -syscall_test(test = "//test/syscalls/linux:ptrace_test") +syscall_test( + test = "//test/syscalls/linux:ptrace_test", + vfs2 = "True", +) syscall_test( size = "medium", shard_count = 5, test = "//test/syscalls/linux:pty_test", + vfs2 = "True", ) syscall_test( test = "//test/syscalls/linux:pty_root_test", + vfs2 = "True", ) syscall_test( add_overlay = True, test = "//test/syscalls/linux:pwritev2_test", + vfs2 = "True", ) syscall_test( add_overlay = True, test = "//test/syscalls/linux:pwrite64_test", + vfs2 = "True", ) -syscall_test(test = "//test/syscalls/linux:raw_socket_hdrincl_test") +syscall_test( + test = "//test/syscalls/linux:raw_socket_hdrincl_test", + vfs2 = "True", +) -syscall_test(test = "//test/syscalls/linux:raw_socket_icmp_test") +syscall_test( + test = "//test/syscalls/linux:raw_socket_icmp_test", + vfs2 = "True", +) -syscall_test(test = "//test/syscalls/linux:raw_socket_ipv4_test") +syscall_test( + test = "//test/syscalls/linux:raw_socket_test", + vfs2 = "True", +) syscall_test( add_overlay = True, test = "//test/syscalls/linux:read_test", + vfs2 = "True", ) syscall_test( add_overlay = True, test = "//test/syscalls/linux:readahead_test", + vfs2 = "True", ) syscall_test( size = "medium", shard_count = 5, test = "//test/syscalls/linux:readv_socket_test", + vfs2 = "True", ) syscall_test( size = "medium", add_overlay = True, test = "//test/syscalls/linux:readv_test", + vfs2 = "True", ) syscall_test( size = "medium", add_overlay = True, test = "//test/syscalls/linux:rename_test", + vfs2 = "True", ) -syscall_test(test = "//test/syscalls/linux:rlimits_test") +syscall_test( + test = "//test/syscalls/linux:rlimits_test", + vfs2 = "True", +) -syscall_test(test = "//test/syscalls/linux:rseq_test") +syscall_test( + test = "//test/syscalls/linux:rseq_test", + vfs2 = "True", +) -syscall_test(test = "//test/syscalls/linux:rtsignal_test") +syscall_test( + test = "//test/syscalls/linux:rtsignal_test", + vfs2 = "True", +) -syscall_test(test = "//test/syscalls/linux:signalfd_test") +syscall_test( + test = "//test/syscalls/linux:signalfd_test", + vfs2 = "True", +) -syscall_test(test = "//test/syscalls/linux:sched_test") +syscall_test( + test = "//test/syscalls/linux:sched_test", + vfs2 = "True", +) -syscall_test(test = "//test/syscalls/linux:sched_yield_test") +syscall_test( + test = "//test/syscalls/linux:sched_yield_test", + vfs2 = "True", +) -syscall_test(test = "//test/syscalls/linux:seccomp_test") +syscall_test( + test = "//test/syscalls/linux:seccomp_test", + vfs2 = "True", +) -syscall_test(test = "//test/syscalls/linux:select_test") +syscall_test( + test = "//test/syscalls/linux:select_test", + vfs2 = "True", +) syscall_test( shard_count = 20, test = "//test/syscalls/linux:semaphore_test", + vfs2 = "True", ) syscall_test( add_overlay = True, test = "//test/syscalls/linux:sendfile_socket_test", + vfs2 = "True", ) syscall_test( add_overlay = True, test = "//test/syscalls/linux:sendfile_test", + vfs2 = "True", ) syscall_test( add_overlay = True, test = "//test/syscalls/linux:splice_test", + vfs2 = "True", ) -syscall_test(test = "//test/syscalls/linux:sigaction_test") +syscall_test( + test = "//test/syscalls/linux:sigaction_test", + vfs2 = "True", +) # TODO(b/119826902): Enable once the test passes in runsc. -# syscall_test(test = "//test/syscalls/linux:sigaltstack_test") +# syscall_test(vfs2="True",test = "//test/syscalls/linux:sigaltstack_test") -syscall_test(test = "//test/syscalls/linux:sigiret_test") +syscall_test( + test = "//test/syscalls/linux:sigiret_test", + vfs2 = "True", +) -syscall_test(test = "//test/syscalls/linux:sigprocmask_test") +syscall_test( + test = "//test/syscalls/linux:sigprocmask_test", + vfs2 = "True", +) syscall_test( size = "medium", test = "//test/syscalls/linux:sigstop_test", + vfs2 = "True", ) -syscall_test(test = "//test/syscalls/linux:sigtimedwait_test") +syscall_test( + test = "//test/syscalls/linux:sigtimedwait_test", + vfs2 = "True", +) syscall_test( size = "medium", test = "//test/syscalls/linux:shm_test", + vfs2 = "True", ) syscall_test( size = "medium", test = "//test/syscalls/linux:socket_abstract_non_blocking_test", + vfs2 = "True", ) syscall_test( size = "large", shard_count = 50, test = "//test/syscalls/linux:socket_abstract_test", + vfs2 = "True", ) syscall_test( size = "medium", test = "//test/syscalls/linux:socket_domain_non_blocking_test", + vfs2 = "True", ) syscall_test( size = "large", shard_count = 50, test = "//test/syscalls/linux:socket_domain_test", + vfs2 = "True", ) syscall_test( size = "medium", add_overlay = True, test = "//test/syscalls/linux:socket_filesystem_non_blocking_test", + vfs2 = "True", ) syscall_test( @@ -477,89 +729,138 @@ syscall_test( add_overlay = True, shard_count = 50, test = "//test/syscalls/linux:socket_filesystem_test", + vfs2 = "True", ) syscall_test( size = "large", shard_count = 50, test = "//test/syscalls/linux:socket_inet_loopback_test", + vfs2 = "True", +) + +syscall_test( + size = "large", + shard_count = 50, + # Takes too long for TSAN. Creates a lot of TCP sockets. + tags = ["nogotsan"], + test = "//test/syscalls/linux:socket_inet_loopback_nogotsan_test", + vfs2 = "True", ) syscall_test( size = "large", shard_count = 50, test = "//test/syscalls/linux:socket_ip_tcp_generic_loopback_test", + vfs2 = "True", ) syscall_test( size = "medium", test = "//test/syscalls/linux:socket_ip_tcp_loopback_non_blocking_test", + vfs2 = "True", ) syscall_test( size = "large", shard_count = 50, test = "//test/syscalls/linux:socket_ip_tcp_loopback_test", + vfs2 = "True", ) syscall_test( size = "medium", shard_count = 50, test = "//test/syscalls/linux:socket_ip_tcp_udp_generic_loopback_test", + vfs2 = "True", ) syscall_test( size = "medium", test = "//test/syscalls/linux:socket_ip_udp_loopback_non_blocking_test", + vfs2 = "True", ) syscall_test( size = "large", shard_count = 50, test = "//test/syscalls/linux:socket_ip_udp_loopback_test", + vfs2 = "True", ) syscall_test( size = "medium", test = "//test/syscalls/linux:socket_ipv4_udp_unbound_loopback_test", + vfs2 = "True", ) -syscall_test(test = "//test/syscalls/linux:socket_ip_unbound_test") +syscall_test( + test = "//test/syscalls/linux:socket_ip_unbound_test", + vfs2 = "True", +) -syscall_test(test = "//test/syscalls/linux:socket_netdevice_test") +syscall_test( + test = "//test/syscalls/linux:socket_netdevice_test", + vfs2 = "True", +) -syscall_test(test = "//test/syscalls/linux:socket_netlink_test") +syscall_test( + test = "//test/syscalls/linux:socket_netlink_test", + vfs2 = "True", +) -syscall_test(test = "//test/syscalls/linux:socket_netlink_route_test") +syscall_test( + test = "//test/syscalls/linux:socket_netlink_route_test", + vfs2 = "True", +) -syscall_test(test = "//test/syscalls/linux:socket_netlink_uevent_test") +syscall_test( + test = "//test/syscalls/linux:socket_netlink_uevent_test", + vfs2 = "True", +) -syscall_test(test = "//test/syscalls/linux:socket_blocking_local_test") +syscall_test( + test = "//test/syscalls/linux:socket_blocking_local_test", + vfs2 = "True", +) -syscall_test(test = "//test/syscalls/linux:socket_blocking_ip_test") +syscall_test( + test = "//test/syscalls/linux:socket_blocking_ip_test", + vfs2 = "True", +) -syscall_test(test = "//test/syscalls/linux:socket_non_stream_blocking_local_test") +syscall_test( + test = "//test/syscalls/linux:socket_non_stream_blocking_local_test", + vfs2 = "True", +) -syscall_test(test = "//test/syscalls/linux:socket_non_stream_blocking_udp_test") +syscall_test( + test = "//test/syscalls/linux:socket_non_stream_blocking_udp_test", + vfs2 = "True", +) syscall_test( size = "large", test = "//test/syscalls/linux:socket_stream_blocking_local_test", + vfs2 = "True", ) syscall_test( size = "large", test = "//test/syscalls/linux:socket_stream_blocking_tcp_test", + vfs2 = "True", ) syscall_test( size = "medium", test = "//test/syscalls/linux:socket_stream_local_test", + vfs2 = "True", ) syscall_test( size = "medium", test = "//test/syscalls/linux:socket_stream_nonblock_local_test", + vfs2 = "True", ) syscall_test( @@ -567,11 +868,13 @@ syscall_test( size = "enormous", shard_count = 5, test = "//test/syscalls/linux:socket_unix_dgram_local_test", + vfs2 = "True", ) syscall_test( size = "medium", test = "//test/syscalls/linux:socket_unix_dgram_non_blocking_test", + vfs2 = "True", ) syscall_test( @@ -579,6 +882,7 @@ syscall_test( add_overlay = True, shard_count = 50, test = "//test/syscalls/linux:socket_unix_pair_test", + vfs2 = "True", ) syscall_test( @@ -586,155 +890,236 @@ syscall_test( size = "enormous", shard_count = 5, test = "//test/syscalls/linux:socket_unix_seqpacket_local_test", + vfs2 = "True", ) syscall_test( size = "medium", test = "//test/syscalls/linux:socket_unix_stream_test", + vfs2 = "True", ) syscall_test( size = "medium", test = "//test/syscalls/linux:socket_unix_unbound_abstract_test", + vfs2 = "True", ) syscall_test( size = "medium", test = "//test/syscalls/linux:socket_unix_unbound_dgram_test", + vfs2 = "True", ) syscall_test( size = "medium", test = "//test/syscalls/linux:socket_unix_unbound_filesystem_test", + vfs2 = "True", ) syscall_test( size = "medium", shard_count = 10, test = "//test/syscalls/linux:socket_unix_unbound_seqpacket_test", + vfs2 = "True", ) syscall_test( size = "large", shard_count = 50, test = "//test/syscalls/linux:socket_unix_unbound_stream_test", + vfs2 = "True", ) syscall_test( add_overlay = True, test = "//test/syscalls/linux:statfs_test", + vfs2 = "True", ) syscall_test( add_overlay = True, test = "//test/syscalls/linux:stat_test", + vfs2 = "True", ) syscall_test( add_overlay = True, test = "//test/syscalls/linux:stat_times_test", + vfs2 = "True", ) syscall_test( add_overlay = True, test = "//test/syscalls/linux:sticky_test", + vfs2 = "True", ) syscall_test( add_overlay = True, test = "//test/syscalls/linux:symlink_test", + vfs2 = "True", ) syscall_test( add_overlay = True, test = "//test/syscalls/linux:sync_test", + vfs2 = "True", ) syscall_test( add_overlay = True, test = "//test/syscalls/linux:sync_file_range_test", + vfs2 = "True", ) -syscall_test(test = "//test/syscalls/linux:sysinfo_test") +syscall_test( + test = "//test/syscalls/linux:sysinfo_test", + vfs2 = "True", +) -syscall_test(test = "//test/syscalls/linux:syslog_test") +syscall_test( + test = "//test/syscalls/linux:syslog_test", + vfs2 = "True", +) -syscall_test(test = "//test/syscalls/linux:sysret_test") +syscall_test( + test = "//test/syscalls/linux:sysret_test", + vfs2 = "True", +) syscall_test( size = "medium", shard_count = 10, test = "//test/syscalls/linux:tcp_socket_test", + vfs2 = "True", ) -syscall_test(test = "//test/syscalls/linux:tgkill_test") +syscall_test( + test = "//test/syscalls/linux:tgkill_test", + vfs2 = "True", +) -syscall_test(test = "//test/syscalls/linux:timerfd_test") +syscall_test( + test = "//test/syscalls/linux:timerfd_test", + vfs2 = "True", +) -syscall_test(test = "//test/syscalls/linux:timers_test") +syscall_test( + test = "//test/syscalls/linux:timers_test", + vfs2 = "True", +) -syscall_test(test = "//test/syscalls/linux:time_test") +syscall_test( + test = "//test/syscalls/linux:time_test", + vfs2 = "True", +) -syscall_test(test = "//test/syscalls/linux:tkill_test") +syscall_test( + test = "//test/syscalls/linux:tkill_test", + vfs2 = "True", +) syscall_test( add_overlay = True, test = "//test/syscalls/linux:truncate_test", ) -syscall_test(test = "//test/syscalls/linux:tuntap_test") +syscall_test( + test = "//test/syscalls/linux:tuntap_test", + vfs2 = "True", +) syscall_test( add_hostinet = True, test = "//test/syscalls/linux:tuntap_hostinet_test", + vfs2 = "True", ) -syscall_test(test = "//test/syscalls/linux:udp_bind_test") +syscall_test( + test = "//test/syscalls/linux:udp_bind_test", + vfs2 = "True", +) syscall_test( size = "medium", add_hostinet = True, shard_count = 10, test = "//test/syscalls/linux:udp_socket_test", + vfs2 = "True", ) -syscall_test(test = "//test/syscalls/linux:uidgid_test") +syscall_test( + test = "//test/syscalls/linux:uidgid_test", + vfs2 = "True", +) -syscall_test(test = "//test/syscalls/linux:uname_test") +syscall_test( + test = "//test/syscalls/linux:uname_test", + vfs2 = "True", +) syscall_test( add_overlay = True, test = "//test/syscalls/linux:unlink_test", + vfs2 = "True", ) -syscall_test(test = "//test/syscalls/linux:unshare_test") +syscall_test( + test = "//test/syscalls/linux:unshare_test", + vfs2 = "True", +) -syscall_test(test = "//test/syscalls/linux:utimes_test") +syscall_test( + test = "//test/syscalls/linux:utimes_test", + vfs2 = "True", +) syscall_test( size = "medium", test = "//test/syscalls/linux:vdso_clock_gettime_test", + vfs2 = "True", ) -syscall_test(test = "//test/syscalls/linux:vdso_test") +syscall_test( + test = "//test/syscalls/linux:vdso_test", + vfs2 = "True", +) -syscall_test(test = "//test/syscalls/linux:vsyscall_test") +syscall_test( + test = "//test/syscalls/linux:vsyscall_test", + vfs2 = "True", +) -syscall_test(test = "//test/syscalls/linux:vfork_test") +syscall_test( + test = "//test/syscalls/linux:vfork_test", + vfs2 = "True", +) syscall_test( size = "medium", shard_count = 5, test = "//test/syscalls/linux:wait_test", + vfs2 = "True", ) syscall_test( add_overlay = True, test = "//test/syscalls/linux:write_test", + vfs2 = "True", ) -syscall_test(test = "//test/syscalls/linux:proc_net_unix_test") +syscall_test( + test = "//test/syscalls/linux:proc_net_unix_test", + vfs2 = "True", +) -syscall_test(test = "//test/syscalls/linux:proc_net_tcp_test") +syscall_test( + test = "//test/syscalls/linux:proc_net_tcp_test", + vfs2 = "True", +) -syscall_test(test = "//test/syscalls/linux:proc_net_udp_test") +syscall_test( + test = "//test/syscalls/linux:proc_net_udp_test", + vfs2 = "True", +) diff --git a/test/syscalls/linux/BUILD b/test/syscalls/linux/BUILD index 5acdb8438..66a31cd28 100644 --- a/test/syscalls/linux/BUILD +++ b/test/syscalls/linux/BUILD @@ -363,8 +363,8 @@ cc_binary( linkstatic = 1, deps = [ ":socket_test_util", - "//test/util:file_descriptor", gtest, + "//test/util:file_descriptor", "//test/util:temp_umask", "//test/util:test_main", "//test/util:test_util", @@ -372,6 +372,21 @@ cc_binary( ) cc_binary( + name = "socket_capability_test", + testonly = 1, + srcs = ["socket_capability.cc"], + linkstatic = 1, + deps = [ + ":socket_test_util", + "//test/util:capability_util", + "//test/util:file_descriptor", + gtest, + "//test/util:test_main", + "//test/util:test_util", + ], +) + +cc_binary( name = "brk_test", testonly = 1, srcs = ["brk.cc"], @@ -733,9 +748,14 @@ cc_binary( linkstatic = 1, deps = [ ":file_base", + ":socket_test_util", "//test/util:cleanup", + "//test/util:eventfd_util", "//test/util:file_descriptor", + "@com_google_absl//absl/strings", + "@com_google_absl//absl/time", gtest, + "//test/util:posix_error", "//test/util:temp_path", "//test/util:test_main", "//test/util:test_util", @@ -776,6 +796,7 @@ cc_binary( deps = [ ":socket_test_util", "//test/util:cleanup", + "//test/util:epoll_util", "//test/util:eventfd_util", "//test/util:fs_util", "@com_google_absl//absl/base:core_headers", @@ -789,6 +810,7 @@ cc_binary( "//test/util:save_util", "//test/util:temp_path", "//test/util:test_util", + "//test/util:thread_util", "//test/util:timer_util", ], ) @@ -802,10 +824,13 @@ cc_binary( ], linkstatic = 1, deps = [ + ":socket_test_util", "//test/util:file_descriptor", "@com_google_absl//absl/strings", "@com_google_absl//absl/time", gtest, + "//test/util:epoll_util", + "//test/util:eventfd_util", "//test/util:posix_error", "//test/util:temp_path", "//test/util:test_main", @@ -849,10 +874,7 @@ cc_binary( cc_binary( name = "fpsig_nested_test", testonly = 1, - srcs = select_arch( - amd64 = ["fpsig_nested.cc"], - arm64 = [], - ), + srcs = ["fpsig_nested.cc"], linkstatic = 1, deps = [ gtest, @@ -921,6 +943,7 @@ cc_binary( "//test/util:eventfd_util", "//test/util:file_descriptor", "//test/util:fs_util", + "@com_google_absl//absl/container:node_hash_set", "@com_google_absl//absl/strings", gtest, "//test/util:posix_error", @@ -951,12 +974,14 @@ cc_binary( "//test/util:epoll_util", "//test/util:file_descriptor", "//test/util:fs_util", + "//test/util:posix_error", "//test/util:temp_path", "//test/util:test_main", "//test/util:test_util", "//test/util:thread_util", "@com_google_absl//absl/strings", "@com_google_absl//absl/strings:str_format", + "@com_google_absl//absl/synchronization", "@com_google_absl//absl/time", ], ) @@ -1306,6 +1331,7 @@ cc_binary( name = "packet_socket_raw_test", testonly = 1, srcs = ["packet_socket_raw.cc"], + defines = select_system(), linkstatic = 1, deps = [ ":socket_test_util", @@ -1382,7 +1408,7 @@ cc_binary( srcs = ["partial_bad_buffer.cc"], linkstatic = 1, deps = [ - "//test/syscalls/linux:socket_test_util", + ":socket_test_util", "//test/util:file_descriptor", "//test/util:fs_util", "@com_google_absl//absl/time", @@ -1411,6 +1437,21 @@ cc_binary( ) cc_binary( + name = "ping_socket_test", + testonly = 1, + srcs = ["ping_socket.cc"], + linkstatic = 1, + deps = [ + ":socket_test_util", + "//test/util:file_descriptor", + gtest, + "//test/util:save_util", + "//test/util:test_main", + "//test/util:test_util", + ], +) + +cc_binary( name = "pipe_test", testonly = 1, srcs = ["pipe.cc"], @@ -1767,9 +1808,10 @@ cc_binary( ) cc_binary( - name = "raw_socket_ipv4_test", + name = "raw_socket_test", testonly = 1, - srcs = ["raw_socket_ipv4.cc"], + srcs = ["raw_socket.cc"], + defines = select_system(), linkstatic = 1, deps = [ ":socket_test_util", @@ -2780,6 +2822,26 @@ cc_binary( ) cc_binary( + name = "socket_inet_loopback_nogotsan_test", + testonly = 1, + srcs = ["socket_inet_loopback_nogotsan.cc"], + linkstatic = 1, + deps = [ + ":ip_socket_test_util", + ":socket_test_util", + "//test/util:file_descriptor", + "@com_google_absl//absl/memory", + "@com_google_absl//absl/strings", + gtest, + "//test/util:posix_error", + "//test/util:save_util", + "//test/util:test_main", + "//test/util:test_util", + "//test/util:thread_util", + ], +) + +cc_binary( name = "socket_netlink_test", testonly = 1, srcs = ["socket_netlink.cc"], @@ -3348,6 +3410,7 @@ cc_binary( name = "tcp_socket_test", testonly = 1, srcs = ["tcp_socket.cc"], + defines = select_system(), linkstatic = 1, deps = [ ":socket_test_util", @@ -3461,7 +3524,7 @@ cc_binary( deps = [ ":socket_test_util", gtest, - "//test/syscalls/linux:socket_netlink_route_util", + ":socket_netlink_route_util", "//test/util:capability_util", "//test/util:file_descriptor", "//test/util:fs_util", @@ -3494,11 +3557,15 @@ cc_library( hdrs = ["udp_socket_test_cases.h"], defines = select_system(), deps = [ + ":ip_socket_test_util", ":socket_test_util", ":unix_domain_socket_test_util", "@com_google_absl//absl/base:core_headers", + "@com_google_absl//absl/strings:str_format", "@com_google_absl//absl/time", gtest, + "//test/util:file_descriptor", + "//test/util:posix_error", "//test/util:test_main", "//test/util:test_util", "//test/util:thread_util", diff --git a/test/syscalls/linux/accept_bind.cc b/test/syscalls/linux/accept_bind.cc index e08c578f0..f65a14fb8 100644 --- a/test/syscalls/linux/accept_bind.cc +++ b/test/syscalls/linux/accept_bind.cc @@ -13,6 +13,7 @@ // limitations under the License. #include <stdio.h> +#include <sys/socket.h> #include <sys/un.h> #include <algorithm> @@ -141,6 +142,47 @@ TEST_P(AllSocketPairTest, Connect) { SyscallSucceeds()); } +TEST_P(AllSocketPairTest, ConnectWithWrongType) { + auto sockets = ASSERT_NO_ERRNO_AND_VALUE(NewSocketPair()); + + int type; + socklen_t typelen = sizeof(type); + EXPECT_THAT( + getsockopt(sockets->first_fd(), SOL_SOCKET, SO_TYPE, &type, &typelen), + SyscallSucceeds()); + switch (type) { + case SOCK_STREAM: + type = SOCK_SEQPACKET; + break; + case SOCK_SEQPACKET: + type = SOCK_STREAM; + break; + } + + const FileDescriptor another_socket = + ASSERT_NO_ERRNO_AND_VALUE(Socket(AF_UNIX, type, 0)); + + ASSERT_THAT(bind(sockets->first_fd(), sockets->first_addr(), + sockets->first_addr_size()), + SyscallSucceeds()); + + ASSERT_THAT(listen(sockets->first_fd(), 5), SyscallSucceeds()); + + if (sockets->first_addr()->sa_data[0] != 0) { + ASSERT_THAT(connect(another_socket.get(), sockets->first_addr(), + sockets->first_addr_size()), + SyscallFailsWithErrno(EPROTOTYPE)); + } else { + ASSERT_THAT(connect(another_socket.get(), sockets->first_addr(), + sockets->first_addr_size()), + SyscallFailsWithErrno(ECONNREFUSED)); + } + + ASSERT_THAT(connect(sockets->second_fd(), sockets->first_addr(), + sockets->first_addr_size()), + SyscallSucceeds()); +} + TEST_P(AllSocketPairTest, ConnectNonListening) { auto sockets = ASSERT_NO_ERRNO_AND_VALUE(NewSocketPair()); diff --git a/test/syscalls/linux/dev.cc b/test/syscalls/linux/dev.cc index 4dd302eed..1d0d584cd 100644 --- a/test/syscalls/linux/dev.cc +++ b/test/syscalls/linux/dev.cc @@ -153,6 +153,27 @@ TEST(DevTest, TTYExists) { EXPECT_EQ(statbuf.st_mode, S_IFCHR | 0666); } +TEST(DevTest, OpenDevFuse) { + // Note(gvisor.dev/issue/3076) This won't work in the sentry until the new + // device registration is complete. + SKIP_IF(IsRunningWithVFS1() || IsRunningOnGvisor() || !IsFUSEEnabled()); + + ASSERT_NO_ERRNO_AND_VALUE(Open("/dev/fuse", O_RDONLY)); +} + +TEST(DevTest, ReadDevFuseWithoutMount) { + // Note(gvisor.dev/issue/3076) This won't work in the sentry until the new + // device registration is complete. + SKIP_IF(IsRunningWithVFS1() || IsRunningOnGvisor()); + + const FileDescriptor fd = + ASSERT_NO_ERRNO_AND_VALUE(Open("/dev/fuse", O_RDONLY)); + + std::vector<char> buf(1); + EXPECT_THAT(ReadFd(fd.get(), buf.data(), sizeof(buf)), + SyscallFailsWithErrno(EPERM)); +} + } // namespace } // namespace testing diff --git a/test/syscalls/linux/epoll.cc b/test/syscalls/linux/epoll.cc index f57d38dc7..2101e5c9f 100644 --- a/test/syscalls/linux/epoll.cc +++ b/test/syscalls/linux/epoll.cc @@ -422,6 +422,28 @@ TEST(EpollTest, CloseFile) { SyscallSucceedsWithValue(0)); } +TEST(EpollTest, PipeReaderHupAfterWriterClosed) { + auto epollfd = ASSERT_NO_ERRNO_AND_VALUE(NewEpollFD()); + int pipefds[2]; + ASSERT_THAT(pipe(pipefds), SyscallSucceeds()); + FileDescriptor rfd(pipefds[0]); + FileDescriptor wfd(pipefds[1]); + + ASSERT_NO_ERRNO(RegisterEpollFD(epollfd.get(), rfd.get(), 0, kMagicConstant)); + struct epoll_event result[kFDsPerEpoll]; + // Initially, rfd should not generate any events of interest. + ASSERT_THAT(epoll_wait(epollfd.get(), result, kFDsPerEpoll, 0), + SyscallSucceedsWithValue(0)); + // Close the write end of the pipe. + wfd.reset(); + // rfd should now generate EPOLLHUP, which EPOLL_CTL_ADD unconditionally adds + // to the set of events of interest. + ASSERT_THAT(epoll_wait(epollfd.get(), result, kFDsPerEpoll, 0), + SyscallSucceedsWithValue(1)); + EXPECT_EQ(result[0].events, EPOLLHUP); + EXPECT_EQ(result[0].data.u64, kMagicConstant); +} + } // namespace } // namespace testing diff --git a/test/syscalls/linux/eventfd.cc b/test/syscalls/linux/eventfd.cc index 548b05a64..dc794415e 100644 --- a/test/syscalls/linux/eventfd.cc +++ b/test/syscalls/linux/eventfd.cc @@ -100,20 +100,21 @@ TEST(EventfdTest, SmallRead) { ASSERT_THAT(read(efd.get(), &l, 4), SyscallFailsWithErrno(EINVAL)); } -TEST(EventfdTest, PreadIllegalSeek) { - FileDescriptor efd = - ASSERT_NO_ERRNO_AND_VALUE(NewEventFD(0, EFD_NONBLOCK | EFD_SEMAPHORE)); - - uint64_t l = 0; - ASSERT_THAT(pread(efd.get(), &l, 4, 0), SyscallFailsWithErrno(ESPIPE)); +TEST(EventfdTest, IllegalSeek) { + FileDescriptor efd = ASSERT_NO_ERRNO_AND_VALUE(NewEventFD(0, 0)); + EXPECT_THAT(lseek(efd.get(), 0, SEEK_SET), SyscallFailsWithErrno(ESPIPE)); } -TEST(EventfdTest, PwriteIllegalSeek) { - FileDescriptor efd = - ASSERT_NO_ERRNO_AND_VALUE(NewEventFD(0, EFD_NONBLOCK | EFD_SEMAPHORE)); +TEST(EventfdTest, IllegalPread) { + FileDescriptor efd = ASSERT_NO_ERRNO_AND_VALUE(NewEventFD(0, 0)); + int l; + EXPECT_THAT(pread(efd.get(), &l, sizeof(l), 0), + SyscallFailsWithErrno(ESPIPE)); +} - uint64_t l = 0; - ASSERT_THAT(pwrite(efd.get(), &l, 4, 0), SyscallFailsWithErrno(ESPIPE)); +TEST(EventfdTest, IllegalPwrite) { + FileDescriptor efd = ASSERT_NO_ERRNO_AND_VALUE(NewEventFD(0, 0)); + EXPECT_THAT(pwrite(efd.get(), "x", 1, 0), SyscallFailsWithErrno(ESPIPE)); } TEST(EventfdTest, BigWrite) { diff --git a/test/syscalls/linux/exec.cc b/test/syscalls/linux/exec.cc index 12c9b05ca..c5acfc794 100644 --- a/test/syscalls/linux/exec.cc +++ b/test/syscalls/linux/exec.cc @@ -553,7 +553,12 @@ TEST(ExecTest, SymlinkLimitRefreshedForInterpreter) { // Hold onto TempPath objects so they are not destructed prematurely. std::vector<TempPath> interpreter_symlinks; std::vector<TempPath> script_symlinks; - for (int i = 0; i < kLinuxMaxSymlinks; i++) { + // Replace both the interpreter and script paths with symlink chains of just + // over half the symlink limit each; this is the minimum required to test that + // the symlink limit applies separately to each traversal, while tolerating + // some symlinks in the resolution of (the original) interpreter_path and + // script_path. + for (int i = 0; i < (kLinuxMaxSymlinks / 2) + 1; i++) { interpreter_symlinks.push_back(ASSERT_NO_ERRNO_AND_VALUE( TempPath::CreateSymlinkTo(tmp_dir, interpreter_path))); interpreter_path = interpreter_symlinks[i].path(); @@ -673,6 +678,31 @@ TEST(ExecveatTest, SymlinkNoFollowWithRelativePath) { EXPECT_EQ(execve_errno, ELOOP); } +TEST(ExecveatTest, UnshareFiles) { + TempPath tempFile = ASSERT_NO_ERRNO_AND_VALUE( + TempPath::CreateFileWith(GetAbsoluteTestTmpdir(), "bar", 0755)); + const FileDescriptor fd_closed_on_exec = + ASSERT_NO_ERRNO_AND_VALUE(Open(tempFile.path(), O_RDONLY | O_CLOEXEC)); + + ExecveArray argv = {"test"}; + ExecveArray envp; + std::string child_path = RunfilePath(kBasicWorkload); + pid_t child = + syscall(__NR_clone, SIGCHLD | CLONE_VFORK | CLONE_FILES, 0, 0, 0, 0); + if (child == 0) { + execve(child_path.c_str(), argv.get(), envp.get()); + _exit(1); + } + ASSERT_THAT(child, SyscallSucceeds()); + + int status; + ASSERT_THAT(RetryEINTR(waitpid)(child, &status, 0), SyscallSucceeds()); + EXPECT_EQ(status, 0); + + struct stat st; + EXPECT_THAT(fstat(fd_closed_on_exec.get(), &st), SyscallSucceeds()); +} + TEST(ExecveatTest, SymlinkNoFollowWithAbsolutePath) { std::string parent_dir = "/tmp"; TempPath link = ASSERT_NO_ERRNO_AND_VALUE( diff --git a/test/syscalls/linux/exec_binary.cc b/test/syscalls/linux/exec_binary.cc index 1a9f203b9..18d2f22c1 100644 --- a/test/syscalls/linux/exec_binary.cc +++ b/test/syscalls/linux/exec_binary.cc @@ -438,7 +438,12 @@ TEST(ElfTest, MissingText) { ASSERT_THAT(RetryEINTR(waitpid)(child, &status, 0), SyscallSucceedsWithValue(child)); // It runs off the end of the zeroes filling the end of the page. +#if defined(__x86_64__) EXPECT_TRUE(WIFSIGNALED(status) && WTERMSIG(status) == SIGSEGV) << status; +#elif defined(__aarch64__) + // 0 is an invalid instruction opcode on arm64. + EXPECT_TRUE(WIFSIGNALED(status) && WTERMSIG(status) == SIGILL) << status; +#endif } // Typical ELF with a data + bss segment diff --git a/test/syscalls/linux/fallocate.cc b/test/syscalls/linux/fallocate.cc index 7819f4ac3..cabc2b751 100644 --- a/test/syscalls/linux/fallocate.cc +++ b/test/syscalls/linux/fallocate.cc @@ -15,16 +15,27 @@ #include <errno.h> #include <fcntl.h> #include <signal.h> +#include <sys/eventfd.h> #include <sys/resource.h> +#include <sys/signalfd.h> +#include <sys/socket.h> #include <sys/stat.h> +#include <sys/timerfd.h> #include <syscall.h> #include <time.h> #include <unistd.h> +#include <ctime> + #include "gtest/gtest.h" +#include "absl/strings/str_cat.h" +#include "absl/time/time.h" #include "test/syscalls/linux/file_base.h" +#include "test/syscalls/linux/socket_test_util.h" #include "test/util/cleanup.h" +#include "test/util/eventfd_util.h" #include "test/util/file_descriptor.h" +#include "test/util/posix_error.h" #include "test/util/temp_path.h" #include "test/util/test_util.h" @@ -70,6 +81,12 @@ TEST_F(AllocateTest, Fallocate) { ASSERT_THAT(fallocate(test_file_fd_.get(), 0, 39, 1), SyscallSucceeds()); ASSERT_THAT(fstat(test_file_fd_.get(), &buf), SyscallSucceeds()); EXPECT_EQ(buf.st_size, 40); + + // Given length 0 should fail with EINVAL. + ASSERT_THAT(fallocate(test_file_fd_.get(), 0, 50, 0), + SyscallFailsWithErrno(EINVAL)); + ASSERT_THAT(fstat(test_file_fd_.get(), &buf), SyscallSucceeds()); + EXPECT_EQ(buf.st_size, 40); } TEST_F(AllocateTest, FallocateInvalid) { @@ -136,6 +153,34 @@ TEST_F(AllocateTest, FallocateRlimit) { ASSERT_THAT(sigprocmask(SIG_UNBLOCK, &new_mask, nullptr), SyscallSucceeds()); } +TEST_F(AllocateTest, FallocateOtherFDs) { + int fd; + ASSERT_THAT(fd = timerfd_create(CLOCK_MONOTONIC, 0), SyscallSucceeds()); + auto timer_fd = FileDescriptor(fd); + EXPECT_THAT(fallocate(timer_fd.get(), 0, 0, 10), + SyscallFailsWithErrno(ENODEV)); + + sigset_t mask; + sigemptyset(&mask); + ASSERT_THAT(fd = signalfd(-1, &mask, 0), SyscallSucceeds()); + auto sfd = FileDescriptor(fd); + EXPECT_THAT(fallocate(sfd.get(), 0, 0, 10), SyscallFailsWithErrno(ENODEV)); + + auto efd = + ASSERT_NO_ERRNO_AND_VALUE(NewEventFD(0, EFD_NONBLOCK | EFD_SEMAPHORE)); + EXPECT_THAT(fallocate(efd.get(), 0, 0, 10), SyscallFailsWithErrno(ENODEV)); + + auto sockfd = ASSERT_NO_ERRNO_AND_VALUE(Socket(AF_INET, SOCK_DGRAM, 0)); + EXPECT_THAT(fallocate(sockfd.get(), 0, 0, 10), SyscallFailsWithErrno(ENODEV)); + + int socks[2]; + ASSERT_THAT(socketpair(AF_UNIX, SOCK_STREAM, PF_UNIX, socks), + SyscallSucceeds()); + auto sock0 = FileDescriptor(socks[0]); + auto sock1 = FileDescriptor(socks[1]); + EXPECT_THAT(fallocate(sock0.get(), 0, 0, 10), SyscallFailsWithErrno(ENODEV)); +} + } // namespace } // namespace testing } // namespace gvisor diff --git a/test/syscalls/linux/fcntl.cc b/test/syscalls/linux/fcntl.cc index c7cc5816e..34016d4bd 100644 --- a/test/syscalls/linux/fcntl.cc +++ b/test/syscalls/linux/fcntl.cc @@ -18,7 +18,10 @@ #include <syscall.h> #include <unistd.h> +#include <iostream> +#include <list> #include <string> +#include <vector> #include "gtest/gtest.h" #include "absl/base/macros.h" @@ -37,6 +40,7 @@ #include "test/util/save_util.h" #include "test/util/temp_path.h" #include "test/util/test_util.h" +#include "test/util/thread_util.h" #include "test/util/timer_util.h" ABSL_FLAG(std::string, child_setlock_on, "", @@ -115,6 +119,15 @@ PosixErrorOr<Cleanup> SubprocessLock(std::string const& path, bool for_write, return std::move(cleanup); } +TEST(FcntlTest, SetCloExecBadFD) { + // Open an eventfd file descriptor with FD_CLOEXEC descriptor flag not set. + FileDescriptor f = ASSERT_NO_ERRNO_AND_VALUE(NewEventFD(0, 0)); + auto fd = f.get(); + f.reset(); + ASSERT_THAT(fcntl(fd, F_GETFD), SyscallFailsWithErrno(EBADF)); + ASSERT_THAT(fcntl(fd, F_SETFD, FD_CLOEXEC), SyscallFailsWithErrno(EBADF)); +} + TEST(FcntlTest, SetCloExec) { // Open an eventfd file descriptor with FD_CLOEXEC descriptor flag not set. FileDescriptor fd = ASSERT_NO_ERRNO_AND_VALUE(NewEventFD(0, 0)); @@ -182,45 +195,85 @@ TEST(FcntlTest, SetFlags) { EXPECT_EQ(rflags, expected); } -TEST_F(FcntlLockTest, SetLockBadFd) { +void TestLock(int fd, short lock_type = F_RDLCK) { // NOLINT, type in flock struct flock fl; - fl.l_type = F_WRLCK; + fl.l_type = lock_type; fl.l_whence = SEEK_SET; fl.l_start = 0; - // len 0 has a special meaning: lock all bytes despite how - // large the file grows. + // len 0 locks all bytes despite how large the file grows. fl.l_len = 0; - EXPECT_THAT(fcntl(-1, F_SETLK, &fl), SyscallFailsWithErrno(EBADF)); + EXPECT_THAT(fcntl(fd, F_SETLK, &fl), SyscallSucceeds()); } -TEST_F(FcntlLockTest, SetLockPipe) { - int fds[2]; - ASSERT_THAT(pipe(fds), SyscallSucceeds()); - +void TestLockBadFD(int fd, + short lock_type = F_RDLCK) { // NOLINT, type in flock struct flock fl; - fl.l_type = F_WRLCK; + fl.l_type = lock_type; fl.l_whence = SEEK_SET; fl.l_start = 0; - // Same as SetLockBadFd, but doesn't matter, we expect this to fail. + // len 0 locks all bytes despite how large the file grows. fl.l_len = 0; - EXPECT_THAT(fcntl(fds[0], F_SETLK, &fl), SyscallFailsWithErrno(EBADF)); - EXPECT_THAT(close(fds[0]), SyscallSucceeds()); - EXPECT_THAT(close(fds[1]), SyscallSucceeds()); + EXPECT_THAT(fcntl(fd, F_SETLK, &fl), SyscallFailsWithErrno(EBADF)); } +TEST_F(FcntlLockTest, SetLockBadFd) { TestLockBadFD(-1); } + TEST_F(FcntlLockTest, SetLockDir) { auto dir = ASSERT_NO_ERRNO_AND_VALUE(TempPath::CreateDir()); - FileDescriptor fd = - ASSERT_NO_ERRNO_AND_VALUE(Open(dir.path(), O_RDONLY, 0666)); + auto fd = ASSERT_NO_ERRNO_AND_VALUE(Open(dir.path(), O_RDONLY, 0000)); + TestLock(fd.get()); +} - struct flock fl; - fl.l_type = F_RDLCK; - fl.l_whence = SEEK_SET; - fl.l_start = 0; - // Same as SetLockBadFd. - fl.l_len = 0; +TEST_F(FcntlLockTest, SetLockSymlink) { + // TODO(gvisor.dev/issue/2782): Replace with IsRunningWithVFS1() when O_PATH + // is supported. + SKIP_IF(IsRunningOnGvisor()); - EXPECT_THAT(fcntl(fd.get(), F_SETLK, &fl), SyscallSucceeds()); + auto file = ASSERT_NO_ERRNO_AND_VALUE(TempPath::CreateFile()); + auto symlink = ASSERT_NO_ERRNO_AND_VALUE( + TempPath::CreateSymlinkTo(GetAbsoluteTestTmpdir(), file.path())); + + auto fd = + ASSERT_NO_ERRNO_AND_VALUE(Open(symlink.path(), O_RDONLY | O_PATH, 0000)); + TestLockBadFD(fd.get()); +} + +TEST_F(FcntlLockTest, SetLockProc) { + auto fd = + ASSERT_NO_ERRNO_AND_VALUE(Open("/proc/self/status", O_RDONLY, 0000)); + TestLock(fd.get()); +} + +TEST_F(FcntlLockTest, SetLockPipe) { + SKIP_IF(IsRunningWithVFS1()); + + int fds[2]; + ASSERT_THAT(pipe(fds), SyscallSucceeds()); + + TestLock(fds[0]); + TestLockBadFD(fds[0], F_WRLCK); + + TestLock(fds[1], F_WRLCK); + TestLockBadFD(fds[1]); + + EXPECT_THAT(close(fds[0]), SyscallSucceeds()); + EXPECT_THAT(close(fds[1]), SyscallSucceeds()); +} + +TEST_F(FcntlLockTest, SetLockSocket) { + SKIP_IF(IsRunningWithVFS1()); + + int sock = socket(AF_UNIX, SOCK_STREAM, 0); + ASSERT_THAT(sock, SyscallSucceeds()); + + struct sockaddr_un addr = + ASSERT_NO_ERRNO_AND_VALUE(UniqueUnixAddr(true /* abstract */, AF_UNIX)); + ASSERT_THAT( + bind(sock, reinterpret_cast<struct sockaddr*>(&addr), sizeof(addr)), + SyscallSucceeds()); + + TestLock(sock); + EXPECT_THAT(close(sock), SyscallSucceeds()); } TEST_F(FcntlLockTest, SetLockBadOpenFlagsWrite) { @@ -232,8 +285,7 @@ TEST_F(FcntlLockTest, SetLockBadOpenFlagsWrite) { fl0.l_type = F_WRLCK; fl0.l_whence = SEEK_SET; fl0.l_start = 0; - // Same as SetLockBadFd. - fl0.l_len = 0; + fl0.l_len = 0; // Lock all file // Expect that setting a write lock using a read only file descriptor // won't work. @@ -695,7 +747,7 @@ TEST_F(FcntlLockTest, SetWriteLockThenBlockingWriteLock) { << "Exited with code: " << status; } -// This test will veirfy that blocking works as expected when another process +// This test will verify that blocking works as expected when another process // holds a read lock when obtaining a write lock. This test will hold the lock // for some amount of time and then wait for the second process to send over the // socket_fd the amount of time it was blocked for before the lock succeeded. @@ -905,15 +957,18 @@ TEST(FcntlTest, DupAfterO_ASYNC) { EXPECT_EQ(after & O_ASYNC, O_ASYNC); } -TEST(FcntlTest, GetOwn) { +TEST(FcntlTest, GetOwnNone) { FileDescriptor s = ASSERT_NO_ERRNO_AND_VALUE( Socket(AF_UNIX, SOCK_SEQPACKET | SOCK_NONBLOCK | SOCK_CLOEXEC, 0)); - EXPECT_EQ(syscall(__NR_fcntl, s.get(), F_GETOWN), 0); + // Use the raw syscall because the glibc wrapper may convert F_{GET,SET}OWN + // into F_{GET,SET}OWN_EX. + EXPECT_THAT(syscall(__NR_fcntl, s.get(), F_GETOWN), + SyscallSucceedsWithValue(0)); MaybeSave(); } -TEST(FcntlTest, GetOwnEx) { +TEST(FcntlTest, GetOwnExNone) { FileDescriptor s = ASSERT_NO_ERRNO_AND_VALUE( Socket(AF_UNIX, SOCK_SEQPACKET | SOCK_NONBLOCK | SOCK_CLOEXEC, 0)); @@ -922,6 +977,100 @@ TEST(FcntlTest, GetOwnEx) { SyscallSucceedsWithValue(0)); } +TEST(FcntlTest, SetOwnInvalidPid) { + SKIP_IF(IsRunningWithVFS1()); + + FileDescriptor s = ASSERT_NO_ERRNO_AND_VALUE( + Socket(AF_UNIX, SOCK_SEQPACKET | SOCK_NONBLOCK | SOCK_CLOEXEC, 0)); + + EXPECT_THAT(syscall(__NR_fcntl, s.get(), F_SETOWN, 12345678), + SyscallFailsWithErrno(ESRCH)); +} + +TEST(FcntlTest, SetOwnInvalidPgrp) { + SKIP_IF(IsRunningWithVFS1()); + + FileDescriptor s = ASSERT_NO_ERRNO_AND_VALUE( + Socket(AF_UNIX, SOCK_SEQPACKET | SOCK_NONBLOCK | SOCK_CLOEXEC, 0)); + + EXPECT_THAT(syscall(__NR_fcntl, s.get(), F_SETOWN, -12345678), + SyscallFailsWithErrno(ESRCH)); +} + +TEST(FcntlTest, SetOwnPid) { + FileDescriptor s = ASSERT_NO_ERRNO_AND_VALUE( + Socket(AF_UNIX, SOCK_SEQPACKET | SOCK_NONBLOCK | SOCK_CLOEXEC, 0)); + + pid_t pid; + EXPECT_THAT(pid = getpid(), SyscallSucceeds()); + + ASSERT_THAT(syscall(__NR_fcntl, s.get(), F_SETOWN, pid), + SyscallSucceedsWithValue(0)); + + EXPECT_THAT(syscall(__NR_fcntl, s.get(), F_GETOWN), + SyscallSucceedsWithValue(pid)); + MaybeSave(); +} + +TEST(FcntlTest, SetOwnPgrp) { + FileDescriptor s = ASSERT_NO_ERRNO_AND_VALUE( + Socket(AF_UNIX, SOCK_SEQPACKET | SOCK_NONBLOCK | SOCK_CLOEXEC, 0)); + + pid_t pgid; + EXPECT_THAT(pgid = getpgrp(), SyscallSucceeds()); + + ASSERT_THAT(syscall(__NR_fcntl, s.get(), F_SETOWN, -pgid), + SyscallSucceedsWithValue(0)); + + // Verify with F_GETOWN_EX; using F_GETOWN on Linux may incorrectly treat the + // negative return value as an error, converting the return value to -1 and + // setting errno accordingly. + f_owner_ex got_owner = {}; + ASSERT_THAT(syscall(__NR_fcntl, s.get(), F_GETOWN_EX, &got_owner), + SyscallSucceedsWithValue(0)); + EXPECT_EQ(got_owner.type, F_OWNER_PGRP); + EXPECT_EQ(got_owner.pid, pgid); + MaybeSave(); +} + +TEST(FcntlTest, SetOwnUnset) { + FileDescriptor s = ASSERT_NO_ERRNO_AND_VALUE( + Socket(AF_UNIX, SOCK_SEQPACKET | SOCK_NONBLOCK | SOCK_CLOEXEC, 0)); + + // Set and unset pid. + pid_t pid; + EXPECT_THAT(pid = getpid(), SyscallSucceeds()); + ASSERT_THAT(syscall(__NR_fcntl, s.get(), F_SETOWN, pid), + SyscallSucceedsWithValue(0)); + ASSERT_THAT(syscall(__NR_fcntl, s.get(), F_SETOWN, 0), + SyscallSucceedsWithValue(0)); + + EXPECT_THAT(syscall(__NR_fcntl, s.get(), F_GETOWN), + SyscallSucceedsWithValue(0)); + + // Set and unset pgid. + pid_t pgid; + EXPECT_THAT(pgid = getpgrp(), SyscallSucceeds()); + ASSERT_THAT(syscall(__NR_fcntl, s.get(), F_SETOWN, -pgid), + SyscallSucceedsWithValue(0)); + ASSERT_THAT(syscall(__NR_fcntl, s.get(), F_SETOWN, 0), + SyscallSucceedsWithValue(0)); + + EXPECT_THAT(syscall(__NR_fcntl, s.get(), F_GETOWN), + SyscallSucceedsWithValue(0)); + MaybeSave(); +} + +// F_SETOWN flips the sign of negative values, an operation that is guarded +// against overflow. +TEST(FcntlTest, SetOwnOverflow) { + FileDescriptor s = ASSERT_NO_ERRNO_AND_VALUE( + Socket(AF_UNIX, SOCK_SEQPACKET | SOCK_NONBLOCK | SOCK_CLOEXEC, 0)); + + EXPECT_THAT(syscall(__NR_fcntl, s.get(), F_SETOWN, INT_MIN), + SyscallFailsWithErrno(EINVAL)); +} + TEST(FcntlTest, SetOwnExInvalidType) { FileDescriptor s = ASSERT_NO_ERRNO_AND_VALUE( Socket(AF_UNIX, SOCK_SEQPACKET | SOCK_NONBLOCK | SOCK_CLOEXEC, 0)); @@ -977,9 +1126,10 @@ TEST(FcntlTest, SetOwnExTid) { EXPECT_THAT(owner.pid = syscall(__NR_gettid), SyscallSucceeds()); ASSERT_THAT(syscall(__NR_fcntl, s.get(), F_SETOWN_EX, &owner), - SyscallSucceeds()); + SyscallSucceedsWithValue(0)); - EXPECT_EQ(syscall(__NR_fcntl, s.get(), F_GETOWN), owner.pid); + EXPECT_THAT(syscall(__NR_fcntl, s.get(), F_GETOWN), + SyscallSucceedsWithValue(owner.pid)); MaybeSave(); } @@ -992,9 +1142,10 @@ TEST(FcntlTest, SetOwnExPid) { EXPECT_THAT(owner.pid = getpid(), SyscallSucceeds()); ASSERT_THAT(syscall(__NR_fcntl, s.get(), F_SETOWN_EX, &owner), - SyscallSucceeds()); + SyscallSucceedsWithValue(0)); - EXPECT_EQ(syscall(__NR_fcntl, s.get(), F_GETOWN), owner.pid); + EXPECT_THAT(syscall(__NR_fcntl, s.get(), F_GETOWN), + SyscallSucceedsWithValue(owner.pid)); MaybeSave(); } @@ -1002,18 +1153,54 @@ TEST(FcntlTest, SetOwnExPgrp) { FileDescriptor s = ASSERT_NO_ERRNO_AND_VALUE( Socket(AF_UNIX, SOCK_SEQPACKET | SOCK_NONBLOCK | SOCK_CLOEXEC, 0)); + f_owner_ex set_owner = {}; + set_owner.type = F_OWNER_PGRP; + EXPECT_THAT(set_owner.pid = getpgrp(), SyscallSucceeds()); + + ASSERT_THAT(syscall(__NR_fcntl, s.get(), F_SETOWN_EX, &set_owner), + SyscallSucceedsWithValue(0)); + + // Verify with F_GETOWN_EX; using F_GETOWN on Linux may incorrectly treat the + // negative return value as an error, converting the return value to -1 and + // setting errno accordingly. + f_owner_ex got_owner = {}; + ASSERT_THAT(syscall(__NR_fcntl, s.get(), F_GETOWN_EX, &got_owner), + SyscallSucceedsWithValue(0)); + EXPECT_EQ(got_owner.type, set_owner.type); + EXPECT_EQ(got_owner.pid, set_owner.pid); + MaybeSave(); +} + +TEST(FcntlTest, SetOwnExUnset) { + SKIP_IF(IsRunningWithVFS1()); + + FileDescriptor s = ASSERT_NO_ERRNO_AND_VALUE( + Socket(AF_UNIX, SOCK_SEQPACKET | SOCK_NONBLOCK | SOCK_CLOEXEC, 0)); + + // Set and unset pid. f_owner_ex owner = {}; + owner.type = F_OWNER_PID; + EXPECT_THAT(owner.pid = getpid(), SyscallSucceeds()); + ASSERT_THAT(syscall(__NR_fcntl, s.get(), F_SETOWN_EX, &owner), + SyscallSucceedsWithValue(0)); + owner.pid = 0; + ASSERT_THAT(syscall(__NR_fcntl, s.get(), F_SETOWN_EX, &owner), + SyscallSucceedsWithValue(0)); + + EXPECT_THAT(syscall(__NR_fcntl, s.get(), F_GETOWN), + SyscallSucceedsWithValue(0)); + + // Set and unset pgid. owner.type = F_OWNER_PGRP; EXPECT_THAT(owner.pid = getpgrp(), SyscallSucceeds()); - ASSERT_THAT(syscall(__NR_fcntl, s.get(), F_SETOWN_EX, &owner), - SyscallSucceeds()); - - // NOTE(igudger): I don't understand why, but this is flaky on Linux. - // GetOwnExPgrp (below) does not have this issue. - SKIP_IF(!IsRunningOnGvisor()); + SyscallSucceedsWithValue(0)); + owner.pid = 0; + ASSERT_THAT(syscall(__NR_fcntl, s.get(), F_SETOWN_EX, &owner), + SyscallSucceedsWithValue(0)); - EXPECT_EQ(syscall(__NR_fcntl, s.get(), F_GETOWN), -owner.pid); + EXPECT_THAT(syscall(__NR_fcntl, s.get(), F_GETOWN), + SyscallSucceedsWithValue(0)); MaybeSave(); } @@ -1026,7 +1213,7 @@ TEST(FcntlTest, GetOwnExTid) { EXPECT_THAT(set_owner.pid = syscall(__NR_gettid), SyscallSucceeds()); ASSERT_THAT(syscall(__NR_fcntl, s.get(), F_SETOWN_EX, &set_owner), - SyscallSucceeds()); + SyscallSucceedsWithValue(0)); f_owner_ex got_owner = {}; ASSERT_THAT(syscall(__NR_fcntl, s.get(), F_GETOWN_EX, &got_owner), @@ -1044,7 +1231,7 @@ TEST(FcntlTest, GetOwnExPid) { EXPECT_THAT(set_owner.pid = getpid(), SyscallSucceeds()); ASSERT_THAT(syscall(__NR_fcntl, s.get(), F_SETOWN_EX, &set_owner), - SyscallSucceeds()); + SyscallSucceedsWithValue(0)); f_owner_ex got_owner = {}; ASSERT_THAT(syscall(__NR_fcntl, s.get(), F_GETOWN_EX, &got_owner), @@ -1062,7 +1249,7 @@ TEST(FcntlTest, GetOwnExPgrp) { EXPECT_THAT(set_owner.pid = getpgrp(), SyscallSucceeds()); ASSERT_THAT(syscall(__NR_fcntl, s.get(), F_SETOWN_EX, &set_owner), - SyscallSucceeds()); + SyscallSucceedsWithValue(0)); f_owner_ex got_owner = {}; ASSERT_THAT(syscall(__NR_fcntl, s.get(), F_GETOWN_EX, &got_owner), @@ -1071,6 +1258,45 @@ TEST(FcntlTest, GetOwnExPgrp) { EXPECT_EQ(got_owner.pid, set_owner.pid); } +// Make sure that making multiple concurrent changes to async signal generation +// does not cause any race issues. +TEST(FcntlTest, SetFlSetOwnDoNotRace) { + FileDescriptor s = ASSERT_NO_ERRNO_AND_VALUE( + Socket(AF_UNIX, SOCK_SEQPACKET | SOCK_NONBLOCK | SOCK_CLOEXEC, 0)); + + pid_t pid; + EXPECT_THAT(pid = getpid(), SyscallSucceeds()); + + constexpr absl::Duration runtime = absl::Milliseconds(300); + auto setAsync = [&s, &runtime] { + for (auto start = absl::Now(); absl::Now() - start < runtime;) { + ASSERT_THAT(syscall(__NR_fcntl, s.get(), F_SETFL, O_ASYNC), + SyscallSucceeds()); + sched_yield(); + } + }; + auto resetAsync = [&s, &runtime] { + for (auto start = absl::Now(); absl::Now() - start < runtime;) { + ASSERT_THAT(syscall(__NR_fcntl, s.get(), F_SETFL, 0), SyscallSucceeds()); + sched_yield(); + } + }; + auto setOwn = [&s, &pid, &runtime] { + for (auto start = absl::Now(); absl::Now() - start < runtime;) { + ASSERT_THAT(syscall(__NR_fcntl, s.get(), F_SETOWN, pid), + SyscallSucceeds()); + sched_yield(); + } + }; + + std::list<ScopedThread> threads; + for (int i = 0; i < 10; i++) { + threads.emplace_back(setAsync); + threads.emplace_back(resetAsync); + threads.emplace_back(setOwn); + } +} + } // namespace } // namespace testing @@ -1100,8 +1326,7 @@ int main(int argc, char** argv) { fl.l_start = absl::GetFlag(FLAGS_child_setlock_start); fl.l_len = absl::GetFlag(FLAGS_child_setlock_len); - // Test the fcntl, no need to log, the error is unambiguously - // from fcntl at this point. + // Test the fcntl. int err = 0; int ret = 0; @@ -1114,6 +1339,8 @@ int main(int argc, char** argv) { if (ret == -1 && errno != 0) { err = errno; + std::cerr << "CHILD lock " << setlock_on << " failed " << err + << std::endl; } // If there is a socket fd let's send back the time in microseconds it took diff --git a/test/syscalls/linux/flock.cc b/test/syscalls/linux/flock.cc index 3ecb8db8e..638a93979 100644 --- a/test/syscalls/linux/flock.cc +++ b/test/syscalls/linux/flock.cc @@ -21,6 +21,7 @@ #include "absl/time/clock.h" #include "absl/time/time.h" #include "test/syscalls/linux/file_base.h" +#include "test/syscalls/linux/socket_test_util.h" #include "test/util/file_descriptor.h" #include "test/util/temp_path.h" #include "test/util/test_util.h" @@ -34,11 +35,6 @@ namespace { class FlockTest : public FileTest {}; -TEST_F(FlockTest, BadFD) { - // EBADF: fd is not an open file descriptor. - ASSERT_THAT(flock(-1, 0), SyscallFailsWithErrno(EBADF)); -} - TEST_F(FlockTest, InvalidOpCombinations) { // The operation cannot be both exclusive and shared. EXPECT_THAT(flock(test_file_fd_.get(), LOCK_EX | LOCK_SH | LOCK_NB), @@ -57,15 +53,6 @@ TEST_F(FlockTest, NoOperationSpecified) { SyscallFailsWithErrno(EINVAL)); } -TEST(FlockTestNoFixture, FlockSupportsPipes) { - int fds[2]; - ASSERT_THAT(pipe(fds), SyscallSucceeds()); - - EXPECT_THAT(flock(fds[0], LOCK_EX | LOCK_NB), SyscallSucceeds()); - EXPECT_THAT(close(fds[0]), SyscallSucceeds()); - EXPECT_THAT(close(fds[1]), SyscallSucceeds()); -} - TEST_F(FlockTest, TestSimpleExLock) { // Test that we can obtain an exclusive lock (no other holders) // and that we can unlock it. @@ -583,6 +570,66 @@ TEST_F(FlockTest, BlockingLockFirstExclusiveSecondExclusive_NoRandomSave) { EXPECT_THAT(flock(test_file_fd_.get(), LOCK_UN), SyscallSucceeds()); } +TEST(FlockTestNoFixture, BadFD) { + // EBADF: fd is not an open file descriptor. + ASSERT_THAT(flock(-1, 0), SyscallFailsWithErrno(EBADF)); +} + +TEST(FlockTestNoFixture, FlockDir) { + auto dir = ASSERT_NO_ERRNO_AND_VALUE(TempPath::CreateDir()); + auto fd = ASSERT_NO_ERRNO_AND_VALUE(Open(dir.path(), O_RDONLY, 0000)); + EXPECT_THAT(flock(fd.get(), LOCK_EX | LOCK_NB), SyscallSucceeds()); +} + +TEST(FlockTestNoFixture, FlockSymlink) { + // TODO(gvisor.dev/issue/2782): Replace with IsRunningWithVFS1() when O_PATH + // is supported. + SKIP_IF(IsRunningOnGvisor()); + + auto file = ASSERT_NO_ERRNO_AND_VALUE(TempPath::CreateFile()); + auto symlink = ASSERT_NO_ERRNO_AND_VALUE( + TempPath::CreateSymlinkTo(GetAbsoluteTestTmpdir(), file.path())); + + auto fd = + ASSERT_NO_ERRNO_AND_VALUE(Open(symlink.path(), O_RDONLY | O_PATH, 0000)); + EXPECT_THAT(flock(fd.get(), LOCK_EX | LOCK_NB), SyscallFailsWithErrno(EBADF)); +} + +TEST(FlockTestNoFixture, FlockProc) { + auto fd = + ASSERT_NO_ERRNO_AND_VALUE(Open("/proc/self/status", O_RDONLY, 0000)); + EXPECT_THAT(flock(fd.get(), LOCK_EX | LOCK_NB), SyscallSucceeds()); +} + +TEST(FlockTestNoFixture, FlockPipe) { + int fds[2]; + ASSERT_THAT(pipe(fds), SyscallSucceeds()); + + EXPECT_THAT(flock(fds[0], LOCK_EX | LOCK_NB), SyscallSucceeds()); + // Check that the pipe was locked above. + EXPECT_THAT(flock(fds[1], LOCK_EX | LOCK_NB), SyscallFailsWithErrno(EAGAIN)); + + EXPECT_THAT(flock(fds[0], LOCK_UN), SyscallSucceeds()); + EXPECT_THAT(flock(fds[1], LOCK_EX | LOCK_NB), SyscallSucceeds()); + + EXPECT_THAT(close(fds[0]), SyscallSucceeds()); + EXPECT_THAT(close(fds[1]), SyscallSucceeds()); +} + +TEST(FlockTestNoFixture, FlockSocket) { + int sock = socket(AF_UNIX, SOCK_STREAM, 0); + ASSERT_THAT(sock, SyscallSucceeds()); + + struct sockaddr_un addr = + ASSERT_NO_ERRNO_AND_VALUE(UniqueUnixAddr(true /* abstract */, AF_UNIX)); + ASSERT_THAT( + bind(sock, reinterpret_cast<struct sockaddr*>(&addr), sizeof(addr)), + SyscallSucceeds()); + + EXPECT_THAT(flock(sock, LOCK_EX | LOCK_NB), SyscallSucceeds()); + EXPECT_THAT(close(sock), SyscallSucceeds()); +} + } // namespace } // namespace testing diff --git a/test/syscalls/linux/fpsig_fork.cc b/test/syscalls/linux/fpsig_fork.cc index d08111cd3..c47567b4e 100644 --- a/test/syscalls/linux/fpsig_fork.cc +++ b/test/syscalls/linux/fpsig_fork.cc @@ -37,11 +37,11 @@ namespace { #define __stringify_1(x...) #x #define __stringify(x...) __stringify_1(x) #define GET_FPREG(var, regname) \ - asm volatile("str "__stringify(regname) ", %0" : "=m"(var)) + asm volatile("str " __stringify(regname) ", %0" : "=m"(var)) #define SET_FPREG(var, regname) \ - asm volatile("ldr "__stringify(regname) ", %0" : "=m"(var)) + asm volatile("ldr " __stringify(regname) ", %0" : "=m"(var)) #define GET_FP0(var) GET_FPREG(var, d0) -#define SET_FP0(var) GET_FPREG(var, d0) +#define SET_FP0(var) SET_FPREG(var, d0) #endif int parent, child; diff --git a/test/syscalls/linux/fpsig_nested.cc b/test/syscalls/linux/fpsig_nested.cc index c476a8e7a..302d928d1 100644 --- a/test/syscalls/linux/fpsig_nested.cc +++ b/test/syscalls/linux/fpsig_nested.cc @@ -26,9 +26,22 @@ namespace testing { namespace { +#ifdef __x86_64__ #define GET_XMM(__var, __xmm) \ asm volatile("movq %%" #__xmm ", %0" : "=r"(__var)) #define SET_XMM(__var, __xmm) asm volatile("movq %0, %%" #__xmm : : "r"(__var)) +#define GET_FP0(__var) GET_XMM(__var, xmm0) +#define SET_FP0(__var) SET_XMM(__var, xmm0) +#elif __aarch64__ +#define __stringify_1(x...) #x +#define __stringify(x...) __stringify_1(x) +#define GET_FPREG(var, regname) \ + asm volatile("str " __stringify(regname) ", %0" : "=m"(var)) +#define SET_FPREG(var, regname) \ + asm volatile("ldr " __stringify(regname) ", %0" : "=m"(var)) +#define GET_FP0(var) GET_FPREG(var, d0) +#define SET_FP0(var) SET_FPREG(var, d0) +#endif int pid; int tid; @@ -40,20 +53,21 @@ void sigusr2(int s, siginfo_t* siginfo, void* _uc) { uint64_t val = SIGUSR2; // Record the value of %xmm0 on entry and then clobber it. - GET_XMM(entryxmm[1], xmm0); - SET_XMM(val, xmm0); - GET_XMM(exitxmm[1], xmm0); + GET_FP0(entryxmm[1]); + SET_FP0(val); + GET_FP0(exitxmm[1]); } void sigusr1(int s, siginfo_t* siginfo, void* _uc) { uint64_t val = SIGUSR1; // Record the value of %xmm0 on entry and then clobber it. - GET_XMM(entryxmm[0], xmm0); - SET_XMM(val, xmm0); + GET_FP0(entryxmm[0]); + SET_FP0(val); // Send a SIGUSR2 to ourself. The signal mask is configured such that // the SIGUSR2 handler will run before this handler returns. +#ifdef __x86_64__ asm volatile( "movl %[killnr], %%eax;" "movl %[pid], %%edi;" @@ -66,10 +80,19 @@ void sigusr1(int s, siginfo_t* siginfo, void* _uc) { : "rax", "rdi", "rsi", "rdx", // Clobbered by syscall. "rcx", "r11"); +#elif __aarch64__ + asm volatile( + "mov x8, %0\n" + "mov x0, %1\n" + "mov x1, %2\n" + "mov x2, %3\n" + "svc #0\n" ::"r"(__NR_tgkill), + "r"(pid), "r"(tid), "r"(SIGUSR2)); +#endif // Record value of %xmm0 again to verify that the nested signal handler // does not clobber it. - GET_XMM(exitxmm[0], xmm0); + GET_FP0(exitxmm[0]); } TEST(FPSigTest, NestedSignals) { @@ -98,8 +121,9 @@ TEST(FPSigTest, NestedSignals) { // to signal the current thread ensures that this is the clobbered thread. uint64_t expected = 0xdeadbeeffacefeed; - SET_XMM(expected, xmm0); + SET_FP0(expected); +#ifdef __x86_64__ asm volatile( "movl %[killnr], %%eax;" "movl %[pid], %%edi;" @@ -112,9 +136,18 @@ TEST(FPSigTest, NestedSignals) { : "rax", "rdi", "rsi", "rdx", // Clobbered by syscall. "rcx", "r11"); +#elif __aarch64__ + asm volatile( + "mov x8, %0\n" + "mov x0, %1\n" + "mov x1, %2\n" + "mov x2, %3\n" + "svc #0\n" ::"r"(__NR_tgkill), + "r"(pid), "r"(tid), "r"(SIGUSR1)); +#endif uint64_t got; - GET_XMM(got, xmm0); + GET_FP0(got); // // The checks below verifies the following: diff --git a/test/syscalls/linux/futex.cc b/test/syscalls/linux/futex.cc index 40c80a6e1..90b1f0508 100644 --- a/test/syscalls/linux/futex.cc +++ b/test/syscalls/linux/futex.cc @@ -18,6 +18,7 @@ #include <sys/syscall.h> #include <sys/time.h> #include <sys/types.h> +#include <syscall.h> #include <unistd.h> #include <algorithm> @@ -737,6 +738,97 @@ TEST_P(PrivateAndSharedFutexTest, PITryLockConcurrency_NoRandomSave) { } } +int get_robust_list(int pid, struct robust_list_head** head_ptr, + size_t* len_ptr) { + return syscall(__NR_get_robust_list, pid, head_ptr, len_ptr); +} + +int set_robust_list(struct robust_list_head* head, size_t len) { + return syscall(__NR_set_robust_list, head, len); +} + +TEST(RobustFutexTest, BasicSetGet) { + struct robust_list_head hd = {}; + struct robust_list_head* hd_ptr = &hd; + + // Set! + EXPECT_THAT(set_robust_list(hd_ptr, sizeof(hd)), SyscallSucceedsWithValue(0)); + + // Get! + struct robust_list_head* new_hd_ptr = hd_ptr; + size_t len; + EXPECT_THAT(get_robust_list(0, &new_hd_ptr, &len), + SyscallSucceedsWithValue(0)); + EXPECT_EQ(new_hd_ptr, hd_ptr); + EXPECT_EQ(len, sizeof(hd)); +} + +TEST(RobustFutexTest, GetFromOtherTid) { + // Get the current tid and list head. + pid_t tid = gettid(); + struct robust_list_head* hd_ptr = {}; + size_t len; + EXPECT_THAT(get_robust_list(0, &hd_ptr, &len), SyscallSucceedsWithValue(0)); + + // Create a new thread. + ScopedThread t([&] { + // Current tid list head should be different from parent tid. + struct robust_list_head* got_hd_ptr = {}; + EXPECT_THAT(get_robust_list(0, &got_hd_ptr, &len), + SyscallSucceedsWithValue(0)); + EXPECT_NE(hd_ptr, got_hd_ptr); + + // Get the parent list head by passing its tid. + EXPECT_THAT(get_robust_list(tid, &got_hd_ptr, &len), + SyscallSucceedsWithValue(0)); + EXPECT_EQ(hd_ptr, got_hd_ptr); + }); + + // Wait for thread. + t.Join(); +} + +TEST(RobustFutexTest, InvalidSize) { + struct robust_list_head* hd = {}; + EXPECT_THAT(set_robust_list(hd, sizeof(*hd) + 1), + SyscallFailsWithErrno(EINVAL)); +} + +TEST(RobustFutexTest, PthreadMutexAttr) { + constexpr int kNumMutexes = 3; + + // Create a bunch of robust mutexes. + pthread_mutexattr_t attrs[kNumMutexes]; + pthread_mutex_t mtxs[kNumMutexes]; + for (int i = 0; i < kNumMutexes; i++) { + TEST_PCHECK(pthread_mutexattr_init(&attrs[i]) == 0); + TEST_PCHECK(pthread_mutexattr_setrobust(&attrs[i], PTHREAD_MUTEX_ROBUST) == + 0); + TEST_PCHECK(pthread_mutex_init(&mtxs[i], &attrs[i]) == 0); + } + + // Start thread to lock the mutexes and then exit. + ScopedThread t([&] { + for (int i = 0; i < kNumMutexes; i++) { + TEST_PCHECK(pthread_mutex_lock(&mtxs[i]) == 0); + } + pthread_exit(NULL); + }); + + // Wait for thread. + t.Join(); + + // Now try to take the mutexes. + for (int i = 0; i < kNumMutexes; i++) { + // Should get EOWNERDEAD. + EXPECT_EQ(pthread_mutex_lock(&mtxs[i]), EOWNERDEAD); + // Make the mutex consistent. + EXPECT_EQ(pthread_mutex_consistent(&mtxs[i]), 0); + // Unlock. + EXPECT_EQ(pthread_mutex_unlock(&mtxs[i]), 0); + } +} + } // namespace } // namespace testing } // namespace gvisor diff --git a/test/syscalls/linux/getdents.cc b/test/syscalls/linux/getdents.cc index b147d6181..b040cdcf7 100644 --- a/test/syscalls/linux/getdents.cc +++ b/test/syscalls/linux/getdents.cc @@ -32,6 +32,7 @@ #include "gmock/gmock.h" #include "gtest/gtest.h" +#include "absl/container/node_hash_set.h" #include "absl/strings/numbers.h" #include "absl/strings/str_cat.h" #include "test/util/eventfd_util.h" @@ -393,7 +394,7 @@ TYPED_TEST(GetdentsTest, ProcSelfFd) { // Make the buffer very small since we want to iterate. typename TestFixture::DirentBufferType dirents( 2 * sizeof(typename TestFixture::LinuxDirentType)); - std::unordered_set<int> prev_fds; + absl::node_hash_set<int> prev_fds; while (true) { dirents.Reset(); int rv; diff --git a/test/syscalls/linux/inotify.cc b/test/syscalls/linux/inotify.cc index 0e13ad190..6cf398097 100644 --- a/test/syscalls/linux/inotify.cc +++ b/test/syscalls/linux/inotify.cc @@ -18,7 +18,9 @@ #include <sys/epoll.h> #include <sys/inotify.h> #include <sys/ioctl.h> +#include <sys/sendfile.h> #include <sys/time.h> +#include <sys/xattr.h> #include <atomic> #include <list> @@ -28,11 +30,13 @@ #include "absl/strings/str_cat.h" #include "absl/strings/str_format.h" #include "absl/strings/str_join.h" +#include "absl/synchronization/mutex.h" #include "absl/time/clock.h" #include "absl/time/time.h" #include "test/util/epoll_util.h" #include "test/util/file_descriptor.h" #include "test/util/fs_util.h" +#include "test/util/posix_error.h" #include "test/util/temp_path.h" #include "test/util/test_util.h" #include "test/util/thread_util.h" @@ -330,9 +334,32 @@ PosixErrorOr<int> InotifyAddWatch(int fd, const std::string& path, return wd; } -TEST(Inotify, InotifyFdNotWritable) { +TEST(Inotify, IllegalSeek) { const FileDescriptor fd = ASSERT_NO_ERRNO_AND_VALUE(InotifyInit1(0)); - EXPECT_THAT(write(fd.get(), "x", 1), SyscallFailsWithErrno(EBADF)); + EXPECT_THAT(lseek(fd.get(), 0, SEEK_SET), SyscallFailsWithErrno(ESPIPE)); +} + +TEST(Inotify, IllegalPread) { + const FileDescriptor fd = ASSERT_NO_ERRNO_AND_VALUE(InotifyInit1(0)); + int val; + EXPECT_THAT(pread(fd.get(), &val, sizeof(val), 0), + SyscallFailsWithErrno(ESPIPE)); +} + +TEST(Inotify, IllegalPwrite) { + const FileDescriptor fd = ASSERT_NO_ERRNO_AND_VALUE(InotifyInit1(0)); + EXPECT_THAT(pwrite(fd.get(), "x", 1, 0), SyscallFailsWithErrno(ESPIPE)); +} + +TEST(Inotify, IllegalWrite) { + const FileDescriptor fd = ASSERT_NO_ERRNO_AND_VALUE(InotifyInit1(0)); + int val = 0; + EXPECT_THAT(write(fd.get(), &val, sizeof(val)), SyscallFailsWithErrno(EBADF)); +} + +TEST(Inotify, InitFlags) { + EXPECT_THAT(inotify_init1(IN_NONBLOCK | IN_CLOEXEC), SyscallSucceeds()); + EXPECT_THAT(inotify_init1(12345), SyscallFailsWithErrno(EINVAL)); } TEST(Inotify, NonBlockingReadReturnsEagain) { @@ -395,7 +422,7 @@ TEST(Inotify, CanDeleteFileAfterRemovingWatch) { file1.reset(); } -TEST(Inotify, CanRemoveWatchAfterDeletingFile) { +TEST(Inotify, RemoveWatchAfterDeletingFileFails) { const TempPath root = ASSERT_NO_ERRNO_AND_VALUE(TempPath::CreateDir()); TempPath file1 = ASSERT_NO_ERRNO_AND_VALUE(TempPath::CreateFileIn(root.path())); @@ -491,17 +518,23 @@ TEST(Inotify, DeletingChildGeneratesEvents) { Event(IN_DELETE, root_wd, Basename(file1_path))})); } +// Creating a file in "parent/child" should generate events for child, but not +// parent. TEST(Inotify, CreatingFileGeneratesEvents) { - const TempPath root = ASSERT_NO_ERRNO_AND_VALUE(TempPath::CreateDir()); + const TempPath parent = ASSERT_NO_ERRNO_AND_VALUE(TempPath::CreateDir()); + const TempPath child = + ASSERT_NO_ERRNO_AND_VALUE(TempPath::CreateDirIn(parent.path())); const FileDescriptor fd = ASSERT_NO_ERRNO_AND_VALUE(InotifyInit1(IN_NONBLOCK)); + ASSERT_NO_ERRNO_AND_VALUE( + InotifyAddWatch(fd.get(), parent.path(), IN_ALL_EVENTS)); const int wd = ASSERT_NO_ERRNO_AND_VALUE( - InotifyAddWatch(fd.get(), root.path(), IN_ALL_EVENTS)); + InotifyAddWatch(fd.get(), child.path(), IN_ALL_EVENTS)); // Create a new file in the directory. const TempPath file1 = - ASSERT_NO_ERRNO_AND_VALUE(TempPath::CreateFileIn(root.path())); + ASSERT_NO_ERRNO_AND_VALUE(TempPath::CreateFileIn(child.path())); const std::vector<Event> events = ASSERT_NO_ERRNO_AND_VALUE(DrainEvents(fd.get())); @@ -554,6 +587,47 @@ TEST(Inotify, WritingFileGeneratesModifyEvent) { ASSERT_THAT(events, Are({Event(IN_MODIFY, wd, Basename(file1.path()))})); } +TEST(Inotify, SizeZeroReadWriteGeneratesNothing) { + const TempPath root = ASSERT_NO_ERRNO_AND_VALUE(TempPath::CreateDir()); + const FileDescriptor fd = + ASSERT_NO_ERRNO_AND_VALUE(InotifyInit1(IN_NONBLOCK)); + const TempPath file1 = + ASSERT_NO_ERRNO_AND_VALUE(TempPath::CreateFileIn(root.path())); + + const FileDescriptor file1_fd = + ASSERT_NO_ERRNO_AND_VALUE(Open(file1.path(), O_RDWR)); + ASSERT_NO_ERRNO_AND_VALUE( + InotifyAddWatch(fd.get(), root.path(), IN_ALL_EVENTS)); + + // Read from the empty file. + int val; + ASSERT_THAT(read(file1_fd.get(), &val, sizeof(val)), + SyscallSucceedsWithValue(0)); + + // Write zero bytes. + ASSERT_THAT(write(file1_fd.get(), "", 0), SyscallSucceedsWithValue(0)); + + const std::vector<Event> events = + ASSERT_NO_ERRNO_AND_VALUE(DrainEvents(fd.get())); + ASSERT_THAT(events, Are({})); +} + +TEST(Inotify, FailedFileCreationGeneratesNoEvents) { + const TempPath dir = ASSERT_NO_ERRNO_AND_VALUE(TempPath::CreateDir()); + const std::string dir_path = dir.path(); + const FileDescriptor fd = + ASSERT_NO_ERRNO_AND_VALUE(InotifyInit1(IN_NONBLOCK)); + ASSERT_NO_ERRNO_AND_VALUE(InotifyAddWatch(fd.get(), dir_path, IN_ALL_EVENTS)); + + const char* p = dir_path.c_str(); + ASSERT_THAT(mkdir(p, 0777), SyscallFails()); + ASSERT_THAT(mknod(p, S_IFIFO, 0777), SyscallFails()); + ASSERT_THAT(symlink(p, p), SyscallFails()); + ASSERT_THAT(link(p, p), SyscallFails()); + std::vector<Event> events = ASSERT_NO_ERRNO_AND_VALUE(DrainEvents(fd.get())); + ASSERT_THAT(events, Are({})); +} + TEST(Inotify, WatchSetAfterOpenReportsCloseFdEvent) { const TempPath root = ASSERT_NO_ERRNO_AND_VALUE(TempPath::CreateDir()); const FileDescriptor fd = @@ -602,7 +676,7 @@ TEST(Inotify, ChildrenDeletionInWatchedDirGeneratesEvent) { Event(IN_DELETE | IN_ISDIR, wd, Basename(dir1_path))})); } -TEST(Inotify, WatchTargetDeletionGeneratesEvent) { +TEST(Inotify, RmdirOnWatchedTargetGeneratesEvent) { const TempPath root = ASSERT_NO_ERRNO_AND_VALUE(TempPath::CreateDir()); const FileDescriptor fd = ASSERT_NO_ERRNO_AND_VALUE(InotifyInit1(IN_NONBLOCK)); @@ -1228,7 +1302,7 @@ TEST(Inotify, LinkGeneratesAttribAndCreateEvents) { InotifyAddWatch(fd.get(), file1.path(), IN_ALL_EVENTS)); const int rc = link(file1.path().c_str(), link1.path().c_str()); - // link(2) is only supported on tmpfs in the sandbox. + // NOTE(b/34861058): link(2) is only supported on tmpfs in the sandbox. SKIP_IF(IsRunningOnGvisor() && rc != 0 && (errno == EPERM || errno == ENOENT)); ASSERT_THAT(rc, SyscallSucceeds()); @@ -1251,7 +1325,7 @@ TEST(Inotify, UtimesGeneratesAttribEvent) { const int wd = ASSERT_NO_ERRNO_AND_VALUE( InotifyAddWatch(fd.get(), root.path(), IN_ALL_EVENTS)); - struct timeval times[2] = {{1, 0}, {2, 0}}; + const struct timeval times[2] = {{1, 0}, {2, 0}}; EXPECT_THAT(futimes(file1_fd.get(), times), SyscallSucceeds()); const std::vector<Event> events = @@ -1322,21 +1396,27 @@ TEST(Inotify, HardlinksReuseSameWatch) { Event(IN_DELETE, root_wd, Basename(file1_path))})); } +// Calling mkdir within "parent/child" should generate an event for child, but +// not parent. TEST(Inotify, MkdirGeneratesCreateEventWithDirFlag) { - const TempPath root = ASSERT_NO_ERRNO_AND_VALUE(TempPath::CreateDir()); + const TempPath parent = ASSERT_NO_ERRNO_AND_VALUE(TempPath::CreateDir()); + const TempPath child = + ASSERT_NO_ERRNO_AND_VALUE(TempPath::CreateDirIn(parent.path())); const FileDescriptor fd = ASSERT_NO_ERRNO_AND_VALUE(InotifyInit1(IN_NONBLOCK)); - const int root_wd = ASSERT_NO_ERRNO_AND_VALUE( - InotifyAddWatch(fd.get(), root.path(), IN_ALL_EVENTS)); + ASSERT_NO_ERRNO_AND_VALUE( + InotifyAddWatch(fd.get(), parent.path(), IN_ALL_EVENTS)); + const int child_wd = ASSERT_NO_ERRNO_AND_VALUE( + InotifyAddWatch(fd.get(), child.path(), IN_ALL_EVENTS)); - const TempPath dir1(NewTempAbsPathInDir(root.path())); + const TempPath dir1(NewTempAbsPathInDir(child.path())); ASSERT_THAT(mkdir(dir1.path().c_str(), 0777), SyscallSucceeds()); const std::vector<Event> events = ASSERT_NO_ERRNO_AND_VALUE(DrainEvents(fd.get())); ASSERT_THAT( events, - Are({Event(IN_CREATE | IN_ISDIR, root_wd, Basename(dir1.path()))})); + Are({Event(IN_CREATE | IN_ISDIR, child_wd, Basename(dir1.path()))})); } TEST(Inotify, MultipleInotifyInstancesAndWatchesAllGetEvents) { @@ -1424,20 +1504,26 @@ TEST(Inotify, DuplicateWatchReturnsSameWatchDescriptor) { TEST(Inotify, UnmatchedEventsAreDiscarded) { const TempPath root = ASSERT_NO_ERRNO_AND_VALUE(TempPath::CreateDir()); - const TempPath file1 = + TempPath file1 = ASSERT_NO_ERRNO_AND_VALUE(TempPath::CreateFileIn(root.path())); const FileDescriptor fd = ASSERT_NO_ERRNO_AND_VALUE(InotifyInit1(IN_NONBLOCK)); - ASSERT_NO_ERRNO_AND_VALUE(InotifyAddWatch(fd.get(), file1.path(), IN_ACCESS)); + const int wd = ASSERT_NO_ERRNO_AND_VALUE( + InotifyAddWatch(fd.get(), file1.path(), IN_ACCESS)); - const FileDescriptor file1_fd = + FileDescriptor file1_fd = ASSERT_NO_ERRNO_AND_VALUE(Open(file1.path(), O_WRONLY)); - const std::vector<Event> events = - ASSERT_NO_ERRNO_AND_VALUE(DrainEvents(fd.get())); + std::vector<Event> events = ASSERT_NO_ERRNO_AND_VALUE(DrainEvents(fd.get())); // We only asked for access events, the open event should be discarded. ASSERT_THAT(events, Are({})); + + // IN_IGNORED events are always generated, regardless of the mask. + file1_fd.reset(); + file1.reset(); + events = ASSERT_NO_ERRNO_AND_VALUE(DrainEvents(fd.get())); + ASSERT_THAT(events, Are({Event(IN_IGNORED, wd)})); } TEST(Inotify, AddWatchWithInvalidEventMaskFails) { @@ -1596,7 +1682,98 @@ TEST(Inotify, EpollNoDeadlock) { } } -TEST(Inotify, SpliceEvent) { +TEST(Inotify, Fallocate) { + const TempPath file = ASSERT_NO_ERRNO_AND_VALUE(TempPath::CreateFile()); + const FileDescriptor fd = + ASSERT_NO_ERRNO_AND_VALUE(Open(file.path(), O_RDWR)); + + const FileDescriptor inotify_fd = + ASSERT_NO_ERRNO_AND_VALUE(InotifyInit1(IN_NONBLOCK)); + const int wd = ASSERT_NO_ERRNO_AND_VALUE( + InotifyAddWatch(inotify_fd.get(), file.path(), IN_ALL_EVENTS)); + + // Do an arbitrary modification with fallocate. + ASSERT_THAT(fallocate(fd.get(), 0, 0, 123), SyscallSucceeds()); + std::vector<Event> events = + ASSERT_NO_ERRNO_AND_VALUE(DrainEvents(inotify_fd.get())); + EXPECT_THAT(events, Are({Event(IN_MODIFY, wd)})); +} + +TEST(Inotify, Sendfile) { + SKIP_IF(IsRunningWithVFS1()); + + const TempPath root = ASSERT_NO_ERRNO_AND_VALUE(TempPath::CreateDir()); + const TempPath in_file = ASSERT_NO_ERRNO_AND_VALUE( + TempPath::CreateFileWith(root.path(), "x", 0644)); + const TempPath out_file = ASSERT_NO_ERRNO_AND_VALUE(TempPath::CreateFile()); + const FileDescriptor in = + ASSERT_NO_ERRNO_AND_VALUE(Open(in_file.path(), O_RDONLY)); + const FileDescriptor out = + ASSERT_NO_ERRNO_AND_VALUE(Open(out_file.path(), O_WRONLY)); + + // Create separate inotify instances for the in and out fds. If both watches + // were on the same instance, we would have discrepancies between Linux and + // gVisor (order of events, duplicate events), which is not that important + // since inotify is asynchronous anyway. + const FileDescriptor in_inotify = + ASSERT_NO_ERRNO_AND_VALUE(InotifyInit1(IN_NONBLOCK)); + const FileDescriptor out_inotify = + ASSERT_NO_ERRNO_AND_VALUE(InotifyInit1(IN_NONBLOCK)); + const int in_wd = ASSERT_NO_ERRNO_AND_VALUE( + InotifyAddWatch(in_inotify.get(), in_file.path(), IN_ALL_EVENTS)); + const int out_wd = ASSERT_NO_ERRNO_AND_VALUE( + InotifyAddWatch(out_inotify.get(), out_file.path(), IN_ALL_EVENTS)); + + ASSERT_THAT(sendfile(out.get(), in.get(), /*offset=*/nullptr, 1), + SyscallSucceeds()); + + // Expect a single access event and a single modify event. + std::vector<Event> in_events = + ASSERT_NO_ERRNO_AND_VALUE(DrainEvents(in_inotify.get())); + std::vector<Event> out_events = + ASSERT_NO_ERRNO_AND_VALUE(DrainEvents(out_inotify.get())); + EXPECT_THAT(in_events, Are({Event(IN_ACCESS, in_wd)})); + EXPECT_THAT(out_events, Are({Event(IN_MODIFY, out_wd)})); +} + +// On Linux, inotify behavior is not very consistent with splice(2). We try our +// best to emulate Linux for very basic calls to splice. +TEST(Inotify, SpliceOnWatchTarget) { + int pipes[2]; + ASSERT_THAT(pipe2(pipes, O_NONBLOCK), SyscallSucceeds()); + + const TempPath dir = ASSERT_NO_ERRNO_AND_VALUE(TempPath::CreateDir()); + const FileDescriptor inotify_fd = + ASSERT_NO_ERRNO_AND_VALUE(InotifyInit1(IN_NONBLOCK)); + const TempPath file = ASSERT_NO_ERRNO_AND_VALUE(TempPath::CreateFileWith( + dir.path(), "some content", TempPath::kDefaultFileMode)); + + const FileDescriptor fd = + ASSERT_NO_ERRNO_AND_VALUE(Open(file.path(), O_RDWR)); + const int dir_wd = ASSERT_NO_ERRNO_AND_VALUE( + InotifyAddWatch(inotify_fd.get(), dir.path(), IN_ALL_EVENTS)); + const int file_wd = ASSERT_NO_ERRNO_AND_VALUE( + InotifyAddWatch(inotify_fd.get(), file.path(), IN_ALL_EVENTS)); + + EXPECT_THAT(splice(fd.get(), nullptr, pipes[1], nullptr, 1, /*flags=*/0), + SyscallSucceedsWithValue(1)); + + // Surprisingly, events are not generated in Linux if we read from a file. + std::vector<Event> events = + ASSERT_NO_ERRNO_AND_VALUE(DrainEvents(inotify_fd.get())); + ASSERT_THAT(events, Are({})); + + EXPECT_THAT(splice(pipes[0], nullptr, fd.get(), nullptr, 1, /*flags=*/0), + SyscallSucceedsWithValue(1)); + + events = ASSERT_NO_ERRNO_AND_VALUE(DrainEvents(inotify_fd.get())); + ASSERT_THAT(events, Are({ + Event(IN_MODIFY, dir_wd, Basename(file.path())), + Event(IN_MODIFY, file_wd), + })); +} + +TEST(Inotify, SpliceOnInotifyFD) { int pipes[2]; ASSERT_THAT(pipe2(pipes, O_NONBLOCK), SyscallSucceeds()); @@ -1624,6 +1801,635 @@ TEST(Inotify, SpliceEvent) { ASSERT_THAT(events, Are({Event(IN_ACCESS, watcher)})); } +// Watches on a parent should not be triggered by actions on a hard link to one +// of its children that has a different parent. +TEST(Inotify, LinkOnOtherParent) { + const TempPath dir1 = ASSERT_NO_ERRNO_AND_VALUE(TempPath::CreateDir()); + const TempPath dir2 = ASSERT_NO_ERRNO_AND_VALUE(TempPath::CreateDir()); + const TempPath file = + ASSERT_NO_ERRNO_AND_VALUE(TempPath::CreateFileIn(dir1.path())); + std::string link_path = NewTempAbsPathInDir(dir2.path()); + + const int rc = link(file.path().c_str(), link_path.c_str()); + // NOTE(b/34861058): link(2) is only supported on tmpfs in the sandbox. + SKIP_IF(IsRunningOnGvisor() && rc != 0 && + (errno == EPERM || errno == ENOENT)); + ASSERT_THAT(rc, SyscallSucceeds()); + + const FileDescriptor inotify_fd = + ASSERT_NO_ERRNO_AND_VALUE(InotifyInit1(IN_NONBLOCK)); + ASSERT_NO_ERRNO_AND_VALUE( + InotifyAddWatch(inotify_fd.get(), dir1.path(), IN_ALL_EVENTS)); + + // Perform various actions on the link outside of dir1, which should trigger + // no inotify events. + const FileDescriptor fd = + ASSERT_NO_ERRNO_AND_VALUE(Open(link_path.c_str(), O_RDWR)); + int val = 0; + ASSERT_THAT(write(fd.get(), &val, sizeof(val)), SyscallSucceeds()); + ASSERT_THAT(read(fd.get(), &val, sizeof(val)), SyscallSucceeds()); + ASSERT_THAT(ftruncate(fd.get(), 12345), SyscallSucceeds()); + ASSERT_THAT(unlink(link_path.c_str()), SyscallSucceeds()); + const std::vector<Event> events = + ASSERT_NO_ERRNO_AND_VALUE(DrainEvents(inotify_fd.get())); + EXPECT_THAT(events, Are({})); +} + +TEST(Inotify, Xattr) { + // TODO(gvisor.dev/issue/1636): Support extended attributes in runsc gofer. + SKIP_IF(IsRunningOnGvisor()); + + const TempPath file = ASSERT_NO_ERRNO_AND_VALUE(TempPath::CreateFile()); + const std::string path = file.path(); + const FileDescriptor fd = ASSERT_NO_ERRNO_AND_VALUE(Open(path, O_RDWR)); + const FileDescriptor inotify_fd = + ASSERT_NO_ERRNO_AND_VALUE(InotifyInit1(IN_NONBLOCK)); + const int wd = ASSERT_NO_ERRNO_AND_VALUE( + InotifyAddWatch(inotify_fd.get(), path, IN_ALL_EVENTS)); + + const char* cpath = path.c_str(); + const char* name = "user.test"; + int val = 123; + ASSERT_THAT(setxattr(cpath, name, &val, sizeof(val), /*flags=*/0), + SyscallSucceeds()); + std::vector<Event> events = + ASSERT_NO_ERRNO_AND_VALUE(DrainEvents(inotify_fd.get())); + EXPECT_THAT(events, Are({Event(IN_ATTRIB, wd)})); + + ASSERT_THAT(getxattr(cpath, name, &val, sizeof(val)), SyscallSucceeds()); + events = ASSERT_NO_ERRNO_AND_VALUE(DrainEvents(inotify_fd.get())); + EXPECT_THAT(events, Are({})); + + char list[100]; + ASSERT_THAT(listxattr(cpath, list, sizeof(list)), SyscallSucceeds()); + events = ASSERT_NO_ERRNO_AND_VALUE(DrainEvents(inotify_fd.get())); + EXPECT_THAT(events, Are({})); + + ASSERT_THAT(removexattr(cpath, name), SyscallSucceeds()); + events = ASSERT_NO_ERRNO_AND_VALUE(DrainEvents(inotify_fd.get())); + EXPECT_THAT(events, Are({Event(IN_ATTRIB, wd)})); + + ASSERT_THAT(fsetxattr(fd.get(), name, &val, sizeof(val), /*flags=*/0), + SyscallSucceeds()); + events = ASSERT_NO_ERRNO_AND_VALUE(DrainEvents(inotify_fd.get())); + EXPECT_THAT(events, Are({Event(IN_ATTRIB, wd)})); + + ASSERT_THAT(fgetxattr(fd.get(), name, &val, sizeof(val)), SyscallSucceeds()); + events = ASSERT_NO_ERRNO_AND_VALUE(DrainEvents(inotify_fd.get())); + EXPECT_THAT(events, Are({})); + + ASSERT_THAT(flistxattr(fd.get(), list, sizeof(list)), SyscallSucceeds()); + events = ASSERT_NO_ERRNO_AND_VALUE(DrainEvents(inotify_fd.get())); + EXPECT_THAT(events, Are({})); + + ASSERT_THAT(fremovexattr(fd.get(), name), SyscallSucceeds()); + events = ASSERT_NO_ERRNO_AND_VALUE(DrainEvents(inotify_fd.get())); + EXPECT_THAT(events, Are({Event(IN_ATTRIB, wd)})); +} + +TEST(Inotify, Exec) { + const TempPath dir = ASSERT_NO_ERRNO_AND_VALUE(TempPath::CreateDir()); + const TempPath bin = ASSERT_NO_ERRNO_AND_VALUE( + TempPath::CreateSymlinkTo(dir.path(), "/bin/true")); + + const FileDescriptor fd = + ASSERT_NO_ERRNO_AND_VALUE(InotifyInit1(IN_NONBLOCK)); + const int wd = ASSERT_NO_ERRNO_AND_VALUE( + InotifyAddWatch(fd.get(), bin.path(), IN_ALL_EVENTS)); + + // Perform exec. + ScopedThread t([&bin]() { + ASSERT_THAT(execl(bin.path().c_str(), bin.path().c_str(), (char*)nullptr), + SyscallSucceeds()); + }); + t.Join(); + + std::vector<Event> events = ASSERT_NO_ERRNO_AND_VALUE(DrainEvents(fd.get())); + EXPECT_THAT(events, Are({Event(IN_OPEN, wd), Event(IN_ACCESS, wd)})); +} + +// Watches without IN_EXCL_UNLINK, should continue to emit events for file +// descriptors after their corresponding files have been unlinked. +// +// We need to disable S/R because there are filesystems where we cannot re-open +// fds to an unlinked file across S/R, e.g. gofer-backed filesytems. +TEST(Inotify, IncludeUnlinkedFile_NoRandomSave) { + const DisableSave ds; + + const TempPath dir = ASSERT_NO_ERRNO_AND_VALUE(TempPath::CreateDir()); + const TempPath file = ASSERT_NO_ERRNO_AND_VALUE( + TempPath::CreateFileWith(dir.path(), "123", TempPath::kDefaultFileMode)); + FileDescriptor fd = ASSERT_NO_ERRNO_AND_VALUE(Open(file.path(), O_RDWR)); + + const FileDescriptor inotify_fd = + ASSERT_NO_ERRNO_AND_VALUE(InotifyInit1(IN_NONBLOCK)); + const int dir_wd = ASSERT_NO_ERRNO_AND_VALUE( + InotifyAddWatch(inotify_fd.get(), dir.path(), IN_ALL_EVENTS)); + const int file_wd = ASSERT_NO_ERRNO_AND_VALUE( + InotifyAddWatch(inotify_fd.get(), file.path(), IN_ALL_EVENTS)); + + ASSERT_THAT(unlink(file.path().c_str()), SyscallSucceeds()); + int val = 0; + ASSERT_THAT(read(fd.get(), &val, sizeof(val)), SyscallSucceeds()); + ASSERT_THAT(write(fd.get(), &val, sizeof(val)), SyscallSucceeds()); + std::vector<Event> events = + ASSERT_NO_ERRNO_AND_VALUE(DrainEvents(inotify_fd.get())); + EXPECT_THAT(events, Are({ + Event(IN_ATTRIB, file_wd), + Event(IN_DELETE, dir_wd, Basename(file.path())), + Event(IN_ACCESS, dir_wd, Basename(file.path())), + Event(IN_ACCESS, file_wd), + Event(IN_MODIFY, dir_wd, Basename(file.path())), + Event(IN_MODIFY, file_wd), + })); + + fd.reset(); + events = ASSERT_NO_ERRNO_AND_VALUE(DrainEvents(inotify_fd.get())); + EXPECT_THAT(events, Are({ + Event(IN_CLOSE_WRITE, dir_wd, Basename(file.path())), + Event(IN_CLOSE_WRITE, file_wd), + Event(IN_DELETE_SELF, file_wd), + Event(IN_IGNORED, file_wd), + })); +} + +// Watches created with IN_EXCL_UNLINK will stop emitting events on fds for +// children that have already been unlinked. +// +// We need to disable S/R because there are filesystems where we cannot re-open +// fds to an unlinked file across S/R, e.g. gofer-backed filesytems. +TEST(Inotify, ExcludeUnlink_NoRandomSave) { + const DisableSave ds; + // TODO(gvisor.dev/issue/1624): This test fails on VFS1. + SKIP_IF(IsRunningWithVFS1()); + + const TempPath dir = ASSERT_NO_ERRNO_AND_VALUE(TempPath::CreateDir()); + const TempPath file = + ASSERT_NO_ERRNO_AND_VALUE(TempPath::CreateFileIn(dir.path())); + + FileDescriptor fd = ASSERT_NO_ERRNO_AND_VALUE(Open(file.path(), O_RDWR)); + + const FileDescriptor inotify_fd = + ASSERT_NO_ERRNO_AND_VALUE(InotifyInit1(IN_NONBLOCK)); + const int dir_wd = ASSERT_NO_ERRNO_AND_VALUE(InotifyAddWatch( + inotify_fd.get(), dir.path(), IN_ALL_EVENTS | IN_EXCL_UNLINK)); + const int file_wd = ASSERT_NO_ERRNO_AND_VALUE(InotifyAddWatch( + inotify_fd.get(), file.path(), IN_ALL_EVENTS | IN_EXCL_UNLINK)); + + // Unlink the child, which should cause further operations on the open file + // descriptor to be ignored. + ASSERT_THAT(unlink(file.path().c_str()), SyscallSucceeds()); + int val = 0; + ASSERT_THAT(write(fd.get(), &val, sizeof(val)), SyscallSucceeds()); + ASSERT_THAT(read(fd.get(), &val, sizeof(val)), SyscallSucceeds()); + std::vector<Event> events = + ASSERT_NO_ERRNO_AND_VALUE(DrainEvents(inotify_fd.get())); + EXPECT_THAT(events, Are({ + Event(IN_ATTRIB, file_wd), + Event(IN_DELETE, dir_wd, Basename(file.path())), + })); + + fd.reset(); + events = ASSERT_NO_ERRNO_AND_VALUE(DrainEvents(inotify_fd.get())); + ASSERT_THAT(events, Are({ + Event(IN_DELETE_SELF, file_wd), + Event(IN_IGNORED, file_wd), + })); +} + +// We need to disable S/R because there are filesystems where we cannot re-open +// fds to an unlinked file across S/R, e.g. gofer-backed filesytems. +TEST(Inotify, ExcludeUnlinkDirectory_NoRandomSave) { + // TODO(gvisor.dev/issue/1624): This test fails on VFS1. Remove once VFS1 is + // deleted. + SKIP_IF(IsRunningWithVFS1()); + + const DisableSave ds; + + const TempPath parent = ASSERT_NO_ERRNO_AND_VALUE(TempPath::CreateDir()); + TempPath dir = + ASSERT_NO_ERRNO_AND_VALUE(TempPath::CreateDirIn(parent.path())); + std::string dirPath = dir.path(); + const FileDescriptor inotify_fd = + ASSERT_NO_ERRNO_AND_VALUE(InotifyInit1(IN_NONBLOCK)); + + FileDescriptor fd = + ASSERT_NO_ERRNO_AND_VALUE(Open(dirPath.c_str(), O_RDONLY | O_DIRECTORY)); + const int parent_wd = ASSERT_NO_ERRNO_AND_VALUE(InotifyAddWatch( + inotify_fd.get(), parent.path(), IN_ALL_EVENTS | IN_EXCL_UNLINK)); + const int self_wd = ASSERT_NO_ERRNO_AND_VALUE(InotifyAddWatch( + inotify_fd.get(), dir.path(), IN_ALL_EVENTS | IN_EXCL_UNLINK)); + + // Unlink the dir, and then close the open fd. + ASSERT_THAT(rmdir(dirPath.c_str()), SyscallSucceeds()); + dir.reset(); + + std::vector<Event> events = + ASSERT_NO_ERRNO_AND_VALUE(DrainEvents(inotify_fd.get())); + // No close event should appear. + ASSERT_THAT(events, + Are({Event(IN_DELETE | IN_ISDIR, parent_wd, Basename(dirPath))})); + + fd.reset(); + events = ASSERT_NO_ERRNO_AND_VALUE(DrainEvents(inotify_fd.get())); + ASSERT_THAT(events, Are({ + Event(IN_DELETE_SELF, self_wd), + Event(IN_IGNORED, self_wd), + })); +} + +// If "dir/child" and "dir/child2" are links to the same file, and "dir/child" +// is unlinked, a watch on "dir" with IN_EXCL_UNLINK will exclude future events +// for fds on "dir/child" but not "dir/child2". +// +// We need to disable S/R because there are filesystems where we cannot re-open +// fds to an unlinked file across S/R, e.g. gofer-backed filesytems. +TEST(Inotify, ExcludeUnlinkMultipleChildren_NoRandomSave) { + const DisableSave ds; + // TODO(gvisor.dev/issue/1624): This test fails on VFS1. + SKIP_IF(IsRunningWithVFS1()); + + const TempPath dir = ASSERT_NO_ERRNO_AND_VALUE(TempPath::CreateDir()); + const TempPath file = + ASSERT_NO_ERRNO_AND_VALUE(TempPath::CreateFileIn(dir.path())); + std::string path1 = file.path(); + std::string path2 = NewTempAbsPathInDir(dir.path()); + + const int rc = link(path1.c_str(), path2.c_str()); + // NOTE(b/34861058): link(2) is only supported on tmpfs in the sandbox. + SKIP_IF(IsRunningOnGvisor() && rc != 0 && + (errno == EPERM || errno == ENOENT)); + ASSERT_THAT(rc, SyscallSucceeds()); + const FileDescriptor fd1 = + ASSERT_NO_ERRNO_AND_VALUE(Open(path1.c_str(), O_RDWR)); + const FileDescriptor fd2 = + ASSERT_NO_ERRNO_AND_VALUE(Open(path2.c_str(), O_RDWR)); + + const FileDescriptor inotify_fd = + ASSERT_NO_ERRNO_AND_VALUE(InotifyInit1(IN_NONBLOCK)); + const int wd = ASSERT_NO_ERRNO_AND_VALUE(InotifyAddWatch( + inotify_fd.get(), dir.path(), IN_ALL_EVENTS | IN_EXCL_UNLINK)); + + // After unlinking path1, only events on the fd for path2 should be generated. + ASSERT_THAT(unlink(path1.c_str()), SyscallSucceeds()); + ASSERT_THAT(write(fd1.get(), "x", 1), SyscallSucceeds()); + ASSERT_THAT(write(fd2.get(), "x", 1), SyscallSucceeds()); + + const std::vector<Event> events = + ASSERT_NO_ERRNO_AND_VALUE(DrainEvents(inotify_fd.get())); + EXPECT_THAT(events, Are({ + Event(IN_DELETE, wd, Basename(path1)), + Event(IN_MODIFY, wd, Basename(path2)), + })); +} + +// On native Linux, actions of data type FSNOTIFY_EVENT_INODE are not affected +// by IN_EXCL_UNLINK (see +// fs/notify/inotify/inotify_fsnotify.c:inotify_handle_event). Inode-level +// events include changes to metadata and extended attributes. +// +// We need to disable S/R because there are filesystems where we cannot re-open +// fds to an unlinked file across S/R, e.g. gofer-backed filesytems. +TEST(Inotify, ExcludeUnlinkInodeEvents_NoRandomSave) { + const DisableSave ds; + + const TempPath dir = ASSERT_NO_ERRNO_AND_VALUE(TempPath::CreateDir()); + const TempPath file = + ASSERT_NO_ERRNO_AND_VALUE(TempPath::CreateFileIn(dir.path())); + + const FileDescriptor fd = + ASSERT_NO_ERRNO_AND_VALUE(Open(file.path().c_str(), O_RDWR)); + + // NOTE(b/157163751): Create another link before unlinking. This is needed for + // the gofer filesystem in gVisor, where open fds will not work once the link + // count hits zero. In VFS2, we end up skipping the gofer test anyway, because + // hard links are not supported for gofer fs. + if (IsRunningOnGvisor()) { + std::string link_path = NewTempAbsPath(); + const int rc = link(file.path().c_str(), link_path.c_str()); + // NOTE(b/34861058): link(2) is only supported on tmpfs in the sandbox. + SKIP_IF(rc != 0 && (errno == EPERM || errno == ENOENT)); + ASSERT_THAT(rc, SyscallSucceeds()); + } + + const FileDescriptor inotify_fd = + ASSERT_NO_ERRNO_AND_VALUE(InotifyInit1(IN_NONBLOCK)); + const int dir_wd = ASSERT_NO_ERRNO_AND_VALUE(InotifyAddWatch( + inotify_fd.get(), dir.path(), IN_ALL_EVENTS | IN_EXCL_UNLINK)); + const int file_wd = ASSERT_NO_ERRNO_AND_VALUE(InotifyAddWatch( + inotify_fd.get(), file.path(), IN_ALL_EVENTS | IN_EXCL_UNLINK)); + + // Even after unlinking, inode-level operations will trigger events regardless + // of IN_EXCL_UNLINK. + ASSERT_THAT(unlink(file.path().c_str()), SyscallSucceeds()); + + // Perform various actions on fd. + ASSERT_THAT(ftruncate(fd.get(), 12345), SyscallSucceeds()); + std::vector<Event> events = + ASSERT_NO_ERRNO_AND_VALUE(DrainEvents(inotify_fd.get())); + EXPECT_THAT(events, Are({ + Event(IN_ATTRIB, file_wd), + Event(IN_DELETE, dir_wd, Basename(file.path())), + Event(IN_MODIFY, dir_wd, Basename(file.path())), + Event(IN_MODIFY, file_wd), + })); + + const struct timeval times[2] = {{1, 0}, {2, 0}}; + ASSERT_THAT(futimes(fd.get(), times), SyscallSucceeds()); + events = ASSERT_NO_ERRNO_AND_VALUE(DrainEvents(inotify_fd.get())); + EXPECT_THAT(events, Are({ + Event(IN_ATTRIB, dir_wd, Basename(file.path())), + Event(IN_ATTRIB, file_wd), + })); + + // S/R is disabled on this entire test due to behavior with unlink; it must + // also be disabled after this point because of fchmod. + ASSERT_THAT(fchmod(fd.get(), 0777), SyscallSucceeds()); + events = ASSERT_NO_ERRNO_AND_VALUE(DrainEvents(inotify_fd.get())); + EXPECT_THAT(events, Are({ + Event(IN_ATTRIB, dir_wd, Basename(file.path())), + Event(IN_ATTRIB, file_wd), + })); +} + +TEST(Inotify, OneShot) { + // TODO(gvisor.dev/issue/1624): IN_ONESHOT not supported in VFS1. + SKIP_IF(IsRunningWithVFS1()); + + const TempPath file = ASSERT_NO_ERRNO_AND_VALUE(TempPath::CreateFile()); + const FileDescriptor inotify_fd = + ASSERT_NO_ERRNO_AND_VALUE(InotifyInit1(IN_NONBLOCK)); + + const int wd = ASSERT_NO_ERRNO_AND_VALUE( + InotifyAddWatch(inotify_fd.get(), file.path(), IN_MODIFY | IN_ONESHOT)); + + // Open an fd, write to it, and then close it. + FileDescriptor fd = ASSERT_NO_ERRNO_AND_VALUE(Open(file.path(), O_WRONLY)); + ASSERT_THAT(write(fd.get(), "x", 1), SyscallSucceedsWithValue(1)); + fd.reset(); + + // We should get a single event followed by IN_IGNORED indicating removal + // of the one-shot watch. Prior activity (i.e. open) that is not in the mask + // should not trigger removal, and activity after removal (i.e. close) should + // not generate events. + std::vector<Event> events = + ASSERT_NO_ERRNO_AND_VALUE(DrainEvents(inotify_fd.get())); + EXPECT_THAT(events, Are({ + Event(IN_MODIFY, wd), + Event(IN_IGNORED, wd), + })); + + // The watch should already have been removed. + EXPECT_THAT(inotify_rm_watch(inotify_fd.get(), wd), + SyscallFailsWithErrno(EINVAL)); +} + +// This test helps verify that the lock order of filesystem and inotify locks +// is respected when inotify instances and watch targets are concurrently being +// destroyed. +TEST(InotifyTest, InotifyAndTargetDestructionDoNotDeadlock_NoRandomSave) { + const DisableSave ds; // Too many syscalls. + + // A file descriptor protected by a mutex. This ensures that while a + // descriptor is in use, it cannot be closed and reused for a different file + // description. + struct atomic_fd { + int fd; + absl::Mutex mu; + }; + + // Set up initial inotify instances. + constexpr int num_fds = 3; + std::vector<atomic_fd> fds(num_fds); + for (int i = 0; i < num_fds; i++) { + int fd; + ASSERT_THAT(fd = inotify_init1(IN_NONBLOCK), SyscallSucceeds()); + fds[i].fd = fd; + } + + // Set up initial watch targets. + std::vector<std::string> paths; + for (int i = 0; i < 3; i++) { + paths.push_back(NewTempAbsPath()); + ASSERT_THAT(mknod(paths[i].c_str(), S_IFREG | 0600, 0), SyscallSucceeds()); + } + + constexpr absl::Duration runtime = absl::Seconds(4); + + // Constantly replace each inotify instance with a new one. + auto replace_fds = [&] { + for (auto start = absl::Now(); absl::Now() - start < runtime;) { + for (auto& afd : fds) { + int new_fd; + ASSERT_THAT(new_fd = inotify_init1(IN_NONBLOCK), SyscallSucceeds()); + absl::MutexLock l(&afd.mu); + ASSERT_THAT(close(afd.fd), SyscallSucceeds()); + afd.fd = new_fd; + for (auto& p : paths) { + // inotify_add_watch may fail if the file at p was deleted. + ASSERT_THAT(inotify_add_watch(afd.fd, p.c_str(), IN_ALL_EVENTS), + AnyOf(SyscallSucceeds(), SyscallFailsWithErrno(ENOENT))); + } + } + sched_yield(); + } + }; + + std::list<ScopedThread> ts; + for (int i = 0; i < 3; i++) { + ts.emplace_back(replace_fds); + } + + // Constantly replace each watch target with a new one. + for (auto start = absl::Now(); absl::Now() - start < runtime;) { + for (auto& p : paths) { + ASSERT_THAT(unlink(p.c_str()), SyscallSucceeds()); + ASSERT_THAT(mknod(p.c_str(), S_IFREG | 0600, 0), SyscallSucceeds()); + } + sched_yield(); + } +} + +// This test helps verify that the lock order of filesystem and inotify locks +// is respected when adding/removing watches occurs concurrently with the +// removal of their targets. +TEST(InotifyTest, AddRemoveUnlinkDoNotDeadlock_NoRandomSave) { + const DisableSave ds; // Too many syscalls. + + // Set up inotify instances. + constexpr int num_fds = 3; + std::vector<int> fds(num_fds); + for (int i = 0; i < num_fds; i++) { + ASSERT_THAT(fds[i] = inotify_init1(IN_NONBLOCK), SyscallSucceeds()); + } + + // Set up initial watch targets. + std::vector<std::string> paths; + for (int i = 0; i < 3; i++) { + paths.push_back(NewTempAbsPath()); + ASSERT_THAT(mknod(paths[i].c_str(), S_IFREG | 0600, 0), SyscallSucceeds()); + } + + constexpr absl::Duration runtime = absl::Seconds(1); + + // Constantly add/remove watches for each inotify instance/watch target pair. + auto add_remove_watches = [&] { + for (auto start = absl::Now(); absl::Now() - start < runtime;) { + for (int fd : fds) { + for (auto& p : paths) { + // Do not assert on inotify_add_watch and inotify_rm_watch. They may + // fail if the file at p was deleted. inotify_add_watch may also fail + // if another thread beat us to adding a watch. + const int wd = inotify_add_watch(fd, p.c_str(), IN_ALL_EVENTS); + if (wd > 0) { + inotify_rm_watch(fd, wd); + } + } + } + sched_yield(); + } + }; + + std::list<ScopedThread> ts; + for (int i = 0; i < 15; i++) { + ts.emplace_back(add_remove_watches); + } + + // Constantly replace each watch target with a new one. + for (auto start = absl::Now(); absl::Now() - start < runtime;) { + for (auto& p : paths) { + ASSERT_THAT(unlink(p.c_str()), SyscallSucceeds()); + ASSERT_THAT(mknod(p.c_str(), S_IFREG | 0600, 0), SyscallSucceeds()); + } + sched_yield(); + } +} + +// This test helps verify that the lock order of filesystem and inotify locks +// is respected when many inotify events and filesystem operations occur +// simultaneously. +TEST(InotifyTest, NotifyNoDeadlock_NoRandomSave) { + const DisableSave ds; // Too many syscalls. + + const TempPath parent = ASSERT_NO_ERRNO_AND_VALUE(TempPath::CreateDir()); + const std::string dir = parent.path(); + + // mu protects file, which will change on rename. + absl::Mutex mu; + std::string file = NewTempAbsPathInDir(dir); + ASSERT_THAT(mknod(file.c_str(), 0644 | S_IFREG, 0), SyscallSucceeds()); + + const absl::Duration runtime = absl::Milliseconds(300); + + // Add/remove watches on dir and file. + ScopedThread add_remove_watches([&] { + const FileDescriptor ifd = + ASSERT_NO_ERRNO_AND_VALUE(InotifyInit1(IN_NONBLOCK)); + int dir_wd = ASSERT_NO_ERRNO_AND_VALUE( + InotifyAddWatch(ifd.get(), dir, IN_ALL_EVENTS)); + int file_wd; + { + absl::ReaderMutexLock l(&mu); + file_wd = ASSERT_NO_ERRNO_AND_VALUE( + InotifyAddWatch(ifd.get(), file, IN_ALL_EVENTS)); + } + for (auto start = absl::Now(); absl::Now() - start < runtime;) { + ASSERT_THAT(inotify_rm_watch(ifd.get(), file_wd), SyscallSucceeds()); + ASSERT_THAT(inotify_rm_watch(ifd.get(), dir_wd), SyscallSucceeds()); + dir_wd = ASSERT_NO_ERRNO_AND_VALUE( + InotifyAddWatch(ifd.get(), dir, IN_ALL_EVENTS)); + { + absl::ReaderMutexLock l(&mu); + file_wd = ASSERT_NO_ERRNO_AND_VALUE( + InotifyAddWatch(ifd.get(), file, IN_ALL_EVENTS)); + } + sched_yield(); + } + }); + + // Modify attributes on dir and file. + ScopedThread stats([&] { + int fd, dir_fd; + { + absl::ReaderMutexLock l(&mu); + ASSERT_THAT(fd = open(file.c_str(), O_RDONLY), SyscallSucceeds()); + } + ASSERT_THAT(dir_fd = open(dir.c_str(), O_RDONLY | O_DIRECTORY), + SyscallSucceeds()); + const struct timeval times[2] = {{1, 0}, {2, 0}}; + + for (auto start = absl::Now(); absl::Now() - start < runtime;) { + { + absl::ReaderMutexLock l(&mu); + EXPECT_THAT(utimes(file.c_str(), times), SyscallSucceeds()); + } + EXPECT_THAT(futimes(fd, times), SyscallSucceeds()); + EXPECT_THAT(utimes(dir.c_str(), times), SyscallSucceeds()); + EXPECT_THAT(futimes(dir_fd, times), SyscallSucceeds()); + sched_yield(); + } + }); + + // Modify extended attributes on dir and file. + ScopedThread xattrs([&] { + // TODO(gvisor.dev/issue/1636): Support extended attributes in runsc gofer. + if (!IsRunningOnGvisor()) { + int fd; + { + absl::ReaderMutexLock l(&mu); + ASSERT_THAT(fd = open(file.c_str(), O_RDONLY), SyscallSucceeds()); + } + + const char* name = "user.test"; + int val = 123; + for (auto start = absl::Now(); absl::Now() - start < runtime;) { + { + absl::ReaderMutexLock l(&mu); + ASSERT_THAT( + setxattr(file.c_str(), name, &val, sizeof(val), /*flags=*/0), + SyscallSucceeds()); + ASSERT_THAT(removexattr(file.c_str(), name), SyscallSucceeds()); + } + + ASSERT_THAT(fsetxattr(fd, name, &val, sizeof(val), /*flags=*/0), + SyscallSucceeds()); + ASSERT_THAT(fremovexattr(fd, name), SyscallSucceeds()); + sched_yield(); + } + } + }); + + // Read and write file's contents. Read and write dir's entries. + ScopedThread read_write([&] { + int fd; + { + absl::ReaderMutexLock l(&mu); + ASSERT_THAT(fd = open(file.c_str(), O_RDWR), SyscallSucceeds()); + } + for (auto start = absl::Now(); absl::Now() - start < runtime;) { + int val = 123; + ASSERT_THAT(write(fd, &val, sizeof(val)), SyscallSucceeds()); + ASSERT_THAT(read(fd, &val, sizeof(val)), SyscallSucceeds()); + TempPath new_file = + ASSERT_NO_ERRNO_AND_VALUE(TempPath::CreateFileIn(dir)); + ASSERT_NO_ERRNO(ListDir(dir, false)); + new_file.reset(); + sched_yield(); + } + }); + + // Rename file. + for (auto start = absl::Now(); absl::Now() - start < runtime;) { + const std::string new_path = NewTempAbsPathInDir(dir); + { + absl::WriterMutexLock l(&mu); + ASSERT_THAT(rename(file.c_str(), new_path.c_str()), SyscallSucceeds()); + file = new_path; + } + sched_yield(); + } +} + } // namespace } // namespace testing } // namespace gvisor diff --git a/test/syscalls/linux/link.cc b/test/syscalls/linux/link.cc index e74fa2ed5..544681168 100644 --- a/test/syscalls/linux/link.cc +++ b/test/syscalls/linux/link.cc @@ -79,8 +79,13 @@ TEST(LinkTest, PermissionDenied) { // Make the file "unsafe" to link by making it only readable, but not // writable. - const auto oldfile = + const auto unwriteable_file = ASSERT_NO_ERRNO_AND_VALUE(TempPath::CreateFileMode(0400)); + const std::string special_path = NewTempAbsPath(); + ASSERT_THAT(mkfifo(special_path.c_str(), 0666), SyscallSucceeds()); + const auto setuid_file = + ASSERT_NO_ERRNO_AND_VALUE(TempPath::CreateFileMode(0666 | S_ISUID)); + const std::string newname = NewTempAbsPath(); // Do setuid in a separate thread so that after finishing this test, the @@ -97,8 +102,14 @@ TEST(LinkTest, PermissionDenied) { EXPECT_THAT(syscall(SYS_setuid, absl::GetFlag(FLAGS_scratch_uid)), SyscallSucceeds()); - EXPECT_THAT(link(oldfile.path().c_str(), newname.c_str()), + EXPECT_THAT(link(unwriteable_file.path().c_str(), newname.c_str()), + SyscallFailsWithErrno(EPERM)); + EXPECT_THAT(link(special_path.c_str(), newname.c_str()), SyscallFailsWithErrno(EPERM)); + if (!IsRunningWithVFS1()) { + EXPECT_THAT(link(setuid_file.path().c_str(), newname.c_str()), + SyscallFailsWithErrno(EPERM)); + } }); } diff --git a/test/syscalls/linux/mknod.cc b/test/syscalls/linux/mknod.cc index 4c45766c7..05dfb375a 100644 --- a/test/syscalls/linux/mknod.cc +++ b/test/syscalls/linux/mknod.cc @@ -15,6 +15,7 @@ #include <errno.h> #include <fcntl.h> #include <sys/stat.h> +#include <sys/types.h> #include <sys/un.h> #include <unistd.h> @@ -39,7 +40,28 @@ TEST(MknodTest, RegularFile) { EXPECT_THAT(mknod(node1.c_str(), 0, 0), SyscallSucceeds()); } -TEST(MknodTest, MknodAtRegularFile) { +TEST(MknodTest, RegularFilePermissions) { + const std::string node = NewTempAbsPath(); + mode_t newUmask = 0077; + umask(newUmask); + + // Attempt to open file with mode 0777. Not specifying file type should create + // a regualar file. + mode_t perms = S_IRWXU | S_IRWXG | S_IRWXO; + EXPECT_THAT(mknod(node.c_str(), perms, 0), SyscallSucceeds()); + + // In the absence of a default ACL, the permissions of the created node are + // (mode & ~umask). -- mknod(2) + mode_t wantPerms = perms & ~newUmask; + struct stat st; + ASSERT_THAT(stat(node.c_str(), &st), SyscallSucceeds()); + ASSERT_EQ(st.st_mode & 0777, wantPerms); + + // "Zero file type is equivalent to type S_IFREG." - mknod(2) + ASSERT_EQ(st.st_mode & S_IFMT, S_IFREG); +} + +TEST(MknodTest, MknodAtFIFO) { const TempPath dir = ASSERT_NO_ERRNO_AND_VALUE(TempPath::CreateDir()); const std::string fifo_relpath = NewTempRelPath(); const std::string fifo = JoinPath(dir.path(), fifo_relpath); @@ -72,7 +94,7 @@ TEST(MknodTest, MknodOnExistingPathFails) { TEST(MknodTest, UnimplementedTypesReturnError) { const std::string path = NewTempAbsPath(); - if (IsRunningOnGvisor()) { + if (IsRunningWithVFS1()) { ASSERT_THAT(mknod(path.c_str(), S_IFSOCK, 0), SyscallFailsWithErrno(EOPNOTSUPP)); } diff --git a/test/syscalls/linux/mount.cc b/test/syscalls/linux/mount.cc index a3e9745cf..46b6f38db 100644 --- a/test/syscalls/linux/mount.cc +++ b/test/syscalls/linux/mount.cc @@ -321,6 +321,42 @@ TEST(MountTest, RenameRemoveMountPoint) { ASSERT_THAT(rmdir(dir.path().c_str()), SyscallFailsWithErrno(EBUSY)); } +TEST(MountTest, MountFuseFilesystemNoDevice) { + SKIP_IF(!ASSERT_NO_ERRNO_AND_VALUE(HaveCapability(CAP_SYS_ADMIN))); + SKIP_IF(IsRunningOnGvisor() && !IsFUSEEnabled()); + + auto const dir = ASSERT_NO_ERRNO_AND_VALUE(TempPath::CreateDir()); + + // Before kernel version 4.16-rc6, FUSE mount is protected by + // capable(CAP_SYS_ADMIN). After this version, it uses + // ns_capable(CAP_SYS_ADMIN) to protect. Before the 4.16 kernel, it was not + // allowed to mount fuse file systems without the global CAP_SYS_ADMIN. + int res = mount("", dir.path().c_str(), "fuse", 0, ""); + SKIP_IF(!IsRunningOnGvisor() && res == -1 && errno == EPERM); + + EXPECT_THAT(mount("", dir.path().c_str(), "fuse", 0, ""), + SyscallFailsWithErrno(EINVAL)); +} + +TEST(MountTest, MountFuseFilesystem) { + SKIP_IF(!ASSERT_NO_ERRNO_AND_VALUE(HaveCapability(CAP_SYS_ADMIN))); + SKIP_IF(IsRunningOnGvisor() && !IsFUSEEnabled()); + + const FileDescriptor fd = + ASSERT_NO_ERRNO_AND_VALUE(Open("/dev/fuse", O_WRONLY)); + std::string mopts = "fd=" + std::to_string(fd.get()); + + auto const dir = ASSERT_NO_ERRNO_AND_VALUE(TempPath::CreateDir()); + + // See comments in MountFuseFilesystemNoDevice for the reason why we skip + // EPERM when running on Linux. + int res = mount("", dir.path().c_str(), "fuse", 0, ""); + SKIP_IF(!IsRunningOnGvisor() && res == -1 && errno == EPERM); + + auto const mount = + ASSERT_NO_ERRNO_AND_VALUE(Mount("", dir.path(), "fuse", 0, mopts, 0)); +} + } // namespace } // namespace testing diff --git a/test/syscalls/linux/open.cc b/test/syscalls/linux/open.cc index 640fe6bfc..bf350946b 100644 --- a/test/syscalls/linux/open.cc +++ b/test/syscalls/linux/open.cc @@ -235,7 +235,7 @@ TEST_F(OpenTest, AppendOnly) { ASSERT_NO_ERRNO_AND_VALUE(Open(test_file_name_, O_RDWR | O_APPEND)); EXPECT_THAT(lseek(fd2.get(), 0, SEEK_CUR), SyscallSucceedsWithValue(0)); - // Then try to write to the first file and make sure the bytes are appended. + // Then try to write to the first fd and make sure the bytes are appended. EXPECT_THAT(WriteFd(fd1.get(), buf.data(), buf.size()), SyscallSucceedsWithValue(buf.size())); @@ -247,7 +247,7 @@ TEST_F(OpenTest, AppendOnly) { EXPECT_THAT(lseek(fd1.get(), 0, SEEK_CUR), SyscallSucceedsWithValue(kBufSize * 2)); - // Then try to write to the second file and make sure the bytes are appended. + // Then try to write to the second fd and make sure the bytes are appended. EXPECT_THAT(WriteFd(fd2.get(), buf.data(), buf.size()), SyscallSucceedsWithValue(buf.size())); @@ -416,6 +416,35 @@ TEST_F(OpenTest, CanTruncateWriteOnlyNoReadPermission_NoRandomSave) { EXPECT_EQ(stat.st_size, 0); } +TEST_F(OpenTest, CanTruncateWithStrangePermissions) { + ASSERT_NO_ERRNO(SetCapability(CAP_DAC_OVERRIDE, false)); + ASSERT_NO_ERRNO(SetCapability(CAP_DAC_READ_SEARCH, false)); + const DisableSave ds; // Permissions are dropped. + std::string path = NewTempAbsPath(); + int fd; + // Create a file without user permissions. + EXPECT_THAT( // SAVE_BELOW + fd = open(path.c_str(), O_CREAT | O_TRUNC | O_WRONLY, 055), + SyscallSucceeds()); + EXPECT_THAT(close(fd), SyscallSucceeds()); + + // Cannot open file because we are owner and have no permissions set. + EXPECT_THAT(open(path.c_str(), O_RDONLY), SyscallFailsWithErrno(EACCES)); + + // We *can* chmod the file, because we are the owner. + EXPECT_THAT(chmod(path.c_str(), 0755), SyscallSucceeds()); + + // Now we can open the file again. + EXPECT_THAT(fd = open(path.c_str(), O_RDWR), SyscallSucceeds()); + EXPECT_THAT(close(fd), SyscallSucceeds()); +} + +TEST_F(OpenTest, OpenNonDirectoryWithTrailingSlash) { + const TempPath file = ASSERT_NO_ERRNO_AND_VALUE(TempPath::CreateFile()); + const std::string bad_path = file.path() + "/"; + EXPECT_THAT(open(bad_path.c_str(), O_RDONLY), SyscallFailsWithErrno(ENOTDIR)); +} + } // namespace } // namespace testing diff --git a/test/syscalls/linux/packet_socket.cc b/test/syscalls/linux/packet_socket.cc index 5ac68feb4..861617ff7 100644 --- a/test/syscalls/linux/packet_socket.cc +++ b/test/syscalls/linux/packet_socket.cc @@ -188,11 +188,12 @@ void ReceiveMessage(int sock, int ifindex) { // sizeof(sockaddr_ll). ASSERT_THAT(src_len, AnyOf(Eq(sizeof(src)), Eq(sizeof(src) - 2))); - // TODO(b/129292371): Verify protocol once we return it. + // TODO(gvisor.dev/issue/173): Verify protocol once we return it. // Verify the source address. EXPECT_EQ(src.sll_family, AF_PACKET); EXPECT_EQ(src.sll_ifindex, ifindex); EXPECT_EQ(src.sll_halen, ETH_ALEN); + EXPECT_EQ(ntohs(src.sll_protocol), ETH_P_IP); // This came from the loopback device, so the address is all 0s. for (int i = 0; i < src.sll_halen; i++) { EXPECT_EQ(src.sll_addr[i], 0); @@ -233,7 +234,7 @@ TEST_P(CookedPacketTest, Receive) { // Send via a packet socket. TEST_P(CookedPacketTest, Send) { - // TODO(b/129292371): Remove once we support packet socket writing. + // TODO(gvisor.dev/issue/173): Remove once we support packet socket writing. SKIP_IF(IsRunningOnGvisor()); // Let's send a UDP packet and receive it using a regular UDP socket. @@ -343,7 +344,7 @@ TEST_P(CookedPacketTest, BindReceive) { } // Double Bind socket. -TEST_P(CookedPacketTest, DoubleBind) { +TEST_P(CookedPacketTest, DoubleBindSucceeds) { struct sockaddr_ll bind_addr = {}; bind_addr.sll_family = AF_PACKET; bind_addr.sll_protocol = htons(GetParam()); @@ -354,12 +355,11 @@ TEST_P(CookedPacketTest, DoubleBind) { SyscallSucceeds()); // Binding socket again should fail. - ASSERT_THAT( - bind(socket_, reinterpret_cast<struct sockaddr*>(&bind_addr), - sizeof(bind_addr)), - // Linux 4.09 returns EINVAL here, but some time before 4.19 it switched - // to EADDRINUSE. - AnyOf(SyscallFailsWithErrno(EADDRINUSE), SyscallFailsWithErrno(EINVAL))); + ASSERT_THAT(bind(socket_, reinterpret_cast<struct sockaddr*>(&bind_addr), + sizeof(bind_addr)), + // Linux 4.09 returns EINVAL here, but some time before 4.19 it + // switched to EADDRINUSE. + SyscallSucceeds()); } // Bind and verify we do not receive data on interface which is not bound @@ -417,6 +417,122 @@ TEST_P(CookedPacketTest, BindDrop) { EXPECT_THAT(RetryEINTR(poll)(&pfd, 1, 1000), SyscallSucceedsWithValue(0)); } +// Verify that we receive outbound packets. This test requires at least one +// non loopback interface so that we can actually capture an outgoing packet. +TEST_P(CookedPacketTest, ReceiveOutbound) { + // Only ETH_P_ALL sockets can receive outbound packets on linux. + SKIP_IF(GetParam() != ETH_P_ALL); + + // Let's use a simple IP payload: a UDP datagram. + FileDescriptor udp_sock = + ASSERT_NO_ERRNO_AND_VALUE(Socket(AF_INET, SOCK_DGRAM, 0)); + + struct ifaddrs* if_addr_list = nullptr; + auto cleanup = Cleanup([&if_addr_list]() { freeifaddrs(if_addr_list); }); + + ASSERT_THAT(getifaddrs(&if_addr_list), SyscallSucceeds()); + + // Get interface other than loopback. + struct ifreq ifr = {}; + for (struct ifaddrs* i = if_addr_list; i; i = i->ifa_next) { + if (strcmp(i->ifa_name, "lo") != 0) { + strncpy(ifr.ifr_name, i->ifa_name, sizeof(ifr.ifr_name)); + break; + } + } + + // Skip if no interface is available other than loopback. + if (strlen(ifr.ifr_name) == 0) { + GTEST_SKIP(); + } + + // Get interface index and name. + EXPECT_THAT(ioctl(socket_, SIOCGIFINDEX, &ifr), SyscallSucceeds()); + EXPECT_NE(ifr.ifr_ifindex, 0); + int ifindex = ifr.ifr_ifindex; + + constexpr int kMACSize = 6; + char hwaddr[kMACSize]; + // Get interface address. + ASSERT_THAT(ioctl(socket_, SIOCGIFHWADDR, &ifr), SyscallSucceeds()); + ASSERT_THAT(ifr.ifr_hwaddr.sa_family, + AnyOf(Eq(ARPHRD_NONE), Eq(ARPHRD_ETHER))); + memcpy(hwaddr, ifr.ifr_hwaddr.sa_data, kMACSize); + + // Just send it to the google dns server 8.8.8.8. It's UDP we don't care + // if it actually gets to the DNS Server we just want to see that we receive + // it on our AF_PACKET socket. + // + // NOTE: We just want to pick an IP that is non-local to avoid having to + // handle ARP as this should cause the UDP packet to be sent to the default + // gateway configured for the system under test. Otherwise the only packet we + // will see is the ARP query unless we picked an IP which will actually + // resolve. The test is a bit brittle but this was the best compromise for + // now. + struct sockaddr_in dest = {}; + ASSERT_EQ(inet_pton(AF_INET, "8.8.8.8", &dest.sin_addr.s_addr), 1); + dest.sin_family = AF_INET; + dest.sin_port = kPort; + EXPECT_THAT(sendto(udp_sock.get(), kMessage, sizeof(kMessage), 0, + reinterpret_cast<struct sockaddr*>(&dest), sizeof(dest)), + SyscallSucceedsWithValue(sizeof(kMessage))); + + // Wait and make sure the socket receives the data. + struct pollfd pfd = {}; + pfd.fd = socket_; + pfd.events = POLLIN; + EXPECT_THAT(RetryEINTR(poll)(&pfd, 1, 1000), SyscallSucceedsWithValue(1)); + + // Now read and check that the packet is the one we just sent. + // Read and verify the data. + constexpr size_t packet_size = + sizeof(struct iphdr) + sizeof(struct udphdr) + sizeof(kMessage); + char buf[64]; + struct sockaddr_ll src = {}; + socklen_t src_len = sizeof(src); + ASSERT_THAT(recvfrom(socket_, buf, sizeof(buf), 0, + reinterpret_cast<struct sockaddr*>(&src), &src_len), + SyscallSucceedsWithValue(packet_size)); + + // sockaddr_ll ends with an 8 byte physical address field, but ethernet + // addresses only use 6 bytes. Linux used to return sizeof(sockaddr_ll)-2 + // here, but since commit b2cf86e1563e33a14a1c69b3e508d15dc12f804c returns + // sizeof(sockaddr_ll). + ASSERT_THAT(src_len, AnyOf(Eq(sizeof(src)), Eq(sizeof(src) - 2))); + + // Verify the source address. + EXPECT_EQ(src.sll_family, AF_PACKET); + EXPECT_EQ(src.sll_ifindex, ifindex); + EXPECT_EQ(src.sll_halen, ETH_ALEN); + EXPECT_EQ(ntohs(src.sll_protocol), ETH_P_IP); + EXPECT_EQ(src.sll_pkttype, PACKET_OUTGOING); + // Verify the link address of the interface matches that of the non + // non loopback interface address we stored above. + for (int i = 0; i < src.sll_halen; i++) { + EXPECT_EQ(src.sll_addr[i], hwaddr[i]); + } + + // Verify the IP header. + struct iphdr ip = {}; + memcpy(&ip, buf, sizeof(ip)); + EXPECT_EQ(ip.ihl, 5); + EXPECT_EQ(ip.version, 4); + EXPECT_EQ(ip.tot_len, htons(packet_size)); + EXPECT_EQ(ip.protocol, IPPROTO_UDP); + EXPECT_EQ(ip.daddr, dest.sin_addr.s_addr); + EXPECT_NE(ip.saddr, htonl(INADDR_LOOPBACK)); + + // Verify the UDP header. + struct udphdr udp = {}; + memcpy(&udp, buf + sizeof(iphdr), sizeof(udp)); + EXPECT_EQ(udp.dest, kPort); + EXPECT_EQ(udp.len, htons(sizeof(udphdr) + sizeof(kMessage))); + + // Verify the payload. + char* payload = reinterpret_cast<char*>(buf + sizeof(iphdr) + sizeof(udphdr)); + EXPECT_EQ(strncmp(payload, kMessage, sizeof(kMessage)), 0); +} + // Bind with invalid address. TEST_P(CookedPacketTest, BindFail) { // Null address. diff --git a/test/syscalls/linux/packet_socket_raw.cc b/test/syscalls/linux/packet_socket_raw.cc index d258d353c..a11a03415 100644 --- a/test/syscalls/linux/packet_socket_raw.cc +++ b/test/syscalls/linux/packet_socket_raw.cc @@ -14,6 +14,9 @@ #include <arpa/inet.h> #include <linux/capability.h> +#ifndef __fuchsia__ +#include <linux/filter.h> +#endif // __fuchsia__ #include <linux/if_arp.h> #include <linux/if_packet.h> #include <net/ethernet.h> @@ -97,7 +100,7 @@ class RawPacketTest : public ::testing::TestWithParam<int> { int GetLoopbackIndex(); // The socket used for both reading and writing. - int socket_; + int s_; }; void RawPacketTest::SetUp() { @@ -108,34 +111,58 @@ void RawPacketTest::SetUp() { } if (!IsRunningOnGvisor()) { + // Ensure that looped back packets aren't rejected by the kernel. FileDescriptor acceptLocal = ASSERT_NO_ERRNO_AND_VALUE( - Open("/proc/sys/net/ipv4/conf/lo/accept_local", O_RDONLY)); + Open("/proc/sys/net/ipv4/conf/lo/accept_local", O_RDWR)); FileDescriptor routeLocalnet = ASSERT_NO_ERRNO_AND_VALUE( - Open("/proc/sys/net/ipv4/conf/lo/route_localnet", O_RDONLY)); + Open("/proc/sys/net/ipv4/conf/lo/route_localnet", O_RDWR)); char enabled; ASSERT_THAT(read(acceptLocal.get(), &enabled, 1), SyscallSucceedsWithValue(1)); - ASSERT_EQ(enabled, '1'); + if (enabled != '1') { + enabled = '1'; + ASSERT_THAT(lseek(acceptLocal.get(), 0, SEEK_SET), + SyscallSucceedsWithValue(0)); + ASSERT_THAT(write(acceptLocal.get(), &enabled, 1), + SyscallSucceedsWithValue(1)); + ASSERT_THAT(lseek(acceptLocal.get(), 0, SEEK_SET), + SyscallSucceedsWithValue(0)); + ASSERT_THAT(read(acceptLocal.get(), &enabled, 1), + SyscallSucceedsWithValue(1)); + ASSERT_EQ(enabled, '1'); + } + ASSERT_THAT(read(routeLocalnet.get(), &enabled, 1), SyscallSucceedsWithValue(1)); - ASSERT_EQ(enabled, '1'); + if (enabled != '1') { + enabled = '1'; + ASSERT_THAT(lseek(routeLocalnet.get(), 0, SEEK_SET), + SyscallSucceedsWithValue(0)); + ASSERT_THAT(write(routeLocalnet.get(), &enabled, 1), + SyscallSucceedsWithValue(1)); + ASSERT_THAT(lseek(routeLocalnet.get(), 0, SEEK_SET), + SyscallSucceedsWithValue(0)); + ASSERT_THAT(read(routeLocalnet.get(), &enabled, 1), + SyscallSucceedsWithValue(1)); + ASSERT_EQ(enabled, '1'); + } } - ASSERT_THAT(socket_ = socket(AF_PACKET, SOCK_RAW, htons(GetParam())), + ASSERT_THAT(s_ = socket(AF_PACKET, SOCK_RAW, htons(GetParam())), SyscallSucceeds()); } void RawPacketTest::TearDown() { // TearDown will be run even if we skip the test. if (ASSERT_NO_ERRNO_AND_VALUE(HaveCapability(CAP_NET_RAW))) { - EXPECT_THAT(close(socket_), SyscallSucceeds()); + EXPECT_THAT(close(s_), SyscallSucceeds()); } } int RawPacketTest::GetLoopbackIndex() { struct ifreq ifr; snprintf(ifr.ifr_name, IFNAMSIZ, "lo"); - EXPECT_THAT(ioctl(socket_, SIOCGIFINDEX, &ifr), SyscallSucceeds()); + EXPECT_THAT(ioctl(s_, SIOCGIFINDEX, &ifr), SyscallSucceeds()); EXPECT_NE(ifr.ifr_ifindex, 0); return ifr.ifr_ifindex; } @@ -149,7 +176,7 @@ TEST_P(RawPacketTest, Receive) { // Wait for the socket to become readable. struct pollfd pfd = {}; - pfd.fd = socket_; + pfd.fd = s_; pfd.events = POLLIN; EXPECT_THAT(RetryEINTR(poll)(&pfd, 1, 2000), SyscallSucceedsWithValue(1)); @@ -159,7 +186,7 @@ TEST_P(RawPacketTest, Receive) { char buf[64]; struct sockaddr_ll src = {}; socklen_t src_len = sizeof(src); - ASSERT_THAT(recvfrom(socket_, buf, sizeof(buf), 0, + ASSERT_THAT(recvfrom(s_, buf, sizeof(buf), 0, reinterpret_cast<struct sockaddr*>(&src), &src_len), SyscallSucceedsWithValue(packet_size)); // sockaddr_ll ends with an 8 byte physical address field, but ethernet @@ -168,11 +195,12 @@ TEST_P(RawPacketTest, Receive) { // sizeof(sockaddr_ll). ASSERT_THAT(src_len, AnyOf(Eq(sizeof(src)), Eq(sizeof(src) - 2))); - // TODO(b/129292371): Verify protocol once we return it. + // TODO(gvisor.dev/issue/173): Verify protocol once we return it. // Verify the source address. EXPECT_EQ(src.sll_family, AF_PACKET); EXPECT_EQ(src.sll_ifindex, GetLoopbackIndex()); EXPECT_EQ(src.sll_halen, ETH_ALEN); + EXPECT_EQ(ntohs(src.sll_protocol), ETH_P_IP); // This came from the loopback device, so the address is all 0s. for (int i = 0; i < src.sll_halen; i++) { EXPECT_EQ(src.sll_addr[i], 0); @@ -212,7 +240,7 @@ TEST_P(RawPacketTest, Receive) { // Send via a packet socket. TEST_P(RawPacketTest, Send) { - // TODO(b/129292371): Remove once we support packet socket writing. + // TODO(gvisor.dev/issue/173): Remove once we support packet socket writing. SKIP_IF(IsRunningOnGvisor()); // Let's send a UDP packet and receive it using a regular UDP socket. @@ -277,7 +305,7 @@ TEST_P(RawPacketTest, Send) { sizeof(kMessage)); // Send it. - ASSERT_THAT(sendto(socket_, send_buf, sizeof(send_buf), 0, + ASSERT_THAT(sendto(s_, send_buf, sizeof(send_buf), 0, reinterpret_cast<struct sockaddr*>(&dest), sizeof(dest)), SyscallSucceedsWithValue(sizeof(send_buf))); @@ -286,13 +314,13 @@ TEST_P(RawPacketTest, Send) { pfd.fd = udp_sock.get(); pfd.events = POLLIN; ASSERT_THAT(RetryEINTR(poll)(&pfd, 1, 5000), SyscallSucceedsWithValue(1)); - pfd.fd = socket_; + pfd.fd = s_; pfd.events = POLLIN; ASSERT_THAT(RetryEINTR(poll)(&pfd, 1, 5000), SyscallSucceedsWithValue(1)); // Receive on the packet socket. char recv_buf[sizeof(send_buf)]; - ASSERT_THAT(recv(socket_, recv_buf, sizeof(recv_buf), 0), + ASSERT_THAT(recv(s_, recv_buf, sizeof(recv_buf), 0), SyscallSucceedsWithValue(sizeof(recv_buf))); ASSERT_EQ(memcmp(recv_buf, send_buf, sizeof(send_buf)), 0); @@ -309,6 +337,318 @@ TEST_P(RawPacketTest, Send) { EXPECT_EQ(src.sin_addr.s_addr, htonl(INADDR_LOOPBACK)); } +// Check that setting SO_RCVBUF below min is clamped to the minimum +// receive buffer size. +TEST_P(RawPacketTest, SetSocketRecvBufBelowMin) { + SKIP_IF(!ASSERT_NO_ERRNO_AND_VALUE(HaveCapability(CAP_NET_RAW))); + + // Discover minimum receive buf size by trying to set it to zero. + // See: + // https://github.com/torvalds/linux/blob/a5dc8300df75e8b8384b4c82225f1e4a0b4d9b55/net/core/sock.c#L820 + constexpr int kRcvBufSz = 0; + ASSERT_THAT( + setsockopt(s_, SOL_SOCKET, SO_RCVBUF, &kRcvBufSz, sizeof(kRcvBufSz)), + SyscallSucceeds()); + + int min = 0; + socklen_t min_len = sizeof(min); + ASSERT_THAT(getsockopt(s_, SOL_SOCKET, SO_RCVBUF, &min, &min_len), + SyscallSucceeds()); + + // Linux doubles the value so let's use a value that when doubled will still + // be smaller than min. + int below_min = min / 2 - 1; + ASSERT_THAT( + setsockopt(s_, SOL_SOCKET, SO_RCVBUF, &below_min, sizeof(below_min)), + SyscallSucceeds()); + + int val = 0; + socklen_t val_len = sizeof(val); + ASSERT_THAT(getsockopt(s_, SOL_SOCKET, SO_RCVBUF, &val, &val_len), + SyscallSucceeds()); + + ASSERT_EQ(min, val); +} + +// Check that setting SO_RCVBUF above max is clamped to the maximum +// receive buffer size. +TEST_P(RawPacketTest, SetSocketRecvBufAboveMax) { + SKIP_IF(!ASSERT_NO_ERRNO_AND_VALUE(HaveCapability(CAP_NET_RAW))); + + // Discover max buf size by trying to set the largest possible buffer size. + constexpr int kRcvBufSz = 0xffffffff; + ASSERT_THAT( + setsockopt(s_, SOL_SOCKET, SO_RCVBUF, &kRcvBufSz, sizeof(kRcvBufSz)), + SyscallSucceeds()); + + int max = 0; + socklen_t max_len = sizeof(max); + ASSERT_THAT(getsockopt(s_, SOL_SOCKET, SO_RCVBUF, &max, &max_len), + SyscallSucceeds()); + + int above_max = max + 1; + ASSERT_THAT( + setsockopt(s_, SOL_SOCKET, SO_RCVBUF, &above_max, sizeof(above_max)), + SyscallSucceeds()); + + int val = 0; + socklen_t val_len = sizeof(val); + ASSERT_THAT(getsockopt(s_, SOL_SOCKET, SO_RCVBUF, &val, &val_len), + SyscallSucceeds()); + ASSERT_EQ(max, val); +} + +// Check that setting SO_RCVBUF min <= kRcvBufSz <= max is honored. +TEST_P(RawPacketTest, SetSocketRecvBuf) { + SKIP_IF(!ASSERT_NO_ERRNO_AND_VALUE(HaveCapability(CAP_NET_RAW))); + + int max = 0; + int min = 0; + { + // Discover max buf size by trying to set a really large buffer size. + constexpr int kRcvBufSz = 0xffffffff; + ASSERT_THAT( + setsockopt(s_, SOL_SOCKET, SO_RCVBUF, &kRcvBufSz, sizeof(kRcvBufSz)), + SyscallSucceeds()); + + max = 0; + socklen_t max_len = sizeof(max); + ASSERT_THAT(getsockopt(s_, SOL_SOCKET, SO_RCVBUF, &max, &max_len), + SyscallSucceeds()); + } + + { + // Discover minimum buffer size by trying to set a zero size receive buffer + // size. + // See: + // https://github.com/torvalds/linux/blob/a5dc8300df75e8b8384b4c82225f1e4a0b4d9b55/net/core/sock.c#L820 + constexpr int kRcvBufSz = 0; + ASSERT_THAT( + setsockopt(s_, SOL_SOCKET, SO_RCVBUF, &kRcvBufSz, sizeof(kRcvBufSz)), + SyscallSucceeds()); + + socklen_t min_len = sizeof(min); + ASSERT_THAT(getsockopt(s_, SOL_SOCKET, SO_RCVBUF, &min, &min_len), + SyscallSucceeds()); + } + + int quarter_sz = min + (max - min) / 4; + ASSERT_THAT( + setsockopt(s_, SOL_SOCKET, SO_RCVBUF, &quarter_sz, sizeof(quarter_sz)), + SyscallSucceeds()); + + int val = 0; + socklen_t val_len = sizeof(val); + ASSERT_THAT(getsockopt(s_, SOL_SOCKET, SO_RCVBUF, &val, &val_len), + SyscallSucceeds()); + + // Linux doubles the value set by SO_SNDBUF/SO_RCVBUF. + // TODO(gvisor.dev/issue/2926): Remove when Netstack matches linux behavior. + if (!IsRunningOnGvisor()) { + quarter_sz *= 2; + } + ASSERT_EQ(quarter_sz, val); +} + +// Check that setting SO_SNDBUF below min is clamped to the minimum +// receive buffer size. +TEST_P(RawPacketTest, SetSocketSendBufBelowMin) { + SKIP_IF(!ASSERT_NO_ERRNO_AND_VALUE(HaveCapability(CAP_NET_RAW))); + + // Discover minimum buffer size by trying to set it to zero. + constexpr int kSndBufSz = 0; + ASSERT_THAT( + setsockopt(s_, SOL_SOCKET, SO_SNDBUF, &kSndBufSz, sizeof(kSndBufSz)), + SyscallSucceeds()); + + int min = 0; + socklen_t min_len = sizeof(min); + ASSERT_THAT(getsockopt(s_, SOL_SOCKET, SO_SNDBUF, &min, &min_len), + SyscallSucceeds()); + + // Linux doubles the value so let's use a value that when doubled will still + // be smaller than min. + int below_min = min / 2 - 1; + ASSERT_THAT( + setsockopt(s_, SOL_SOCKET, SO_SNDBUF, &below_min, sizeof(below_min)), + SyscallSucceeds()); + + int val = 0; + socklen_t val_len = sizeof(val); + ASSERT_THAT(getsockopt(s_, SOL_SOCKET, SO_SNDBUF, &val, &val_len), + SyscallSucceeds()); + + ASSERT_EQ(min, val); +} + +// Check that setting SO_SNDBUF above max is clamped to the maximum +// send buffer size. +TEST_P(RawPacketTest, SetSocketSendBufAboveMax) { + SKIP_IF(!ASSERT_NO_ERRNO_AND_VALUE(HaveCapability(CAP_NET_RAW))); + + // Discover maximum buffer size by trying to set it to a large value. + constexpr int kSndBufSz = 0xffffffff; + ASSERT_THAT( + setsockopt(s_, SOL_SOCKET, SO_SNDBUF, &kSndBufSz, sizeof(kSndBufSz)), + SyscallSucceeds()); + + int max = 0; + socklen_t max_len = sizeof(max); + ASSERT_THAT(getsockopt(s_, SOL_SOCKET, SO_SNDBUF, &max, &max_len), + SyscallSucceeds()); + + int above_max = max + 1; + ASSERT_THAT( + setsockopt(s_, SOL_SOCKET, SO_SNDBUF, &above_max, sizeof(above_max)), + SyscallSucceeds()); + + int val = 0; + socklen_t val_len = sizeof(val); + ASSERT_THAT(getsockopt(s_, SOL_SOCKET, SO_SNDBUF, &val, &val_len), + SyscallSucceeds()); + ASSERT_EQ(max, val); +} + +// Check that setting SO_SNDBUF min <= kSndBufSz <= max is honored. +TEST_P(RawPacketTest, SetSocketSendBuf) { + SKIP_IF(!ASSERT_NO_ERRNO_AND_VALUE(HaveCapability(CAP_NET_RAW))); + + int max = 0; + int min = 0; + { + // Discover maximum buffer size by trying to set it to a large value. + constexpr int kSndBufSz = 0xffffffff; + ASSERT_THAT( + setsockopt(s_, SOL_SOCKET, SO_SNDBUF, &kSndBufSz, sizeof(kSndBufSz)), + SyscallSucceeds()); + + max = 0; + socklen_t max_len = sizeof(max); + ASSERT_THAT(getsockopt(s_, SOL_SOCKET, SO_SNDBUF, &max, &max_len), + SyscallSucceeds()); + } + + { + // Discover minimum buffer size by trying to set it to zero. + constexpr int kSndBufSz = 0; + ASSERT_THAT( + setsockopt(s_, SOL_SOCKET, SO_SNDBUF, &kSndBufSz, sizeof(kSndBufSz)), + SyscallSucceeds()); + + socklen_t min_len = sizeof(min); + ASSERT_THAT(getsockopt(s_, SOL_SOCKET, SO_SNDBUF, &min, &min_len), + SyscallSucceeds()); + } + + int quarter_sz = min + (max - min) / 4; + ASSERT_THAT( + setsockopt(s_, SOL_SOCKET, SO_SNDBUF, &quarter_sz, sizeof(quarter_sz)), + SyscallSucceeds()); + + int val = 0; + socklen_t val_len = sizeof(val); + ASSERT_THAT(getsockopt(s_, SOL_SOCKET, SO_SNDBUF, &val, &val_len), + SyscallSucceeds()); + + // Linux doubles the value set by SO_SNDBUF/SO_RCVBUF. + // TODO(gvisor.dev/issue/2926): Remove the gvisor special casing when Netstack + // matches linux behavior. + if (!IsRunningOnGvisor()) { + quarter_sz *= 2; + } + + ASSERT_EQ(quarter_sz, val); +} + +TEST_P(RawPacketTest, GetSocketError) { + SKIP_IF(!ASSERT_NO_ERRNO_AND_VALUE(HaveCapability(CAP_NET_RAW))); + + int val = 0; + socklen_t val_len = sizeof(val); + ASSERT_THAT(getsockopt(s_, SOL_SOCKET, SO_ERROR, &val, &val_len), + SyscallSucceeds()); + ASSERT_EQ(val, 0); +} + +TEST_P(RawPacketTest, GetSocketErrorBind) { + SKIP_IF(!ASSERT_NO_ERRNO_AND_VALUE(HaveCapability(CAP_NET_RAW))); + + { + // Bind to the loopback device. + struct sockaddr_ll bind_addr = {}; + bind_addr.sll_family = AF_PACKET; + bind_addr.sll_protocol = htons(GetParam()); + bind_addr.sll_ifindex = GetLoopbackIndex(); + + ASSERT_THAT(bind(s_, reinterpret_cast<struct sockaddr*>(&bind_addr), + sizeof(bind_addr)), + SyscallSucceeds()); + + // SO_ERROR should return no errors. + int val = 0; + socklen_t val_len = sizeof(val); + ASSERT_THAT(getsockopt(s_, SOL_SOCKET, SO_ERROR, &val, &val_len), + SyscallSucceeds()); + ASSERT_EQ(val, 0); + } + + { + // Now try binding to an invalid interface. + struct sockaddr_ll bind_addr = {}; + bind_addr.sll_family = AF_PACKET; + bind_addr.sll_protocol = htons(GetParam()); + bind_addr.sll_ifindex = 0xffff; // Just pick a really large number. + + // Binding should fail with EINVAL + ASSERT_THAT(bind(s_, reinterpret_cast<struct sockaddr*>(&bind_addr), + sizeof(bind_addr)), + SyscallFailsWithErrno(ENODEV)); + + // SO_ERROR does not return error when the device is invalid. + // On Linux there is just one odd ball condition where this can return + // an error where the device was valid and then removed or disabled + // between the first check for index and the actual registration of + // the packet endpoint. On Netstack this is not possible as the stack + // global mutex is held during registration and check. + int val = 0; + socklen_t val_len = sizeof(val); + ASSERT_THAT(getsockopt(s_, SOL_SOCKET, SO_ERROR, &val, &val_len), + SyscallSucceeds()); + ASSERT_EQ(val, 0); + } +} + +#ifndef __fuchsia__ + +TEST_P(RawPacketTest, SetSocketDetachFilterNoInstalledFilter) { + // TODO(gvisor.dev/2746): Support SO_ATTACH_FILTER/SO_DETACH_FILTER. + // + // gVisor returns no error on SO_DETACH_FILTER even if there is no filter + // attached unlike linux which does return ENOENT in such cases. This is + // because gVisor doesn't support SO_ATTACH_FILTER and just silently returns + // success. + if (IsRunningOnGvisor()) { + constexpr int val = 0; + ASSERT_THAT(setsockopt(s_, SOL_SOCKET, SO_DETACH_FILTER, &val, sizeof(val)), + SyscallSucceeds()); + return; + } + constexpr int val = 0; + ASSERT_THAT(setsockopt(s_, SOL_SOCKET, SO_DETACH_FILTER, &val, sizeof(val)), + SyscallFailsWithErrno(ENOENT)); +} + +TEST_P(RawPacketTest, GetSocketDetachFilter) { + SKIP_IF(!ASSERT_NO_ERRNO_AND_VALUE(HaveCapability(CAP_NET_RAW))); + + int val = 0; + socklen_t val_len = sizeof(val); + ASSERT_THAT(getsockopt(s_, SOL_SOCKET, SO_DETACH_FILTER, &val, &val_len), + SyscallFailsWithErrno(ENOPROTOOPT)); +} + +#endif // __fuchsia__ + INSTANTIATE_TEST_SUITE_P(AllInetTests, RawPacketTest, ::testing::Values(ETH_P_IP, ETH_P_ALL)); diff --git a/test/syscalls/linux/ping_socket.cc b/test/syscalls/linux/ping_socket.cc new file mode 100644 index 000000000..a9bfdb37b --- /dev/null +++ b/test/syscalls/linux/ping_socket.cc @@ -0,0 +1,91 @@ +// Copyright 2020 The gVisor Authors. +// +// Licensed under the Apache License, Version 2.0 (the "License"); +// you may not use this file except in compliance with the License. +// You may obtain a copy of the License at +// +// http://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, software +// distributed under the License is distributed on an "AS IS" BASIS, +// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +// See the License for the specific language governing permissions and +// limitations under the License. + +#include <netinet/in.h> +#include <netinet/ip.h> +#include <netinet/ip_icmp.h> +#include <sys/socket.h> +#include <sys/types.h> +#include <unistd.h> + +#include <vector> + +#include "gtest/gtest.h" +#include "test/syscalls/linux/socket_test_util.h" +#include "test/util/file_descriptor.h" +#include "test/util/save_util.h" +#include "test/util/test_util.h" + +namespace gvisor { +namespace testing { +namespace { + +class PingSocket : public ::testing::Test { + protected: + // Creates a socket to be used in tests. + void SetUp() override; + + // Closes the socket created by SetUp(). + void TearDown() override; + + // The loopback address. + struct sockaddr_in addr_; +}; + +void PingSocket::SetUp() { + // On some hosts ping sockets are restricted to specific groups using the + // sysctl "ping_group_range". + int s = socket(AF_INET, SOCK_RAW, IPPROTO_ICMP); + if (s < 0 && errno == EPERM) { + GTEST_SKIP(); + } + close(s); + + addr_ = {}; + // Just a random port as the destination port number is irrelevant for ping + // sockets. + addr_.sin_port = 12345; + addr_.sin_addr.s_addr = htonl(INADDR_LOOPBACK); + addr_.sin_family = AF_INET; +} + +void PingSocket::TearDown() {} + +// Test ICMP port exhaustion returns EAGAIN. +// +// We disable both random/cooperative S/R for this test as it makes way too many +// syscalls. +TEST_F(PingSocket, ICMPPortExhaustion_NoRandomSave) { + DisableSave ds; + std::vector<FileDescriptor> sockets; + constexpr int kSockets = 65536; + addr_.sin_port = 0; + for (int i = 0; i < kSockets; i++) { + auto s = + ASSERT_NO_ERRNO_AND_VALUE(Socket(AF_INET, SOCK_DGRAM, IPPROTO_ICMP)); + int ret = connect(s.get(), reinterpret_cast<struct sockaddr*>(&addr_), + sizeof(addr_)); + if (ret == 0) { + sockets.push_back(std::move(s)); + continue; + } + ASSERT_THAT(ret, SyscallFailsWithErrno(EAGAIN)); + break; + } +} + +} // namespace + +} // namespace testing +} // namespace gvisor diff --git a/test/syscalls/linux/poll.cc b/test/syscalls/linux/poll.cc index 1e35a4a8b..7a316427d 100644 --- a/test/syscalls/linux/poll.cc +++ b/test/syscalls/linux/poll.cc @@ -259,9 +259,9 @@ TEST_F(PollTest, Nfds) { TEST_PCHECK(getrlimit(RLIMIT_NOFILE, &rlim) == 0); // gVisor caps the number of FDs that epoll can use beyond RLIMIT_NOFILE. - constexpr rlim_t gVisorMax = 1048576; - if (rlim.rlim_cur > gVisorMax) { - rlim.rlim_cur = gVisorMax; + constexpr rlim_t maxFD = 4096; + if (rlim.rlim_cur > maxFD) { + rlim.rlim_cur = maxFD; TEST_PCHECK(setrlimit(RLIMIT_NOFILE, &rlim) == 0); } diff --git a/test/syscalls/linux/proc.cc b/test/syscalls/linux/proc.cc index 63642880a..d6b875dbf 100644 --- a/test/syscalls/linux/proc.cc +++ b/test/syscalls/linux/proc.cc @@ -754,8 +754,53 @@ TEST(ProcCpuinfo, RequiredFieldsArePresent) { } } -TEST(ProcCpuinfo, DeniesWrite) { - EXPECT_THAT(open("/proc/cpuinfo", O_WRONLY), SyscallFailsWithErrno(EACCES)); +TEST(ProcCpuinfo, DeniesWriteNonRoot) { + SKIP_IF(!ASSERT_NO_ERRNO_AND_VALUE(HaveCapability(CAP_FOWNER))); + + // Do setuid in a separate thread so that after finishing this test, the + // process can still open files the test harness created before starting this + // test. Otherwise, the files are created by root (UID before the test), but + // cannot be opened by the `uid` set below after the test. After calling + // setuid(non-zero-UID), there is no way to get root privileges back. + ScopedThread([&] { + // Use syscall instead of glibc setuid wrapper because we want this setuid + // call to only apply to this task. POSIX threads, however, require that all + // threads have the same UIDs, so using the setuid wrapper sets all threads' + // real UID. + // Also drops capabilities. + constexpr int kNobody = 65534; + EXPECT_THAT(syscall(SYS_setuid, kNobody), SyscallSucceeds()); + EXPECT_THAT(open("/proc/cpuinfo", O_WRONLY), SyscallFailsWithErrno(EACCES)); + // TODO(gvisor.dev/issue/1193): Properly support setting size attributes in + // kernfs. + if (!IsRunningOnGvisor() || IsRunningWithVFS1()) { + EXPECT_THAT(truncate("/proc/cpuinfo", 123), + SyscallFailsWithErrno(EACCES)); + } + }); +} + +// With root privileges, it is possible to open /proc/cpuinfo with write mode, +// but all write operations will return EIO. +TEST(ProcCpuinfo, DeniesWriteRoot) { + // VFS1 does not behave differently for root/non-root. + SKIP_IF(IsRunningWithVFS1()); + SKIP_IF(!ASSERT_NO_ERRNO_AND_VALUE(HaveCapability(CAP_FOWNER))); + + int fd; + EXPECT_THAT(fd = open("/proc/cpuinfo", O_WRONLY), SyscallSucceeds()); + if (fd > 0) { + EXPECT_THAT(write(fd, "x", 1), SyscallFailsWithErrno(EIO)); + EXPECT_THAT(pwrite(fd, "x", 1, 123), SyscallFailsWithErrno(EIO)); + } + // TODO(gvisor.dev/issue/1193): Properly support setting size attributes in + // kernfs. + if (!IsRunningOnGvisor() || IsRunningWithVFS1()) { + if (fd > 0) { + EXPECT_THAT(ftruncate(fd, 123), SyscallFailsWithErrno(EIO)); + } + EXPECT_THAT(truncate("/proc/cpuinfo", 123), SyscallFailsWithErrno(EIO)); + } } // Sanity checks that uptime is present. @@ -1921,7 +1966,7 @@ TEST(ProcSelfMountinfo, RequiredFieldsArePresent) { AllOf( // Root mount. ContainsRegex( - R"([0-9]+ [0-9]+ [0-9]+:[0-9]+ / / (rw|ro).*- \S+ \S+ (rw|ro)\S*)"), + R"([0-9]+ [0-9]+ [0-9]+:[0-9]+ /\S* / (rw|ro).*- \S+ \S+ (rw|ro)\S*)"), // Proc mount - always rw. ContainsRegex( R"([0-9]+ [0-9]+ [0-9]+:[0-9]+ / /proc rw.*- \S+ \S+ rw\S*)"))); @@ -1967,30 +2012,41 @@ void CheckDuplicatesRecursively(std::string path) { errno = 0; struct dirent* dp = readdir(dir); if (dp == nullptr) { + // Linux will return EINVAL when calling getdents on a /proc/tid/net + // file corresponding to a zombie task. + // See fs/proc/proc_net.c:proc_tgid_net_readdir(). + // + // We just ignore the directory in this case. + if (errno == EINVAL && absl::StartsWith(path, "/proc/") && + absl::EndsWith(path, "/net")) { + break; + } + + // Otherwise, no errors are allowed. ASSERT_EQ(errno, 0) << path; break; // We're done. } - if (strcmp(dp->d_name, ".") == 0 || strcmp(dp->d_name, "..") == 0) { + const std::string name = dp->d_name; + + if (name == "." || name == "..") { continue; } // Ignore a duplicate entry if it isn't the last attempt. if (i == max_attempts - 1) { - ASSERT_EQ(children.find(std::string(dp->d_name)), children.end()) - << absl::StrCat(path, "/", dp->d_name); - } else if (children.find(std::string(dp->d_name)) != children.end()) { + ASSERT_EQ(children.find(name), children.end()) + << absl::StrCat(path, "/", name); + } else if (children.find(name) != children.end()) { std::cerr << "Duplicate entry: " << i << ":" - << absl::StrCat(path, "/", dp->d_name) << std::endl; + << absl::StrCat(path, "/", name) << std::endl; success = false; break; } - children.insert(std::string(dp->d_name)); - - ASSERT_NE(dp->d_type, DT_UNKNOWN); + children.insert(name); if (dp->d_type == DT_DIR) { - child_dirs.push_back(std::string(dp->d_name)); + child_dirs.push_back(name); } } if (success) { diff --git a/test/syscalls/linux/proc_net.cc b/test/syscalls/linux/proc_net.cc index cac394910..3377b65cf 100644 --- a/test/syscalls/linux/proc_net.cc +++ b/test/syscalls/linux/proc_net.cc @@ -20,8 +20,13 @@ #include <sys/syscall.h> #include <sys/types.h> +#include <vector> + #include "gtest/gtest.h" +#include "absl/strings/numbers.h" +#include "absl/strings/str_cat.h" #include "absl/strings/str_split.h" +#include "absl/strings/string_view.h" #include "absl/time/clock.h" #include "test/syscalls/linux/socket_test_util.h" #include "test/util/capability_util.h" @@ -404,6 +409,74 @@ TEST(ProcNetSnmp, UdpIn_NoRandomSave) { EXPECT_EQ(oldInDatagrams, newInDatagrams - 1); } +TEST(ProcNetSnmp, CheckNetStat) { + // TODO(b/155123175): SNMP and netstat don't work on gVisor. + SKIP_IF(IsRunningOnGvisor()); + + std::string contents = + ASSERT_NO_ERRNO_AND_VALUE(GetContents("/proc/net/netstat")); + + int name_count = 0; + int value_count = 0; + std::vector<absl::string_view> lines = absl::StrSplit(contents, '\n'); + for (int i = 0; i + 1 < lines.size(); i += 2) { + std::vector<absl::string_view> names = + absl::StrSplit(lines[i], absl::ByAnyChar("\t ")); + std::vector<absl::string_view> values = + absl::StrSplit(lines[i + 1], absl::ByAnyChar("\t ")); + EXPECT_EQ(names.size(), values.size()) << " mismatch in lines '" << lines[i] + << "' and '" << lines[i + 1] << "'"; + for (int j = 0; j < names.size() && j < values.size(); ++j) { + if (names[j] == "TCPOrigDataSent" || names[j] == "TCPSynRetrans" || + names[j] == "TCPDSACKRecv" || names[j] == "TCPDSACKOfoRecv") { + ++name_count; + int64_t val; + if (absl::SimpleAtoi(values[j], &val)) { + ++value_count; + } + } + } + } + EXPECT_EQ(name_count, 4); + EXPECT_EQ(value_count, 4); +} + +TEST(ProcNetSnmp, Stat) { + struct stat st = {}; + ASSERT_THAT(stat("/proc/net/snmp", &st), SyscallSucceeds()); +} + +TEST(ProcNetSnmp, CheckSnmp) { + // TODO(b/155123175): SNMP and netstat don't work on gVisor. + SKIP_IF(IsRunningOnGvisor()); + + std::string contents = + ASSERT_NO_ERRNO_AND_VALUE(GetContents("/proc/net/snmp")); + + int name_count = 0; + int value_count = 0; + std::vector<absl::string_view> lines = absl::StrSplit(contents, '\n'); + for (int i = 0; i + 1 < lines.size(); i += 2) { + std::vector<absl::string_view> names = + absl::StrSplit(lines[i], absl::ByAnyChar("\t ")); + std::vector<absl::string_view> values = + absl::StrSplit(lines[i + 1], absl::ByAnyChar("\t ")); + EXPECT_EQ(names.size(), values.size()) << " mismatch in lines '" << lines[i] + << "' and '" << lines[i + 1] << "'"; + for (int j = 0; j < names.size() && j < values.size(); ++j) { + if (names[j] == "RetransSegs") { + ++name_count; + int64_t val; + if (absl::SimpleAtoi(values[j], &val)) { + ++value_count; + } + } + } + } + EXPECT_EQ(name_count, 1); + EXPECT_EQ(value_count, 1); +} + } // namespace } // namespace testing } // namespace gvisor diff --git a/test/syscalls/linux/pty.cc b/test/syscalls/linux/pty.cc index b8a0159ba..f9392b9e0 100644 --- a/test/syscalls/linux/pty.cc +++ b/test/syscalls/linux/pty.cc @@ -364,6 +364,12 @@ PosixErrorOr<size_t> PollAndReadFd(int fd, void* buf, size_t count, ssize_t n = ReadFd(fd, static_cast<char*>(buf) + completed, count - completed); if (n < 0) { + if (errno == EAGAIN) { + // Linux sometimes returns EAGAIN from this read, despite the fact that + // poll returned success. Let's just do what do as we are told and try + // again. + continue; + } return PosixError(errno, "read failed"); } completed += n; @@ -628,6 +634,11 @@ TEST_F(PtyTest, TermiosAffectsSlave) { // Verify this by setting ICRNL (which rewrites input \r to \n) and verify that // it has no effect on the master. TEST_F(PtyTest, MasterTermiosUnchangable) { + struct kernel_termios master_termios = {}; + EXPECT_THAT(ioctl(master_.get(), TCGETS, &master_termios), SyscallSucceeds()); + master_termios.c_lflag |= ICRNL; + EXPECT_THAT(ioctl(master_.get(), TCSETS, &master_termios), SyscallSucceeds()); + char c = '\r'; ASSERT_THAT(WriteFd(slave_.get(), &c, 1), SyscallSucceedsWithValue(1)); diff --git a/test/syscalls/linux/pty_root.cc b/test/syscalls/linux/pty_root.cc index 14a4af980..1d7dbefdb 100644 --- a/test/syscalls/linux/pty_root.cc +++ b/test/syscalls/linux/pty_root.cc @@ -25,16 +25,26 @@ namespace gvisor { namespace testing { -// These tests should be run as root. namespace { +// StealTTY tests whether privileged processes can steal controlling terminals. +// If the stealing process has CAP_SYS_ADMIN in the root user namespace, the +// test ensures that stealing works. If it has non-root CAP_SYS_ADMIN, it +// ensures stealing fails. TEST(JobControlRootTest, StealTTY) { SKIP_IF(!ASSERT_NO_ERRNO_AND_VALUE(HaveCapability(CAP_SYS_ADMIN))); - // Make this a session leader, which also drops the controlling terminal. - // In the gVisor test environment, this test will be run as the session - // leader already (as the sentry init process). + bool true_root = true; if (!IsRunningOnGvisor()) { + // If running in Linux, we may only have CAP_SYS_ADMIN in a non-root user + // namespace (i.e. we are not truly root). We use init_module as a proxy for + // whether we are true root, as it returns EPERM immediately. + ASSERT_THAT(syscall(SYS_init_module, nullptr, 0, nullptr), SyscallFails()); + true_root = errno != EPERM; + + // Make this a session leader, which also drops the controlling terminal. + // In the gVisor test environment, this test will be run as the session + // leader already (as the sentry init process). ASSERT_THAT(setsid(), SyscallSucceeds()); } @@ -53,8 +63,8 @@ TEST(JobControlRootTest, StealTTY) { ASSERT_THAT(setsid(), SyscallSucceeds()); // We shouldn't be able to steal the terminal with the wrong arg value. TEST_PCHECK(ioctl(slave.get(), TIOCSCTTY, 0)); - // We should be able to steal it here. - TEST_PCHECK(!ioctl(slave.get(), TIOCSCTTY, 1)); + // We should be able to steal it if we are true root. + TEST_PCHECK(true_root == !ioctl(slave.get(), TIOCSCTTY, 1)); _exit(0); } diff --git a/test/syscalls/linux/pwritev2.cc b/test/syscalls/linux/pwritev2.cc index 3fe5a600f..63b686c62 100644 --- a/test/syscalls/linux/pwritev2.cc +++ b/test/syscalls/linux/pwritev2.cc @@ -69,7 +69,7 @@ ssize_t pwritev2(unsigned long fd, const struct iovec* iov, } // This test is the base case where we call pwritev (no offset, no flags). -TEST(Writev2Test, TestBaseCall) { +TEST(Writev2Test, BaseCall) { SKIP_IF(pwritev2(-1, nullptr, 0, 0, 0) < 0 && errno == ENOSYS); const TempPath file = ASSERT_NO_ERRNO_AND_VALUE(TempPath::CreateFileWith( @@ -97,7 +97,7 @@ TEST(Writev2Test, TestBaseCall) { } // This test is where we call pwritev2 with a positive offset and no flags. -TEST(Pwritev2Test, TestValidPositiveOffset) { +TEST(Pwritev2Test, ValidPositiveOffset) { SKIP_IF(pwritev2(-1, nullptr, 0, 0, 0) < 0 && errno == ENOSYS); std::string prefix(kBufSize, '0'); @@ -129,7 +129,7 @@ TEST(Pwritev2Test, TestValidPositiveOffset) { // This test is the base case where we call writev by using -1 as the offset. // The write should use the file offset, so the test increments the file offset // prior to call pwritev2. -TEST(Pwritev2Test, TestNegativeOneOffset) { +TEST(Pwritev2Test, NegativeOneOffset) { SKIP_IF(pwritev2(-1, nullptr, 0, 0, 0) < 0 && errno == ENOSYS); const std::string prefix = "00"; @@ -164,7 +164,7 @@ TEST(Pwritev2Test, TestNegativeOneOffset) { // pwritev2 requires if the RWF_HIPRI flag is passed, the fd must be opened with // O_DIRECT. This test implements a correct call with the RWF_HIPRI flag. -TEST(Pwritev2Test, TestCallWithRWF_HIPRI) { +TEST(Pwritev2Test, CallWithRWF_HIPRI) { SKIP_IF(pwritev2(-1, nullptr, 0, 0, 0) < 0 && errno == ENOSYS); const TempPath file = ASSERT_NO_ERRNO_AND_VALUE(TempPath::CreateFileWith( @@ -189,47 +189,8 @@ TEST(Pwritev2Test, TestCallWithRWF_HIPRI) { EXPECT_EQ(buf, content); } -// This test checks that pwritev2 can be called with valid flags -TEST(Pwritev2Test, TestCallWithValidFlags) { - SKIP_IF(pwritev2(-1, nullptr, 0, 0, 0) < 0 && errno == ENOSYS); - - const TempPath file = ASSERT_NO_ERRNO_AND_VALUE(TempPath::CreateFileWith( - GetAbsoluteTestTmpdir(), "", TempPath::kDefaultFileMode)); - const FileDescriptor fd = - ASSERT_NO_ERRNO_AND_VALUE(Open(file.path(), O_RDWR)); - - std::vector<char> content(kBufSize, '0'); - struct iovec iov; - iov.iov_base = content.data(); - iov.iov_len = content.size(); - - EXPECT_THAT(pwritev2(fd.get(), &iov, /*iovcnt=*/1, - /*offset=*/0, /*flags=*/RWF_DSYNC), - SyscallSucceedsWithValue(kBufSize)); - - std::vector<char> buf(content.size()); - EXPECT_THAT(read(fd.get(), buf.data(), buf.size()), - SyscallSucceedsWithValue(buf.size())); - - EXPECT_EQ(buf, content); - - SetContent(content); - - EXPECT_THAT(pwritev2(fd.get(), &iov, /*iovcnt=*/1, - /*offset=*/0, /*flags=*/0x4), - SyscallSucceedsWithValue(kBufSize)); - - ASSERT_THAT(lseek(fd.get(), 0, SEEK_CUR), - SyscallSucceedsWithValue(content.size())); - - EXPECT_THAT(pread(fd.get(), buf.data(), buf.size(), /*offset=*/0), - SyscallSucceedsWithValue(buf.size())); - - EXPECT_EQ(buf, content); -} - // This test calls pwritev2 with a bad file descriptor. -TEST(Writev2Test, TestBadFile) { +TEST(Writev2Test, BadFile) { SKIP_IF(pwritev2(-1, nullptr, 0, 0, 0) < 0 && errno == ENOSYS); ASSERT_THAT(pwritev2(/*fd=*/-1, /*iov=*/nullptr, /*iovcnt=*/0, /*offset=*/0, /*flags=*/0), @@ -237,7 +198,7 @@ TEST(Writev2Test, TestBadFile) { } // This test calls pwrite2 with an invalid offset. -TEST(Pwritev2Test, TestInvalidOffset) { +TEST(Pwritev2Test, InvalidOffset) { SKIP_IF(pwritev2(-1, nullptr, 0, 0, 0) < 0 && errno == ENOSYS); const TempPath file = ASSERT_NO_ERRNO_AND_VALUE(TempPath::CreateFileWith( @@ -255,7 +216,7 @@ TEST(Pwritev2Test, TestInvalidOffset) { SyscallFailsWithErrno(EINVAL)); } -TEST(Pwritev2Test, TestUnseekableFileValid) { +TEST(Pwritev2Test, UnseekableFileValid) { SKIP_IF(pwritev2(-1, nullptr, 0, 0, 0) < 0 && errno == ENOSYS); int pipe_fds[2]; @@ -285,7 +246,7 @@ TEST(Pwritev2Test, TestUnseekableFileValid) { // Calling pwritev2 with a non-negative offset calls pwritev. Calling pwritev // with an unseekable file is not allowed. A pipe is used for an unseekable // file. -TEST(Pwritev2Test, TestUnseekableFileInValid) { +TEST(Pwritev2Test, UnseekableFileInvalid) { SKIP_IF(pwritev2(-1, nullptr, 0, 0, 0) < 0 && errno == ENOSYS); int pipe_fds[2]; @@ -304,7 +265,7 @@ TEST(Pwritev2Test, TestUnseekableFileInValid) { EXPECT_THAT(close(pipe_fds[1]), SyscallSucceeds()); } -TEST(Pwritev2Test, TestReadOnlyFile) { +TEST(Pwritev2Test, ReadOnlyFile) { SKIP_IF(pwritev2(-1, nullptr, 0, 0, 0) < 0 && errno == ENOSYS); const TempPath file = ASSERT_NO_ERRNO_AND_VALUE(TempPath::CreateFileWith( @@ -323,7 +284,7 @@ TEST(Pwritev2Test, TestReadOnlyFile) { } // This test calls pwritev2 with an invalid flag. -TEST(Pwritev2Test, TestInvalidFlag) { +TEST(Pwritev2Test, InvalidFlag) { SKIP_IF(pwritev2(-1, nullptr, 0, 0, 0) < 0 && errno == ENOSYS); const TempPath file = ASSERT_NO_ERRNO_AND_VALUE(TempPath::CreateFileWith( diff --git a/test/syscalls/linux/raw_socket.cc b/test/syscalls/linux/raw_socket.cc new file mode 100644 index 000000000..8d6e5c913 --- /dev/null +++ b/test/syscalls/linux/raw_socket.cc @@ -0,0 +1,869 @@ +// Copyright 2019 The gVisor Authors. +// +// Licensed under the Apache License, Version 2.0 (the "License"); +// you may not use this file except in compliance with the License. +// You may obtain a copy of the License at +// +// http://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, software +// distributed under the License is distributed on an "AS IS" BASIS, +// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +// See the License for the specific language governing permissions and +// limitations under the License. + +#include <linux/capability.h> +#ifndef __fuchsia__ +#include <linux/filter.h> +#endif // __fuchsia__ +#include <netinet/in.h> +#include <netinet/ip.h> +#include <netinet/ip6.h> +#include <netinet/ip_icmp.h> +#include <poll.h> +#include <sys/socket.h> +#include <sys/types.h> +#include <unistd.h> + +#include <algorithm> + +#include "gtest/gtest.h" +#include "test/syscalls/linux/socket_test_util.h" +#include "test/syscalls/linux/unix_domain_socket_test_util.h" +#include "test/util/capability_util.h" +#include "test/util/file_descriptor.h" +#include "test/util/test_util.h" + +// Note: in order to run these tests, /proc/sys/net/ipv4/ping_group_range will +// need to be configured to let the superuser create ping sockets (see icmp(7)). + +namespace gvisor { +namespace testing { + +namespace { + +// Fixture for tests parameterized by protocol. +class RawSocketTest : public ::testing::TestWithParam<std::tuple<int, int>> { + protected: + // Creates a socket to be used in tests. + void SetUp() override; + + // Closes the socket created by SetUp(). + void TearDown() override; + + // Sends buf via s_. + void SendBuf(const char* buf, int buf_len); + + // Reads from s_ into recv_buf. + void ReceiveBuf(char* recv_buf, size_t recv_buf_len); + + void ReceiveBufFrom(int sock, char* recv_buf, size_t recv_buf_len); + + int Protocol() { return std::get<0>(GetParam()); } + + int Family() { return std::get<1>(GetParam()); } + + socklen_t AddrLen() { + if (Family() == AF_INET) { + return sizeof(sockaddr_in); + } + return sizeof(sockaddr_in6); + } + + int HdrLen() { + if (Family() == AF_INET) { + return sizeof(struct iphdr); + } + // IPv6 raw sockets don't include the header. + return 0; + } + + // The socket used for both reading and writing. + int s_; + + // The loopback address. + struct sockaddr_storage addr_; +}; + +void RawSocketTest::SetUp() { + if (!ASSERT_NO_ERRNO_AND_VALUE(HaveCapability(CAP_NET_RAW))) { + ASSERT_THAT(socket(Family(), SOCK_RAW, Protocol()), + SyscallFailsWithErrno(EPERM)); + GTEST_SKIP(); + } + + ASSERT_THAT(s_ = socket(Family(), SOCK_RAW, Protocol()), SyscallSucceeds()); + + addr_ = {}; + + // We don't set ports because raw sockets don't have a notion of ports. + if (Family() == AF_INET) { + struct sockaddr_in* sin = reinterpret_cast<struct sockaddr_in*>(&addr_); + sin->sin_family = AF_INET; + sin->sin_addr.s_addr = htonl(INADDR_LOOPBACK); + } else { + struct sockaddr_in6* sin6 = reinterpret_cast<struct sockaddr_in6*>(&addr_); + sin6->sin6_family = AF_INET6; + sin6->sin6_addr = in6addr_loopback; + } +} + +void RawSocketTest::TearDown() { + // TearDown will be run even if we skip the test. + if (ASSERT_NO_ERRNO_AND_VALUE(HaveCapability(CAP_NET_RAW))) { + EXPECT_THAT(close(s_), SyscallSucceeds()); + } +} + +// We should be able to create multiple raw sockets for the same protocol. +// BasicRawSocket::Setup creates the first one, so we only have to create one +// more here. +TEST_P(RawSocketTest, MultipleCreation) { + SKIP_IF(!ASSERT_NO_ERRNO_AND_VALUE(HaveCapability(CAP_NET_RAW))); + + int s2; + ASSERT_THAT(s2 = socket(Family(), SOCK_RAW, Protocol()), SyscallSucceeds()); + + ASSERT_THAT(close(s2), SyscallSucceeds()); +} + +// Test that shutting down an unconnected socket fails. +TEST_P(RawSocketTest, FailShutdownWithoutConnect) { + SKIP_IF(!ASSERT_NO_ERRNO_AND_VALUE(HaveCapability(CAP_NET_RAW))); + + ASSERT_THAT(shutdown(s_, SHUT_WR), SyscallFailsWithErrno(ENOTCONN)); + ASSERT_THAT(shutdown(s_, SHUT_RD), SyscallFailsWithErrno(ENOTCONN)); +} + +// Shutdown is a no-op for raw sockets (and datagram sockets in general). +TEST_P(RawSocketTest, ShutdownWriteNoop) { + SKIP_IF(!ASSERT_NO_ERRNO_AND_VALUE(HaveCapability(CAP_NET_RAW))); + + ASSERT_THAT( + connect(s_, reinterpret_cast<struct sockaddr*>(&addr_), AddrLen()), + SyscallSucceeds()); + ASSERT_THAT(shutdown(s_, SHUT_WR), SyscallSucceeds()); + + // Arbitrary. + constexpr char kBuf[] = "noop"; + ASSERT_THAT(RetryEINTR(write)(s_, kBuf, sizeof(kBuf)), + SyscallSucceedsWithValue(sizeof(kBuf))); +} + +// Shutdown is a no-op for raw sockets (and datagram sockets in general). +TEST_P(RawSocketTest, ShutdownReadNoop) { + SKIP_IF(!ASSERT_NO_ERRNO_AND_VALUE(HaveCapability(CAP_NET_RAW))); + + ASSERT_THAT( + connect(s_, reinterpret_cast<struct sockaddr*>(&addr_), AddrLen()), + SyscallSucceeds()); + ASSERT_THAT(shutdown(s_, SHUT_RD), SyscallSucceeds()); + + // Arbitrary. + constexpr char kBuf[] = "gdg"; + ASSERT_NO_FATAL_FAILURE(SendBuf(kBuf, sizeof(kBuf))); + + std::vector<char> c(sizeof(kBuf) + HdrLen()); + ASSERT_THAT(read(s_, c.data(), c.size()), SyscallSucceedsWithValue(c.size())); +} + +// Test that listen() fails. +TEST_P(RawSocketTest, FailListen) { + SKIP_IF(!ASSERT_NO_ERRNO_AND_VALUE(HaveCapability(CAP_NET_RAW))); + + ASSERT_THAT(listen(s_, 1), SyscallFailsWithErrno(ENOTSUP)); +} + +// Test that accept() fails. +TEST_P(RawSocketTest, FailAccept) { + SKIP_IF(!ASSERT_NO_ERRNO_AND_VALUE(HaveCapability(CAP_NET_RAW))); + + struct sockaddr saddr; + socklen_t addrlen; + ASSERT_THAT(accept(s_, &saddr, &addrlen), SyscallFailsWithErrno(ENOTSUP)); +} + +// Test that getpeername() returns nothing before connect(). +TEST_P(RawSocketTest, FailGetPeerNameBeforeConnect) { + SKIP_IF(!ASSERT_NO_ERRNO_AND_VALUE(HaveCapability(CAP_NET_RAW))); + + struct sockaddr saddr; + socklen_t addrlen = sizeof(saddr); + ASSERT_THAT(getpeername(s_, &saddr, &addrlen), + SyscallFailsWithErrno(ENOTCONN)); +} + +// Test that getpeername() returns something after connect(). +TEST_P(RawSocketTest, GetPeerName) { + SKIP_IF(!ASSERT_NO_ERRNO_AND_VALUE(HaveCapability(CAP_NET_RAW))); + + ASSERT_THAT( + connect(s_, reinterpret_cast<struct sockaddr*>(&addr_), AddrLen()), + SyscallSucceeds()); + struct sockaddr saddr; + socklen_t addrlen = sizeof(saddr); + ASSERT_THAT(getpeername(s_, &saddr, &addrlen), + SyscallFailsWithErrno(ENOTCONN)); + ASSERT_GT(addrlen, 0); +} + +// Test that the socket is writable immediately. +TEST_P(RawSocketTest, PollWritableImmediately) { + SKIP_IF(!ASSERT_NO_ERRNO_AND_VALUE(HaveCapability(CAP_NET_RAW))); + + struct pollfd pfd = {}; + pfd.fd = s_; + pfd.events = POLLOUT; + ASSERT_THAT(RetryEINTR(poll)(&pfd, 1, 10000), SyscallSucceedsWithValue(1)); +} + +// Test that the socket isn't readable before receiving anything. +TEST_P(RawSocketTest, PollNotReadableInitially) { + SKIP_IF(!ASSERT_NO_ERRNO_AND_VALUE(HaveCapability(CAP_NET_RAW))); + + // Try to receive data with MSG_DONTWAIT, which returns immediately if there's + // nothing to be read. + char buf[117]; + ASSERT_THAT(RetryEINTR(recv)(s_, buf, sizeof(buf), MSG_DONTWAIT), + SyscallFailsWithErrno(EAGAIN)); +} + +// Test that the socket becomes readable once something is written to it. +TEST_P(RawSocketTest, PollTriggeredOnWrite) { + SKIP_IF(!ASSERT_NO_ERRNO_AND_VALUE(HaveCapability(CAP_NET_RAW))); + + // Write something so that there's data to be read. + // Arbitrary. + constexpr char kBuf[] = "JP5"; + ASSERT_NO_FATAL_FAILURE(SendBuf(kBuf, sizeof(kBuf))); + + struct pollfd pfd = {}; + pfd.fd = s_; + pfd.events = POLLIN; + ASSERT_THAT(RetryEINTR(poll)(&pfd, 1, 10000), SyscallSucceedsWithValue(1)); +} + +// Test that we can connect() to a valid IP (loopback). +TEST_P(RawSocketTest, ConnectToLoopback) { + SKIP_IF(!ASSERT_NO_ERRNO_AND_VALUE(HaveCapability(CAP_NET_RAW))); + + ASSERT_THAT( + connect(s_, reinterpret_cast<struct sockaddr*>(&addr_), AddrLen()), + SyscallSucceeds()); +} + +// Test that calling send() without connect() fails. +TEST_P(RawSocketTest, SendWithoutConnectFails) { + SKIP_IF(!ASSERT_NO_ERRNO_AND_VALUE(HaveCapability(CAP_NET_RAW))); + + // Arbitrary. + constexpr char kBuf[] = "Endgame was good"; + ASSERT_THAT(send(s_, kBuf, sizeof(kBuf), 0), + SyscallFailsWithErrno(EDESTADDRREQ)); +} + +// Wildcard Bind. +TEST_P(RawSocketTest, BindToWildcard) { + SKIP_IF(!ASSERT_NO_ERRNO_AND_VALUE(HaveCapability(CAP_NET_RAW))); + struct sockaddr_storage addr; + addr = {}; + + // We don't set ports because raw sockets don't have a notion of ports. + if (Family() == AF_INET) { + struct sockaddr_in* sin = reinterpret_cast<struct sockaddr_in*>(&addr); + sin->sin_family = AF_INET; + sin->sin_addr.s_addr = htonl(INADDR_ANY); + } else { + struct sockaddr_in6* sin6 = reinterpret_cast<struct sockaddr_in6*>(&addr); + sin6->sin6_family = AF_INET6; + sin6->sin6_addr = in6addr_any; + } + + ASSERT_THAT(bind(s_, reinterpret_cast<struct sockaddr*>(&addr_), AddrLen()), + SyscallSucceeds()); +} + +// Bind to localhost. +TEST_P(RawSocketTest, BindToLocalhost) { + SKIP_IF(!ASSERT_NO_ERRNO_AND_VALUE(HaveCapability(CAP_NET_RAW))); + + ASSERT_THAT( + bind(s_, reinterpret_cast<struct sockaddr*>(&addr_), AddrLen()), + SyscallSucceeds()); +} + +// Bind to a different address. +TEST_P(RawSocketTest, BindToInvalid) { + SKIP_IF(!ASSERT_NO_ERRNO_AND_VALUE(HaveCapability(CAP_NET_RAW))); + + struct sockaddr_storage bind_addr = addr_; + if (Family() == AF_INET) { + struct sockaddr_in* sin = reinterpret_cast<struct sockaddr_in*>(&bind_addr); + sin->sin_addr = {1}; // 1.0.0.0 - An address that we can't bind to. + } else { + struct sockaddr_in6* sin6 = + reinterpret_cast<struct sockaddr_in6*>(&bind_addr); + memset(&sin6->sin6_addr.s6_addr, 0, sizeof(sin6->sin6_addr.s6_addr)); + sin6->sin6_addr.s6_addr[0] = 1; // 1: - An address that we can't bind to. + } + ASSERT_THAT(bind(s_, reinterpret_cast<struct sockaddr*>(&bind_addr), + AddrLen()), SyscallFailsWithErrno(EADDRNOTAVAIL)); +} + +// Send and receive an packet. +TEST_P(RawSocketTest, SendAndReceive) { + SKIP_IF(!ASSERT_NO_ERRNO_AND_VALUE(HaveCapability(CAP_NET_RAW))); + + // Arbitrary. + constexpr char kBuf[] = "TB12"; + ASSERT_NO_FATAL_FAILURE(SendBuf(kBuf, sizeof(kBuf))); + + // Receive the packet and make sure it's identical. + std::vector<char> recv_buf(sizeof(kBuf) + HdrLen()); + ASSERT_NO_FATAL_FAILURE(ReceiveBuf(recv_buf.data(), recv_buf.size())); + EXPECT_EQ(memcmp(recv_buf.data() + HdrLen(), kBuf, sizeof(kBuf)), 0); +} + +// We should be able to create multiple raw sockets for the same protocol and +// receive the same packet on both. +TEST_P(RawSocketTest, MultipleSocketReceive) { + SKIP_IF(!ASSERT_NO_ERRNO_AND_VALUE(HaveCapability(CAP_NET_RAW))); + + int s2; + ASSERT_THAT(s2 = socket(Family(), SOCK_RAW, Protocol()), SyscallSucceeds()); + + // Arbitrary. + constexpr char kBuf[] = "TB10"; + ASSERT_NO_FATAL_FAILURE(SendBuf(kBuf, sizeof(kBuf))); + + // Receive it on socket 1. + std::vector<char> recv_buf1(sizeof(kBuf) + HdrLen()); + ASSERT_NO_FATAL_FAILURE(ReceiveBuf(recv_buf1.data(), recv_buf1.size())); + + // Receive it on socket 2. + std::vector<char> recv_buf2(sizeof(kBuf) + HdrLen()); + ASSERT_NO_FATAL_FAILURE(ReceiveBufFrom(s2, recv_buf2.data(), + recv_buf2.size())); + + EXPECT_EQ(memcmp(recv_buf1.data() + HdrLen(), + recv_buf2.data() + HdrLen(), sizeof(kBuf)), + 0); + + ASSERT_THAT(close(s2), SyscallSucceeds()); +} + +// Test that connect sends packets to the right place. +TEST_P(RawSocketTest, SendAndReceiveViaConnect) { + SKIP_IF(!ASSERT_NO_ERRNO_AND_VALUE(HaveCapability(CAP_NET_RAW))); + + ASSERT_THAT( + connect(s_, reinterpret_cast<struct sockaddr*>(&addr_), AddrLen()), + SyscallSucceeds()); + + // Arbitrary. + constexpr char kBuf[] = "JH4"; + ASSERT_THAT(send(s_, kBuf, sizeof(kBuf), 0), + SyscallSucceedsWithValue(sizeof(kBuf))); + + // Receive the packet and make sure it's identical. + std::vector<char> recv_buf(sizeof(kBuf) + HdrLen()); + ASSERT_NO_FATAL_FAILURE(ReceiveBuf(recv_buf.data(), recv_buf.size())); + EXPECT_EQ(memcmp(recv_buf.data() + HdrLen(), kBuf, sizeof(kBuf)), 0); +} + +// Bind to localhost, then send and receive packets. +TEST_P(RawSocketTest, BindSendAndReceive) { + SKIP_IF(!ASSERT_NO_ERRNO_AND_VALUE(HaveCapability(CAP_NET_RAW))); + + ASSERT_THAT( + bind(s_, reinterpret_cast<struct sockaddr*>(&addr_), AddrLen()), + SyscallSucceeds()); + + // Arbitrary. + constexpr char kBuf[] = "DR16"; + ASSERT_NO_FATAL_FAILURE(SendBuf(kBuf, sizeof(kBuf))); + + // Receive the packet and make sure it's identical. + std::vector<char> recv_buf(sizeof(kBuf) + HdrLen()); + ASSERT_NO_FATAL_FAILURE(ReceiveBuf(recv_buf.data(), recv_buf.size())); + EXPECT_EQ(memcmp(recv_buf.data() + HdrLen(), kBuf, sizeof(kBuf)), 0); +} + +// Bind and connect to localhost and send/receive packets. +TEST_P(RawSocketTest, BindConnectSendAndReceive) { + SKIP_IF(!ASSERT_NO_ERRNO_AND_VALUE(HaveCapability(CAP_NET_RAW))); + + ASSERT_THAT( + bind(s_, reinterpret_cast<struct sockaddr*>(&addr_), AddrLen()), + SyscallSucceeds()); + ASSERT_THAT( + connect(s_, reinterpret_cast<struct sockaddr*>(&addr_), AddrLen()), + SyscallSucceeds()); + + // Arbitrary. + constexpr char kBuf[] = "DG88"; + ASSERT_NO_FATAL_FAILURE(SendBuf(kBuf, sizeof(kBuf))); + + // Receive the packet and make sure it's identical. + std::vector<char> recv_buf(sizeof(kBuf) + HdrLen()); + ASSERT_NO_FATAL_FAILURE(ReceiveBuf(recv_buf.data(), recv_buf.size())); + EXPECT_EQ(memcmp(recv_buf.data() + HdrLen(), kBuf, sizeof(kBuf)), 0); +} + +// Check that setting SO_RCVBUF below min is clamped to the minimum +// receive buffer size. +TEST_P(RawSocketTest, SetSocketRecvBufBelowMin) { + SKIP_IF(!ASSERT_NO_ERRNO_AND_VALUE(HaveCapability(CAP_NET_RAW))); + + // Discover minimum receive buf size by trying to set it to zero. + // See: + // https://github.com/torvalds/linux/blob/a5dc8300df75e8b8384b4c82225f1e4a0b4d9b55/net/core/sock.c#L820 + constexpr int kRcvBufSz = 0; + ASSERT_THAT( + setsockopt(s_, SOL_SOCKET, SO_RCVBUF, &kRcvBufSz, sizeof(kRcvBufSz)), + SyscallSucceeds()); + + int min = 0; + socklen_t min_len = sizeof(min); + ASSERT_THAT(getsockopt(s_, SOL_SOCKET, SO_RCVBUF, &min, &min_len), + SyscallSucceeds()); + + // Linux doubles the value so let's use a value that when doubled will still + // be smaller than min. + int below_min = min / 2 - 1; + ASSERT_THAT( + setsockopt(s_, SOL_SOCKET, SO_RCVBUF, &below_min, sizeof(below_min)), + SyscallSucceeds()); + + int val = 0; + socklen_t val_len = sizeof(val); + ASSERT_THAT(getsockopt(s_, SOL_SOCKET, SO_RCVBUF, &val, &val_len), + SyscallSucceeds()); + + ASSERT_EQ(min, val); +} + +// Check that setting SO_RCVBUF above max is clamped to the maximum +// receive buffer size. +TEST_P(RawSocketTest, SetSocketRecvBufAboveMax) { + SKIP_IF(!ASSERT_NO_ERRNO_AND_VALUE(HaveCapability(CAP_NET_RAW))); + + // Discover max buf size by trying to set the largest possible buffer size. + constexpr int kRcvBufSz = 0xffffffff; + ASSERT_THAT( + setsockopt(s_, SOL_SOCKET, SO_RCVBUF, &kRcvBufSz, sizeof(kRcvBufSz)), + SyscallSucceeds()); + + int max = 0; + socklen_t max_len = sizeof(max); + ASSERT_THAT(getsockopt(s_, SOL_SOCKET, SO_RCVBUF, &max, &max_len), + SyscallSucceeds()); + + int above_max = max + 1; + ASSERT_THAT( + setsockopt(s_, SOL_SOCKET, SO_RCVBUF, &above_max, sizeof(above_max)), + SyscallSucceeds()); + + int val = 0; + socklen_t val_len = sizeof(val); + ASSERT_THAT(getsockopt(s_, SOL_SOCKET, SO_RCVBUF, &val, &val_len), + SyscallSucceeds()); + ASSERT_EQ(max, val); +} + +// Check that setting SO_RCVBUF min <= kRcvBufSz <= max is honored. +TEST_P(RawSocketTest, SetSocketRecvBuf) { + SKIP_IF(!ASSERT_NO_ERRNO_AND_VALUE(HaveCapability(CAP_NET_RAW))); + + int max = 0; + int min = 0; + { + // Discover max buf size by trying to set a really large buffer size. + constexpr int kRcvBufSz = 0xffffffff; + ASSERT_THAT( + setsockopt(s_, SOL_SOCKET, SO_RCVBUF, &kRcvBufSz, sizeof(kRcvBufSz)), + SyscallSucceeds()); + + max = 0; + socklen_t max_len = sizeof(max); + ASSERT_THAT(getsockopt(s_, SOL_SOCKET, SO_RCVBUF, &max, &max_len), + SyscallSucceeds()); + } + + { + // Discover minimum buffer size by trying to set a zero size receive buffer + // size. + // See: + // https://github.com/torvalds/linux/blob/a5dc8300df75e8b8384b4c82225f1e4a0b4d9b55/net/core/sock.c#L820 + constexpr int kRcvBufSz = 0; + ASSERT_THAT( + setsockopt(s_, SOL_SOCKET, SO_RCVBUF, &kRcvBufSz, sizeof(kRcvBufSz)), + SyscallSucceeds()); + + socklen_t min_len = sizeof(min); + ASSERT_THAT(getsockopt(s_, SOL_SOCKET, SO_RCVBUF, &min, &min_len), + SyscallSucceeds()); + } + + int quarter_sz = min + (max - min) / 4; + ASSERT_THAT( + setsockopt(s_, SOL_SOCKET, SO_RCVBUF, &quarter_sz, sizeof(quarter_sz)), + SyscallSucceeds()); + + int val = 0; + socklen_t val_len = sizeof(val); + ASSERT_THAT(getsockopt(s_, SOL_SOCKET, SO_RCVBUF, &val, &val_len), + SyscallSucceeds()); + + // Linux doubles the value set by SO_SNDBUF/SO_RCVBUF. + // TODO(gvisor.dev/issue/2926): Remove when Netstack matches linux behavior. + if (!IsRunningOnGvisor()) { + quarter_sz *= 2; + } + ASSERT_EQ(quarter_sz, val); +} + +// Check that setting SO_SNDBUF below min is clamped to the minimum +// receive buffer size. +TEST_P(RawSocketTest, SetSocketSendBufBelowMin) { + SKIP_IF(!ASSERT_NO_ERRNO_AND_VALUE(HaveCapability(CAP_NET_RAW))); + + // Discover minimum buffer size by trying to set it to zero. + constexpr int kSndBufSz = 0; + ASSERT_THAT( + setsockopt(s_, SOL_SOCKET, SO_SNDBUF, &kSndBufSz, sizeof(kSndBufSz)), + SyscallSucceeds()); + + int min = 0; + socklen_t min_len = sizeof(min); + ASSERT_THAT(getsockopt(s_, SOL_SOCKET, SO_SNDBUF, &min, &min_len), + SyscallSucceeds()); + + // Linux doubles the value so let's use a value that when doubled will still + // be smaller than min. + int below_min = min / 2 - 1; + ASSERT_THAT( + setsockopt(s_, SOL_SOCKET, SO_SNDBUF, &below_min, sizeof(below_min)), + SyscallSucceeds()); + + int val = 0; + socklen_t val_len = sizeof(val); + ASSERT_THAT(getsockopt(s_, SOL_SOCKET, SO_SNDBUF, &val, &val_len), + SyscallSucceeds()); + + ASSERT_EQ(min, val); +} + +// Check that setting SO_SNDBUF above max is clamped to the maximum +// send buffer size. +TEST_P(RawSocketTest, SetSocketSendBufAboveMax) { + SKIP_IF(!ASSERT_NO_ERRNO_AND_VALUE(HaveCapability(CAP_NET_RAW))); + + // Discover maximum buffer size by trying to set it to a large value. + constexpr int kSndBufSz = 0xffffffff; + ASSERT_THAT( + setsockopt(s_, SOL_SOCKET, SO_SNDBUF, &kSndBufSz, sizeof(kSndBufSz)), + SyscallSucceeds()); + + int max = 0; + socklen_t max_len = sizeof(max); + ASSERT_THAT(getsockopt(s_, SOL_SOCKET, SO_SNDBUF, &max, &max_len), + SyscallSucceeds()); + + int above_max = max + 1; + ASSERT_THAT( + setsockopt(s_, SOL_SOCKET, SO_SNDBUF, &above_max, sizeof(above_max)), + SyscallSucceeds()); + + int val = 0; + socklen_t val_len = sizeof(val); + ASSERT_THAT(getsockopt(s_, SOL_SOCKET, SO_SNDBUF, &val, &val_len), + SyscallSucceeds()); + ASSERT_EQ(max, val); +} + +// Check that setting SO_SNDBUF min <= kSndBufSz <= max is honored. +TEST_P(RawSocketTest, SetSocketSendBuf) { + SKIP_IF(!ASSERT_NO_ERRNO_AND_VALUE(HaveCapability(CAP_NET_RAW))); + + int max = 0; + int min = 0; + { + // Discover maximum buffer size by trying to set it to a large value. + constexpr int kSndBufSz = 0xffffffff; + ASSERT_THAT( + setsockopt(s_, SOL_SOCKET, SO_SNDBUF, &kSndBufSz, sizeof(kSndBufSz)), + SyscallSucceeds()); + + max = 0; + socklen_t max_len = sizeof(max); + ASSERT_THAT(getsockopt(s_, SOL_SOCKET, SO_SNDBUF, &max, &max_len), + SyscallSucceeds()); + } + + { + // Discover minimum buffer size by trying to set it to zero. + constexpr int kSndBufSz = 0; + ASSERT_THAT( + setsockopt(s_, SOL_SOCKET, SO_SNDBUF, &kSndBufSz, sizeof(kSndBufSz)), + SyscallSucceeds()); + + socklen_t min_len = sizeof(min); + ASSERT_THAT(getsockopt(s_, SOL_SOCKET, SO_SNDBUF, &min, &min_len), + SyscallSucceeds()); + } + + int quarter_sz = min + (max - min) / 4; + ASSERT_THAT( + setsockopt(s_, SOL_SOCKET, SO_SNDBUF, &quarter_sz, sizeof(quarter_sz)), + SyscallSucceeds()); + + int val = 0; + socklen_t val_len = sizeof(val); + ASSERT_THAT(getsockopt(s_, SOL_SOCKET, SO_SNDBUF, &val, &val_len), + SyscallSucceeds()); + + // Linux doubles the value set by SO_SNDBUF/SO_RCVBUF. + // TODO(gvisor.dev/issue/2926): Remove the gvisor special casing when Netstack + // matches linux behavior. + if (!IsRunningOnGvisor()) { + quarter_sz *= 2; + } + + ASSERT_EQ(quarter_sz, val); +} + +// Test that receive buffer limits are not enforced when the recv buffer is +// empty. +TEST_P(RawSocketTest, RecvBufLimitsEmptyRecvBuffer) { + SKIP_IF(!ASSERT_NO_ERRNO_AND_VALUE(HaveCapability(CAP_NET_RAW))); + + ASSERT_THAT( + bind(s_, reinterpret_cast<struct sockaddr*>(&addr_), AddrLen()), + SyscallSucceeds()); + ASSERT_THAT( + connect(s_, reinterpret_cast<struct sockaddr*>(&addr_), AddrLen()), + SyscallSucceeds()); + + int min = 0; + { + // Discover minimum buffer size by trying to set it to zero. + constexpr int kRcvBufSz = 0; + ASSERT_THAT( + setsockopt(s_, SOL_SOCKET, SO_RCVBUF, &kRcvBufSz, sizeof(kRcvBufSz)), + SyscallSucceeds()); + + socklen_t min_len = sizeof(min); + ASSERT_THAT(getsockopt(s_, SOL_SOCKET, SO_RCVBUF, &min, &min_len), + SyscallSucceeds()); + } + + { + // Send data of size min and verify that it's received. + std::vector<char> buf(min); + RandomizeBuffer(buf.data(), buf.size()); + ASSERT_NO_FATAL_FAILURE(SendBuf(buf.data(), buf.size())); + + // Receive the packet and make sure it's identical. + std::vector<char> recv_buf(buf.size() + HdrLen()); + ASSERT_NO_FATAL_FAILURE(ReceiveBuf(recv_buf.data(), recv_buf.size())); + EXPECT_EQ( + memcmp(recv_buf.data() + HdrLen(), buf.data(), buf.size()), + 0); + } + + { + // Send data of size min + 1 and verify that its received. Both linux and + // Netstack accept a dgram that exceeds rcvBuf limits if the receive buffer + // is currently empty. + std::vector<char> buf(min + 1); + RandomizeBuffer(buf.data(), buf.size()); + ASSERT_NO_FATAL_FAILURE(SendBuf(buf.data(), buf.size())); + // Receive the packet and make sure it's identical. + std::vector<char> recv_buf(buf.size() + HdrLen()); + ASSERT_NO_FATAL_FAILURE(ReceiveBuf(recv_buf.data(), recv_buf.size())); + EXPECT_EQ( + memcmp(recv_buf.data() + HdrLen(), buf.data(), buf.size()), + 0); + } +} + +TEST_P(RawSocketTest, RecvBufLimits) { + // TCP stack generates RSTs for unknown endpoints and it complicates the test + // as we have to deal with the RST packets as well. For testing the raw socket + // endpoints buffer limit enforcement we can just test for UDP. + // + // We don't use SKIP_IF here because root_test_runner explicitly fails if a + // test is skipped. + if (Protocol() == IPPROTO_TCP) { + return; + } + SKIP_IF(!ASSERT_NO_ERRNO_AND_VALUE(HaveCapability(CAP_NET_RAW))); + + ASSERT_THAT( + bind(s_, reinterpret_cast<struct sockaddr*>(&addr_), AddrLen()), + SyscallSucceeds()); + ASSERT_THAT( + connect(s_, reinterpret_cast<struct sockaddr*>(&addr_), AddrLen()), + SyscallSucceeds()); + + int min = 0; + { + // Discover minimum buffer size by trying to set it to zero. + constexpr int kRcvBufSz = 0; + ASSERT_THAT( + setsockopt(s_, SOL_SOCKET, SO_RCVBUF, &kRcvBufSz, sizeof(kRcvBufSz)), + SyscallSucceeds()); + + socklen_t min_len = sizeof(min); + ASSERT_THAT(getsockopt(s_, SOL_SOCKET, SO_RCVBUF, &min, &min_len), + SyscallSucceeds()); + } + + // Now set the limit to min * 2. + int new_rcv_buf_sz = min * 4; + if (!IsRunningOnGvisor()) { + // Linux doubles the value specified so just set to min. + new_rcv_buf_sz = min * 2; + } + + ASSERT_THAT(setsockopt(s_, SOL_SOCKET, SO_RCVBUF, &new_rcv_buf_sz, + sizeof(new_rcv_buf_sz)), + SyscallSucceeds()); + int rcv_buf_sz = 0; + { + socklen_t rcv_buf_len = sizeof(rcv_buf_sz); + ASSERT_THAT( + getsockopt(s_, SOL_SOCKET, SO_RCVBUF, &rcv_buf_sz, &rcv_buf_len), + SyscallSucceeds()); + } + + // Set a receive timeout so that we don't block forever on reads if the test + // fails. + struct timeval tv { + .tv_sec = 1, .tv_usec = 0, + }; + ASSERT_THAT(setsockopt(s_, SOL_SOCKET, SO_RCVTIMEO, &tv, sizeof(tv)), + SyscallSucceeds()); + + { + std::vector<char> buf(min); + RandomizeBuffer(buf.data(), buf.size()); + + ASSERT_NO_FATAL_FAILURE(SendBuf(buf.data(), buf.size())); + ASSERT_NO_FATAL_FAILURE(SendBuf(buf.data(), buf.size())); + ASSERT_NO_FATAL_FAILURE(SendBuf(buf.data(), buf.size())); + ASSERT_NO_FATAL_FAILURE(SendBuf(buf.data(), buf.size())); + int sent = 4; + if (IsRunningOnGvisor()) { + // Linux seems to drop the 4th packet even though technically it should + // fit in the receive buffer. + ASSERT_NO_FATAL_FAILURE(SendBuf(buf.data(), buf.size())); + sent++; + } + + // Verify that the expected number of packets are available to be read. + for (int i = 0; i < sent - 1; i++) { + // Receive the packet and make sure it's identical. + std::vector<char> recv_buf(buf.size() + HdrLen()); + ASSERT_NO_FATAL_FAILURE(ReceiveBuf(recv_buf.data(), recv_buf.size())); + EXPECT_EQ(memcmp(recv_buf.data() + HdrLen(), buf.data(), + buf.size()), + 0); + } + + // Assert that the last packet is dropped because the receive buffer should + // be full after the first four packets. + std::vector<char> recv_buf(buf.size() + HdrLen()); + struct iovec iov = {}; + iov.iov_base = static_cast<void*>(const_cast<char*>(recv_buf.data())); + iov.iov_len = buf.size(); + struct msghdr msg = {}; + msg.msg_iov = &iov; + msg.msg_iovlen = 1; + msg.msg_control = NULL; + msg.msg_controllen = 0; + msg.msg_flags = 0; + ASSERT_THAT(RetryEINTR(recvmsg)(s_, &msg, MSG_DONTWAIT), + SyscallFailsWithErrno(EAGAIN)); + } +} + +void RawSocketTest::SendBuf(const char* buf, int buf_len) { + // It's safe to use const_cast here because sendmsg won't modify the iovec or + // address. + struct iovec iov = {}; + iov.iov_base = static_cast<void*>(const_cast<char*>(buf)); + iov.iov_len = static_cast<size_t>(buf_len); + struct msghdr msg = {}; + msg.msg_name = static_cast<void*>(&addr_); + msg.msg_namelen = AddrLen(); + msg.msg_iov = &iov; + msg.msg_iovlen = 1; + msg.msg_control = NULL; + msg.msg_controllen = 0; + msg.msg_flags = 0; + ASSERT_THAT(sendmsg(s_, &msg, 0), SyscallSucceedsWithValue(buf_len)); +} + +void RawSocketTest::ReceiveBuf(char* recv_buf, size_t recv_buf_len) { + ASSERT_NO_FATAL_FAILURE(ReceiveBufFrom(s_, recv_buf, recv_buf_len)); +} + +void RawSocketTest::ReceiveBufFrom(int sock, char* recv_buf, + size_t recv_buf_len) { + ASSERT_NO_FATAL_FAILURE(RecvNoCmsg(sock, recv_buf, recv_buf_len)); +} + +#ifndef __fuchsia__ + +TEST_P(RawSocketTest, SetSocketDetachFilterNoInstalledFilter) { + // TODO(gvisor.dev/2746): Support SO_ATTACH_FILTER/SO_DETACH_FILTER. + if (IsRunningOnGvisor()) { + constexpr int val = 0; + ASSERT_THAT(setsockopt(s_, SOL_SOCKET, SO_DETACH_FILTER, &val, sizeof(val)), + SyscallSucceeds()); + return; + } + + constexpr int val = 0; + ASSERT_THAT(setsockopt(s_, SOL_SOCKET, SO_DETACH_FILTER, &val, sizeof(val)), + SyscallFailsWithErrno(ENOENT)); +} + +TEST_P(RawSocketTest, GetSocketDetachFilter) { + int val = 0; + socklen_t val_len = sizeof(val); + ASSERT_THAT(getsockopt(s_, SOL_SOCKET, SO_DETACH_FILTER, &val, &val_len), + SyscallFailsWithErrno(ENOPROTOOPT)); +} + +#endif // __fuchsia__ + +// AF_INET6+SOCK_RAW+IPPROTO_RAW sockets can be created, but not written to. +TEST(RawSocketTest, IPv6ProtoRaw) { + SKIP_IF(!ASSERT_NO_ERRNO_AND_VALUE(HaveCapability(CAP_NET_RAW))); + + int sock; + ASSERT_THAT(sock = socket(AF_INET6, SOCK_RAW, IPPROTO_RAW), + SyscallSucceeds()); + + // Verify that writing yields EINVAL. + char buf[] = "This is such a weird little edge case"; + struct sockaddr_in6 sin6 = {}; + sin6.sin6_family = AF_INET6; + sin6.sin6_addr = in6addr_loopback; + ASSERT_THAT(sendto(sock, buf, sizeof(buf), 0 /* flags */, + reinterpret_cast<struct sockaddr*>(&sin6), sizeof(sin6)), + SyscallFailsWithErrno(EINVAL)); +} + +INSTANTIATE_TEST_SUITE_P( + AllInetTests, RawSocketTest, + ::testing::Combine(::testing::Values(IPPROTO_TCP, IPPROTO_UDP), + ::testing::Values(AF_INET, AF_INET6))); + +} // namespace + +} // namespace testing +} // namespace gvisor diff --git a/test/syscalls/linux/raw_socket_hdrincl.cc b/test/syscalls/linux/raw_socket_hdrincl.cc index 0a27506aa..2f25aceb2 100644 --- a/test/syscalls/linux/raw_socket_hdrincl.cc +++ b/test/syscalls/linux/raw_socket_hdrincl.cc @@ -167,7 +167,7 @@ TEST_F(RawHDRINCL, NotReadable) { // nothing to be read. char buf[117]; ASSERT_THAT(RetryEINTR(recv)(socket_, buf, sizeof(buf), MSG_DONTWAIT), - SyscallFailsWithErrno(EINVAL)); + SyscallFailsWithErrno(EAGAIN)); } // Test that we can connect() to a valid IP (loopback). @@ -178,6 +178,9 @@ TEST_F(RawHDRINCL, ConnectToLoopback) { } TEST_F(RawHDRINCL, SendWithoutConnectSucceeds) { + // FIXME(gvisor.dev/issue/3159): Test currently flaky. + SKIP_IF(true); + struct iphdr hdr = LoopbackHeader(); ASSERT_THAT(send(socket_, &hdr, sizeof(hdr), 0), SyscallSucceedsWithValue(sizeof(hdr))); @@ -273,14 +276,17 @@ TEST_F(RawHDRINCL, SendAndReceive) { // The network stack should have set the source address. EXPECT_EQ(src.sin_family, AF_INET); EXPECT_EQ(absl::gbswap_32(src.sin_addr.s_addr), INADDR_LOOPBACK); - // The packet ID should be 0, as the packet is less than 68 bytes. - struct iphdr iphdr = {}; - memcpy(&iphdr, recv_buf, sizeof(iphdr)); - EXPECT_EQ(iphdr.id, 0); + // The packet ID should not be 0, as the packet has DF=0. + struct iphdr* iphdr = reinterpret_cast<struct iphdr*>(recv_buf); + EXPECT_NE(iphdr->id, 0); } -// Send and receive a packet with nonzero IP ID. -TEST_F(RawHDRINCL, SendAndReceiveNonzeroID) { +// Send and receive a packet where the sendto address is not the same as the +// provided destination. +TEST_F(RawHDRINCL, SendAndReceiveDifferentAddress) { + // FIXME(gvisor.dev/issue/3160): Test currently flaky. + SKIP_IF(true); + int port = 40000; if (!IsRunningOnGvisor()) { port = static_cast<short>(ASSERT_NO_ERRNO_AND_VALUE( @@ -292,19 +298,24 @@ TEST_F(RawHDRINCL, SendAndReceiveNonzeroID) { FileDescriptor udp_sock = ASSERT_NO_ERRNO_AND_VALUE(Socket(AF_INET, SOCK_RAW, IPPROTO_UDP)); - // Construct a packet with an IP header, UDP header, and payload. Make the - // payload large enough to force an IP ID to be assigned. - constexpr char kPayload[128] = {}; + // Construct a packet with an IP header, UDP header, and payload. + constexpr char kPayload[] = "toto"; char packet[sizeof(struct iphdr) + sizeof(struct udphdr) + sizeof(kPayload)]; ASSERT_TRUE( FillPacket(packet, sizeof(packet), port, kPayload, sizeof(kPayload))); + // Overwrite the IP destination address with an IP we can't get to. + struct iphdr iphdr = {}; + memcpy(&iphdr, packet, sizeof(iphdr)); + iphdr.daddr = 42; + memcpy(packet, &iphdr, sizeof(iphdr)); socklen_t addrlen = sizeof(addr_); ASSERT_NO_FATAL_FAILURE(sendto(socket_, &packet, sizeof(packet), 0, reinterpret_cast<struct sockaddr*>(&addr_), addrlen)); - // Receive the payload. + // Receive the payload, since sendto should replace the bad destination with + // localhost. char recv_buf[sizeof(packet)]; struct sockaddr_in src; socklen_t src_size = sizeof(src); @@ -318,47 +329,58 @@ TEST_F(RawHDRINCL, SendAndReceiveNonzeroID) { // The network stack should have set the source address. EXPECT_EQ(src.sin_family, AF_INET); EXPECT_EQ(absl::gbswap_32(src.sin_addr.s_addr), INADDR_LOOPBACK); - // The packet ID should not be 0, as the packet was more than 68 bytes. - struct iphdr* iphdr = reinterpret_cast<struct iphdr*>(recv_buf); - EXPECT_NE(iphdr->id, 0); + // The packet ID should not be 0, as the packet has DF=0. + struct iphdr recv_iphdr = {}; + memcpy(&recv_iphdr, recv_buf, sizeof(recv_iphdr)); + EXPECT_NE(recv_iphdr.id, 0); + // The destination address should be localhost, not the bad IP we set + // initially. + EXPECT_EQ(absl::gbswap_32(recv_iphdr.daddr), INADDR_LOOPBACK); } -// Send and receive a packet where the sendto address is not the same as the -// provided destination. -TEST_F(RawHDRINCL, SendAndReceiveDifferentAddress) { +// Send and receive a packet w/ the IP_HDRINCL option set. +TEST_F(RawHDRINCL, SendAndReceiveIPHdrIncl) { int port = 40000; if (!IsRunningOnGvisor()) { port = static_cast<short>(ASSERT_NO_ERRNO_AND_VALUE( PortAvailable(0, AddressFamily::kIpv4, SocketType::kUdp, false))); } - // IPPROTO_RAW sockets are write-only. We'll have to open another socket to - // read what we write. - FileDescriptor udp_sock = + FileDescriptor recv_sock = + ASSERT_NO_ERRNO_AND_VALUE(Socket(AF_INET, SOCK_RAW, IPPROTO_UDP)); + + FileDescriptor send_sock = ASSERT_NO_ERRNO_AND_VALUE(Socket(AF_INET, SOCK_RAW, IPPROTO_UDP)); + // Enable IP_HDRINCL option so that we can build and send w/ an IP + // header. + constexpr int kSockOptOn = 1; + ASSERT_THAT(setsockopt(send_sock.get(), SOL_IP, IP_HDRINCL, &kSockOptOn, + sizeof(kSockOptOn)), + SyscallSucceeds()); + // This is not strictly required but we do it to make sure that setting + // IP_HDRINCL on a non IPPROTO_RAW socket does not prevent it from receiving + // packets. + ASSERT_THAT(setsockopt(recv_sock.get(), SOL_IP, IP_HDRINCL, &kSockOptOn, + sizeof(kSockOptOn)), + SyscallSucceeds()); + // Construct a packet with an IP header, UDP header, and payload. constexpr char kPayload[] = "toto"; char packet[sizeof(struct iphdr) + sizeof(struct udphdr) + sizeof(kPayload)]; ASSERT_TRUE( FillPacket(packet, sizeof(packet), port, kPayload, sizeof(kPayload))); - // Overwrite the IP destination address with an IP we can't get to. - struct iphdr iphdr = {}; - memcpy(&iphdr, packet, sizeof(iphdr)); - iphdr.daddr = 42; - memcpy(packet, &iphdr, sizeof(iphdr)); socklen_t addrlen = sizeof(addr_); - ASSERT_NO_FATAL_FAILURE(sendto(socket_, &packet, sizeof(packet), 0, + ASSERT_NO_FATAL_FAILURE(sendto(send_sock.get(), &packet, sizeof(packet), 0, reinterpret_cast<struct sockaddr*>(&addr_), addrlen)); - // Receive the payload, since sendto should replace the bad destination with - // localhost. + // Receive the payload. char recv_buf[sizeof(packet)]; struct sockaddr_in src; socklen_t src_size = sizeof(src); - ASSERT_THAT(recvfrom(udp_sock.get(), recv_buf, sizeof(recv_buf), 0, + ASSERT_THAT(recvfrom(recv_sock.get(), recv_buf, sizeof(recv_buf), 0, reinterpret_cast<struct sockaddr*>(&src), &src_size), SyscallSucceedsWithValue(sizeof(packet))); EXPECT_EQ( @@ -368,13 +390,20 @@ TEST_F(RawHDRINCL, SendAndReceiveDifferentAddress) { // The network stack should have set the source address. EXPECT_EQ(src.sin_family, AF_INET); EXPECT_EQ(absl::gbswap_32(src.sin_addr.s_addr), INADDR_LOOPBACK); - // The packet ID should be 0, as the packet is less than 68 bytes. - struct iphdr recv_iphdr = {}; - memcpy(&recv_iphdr, recv_buf, sizeof(recv_iphdr)); - EXPECT_EQ(recv_iphdr.id, 0); - // The destination address should be localhost, not the bad IP we set - // initially. - EXPECT_EQ(absl::gbswap_32(recv_iphdr.daddr), INADDR_LOOPBACK); + struct iphdr iphdr = {}; + memcpy(&iphdr, recv_buf, sizeof(iphdr)); + EXPECT_NE(iphdr.id, 0); + + // Also verify that the packet we just sent was not delivered to the + // IPPROTO_RAW socket. + { + char recv_buf[sizeof(packet)]; + struct sockaddr_in src; + socklen_t src_size = sizeof(src); + ASSERT_THAT(recvfrom(socket_, recv_buf, sizeof(recv_buf), MSG_DONTWAIT, + reinterpret_cast<struct sockaddr*>(&src), &src_size), + SyscallFailsWithErrno(EAGAIN)); + } } } // namespace diff --git a/test/syscalls/linux/raw_socket_ipv4.cc b/test/syscalls/linux/raw_socket_ipv4.cc deleted file mode 100644 index cde2f07c9..000000000 --- a/test/syscalls/linux/raw_socket_ipv4.cc +++ /dev/null @@ -1,392 +0,0 @@ -// Copyright 2019 The gVisor Authors. -// -// Licensed under the Apache License, Version 2.0 (the "License"); -// you may not use this file except in compliance with the License. -// You may obtain a copy of the License at -// -// http://www.apache.org/licenses/LICENSE-2.0 -// -// Unless required by applicable law or agreed to in writing, software -// distributed under the License is distributed on an "AS IS" BASIS, -// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -// See the License for the specific language governing permissions and -// limitations under the License. - -#include <linux/capability.h> -#include <netinet/in.h> -#include <netinet/ip.h> -#include <netinet/ip_icmp.h> -#include <poll.h> -#include <sys/socket.h> -#include <sys/types.h> -#include <unistd.h> - -#include <algorithm> - -#include "gtest/gtest.h" -#include "test/syscalls/linux/socket_test_util.h" -#include "test/syscalls/linux/unix_domain_socket_test_util.h" -#include "test/util/capability_util.h" -#include "test/util/file_descriptor.h" -#include "test/util/test_util.h" - -// Note: in order to run these tests, /proc/sys/net/ipv4/ping_group_range will -// need to be configured to let the superuser create ping sockets (see icmp(7)). - -namespace gvisor { -namespace testing { - -namespace { - -// Fixture for tests parameterized by protocol. -class RawSocketTest : public ::testing::TestWithParam<int> { - protected: - // Creates a socket to be used in tests. - void SetUp() override; - - // Closes the socket created by SetUp(). - void TearDown() override; - - // Sends buf via s_. - void SendBuf(const char* buf, int buf_len); - - // Sends buf to the provided address via the provided socket. - void SendBufTo(int sock, const struct sockaddr_in& addr, const char* buf, - int buf_len); - - // Reads from s_ into recv_buf. - void ReceiveBuf(char* recv_buf, size_t recv_buf_len); - - int Protocol() { return GetParam(); } - - // The socket used for both reading and writing. - int s_; - - // The loopback address. - struct sockaddr_in addr_; -}; - -void RawSocketTest::SetUp() { - if (!ASSERT_NO_ERRNO_AND_VALUE(HaveCapability(CAP_NET_RAW))) { - ASSERT_THAT(socket(AF_INET, SOCK_RAW, Protocol()), - SyscallFailsWithErrno(EPERM)); - GTEST_SKIP(); - } - - ASSERT_THAT(s_ = socket(AF_INET, SOCK_RAW, Protocol()), SyscallSucceeds()); - - addr_ = {}; - - // We don't set ports because raw sockets don't have a notion of ports. - addr_.sin_addr.s_addr = htonl(INADDR_LOOPBACK); - addr_.sin_family = AF_INET; -} - -void RawSocketTest::TearDown() { - // TearDown will be run even if we skip the test. - if (ASSERT_NO_ERRNO_AND_VALUE(HaveCapability(CAP_NET_RAW))) { - EXPECT_THAT(close(s_), SyscallSucceeds()); - } -} - -// We should be able to create multiple raw sockets for the same protocol. -// BasicRawSocket::Setup creates the first one, so we only have to create one -// more here. -TEST_P(RawSocketTest, MultipleCreation) { - SKIP_IF(!ASSERT_NO_ERRNO_AND_VALUE(HaveCapability(CAP_NET_RAW))); - - int s2; - ASSERT_THAT(s2 = socket(AF_INET, SOCK_RAW, Protocol()), SyscallSucceeds()); - - ASSERT_THAT(close(s2), SyscallSucceeds()); -} - -// Test that shutting down an unconnected socket fails. -TEST_P(RawSocketTest, FailShutdownWithoutConnect) { - SKIP_IF(!ASSERT_NO_ERRNO_AND_VALUE(HaveCapability(CAP_NET_RAW))); - - ASSERT_THAT(shutdown(s_, SHUT_WR), SyscallFailsWithErrno(ENOTCONN)); - ASSERT_THAT(shutdown(s_, SHUT_RD), SyscallFailsWithErrno(ENOTCONN)); -} - -// Shutdown is a no-op for raw sockets (and datagram sockets in general). -TEST_P(RawSocketTest, ShutdownWriteNoop) { - SKIP_IF(!ASSERT_NO_ERRNO_AND_VALUE(HaveCapability(CAP_NET_RAW))); - - ASSERT_THAT( - connect(s_, reinterpret_cast<struct sockaddr*>(&addr_), sizeof(addr_)), - SyscallSucceeds()); - ASSERT_THAT(shutdown(s_, SHUT_WR), SyscallSucceeds()); - - // Arbitrary. - constexpr char kBuf[] = "noop"; - ASSERT_THAT(RetryEINTR(write)(s_, kBuf, sizeof(kBuf)), - SyscallSucceedsWithValue(sizeof(kBuf))); -} - -// Shutdown is a no-op for raw sockets (and datagram sockets in general). -TEST_P(RawSocketTest, ShutdownReadNoop) { - SKIP_IF(!ASSERT_NO_ERRNO_AND_VALUE(HaveCapability(CAP_NET_RAW))); - - ASSERT_THAT( - connect(s_, reinterpret_cast<struct sockaddr*>(&addr_), sizeof(addr_)), - SyscallSucceeds()); - ASSERT_THAT(shutdown(s_, SHUT_RD), SyscallSucceeds()); - - // Arbitrary. - constexpr char kBuf[] = "gdg"; - ASSERT_NO_FATAL_FAILURE(SendBuf(kBuf, sizeof(kBuf))); - - constexpr size_t kReadSize = sizeof(kBuf) + sizeof(struct iphdr); - char c[kReadSize]; - ASSERT_THAT(read(s_, &c, sizeof(c)), SyscallSucceedsWithValue(kReadSize)); -} - -// Test that listen() fails. -TEST_P(RawSocketTest, FailListen) { - SKIP_IF(!ASSERT_NO_ERRNO_AND_VALUE(HaveCapability(CAP_NET_RAW))); - - ASSERT_THAT(listen(s_, 1), SyscallFailsWithErrno(ENOTSUP)); -} - -// Test that accept() fails. -TEST_P(RawSocketTest, FailAccept) { - SKIP_IF(!ASSERT_NO_ERRNO_AND_VALUE(HaveCapability(CAP_NET_RAW))); - - struct sockaddr saddr; - socklen_t addrlen; - ASSERT_THAT(accept(s_, &saddr, &addrlen), SyscallFailsWithErrno(ENOTSUP)); -} - -// Test that getpeername() returns nothing before connect(). -TEST_P(RawSocketTest, FailGetPeerNameBeforeConnect) { - SKIP_IF(!ASSERT_NO_ERRNO_AND_VALUE(HaveCapability(CAP_NET_RAW))); - - struct sockaddr saddr; - socklen_t addrlen = sizeof(saddr); - ASSERT_THAT(getpeername(s_, &saddr, &addrlen), - SyscallFailsWithErrno(ENOTCONN)); -} - -// Test that getpeername() returns something after connect(). -TEST_P(RawSocketTest, GetPeerName) { - SKIP_IF(!ASSERT_NO_ERRNO_AND_VALUE(HaveCapability(CAP_NET_RAW))); - - ASSERT_THAT( - connect(s_, reinterpret_cast<struct sockaddr*>(&addr_), sizeof(addr_)), - SyscallSucceeds()); - struct sockaddr saddr; - socklen_t addrlen = sizeof(saddr); - ASSERT_THAT(getpeername(s_, &saddr, &addrlen), - SyscallFailsWithErrno(ENOTCONN)); - ASSERT_GT(addrlen, 0); -} - -// Test that the socket is writable immediately. -TEST_P(RawSocketTest, PollWritableImmediately) { - SKIP_IF(!ASSERT_NO_ERRNO_AND_VALUE(HaveCapability(CAP_NET_RAW))); - - struct pollfd pfd = {}; - pfd.fd = s_; - pfd.events = POLLOUT; - ASSERT_THAT(RetryEINTR(poll)(&pfd, 1, 10000), SyscallSucceedsWithValue(1)); -} - -// Test that the socket isn't readable before receiving anything. -TEST_P(RawSocketTest, PollNotReadableInitially) { - SKIP_IF(!ASSERT_NO_ERRNO_AND_VALUE(HaveCapability(CAP_NET_RAW))); - - // Try to receive data with MSG_DONTWAIT, which returns immediately if there's - // nothing to be read. - char buf[117]; - ASSERT_THAT(RetryEINTR(recv)(s_, buf, sizeof(buf), MSG_DONTWAIT), - SyscallFailsWithErrno(EAGAIN)); -} - -// Test that the socket becomes readable once something is written to it. -TEST_P(RawSocketTest, PollTriggeredOnWrite) { - SKIP_IF(!ASSERT_NO_ERRNO_AND_VALUE(HaveCapability(CAP_NET_RAW))); - - // Write something so that there's data to be read. - // Arbitrary. - constexpr char kBuf[] = "JP5"; - ASSERT_NO_FATAL_FAILURE(SendBuf(kBuf, sizeof(kBuf))); - - struct pollfd pfd = {}; - pfd.fd = s_; - pfd.events = POLLIN; - ASSERT_THAT(RetryEINTR(poll)(&pfd, 1, 10000), SyscallSucceedsWithValue(1)); -} - -// Test that we can connect() to a valid IP (loopback). -TEST_P(RawSocketTest, ConnectToLoopback) { - SKIP_IF(!ASSERT_NO_ERRNO_AND_VALUE(HaveCapability(CAP_NET_RAW))); - - ASSERT_THAT( - connect(s_, reinterpret_cast<struct sockaddr*>(&addr_), sizeof(addr_)), - SyscallSucceeds()); -} - -// Test that calling send() without connect() fails. -TEST_P(RawSocketTest, SendWithoutConnectFails) { - SKIP_IF(!ASSERT_NO_ERRNO_AND_VALUE(HaveCapability(CAP_NET_RAW))); - - // Arbitrary. - constexpr char kBuf[] = "Endgame was good"; - ASSERT_THAT(send(s_, kBuf, sizeof(kBuf), 0), - SyscallFailsWithErrno(EDESTADDRREQ)); -} - -// Bind to localhost. -TEST_P(RawSocketTest, BindToLocalhost) { - SKIP_IF(!ASSERT_NO_ERRNO_AND_VALUE(HaveCapability(CAP_NET_RAW))); - - ASSERT_THAT( - bind(s_, reinterpret_cast<struct sockaddr*>(&addr_), sizeof(addr_)), - SyscallSucceeds()); -} - -// Bind to a different address. -TEST_P(RawSocketTest, BindToInvalid) { - SKIP_IF(!ASSERT_NO_ERRNO_AND_VALUE(HaveCapability(CAP_NET_RAW))); - - struct sockaddr_in bind_addr = {}; - bind_addr.sin_family = AF_INET; - bind_addr.sin_addr = {1}; // 1.0.0.0 - An address that we can't bind to. - ASSERT_THAT(bind(s_, reinterpret_cast<struct sockaddr*>(&bind_addr), - sizeof(bind_addr)), - SyscallFailsWithErrno(EADDRNOTAVAIL)); -} - -// Send and receive an packet. -TEST_P(RawSocketTest, SendAndReceive) { - SKIP_IF(!ASSERT_NO_ERRNO_AND_VALUE(HaveCapability(CAP_NET_RAW))); - - // Arbitrary. - constexpr char kBuf[] = "TB12"; - ASSERT_NO_FATAL_FAILURE(SendBuf(kBuf, sizeof(kBuf))); - - // Receive the packet and make sure it's identical. - char recv_buf[sizeof(kBuf) + sizeof(struct iphdr)]; - ASSERT_NO_FATAL_FAILURE(ReceiveBuf(recv_buf, sizeof(recv_buf))); - EXPECT_EQ(memcmp(recv_buf + sizeof(struct iphdr), kBuf, sizeof(kBuf)), 0); -} - -// We should be able to create multiple raw sockets for the same protocol and -// receive the same packet on both. -TEST_P(RawSocketTest, MultipleSocketReceive) { - SKIP_IF(!ASSERT_NO_ERRNO_AND_VALUE(HaveCapability(CAP_NET_RAW))); - - int s2; - ASSERT_THAT(s2 = socket(AF_INET, SOCK_RAW, Protocol()), SyscallSucceeds()); - - // Arbitrary. - constexpr char kBuf[] = "TB10"; - ASSERT_NO_FATAL_FAILURE(SendBuf(kBuf, sizeof(kBuf))); - - // Receive it on socket 1. - char recv_buf1[sizeof(kBuf) + sizeof(struct iphdr)]; - ASSERT_NO_FATAL_FAILURE(ReceiveBuf(recv_buf1, sizeof(recv_buf1))); - - // Receive it on socket 2. - char recv_buf2[sizeof(kBuf) + sizeof(struct iphdr)]; - ASSERT_NO_FATAL_FAILURE(RecvNoCmsg(s2, recv_buf2, sizeof(recv_buf2))); - - EXPECT_EQ(memcmp(recv_buf1 + sizeof(struct iphdr), - recv_buf2 + sizeof(struct iphdr), sizeof(kBuf)), - 0); - - ASSERT_THAT(close(s2), SyscallSucceeds()); -} - -// Test that connect sends packets to the right place. -TEST_P(RawSocketTest, SendAndReceiveViaConnect) { - SKIP_IF(!ASSERT_NO_ERRNO_AND_VALUE(HaveCapability(CAP_NET_RAW))); - - ASSERT_THAT( - connect(s_, reinterpret_cast<struct sockaddr*>(&addr_), sizeof(addr_)), - SyscallSucceeds()); - - // Arbitrary. - constexpr char kBuf[] = "JH4"; - ASSERT_THAT(send(s_, kBuf, sizeof(kBuf), 0), - SyscallSucceedsWithValue(sizeof(kBuf))); - - // Receive the packet and make sure it's identical. - char recv_buf[sizeof(kBuf) + sizeof(struct iphdr)]; - ASSERT_NO_FATAL_FAILURE(ReceiveBuf(recv_buf, sizeof(recv_buf))); - EXPECT_EQ(memcmp(recv_buf + sizeof(struct iphdr), kBuf, sizeof(kBuf)), 0); -} - -// Bind to localhost, then send and receive packets. -TEST_P(RawSocketTest, BindSendAndReceive) { - SKIP_IF(!ASSERT_NO_ERRNO_AND_VALUE(HaveCapability(CAP_NET_RAW))); - - ASSERT_THAT( - bind(s_, reinterpret_cast<struct sockaddr*>(&addr_), sizeof(addr_)), - SyscallSucceeds()); - - // Arbitrary. - constexpr char kBuf[] = "DR16"; - ASSERT_NO_FATAL_FAILURE(SendBuf(kBuf, sizeof(kBuf))); - - // Receive the packet and make sure it's identical. - char recv_buf[sizeof(kBuf) + sizeof(struct iphdr)]; - ASSERT_NO_FATAL_FAILURE(ReceiveBuf(recv_buf, sizeof(recv_buf))); - EXPECT_EQ(memcmp(recv_buf + sizeof(struct iphdr), kBuf, sizeof(kBuf)), 0); -} - -// Bind and connect to localhost and send/receive packets. -TEST_P(RawSocketTest, BindConnectSendAndReceive) { - SKIP_IF(!ASSERT_NO_ERRNO_AND_VALUE(HaveCapability(CAP_NET_RAW))); - - ASSERT_THAT( - bind(s_, reinterpret_cast<struct sockaddr*>(&addr_), sizeof(addr_)), - SyscallSucceeds()); - ASSERT_THAT( - connect(s_, reinterpret_cast<struct sockaddr*>(&addr_), sizeof(addr_)), - SyscallSucceeds()); - - // Arbitrary. - constexpr char kBuf[] = "DG88"; - ASSERT_NO_FATAL_FAILURE(SendBuf(kBuf, sizeof(kBuf))); - - // Receive the packet and make sure it's identical. - char recv_buf[sizeof(kBuf) + sizeof(struct iphdr)]; - ASSERT_NO_FATAL_FAILURE(ReceiveBuf(recv_buf, sizeof(recv_buf))); - EXPECT_EQ(memcmp(recv_buf + sizeof(struct iphdr), kBuf, sizeof(kBuf)), 0); -} - -void RawSocketTest::SendBuf(const char* buf, int buf_len) { - ASSERT_NO_FATAL_FAILURE(SendBufTo(s_, addr_, buf, buf_len)); -} - -void RawSocketTest::SendBufTo(int sock, const struct sockaddr_in& addr, - const char* buf, int buf_len) { - // It's safe to use const_cast here because sendmsg won't modify the iovec or - // address. - struct iovec iov = {}; - iov.iov_base = static_cast<void*>(const_cast<char*>(buf)); - iov.iov_len = static_cast<size_t>(buf_len); - struct msghdr msg = {}; - msg.msg_name = static_cast<void*>(const_cast<struct sockaddr_in*>(&addr)); - msg.msg_namelen = sizeof(addr); - msg.msg_iov = &iov; - msg.msg_iovlen = 1; - msg.msg_control = NULL; - msg.msg_controllen = 0; - msg.msg_flags = 0; - ASSERT_THAT(sendmsg(sock, &msg, 0), SyscallSucceedsWithValue(buf_len)); -} - -void RawSocketTest::ReceiveBuf(char* recv_buf, size_t recv_buf_len) { - ASSERT_NO_FATAL_FAILURE(RecvNoCmsg(s_, recv_buf, recv_buf_len)); -} - -INSTANTIATE_TEST_SUITE_P(AllInetTests, RawSocketTest, - ::testing::Values(IPPROTO_TCP, IPPROTO_UDP)); - -} // namespace - -} // namespace testing -} // namespace gvisor diff --git a/test/syscalls/linux/socket.cc b/test/syscalls/linux/socket.cc index e0a4d0985..c20cd3fcc 100644 --- a/test/syscalls/linux/socket.cc +++ b/test/syscalls/linux/socket.cc @@ -86,37 +86,12 @@ TEST(SocketTest, UnixSocketStat) { EXPECT_EQ(statbuf.st_mode, S_IFSOCK | sock_perm & ~mask); // Timestamps should be equal and non-zero. - EXPECT_NE(statbuf.st_atime, 0); - EXPECT_EQ(statbuf.st_atime, statbuf.st_mtime); - EXPECT_EQ(statbuf.st_atime, statbuf.st_ctime); -} - -TEST(SocketTest, UnixConnectNeedsWritePerm) { - SKIP_IF(IsRunningWithVFS1()); - - FileDescriptor bound = - ASSERT_NO_ERRNO_AND_VALUE(Socket(AF_UNIX, SOCK_STREAM, PF_UNIX)); - - struct sockaddr_un addr = - ASSERT_NO_ERRNO_AND_VALUE(UniqueUnixAddr(/*abstract=*/false, AF_UNIX)); - ASSERT_THAT(bind(bound.get(), reinterpret_cast<struct sockaddr*>(&addr), - sizeof(addr)), - SyscallSucceeds()); - ASSERT_THAT(listen(bound.get(), 1), SyscallSucceeds()); - - // Connect should fail without write perms. - ASSERT_THAT(chmod(addr.sun_path, 0500), SyscallSucceeds()); - FileDescriptor client = - ASSERT_NO_ERRNO_AND_VALUE(Socket(AF_UNIX, SOCK_STREAM, PF_UNIX)); - EXPECT_THAT(connect(client.get(), reinterpret_cast<struct sockaddr*>(&addr), - sizeof(addr)), - SyscallFailsWithErrno(EACCES)); - - // Connect should succeed with write perms. - ASSERT_THAT(chmod(addr.sun_path, 0200), SyscallSucceeds()); - EXPECT_THAT(connect(client.get(), reinterpret_cast<struct sockaddr*>(&addr), - sizeof(addr)), - SyscallSucceeds()); + // TODO(b/158882152): Sockets currently don't implement timestamps. + if (!IsRunningOnGvisor()) { + EXPECT_NE(statbuf.st_atime, 0); + EXPECT_EQ(statbuf.st_atime, statbuf.st_mtime); + EXPECT_EQ(statbuf.st_atime, statbuf.st_ctime); + } } using SocketOpenTest = ::testing::TestWithParam<int>; diff --git a/test/syscalls/linux/socket_bind_to_device_sequence.cc b/test/syscalls/linux/socket_bind_to_device_sequence.cc index 1967329ee..d3cc71dbf 100644 --- a/test/syscalls/linux/socket_bind_to_device_sequence.cc +++ b/test/syscalls/linux/socket_bind_to_device_sequence.cc @@ -180,7 +180,7 @@ class BindToDeviceSequenceTest : public ::testing::TestWithParam<SocketKind> { private: SocketKind socket_factory_; // devices maps from the device id in the test case to the name of the device. - std::unordered_map<int, string> devices_; + absl::node_hash_map<int, string> devices_; // These are the tunnels that were created for the test and will be destroyed // by the destructor. vector<std::unique_ptr<Tunnel>> tunnels_; @@ -363,9 +363,6 @@ TEST_P(BindToDeviceSequenceTest, BindTwiceWithReuseOnce) { } TEST_P(BindToDeviceSequenceTest, BindWithReuseAddr) { - // FIXME(b/129164367): Support SO_REUSEADDR on UDP sockets. - SKIP_IF(IsRunningOnGvisor()); - ASSERT_NO_FATAL_FAILURE( BindSocket(/* reusePort */ false, /* reuse_addr */ true)); ASSERT_NO_FATAL_FAILURE(BindSocket(/* reuse_port */ false, @@ -405,9 +402,6 @@ TEST_P(BindToDeviceSequenceTest, BindReuseAddrReusePortThenReusePort) { } TEST_P(BindToDeviceSequenceTest, BindReuseAddrReusePortThenReuseAddr) { - // FIXME(b/129164367): Support SO_REUSEADDR on UDP sockets. - SKIP_IF(IsRunningOnGvisor()); - ASSERT_NO_FATAL_FAILURE(BindSocket(/* reuse_port */ true, /* reuse_addr */ true, /* bind_to_device */ 0)); @@ -434,9 +428,6 @@ TEST_P(BindToDeviceSequenceTest, BindDoubleReuseAddrReusePortThenReusePort) { } TEST_P(BindToDeviceSequenceTest, BindDoubleReuseAddrReusePortThenReuseAddr) { - // FIXME(b/129164367): Support SO_REUSEADDR on UDP sockets. - SKIP_IF(IsRunningOnGvisor()); - ASSERT_NO_FATAL_FAILURE(BindSocket( /* reuse_port */ true, /* reuse_addr */ true, /* bind_to_device */ 0)); ASSERT_NO_FATAL_FAILURE(BindSocket(/* reuse_port */ true, @@ -469,10 +460,20 @@ TEST_P(BindToDeviceSequenceTest, BindReuseAddrThenReuseAddr) { /* bind_to_device */ 0, EADDRINUSE)); } -// This behavior seems like a bug? TEST_P(BindToDeviceSequenceTest, BindReuseAddrThenReuseAddrReusePortThenReuseAddr) { - // FIXME(b/129164367): Support SO_REUSEADDR on UDP sockets. + // The behavior described in this test seems like a Linux bug. It doesn't + // make any sense and it is unlikely that any applications rely on it. + // + // Both SO_REUSEADDR and SO_REUSEPORT allow binding multiple UDP sockets to + // the same address and deliver each packet to exactly one of the bound + // sockets. If both are enabled, one of the strategies is selected to route + // packets. The strategy is selected dynamically based on the settings of the + // currently bound sockets. Usually, the strategy is selected based on the + // common setting (SO_REUSEADDR or SO_REUSEPORT) amongst the sockets, but for + // some reason, Linux allows binding sets of sockets with no overlapping + // settings in some situations. In this case, it is not obvious which strategy + // would be selected as the configured setting is a contradiction. SKIP_IF(IsRunningOnGvisor()); ASSERT_NO_FATAL_FAILURE(BindSocket( diff --git a/test/syscalls/linux/socket_capability.cc b/test/syscalls/linux/socket_capability.cc new file mode 100644 index 000000000..84b5b2b21 --- /dev/null +++ b/test/syscalls/linux/socket_capability.cc @@ -0,0 +1,61 @@ +// Copyright 2018 The gVisor Authors. +// +// Licensed under the Apache License, Version 2.0 (the "License"); +// you may not use this file except in compliance with the License. +// You may obtain a copy of the License at +// +// http://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, software +// distributed under the License is distributed on an "AS IS" BASIS, +// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +// See the License for the specific language governing permissions and +// limitations under the License. + +// Subset of socket tests that need Linux-specific headers (compared to POSIX +// headers). + +#include "gtest/gtest.h" +#include "test/syscalls/linux/socket_test_util.h" +#include "test/util/capability_util.h" +#include "test/util/file_descriptor.h" +#include "test/util/test_util.h" + +namespace gvisor { +namespace testing { + +TEST(SocketTest, UnixConnectNeedsWritePerm) { + SKIP_IF(IsRunningWithVFS1()); + + FileDescriptor bound = + ASSERT_NO_ERRNO_AND_VALUE(Socket(AF_UNIX, SOCK_STREAM, PF_UNIX)); + + struct sockaddr_un addr = + ASSERT_NO_ERRNO_AND_VALUE(UniqueUnixAddr(/*abstract=*/false, AF_UNIX)); + ASSERT_THAT(bind(bound.get(), reinterpret_cast<struct sockaddr*>(&addr), + sizeof(addr)), + SyscallSucceeds()); + ASSERT_THAT(listen(bound.get(), 1), SyscallSucceeds()); + + // Drop capabilites that allow us to override permision checks. Otherwise if + // the test is run as root, the connect below will bypass permission checks + // and succeed unexpectedly. + ASSERT_NO_ERRNO(SetCapability(CAP_DAC_OVERRIDE, false)); + + // Connect should fail without write perms. + ASSERT_THAT(chmod(addr.sun_path, 0500), SyscallSucceeds()); + FileDescriptor client = + ASSERT_NO_ERRNO_AND_VALUE(Socket(AF_UNIX, SOCK_STREAM, PF_UNIX)); + ASSERT_THAT(connect(client.get(), reinterpret_cast<struct sockaddr*>(&addr), + sizeof(addr)), + SyscallFailsWithErrno(EACCES)); + + // Connect should succeed with write perms. + ASSERT_THAT(chmod(addr.sun_path, 0200), SyscallSucceeds()); + EXPECT_THAT(connect(client.get(), reinterpret_cast<struct sockaddr*>(&addr), + sizeof(addr)), + SyscallSucceeds()); +} + +} // namespace testing +} // namespace gvisor diff --git a/test/syscalls/linux/socket_inet_loopback.cc b/test/syscalls/linux/socket_inet_loopback.cc index fa890ec98..18b9e4b70 100644 --- a/test/syscalls/linux/socket_inet_loopback.cc +++ b/test/syscalls/linux/socket_inet_loopback.cc @@ -17,6 +17,7 @@ #include <netinet/tcp.h> #include <poll.h> #include <string.h> +#include <sys/socket.h> #include <atomic> #include <iostream> @@ -686,24 +687,9 @@ TEST_P(SocketInetLoopbackTest, TCPFinWait2Test_NoRandomSave) { // be restarted causing the final bind/connect to fail. DisableSave ds; - // TODO(gvisor.dev/issue/1030): Portmanager does not track all 5 tuple - // reservations which causes the bind() to succeed on gVisor but connect - // correctly fails. - if (IsRunningOnGvisor()) { - ASSERT_THAT( - bind(conn_fd2.get(), reinterpret_cast<sockaddr*>(&conn_bound_addr), - conn_addrlen), - SyscallSucceeds()); - ASSERT_THAT(RetryEINTR(connect)(conn_fd2.get(), - reinterpret_cast<sockaddr*>(&conn_addr), - conn_addrlen), - SyscallFailsWithErrno(EADDRINUSE)); - } else { - ASSERT_THAT( - bind(conn_fd2.get(), reinterpret_cast<sockaddr*>(&conn_bound_addr), - conn_addrlen), - SyscallFailsWithErrno(EADDRINUSE)); - } + ASSERT_THAT(bind(conn_fd2.get(), + reinterpret_cast<sockaddr*>(&conn_bound_addr), conn_addrlen), + SyscallFailsWithErrno(EADDRINUSE)); // Sleep for a little over the linger timeout to reduce flakiness in // save/restore tests. @@ -1737,6 +1723,171 @@ TEST_P(SocketMultiProtocolInetLoopbackTest, DualStackV6AnyReservesEverything) { ASSERT_THAT(bind(fd_v4.get(), reinterpret_cast<sockaddr*>(&addr_v4), test_addr_v4.addr_len), SyscallFailsWithErrno(EADDRINUSE)); + + // Verify that binding the v4 any on the same port with a v4 socket + // fails. + TestAddress const& test_addr_v4_any = V4Any(); + sockaddr_storage addr_v4_any = test_addr_v4_any.addr; + ASSERT_NO_ERRNO(SetAddrPort(test_addr_v4_any.family(), &addr_v4_any, port)); + const FileDescriptor fd_v4_any = ASSERT_NO_ERRNO_AND_VALUE( + Socket(test_addr_v4_any.family(), param.type, 0)); + ASSERT_THAT(bind(fd_v4_any.get(), reinterpret_cast<sockaddr*>(&addr_v4_any), + test_addr_v4_any.addr_len), + SyscallFailsWithErrno(EADDRINUSE)); +} + +TEST_P(SocketMultiProtocolInetLoopbackTest, + DualStackV6AnyReuseAddrDoesNotReserveV4Any) { + auto const& param = GetParam(); + + // Bind the v6 any on a dual stack socket. + TestAddress const& test_addr_dual = V6Any(); + sockaddr_storage addr_dual = test_addr_dual.addr; + const FileDescriptor fd_dual = + ASSERT_NO_ERRNO_AND_VALUE(Socket(test_addr_dual.family(), param.type, 0)); + ASSERT_THAT(setsockopt(fd_dual.get(), SOL_SOCKET, SO_REUSEADDR, &kSockOptOn, + sizeof(kSockOptOn)), + SyscallSucceeds()); + ASSERT_THAT(bind(fd_dual.get(), reinterpret_cast<sockaddr*>(&addr_dual), + test_addr_dual.addr_len), + SyscallSucceeds()); + + // Get the port that we bound. + socklen_t addrlen = test_addr_dual.addr_len; + ASSERT_THAT(getsockname(fd_dual.get(), + reinterpret_cast<sockaddr*>(&addr_dual), &addrlen), + SyscallSucceeds()); + uint16_t const port = + ASSERT_NO_ERRNO_AND_VALUE(AddrPort(test_addr_dual.family(), addr_dual)); + + // Verify that binding the v4 any on the same port with a v4 socket succeeds. + TestAddress const& test_addr_v4_any = V4Any(); + sockaddr_storage addr_v4_any = test_addr_v4_any.addr; + ASSERT_NO_ERRNO(SetAddrPort(test_addr_v4_any.family(), &addr_v4_any, port)); + const FileDescriptor fd_v4_any = ASSERT_NO_ERRNO_AND_VALUE( + Socket(test_addr_v4_any.family(), param.type, 0)); + ASSERT_THAT(setsockopt(fd_v4_any.get(), SOL_SOCKET, SO_REUSEADDR, &kSockOptOn, + sizeof(kSockOptOn)), + SyscallSucceeds()); + ASSERT_THAT(bind(fd_v4_any.get(), reinterpret_cast<sockaddr*>(&addr_v4_any), + test_addr_v4_any.addr_len), + SyscallSucceeds()); +} + +TEST_P(SocketMultiProtocolInetLoopbackTest, + DualStackV6AnyReuseAddrListenReservesV4Any) { + auto const& param = GetParam(); + + // Only TCP sockets are supported. + SKIP_IF((param.type & SOCK_STREAM) == 0); + + // Bind the v6 any on a dual stack socket. + TestAddress const& test_addr_dual = V6Any(); + sockaddr_storage addr_dual = test_addr_dual.addr; + const FileDescriptor fd_dual = + ASSERT_NO_ERRNO_AND_VALUE(Socket(test_addr_dual.family(), param.type, 0)); + ASSERT_THAT(setsockopt(fd_dual.get(), SOL_SOCKET, SO_REUSEADDR, &kSockOptOn, + sizeof(kSockOptOn)), + SyscallSucceeds()); + ASSERT_THAT(bind(fd_dual.get(), reinterpret_cast<sockaddr*>(&addr_dual), + test_addr_dual.addr_len), + SyscallSucceeds()); + + ASSERT_THAT(listen(fd_dual.get(), 5), SyscallSucceeds()); + + // Get the port that we bound. + socklen_t addrlen = test_addr_dual.addr_len; + ASSERT_THAT(getsockname(fd_dual.get(), + reinterpret_cast<sockaddr*>(&addr_dual), &addrlen), + SyscallSucceeds()); + uint16_t const port = + ASSERT_NO_ERRNO_AND_VALUE(AddrPort(test_addr_dual.family(), addr_dual)); + + // Verify that binding the v4 any on the same port with a v4 socket succeeds. + TestAddress const& test_addr_v4_any = V4Any(); + sockaddr_storage addr_v4_any = test_addr_v4_any.addr; + ASSERT_NO_ERRNO(SetAddrPort(test_addr_v4_any.family(), &addr_v4_any, port)); + const FileDescriptor fd_v4_any = ASSERT_NO_ERRNO_AND_VALUE( + Socket(test_addr_v4_any.family(), param.type, 0)); + ASSERT_THAT(setsockopt(fd_v4_any.get(), SOL_SOCKET, SO_REUSEADDR, &kSockOptOn, + sizeof(kSockOptOn)), + SyscallSucceeds()); + + ASSERT_THAT(bind(fd_v4_any.get(), reinterpret_cast<sockaddr*>(&addr_v4_any), + test_addr_v4_any.addr_len), + SyscallFailsWithErrno(EADDRINUSE)); +} + +TEST_P(SocketMultiProtocolInetLoopbackTest, + DualStackV6AnyWithListenReservesEverything) { + auto const& param = GetParam(); + + // Only TCP sockets are supported. + SKIP_IF((param.type & SOCK_STREAM) == 0); + + // Bind the v6 any on a dual stack socket. + TestAddress const& test_addr_dual = V6Any(); + sockaddr_storage addr_dual = test_addr_dual.addr; + const FileDescriptor fd_dual = + ASSERT_NO_ERRNO_AND_VALUE(Socket(test_addr_dual.family(), param.type, 0)); + ASSERT_THAT(bind(fd_dual.get(), reinterpret_cast<sockaddr*>(&addr_dual), + test_addr_dual.addr_len), + SyscallSucceeds()); + + ASSERT_THAT(listen(fd_dual.get(), 5), SyscallSucceeds()); + + // Get the port that we bound. + socklen_t addrlen = test_addr_dual.addr_len; + ASSERT_THAT(getsockname(fd_dual.get(), + reinterpret_cast<sockaddr*>(&addr_dual), &addrlen), + SyscallSucceeds()); + uint16_t const port = + ASSERT_NO_ERRNO_AND_VALUE(AddrPort(test_addr_dual.family(), addr_dual)); + + // Verify that binding the v6 loopback with the same port fails. + TestAddress const& test_addr_v6 = V6Loopback(); + sockaddr_storage addr_v6 = test_addr_v6.addr; + ASSERT_NO_ERRNO(SetAddrPort(test_addr_v6.family(), &addr_v6, port)); + const FileDescriptor fd_v6 = + ASSERT_NO_ERRNO_AND_VALUE(Socket(test_addr_v6.family(), param.type, 0)); + ASSERT_THAT(bind(fd_v6.get(), reinterpret_cast<sockaddr*>(&addr_v6), + test_addr_v6.addr_len), + SyscallFailsWithErrno(EADDRINUSE)); + + // Verify that binding the v4 loopback on the same port with a v6 socket + // fails. + TestAddress const& test_addr_v4_mapped = V4MappedLoopback(); + sockaddr_storage addr_v4_mapped = test_addr_v4_mapped.addr; + ASSERT_NO_ERRNO( + SetAddrPort(test_addr_v4_mapped.family(), &addr_v4_mapped, port)); + const FileDescriptor fd_v4_mapped = ASSERT_NO_ERRNO_AND_VALUE( + Socket(test_addr_v4_mapped.family(), param.type, 0)); + ASSERT_THAT( + bind(fd_v4_mapped.get(), reinterpret_cast<sockaddr*>(&addr_v4_mapped), + test_addr_v4_mapped.addr_len), + SyscallFailsWithErrno(EADDRINUSE)); + + // Verify that binding the v4 loopback on the same port with a v4 socket + // fails. + TestAddress const& test_addr_v4 = V4Loopback(); + sockaddr_storage addr_v4 = test_addr_v4.addr; + ASSERT_NO_ERRNO(SetAddrPort(test_addr_v4.family(), &addr_v4, port)); + const FileDescriptor fd_v4 = + ASSERT_NO_ERRNO_AND_VALUE(Socket(test_addr_v4.family(), param.type, 0)); + ASSERT_THAT(bind(fd_v4.get(), reinterpret_cast<sockaddr*>(&addr_v4), + test_addr_v4.addr_len), + SyscallFailsWithErrno(EADDRINUSE)); + + // Verify that binding the v4 any on the same port with a v4 socket + // fails. + TestAddress const& test_addr_v4_any = V4Any(); + sockaddr_storage addr_v4_any = test_addr_v4_any.addr; + ASSERT_NO_ERRNO(SetAddrPort(test_addr_v4_any.family(), &addr_v4_any, port)); + const FileDescriptor fd_v4_any = ASSERT_NO_ERRNO_AND_VALUE( + Socket(test_addr_v4_any.family(), param.type, 0)); + ASSERT_THAT(bind(fd_v4_any.get(), reinterpret_cast<sockaddr*>(&addr_v4_any), + test_addr_v4_any.addr_len), + SyscallFailsWithErrno(EADDRINUSE)); } TEST_P(SocketMultiProtocolInetLoopbackTest, V6OnlyV6AnyReservesV6) { @@ -1798,10 +1949,6 @@ TEST_P(SocketMultiProtocolInetLoopbackTest, V6OnlyV6AnyReservesV6) { TEST_P(SocketMultiProtocolInetLoopbackTest, V6EphemeralPortReserved) { auto const& param = GetParam(); - // FIXME(b/76031995): Support disabling SO_REUSEADDR for TCP sockets and make - // it disabled by default. - SKIP_IF(IsRunningOnGvisor() && param.type == SOCK_STREAM); - for (int i = 0; true; i++) { // Bind the v6 loopback on a dual stack socket. TestAddress const& test_addr = V6Loopback(); @@ -1864,17 +2011,6 @@ TEST_P(SocketMultiProtocolInetLoopbackTest, V6EphemeralPortReserved) { test_addr_v6.addr_len), SyscallFailsWithErrno(EADDRINUSE)); - // Verify that binding the v4 any with the same port fails. - TestAddress const& test_addr_v4_any = V4Any(); - sockaddr_storage addr_v4_any = test_addr_v4_any.addr; - ASSERT_NO_ERRNO( - SetAddrPort(test_addr_v4_any.family(), &addr_v4_any, ephemeral_port)); - const FileDescriptor fd_v4_any = ASSERT_NO_ERRNO_AND_VALUE( - Socket(test_addr_v4_any.family(), param.type, 0)); - ASSERT_THAT(bind(fd_v4_any.get(), reinterpret_cast<sockaddr*>(&addr_v4_any), - test_addr_v4_any.addr_len), - SyscallFailsWithErrno(EADDRINUSE)); - // Verify that we can still bind the v4 loopback on the same port. TestAddress const& test_addr_v4_mapped = V4MappedLoopback(); sockaddr_storage addr_v4_mapped = test_addr_v4_mapped.addr; @@ -1900,9 +2036,6 @@ TEST_P(SocketMultiProtocolInetLoopbackTest, V6EphemeralPortReserved) { TEST_P(SocketMultiProtocolInetLoopbackTest, V6EphemeralPortReservedReuseAddr) { auto const& param = GetParam(); - // FIXME(b/129164367): Support SO_REUSEADDR on UDP sockets. - SKIP_IF(IsRunningOnGvisor() && param.type == SOCK_DGRAM); - // Bind the v6 loopback on a dual stack socket. TestAddress const& test_addr = V6Loopback(); sockaddr_storage bound_addr = test_addr.addr; @@ -1966,10 +2099,6 @@ TEST_P(SocketMultiProtocolInetLoopbackTest, V6EphemeralPortReservedReuseAddr) { TEST_P(SocketMultiProtocolInetLoopbackTest, V4MappedEphemeralPortReserved) { auto const& param = GetParam(); - // FIXME(b/76031995): Support disabling SO_REUSEADDR for TCP sockets and make - // it disabled by default. - SKIP_IF(IsRunningOnGvisor() && param.type == SOCK_STREAM); - for (int i = 0; true; i++) { // Bind the v4 loopback on a dual stack socket. TestAddress const& test_addr = V4MappedLoopback(); @@ -2091,9 +2220,6 @@ TEST_P(SocketMultiProtocolInetLoopbackTest, V4MappedEphemeralPortReservedResueAddr) { auto const& param = GetParam(); - // FIXME(b/129164367): Support SO_REUSEADDR on UDP sockets. - SKIP_IF(IsRunningOnGvisor() && param.type == SOCK_DGRAM); - // Bind the v4 loopback on a dual stack socket. TestAddress const& test_addr = V4MappedLoopback(); sockaddr_storage bound_addr = test_addr.addr; @@ -2158,10 +2284,6 @@ TEST_P(SocketMultiProtocolInetLoopbackTest, TEST_P(SocketMultiProtocolInetLoopbackTest, V4EphemeralPortReserved) { auto const& param = GetParam(); - // FIXME(b/76031995): Support disabling SO_REUSEADDR for TCP sockets and make - // it disabled by default. - SKIP_IF(IsRunningOnGvisor() && param.type == SOCK_STREAM); - for (int i = 0; true; i++) { // Bind the v4 loopback on a v4 socket. TestAddress const& test_addr = V4Loopback(); @@ -2283,9 +2405,6 @@ TEST_P(SocketMultiProtocolInetLoopbackTest, V4EphemeralPortReserved) { TEST_P(SocketMultiProtocolInetLoopbackTest, V4EphemeralPortReservedReuseAddr) { auto const& param = GetParam(); - // FIXME(b/129164367): Support SO_REUSEADDR on UDP sockets. - SKIP_IF(IsRunningOnGvisor() && param.type == SOCK_DGRAM); - // Bind the v4 loopback on a v4 socket. TestAddress const& test_addr = V4Loopback(); sockaddr_storage bound_addr = test_addr.addr; diff --git a/test/syscalls/linux/socket_inet_loopback_nogotsan.cc b/test/syscalls/linux/socket_inet_loopback_nogotsan.cc new file mode 100644 index 000000000..791e2bd51 --- /dev/null +++ b/test/syscalls/linux/socket_inet_loopback_nogotsan.cc @@ -0,0 +1,174 @@ +// Copyright 2018 The gVisor Authors. +// +// Licensed under the Apache License, Version 2.0 (the "License"); +// you may not use this file except in compliance with the License. +// You may obtain a copy of the License at +// +// http://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, software +// distributed under the License is distributed on an "AS IS" BASIS, +// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +// See the License for the specific language governing permissions and +// limitations under the License. + +#include <arpa/inet.h> +#include <netinet/in.h> +#include <netinet/tcp.h> +#include <string.h> + +#include <iostream> +#include <memory> +#include <string> +#include <utility> +#include <vector> + +#include "gmock/gmock.h" +#include "gtest/gtest.h" +#include "absl/strings/str_cat.h" +#include "test/syscalls/linux/ip_socket_test_util.h" +#include "test/syscalls/linux/socket_test_util.h" +#include "test/util/file_descriptor.h" +#include "test/util/posix_error.h" +#include "test/util/save_util.h" +#include "test/util/test_util.h" + +namespace gvisor { +namespace testing { + +namespace { + +using ::testing::Gt; + +PosixErrorOr<uint16_t> AddrPort(int family, sockaddr_storage const& addr) { + switch (family) { + case AF_INET: + return static_cast<uint16_t>( + reinterpret_cast<sockaddr_in const*>(&addr)->sin_port); + case AF_INET6: + return static_cast<uint16_t>( + reinterpret_cast<sockaddr_in6 const*>(&addr)->sin6_port); + default: + return PosixError(EINVAL, + absl::StrCat("unknown socket family: ", family)); + } +} + +PosixError SetAddrPort(int family, sockaddr_storage* addr, uint16_t port) { + switch (family) { + case AF_INET: + reinterpret_cast<sockaddr_in*>(addr)->sin_port = port; + return NoError(); + case AF_INET6: + reinterpret_cast<sockaddr_in6*>(addr)->sin6_port = port; + return NoError(); + default: + return PosixError(EINVAL, + absl::StrCat("unknown socket family: ", family)); + } +} + +struct TestParam { + TestAddress listener; + TestAddress connector; +}; + +std::string DescribeTestParam(::testing::TestParamInfo<TestParam> const& info) { + return absl::StrCat("Listen", info.param.listener.description, "_Connect", + info.param.connector.description); +} + +using SocketInetLoopbackTest = ::testing::TestWithParam<TestParam>; + +// This test verifies that connect returns EADDRNOTAVAIL if all local ephemeral +// ports are already in use for a given destination ip/port. +// +// We disable S/R because this test creates a large number of sockets. +// +// FIXME(b/162475855): This test is failing reliably. +TEST_P(SocketInetLoopbackTest, DISABLED_TestTCPPortExhaustion_NoRandomSave) { + auto const& param = GetParam(); + TestAddress const& listener = param.listener; + TestAddress const& connector = param.connector; + + constexpr int kBacklog = 10; + constexpr int kClients = 65536; + + // Create the listening socket. + auto listen_fd = ASSERT_NO_ERRNO_AND_VALUE( + Socket(listener.family(), SOCK_STREAM, IPPROTO_TCP)); + sockaddr_storage listen_addr = listener.addr; + ASSERT_THAT(bind(listen_fd.get(), reinterpret_cast<sockaddr*>(&listen_addr), + listener.addr_len), + SyscallSucceeds()); + ASSERT_THAT(listen(listen_fd.get(), kBacklog), SyscallSucceeds()); + + // Get the port bound by the listening socket. + socklen_t addrlen = listener.addr_len; + ASSERT_THAT(getsockname(listen_fd.get(), + reinterpret_cast<sockaddr*>(&listen_addr), &addrlen), + SyscallSucceeds()); + uint16_t const port = + ASSERT_NO_ERRNO_AND_VALUE(AddrPort(listener.family(), listen_addr)); + + // Disable cooperative S/R as we are making too many syscalls. + DisableSave ds; + + // Now we keep opening connections till we run out of local ephemeral ports. + // and assert the error we get back. + sockaddr_storage conn_addr = connector.addr; + ASSERT_NO_ERRNO(SetAddrPort(connector.family(), &conn_addr, port)); + std::vector<FileDescriptor> clients; + std::vector<FileDescriptor> servers; + + for (int i = 0; i < kClients; i++) { + FileDescriptor client = ASSERT_NO_ERRNO_AND_VALUE( + Socket(connector.family(), SOCK_STREAM, IPPROTO_TCP)); + int ret = connect(client.get(), reinterpret_cast<sockaddr*>(&conn_addr), + connector.addr_len); + if (ret == 0) { + clients.push_back(std::move(client)); + FileDescriptor server = + ASSERT_NO_ERRNO_AND_VALUE(Accept(listen_fd.get(), nullptr, nullptr)); + servers.push_back(std::move(server)); + continue; + } + ASSERT_THAT(ret, SyscallFailsWithErrno(EADDRNOTAVAIL)); + break; + } +} + +INSTANTIATE_TEST_SUITE_P( + All, SocketInetLoopbackTest, + ::testing::Values( + // Listeners bound to IPv4 addresses refuse connections using IPv6 + // addresses. + TestParam{V4Any(), V4Any()}, TestParam{V4Any(), V4Loopback()}, + TestParam{V4Any(), V4MappedAny()}, + TestParam{V4Any(), V4MappedLoopback()}, + TestParam{V4Loopback(), V4Any()}, TestParam{V4Loopback(), V4Loopback()}, + TestParam{V4Loopback(), V4MappedLoopback()}, + TestParam{V4MappedAny(), V4Any()}, + TestParam{V4MappedAny(), V4Loopback()}, + TestParam{V4MappedAny(), V4MappedAny()}, + TestParam{V4MappedAny(), V4MappedLoopback()}, + TestParam{V4MappedLoopback(), V4Any()}, + TestParam{V4MappedLoopback(), V4Loopback()}, + TestParam{V4MappedLoopback(), V4MappedLoopback()}, + + // Listeners bound to IN6ADDR_ANY accept all connections. + TestParam{V6Any(), V4Any()}, TestParam{V6Any(), V4Loopback()}, + TestParam{V6Any(), V4MappedAny()}, + TestParam{V6Any(), V4MappedLoopback()}, TestParam{V6Any(), V6Any()}, + TestParam{V6Any(), V6Loopback()}, + + // Listeners bound to IN6ADDR_LOOPBACK refuse connections using IPv4 + // addresses. + TestParam{V6Loopback(), V6Any()}, + TestParam{V6Loopback(), V6Loopback()}), + DescribeTestParam); + +} // namespace + +} // namespace testing +} // namespace gvisor diff --git a/test/syscalls/linux/socket_ip_tcp_generic.cc b/test/syscalls/linux/socket_ip_tcp_generic.cc index fa81845fd..c2ecb639f 100644 --- a/test/syscalls/linux/socket_ip_tcp_generic.cc +++ b/test/syscalls/linux/socket_ip_tcp_generic.cc @@ -524,6 +524,7 @@ TEST_P(TCPSocketPairTest, SetTCPKeepintvlZero) { // Copied from include/net/tcp.h. constexpr int MAX_TCP_KEEPIDLE = 32767; constexpr int MAX_TCP_KEEPINTVL = 32767; +constexpr int MAX_TCP_KEEPCNT = 127; TEST_P(TCPSocketPairTest, SetTCPKeepidleAboveMax) { auto sockets = ASSERT_NO_ERRNO_AND_VALUE(NewSocketPair()); @@ -575,6 +576,78 @@ TEST_P(TCPSocketPairTest, SetTCPKeepintvlToMax) { EXPECT_EQ(get, MAX_TCP_KEEPINTVL); } +TEST_P(TCPSocketPairTest, TCPKeepcountDefault) { + auto sockets = ASSERT_NO_ERRNO_AND_VALUE(NewSocketPair()); + + int get = -1; + socklen_t get_len = sizeof(get); + EXPECT_THAT( + getsockopt(sockets->first_fd(), IPPROTO_TCP, TCP_KEEPCNT, &get, &get_len), + SyscallSucceedsWithValue(0)); + EXPECT_EQ(get_len, sizeof(get)); + EXPECT_EQ(get, 9); // 9 keepalive probes. +} + +TEST_P(TCPSocketPairTest, SetTCPKeepcountZero) { + auto sockets = ASSERT_NO_ERRNO_AND_VALUE(NewSocketPair()); + + constexpr int kZero = 0; + EXPECT_THAT(setsockopt(sockets->first_fd(), IPPROTO_TCP, TCP_KEEPCNT, &kZero, + sizeof(kZero)), + SyscallFailsWithErrno(EINVAL)); +} + +TEST_P(TCPSocketPairTest, SetTCPKeepcountAboveMax) { + auto sockets = ASSERT_NO_ERRNO_AND_VALUE(NewSocketPair()); + + constexpr int kAboveMax = MAX_TCP_KEEPCNT + 1; + EXPECT_THAT(setsockopt(sockets->first_fd(), IPPROTO_TCP, TCP_KEEPCNT, + &kAboveMax, sizeof(kAboveMax)), + SyscallFailsWithErrno(EINVAL)); +} + +TEST_P(TCPSocketPairTest, SetTCPKeepcountToMax) { + auto sockets = ASSERT_NO_ERRNO_AND_VALUE(NewSocketPair()); + + EXPECT_THAT(setsockopt(sockets->first_fd(), IPPROTO_TCP, TCP_KEEPCNT, + &MAX_TCP_KEEPCNT, sizeof(MAX_TCP_KEEPCNT)), + SyscallSucceedsWithValue(0)); + + int get = -1; + socklen_t get_len = sizeof(get); + EXPECT_THAT( + getsockopt(sockets->first_fd(), IPPROTO_TCP, TCP_KEEPCNT, &get, &get_len), + SyscallSucceedsWithValue(0)); + EXPECT_EQ(get_len, sizeof(get)); + EXPECT_EQ(get, MAX_TCP_KEEPCNT); +} + +TEST_P(TCPSocketPairTest, SetTCPKeepcountToOne) { + auto sockets = ASSERT_NO_ERRNO_AND_VALUE(NewSocketPair()); + + int keepaliveCount = 1; + EXPECT_THAT(setsockopt(sockets->first_fd(), IPPROTO_TCP, TCP_KEEPCNT, + &keepaliveCount, sizeof(keepaliveCount)), + SyscallSucceedsWithValue(0)); + + int get = -1; + socklen_t get_len = sizeof(get); + EXPECT_THAT( + getsockopt(sockets->first_fd(), IPPROTO_TCP, TCP_KEEPCNT, &get, &get_len), + SyscallSucceedsWithValue(0)); + EXPECT_EQ(get_len, sizeof(get)); + EXPECT_EQ(get, keepaliveCount); +} + +TEST_P(TCPSocketPairTest, SetTCPKeepcountToNegative) { + auto sockets = ASSERT_NO_ERRNO_AND_VALUE(NewSocketPair()); + + int keepaliveCount = -5; + EXPECT_THAT(setsockopt(sockets->first_fd(), IPPROTO_TCP, TCP_KEEPCNT, + &keepaliveCount, sizeof(keepaliveCount)), + SyscallFailsWithErrno(EINVAL)); +} + TEST_P(TCPSocketPairTest, SetOOBInline) { auto sockets = ASSERT_NO_ERRNO_AND_VALUE(NewSocketPair()); @@ -921,6 +994,30 @@ TEST_P(TCPSocketPairTest, SetTCPWindowClampBelowMinRcvBufConnectedSocket) { } } +TEST_P(TCPSocketPairTest, IpMulticastTtlDefault) { + auto sockets = ASSERT_NO_ERRNO_AND_VALUE(NewSocketPair()); + + int get = -1; + socklen_t get_len = sizeof(get); + EXPECT_THAT(getsockopt(sockets->first_fd(), IPPROTO_IP, IP_MULTICAST_TTL, + &get, &get_len), + SyscallSucceedsWithValue(0)); + EXPECT_EQ(get_len, sizeof(get)); + EXPECT_GT(get, 0); +} + +TEST_P(TCPSocketPairTest, IpMulticastLoopDefault) { + auto sockets = ASSERT_NO_ERRNO_AND_VALUE(NewSocketPair()); + + int get = -1; + socklen_t get_len = sizeof(get); + EXPECT_THAT(getsockopt(sockets->first_fd(), IPPROTO_IP, IP_MULTICAST_LOOP, + &get, &get_len), + SyscallSucceedsWithValue(0)); + EXPECT_EQ(get_len, sizeof(get)); + EXPECT_EQ(get, 1); +} + TEST_P(TCPSocketPairTest, TCPResetDuringClose_NoRandomSave) { DisableSave ds; // Too many syscalls. constexpr int kThreadCount = 1000; diff --git a/test/syscalls/linux/socket_ip_udp_generic.cc b/test/syscalls/linux/socket_ip_udp_generic.cc index 1c533fdf2..edb86aded 100644 --- a/test/syscalls/linux/socket_ip_udp_generic.cc +++ b/test/syscalls/linux/socket_ip_udp_generic.cc @@ -225,9 +225,6 @@ TEST_P(UDPSocketPairTest, ReuseAddrDefault) { TEST_P(UDPSocketPairTest, SetReuseAddr) { auto sockets = ASSERT_NO_ERRNO_AND_VALUE(NewSocketPair()); - // FIXME(b/129164367): Support SO_REUSEADDR on UDP sockets. - SKIP_IF(IsRunningOnGvisor()); - ASSERT_THAT(setsockopt(sockets->first_fd(), SOL_SOCKET, SO_REUSEADDR, &kSockOptOn, sizeof(kSockOptOn)), SyscallSucceeds()); @@ -292,9 +289,6 @@ TEST_P(UDPSocketPairTest, SetReusePort) { TEST_P(UDPSocketPairTest, SetReuseAddrReusePort) { auto sockets = ASSERT_NO_ERRNO_AND_VALUE(NewSocketPair()); - // FIXME(b/129164367): Support SO_REUSEADDR on UDP sockets. - SKIP_IF(IsRunningOnGvisor()); - ASSERT_THAT(setsockopt(sockets->first_fd(), SOL_SOCKET, SO_REUSEADDR, &kSockOptOn, sizeof(kSockOptOn)), SyscallSucceeds()); diff --git a/test/syscalls/linux/socket_ip_unbound.cc b/test/syscalls/linux/socket_ip_unbound.cc index ca597e267..1c7b0cf90 100644 --- a/test/syscalls/linux/socket_ip_unbound.cc +++ b/test/syscalls/linux/socket_ip_unbound.cc @@ -40,7 +40,7 @@ TEST_P(IPUnboundSocketTest, TtlDefault) { socklen_t get_sz = sizeof(get); EXPECT_THAT(getsockopt(socket->get(), IPPROTO_IP, IP_TTL, &get, &get_sz), SyscallSucceedsWithValue(0)); - EXPECT_EQ(get, 64); + EXPECT_TRUE(get == 64 || get == 127); EXPECT_EQ(get_sz, sizeof(get)); } @@ -157,7 +157,7 @@ TEST_P(IPUnboundSocketTest, TOSDefault) { int get = -1; socklen_t get_sz = sizeof(get); constexpr int kDefaultTOS = 0; - EXPECT_THAT(getsockopt(socket->get(), t.level, t.option, &get, &get_sz), + ASSERT_THAT(getsockopt(socket->get(), t.level, t.option, &get, &get_sz), SyscallSucceedsWithValue(0)); EXPECT_EQ(get_sz, sizeof(get)); EXPECT_EQ(get, kDefaultTOS); @@ -173,7 +173,7 @@ TEST_P(IPUnboundSocketTest, SetTOS) { int get = -1; socklen_t get_sz = sizeof(get); - EXPECT_THAT(getsockopt(socket->get(), t.level, t.option, &get, &get_sz), + ASSERT_THAT(getsockopt(socket->get(), t.level, t.option, &get, &get_sz), SyscallSucceedsWithValue(0)); EXPECT_EQ(get_sz, sizeof(get)); EXPECT_EQ(get, set); @@ -188,7 +188,7 @@ TEST_P(IPUnboundSocketTest, ZeroTOS) { SyscallSucceedsWithValue(0)); int get = -1; socklen_t get_sz = sizeof(get); - EXPECT_THAT(getsockopt(socket->get(), t.level, t.option, &get, &get_sz), + ASSERT_THAT(getsockopt(socket->get(), t.level, t.option, &get, &get_sz), SyscallSucceedsWithValue(0)); EXPECT_EQ(get_sz, sizeof(get)); EXPECT_EQ(get, set); @@ -210,7 +210,7 @@ TEST_P(IPUnboundSocketTest, InvalidLargeTOS) { } int get = -1; socklen_t get_sz = sizeof(get); - EXPECT_THAT(getsockopt(socket->get(), t.level, t.option, &get, &get_sz), + ASSERT_THAT(getsockopt(socket->get(), t.level, t.option, &get, &get_sz), SyscallSucceedsWithValue(0)); EXPECT_EQ(get_sz, sizeof(get)); EXPECT_EQ(get, kDefaultTOS); @@ -229,7 +229,7 @@ TEST_P(IPUnboundSocketTest, CheckSkipECN) { } int get = -1; socklen_t get_sz = sizeof(get); - EXPECT_THAT(getsockopt(socket->get(), t.level, t.option, &get, &get_sz), + ASSERT_THAT(getsockopt(socket->get(), t.level, t.option, &get, &get_sz), SyscallSucceedsWithValue(0)); EXPECT_EQ(get_sz, sizeof(get)); EXPECT_EQ(get, expect); @@ -249,7 +249,7 @@ TEST_P(IPUnboundSocketTest, ZeroTOSOptionSize) { } int get = -1; socklen_t get_sz = 0; - EXPECT_THAT(getsockopt(socket->get(), t.level, t.option, &get, &get_sz), + ASSERT_THAT(getsockopt(socket->get(), t.level, t.option, &get, &get_sz), SyscallSucceedsWithValue(0)); EXPECT_EQ(get_sz, 0); EXPECT_EQ(get, -1); @@ -276,7 +276,7 @@ TEST_P(IPUnboundSocketTest, SmallTOSOptionSize) { } uint get = -1; socklen_t get_sz = i; - EXPECT_THAT(getsockopt(socket->get(), t.level, t.option, &get, &get_sz), + ASSERT_THAT(getsockopt(socket->get(), t.level, t.option, &get, &get_sz), SyscallSucceedsWithValue(0)); EXPECT_EQ(get_sz, expect_sz); // Account for partial copies by getsockopt, retrieve the lower @@ -297,7 +297,7 @@ TEST_P(IPUnboundSocketTest, LargeTOSOptionSize) { // We expect the system call handler to only copy atmost sizeof(int) bytes // as asserted by the check below. Hence, we do not expect the copy to // overflow in getsockopt. - EXPECT_THAT(getsockopt(socket->get(), t.level, t.option, &get, &get_sz), + ASSERT_THAT(getsockopt(socket->get(), t.level, t.option, &get, &get_sz), SyscallSucceedsWithValue(0)); EXPECT_EQ(get_sz, sizeof(int)); EXPECT_EQ(get, set); @@ -325,7 +325,7 @@ TEST_P(IPUnboundSocketTest, NegativeTOS) { } int get = -1; socklen_t get_sz = sizeof(get); - EXPECT_THAT(getsockopt(socket->get(), t.level, t.option, &get, &get_sz), + ASSERT_THAT(getsockopt(socket->get(), t.level, t.option, &get, &get_sz), SyscallSucceedsWithValue(0)); EXPECT_EQ(get_sz, sizeof(get)); EXPECT_EQ(get, expect); @@ -338,20 +338,20 @@ TEST_P(IPUnboundSocketTest, InvalidNegativeTOS) { TOSOption t = GetTOSOption(GetParam().domain); int expect; if (GetParam().domain == AF_INET) { - EXPECT_THAT(setsockopt(socket->get(), t.level, t.option, &set, set_sz), + ASSERT_THAT(setsockopt(socket->get(), t.level, t.option, &set, set_sz), SyscallSucceedsWithValue(0)); expect = static_cast<uint8_t>(set); if (GetParam().protocol == IPPROTO_TCP) { expect &= ~INET_ECN_MASK; } } else { - EXPECT_THAT(setsockopt(socket->get(), t.level, t.option, &set, set_sz), + ASSERT_THAT(setsockopt(socket->get(), t.level, t.option, &set, set_sz), SyscallFailsWithErrno(EINVAL)); expect = 0; } int get = 0; socklen_t get_sz = sizeof(get); - EXPECT_THAT(getsockopt(socket->get(), t.level, t.option, &get, &get_sz), + ASSERT_THAT(getsockopt(socket->get(), t.level, t.option, &get, &get_sz), SyscallSucceedsWithValue(0)); EXPECT_EQ(get_sz, sizeof(get)); EXPECT_EQ(get, expect); @@ -377,8 +377,11 @@ TEST_P(IPUnboundSocketTest, NullTOS) { // // Linux's implementation would need fixing as passing a nullptr as optval // and non-zero optlen may not be valid. - EXPECT_THAT(setsockopt(socket->get(), t.level, t.option, nullptr, set_sz), - SyscallSucceedsWithValue(0)); + // TODO(b/158666797): Combine the gVisor and linux cases for IPv6. + // Some kernel versions return EFAULT, so we handle both. + EXPECT_THAT( + setsockopt(socket->get(), t.level, t.option, nullptr, set_sz), + AnyOf(SyscallFailsWithErrno(EFAULT), SyscallSucceedsWithValue(0))); } } socklen_t get_sz = sizeof(int); @@ -419,6 +422,34 @@ TEST_P(IPUnboundSocketTest, InsufficientBufferTOS) { EXPECT_THAT(sendmsg(socket->get(), &msg, 0), SyscallFailsWithErrno(EINVAL)); } +TEST_P(IPUnboundSocketTest, ReuseAddrDefault) { + auto socket = ASSERT_NO_ERRNO_AND_VALUE(NewSocket()); + + int get = -1; + socklen_t get_sz = sizeof(get); + ASSERT_THAT( + getsockopt(socket->get(), SOL_SOCKET, SO_REUSEADDR, &get, &get_sz), + SyscallSucceedsWithValue(0)); + EXPECT_EQ(get, kSockOptOff); + EXPECT_EQ(get_sz, sizeof(get)); +} + +TEST_P(IPUnboundSocketTest, SetReuseAddr) { + auto socket = ASSERT_NO_ERRNO_AND_VALUE(NewSocket()); + + ASSERT_THAT(setsockopt(socket->get(), SOL_SOCKET, SO_REUSEADDR, &kSockOptOn, + sizeof(kSockOptOn)), + SyscallSucceedsWithValue(0)); + + int get = -1; + socklen_t get_sz = sizeof(get); + ASSERT_THAT( + getsockopt(socket->get(), SOL_SOCKET, SO_REUSEADDR, &get, &get_sz), + SyscallSucceedsWithValue(0)); + EXPECT_EQ(get, kSockOptOn); + EXPECT_EQ(get_sz, sizeof(get)); +} + INSTANTIATE_TEST_SUITE_P( IPUnboundSockets, IPUnboundSocketTest, ::testing::ValuesIn(VecCat<SocketKind>(VecCat<SocketKind>( diff --git a/test/syscalls/linux/socket_ipv4_udp_unbound.cc b/test/syscalls/linux/socket_ipv4_udp_unbound.cc index bc4b07a62..de0f5f01b 100644 --- a/test/syscalls/linux/socket_ipv4_udp_unbound.cc +++ b/test/syscalls/linux/socket_ipv4_udp_unbound.cc @@ -18,6 +18,7 @@ #include <net/if.h> #include <sys/ioctl.h> #include <sys/socket.h> +#include <sys/types.h> #include <sys/un.h> #include <cstdio> @@ -1672,10 +1673,10 @@ TEST_P(IPv4UDPUnboundSocketTest, TestBindToBcastThenSend) { } // Check that SO_REUSEADDR always delivers to the most recently bound socket. -TEST_P(IPv4UDPUnboundSocketTest, ReuseAddrDistribution) { - // FIXME(b/129164367): Support SO_REUSEADDR on UDP sockets. - SKIP_IF(IsRunningOnGvisor()); - +// +// FIXME(gvisor.dev/issue/873): Endpoint order is not restored correctly. Enable +// random and co-op save (below) once that is fixed. +TEST_P(IPv4UDPUnboundSocketTest, ReuseAddrDistribution_NoRandomSave) { std::vector<std::unique_ptr<FileDescriptor>> sockets; sockets.emplace_back(ASSERT_NO_ERRNO_AND_VALUE(NewSocket())); @@ -1696,6 +1697,9 @@ TEST_P(IPv4UDPUnboundSocketTest, ReuseAddrDistribution) { constexpr int kMessageSize = 200; + // FIXME(gvisor.dev/issue/873): Endpoint order is not restored correctly. + const DisableSave ds; + for (int i = 0; i < 10; i++) { // Add a new receiver. sockets.emplace_back(ASSERT_NO_ERRNO_AND_VALUE(NewSocket())); @@ -1836,9 +1840,6 @@ TEST_P(IPv4UDPUnboundSocketTest, BindReuseAddrReusePortConvertibleToReusePort) { } TEST_P(IPv4UDPUnboundSocketTest, BindReuseAddrReusePortConvertibleToReuseAddr) { - // FIXME(b/129164367): Support SO_REUSEADDR on UDP sockets. - SKIP_IF(IsRunningOnGvisor()); - auto socket1 = ASSERT_NO_ERRNO_AND_VALUE(NewSocket()); auto socket2 = ASSERT_NO_ERRNO_AND_VALUE(NewSocket()); auto socket3 = ASSERT_NO_ERRNO_AND_VALUE(NewSocket()); @@ -1880,9 +1881,6 @@ TEST_P(IPv4UDPUnboundSocketTest, BindReuseAddrReusePortConvertibleToReuseAddr) { } TEST_P(IPv4UDPUnboundSocketTest, BindReuseAddrReusePortConversionReversable1) { - // FIXME(b/129164367): Support SO_REUSEADDR on UDP sockets. - SKIP_IF(IsRunningOnGvisor()); - auto socket1 = ASSERT_NO_ERRNO_AND_VALUE(NewSocket()); auto socket2 = ASSERT_NO_ERRNO_AND_VALUE(NewSocket()); auto socket3 = ASSERT_NO_ERRNO_AND_VALUE(NewSocket()); @@ -1927,9 +1925,6 @@ TEST_P(IPv4UDPUnboundSocketTest, BindReuseAddrReusePortConversionReversable1) { } TEST_P(IPv4UDPUnboundSocketTest, BindReuseAddrReusePortConversionReversable2) { - // FIXME(b/129164367): Support SO_REUSEADDR on UDP sockets. - SKIP_IF(IsRunningOnGvisor()); - auto socket1 = ASSERT_NO_ERRNO_AND_VALUE(NewSocket()); auto socket2 = ASSERT_NO_ERRNO_AND_VALUE(NewSocket()); auto socket3 = ASSERT_NO_ERRNO_AND_VALUE(NewSocket()); @@ -2020,9 +2015,6 @@ TEST_P(IPv4UDPUnboundSocketTest, BindDoubleReuseAddrReusePortThenReusePort) { } TEST_P(IPv4UDPUnboundSocketTest, BindDoubleReuseAddrReusePortThenReuseAddr) { - // FIXME(b/129164367): Support SO_REUSEADDR on UDP sockets. - SKIP_IF(IsRunningOnGvisor()); - auto socket1 = ASSERT_NO_ERRNO_AND_VALUE(NewSocket()); auto socket2 = ASSERT_NO_ERRNO_AND_VALUE(NewSocket()); auto socket3 = ASSERT_NO_ERRNO_AND_VALUE(NewSocket()); @@ -2129,6 +2121,39 @@ TEST_P(IPv4UDPUnboundSocketTest, ReuseAddrReusePortDistribution) { SyscallSucceedsWithValue(kMessageSize)); } +// Check that connect returns EADDRNOTAVAIL when out of local ephemeral ports. +// We disable S/R because this test creates a large number of sockets. +TEST_P(IPv4UDPUnboundSocketTest, UDPConnectPortExhaustion_NoRandomSave) { + auto receiver1 = ASSERT_NO_ERRNO_AND_VALUE(NewSocket()); + constexpr int kClients = 65536; + // Bind the first socket to the loopback and take note of the selected port. + auto addr = V4Loopback(); + ASSERT_THAT(bind(receiver1->get(), reinterpret_cast<sockaddr*>(&addr.addr), + addr.addr_len), + SyscallSucceeds()); + socklen_t addr_len = addr.addr_len; + ASSERT_THAT(getsockname(receiver1->get(), + reinterpret_cast<sockaddr*>(&addr.addr), &addr_len), + SyscallSucceeds()); + EXPECT_EQ(addr_len, addr.addr_len); + + // Disable cooperative S/R as we are making too many syscalls. + DisableSave ds; + std::vector<std::unique_ptr<FileDescriptor>> sockets; + for (int i = 0; i < kClients; i++) { + auto s = ASSERT_NO_ERRNO_AND_VALUE(NewSocket()); + + int ret = connect(s->get(), reinterpret_cast<sockaddr*>(&addr.addr), + addr.addr_len); + if (ret == 0) { + sockets.push_back(std::move(s)); + continue; + } + ASSERT_THAT(ret, SyscallFailsWithErrno(EAGAIN)); + break; + } +} + // Test that socket will receive packet info control message. TEST_P(IPv4UDPUnboundSocketTest, SetAndReceiveIPPKTINFO) { // TODO(gvisor.dev/issue/1202): ioctl() is not supported by hostinet. @@ -2212,5 +2237,220 @@ TEST_P(IPv4UDPUnboundSocketTest, SetAndReceiveIPPKTINFO) { EXPECT_EQ(received_pktinfo.ipi_spec_dst.s_addr, htonl(INADDR_LOOPBACK)); EXPECT_EQ(received_pktinfo.ipi_addr.s_addr, htonl(INADDR_LOOPBACK)); } + +// Check that setting SO_RCVBUF below min is clamped to the minimum +// receive buffer size. +TEST_P(IPv4UDPUnboundSocketTest, SetSocketRecvBufBelowMin) { + auto s = ASSERT_NO_ERRNO_AND_VALUE(NewSocket()); + + // Discover minimum buffer size by setting it to zero. + constexpr int kRcvBufSz = 0; + ASSERT_THAT(setsockopt(s->get(), SOL_SOCKET, SO_RCVBUF, &kRcvBufSz, + sizeof(kRcvBufSz)), + SyscallSucceeds()); + + int min = 0; + socklen_t min_len = sizeof(min); + ASSERT_THAT(getsockopt(s->get(), SOL_SOCKET, SO_RCVBUF, &min, &min_len), + SyscallSucceeds()); + + // Linux doubles the value so let's use a value that when doubled will still + // be smaller than min. + int below_min = min / 2 - 1; + ASSERT_THAT(setsockopt(s->get(), SOL_SOCKET, SO_RCVBUF, &below_min, + sizeof(below_min)), + SyscallSucceeds()); + + int val = 0; + socklen_t val_len = sizeof(val); + ASSERT_THAT(getsockopt(s->get(), SOL_SOCKET, SO_RCVBUF, &val, &val_len), + SyscallSucceeds()); + + ASSERT_EQ(min, val); +} + +// Check that setting SO_RCVBUF above max is clamped to the maximum +// receive buffer size. +TEST_P(IPv4UDPUnboundSocketTest, SetSocketRecvBufAboveMax) { + auto s = ASSERT_NO_ERRNO_AND_VALUE(NewSocket()); + + // Discover maxmimum buffer size by setting to a really large value. + constexpr int kRcvBufSz = 0xffffffff; + ASSERT_THAT(setsockopt(s->get(), SOL_SOCKET, SO_RCVBUF, &kRcvBufSz, + sizeof(kRcvBufSz)), + SyscallSucceeds()); + + int max = 0; + socklen_t max_len = sizeof(max); + ASSERT_THAT(getsockopt(s->get(), SOL_SOCKET, SO_RCVBUF, &max, &max_len), + SyscallSucceeds()); + + int above_max = max + 1; + ASSERT_THAT(setsockopt(s->get(), SOL_SOCKET, SO_RCVBUF, &above_max, + sizeof(above_max)), + SyscallSucceeds()); + + int val = 0; + socklen_t val_len = sizeof(val); + ASSERT_THAT(getsockopt(s->get(), SOL_SOCKET, SO_RCVBUF, &val, &val_len), + SyscallSucceeds()); + ASSERT_EQ(max, val); +} + +// Check that setting SO_RCVBUF min <= rcvBufSz <= max is honored. +TEST_P(IPv4UDPUnboundSocketTest, SetSocketRecvBuf) { + auto s = ASSERT_NO_ERRNO_AND_VALUE(NewSocket()); + + int max = 0; + int min = 0; + { + // Discover maxmimum buffer size by setting to a really large value. + constexpr int kRcvBufSz = 0xffffffff; + ASSERT_THAT(setsockopt(s->get(), SOL_SOCKET, SO_RCVBUF, &kRcvBufSz, + sizeof(kRcvBufSz)), + SyscallSucceeds()); + + max = 0; + socklen_t max_len = sizeof(max); + ASSERT_THAT(getsockopt(s->get(), SOL_SOCKET, SO_RCVBUF, &max, &max_len), + SyscallSucceeds()); + } + + { + // Discover minimum buffer size by setting it to zero. + constexpr int kRcvBufSz = 0; + ASSERT_THAT(setsockopt(s->get(), SOL_SOCKET, SO_RCVBUF, &kRcvBufSz, + sizeof(kRcvBufSz)), + SyscallSucceeds()); + + socklen_t min_len = sizeof(min); + ASSERT_THAT(getsockopt(s->get(), SOL_SOCKET, SO_RCVBUF, &min, &min_len), + SyscallSucceeds()); + } + + int quarter_sz = min + (max - min) / 4; + ASSERT_THAT(setsockopt(s->get(), SOL_SOCKET, SO_RCVBUF, &quarter_sz, + sizeof(quarter_sz)), + SyscallSucceeds()); + + int val = 0; + socklen_t val_len = sizeof(val); + ASSERT_THAT(getsockopt(s->get(), SOL_SOCKET, SO_RCVBUF, &val, &val_len), + SyscallSucceeds()); + + // Linux doubles the value set by SO_SNDBUF/SO_RCVBUF. + if (!IsRunningOnGvisor()) { + quarter_sz *= 2; + } + ASSERT_EQ(quarter_sz, val); +} + +// Check that setting SO_SNDBUF below min is clamped to the minimum +// send buffer size. +TEST_P(IPv4UDPUnboundSocketTest, SetSocketSendBufBelowMin) { + auto s = ASSERT_NO_ERRNO_AND_VALUE(NewSocket()); + + // Discover minimum buffer size by setting it to zero. + constexpr int kSndBufSz = 0; + ASSERT_THAT(setsockopt(s->get(), SOL_SOCKET, SO_SNDBUF, &kSndBufSz, + sizeof(kSndBufSz)), + SyscallSucceeds()); + + int min = 0; + socklen_t min_len = sizeof(min); + ASSERT_THAT(getsockopt(s->get(), SOL_SOCKET, SO_SNDBUF, &min, &min_len), + SyscallSucceeds()); + + // Linux doubles the value so let's use a value that when doubled will still + // be smaller than min. + int below_min = min / 2 - 1; + ASSERT_THAT(setsockopt(s->get(), SOL_SOCKET, SO_SNDBUF, &below_min, + sizeof(below_min)), + SyscallSucceeds()); + + int val = 0; + socklen_t val_len = sizeof(val); + ASSERT_THAT(getsockopt(s->get(), SOL_SOCKET, SO_SNDBUF, &val, &val_len), + SyscallSucceeds()); + + ASSERT_EQ(min, val); +} + +// Check that setting SO_SNDBUF above max is clamped to the maximum +// send buffer size. +TEST_P(IPv4UDPUnboundSocketTest, SetSocketSendBufAboveMax) { + auto s = ASSERT_NO_ERRNO_AND_VALUE(NewSocket()); + + // Discover maxmimum buffer size by setting to a really large value. + constexpr int kSndBufSz = 0xffffffff; + ASSERT_THAT(setsockopt(s->get(), SOL_SOCKET, SO_SNDBUF, &kSndBufSz, + sizeof(kSndBufSz)), + SyscallSucceeds()); + + int max = 0; + socklen_t max_len = sizeof(max); + ASSERT_THAT(getsockopt(s->get(), SOL_SOCKET, SO_SNDBUF, &max, &max_len), + SyscallSucceeds()); + + int above_max = max + 1; + ASSERT_THAT(setsockopt(s->get(), SOL_SOCKET, SO_SNDBUF, &above_max, + sizeof(above_max)), + SyscallSucceeds()); + + int val = 0; + socklen_t val_len = sizeof(val); + ASSERT_THAT(getsockopt(s->get(), SOL_SOCKET, SO_SNDBUF, &val, &val_len), + SyscallSucceeds()); + ASSERT_EQ(max, val); +} + +// Check that setting SO_SNDBUF min <= kSndBufSz <= max is honored. +TEST_P(IPv4UDPUnboundSocketTest, SetSocketSendBuf) { + auto s = ASSERT_NO_ERRNO_AND_VALUE(NewSocket()); + + int max = 0; + int min = 0; + { + // Discover maxmimum buffer size by setting to a really large value. + constexpr int kSndBufSz = 0xffffffff; + ASSERT_THAT(setsockopt(s->get(), SOL_SOCKET, SO_SNDBUF, &kSndBufSz, + sizeof(kSndBufSz)), + SyscallSucceeds()); + + max = 0; + socklen_t max_len = sizeof(max); + ASSERT_THAT(getsockopt(s->get(), SOL_SOCKET, SO_SNDBUF, &max, &max_len), + SyscallSucceeds()); + } + + { + // Discover minimum buffer size by setting it to zero. + constexpr int kSndBufSz = 0; + ASSERT_THAT(setsockopt(s->get(), SOL_SOCKET, SO_SNDBUF, &kSndBufSz, + sizeof(kSndBufSz)), + SyscallSucceeds()); + + socklen_t min_len = sizeof(min); + ASSERT_THAT(getsockopt(s->get(), SOL_SOCKET, SO_SNDBUF, &min, &min_len), + SyscallSucceeds()); + } + + int quarter_sz = min + (max - min) / 4; + ASSERT_THAT(setsockopt(s->get(), SOL_SOCKET, SO_SNDBUF, &quarter_sz, + sizeof(quarter_sz)), + SyscallSucceeds()); + + int val = 0; + socklen_t val_len = sizeof(val); + ASSERT_THAT(getsockopt(s->get(), SOL_SOCKET, SO_SNDBUF, &val, &val_len), + SyscallSucceeds()); + + // Linux doubles the value set by SO_SNDBUF/SO_RCVBUF. + if (!IsRunningOnGvisor()) { + quarter_sz *= 2; + } + + ASSERT_EQ(quarter_sz, val); +} } // namespace testing } // namespace gvisor diff --git a/test/syscalls/linux/socket_netdevice.cc b/test/syscalls/linux/socket_netdevice.cc index 15d4b85a7..5f8d7f981 100644 --- a/test/syscalls/linux/socket_netdevice.cc +++ b/test/syscalls/linux/socket_netdevice.cc @@ -12,6 +12,7 @@ // See the License for the specific language governing permissions and // limitations under the License. +#include <linux/ethtool.h> #include <linux/netlink.h> #include <linux/rtnetlink.h> #include <linux/sockios.h> @@ -49,6 +50,7 @@ TEST(NetdeviceTest, Loopback) { // Check that the loopback is zero hardware address. ASSERT_THAT(ioctl(sock.get(), SIOCGIFHWADDR, &ifr), SyscallSucceeds()); + EXPECT_EQ(ifr.ifr_hwaddr.sa_family, ARPHRD_LOOPBACK); EXPECT_EQ(ifr.ifr_hwaddr.sa_data[0], 0); EXPECT_EQ(ifr.ifr_hwaddr.sa_data[1], 0); EXPECT_EQ(ifr.ifr_hwaddr.sa_data[2], 0); @@ -178,6 +180,27 @@ TEST(NetdeviceTest, InterfaceMTU) { EXPECT_GT(ifr.ifr_mtu, 0); } +TEST(NetdeviceTest, EthtoolGetTSInfo) { + FileDescriptor sock = + ASSERT_NO_ERRNO_AND_VALUE(Socket(AF_INET, SOCK_DGRAM, 0)); + + struct ethtool_ts_info tsi = {}; + tsi.cmd = ETHTOOL_GET_TS_INFO; // Get NIC's Timestamping capabilities. + + // Prepare the request. + struct ifreq ifr = {}; + snprintf(ifr.ifr_name, IFNAMSIZ, "lo"); + ifr.ifr_data = (void*)&tsi; + + // Check that SIOCGIFMTU returns a nonzero MTU. + if (IsRunningOnGvisor()) { + ASSERT_THAT(ioctl(sock.get(), SIOCETHTOOL, &ifr), + SyscallFailsWithErrno(EOPNOTSUPP)); + return; + } + ASSERT_THAT(ioctl(sock.get(), SIOCETHTOOL, &ifr), SyscallSucceeds()); +} + } // namespace } // namespace testing diff --git a/test/syscalls/linux/socket_netlink_route.cc b/test/syscalls/linux/socket_netlink_route.cc index fbe61c5a0..b3fcf8e7c 100644 --- a/test/syscalls/linux/socket_netlink_route.cc +++ b/test/syscalls/linux/socket_netlink_route.cc @@ -577,7 +577,10 @@ TEST(NetlinkRouteTest, GetRouteDump) { std::cout << std::endl; - if (msg->rtm_table == RT_TABLE_MAIN) { + // If the test is running in a new network namespace, it will have only + // the local route table. + if (msg->rtm_table == RT_TABLE_MAIN || + (!IsRunningOnGvisor() && msg->rtm_table == RT_TABLE_LOCAL)) { routeFound = true; dstFound = rtDstFound && dstFound; } @@ -595,7 +598,7 @@ TEST(NetlinkRouteTest, GetRouteRequest) { ASSERT_NO_ERRNO_AND_VALUE(NetlinkBoundSocket(NETLINK_ROUTE)); uint32_t port = ASSERT_NO_ERRNO_AND_VALUE(NetlinkPortID(fd.get())); - struct __attribute__((__packed__)) request { + struct request { struct nlmsghdr hdr; struct rtmsg rtm; struct nlattr nla; diff --git a/test/syscalls/linux/socket_unix.cc b/test/syscalls/linux/socket_unix.cc index 8bf663e8b..591cab3fd 100644 --- a/test/syscalls/linux/socket_unix.cc +++ b/test/syscalls/linux/socket_unix.cc @@ -256,10 +256,9 @@ TEST_P(UnixSocketPairTest, ShutdownWrite) { } TEST_P(UnixSocketPairTest, SocketReopenFromProcfs) { - // TODO(b/122310852): We should be returning ENXIO and NOT EIO. - // TODO(github.dev/issue/1624): This should be resolved in VFS2. Verify - // that this is the case and delete the SKIP_IF once we delete VFS1. - SKIP_IF(IsRunningOnGvisor()); + // TODO(gvisor.dev/issue/1624): In VFS1, we return EIO instead of ENXIO (see + // b/122310852). Remove this skip once VFS1 is deleted. + SKIP_IF(IsRunningWithVFS1()); auto sockets = ASSERT_NO_ERRNO_AND_VALUE(NewSocketPair()); // Opening a socket pair via /proc/self/fd/X is a ENXIO. diff --git a/test/syscalls/linux/socket_unix_seqpacket.cc b/test/syscalls/linux/socket_unix_seqpacket.cc index 84d3a569e..6d03df4d9 100644 --- a/test/syscalls/linux/socket_unix_seqpacket.cc +++ b/test/syscalls/linux/socket_unix_seqpacket.cc @@ -43,6 +43,24 @@ TEST_P(SeqpacketUnixSocketPairTest, ReadOneSideClosed) { SyscallSucceedsWithValue(0)); } +TEST_P(SeqpacketUnixSocketPairTest, Sendto) { + auto sockets = ASSERT_NO_ERRNO_AND_VALUE(NewSocketPair()); + + struct sockaddr_un addr = {}; + addr.sun_family = AF_UNIX; + constexpr char kPath[] = "\0nonexistent"; + memcpy(addr.sun_path, kPath, sizeof(kPath)); + + constexpr char kStr[] = "abc"; + ASSERT_THAT(sendto(sockets->second_fd(), kStr, 3, 0, (struct sockaddr*)&addr, + sizeof(addr)), + SyscallSucceedsWithValue(3)); + + char data[10] = {}; + ASSERT_THAT(read(sockets->first_fd(), data, sizeof(data)), + SyscallSucceedsWithValue(3)); +} + } // namespace } // namespace testing diff --git a/test/syscalls/linux/socket_unix_stream.cc b/test/syscalls/linux/socket_unix_stream.cc index 563467365..99e77b89e 100644 --- a/test/syscalls/linux/socket_unix_stream.cc +++ b/test/syscalls/linux/socket_unix_stream.cc @@ -89,6 +89,20 @@ TEST_P(StreamUnixSocketPairTest, ReadOneSideClosedWithUnreadData) { SyscallFailsWithErrno(ECONNRESET)); } +TEST_P(StreamUnixSocketPairTest, Sendto) { + auto sockets = ASSERT_NO_ERRNO_AND_VALUE(NewSocketPair()); + + struct sockaddr_un addr = {}; + addr.sun_family = AF_UNIX; + constexpr char kPath[] = "\0nonexistent"; + memcpy(addr.sun_path, kPath, sizeof(kPath)); + + constexpr char kStr[] = "abc"; + ASSERT_THAT(sendto(sockets->second_fd(), kStr, 3, 0, (struct sockaddr*)&addr, + sizeof(addr)), + SyscallFailsWithErrno(EISCONN)); +} + INSTANTIATE_TEST_SUITE_P( AllUnixDomainSockets, StreamUnixSocketPairTest, ::testing::ValuesIn(IncludeReversals(VecCat<SocketPairKind>( diff --git a/test/syscalls/linux/splice.cc b/test/syscalls/linux/splice.cc index f103e2e56..08fc4b1b7 100644 --- a/test/syscalls/linux/splice.cc +++ b/test/syscalls/linux/splice.cc @@ -430,6 +430,55 @@ TEST(SpliceTest, TwoPipes) { EXPECT_EQ(memcmp(rbuf.data(), buf.data(), kPageSize), 0); } +TEST(SpliceTest, TwoPipesCircular) { + // This test deadlocks the sentry on VFS1 because VFS1 splice ordering is + // based on fs.File.UniqueID, which does not prevent circular ordering between + // e.g. inode-level locks taken by fs.FileOperations. + SKIP_IF(IsRunningWithVFS1()); + + // Create two pipes. + int fds[2]; + ASSERT_THAT(pipe(fds), SyscallSucceeds()); + const FileDescriptor first_rfd(fds[0]); + const FileDescriptor first_wfd(fds[1]); + ASSERT_THAT(pipe(fds), SyscallSucceeds()); + const FileDescriptor second_rfd(fds[0]); + const FileDescriptor second_wfd(fds[1]); + + // On Linux, each pipe is normally limited to + // include/linux/pipe_fs_i.h:PIPE_DEF_BUFFERS buffers worth of data. + constexpr size_t PIPE_DEF_BUFFERS = 16; + + // Write some data to each pipe. Below we splice 1 byte at a time between + // pipes, which very quickly causes each byte to be stored in a separate + // buffer, so we must ensure that the total amount of data in the system is <= + // PIPE_DEF_BUFFERS bytes. + std::vector<char> buf(PIPE_DEF_BUFFERS / 2); + RandomizeBuffer(buf.data(), buf.size()); + ASSERT_THAT(write(first_wfd.get(), buf.data(), buf.size()), + SyscallSucceedsWithValue(buf.size())); + ASSERT_THAT(write(second_wfd.get(), buf.data(), buf.size()), + SyscallSucceedsWithValue(buf.size())); + + // Have another thread splice from the second pipe to the first, while we + // splice from the first to the second. The test passes if this does not + // deadlock. + const int kIterations = 1000; + DisableSave ds; + ScopedThread t([&]() { + for (int i = 0; i < kIterations; i++) { + ASSERT_THAT( + splice(second_rfd.get(), nullptr, first_wfd.get(), nullptr, 1, 0), + SyscallSucceedsWithValue(1)); + } + }); + for (int i = 0; i < kIterations; i++) { + ASSERT_THAT( + splice(first_rfd.get(), nullptr, second_wfd.get(), nullptr, 1, 0), + SyscallSucceedsWithValue(1)); + } +} + TEST(SpliceTest, Blocking) { // Create two new pipes. int first[2], second[2]; diff --git a/test/syscalls/linux/sticky.cc b/test/syscalls/linux/sticky.cc index 92eec0449..4afed6d08 100644 --- a/test/syscalls/linux/sticky.cc +++ b/test/syscalls/linux/sticky.cc @@ -40,11 +40,17 @@ namespace { TEST(StickyTest, StickyBitPermDenied) { SKIP_IF(!ASSERT_NO_ERRNO_AND_VALUE(HaveCapability(CAP_SETUID))); - auto dir = ASSERT_NO_ERRNO_AND_VALUE(TempPath::CreateDir()); - EXPECT_THAT(chmod(dir.path().c_str(), 0777 | S_ISVTX), SyscallSucceeds()); - const FileDescriptor dirfd = - ASSERT_NO_ERRNO_AND_VALUE(Open(dir.path(), O_DIRECTORY)); - ASSERT_THAT(mkdirat(dirfd.get(), "NewDir", 0755), SyscallSucceeds()); + const TempPath parent = ASSERT_NO_ERRNO_AND_VALUE(TempPath::CreateDir()); + EXPECT_THAT(chmod(parent.path().c_str(), 0777 | S_ISVTX), SyscallSucceeds()); + + // After changing credentials below, we need to use an open fd to make + // modifications in the parent dir, because there is no guarantee that we will + // still have the ability to open it. + const FileDescriptor parent_fd = + ASSERT_NO_ERRNO_AND_VALUE(Open(parent.path(), O_DIRECTORY)); + ASSERT_THAT(openat(parent_fd.get(), "file", O_CREAT), SyscallSucceeds()); + ASSERT_THAT(mkdirat(parent_fd.get(), "dir", 0777), SyscallSucceeds()); + ASSERT_THAT(symlinkat("xyz", parent_fd.get(), "link"), SyscallSucceeds()); // Drop privileges and change IDs only in child thread, or else this parent // thread won't be able to open some log files after the test ends. @@ -62,7 +68,13 @@ TEST(StickyTest, StickyBitPermDenied) { syscall(SYS_setresuid, -1, absl::GetFlag(FLAGS_scratch_uid), -1), SyscallSucceeds()); - EXPECT_THAT(unlinkat(dirfd.get(), "NewDir", AT_REMOVEDIR), + EXPECT_THAT(renameat(parent_fd.get(), "file", parent_fd.get(), "file2"), + SyscallFailsWithErrno(EPERM)); + EXPECT_THAT(unlinkat(parent_fd.get(), "file", 0), + SyscallFailsWithErrno(EPERM)); + EXPECT_THAT(unlinkat(parent_fd.get(), "dir", AT_REMOVEDIR), + SyscallFailsWithErrno(EPERM)); + EXPECT_THAT(unlinkat(parent_fd.get(), "link", 0), SyscallFailsWithErrno(EPERM)); }); } @@ -70,10 +82,17 @@ TEST(StickyTest, StickyBitPermDenied) { TEST(StickyTest, StickyBitSameUID) { SKIP_IF(!ASSERT_NO_ERRNO_AND_VALUE(HaveCapability(CAP_SETUID))); - auto dir = ASSERT_NO_ERRNO_AND_VALUE(TempPath::CreateDir()); - EXPECT_THAT(chmod(dir.path().c_str(), 0777 | S_ISVTX), SyscallSucceeds()); - std::string path = JoinPath(dir.path(), "NewDir"); - ASSERT_THAT(mkdir(path.c_str(), 0755), SyscallSucceeds()); + const TempPath parent = ASSERT_NO_ERRNO_AND_VALUE(TempPath::CreateDir()); + EXPECT_THAT(chmod(parent.path().c_str(), 0777 | S_ISVTX), SyscallSucceeds()); + + // After changing credentials below, we need to use an open fd to make + // modifications in the parent dir, because there is no guarantee that we will + // still have the ability to open it. + const FileDescriptor parent_fd = + ASSERT_NO_ERRNO_AND_VALUE(Open(parent.path(), O_DIRECTORY)); + ASSERT_THAT(openat(parent_fd.get(), "file", O_CREAT), SyscallSucceeds()); + ASSERT_THAT(mkdirat(parent_fd.get(), "dir", 0777), SyscallSucceeds()); + ASSERT_THAT(symlinkat("xyz", parent_fd.get(), "link"), SyscallSucceeds()); // Drop privileges and change IDs only in child thread, or else this parent // thread won't be able to open some log files after the test ends. @@ -89,18 +108,29 @@ TEST(StickyTest, StickyBitSameUID) { SyscallSucceeds()); // We still have the same EUID. - EXPECT_THAT(rmdir(path.c_str()), SyscallSucceeds()); + EXPECT_THAT(renameat(parent_fd.get(), "file", parent_fd.get(), "file2"), + SyscallSucceeds()); + EXPECT_THAT(unlinkat(parent_fd.get(), "file2", 0), SyscallSucceeds()); + EXPECT_THAT(unlinkat(parent_fd.get(), "dir", AT_REMOVEDIR), + SyscallSucceeds()); + EXPECT_THAT(unlinkat(parent_fd.get(), "link", 0), SyscallSucceeds()); }); } TEST(StickyTest, StickyBitCapFOWNER) { SKIP_IF(!ASSERT_NO_ERRNO_AND_VALUE(HaveCapability(CAP_SETUID))); - auto dir = ASSERT_NO_ERRNO_AND_VALUE(TempPath::CreateDir()); - EXPECT_THAT(chmod(dir.path().c_str(), 0777 | S_ISVTX), SyscallSucceeds()); - const FileDescriptor dirfd = - ASSERT_NO_ERRNO_AND_VALUE(Open(dir.path(), O_DIRECTORY)); - ASSERT_THAT(mkdirat(dirfd.get(), "NewDir", 0755), SyscallSucceeds()); + const TempPath parent = ASSERT_NO_ERRNO_AND_VALUE(TempPath::CreateDir()); + EXPECT_THAT(chmod(parent.path().c_str(), 0777 | S_ISVTX), SyscallSucceeds()); + + // After changing credentials below, we need to use an open fd to make + // modifications in the parent dir, because there is no guarantee that we will + // still have the ability to open it. + const FileDescriptor parent_fd = + ASSERT_NO_ERRNO_AND_VALUE(Open(parent.path(), O_DIRECTORY)); + ASSERT_THAT(openat(parent_fd.get(), "file", O_CREAT), SyscallSucceeds()); + ASSERT_THAT(mkdirat(parent_fd.get(), "dir", 0777), SyscallSucceeds()); + ASSERT_THAT(symlinkat("xyz", parent_fd.get(), "link"), SyscallSucceeds()); // Drop privileges and change IDs only in child thread, or else this parent // thread won't be able to open some log files after the test ends. @@ -117,8 +147,12 @@ TEST(StickyTest, StickyBitCapFOWNER) { SyscallSucceeds()); EXPECT_NO_ERRNO(SetCapability(CAP_FOWNER, true)); - EXPECT_THAT(unlinkat(dirfd.get(), "NewDir", AT_REMOVEDIR), + EXPECT_THAT(renameat(parent_fd.get(), "file", parent_fd.get(), "file2"), + SyscallSucceeds()); + EXPECT_THAT(unlinkat(parent_fd.get(), "file2", 0), SyscallSucceeds()); + EXPECT_THAT(unlinkat(parent_fd.get(), "dir", AT_REMOVEDIR), SyscallSucceeds()); + EXPECT_THAT(unlinkat(parent_fd.get(), "link", 0), SyscallSucceeds()); }); } } // namespace diff --git a/test/syscalls/linux/tcp_socket.cc b/test/syscalls/linux/tcp_socket.cc index a4d2953e1..0cea7d11f 100644 --- a/test/syscalls/linux/tcp_socket.cc +++ b/test/syscalls/linux/tcp_socket.cc @@ -13,6 +13,9 @@ // limitations under the License. #include <fcntl.h> +#ifndef __fuchsia__ +#include <linux/filter.h> +#endif // __fuchsia__ #include <netinet/in.h> #include <netinet/tcp.h> #include <poll.h> @@ -1559,6 +1562,63 @@ TEST_P(SimpleTcpSocketTest, SetTCPWindowClampAboveHalfMinRcvBuf) { } } +#ifndef __fuchsia__ + +// TODO(gvisor.dev/2746): Support SO_ATTACH_FILTER/SO_DETACH_FILTER. +// gVisor currently silently ignores attaching a filter. +TEST_P(SimpleTcpSocketTest, SetSocketAttachDetachFilter) { + FileDescriptor s = + ASSERT_NO_ERRNO_AND_VALUE(Socket(GetParam(), SOCK_STREAM, IPPROTO_TCP)); + // Program generated using sudo tcpdump -i lo tcp and port 1234 -dd + struct sock_filter code[] = { + {0x28, 0, 0, 0x0000000c}, {0x15, 0, 6, 0x000086dd}, + {0x30, 0, 0, 0x00000014}, {0x15, 0, 15, 0x00000006}, + {0x28, 0, 0, 0x00000036}, {0x15, 12, 0, 0x000004d2}, + {0x28, 0, 0, 0x00000038}, {0x15, 10, 11, 0x000004d2}, + {0x15, 0, 10, 0x00000800}, {0x30, 0, 0, 0x00000017}, + {0x15, 0, 8, 0x00000006}, {0x28, 0, 0, 0x00000014}, + {0x45, 6, 0, 0x00001fff}, {0xb1, 0, 0, 0x0000000e}, + {0x48, 0, 0, 0x0000000e}, {0x15, 2, 0, 0x000004d2}, + {0x48, 0, 0, 0x00000010}, {0x15, 0, 1, 0x000004d2}, + {0x6, 0, 0, 0x00040000}, {0x6, 0, 0, 0x00000000}, + }; + struct sock_fprog bpf = { + .len = ABSL_ARRAYSIZE(code), + .filter = code, + }; + ASSERT_THAT( + setsockopt(s.get(), SOL_SOCKET, SO_ATTACH_FILTER, &bpf, sizeof(bpf)), + SyscallSucceeds()); + + constexpr int val = 0; + ASSERT_THAT( + setsockopt(s.get(), SOL_SOCKET, SO_DETACH_FILTER, &val, sizeof(val)), + SyscallSucceeds()); +} + +TEST_P(SimpleTcpSocketTest, SetSocketDetachFilterNoInstalledFilter) { + // TODO(gvisor.dev/2746): Support SO_ATTACH_FILTER/SO_DETACH_FILTER. + SKIP_IF(IsRunningOnGvisor()); + FileDescriptor s = + ASSERT_NO_ERRNO_AND_VALUE(Socket(GetParam(), SOCK_STREAM, IPPROTO_TCP)); + constexpr int val = 0; + ASSERT_THAT( + setsockopt(s.get(), SOL_SOCKET, SO_DETACH_FILTER, &val, sizeof(val)), + SyscallFailsWithErrno(ENOENT)); +} + +TEST_P(SimpleTcpSocketTest, GetSocketDetachFilter) { + FileDescriptor s = + ASSERT_NO_ERRNO_AND_VALUE(Socket(GetParam(), SOCK_STREAM, IPPROTO_TCP)); + + int val = 0; + socklen_t val_len = sizeof(val); + ASSERT_THAT(getsockopt(s.get(), SOL_SOCKET, SO_DETACH_FILTER, &val, &val_len), + SyscallFailsWithErrno(ENOPROTOOPT)); +} + +#endif // __fuchsia__ + INSTANTIATE_TEST_SUITE_P(AllInetTests, SimpleTcpSocketTest, ::testing::Values(AF_INET, AF_INET6)); diff --git a/test/syscalls/linux/timerfd.cc b/test/syscalls/linux/timerfd.cc index 86ed87b7c..c4f8fdd7a 100644 --- a/test/syscalls/linux/timerfd.cc +++ b/test/syscalls/linux/timerfd.cc @@ -204,16 +204,33 @@ TEST_P(TimerfdTest, SetAbsoluteTime) { EXPECT_EQ(1, val); } -TEST_P(TimerfdTest, IllegalReadWrite) { +TEST_P(TimerfdTest, IllegalSeek) { + auto const tfd = ASSERT_NO_ERRNO_AND_VALUE(TimerfdCreate(GetParam(), 0)); + if (!IsRunningWithVFS1()) { + EXPECT_THAT(lseek(tfd.get(), 0, SEEK_SET), SyscallFailsWithErrno(ESPIPE)); + } +} + +TEST_P(TimerfdTest, IllegalPread) { + auto const tfd = ASSERT_NO_ERRNO_AND_VALUE(TimerfdCreate(GetParam(), 0)); + int val; + EXPECT_THAT(pread(tfd.get(), &val, sizeof(val), 0), + SyscallFailsWithErrno(ESPIPE)); +} + +TEST_P(TimerfdTest, IllegalPwrite) { + auto const tfd = ASSERT_NO_ERRNO_AND_VALUE(TimerfdCreate(GetParam(), 0)); + EXPECT_THAT(pwrite(tfd.get(), "x", 1, 0), SyscallFailsWithErrno(ESPIPE)); + if (!IsRunningWithVFS1()) { + } +} + +TEST_P(TimerfdTest, IllegalWrite) { auto const tfd = ASSERT_NO_ERRNO_AND_VALUE(TimerfdCreate(GetParam(), TFD_NONBLOCK)); uint64_t val = 0; - EXPECT_THAT(PreadFd(tfd.get(), &val, sizeof(val), 0), - SyscallFailsWithErrno(ESPIPE)); - EXPECT_THAT(WriteFd(tfd.get(), &val, sizeof(val)), + EXPECT_THAT(write(tfd.get(), &val, sizeof(val)), SyscallFailsWithErrno(EINVAL)); - EXPECT_THAT(PwriteFd(tfd.get(), &val, sizeof(val), 0), - SyscallFailsWithErrno(ESPIPE)); } std::string PrintClockId(::testing::TestParamInfo<int> info) { diff --git a/test/syscalls/linux/tuntap.cc b/test/syscalls/linux/tuntap.cc index 6195b11e1..97d554e72 100644 --- a/test/syscalls/linux/tuntap.cc +++ b/test/syscalls/linux/tuntap.cc @@ -398,5 +398,25 @@ TEST_F(TuntapTest, SendUdpTriggersArpResolution) { } } +// Write hang bug found by syskaller: b/155928773 +// https://syzkaller.appspot.com/bug?id=065b893bd8d1d04a4e0a1d53c578537cde1efe99 +TEST_F(TuntapTest, WriteHangBug155928773) { + SKIP_IF(!ASSERT_NO_ERRNO_AND_VALUE(HaveCapability(CAP_NET_ADMIN))); + + FileDescriptor fd = + ASSERT_NO_ERRNO_AND_VALUE(OpenAndAttachTap(kTapName, "10.0.0.1")); + + int sock = socket(AF_INET, SOCK_DGRAM, 0); + ASSERT_THAT(sock, SyscallSucceeds()); + + struct sockaddr_in remote = {}; + remote.sin_family = AF_INET; + remote.sin_port = htons(42); + inet_pton(AF_INET, "10.0.0.1", &remote.sin_addr); + // Return values do not matter in this test. + connect(sock, reinterpret_cast<struct sockaddr*>(&remote), sizeof(remote)); + write(sock, "hello", 5); +} + } // namespace testing } // namespace gvisor diff --git a/test/syscalls/linux/udp_socket_errqueue_test_case.cc b/test/syscalls/linux/udp_socket_errqueue_test_case.cc index fcdba7279..54a0594f7 100644 --- a/test/syscalls/linux/udp_socket_errqueue_test_case.cc +++ b/test/syscalls/linux/udp_socket_errqueue_test_case.cc @@ -47,7 +47,7 @@ TEST_P(UdpSocketTest, ErrorQueue) { msg.msg_controllen = sizeof(cmsgbuf); // recv*(MSG_ERRQUEUE) never blocks, even without MSG_DONTWAIT. - EXPECT_THAT(RetryEINTR(recvmsg)(s_, &msg, MSG_ERRQUEUE), + EXPECT_THAT(RetryEINTR(recvmsg)(bind_.get(), &msg, MSG_ERRQUEUE), SyscallFailsWithErrno(EAGAIN)); } diff --git a/test/syscalls/linux/udp_socket_test_cases.cc b/test/syscalls/linux/udp_socket_test_cases.cc index 42521efef..60c48ed6e 100644 --- a/test/syscalls/linux/udp_socket_test_cases.cc +++ b/test/syscalls/linux/udp_socket_test_cases.cc @@ -16,12 +16,16 @@ #include <arpa/inet.h> #include <fcntl.h> +#ifndef __fuchsia__ +#include <linux/filter.h> +#endif // __fuchsia__ #include <netinet/in.h> #include <poll.h> #include <sys/ioctl.h> #include <sys/socket.h> #include <sys/types.h> +#include "absl/strings/str_format.h" #ifndef SIOCGSTAMP #include <linux/sockios.h> #endif @@ -30,8 +34,11 @@ #include "absl/base/macros.h" #include "absl/time/clock.h" #include "absl/time/time.h" +#include "test/syscalls/linux/ip_socket_test_util.h" #include "test/syscalls/linux/socket_test_util.h" #include "test/syscalls/linux/unix_domain_socket_test_util.h" +#include "test/util/file_descriptor.h" +#include "test/util/posix_error.h" #include "test/util/test_util.h" #include "test/util/thread_util.h" @@ -54,517 +61,470 @@ uint16_t* Port(struct sockaddr_storage* addr) { return nullptr; } +// Sets addr port to "port". +void SetPort(struct sockaddr_storage* addr, uint16_t port) { + switch (addr->ss_family) { + case AF_INET: { + auto sin = reinterpret_cast<struct sockaddr_in*>(addr); + sin->sin_port = port; + break; + } + case AF_INET6: { + auto sin6 = reinterpret_cast<struct sockaddr_in6*>(addr); + sin6->sin6_port = port; + break; + } + } +} + void UdpSocketTest::SetUp() { - int type; + addrlen_ = GetAddrLength(); + + bind_ = + ASSERT_NO_ERRNO_AND_VALUE(Socket(GetFamily(), SOCK_DGRAM, IPPROTO_UDP)); + memset(&bind_addr_storage_, 0, sizeof(bind_addr_storage_)); + bind_addr_ = reinterpret_cast<struct sockaddr*>(&bind_addr_storage_); + + sock_ = + ASSERT_NO_ERRNO_AND_VALUE(Socket(GetFamily(), SOCK_DGRAM, IPPROTO_UDP)); +} + +int UdpSocketTest::GetFamily() { if (GetParam() == AddressFamily::kIpv4) { - type = AF_INET; - auto sin = reinterpret_cast<struct sockaddr_in*>(&anyaddr_storage_); - addrlen_ = sizeof(*sin); - sin->sin_addr.s_addr = htonl(INADDR_ANY); - } else { - type = AF_INET6; - auto sin6 = reinterpret_cast<struct sockaddr_in6*>(&anyaddr_storage_); - addrlen_ = sizeof(*sin6); - if (GetParam() == AddressFamily::kIpv6) { - sin6->sin6_addr = IN6ADDR_ANY_INIT; - } else { - TestAddress const& v4_mapped_any = V4MappedAny(); - sin6->sin6_addr = - reinterpret_cast<const struct sockaddr_in6*>(&v4_mapped_any.addr) - ->sin6_addr; - } + return AF_INET; } - ASSERT_THAT(s_ = socket(type, SOCK_DGRAM, IPPROTO_UDP), SyscallSucceeds()); + return AF_INET6; +} - ASSERT_THAT(t_ = socket(type, SOCK_DGRAM, IPPROTO_UDP), SyscallSucceeds()); +PosixError UdpSocketTest::BindLoopback() { + bind_addr_storage_ = InetLoopbackAddr(); + struct sockaddr* bind_addr_ = + reinterpret_cast<struct sockaddr*>(&bind_addr_storage_); + return BindSocket(bind_.get(), bind_addr_); +} - memset(&anyaddr_storage_, 0, sizeof(anyaddr_storage_)); - anyaddr_ = reinterpret_cast<struct sockaddr*>(&anyaddr_storage_); - anyaddr_->sa_family = type; +PosixError UdpSocketTest::BindAny() { + bind_addr_storage_ = InetAnyAddr(); + struct sockaddr* bind_addr_ = + reinterpret_cast<struct sockaddr*>(&bind_addr_storage_); + return BindSocket(bind_.get(), bind_addr_); +} - if (gvisor::testing::IsRunningOnGvisor()) { - for (size_t i = 0; i < ABSL_ARRAYSIZE(ports_); ++i) { - ports_[i] = TestPort + i; - } - } else { - // When not under gvisor, use utility function to pick port. Assert that - // all ports are different. - std::string error; - for (size_t i = 0; i < ABSL_ARRAYSIZE(ports_); ++i) { - // Find an unused port, we specify port 0 to allow the kernel to provide - // the port. - bool unique = true; - do { - ports_[i] = ASSERT_NO_ERRNO_AND_VALUE(PortAvailable( - 0, AddressFamily::kDualStack, SocketType::kUdp, false)); - ASSERT_GT(ports_[i], 0); - for (size_t j = 0; j < i; ++j) { - if (ports_[j] == ports_[i]) { - unique = false; - break; - } - } - } while (!unique); - } +PosixError UdpSocketTest::BindSocket(int socket, struct sockaddr* addr) { + socklen_t len = sizeof(bind_addr_storage_); + + // Bind, then check that we get the right address. + RETURN_ERROR_IF_SYSCALL_FAIL(bind(socket, addr, addrlen_)); + + RETURN_ERROR_IF_SYSCALL_FAIL(getsockname(socket, addr, &len)); + + if (addrlen_ != len) { + return PosixError( + EINVAL, + absl::StrFormat("getsockname len: %u expected: %u", len, addrlen_)); } + return PosixError(0); +} - // Initialize the sockaddrs. - for (size_t i = 0; i < ABSL_ARRAYSIZE(addr_); ++i) { - memset(&addr_storage_[i], 0, sizeof(addr_storage_[i])); - - addr_[i] = reinterpret_cast<struct sockaddr*>(&addr_storage_[i]); - addr_[i]->sa_family = type; - - switch (type) { - case AF_INET: { - auto sin = reinterpret_cast<struct sockaddr_in*>(addr_[i]); - sin->sin_addr.s_addr = htonl(INADDR_LOOPBACK); - sin->sin_port = htons(ports_[i]); - break; - } - case AF_INET6: { - auto sin6 = reinterpret_cast<struct sockaddr_in6*>(addr_[i]); - sin6->sin6_addr = in6addr_loopback; - sin6->sin6_port = htons(ports_[i]); - break; - } - } +socklen_t UdpSocketTest::GetAddrLength() { + struct sockaddr_storage addr; + if (GetFamily() == AF_INET) { + auto sin = reinterpret_cast<struct sockaddr_in*>(&addr); + return sizeof(*sin); } + + auto sin6 = reinterpret_cast<struct sockaddr_in6*>(&addr); + return sizeof(*sin6); } -TEST_P(UdpSocketTest, Creation) { - int type = AF_INET6; - if (GetParam() == AddressFamily::kIpv4) { - type = AF_INET; +sockaddr_storage UdpSocketTest::InetAnyAddr() { + struct sockaddr_storage addr; + memset(&addr, 0, sizeof(addr)); + reinterpret_cast<struct sockaddr*>(&addr)->sa_family = GetFamily(); + + if (GetFamily() == AF_INET) { + auto sin = reinterpret_cast<struct sockaddr_in*>(&addr); + sin->sin_addr.s_addr = htonl(INADDR_ANY); + sin->sin_port = htons(0); + return addr; } - int s_; + auto sin6 = reinterpret_cast<struct sockaddr_in6*>(&addr); + sin6->sin6_addr = IN6ADDR_ANY_INIT; + sin6->sin6_port = htons(0); + return addr; +} + +sockaddr_storage UdpSocketTest::InetLoopbackAddr() { + struct sockaddr_storage addr; + memset(&addr, 0, sizeof(addr)); + reinterpret_cast<struct sockaddr*>(&addr)->sa_family = GetFamily(); + + if (GetFamily() == AF_INET) { + auto sin = reinterpret_cast<struct sockaddr_in*>(&addr); + sin->sin_addr.s_addr = htonl(INADDR_LOOPBACK); + sin->sin_port = htons(0); + return addr; + } + auto sin6 = reinterpret_cast<struct sockaddr_in6*>(&addr); + sin6->sin6_addr = in6addr_loopback; + sin6->sin6_port = htons(0); + return addr; +} + +void UdpSocketTest::Disconnect(int sockfd) { + sockaddr_storage addr_storage = InetAnyAddr(); + sockaddr* addr = reinterpret_cast<struct sockaddr*>(&addr_storage); + socklen_t addrlen = sizeof(addr_storage); + + addr->sa_family = AF_UNSPEC; + ASSERT_THAT(connect(sockfd, addr, addrlen), SyscallSucceeds()); + + // Check that after disconnect the socket is bound to the ANY address. + EXPECT_THAT(getsockname(sockfd, addr, &addrlen), SyscallSucceeds()); + if (GetParam() == AddressFamily::kIpv4) { + auto addr_out = reinterpret_cast<struct sockaddr_in*>(addr); + EXPECT_EQ(addrlen, sizeof(*addr_out)); + EXPECT_EQ(addr_out->sin_addr.s_addr, htonl(INADDR_ANY)); + } else { + auto addr_out = reinterpret_cast<struct sockaddr_in6*>(addr); + EXPECT_EQ(addrlen, sizeof(*addr_out)); + struct in6_addr loopback = IN6ADDR_ANY_INIT; + + EXPECT_EQ(memcmp(&addr_out->sin6_addr, &loopback, sizeof(in6_addr)), 0); + } +} - ASSERT_THAT(s_ = socket(type, SOCK_DGRAM, IPPROTO_UDP), SyscallSucceeds()); - EXPECT_THAT(close(s_), SyscallSucceeds()); +TEST_P(UdpSocketTest, Creation) { + FileDescriptor sock = + ASSERT_NO_ERRNO_AND_VALUE(Socket(GetFamily(), SOCK_DGRAM, IPPROTO_UDP)); + EXPECT_THAT(close(sock.release()), SyscallSucceeds()); - ASSERT_THAT(s_ = socket(type, SOCK_DGRAM, 0), SyscallSucceeds()); - EXPECT_THAT(close(s_), SyscallSucceeds()); + sock = ASSERT_NO_ERRNO_AND_VALUE(Socket(GetFamily(), SOCK_DGRAM, 0)); + EXPECT_THAT(close(sock.release()), SyscallSucceeds()); - ASSERT_THAT(s_ = socket(type, SOCK_STREAM, IPPROTO_UDP), SyscallFails()); + ASSERT_THAT(socket(GetFamily(), SOCK_STREAM, IPPROTO_UDP), SyscallFails()); } TEST_P(UdpSocketTest, Getsockname) { // Check that we're not bound. struct sockaddr_storage addr; socklen_t addrlen = sizeof(addr); - EXPECT_THAT(getsockname(s_, reinterpret_cast<sockaddr*>(&addr), &addrlen), - SyscallSucceeds()); + EXPECT_THAT( + getsockname(bind_.get(), reinterpret_cast<sockaddr*>(&addr), &addrlen), + SyscallSucceeds()); EXPECT_EQ(addrlen, addrlen_); - EXPECT_EQ(memcmp(&addr, anyaddr_, addrlen_), 0); + struct sockaddr_storage any = InetAnyAddr(); + EXPECT_EQ(memcmp(&addr, reinterpret_cast<struct sockaddr*>(&any), addrlen_), + 0); - // Bind, then check that we get the right address. - ASSERT_THAT(bind(s_, addr_[0], addrlen_), SyscallSucceeds()); + ASSERT_NO_ERRNO(BindLoopback()); + + EXPECT_THAT( + getsockname(bind_.get(), reinterpret_cast<sockaddr*>(&addr), &addrlen), + SyscallSucceeds()); - addrlen = sizeof(addr); - EXPECT_THAT(getsockname(s_, reinterpret_cast<sockaddr*>(&addr), &addrlen), - SyscallSucceeds()); EXPECT_EQ(addrlen, addrlen_); - EXPECT_EQ(memcmp(&addr, addr_[0], addrlen_), 0); + EXPECT_EQ(memcmp(&addr, bind_addr_, addrlen_), 0); } TEST_P(UdpSocketTest, Getpeername) { + ASSERT_NO_ERRNO(BindLoopback()); + // Check that we're not connected. struct sockaddr_storage addr; socklen_t addrlen = sizeof(addr); - EXPECT_THAT(getpeername(s_, reinterpret_cast<sockaddr*>(&addr), &addrlen), - SyscallFailsWithErrno(ENOTCONN)); + EXPECT_THAT( + getpeername(sock_.get(), reinterpret_cast<sockaddr*>(&addr), &addrlen), + SyscallFailsWithErrno(ENOTCONN)); // Connect, then check that we get the right address. - ASSERT_THAT(connect(s_, addr_[0], addrlen_), SyscallSucceeds()); + ASSERT_THAT(connect(sock_.get(), bind_addr_, addrlen_), SyscallSucceeds()); addrlen = sizeof(addr); - EXPECT_THAT(getpeername(s_, reinterpret_cast<sockaddr*>(&addr), &addrlen), - SyscallSucceeds()); + EXPECT_THAT( + getpeername(sock_.get(), reinterpret_cast<sockaddr*>(&addr), &addrlen), + SyscallSucceeds()); EXPECT_EQ(addrlen, addrlen_); - EXPECT_EQ(memcmp(&addr, addr_[0], addrlen_), 0); + EXPECT_EQ(memcmp(&addr, bind_addr_, addrlen_), 0); } TEST_P(UdpSocketTest, SendNotConnected) { + ASSERT_NO_ERRNO(BindLoopback()); + // Do send & write, they must fail. char buf[512]; - EXPECT_THAT(send(s_, buf, sizeof(buf), 0), + EXPECT_THAT(send(sock_.get(), buf, sizeof(buf), 0), SyscallFailsWithErrno(EDESTADDRREQ)); - EXPECT_THAT(write(s_, buf, sizeof(buf)), SyscallFailsWithErrno(EDESTADDRREQ)); + EXPECT_THAT(write(sock_.get(), buf, sizeof(buf)), + SyscallFailsWithErrno(EDESTADDRREQ)); // Use sendto. - ASSERT_THAT(sendto(s_, buf, sizeof(buf), 0, addr_[0], addrlen_), + ASSERT_THAT(sendto(sock_.get(), buf, sizeof(buf), 0, bind_addr_, addrlen_), SyscallSucceedsWithValue(sizeof(buf))); // Check that we're bound now. struct sockaddr_storage addr; socklen_t addrlen = sizeof(addr); - EXPECT_THAT(getsockname(s_, reinterpret_cast<sockaddr*>(&addr), &addrlen), - SyscallSucceeds()); + EXPECT_THAT( + getsockname(sock_.get(), reinterpret_cast<sockaddr*>(&addr), &addrlen), + SyscallSucceeds()); EXPECT_EQ(addrlen, addrlen_); EXPECT_NE(*Port(&addr), 0); } TEST_P(UdpSocketTest, ConnectBinds) { + ASSERT_NO_ERRNO(BindLoopback()); + // Connect the socket. - ASSERT_THAT(connect(s_, addr_[0], addrlen_), SyscallSucceeds()); + ASSERT_THAT(connect(sock_.get(), bind_addr_, addrlen_), SyscallSucceeds()); // Check that we're bound now. struct sockaddr_storage addr; socklen_t addrlen = sizeof(addr); - EXPECT_THAT(getsockname(s_, reinterpret_cast<sockaddr*>(&addr), &addrlen), - SyscallSucceeds()); + EXPECT_THAT( + getsockname(sock_.get(), reinterpret_cast<sockaddr*>(&addr), &addrlen), + SyscallSucceeds()); EXPECT_EQ(addrlen, addrlen_); EXPECT_NE(*Port(&addr), 0); } TEST_P(UdpSocketTest, ReceiveNotBound) { char buf[512]; - EXPECT_THAT(recv(s_, buf, sizeof(buf), MSG_DONTWAIT), + EXPECT_THAT(recv(sock_.get(), buf, sizeof(buf), MSG_DONTWAIT), SyscallFailsWithErrno(EWOULDBLOCK)); } TEST_P(UdpSocketTest, Bind) { - ASSERT_THAT(bind(s_, addr_[0], addrlen_), SyscallSucceeds()); + ASSERT_NO_ERRNO(BindLoopback()); // Try to bind again. - EXPECT_THAT(bind(s_, addr_[1], addrlen_), SyscallFailsWithErrno(EINVAL)); + EXPECT_THAT(bind(bind_.get(), bind_addr_, addrlen_), + SyscallFailsWithErrno(EINVAL)); // Check that we're still bound to the original address. struct sockaddr_storage addr; socklen_t addrlen = sizeof(addr); - EXPECT_THAT(getsockname(s_, reinterpret_cast<sockaddr*>(&addr), &addrlen), - SyscallSucceeds()); + EXPECT_THAT( + getsockname(bind_.get(), reinterpret_cast<sockaddr*>(&addr), &addrlen), + SyscallSucceeds()); EXPECT_EQ(addrlen, addrlen_); - EXPECT_EQ(memcmp(&addr, addr_[0], addrlen_), 0); + EXPECT_EQ(memcmp(&addr, bind_addr_, addrlen_), 0); } TEST_P(UdpSocketTest, BindInUse) { - ASSERT_THAT(bind(s_, addr_[0], addrlen_), SyscallSucceeds()); + ASSERT_NO_ERRNO(BindLoopback()); // Try to bind again. - EXPECT_THAT(bind(t_, addr_[0], addrlen_), SyscallFailsWithErrno(EADDRINUSE)); + EXPECT_THAT(bind(sock_.get(), bind_addr_, addrlen_), + SyscallFailsWithErrno(EADDRINUSE)); } TEST_P(UdpSocketTest, ReceiveAfterConnect) { - // Connect s_ to loopback:TestPort, and bind t_ to loopback:TestPort. - ASSERT_THAT(connect(s_, addr_[0], addrlen_), SyscallSucceeds()); - ASSERT_THAT(bind(t_, addr_[0], addrlen_), SyscallSucceeds()); + ASSERT_NO_ERRNO(BindLoopback()); + ASSERT_THAT(connect(sock_.get(), bind_addr_, addrlen_), SyscallSucceeds()); - // Get the address s_ was bound to during connect. - struct sockaddr_storage addr; - socklen_t addrlen = sizeof(addr); - EXPECT_THAT(getsockname(s_, reinterpret_cast<sockaddr*>(&addr), &addrlen), - SyscallSucceeds()); - EXPECT_EQ(addrlen, addrlen_); - - // Send from t_ to s_. + // Send from sock_ to bind_ char buf[512]; RandomizeBuffer(buf, sizeof(buf)); - ASSERT_THAT(sendto(t_, buf, sizeof(buf), 0, - reinterpret_cast<sockaddr*>(&addr), addrlen), + ASSERT_THAT(sendto(sock_.get(), buf, sizeof(buf), 0, bind_addr_, addrlen_), SyscallSucceedsWithValue(sizeof(buf))); // Receive the data. char received[sizeof(buf)]; - EXPECT_THAT(recv(s_, received, sizeof(received), 0), + EXPECT_THAT(recv(bind_.get(), received, sizeof(received), 0), SyscallSucceedsWithValue(sizeof(received))); EXPECT_EQ(memcmp(buf, received, sizeof(buf)), 0); } TEST_P(UdpSocketTest, ReceiveAfterDisconnect) { - // Connect s_ to loopback:TestPort, and bind t_ to loopback:TestPort. - ASSERT_THAT(connect(s_, addr_[0], addrlen_), SyscallSucceeds()); - ASSERT_THAT(bind(t_, addr_[0], addrlen_), SyscallSucceeds()); - ASSERT_THAT(connect(t_, addr_[1], addrlen_), SyscallSucceeds()); - - // Get the address s_ was bound to during connect. - struct sockaddr_storage addr; - socklen_t addrlen = sizeof(addr); - EXPECT_THAT(getsockname(s_, reinterpret_cast<sockaddr*>(&addr), &addrlen), - SyscallSucceeds()); - EXPECT_EQ(addrlen, addrlen_); + ASSERT_NO_ERRNO(BindLoopback()); for (int i = 0; i < 2; i++) { - // Send from t_ to s_. + // Connet sock_ to bound address. + ASSERT_THAT(connect(sock_.get(), bind_addr_, addrlen_), SyscallSucceeds()); + + struct sockaddr_storage addr; + socklen_t addrlen = sizeof(addr); + EXPECT_THAT( + getsockname(sock_.get(), reinterpret_cast<sockaddr*>(&addr), &addrlen), + SyscallSucceeds()); + EXPECT_EQ(addrlen, addrlen_); + + // Send from sock to bind_. char buf[512]; RandomizeBuffer(buf, sizeof(buf)); - EXPECT_THAT(getsockname(s_, reinterpret_cast<sockaddr*>(&addr), &addrlen), - SyscallSucceeds()); - ASSERT_THAT(sendto(t_, buf, sizeof(buf), 0, + + ASSERT_THAT(sendto(bind_.get(), buf, sizeof(buf), 0, reinterpret_cast<sockaddr*>(&addr), addrlen), SyscallSucceedsWithValue(sizeof(buf))); // Receive the data. char received[sizeof(buf)]; - EXPECT_THAT(recv(s_, received, sizeof(received), 0), + EXPECT_THAT(recv(sock_.get(), received, sizeof(received), 0), SyscallSucceedsWithValue(sizeof(received))); EXPECT_EQ(memcmp(buf, received, sizeof(buf)), 0); - // Disconnect s_. - struct sockaddr addr = {}; - addr.sa_family = AF_UNSPEC; - ASSERT_THAT(connect(s_, &addr, sizeof(addr.sa_family)), SyscallSucceeds()); - // Connect s_ loopback:TestPort. - ASSERT_THAT(connect(s_, addr_[0], addrlen_), SyscallSucceeds()); + // Disconnect sock_. + struct sockaddr unspec = {}; + unspec.sa_family = AF_UNSPEC; + ASSERT_THAT(connect(sock_.get(), &unspec, sizeof(unspec.sa_family)), + SyscallSucceeds()); } } TEST_P(UdpSocketTest, Connect) { - ASSERT_THAT(connect(s_, addr_[0], addrlen_), SyscallSucceeds()); + ASSERT_NO_ERRNO(BindLoopback()); + ASSERT_THAT(connect(sock_.get(), bind_addr_, addrlen_), SyscallSucceeds()); // Check that we're connected to the right peer. struct sockaddr_storage peer; socklen_t peerlen = sizeof(peer); - EXPECT_THAT(getpeername(s_, reinterpret_cast<sockaddr*>(&peer), &peerlen), - SyscallSucceeds()); + EXPECT_THAT( + getpeername(sock_.get(), reinterpret_cast<sockaddr*>(&peer), &peerlen), + SyscallSucceeds()); EXPECT_EQ(peerlen, addrlen_); - EXPECT_EQ(memcmp(&peer, addr_[0], addrlen_), 0); + EXPECT_EQ(memcmp(&peer, bind_addr_, addrlen_), 0); // Try to bind after connect. - EXPECT_THAT(bind(s_, addr_[1], addrlen_), SyscallFailsWithErrno(EINVAL)); + struct sockaddr_storage any = InetAnyAddr(); + EXPECT_THAT( + bind(sock_.get(), reinterpret_cast<struct sockaddr*>(&any), addrlen_), + SyscallFailsWithErrno(EINVAL)); + + struct sockaddr_storage bind2_storage = InetLoopbackAddr(); + struct sockaddr* bind2_addr = + reinterpret_cast<struct sockaddr*>(&bind2_storage); + FileDescriptor bind2 = + ASSERT_NO_ERRNO_AND_VALUE(Socket(GetFamily(), SOCK_DGRAM, IPPROTO_UDP)); + ASSERT_NO_ERRNO(BindSocket(bind2.get(), bind2_addr)); // Try to connect again. - EXPECT_THAT(connect(s_, addr_[2], addrlen_), SyscallSucceeds()); + EXPECT_THAT(connect(sock_.get(), bind2_addr, addrlen_), SyscallSucceeds()); // Check that peer name changed. peerlen = sizeof(peer); - EXPECT_THAT(getpeername(s_, reinterpret_cast<sockaddr*>(&peer), &peerlen), - SyscallSucceeds()); + EXPECT_THAT( + getpeername(sock_.get(), reinterpret_cast<sockaddr*>(&peer), &peerlen), + SyscallSucceeds()); EXPECT_EQ(peerlen, addrlen_); - EXPECT_EQ(memcmp(&peer, addr_[2], addrlen_), 0); + EXPECT_EQ(memcmp(&peer, bind2_addr, addrlen_), 0); } -void ConnectAny(AddressFamily family, int sockfd, uint16_t port) { - struct sockaddr_storage addr = {}; +TEST_P(UdpSocketTest, ConnectAnyZero) { + // TODO(138658473): Enable when we can connect to port 0 with gVisor. + SKIP_IF(IsRunningOnGvisor()); - // Precondition check. - { - socklen_t addrlen = sizeof(addr); - EXPECT_THAT( - getsockname(sockfd, reinterpret_cast<sockaddr*>(&addr), &addrlen), - SyscallSucceeds()); - - if (family == AddressFamily::kIpv4) { - auto addr_out = reinterpret_cast<struct sockaddr_in*>(&addr); - EXPECT_EQ(addrlen, sizeof(*addr_out)); - EXPECT_EQ(addr_out->sin_addr.s_addr, htonl(INADDR_ANY)); - } else { - auto addr_out = reinterpret_cast<struct sockaddr_in6*>(&addr); - EXPECT_EQ(addrlen, sizeof(*addr_out)); - struct in6_addr any = IN6ADDR_ANY_INIT; - EXPECT_EQ(memcmp(&addr_out->sin6_addr, &any, sizeof(in6_addr)), 0); - } - - { - socklen_t addrlen = sizeof(addr); - EXPECT_THAT( - getpeername(sockfd, reinterpret_cast<sockaddr*>(&addr), &addrlen), - SyscallFailsWithErrno(ENOTCONN)); - } - - struct sockaddr_storage baddr = {}; - if (family == AddressFamily::kIpv4) { - auto addr_in = reinterpret_cast<struct sockaddr_in*>(&baddr); - addrlen = sizeof(*addr_in); - addr_in->sin_family = AF_INET; - addr_in->sin_addr.s_addr = htonl(INADDR_ANY); - addr_in->sin_port = port; - } else { - auto addr_in = reinterpret_cast<struct sockaddr_in6*>(&baddr); - addrlen = sizeof(*addr_in); - addr_in->sin6_family = AF_INET6; - addr_in->sin6_port = port; - if (family == AddressFamily::kIpv6) { - addr_in->sin6_addr = IN6ADDR_ANY_INIT; - } else { - TestAddress const& v4_mapped_any = V4MappedAny(); - addr_in->sin6_addr = - reinterpret_cast<const struct sockaddr_in6*>(&v4_mapped_any.addr) - ->sin6_addr; - } - } - - // TODO(b/138658473): gVisor doesn't allow connecting to the zero port. - if (port == 0) { - SKIP_IF(IsRunningOnGvisor()); - } - - ASSERT_THAT(connect(sockfd, reinterpret_cast<sockaddr*>(&baddr), addrlen), - SyscallSucceeds()); - } - - // Postcondition check. - { - socklen_t addrlen = sizeof(addr); - EXPECT_THAT( - getsockname(sockfd, reinterpret_cast<sockaddr*>(&addr), &addrlen), - SyscallSucceeds()); - - if (family == AddressFamily::kIpv4) { - auto addr_out = reinterpret_cast<struct sockaddr_in*>(&addr); - EXPECT_EQ(addrlen, sizeof(*addr_out)); - EXPECT_EQ(addr_out->sin_addr.s_addr, htonl(INADDR_LOOPBACK)); - } else { - auto addr_out = reinterpret_cast<struct sockaddr_in6*>(&addr); - EXPECT_EQ(addrlen, sizeof(*addr_out)); - struct in6_addr loopback; - if (family == AddressFamily::kIpv6) { - loopback = IN6ADDR_LOOPBACK_INIT; - } else { - TestAddress const& v4_mapped_loopback = V4MappedLoopback(); - loopback = reinterpret_cast<const struct sockaddr_in6*>( - &v4_mapped_loopback.addr) - ->sin6_addr; - } - - EXPECT_EQ(memcmp(&addr_out->sin6_addr, &loopback, sizeof(in6_addr)), 0); - } + struct sockaddr_storage any = InetAnyAddr(); + EXPECT_THAT( + connect(sock_.get(), reinterpret_cast<struct sockaddr*>(&any), addrlen_), + SyscallSucceeds()); - addrlen = sizeof(addr); - if (port == 0) { - EXPECT_THAT( - getpeername(sockfd, reinterpret_cast<sockaddr*>(&addr), &addrlen), - SyscallFailsWithErrno(ENOTCONN)); - } else { - EXPECT_THAT( - getpeername(sockfd, reinterpret_cast<sockaddr*>(&addr), &addrlen), - SyscallSucceeds()); - } - } + struct sockaddr_storage addr; + socklen_t addrlen = sizeof(addr); + EXPECT_THAT( + getpeername(sock_.get(), reinterpret_cast<sockaddr*>(&addr), &addrlen), + SyscallFailsWithErrno(ENOTCONN)); } -TEST_P(UdpSocketTest, ConnectAny) { ConnectAny(GetParam(), s_, 0); } - TEST_P(UdpSocketTest, ConnectAnyWithPort) { - auto port = *Port(reinterpret_cast<struct sockaddr_storage*>(addr_[1])); - ConnectAny(GetParam(), s_, port); + ASSERT_NO_ERRNO(BindAny()); + ASSERT_THAT(connect(sock_.get(), bind_addr_, addrlen_), SyscallSucceeds()); + + struct sockaddr_storage addr; + socklen_t addrlen = sizeof(addr); + EXPECT_THAT( + getpeername(sock_.get(), reinterpret_cast<sockaddr*>(&addr), &addrlen), + SyscallSucceeds()); } -void DisconnectAfterConnectAny(AddressFamily family, int sockfd, int port) { - struct sockaddr_storage addr = {}; +TEST_P(UdpSocketTest, DisconnectAfterConnectAny) { + // TODO(138658473): Enable when we can connect to port 0 with gVisor. + SKIP_IF(IsRunningOnGvisor()); + struct sockaddr_storage any = InetAnyAddr(); + EXPECT_THAT( + connect(sock_.get(), reinterpret_cast<struct sockaddr*>(&any), addrlen_), + SyscallSucceeds()); + struct sockaddr_storage addr; socklen_t addrlen = sizeof(addr); - struct sockaddr_storage baddr = {}; - if (family == AddressFamily::kIpv4) { - auto addr_in = reinterpret_cast<struct sockaddr_in*>(&baddr); - addrlen = sizeof(*addr_in); - addr_in->sin_family = AF_INET; - addr_in->sin_addr.s_addr = htonl(INADDR_ANY); - addr_in->sin_port = port; - } else { - auto addr_in = reinterpret_cast<struct sockaddr_in6*>(&baddr); - addrlen = sizeof(*addr_in); - addr_in->sin6_family = AF_INET6; - addr_in->sin6_port = port; - if (family == AddressFamily::kIpv6) { - addr_in->sin6_addr = IN6ADDR_ANY_INIT; - } else { - TestAddress const& v4_mapped_any = V4MappedAny(); - addr_in->sin6_addr = - reinterpret_cast<const struct sockaddr_in6*>(&v4_mapped_any.addr) - ->sin6_addr; - } - } + EXPECT_THAT( + getpeername(sock_.get(), reinterpret_cast<sockaddr*>(&addr), &addrlen), + SyscallFailsWithErrno(ENOTCONN)); - // TODO(b/138658473): gVisor doesn't allow connecting to the zero port. - if (port == 0) { - SKIP_IF(IsRunningOnGvisor()); - } + Disconnect(sock_.get()); +} - ASSERT_THAT(connect(sockfd, reinterpret_cast<sockaddr*>(&baddr), addrlen), - SyscallSucceeds()); - // Now the socket is bound to the loopback address. +TEST_P(UdpSocketTest, DisconnectAfterConnectAnyWithPort) { + ASSERT_NO_ERRNO(BindAny()); + EXPECT_THAT(connect(sock_.get(), bind_addr_, addrlen_), SyscallSucceeds()); - // Disconnect - addrlen = sizeof(addr); - addr.ss_family = AF_UNSPEC; - ASSERT_THAT(connect(sockfd, reinterpret_cast<sockaddr*>(&addr), addrlen), - SyscallSucceeds()); + struct sockaddr_storage addr; + socklen_t addrlen = sizeof(addr); + EXPECT_THAT( + getpeername(sock_.get(), reinterpret_cast<sockaddr*>(&addr), &addrlen), + SyscallSucceeds()); - // Check that after disconnect the socket is bound to the ANY address. - EXPECT_THAT(getsockname(sockfd, reinterpret_cast<sockaddr*>(&addr), &addrlen), - SyscallSucceeds()); - if (family == AddressFamily::kIpv4) { - auto addr_out = reinterpret_cast<struct sockaddr_in*>(&addr); - EXPECT_EQ(addrlen, sizeof(*addr_out)); - EXPECT_EQ(addr_out->sin_addr.s_addr, htonl(INADDR_ANY)); - } else { - auto addr_out = reinterpret_cast<struct sockaddr_in6*>(&addr); - EXPECT_EQ(addrlen, sizeof(*addr_out)); - struct in6_addr loopback = IN6ADDR_ANY_INIT; + EXPECT_EQ(addrlen, addrlen_); + EXPECT_EQ(*Port(&bind_addr_storage_), *Port(&addr)); - EXPECT_EQ(memcmp(&addr_out->sin6_addr, &loopback, sizeof(in6_addr)), 0); - } + Disconnect(sock_.get()); } -TEST_P(UdpSocketTest, DisconnectAfterConnectAny) { - DisconnectAfterConnectAny(GetParam(), s_, 0); -} +TEST_P(UdpSocketTest, DisconnectAfterBind) { + ASSERT_NO_ERRNO(BindLoopback()); -TEST_P(UdpSocketTest, DisconnectAfterConnectAnyWithPort) { - auto port = *Port(reinterpret_cast<struct sockaddr_storage*>(addr_[1])); - DisconnectAfterConnectAny(GetParam(), s_, port); -} + // Bind to the next port above bind_. + struct sockaddr_storage addr_storage = InetLoopbackAddr(); + struct sockaddr* addr = reinterpret_cast<struct sockaddr*>(&addr_storage); + SetPort(&addr_storage, *Port(&bind_addr_storage_) + 1); + ASSERT_NO_ERRNO(BindSocket(sock_.get(), addr)); -TEST_P(UdpSocketTest, DisconnectAfterBind) { - ASSERT_THAT(bind(s_, addr_[1], addrlen_), SyscallSucceeds()); // Connect the socket. - ASSERT_THAT(connect(s_, addr_[0], addrlen_), SyscallSucceeds()); + ASSERT_THAT(connect(sock_.get(), bind_addr_, addrlen_), SyscallSucceeds()); - struct sockaddr_storage addr = {}; - addr.ss_family = AF_UNSPEC; - EXPECT_THAT( - connect(s_, reinterpret_cast<sockaddr*>(&addr), sizeof(addr.ss_family)), - SyscallSucceeds()); + struct sockaddr_storage unspec = {}; + unspec.ss_family = AF_UNSPEC; + EXPECT_THAT(connect(sock_.get(), reinterpret_cast<sockaddr*>(&unspec), + sizeof(unspec.ss_family)), + SyscallSucceeds()); // Check that we're still bound. - socklen_t addrlen = sizeof(addr); - EXPECT_THAT(getsockname(s_, reinterpret_cast<sockaddr*>(&addr), &addrlen), - SyscallSucceeds()); + socklen_t addrlen = sizeof(unspec); + EXPECT_THAT( + getsockname(sock_.get(), reinterpret_cast<sockaddr*>(&unspec), &addrlen), + SyscallSucceeds()); EXPECT_EQ(addrlen, addrlen_); - EXPECT_EQ(memcmp(&addr, addr_[1], addrlen_), 0); + EXPECT_EQ(memcmp(addr, &unspec, addrlen_), 0); addrlen = sizeof(addr); - EXPECT_THAT(getpeername(s_, reinterpret_cast<sockaddr*>(&addr), &addrlen), + EXPECT_THAT(getpeername(sock_.get(), addr, &addrlen), SyscallFailsWithErrno(ENOTCONN)); } TEST_P(UdpSocketTest, BindToAnyConnnectToLocalhost) { - struct sockaddr_storage baddr = {}; - auto port = *Port(reinterpret_cast<struct sockaddr_storage*>(addr_[1])); - if (GetParam() == AddressFamily::kIpv4) { - auto addr_in = reinterpret_cast<struct sockaddr_in*>(&baddr); - addr_in->sin_family = AF_INET; - addr_in->sin_port = port; - addr_in->sin_addr.s_addr = htonl(INADDR_ANY); - } else { - auto addr_in = reinterpret_cast<struct sockaddr_in6*>(&baddr); - addr_in->sin6_family = AF_INET6; - addr_in->sin6_port = port; - addr_in->sin6_scope_id = 0; - addr_in->sin6_addr = IN6ADDR_ANY_INIT; - } - ASSERT_THAT(bind(s_, reinterpret_cast<sockaddr*>(&baddr), addrlen_), - SyscallSucceeds()); - // Connect the socket. - ASSERT_THAT(connect(s_, addr_[0], addrlen_), SyscallSucceeds()); + ASSERT_NO_ERRNO(BindAny()); - struct sockaddr_storage addr = {}; + struct sockaddr_storage addr_storage = InetLoopbackAddr(); + struct sockaddr* addr = reinterpret_cast<struct sockaddr*>(&addr_storage); + SetPort(&addr_storage, *Port(&bind_addr_storage_) + 1); socklen_t addrlen = sizeof(addr); - EXPECT_THAT(getsockname(s_, reinterpret_cast<sockaddr*>(&addr), &addrlen), - SyscallSucceeds()); + + // Connect the socket. + ASSERT_THAT(connect(bind_.get(), addr, addrlen_), SyscallSucceeds()); + + EXPECT_THAT(getsockname(bind_.get(), addr, &addrlen), SyscallSucceeds()); // If the socket is bound to ANY and connected to a loopback address, // getsockname() has to return the loopback address. if (GetParam() == AddressFamily::kIpv4) { - auto addr_out = reinterpret_cast<struct sockaddr_in*>(&addr); + auto addr_out = reinterpret_cast<struct sockaddr_in*>(addr); EXPECT_EQ(addrlen, sizeof(*addr_out)); EXPECT_EQ(addr_out->sin_addr.s_addr, htonl(INADDR_LOOPBACK)); } else { - auto addr_out = reinterpret_cast<struct sockaddr_in6*>(&addr); + auto addr_out = reinterpret_cast<struct sockaddr_in6*>(addr); struct in6_addr loopback = IN6ADDR_LOOPBACK_INIT; EXPECT_EQ(addrlen, sizeof(*addr_out)); EXPECT_EQ(memcmp(&addr_out->sin6_addr, &loopback, sizeof(in6_addr)), 0); @@ -572,91 +532,97 @@ TEST_P(UdpSocketTest, BindToAnyConnnectToLocalhost) { } TEST_P(UdpSocketTest, DisconnectAfterBindToAny) { - struct sockaddr_storage baddr = {}; - socklen_t addrlen; - auto port = *Port(reinterpret_cast<struct sockaddr_storage*>(addr_[1])); - if (GetParam() == AddressFamily::kIpv4) { - auto addr_in = reinterpret_cast<struct sockaddr_in*>(&baddr); - addr_in->sin_family = AF_INET; - addr_in->sin_port = port; - addr_in->sin_addr.s_addr = htonl(INADDR_ANY); - } else { - auto addr_in = reinterpret_cast<struct sockaddr_in6*>(&baddr); - addr_in->sin6_family = AF_INET6; - addr_in->sin6_port = port; - addr_in->sin6_scope_id = 0; - addr_in->sin6_addr = IN6ADDR_ANY_INIT; - } - ASSERT_THAT(bind(s_, reinterpret_cast<sockaddr*>(&baddr), addrlen_), - SyscallSucceeds()); + ASSERT_NO_ERRNO(BindLoopback()); + + struct sockaddr_storage any_storage = InetAnyAddr(); + struct sockaddr* any = reinterpret_cast<struct sockaddr*>(&any_storage); + SetPort(&any_storage, *Port(&bind_addr_storage_) + 1); + + ASSERT_NO_ERRNO(BindSocket(sock_.get(), any)); + // Connect the socket. - ASSERT_THAT(connect(s_, addr_[0], addrlen_), SyscallSucceeds()); + ASSERT_THAT(connect(sock_.get(), bind_addr_, addrlen_), SyscallSucceeds()); - struct sockaddr_storage addr = {}; - addr.ss_family = AF_UNSPEC; - EXPECT_THAT( - connect(s_, reinterpret_cast<sockaddr*>(&addr), sizeof(addr.ss_family)), - SyscallSucceeds()); + Disconnect(sock_.get()); // Check that we're still bound. - addrlen = sizeof(addr); - EXPECT_THAT(getsockname(s_, reinterpret_cast<sockaddr*>(&addr), &addrlen), - SyscallSucceeds()); + struct sockaddr_storage addr; + socklen_t addrlen = sizeof(addr); + EXPECT_THAT( + getsockname(sock_.get(), reinterpret_cast<sockaddr*>(&addr), &addrlen), + SyscallSucceeds()); EXPECT_EQ(addrlen, addrlen_); - EXPECT_EQ(memcmp(&addr, &baddr, addrlen), 0); + EXPECT_EQ(memcmp(&addr, any, addrlen), 0); addrlen = sizeof(addr); - EXPECT_THAT(getpeername(s_, reinterpret_cast<sockaddr*>(&addr), &addrlen), - SyscallFailsWithErrno(ENOTCONN)); + EXPECT_THAT( + getpeername(sock_.get(), reinterpret_cast<sockaddr*>(&addr), &addrlen), + SyscallFailsWithErrno(ENOTCONN)); } TEST_P(UdpSocketTest, Disconnect) { + ASSERT_NO_ERRNO(BindLoopback()); + + struct sockaddr_storage any_storage = InetAnyAddr(); + struct sockaddr* any = reinterpret_cast<struct sockaddr*>(&any_storage); + SetPort(&any_storage, *Port(&bind_addr_storage_) + 1); + ASSERT_NO_ERRNO(BindSocket(sock_.get(), any)); + for (int i = 0; i < 2; i++) { // Try to connect again. - EXPECT_THAT(connect(s_, addr_[2], addrlen_), SyscallSucceeds()); + EXPECT_THAT(connect(sock_.get(), bind_addr_, addrlen_), SyscallSucceeds()); // Check that we're connected to the right peer. struct sockaddr_storage peer; socklen_t peerlen = sizeof(peer); - EXPECT_THAT(getpeername(s_, reinterpret_cast<sockaddr*>(&peer), &peerlen), - SyscallSucceeds()); + EXPECT_THAT( + getpeername(sock_.get(), reinterpret_cast<sockaddr*>(&peer), &peerlen), + SyscallSucceeds()); EXPECT_EQ(peerlen, addrlen_); - EXPECT_EQ(memcmp(&peer, addr_[2], addrlen_), 0); + EXPECT_EQ(memcmp(&peer, bind_addr_, addrlen_), 0); // Try to disconnect. struct sockaddr_storage addr = {}; addr.ss_family = AF_UNSPEC; - EXPECT_THAT( - connect(s_, reinterpret_cast<sockaddr*>(&addr), sizeof(addr.ss_family)), - SyscallSucceeds()); + EXPECT_THAT(connect(sock_.get(), reinterpret_cast<sockaddr*>(&addr), + sizeof(addr.ss_family)), + SyscallSucceeds()); peerlen = sizeof(peer); - EXPECT_THAT(getpeername(s_, reinterpret_cast<sockaddr*>(&peer), &peerlen), - SyscallFailsWithErrno(ENOTCONN)); + EXPECT_THAT( + getpeername(sock_.get(), reinterpret_cast<sockaddr*>(&peer), &peerlen), + SyscallFailsWithErrno(ENOTCONN)); // Check that we're still bound. socklen_t addrlen = sizeof(addr); - EXPECT_THAT(getsockname(s_, reinterpret_cast<sockaddr*>(&addr), &addrlen), - SyscallSucceeds()); + EXPECT_THAT( + getsockname(sock_.get(), reinterpret_cast<sockaddr*>(&addr), &addrlen), + SyscallSucceeds()); EXPECT_EQ(addrlen, addrlen_); - EXPECT_EQ(*Port(&addr), 0); + EXPECT_EQ(*Port(&addr), *Port(&any_storage)); } } TEST_P(UdpSocketTest, ConnectBadAddress) { struct sockaddr addr = {}; - addr.sa_family = addr_[0]->sa_family; - ASSERT_THAT(connect(s_, &addr, sizeof(addr.sa_family)), + addr.sa_family = GetFamily(); + ASSERT_THAT(connect(sock_.get(), &addr, sizeof(addr.sa_family)), SyscallFailsWithErrno(EINVAL)); } TEST_P(UdpSocketTest, SendToAddressOtherThanConnected) { - ASSERT_THAT(connect(s_, addr_[0], addrlen_), SyscallSucceeds()); + ASSERT_NO_ERRNO(BindLoopback()); + + struct sockaddr_storage addr_storage = InetAnyAddr(); + struct sockaddr* addr = reinterpret_cast<struct sockaddr*>(&addr_storage); + SetPort(&addr_storage, *Port(&bind_addr_storage_) + 1); + + ASSERT_THAT(connect(sock_.get(), bind_addr_, addrlen_), SyscallSucceeds()); // Send to a different destination than we're connected to. char buf[512]; - EXPECT_THAT(sendto(s_, buf, sizeof(buf), 0, addr_[1], addrlen_), + EXPECT_THAT(sendto(sock_.get(), buf, sizeof(buf), 0, addr, addrlen_), SyscallSucceedsWithValue(sizeof(buf))); } @@ -664,24 +630,27 @@ TEST_P(UdpSocketTest, ZerolengthWriteAllowed) { // TODO(gvisor.dev/issue/1202): Hostinet does not support zero length writes. SKIP_IF(IsRunningWithHostinet()); - // Bind s_ to loopback:TestPort, and connect to loopback:TestPort+1. - ASSERT_THAT(bind(s_, addr_[0], addrlen_), SyscallSucceeds()); - ASSERT_THAT(connect(s_, addr_[1], addrlen_), SyscallSucceeds()); + ASSERT_NO_ERRNO(BindLoopback()); + // Connect to loopback:bind_addr_+1. + struct sockaddr_storage addr_storage = InetLoopbackAddr(); + struct sockaddr* addr = reinterpret_cast<struct sockaddr*>(&addr_storage); + SetPort(&addr_storage, *Port(&bind_addr_storage_) + 1); + ASSERT_THAT(connect(bind_.get(), addr, addrlen_), SyscallSucceeds()); - // Bind t_ to loopback:TestPort+1. - ASSERT_THAT(bind(t_, addr_[1], addrlen_), SyscallSucceeds()); + // Bind sock to loopback:bind_addr_+1. + ASSERT_THAT(bind(sock_.get(), addr, addrlen_), SyscallSucceeds()); char buf[3]; - // Send zero length packet from s_ to t_. - ASSERT_THAT(write(s_, buf, 0), SyscallSucceedsWithValue(0)); + // Send zero length packet from bind_ to sock_. + ASSERT_THAT(write(bind_.get(), buf, 0), SyscallSucceedsWithValue(0)); - struct pollfd pfd = {t_, POLLIN, 0}; - ASSERT_THAT(RetryEINTR(poll)(&pfd, 1, /*timeout=*/1000), + struct pollfd pfd = {sock_.get(), POLLIN, 0}; + ASSERT_THAT(RetryEINTR(poll)(&pfd, 1, /*timeout*/ 1000), SyscallSucceedsWithValue(1)); // Receive the packet. char received[3]; - EXPECT_THAT(read(t_, received, sizeof(received)), + EXPECT_THAT(read(sock_.get(), received, sizeof(received)), SyscallSucceedsWithValue(0)); } @@ -689,216 +658,252 @@ TEST_P(UdpSocketTest, ZerolengthWriteAllowedNonBlockRead) { // TODO(gvisor.dev/issue/1202): Hostinet does not support zero length writes. SKIP_IF(IsRunningWithHostinet()); - // Bind s_ to loopback:TestPort, and connect to loopback:TestPort+1. - ASSERT_THAT(bind(s_, addr_[0], addrlen_), SyscallSucceeds()); - ASSERT_THAT(connect(s_, addr_[1], addrlen_), SyscallSucceeds()); + ASSERT_NO_ERRNO(BindLoopback()); - // Bind t_ to loopback:TestPort+1. - ASSERT_THAT(bind(t_, addr_[1], addrlen_), SyscallSucceeds()); + // Connect to loopback:bind_addr_port+1. + struct sockaddr_storage addr_storage = InetLoopbackAddr(); + struct sockaddr* addr = reinterpret_cast<struct sockaddr*>(&addr_storage); + SetPort(&addr_storage, *Port(&bind_addr_storage_) + 1); + ASSERT_THAT(connect(bind_.get(), addr, addrlen_), SyscallSucceeds()); - // Set t_ to non-blocking. + // Bind sock to loopback:bind_addr_port+1. + ASSERT_THAT(bind(sock_.get(), addr, addrlen_), SyscallSucceeds()); + + // Set sock to non-blocking. int opts = 0; - ASSERT_THAT(opts = fcntl(t_, F_GETFL), SyscallSucceeds()); - ASSERT_THAT(fcntl(t_, F_SETFL, opts | O_NONBLOCK), SyscallSucceeds()); + ASSERT_THAT(opts = fcntl(sock_.get(), F_GETFL), SyscallSucceeds()); + ASSERT_THAT(fcntl(sock_.get(), F_SETFL, opts | O_NONBLOCK), + SyscallSucceeds()); char buf[3]; - // Send zero length packet from s_ to t_. - ASSERT_THAT(write(s_, buf, 0), SyscallSucceedsWithValue(0)); + // Send zero length packet from bind_ to sock_. + ASSERT_THAT(write(bind_.get(), buf, 0), SyscallSucceedsWithValue(0)); - struct pollfd pfd = {t_, POLLIN, 0}; + struct pollfd pfd = {sock_.get(), POLLIN, 0}; ASSERT_THAT(RetryEINTR(poll)(&pfd, 1, /*timeout=*/1000), SyscallSucceedsWithValue(1)); // Receive the packet. char received[3]; - EXPECT_THAT(read(t_, received, sizeof(received)), + EXPECT_THAT(read(sock_.get(), received, sizeof(received)), SyscallSucceedsWithValue(0)); - EXPECT_THAT(read(t_, received, sizeof(received)), + EXPECT_THAT(read(sock_.get(), received, sizeof(received)), SyscallFailsWithErrno(EAGAIN)); } TEST_P(UdpSocketTest, SendAndReceiveNotConnected) { - // Bind s_ to loopback. - ASSERT_THAT(bind(s_, addr_[0], addrlen_), SyscallSucceeds()); + ASSERT_NO_ERRNO(BindLoopback()); - // Send some data to s_. + // Send some data to bind_. char buf[512]; RandomizeBuffer(buf, sizeof(buf)); - ASSERT_THAT(sendto(t_, buf, sizeof(buf), 0, addr_[0], addrlen_), + ASSERT_THAT(sendto(sock_.get(), buf, sizeof(buf), 0, bind_addr_, addrlen_), SyscallSucceedsWithValue(sizeof(buf))); // Receive the data. char received[sizeof(buf)]; - EXPECT_THAT(recv(s_, received, sizeof(received), 0), + EXPECT_THAT(recv(bind_.get(), received, sizeof(received), 0), SyscallSucceedsWithValue(sizeof(received))); EXPECT_EQ(memcmp(buf, received, sizeof(buf)), 0); } TEST_P(UdpSocketTest, SendAndReceiveConnected) { - // Bind s_ to loopback:TestPort, and connect to loopback:TestPort+1. - ASSERT_THAT(bind(s_, addr_[0], addrlen_), SyscallSucceeds()); - ASSERT_THAT(connect(s_, addr_[1], addrlen_), SyscallSucceeds()); + ASSERT_NO_ERRNO(BindLoopback()); - // Bind t_ to loopback:TestPort+1. - ASSERT_THAT(bind(t_, addr_[1], addrlen_), SyscallSucceeds()); + // Connect to loopback:bind_addr_port+1. + struct sockaddr_storage addr_storage = InetLoopbackAddr(); + struct sockaddr* addr = reinterpret_cast<struct sockaddr*>(&addr_storage); + SetPort(&addr_storage, *Port(&bind_addr_storage_) + 1); + ASSERT_THAT(connect(bind_.get(), addr, addrlen_), SyscallSucceeds()); - // Send some data from t_ to s_. + // Bind sock to loopback:TestPort+1. + ASSERT_THAT(bind(sock_.get(), addr, addrlen_), SyscallSucceeds()); + + // Send some data from sock to bind_. char buf[512]; RandomizeBuffer(buf, sizeof(buf)); - ASSERT_THAT(sendto(t_, buf, sizeof(buf), 0, addr_[0], addrlen_), + ASSERT_THAT(sendto(sock_.get(), buf, sizeof(buf), 0, bind_addr_, addrlen_), SyscallSucceedsWithValue(sizeof(buf))); // Receive the data. char received[sizeof(buf)]; - EXPECT_THAT(recv(s_, received, sizeof(received), 0), + EXPECT_THAT(recv(bind_.get(), received, sizeof(received), 0), SyscallSucceedsWithValue(sizeof(received))); EXPECT_EQ(memcmp(buf, received, sizeof(buf)), 0); } TEST_P(UdpSocketTest, ReceiveFromNotConnected) { - // Bind s_ to loopback:TestPort, and connect to loopback:TestPort+1. - ASSERT_THAT(bind(s_, addr_[0], addrlen_), SyscallSucceeds()); - ASSERT_THAT(connect(s_, addr_[1], addrlen_), SyscallSucceeds()); + ASSERT_NO_ERRNO(BindLoopback()); + + // Connect to loopback:bind_addr_port+1. + struct sockaddr_storage addr_storage = InetLoopbackAddr(); + struct sockaddr* addr = reinterpret_cast<struct sockaddr*>(&addr_storage); + SetPort(&addr_storage, *Port(&bind_addr_storage_) + 1); + ASSERT_THAT(connect(bind_.get(), addr, addrlen_), SyscallSucceeds()); - // Bind t_ to loopback:TestPort+2. - ASSERT_THAT(bind(t_, addr_[2], addrlen_), SyscallSucceeds()); + // Bind sock to loopback:bind_addr_port+2. + struct sockaddr_storage addr2_storage = InetLoopbackAddr(); + struct sockaddr* addr2 = reinterpret_cast<struct sockaddr*>(&addr2_storage); + SetPort(&addr2_storage, *Port(&bind_addr_storage_) + 2); + ASSERT_THAT(bind(sock_.get(), addr2, addrlen_), SyscallSucceeds()); - // Send some data from t_ to s_. + // Send some data from sock to bind_. char buf[512]; - ASSERT_THAT(sendto(t_, buf, sizeof(buf), 0, addr_[0], addrlen_), + ASSERT_THAT(sendto(sock_.get(), buf, sizeof(buf), 0, bind_addr_, addrlen_), SyscallSucceedsWithValue(sizeof(buf))); - // Check that the data isn't_ received because it was sent from a different + // Check that the data isn't received because it was sent from a different // address than we're connected. - EXPECT_THAT(recv(s_, buf, sizeof(buf), MSG_DONTWAIT), + EXPECT_THAT(recv(sock_.get(), buf, sizeof(buf), MSG_DONTWAIT), SyscallFailsWithErrno(EWOULDBLOCK)); } TEST_P(UdpSocketTest, ReceiveBeforeConnect) { - // Bind s_ to loopback:TestPort. - ASSERT_THAT(bind(s_, addr_[0], addrlen_), SyscallSucceeds()); + ASSERT_NO_ERRNO(BindLoopback()); - // Bind t_ to loopback:TestPort+2. - ASSERT_THAT(bind(t_, addr_[2], addrlen_), SyscallSucceeds()); + // Bind sock to loopback:bind_addr_port+2. + struct sockaddr_storage addr2_storage = InetLoopbackAddr(); + struct sockaddr* addr2 = reinterpret_cast<struct sockaddr*>(&addr2_storage); + SetPort(&addr2_storage, *Port(&bind_addr_storage_) + 2); + ASSERT_THAT(bind(sock_.get(), addr2, addrlen_), SyscallSucceeds()); - // Send some data from t_ to s_. + // Send some data from sock to bind_. char buf[512]; RandomizeBuffer(buf, sizeof(buf)); - ASSERT_THAT(sendto(t_, buf, sizeof(buf), 0, addr_[0], addrlen_), + ASSERT_THAT(sendto(sock_.get(), buf, sizeof(buf), 0, bind_addr_, addrlen_), SyscallSucceedsWithValue(sizeof(buf))); // Connect to loopback:TestPort+1. - ASSERT_THAT(connect(s_, addr_[1], addrlen_), SyscallSucceeds()); + struct sockaddr_storage addr_storage = InetLoopbackAddr(); + struct sockaddr* addr = reinterpret_cast<struct sockaddr*>(&addr_storage); + SetPort(&addr_storage, *Port(&bind_addr_storage_) + 1); + ASSERT_THAT(connect(bind_.get(), addr, addrlen_), SyscallSucceeds()); // Receive the data. It works because it was sent before the connect. char received[sizeof(buf)]; - EXPECT_THAT(recv(s_, received, sizeof(received), 0), + EXPECT_THAT(recv(bind_.get(), received, sizeof(received), 0), SyscallSucceedsWithValue(sizeof(received))); EXPECT_EQ(memcmp(buf, received, sizeof(buf)), 0); // Send again. This time it should not be received. - ASSERT_THAT(sendto(t_, buf, sizeof(buf), 0, addr_[0], addrlen_), + ASSERT_THAT(sendto(sock_.get(), buf, sizeof(buf), 0, bind_addr_, addrlen_), SyscallSucceedsWithValue(sizeof(buf))); - EXPECT_THAT(recv(s_, buf, sizeof(buf), MSG_DONTWAIT), + EXPECT_THAT(recv(bind_.get(), buf, sizeof(buf), MSG_DONTWAIT), SyscallFailsWithErrno(EWOULDBLOCK)); } TEST_P(UdpSocketTest, ReceiveFrom) { - // Bind s_ to loopback:TestPort, and connect to loopback:TestPort+1. - ASSERT_THAT(bind(s_, addr_[0], addrlen_), SyscallSucceeds()); - ASSERT_THAT(connect(s_, addr_[1], addrlen_), SyscallSucceeds()); + ASSERT_NO_ERRNO(BindLoopback()); + + // Connect to loopback:bind_addr_port+1. + struct sockaddr_storage addr_storage = InetLoopbackAddr(); + struct sockaddr* addr = reinterpret_cast<struct sockaddr*>(&addr_storage); + SetPort(&addr_storage, *Port(&bind_addr_storage_) + 1); + ASSERT_THAT(connect(bind_.get(), addr, addrlen_), SyscallSucceeds()); - // Bind t_ to loopback:TestPort+1. - ASSERT_THAT(bind(t_, addr_[1], addrlen_), SyscallSucceeds()); + // Bind sock to loopback:TestPort+1. + ASSERT_THAT(bind(sock_.get(), addr, addrlen_), SyscallSucceeds()); - // Send some data from t_ to s_. + // Send some data from sock to bind_. char buf[512]; RandomizeBuffer(buf, sizeof(buf)); - ASSERT_THAT(sendto(t_, buf, sizeof(buf), 0, addr_[0], addrlen_), + ASSERT_THAT(sendto(sock_.get(), buf, sizeof(buf), 0, bind_addr_, addrlen_), SyscallSucceedsWithValue(sizeof(buf))); // Receive the data and sender address. char received[sizeof(buf)]; - struct sockaddr_storage addr; - socklen_t addrlen = sizeof(addr); - EXPECT_THAT(recvfrom(s_, received, sizeof(received), 0, - reinterpret_cast<sockaddr*>(&addr), &addrlen), + struct sockaddr_storage addr2; + socklen_t addr2len = sizeof(addr2); + EXPECT_THAT(recvfrom(bind_.get(), received, sizeof(received), 0, + reinterpret_cast<sockaddr*>(&addr2), &addr2len), SyscallSucceedsWithValue(sizeof(received))); EXPECT_EQ(memcmp(buf, received, sizeof(buf)), 0); - EXPECT_EQ(addrlen, addrlen_); - EXPECT_EQ(memcmp(&addr, addr_[1], addrlen_), 0); + EXPECT_EQ(addr2len, addrlen_); + EXPECT_EQ(memcmp(addr, &addr2, addrlen_), 0); } TEST_P(UdpSocketTest, Listen) { - ASSERT_THAT(listen(s_, SOMAXCONN), SyscallFailsWithErrno(EOPNOTSUPP)); + ASSERT_THAT(listen(sock_.get(), SOMAXCONN), + SyscallFailsWithErrno(EOPNOTSUPP)); } TEST_P(UdpSocketTest, Accept) { - ASSERT_THAT(accept(s_, nullptr, nullptr), SyscallFailsWithErrno(EOPNOTSUPP)); + ASSERT_THAT(accept(sock_.get(), nullptr, nullptr), + SyscallFailsWithErrno(EOPNOTSUPP)); } // This test validates that a read shutdown with pending data allows the read // to proceed with the data before returning EAGAIN. TEST_P(UdpSocketTest, ReadShutdownNonblockPendingData) { - char received[512]; + ASSERT_NO_ERRNO(BindLoopback()); - // Bind t_ to loopback:TestPort+2. - ASSERT_THAT(bind(t_, addr_[2], addrlen_), SyscallSucceeds()); - ASSERT_THAT(connect(t_, addr_[1], addrlen_), SyscallSucceeds()); + // Connect to loopback:bind_addr_port+1. + struct sockaddr_storage addr_storage = InetLoopbackAddr(); + struct sockaddr* addr = reinterpret_cast<struct sockaddr*>(&addr_storage); + SetPort(&addr_storage, *Port(&bind_addr_storage_) + 1); + ASSERT_THAT(connect(bind_.get(), addr, addrlen_), SyscallSucceeds()); - // Connect the socket, then try to shutdown again. - ASSERT_THAT(bind(s_, addr_[1], addrlen_), SyscallSucceeds()); - ASSERT_THAT(connect(s_, addr_[2], addrlen_), SyscallSucceeds()); + // Bind to loopback:bind_addr_port+1 and connect to bind_addr_. + ASSERT_THAT(bind(sock_.get(), addr, addrlen_), SyscallSucceeds()); + ASSERT_THAT(connect(sock_.get(), bind_addr_, addrlen_), SyscallSucceeds()); // Verify that we get EWOULDBLOCK when there is nothing to read. - EXPECT_THAT(recv(s_, received, sizeof(received), MSG_DONTWAIT), + char received[512]; + EXPECT_THAT(recv(bind_.get(), received, sizeof(received), MSG_DONTWAIT), SyscallFailsWithErrno(EWOULDBLOCK)); const char* buf = "abc"; - EXPECT_THAT(write(t_, buf, 3), SyscallSucceedsWithValue(3)); + EXPECT_THAT(write(sock_.get(), buf, 3), SyscallSucceedsWithValue(3)); int opts = 0; - ASSERT_THAT(opts = fcntl(s_, F_GETFL), SyscallSucceeds()); - ASSERT_THAT(fcntl(s_, F_SETFL, opts | O_NONBLOCK), SyscallSucceeds()); - ASSERT_THAT(opts = fcntl(s_, F_GETFL), SyscallSucceeds()); + ASSERT_THAT(opts = fcntl(bind_.get(), F_GETFL), SyscallSucceeds()); + ASSERT_THAT(fcntl(bind_.get(), F_SETFL, opts | O_NONBLOCK), + SyscallSucceeds()); + ASSERT_THAT(opts = fcntl(bind_.get(), F_GETFL), SyscallSucceeds()); ASSERT_NE(opts & O_NONBLOCK, 0); - EXPECT_THAT(shutdown(s_, SHUT_RD), SyscallSucceeds()); + EXPECT_THAT(shutdown(bind_.get(), SHUT_RD), SyscallSucceeds()); - struct pollfd pfd = {s_, POLLIN, 0}; + struct pollfd pfd = {bind_.get(), POLLIN, 0}; ASSERT_THAT(RetryEINTR(poll)(&pfd, 1, /*timeout=*/1000), SyscallSucceedsWithValue(1)); // We should get the data even though read has been shutdown. - EXPECT_THAT(recv(s_, received, 2, 0), SyscallSucceedsWithValue(2)); + EXPECT_THAT(recv(bind_.get(), received, 2, 0), SyscallSucceedsWithValue(2)); // Because we read less than the entire packet length, since it's a packet // based socket any subsequent reads should return EWOULDBLOCK. - EXPECT_THAT(recv(s_, received, 1, 0), SyscallFailsWithErrno(EWOULDBLOCK)); + EXPECT_THAT(recv(bind_.get(), received, 1, 0), + SyscallFailsWithErrno(EWOULDBLOCK)); } // This test is validating that even after a socket is shutdown if it's // reconnected it will reset the shutdown state. TEST_P(UdpSocketTest, ReadShutdownSameSocketResetsShutdownState) { char received[512]; - EXPECT_THAT(recv(s_, received, sizeof(received), MSG_DONTWAIT), + EXPECT_THAT(recv(bind_.get(), received, sizeof(received), MSG_DONTWAIT), SyscallFailsWithErrno(EWOULDBLOCK)); - EXPECT_THAT(shutdown(s_, SHUT_RD), SyscallFailsWithErrno(ENOTCONN)); + EXPECT_THAT(shutdown(bind_.get(), SHUT_RD), SyscallFailsWithErrno(ENOTCONN)); - EXPECT_THAT(recv(s_, received, sizeof(received), MSG_DONTWAIT), + EXPECT_THAT(recv(bind_.get(), received, sizeof(received), MSG_DONTWAIT), SyscallFailsWithErrno(EWOULDBLOCK)); // Connect the socket, then try to shutdown again. - ASSERT_THAT(bind(s_, addr_[1], addrlen_), SyscallSucceeds()); - ASSERT_THAT(connect(s_, addr_[2], addrlen_), SyscallSucceeds()); + ASSERT_NO_ERRNO(BindLoopback()); + + // Connect to loopback:bind_addr_port+1. + struct sockaddr_storage addr_storage = InetLoopbackAddr(); + struct sockaddr* addr = reinterpret_cast<struct sockaddr*>(&addr_storage); + SetPort(&addr_storage, *Port(&bind_addr_storage_) + 1); + ASSERT_THAT(connect(bind_.get(), addr, addrlen_), SyscallSucceeds()); - EXPECT_THAT(recv(s_, received, sizeof(received), MSG_DONTWAIT), + EXPECT_THAT(recv(bind_.get(), received, sizeof(received), MSG_DONTWAIT), SyscallFailsWithErrno(EWOULDBLOCK)); } @@ -907,24 +912,26 @@ TEST_P(UdpSocketTest, ReadShutdown) { // MSG_DONTWAIT blocks indefinitely. SKIP_IF(IsRunningWithHostinet()); + ASSERT_NO_ERRNO(BindLoopback()); + char received[512]; - EXPECT_THAT(recv(s_, received, sizeof(received), MSG_DONTWAIT), + EXPECT_THAT(recv(sock_.get(), received, sizeof(received), MSG_DONTWAIT), SyscallFailsWithErrno(EWOULDBLOCK)); - EXPECT_THAT(shutdown(s_, SHUT_RD), SyscallFailsWithErrno(ENOTCONN)); + EXPECT_THAT(shutdown(sock_.get(), SHUT_RD), SyscallFailsWithErrno(ENOTCONN)); - EXPECT_THAT(recv(s_, received, sizeof(received), MSG_DONTWAIT), + EXPECT_THAT(recv(sock_.get(), received, sizeof(received), MSG_DONTWAIT), SyscallFailsWithErrno(EWOULDBLOCK)); // Connect the socket, then try to shutdown again. - ASSERT_THAT(connect(s_, addr_[0], addrlen_), SyscallSucceeds()); + ASSERT_THAT(connect(sock_.get(), bind_addr_, addrlen_), SyscallSucceeds()); - EXPECT_THAT(recv(s_, received, sizeof(received), MSG_DONTWAIT), + EXPECT_THAT(recv(sock_.get(), received, sizeof(received), MSG_DONTWAIT), SyscallFailsWithErrno(EWOULDBLOCK)); - EXPECT_THAT(shutdown(s_, SHUT_RD), SyscallSucceeds()); + EXPECT_THAT(shutdown(sock_.get(), SHUT_RD), SyscallSucceeds()); - EXPECT_THAT(recv(s_, received, sizeof(received), 0), + EXPECT_THAT(recv(sock_.get(), received, sizeof(received), 0), SyscallSucceedsWithValue(0)); } @@ -932,93 +939,94 @@ TEST_P(UdpSocketTest, ReadShutdownDifferentThread) { // TODO(gvisor.dev/issue/1202): Calling recv() after shutdown without // MSG_DONTWAIT blocks indefinitely. SKIP_IF(IsRunningWithHostinet()); + ASSERT_NO_ERRNO(BindLoopback()); char received[512]; - EXPECT_THAT(recv(s_, received, sizeof(received), MSG_DONTWAIT), + EXPECT_THAT(recv(sock_.get(), received, sizeof(received), MSG_DONTWAIT), SyscallFailsWithErrno(EWOULDBLOCK)); // Connect the socket, then shutdown from another thread. - ASSERT_THAT(connect(s_, addr_[0], addrlen_), SyscallSucceeds()); + ASSERT_THAT(connect(sock_.get(), bind_addr_, addrlen_), SyscallSucceeds()); - EXPECT_THAT(recv(s_, received, sizeof(received), MSG_DONTWAIT), + EXPECT_THAT(recv(sock_.get(), received, sizeof(received), MSG_DONTWAIT), SyscallFailsWithErrno(EWOULDBLOCK)); ScopedThread t([&] { absl::SleepFor(absl::Milliseconds(200)); - EXPECT_THAT(shutdown(this->s_, SHUT_RD), SyscallSucceeds()); + EXPECT_THAT(shutdown(sock_.get(), SHUT_RD), SyscallSucceeds()); }); - EXPECT_THAT(RetryEINTR(recv)(s_, received, sizeof(received), 0), + EXPECT_THAT(RetryEINTR(recv)(sock_.get(), received, sizeof(received), 0), SyscallSucceedsWithValue(0)); t.Join(); - EXPECT_THAT(RetryEINTR(recv)(s_, received, sizeof(received), 0), + EXPECT_THAT(RetryEINTR(recv)(sock_.get(), received, sizeof(received), 0), SyscallSucceedsWithValue(0)); } TEST_P(UdpSocketTest, WriteShutdown) { - EXPECT_THAT(shutdown(s_, SHUT_WR), SyscallFailsWithErrno(ENOTCONN)); - ASSERT_THAT(connect(s_, addr_[0], addrlen_), SyscallSucceeds()); - EXPECT_THAT(shutdown(s_, SHUT_WR), SyscallSucceeds()); + ASSERT_NO_ERRNO(BindLoopback()); + EXPECT_THAT(shutdown(sock_.get(), SHUT_WR), SyscallFailsWithErrno(ENOTCONN)); + ASSERT_THAT(connect(sock_.get(), bind_addr_, addrlen_), SyscallSucceeds()); + EXPECT_THAT(shutdown(sock_.get(), SHUT_WR), SyscallSucceeds()); } TEST_P(UdpSocketTest, SynchronousReceive) { - // Bind s_ to loopback. - ASSERT_THAT(bind(s_, addr_[0], addrlen_), SyscallSucceeds()); + ASSERT_NO_ERRNO(BindLoopback()); - // Send some data to s_ from another thread. + // Send some data to bind_ from another thread. char buf[512]; RandomizeBuffer(buf, sizeof(buf)); // Receive the data prior to actually starting the other thread. char received[512]; - EXPECT_THAT(RetryEINTR(recv)(s_, received, sizeof(received), MSG_DONTWAIT), - SyscallFailsWithErrno(EWOULDBLOCK)); + EXPECT_THAT( + RetryEINTR(recv)(bind_.get(), received, sizeof(received), MSG_DONTWAIT), + SyscallFailsWithErrno(EWOULDBLOCK)); // Start the thread. ScopedThread t([&] { absl::SleepFor(absl::Milliseconds(200)); - ASSERT_THAT( - sendto(this->t_, buf, sizeof(buf), 0, this->addr_[0], this->addrlen_), - SyscallSucceedsWithValue(sizeof(buf))); + ASSERT_THAT(sendto(sock_.get(), buf, sizeof(buf), 0, this->bind_addr_, + this->addrlen_), + SyscallSucceedsWithValue(sizeof(buf))); }); - EXPECT_THAT(RetryEINTR(recv)(s_, received, sizeof(received), 0), + EXPECT_THAT(RetryEINTR(recv)(bind_.get(), received, sizeof(received), 0), SyscallSucceedsWithValue(512)); EXPECT_EQ(memcmp(buf, received, sizeof(buf)), 0); } TEST_P(UdpSocketTest, BoundaryPreserved_SendRecv) { - // Bind s_ to loopback:TestPort. - ASSERT_THAT(bind(s_, addr_[0], addrlen_), SyscallSucceeds()); + ASSERT_NO_ERRNO(BindLoopback()); - // Send 3 packets from t_ to s_. + // Send 3 packets from sock to bind_. constexpr int psize = 100; char buf[3 * psize]; RandomizeBuffer(buf, sizeof(buf)); for (int i = 0; i < 3; ++i) { - ASSERT_THAT(sendto(t_, buf + i * psize, psize, 0, addr_[0], addrlen_), - SyscallSucceedsWithValue(psize)); + ASSERT_THAT( + sendto(sock_.get(), buf + i * psize, psize, 0, bind_addr_, addrlen_), + SyscallSucceedsWithValue(psize)); } // Receive the data as 3 separate packets. char received[6 * psize]; for (int i = 0; i < 3; ++i) { - EXPECT_THAT(recv(s_, received + i * psize, 3 * psize, 0), + EXPECT_THAT(recv(bind_.get(), received + i * psize, 3 * psize, 0), SyscallSucceedsWithValue(psize)); } EXPECT_EQ(memcmp(buf, received, 3 * psize), 0); } TEST_P(UdpSocketTest, BoundaryPreserved_WritevReadv) { - // Bind s_ to loopback:TestPort. - ASSERT_THAT(bind(s_, addr_[0], addrlen_), SyscallSucceeds()); + ASSERT_NO_ERRNO(BindLoopback()); - // Direct writes from t_ to s_. - ASSERT_THAT(connect(t_, addr_[0], addrlen_), SyscallSucceeds()); + // Direct writes from sock to bind_. + ASSERT_THAT(connect(sock_.get(), bind_addr_, addrlen_), SyscallSucceeds()); - // Send 2 packets from t_ to s_, where each packet's data consists of 2 - // discontiguous iovecs. + // Send 2 packets from sock to bind_, where each packet's data consists of + // 2 discontiguous iovecs. constexpr size_t kPieceSize = 100; char buf[4 * kPieceSize]; RandomizeBuffer(buf, sizeof(buf)); @@ -1030,7 +1038,8 @@ TEST_P(UdpSocketTest, BoundaryPreserved_WritevReadv) { reinterpret_cast<uintptr_t>(buf) + (i + 2 * j) * kPieceSize); iov[j].iov_len = kPieceSize; } - ASSERT_THAT(writev(t_, iov, 2), SyscallSucceedsWithValue(2 * kPieceSize)); + ASSERT_THAT(writev(sock_.get(), iov, 2), + SyscallSucceedsWithValue(2 * kPieceSize)); } // Receive the data as 2 separate packets. @@ -1042,17 +1051,17 @@ TEST_P(UdpSocketTest, BoundaryPreserved_WritevReadv) { reinterpret_cast<uintptr_t>(received) + (i + 2 * j) * kPieceSize); iov[j].iov_len = kPieceSize; } - ASSERT_THAT(readv(s_, iov, 3), SyscallSucceedsWithValue(2 * kPieceSize)); + ASSERT_THAT(readv(bind_.get(), iov, 3), + SyscallSucceedsWithValue(2 * kPieceSize)); } EXPECT_EQ(memcmp(buf, received, 4 * kPieceSize), 0); } TEST_P(UdpSocketTest, BoundaryPreserved_SendMsgRecvMsg) { - // Bind s_ to loopback:TestPort. - ASSERT_THAT(bind(s_, addr_[0], addrlen_), SyscallSucceeds()); + ASSERT_NO_ERRNO(BindLoopback()); - // Send 2 packets from t_ to s_, where each packet's data consists of 2 - // discontiguous iovecs. + // Send 2 packets from sock to bind_, where each packet's data consists of + // 2 discontiguous iovecs. constexpr size_t kPieceSize = 100; char buf[4 * kPieceSize]; RandomizeBuffer(buf, sizeof(buf)); @@ -1065,11 +1074,12 @@ TEST_P(UdpSocketTest, BoundaryPreserved_SendMsgRecvMsg) { iov[j].iov_len = kPieceSize; } struct msghdr msg = {}; - msg.msg_name = addr_[0]; + msg.msg_name = bind_addr_; msg.msg_namelen = addrlen_; msg.msg_iov = iov; msg.msg_iovlen = 2; - ASSERT_THAT(sendmsg(t_, &msg, 0), SyscallSucceedsWithValue(2 * kPieceSize)); + ASSERT_THAT(sendmsg(sock_.get(), &msg, 0), + SyscallSucceedsWithValue(2 * kPieceSize)); } // Receive the data as 2 separate packets. @@ -1084,85 +1094,87 @@ TEST_P(UdpSocketTest, BoundaryPreserved_SendMsgRecvMsg) { struct msghdr msg = {}; msg.msg_iov = iov; msg.msg_iovlen = 3; - ASSERT_THAT(recvmsg(s_, &msg, 0), SyscallSucceedsWithValue(2 * kPieceSize)); + ASSERT_THAT(recvmsg(bind_.get(), &msg, 0), + SyscallSucceedsWithValue(2 * kPieceSize)); } EXPECT_EQ(memcmp(buf, received, 4 * kPieceSize), 0); } TEST_P(UdpSocketTest, FIONREADShutdown) { + ASSERT_NO_ERRNO(BindLoopback()); + int n = -1; - EXPECT_THAT(ioctl(s_, FIONREAD, &n), SyscallSucceedsWithValue(0)); + EXPECT_THAT(ioctl(sock_.get(), FIONREAD, &n), SyscallSucceedsWithValue(0)); EXPECT_EQ(n, 0); // A UDP socket must be connected before it can be shutdown. - ASSERT_THAT(connect(s_, addr_[0], addrlen_), SyscallSucceeds()); + ASSERT_THAT(connect(sock_.get(), bind_addr_, addrlen_), SyscallSucceeds()); n = -1; - EXPECT_THAT(ioctl(s_, FIONREAD, &n), SyscallSucceedsWithValue(0)); + EXPECT_THAT(ioctl(sock_.get(), FIONREAD, &n), SyscallSucceedsWithValue(0)); EXPECT_EQ(n, 0); - EXPECT_THAT(shutdown(s_, SHUT_RD), SyscallSucceeds()); + EXPECT_THAT(shutdown(sock_.get(), SHUT_RD), SyscallSucceeds()); n = -1; - EXPECT_THAT(ioctl(s_, FIONREAD, &n), SyscallSucceedsWithValue(0)); + EXPECT_THAT(ioctl(sock_.get(), FIONREAD, &n), SyscallSucceedsWithValue(0)); EXPECT_EQ(n, 0); } TEST_P(UdpSocketTest, FIONREADWriteShutdown) { int n = -1; - EXPECT_THAT(ioctl(s_, FIONREAD, &n), SyscallSucceedsWithValue(0)); + EXPECT_THAT(ioctl(bind_.get(), FIONREAD, &n), SyscallSucceedsWithValue(0)); EXPECT_EQ(n, 0); - // Bind s_ to loopback:TestPort. - ASSERT_THAT(bind(s_, addr_[0], addrlen_), SyscallSucceeds()); + ASSERT_NO_ERRNO(BindLoopback()); // A UDP socket must be connected before it can be shutdown. - ASSERT_THAT(connect(s_, addr_[0], addrlen_), SyscallSucceeds()); + ASSERT_THAT(connect(bind_.get(), bind_addr_, addrlen_), SyscallSucceeds()); n = -1; - EXPECT_THAT(ioctl(s_, FIONREAD, &n), SyscallSucceedsWithValue(0)); + EXPECT_THAT(ioctl(bind_.get(), FIONREAD, &n), SyscallSucceedsWithValue(0)); EXPECT_EQ(n, 0); const char str[] = "abc"; - ASSERT_THAT(send(s_, str, sizeof(str), 0), + ASSERT_THAT(send(bind_.get(), str, sizeof(str), 0), SyscallSucceedsWithValue(sizeof(str))); - struct pollfd pfd = {s_, POLLIN, 0}; + struct pollfd pfd = {bind_.get(), POLLIN, 0}; ASSERT_THAT(RetryEINTR(poll)(&pfd, 1, /*timeout=*/1000), SyscallSucceedsWithValue(1)); n = -1; - EXPECT_THAT(ioctl(s_, FIONREAD, &n), SyscallSucceedsWithValue(0)); + EXPECT_THAT(ioctl(bind_.get(), FIONREAD, &n), SyscallSucceedsWithValue(0)); EXPECT_EQ(n, sizeof(str)); - EXPECT_THAT(shutdown(s_, SHUT_RD), SyscallSucceeds()); + EXPECT_THAT(shutdown(bind_.get(), SHUT_RD), SyscallSucceeds()); n = -1; - EXPECT_THAT(ioctl(s_, FIONREAD, &n), SyscallSucceedsWithValue(0)); + EXPECT_THAT(ioctl(bind_.get(), FIONREAD, &n), SyscallSucceedsWithValue(0)); EXPECT_EQ(n, sizeof(str)); } // NOTE: Do not use `FIONREAD` as test name because it will be replaced by the // corresponding macro and become `0x541B`. TEST_P(UdpSocketTest, Fionread) { - // Bind s_ to loopback:TestPort. - ASSERT_THAT(bind(s_, addr_[0], addrlen_), SyscallSucceeds()); + ASSERT_NO_ERRNO(BindLoopback()); // Check that the bound socket with an empty buffer reports an empty first // packet. int n = -1; - EXPECT_THAT(ioctl(s_, FIONREAD, &n), SyscallSucceedsWithValue(0)); + EXPECT_THAT(ioctl(bind_.get(), FIONREAD, &n), SyscallSucceedsWithValue(0)); EXPECT_EQ(n, 0); - // Send 3 packets from t_ to s_. + // Send 3 packets from sock to bind_. constexpr int psize = 100; char buf[3 * psize]; RandomizeBuffer(buf, sizeof(buf)); - struct pollfd pfd = {s_, POLLIN, 0}; + struct pollfd pfd = {bind_.get(), POLLIN, 0}; for (int i = 0; i < 3; ++i) { - ASSERT_THAT(sendto(t_, buf + i * psize, psize, 0, addr_[0], addrlen_), - SyscallSucceedsWithValue(psize)); + ASSERT_THAT( + sendto(sock_.get(), buf + i * psize, psize, 0, bind_addr_, addrlen_), + SyscallSucceedsWithValue(psize)); ASSERT_THAT(RetryEINTR(poll)(&pfd, 1, /*timeout=*/1000), SyscallSucceedsWithValue(1)); @@ -1170,30 +1182,30 @@ TEST_P(UdpSocketTest, Fionread) { // Check that regardless of how many packets are in the queue, the size // reported is that of a single packet. n = -1; - EXPECT_THAT(ioctl(s_, FIONREAD, &n), SyscallSucceedsWithValue(0)); + EXPECT_THAT(ioctl(bind_.get(), FIONREAD, &n), SyscallSucceedsWithValue(0)); EXPECT_EQ(n, psize); } } TEST_P(UdpSocketTest, FIONREADZeroLengthPacket) { - // Bind s_ to loopback:TestPort. - ASSERT_THAT(bind(s_, addr_[0], addrlen_), SyscallSucceeds()); + ASSERT_NO_ERRNO(BindLoopback()); // Check that the bound socket with an empty buffer reports an empty first // packet. int n = -1; - EXPECT_THAT(ioctl(s_, FIONREAD, &n), SyscallSucceedsWithValue(0)); + EXPECT_THAT(ioctl(bind_.get(), FIONREAD, &n), SyscallSucceedsWithValue(0)); EXPECT_EQ(n, 0); - // Send 3 packets from t_ to s_. + // Send 3 packets from sock to bind_. constexpr int psize = 100; char buf[3 * psize]; RandomizeBuffer(buf, sizeof(buf)); - struct pollfd pfd = {s_, POLLIN, 0}; + struct pollfd pfd = {bind_.get(), POLLIN, 0}; for (int i = 0; i < 3; ++i) { - ASSERT_THAT(sendto(t_, buf + i * psize, 0, 0, addr_[0], addrlen_), - SyscallSucceedsWithValue(0)); + ASSERT_THAT( + sendto(sock_.get(), buf + i * psize, 0, 0, bind_addr_, addrlen_), + SyscallSucceedsWithValue(0)); // TODO(gvisor.dev/issue/2726): sending a zero-length message to a hostinet // socket does not cause a poll event to be triggered. @@ -1205,40 +1217,77 @@ TEST_P(UdpSocketTest, FIONREADZeroLengthPacket) { // Check that regardless of how many packets are in the queue, the size // reported is that of a single packet. n = -1; - EXPECT_THAT(ioctl(s_, FIONREAD, &n), SyscallSucceedsWithValue(0)); + EXPECT_THAT(ioctl(bind_.get(), FIONREAD, &n), SyscallSucceedsWithValue(0)); EXPECT_EQ(n, 0); } } TEST_P(UdpSocketTest, FIONREADZeroLengthWriteShutdown) { int n = -1; - EXPECT_THAT(ioctl(s_, FIONREAD, &n), SyscallSucceedsWithValue(0)); + EXPECT_THAT(ioctl(bind_.get(), FIONREAD, &n), SyscallSucceedsWithValue(0)); EXPECT_EQ(n, 0); - // Bind s_ to loopback:TestPort. - ASSERT_THAT(bind(s_, addr_[0], addrlen_), SyscallSucceeds()); + ASSERT_NO_ERRNO(BindLoopback()); // A UDP socket must be connected before it can be shutdown. - ASSERT_THAT(connect(s_, addr_[0], addrlen_), SyscallSucceeds()); + ASSERT_THAT(connect(bind_.get(), bind_addr_, addrlen_), SyscallSucceeds()); n = -1; - EXPECT_THAT(ioctl(s_, FIONREAD, &n), SyscallSucceedsWithValue(0)); + EXPECT_THAT(ioctl(bind_.get(), FIONREAD, &n), SyscallSucceedsWithValue(0)); EXPECT_EQ(n, 0); const char str[] = "abc"; - ASSERT_THAT(send(s_, str, 0, 0), SyscallSucceedsWithValue(0)); + ASSERT_THAT(send(bind_.get(), str, 0, 0), SyscallSucceedsWithValue(0)); n = -1; - EXPECT_THAT(ioctl(s_, FIONREAD, &n), SyscallSucceedsWithValue(0)); + EXPECT_THAT(ioctl(bind_.get(), FIONREAD, &n), SyscallSucceedsWithValue(0)); EXPECT_EQ(n, 0); - EXPECT_THAT(shutdown(s_, SHUT_RD), SyscallSucceeds()); + EXPECT_THAT(shutdown(bind_.get(), SHUT_RD), SyscallSucceeds()); n = -1; - EXPECT_THAT(ioctl(s_, FIONREAD, &n), SyscallSucceedsWithValue(0)); + EXPECT_THAT(ioctl(bind_.get(), FIONREAD, &n), SyscallSucceedsWithValue(0)); EXPECT_EQ(n, 0); } +TEST_P(UdpSocketTest, SoNoCheckOffByDefault) { + // TODO(gvisor.dev/issue/1202): SO_NO_CHECK socket option not supported by + // hostinet. + SKIP_IF(IsRunningWithHostinet()); + + int v = -1; + socklen_t optlen = sizeof(v); + ASSERT_THAT(getsockopt(bind_.get(), SOL_SOCKET, SO_NO_CHECK, &v, &optlen), + SyscallSucceeds()); + ASSERT_EQ(v, kSockOptOff); + ASSERT_EQ(optlen, sizeof(v)); +} + +TEST_P(UdpSocketTest, SoNoCheck) { + // TODO(gvisor.dev/issue/1202): SO_NO_CHECK socket option not supported by + // hostinet. + SKIP_IF(IsRunningWithHostinet()); + + int v = kSockOptOn; + socklen_t optlen = sizeof(v); + ASSERT_THAT(setsockopt(bind_.get(), SOL_SOCKET, SO_NO_CHECK, &v, optlen), + SyscallSucceeds()); + v = -1; + ASSERT_THAT(getsockopt(bind_.get(), SOL_SOCKET, SO_NO_CHECK, &v, &optlen), + SyscallSucceeds()); + ASSERT_EQ(v, kSockOptOn); + ASSERT_EQ(optlen, sizeof(v)); + + v = kSockOptOff; + ASSERT_THAT(setsockopt(bind_.get(), SOL_SOCKET, SO_NO_CHECK, &v, optlen), + SyscallSucceeds()); + v = -1; + ASSERT_THAT(getsockopt(bind_.get(), SOL_SOCKET, SO_NO_CHECK, &v, &optlen), + SyscallSucceeds()); + ASSERT_EQ(v, kSockOptOff); + ASSERT_EQ(optlen, sizeof(v)); +} + TEST_P(UdpSocketTest, SoTimestampOffByDefault) { // TODO(gvisor.dev/issue/1202): SO_TIMESTAMP socket option not supported by // hostinet. @@ -1246,7 +1295,7 @@ TEST_P(UdpSocketTest, SoTimestampOffByDefault) { int v = -1; socklen_t optlen = sizeof(v); - ASSERT_THAT(getsockopt(s_, SOL_SOCKET, SO_TIMESTAMP, &v, &optlen), + ASSERT_THAT(getsockopt(bind_.get(), SOL_SOCKET, SO_TIMESTAMP, &v, &optlen), SyscallSucceeds()); ASSERT_EQ(v, kSockOptOff); ASSERT_EQ(optlen, sizeof(v)); @@ -1257,18 +1306,19 @@ TEST_P(UdpSocketTest, SoTimestamp) { // supported by hostinet. SKIP_IF(IsRunningWithHostinet()); - ASSERT_THAT(bind(s_, addr_[0], addrlen_), SyscallSucceeds()); - ASSERT_THAT(connect(t_, addr_[0], addrlen_), SyscallSucceeds()); + ASSERT_NO_ERRNO(BindLoopback()); + ASSERT_THAT(connect(sock_.get(), bind_addr_, addrlen_), SyscallSucceeds()); int v = 1; - ASSERT_THAT(setsockopt(s_, SOL_SOCKET, SO_TIMESTAMP, &v, sizeof(v)), + ASSERT_THAT(setsockopt(bind_.get(), SOL_SOCKET, SO_TIMESTAMP, &v, sizeof(v)), SyscallSucceeds()); char buf[3]; - // Send zero length packet from t_ to s_. - ASSERT_THAT(RetryEINTR(write)(t_, buf, 0), SyscallSucceedsWithValue(0)); + // Send zero length packet from sock to bind_. + ASSERT_THAT(RetryEINTR(write)(sock_.get(), buf, 0), + SyscallSucceedsWithValue(0)); - struct pollfd pfd = {s_, POLLIN, 0}; + struct pollfd pfd = {bind_.get(), POLLIN, 0}; ASSERT_THAT(RetryEINTR(poll)(&pfd, 1, /*timeout=*/1000), SyscallSucceedsWithValue(1)); @@ -1282,7 +1332,8 @@ TEST_P(UdpSocketTest, SoTimestamp) { msg.msg_control = cmsgbuf; msg.msg_controllen = sizeof(cmsgbuf); - ASSERT_THAT(RetryEINTR(recvmsg)(s_, &msg, 0), SyscallSucceedsWithValue(0)); + ASSERT_THAT(RetryEINTR(recvmsg)(bind_.get(), &msg, 0), + SyscallSucceedsWithValue(0)); struct cmsghdr* cmsg = CMSG_FIRSTHDR(&msg); ASSERT_NE(cmsg, nullptr); @@ -1296,36 +1347,37 @@ TEST_P(UdpSocketTest, SoTimestamp) { ASSERT_TRUE(tv.tv_sec != 0 || tv.tv_usec != 0); // There should be nothing to get via ioctl. - ASSERT_THAT(ioctl(s_, SIOCGSTAMP, &tv), SyscallFailsWithErrno(ENOENT)); + ASSERT_THAT(ioctl(bind_.get(), SIOCGSTAMP, &tv), + SyscallFailsWithErrno(ENOENT)); } TEST_P(UdpSocketTest, WriteShutdownNotConnected) { - EXPECT_THAT(shutdown(s_, SHUT_WR), SyscallFailsWithErrno(ENOTCONN)); + EXPECT_THAT(shutdown(bind_.get(), SHUT_WR), SyscallFailsWithErrno(ENOTCONN)); } TEST_P(UdpSocketTest, TimestampIoctl) { // TODO(gvisor.dev/issue/1202): ioctl() is not supported by hostinet. SKIP_IF(IsRunningWithHostinet()); - ASSERT_THAT(bind(s_, addr_[0], addrlen_), SyscallSucceeds()); - ASSERT_THAT(connect(t_, addr_[0], addrlen_), SyscallSucceeds()); + ASSERT_NO_ERRNO(BindLoopback()); + ASSERT_THAT(connect(sock_.get(), bind_addr_, addrlen_), SyscallSucceeds()); char buf[3]; - // Send packet from t_ to s_. - ASSERT_THAT(RetryEINTR(write)(t_, buf, sizeof(buf)), + // Send packet from sock to bind_. + ASSERT_THAT(RetryEINTR(write)(sock_.get(), buf, sizeof(buf)), SyscallSucceedsWithValue(sizeof(buf))); - struct pollfd pfd = {s_, POLLIN, 0}; + struct pollfd pfd = {bind_.get(), POLLIN, 0}; ASSERT_THAT(RetryEINTR(poll)(&pfd, 1, /*timeout=*/1000), SyscallSucceedsWithValue(1)); // There should be no control messages. char recv_buf[sizeof(buf)]; - ASSERT_NO_FATAL_FAILURE(RecvNoCmsg(s_, recv_buf, sizeof(recv_buf))); + ASSERT_NO_FATAL_FAILURE(RecvNoCmsg(bind_.get(), recv_buf, sizeof(recv_buf))); // A nonzero timeval should be available via ioctl. struct timeval tv = {}; - ASSERT_THAT(ioctl(s_, SIOCGSTAMP, &tv), SyscallSucceeds()); + ASSERT_THAT(ioctl(bind_.get(), SIOCGSTAMP, &tv), SyscallSucceeds()); ASSERT_TRUE(tv.tv_sec != 0 || tv.tv_usec != 0); } @@ -1333,11 +1385,12 @@ TEST_P(UdpSocketTest, TimestampIoctlNothingRead) { // TODO(gvisor.dev/issue/1202): ioctl() is not supported by hostinet. SKIP_IF(IsRunningWithHostinet()); - ASSERT_THAT(bind(s_, addr_[0], addrlen_), SyscallSucceeds()); - ASSERT_THAT(connect(t_, addr_[0], addrlen_), SyscallSucceeds()); + ASSERT_NO_ERRNO(BindLoopback()); + ASSERT_THAT(connect(sock_.get(), bind_addr_, addrlen_), SyscallSucceeds()); struct timeval tv = {}; - ASSERT_THAT(ioctl(s_, SIOCGSTAMP, &tv), SyscallFailsWithErrno(ENOENT)); + ASSERT_THAT(ioctl(sock_.get(), SIOCGSTAMP, &tv), + SyscallFailsWithErrno(ENOENT)); } // Test that the timestamp accessed via SIOCGSTAMP is still accessible after @@ -1347,33 +1400,35 @@ TEST_P(UdpSocketTest, TimestampIoctlPersistence) { // supported by hostinet. SKIP_IF(IsRunningWithHostinet()); - ASSERT_THAT(bind(s_, addr_[0], addrlen_), SyscallSucceeds()); - ASSERT_THAT(connect(t_, addr_[0], addrlen_), SyscallSucceeds()); + ASSERT_NO_ERRNO(BindLoopback()); + ASSERT_THAT(connect(sock_.get(), bind_addr_, addrlen_), SyscallSucceeds()); char buf[3]; - // Send packet from t_ to s_. - ASSERT_THAT(RetryEINTR(write)(t_, buf, sizeof(buf)), + // Send packet from sock to bind_. + ASSERT_THAT(RetryEINTR(write)(sock_.get(), buf, sizeof(buf)), SyscallSucceedsWithValue(sizeof(buf))); - ASSERT_THAT(RetryEINTR(write)(t_, buf, 0), SyscallSucceedsWithValue(0)); + ASSERT_THAT(RetryEINTR(write)(sock_.get(), buf, 0), + SyscallSucceedsWithValue(0)); - struct pollfd pfd = {s_, POLLIN, 0}; + struct pollfd pfd = {bind_.get(), POLLIN, 0}; ASSERT_THAT(RetryEINTR(poll)(&pfd, 1, /*timeout=*/1000), SyscallSucceedsWithValue(1)); // There should be no control messages. char recv_buf[sizeof(buf)]; - ASSERT_NO_FATAL_FAILURE(RecvNoCmsg(s_, recv_buf, sizeof(recv_buf))); + ASSERT_NO_FATAL_FAILURE(RecvNoCmsg(bind_.get(), recv_buf, sizeof(recv_buf))); // A nonzero timeval should be available via ioctl. struct timeval tv = {}; - ASSERT_THAT(ioctl(s_, SIOCGSTAMP, &tv), SyscallSucceeds()); + ASSERT_THAT(ioctl(bind_.get(), SIOCGSTAMP, &tv), SyscallSucceeds()); ASSERT_TRUE(tv.tv_sec != 0 || tv.tv_usec != 0); // Enable SO_TIMESTAMP and send a message. int v = 1; - EXPECT_THAT(setsockopt(s_, SOL_SOCKET, SO_TIMESTAMP, &v, sizeof(v)), + EXPECT_THAT(setsockopt(bind_.get(), SOL_SOCKET, SO_TIMESTAMP, &v, sizeof(v)), SyscallSucceeds()); - ASSERT_THAT(RetryEINTR(write)(t_, buf, 0), SyscallSucceedsWithValue(0)); + ASSERT_THAT(RetryEINTR(write)(sock_.get(), buf, 0), + SyscallSucceedsWithValue(0)); ASSERT_THAT(RetryEINTR(poll)(&pfd, 1, /*timeout=*/1000), SyscallSucceedsWithValue(1)); @@ -1386,13 +1441,14 @@ TEST_P(UdpSocketTest, TimestampIoctlPersistence) { msg.msg_iovlen = 1; msg.msg_control = cmsgbuf; msg.msg_controllen = sizeof(cmsgbuf); - ASSERT_THAT(RetryEINTR(recvmsg)(s_, &msg, 0), SyscallSucceedsWithValue(0)); + ASSERT_THAT(RetryEINTR(recvmsg)(bind_.get(), &msg, 0), + SyscallSucceedsWithValue(0)); struct cmsghdr* cmsg = CMSG_FIRSTHDR(&msg); ASSERT_NE(cmsg, nullptr); // The ioctl should return the exact same values as before. struct timeval tv2 = {}; - ASSERT_THAT(ioctl(s_, SIOCGSTAMP, &tv2), SyscallSucceeds()); + ASSERT_THAT(ioctl(bind_.get(), SIOCGSTAMP, &tv2), SyscallSucceeds()); ASSERT_EQ(tv.tv_sec, tv2.tv_sec); ASSERT_EQ(tv.tv_usec, tv2.tv_usec); } @@ -1401,8 +1457,8 @@ TEST_P(UdpSocketTest, TimestampIoctlPersistence) { // outgoing packets, and that a receiving socket with IP_RECVTOS or // IPV6_RECVTCLASS will create the corresponding control message. TEST_P(UdpSocketTest, SetAndReceiveTOS) { - ASSERT_THAT(bind(s_, addr_[0], addrlen_), SyscallSucceeds()); - ASSERT_THAT(connect(t_, addr_[0], addrlen_), SyscallSucceeds()); + ASSERT_NO_ERRNO(BindLoopback()); + ASSERT_THAT(connect(sock_.get(), bind_addr_, addrlen_), SyscallSucceeds()); // Allow socket to receive control message. int recv_level = SOL_IP; @@ -1411,9 +1467,9 @@ TEST_P(UdpSocketTest, SetAndReceiveTOS) { recv_level = SOL_IPV6; recv_type = IPV6_RECVTCLASS; } - ASSERT_THAT( - setsockopt(s_, recv_level, recv_type, &kSockOptOn, sizeof(kSockOptOn)), - SyscallSucceeds()); + ASSERT_THAT(setsockopt(bind_.get(), recv_level, recv_type, &kSockOptOn, + sizeof(kSockOptOn)), + SyscallSucceeds()); // Set socket TOS. int sent_level = recv_level; @@ -1422,9 +1478,9 @@ TEST_P(UdpSocketTest, SetAndReceiveTOS) { sent_type = IPV6_TCLASS; } int sent_tos = IPTOS_LOWDELAY; // Choose some TOS value. - ASSERT_THAT( - setsockopt(t_, sent_level, sent_type, &sent_tos, sizeof(sent_tos)), - SyscallSucceeds()); + ASSERT_THAT(setsockopt(sock_.get(), sent_level, sent_type, &sent_tos, + sizeof(sent_tos)), + SyscallSucceeds()); // Prepare message to send. constexpr size_t kDataLength = 1024; @@ -1436,7 +1492,7 @@ TEST_P(UdpSocketTest, SetAndReceiveTOS) { sent_msg.msg_iov = &sent_iov; sent_msg.msg_iovlen = 1; - ASSERT_THAT(RetryEINTR(sendmsg)(t_, &sent_msg, 0), + ASSERT_THAT(RetryEINTR(sendmsg)(sock_.get(), &sent_msg, 0), SyscallSucceedsWithValue(kDataLength)); // Receive message. @@ -1454,7 +1510,7 @@ TEST_P(UdpSocketTest, SetAndReceiveTOS) { std::vector<char> received_cmsgbuf(CMSG_SPACE(cmsg_data_len)); received_msg.msg_control = &received_cmsgbuf[0]; received_msg.msg_controllen = received_cmsgbuf.size(); - ASSERT_THAT(RetryEINTR(recvmsg)(s_, &received_msg, 0), + ASSERT_THAT(RetryEINTR(recvmsg)(bind_.get(), &received_msg, 0), SyscallSucceedsWithValue(kDataLength)); struct cmsghdr* cmsg = CMSG_FIRSTHDR(&received_msg); @@ -1473,8 +1529,9 @@ TEST_P(UdpSocketTest, SetAndReceiveTOS) { TEST_P(UdpSocketTest, SendAndReceiveTOS) { // TODO(b/146661005): Setting TOS via cmsg not supported for netstack. SKIP_IF(IsRunningOnGvisor() && !IsRunningWithHostinet()); - ASSERT_THAT(bind(s_, addr_[0], addrlen_), SyscallSucceeds()); - ASSERT_THAT(connect(t_, addr_[0], addrlen_), SyscallSucceeds()); + + ASSERT_NO_ERRNO(BindLoopback()); + ASSERT_THAT(connect(sock_.get(), bind_addr_, addrlen_), SyscallSucceeds()); // Allow socket to receive control message. int recv_level = SOL_IP; @@ -1484,9 +1541,9 @@ TEST_P(UdpSocketTest, SendAndReceiveTOS) { recv_type = IPV6_RECVTCLASS; } int recv_opt = kSockOptOn; - ASSERT_THAT( - setsockopt(s_, recv_level, recv_type, &recv_opt, sizeof(recv_opt)), - SyscallSucceeds()); + ASSERT_THAT(setsockopt(bind_.get(), recv_level, recv_type, &recv_opt, + sizeof(recv_opt)), + SyscallSucceeds()); // Prepare message to send. constexpr size_t kDataLength = 1024; @@ -1517,7 +1574,7 @@ TEST_P(UdpSocketTest, SendAndReceiveTOS) { sent_cmsg->cmsg_type = sent_type; *(int8_t*)CMSG_DATA(sent_cmsg) = sent_tos; - ASSERT_THAT(RetryEINTR(sendmsg)(t_, &sent_msg, 0), + ASSERT_THAT(RetryEINTR(sendmsg)(sock_.get(), &sent_msg, 0), SyscallSucceedsWithValue(kDataLength)); // Receive message. @@ -1531,7 +1588,7 @@ TEST_P(UdpSocketTest, SendAndReceiveTOS) { std::vector<char> received_cmsgbuf(CMSG_SPACE(cmsg_data_len)); received_msg.msg_control = &received_cmsgbuf[0]; received_msg.msg_controllen = CMSG_LEN(cmsg_data_len); - ASSERT_THAT(RetryEINTR(recvmsg)(s_, &received_msg, 0), + ASSERT_THAT(RetryEINTR(recvmsg)(bind_.get(), &received_msg, 0), SyscallSucceedsWithValue(kDataLength)); struct cmsghdr* cmsg = CMSG_FIRSTHDR(&received_msg); @@ -1543,5 +1600,182 @@ TEST_P(UdpSocketTest, SendAndReceiveTOS) { memcpy(&received_tos, CMSG_DATA(cmsg), sizeof(received_tos)); EXPECT_EQ(received_tos, sent_tos); } + +TEST_P(UdpSocketTest, RecvBufLimitsEmptyRcvBuf) { + // Discover minimum buffer size by setting it to zero. + constexpr int kRcvBufSz = 0; + ASSERT_THAT(setsockopt(bind_.get(), SOL_SOCKET, SO_RCVBUF, &kRcvBufSz, + sizeof(kRcvBufSz)), + SyscallSucceeds()); + + int min = 0; + socklen_t min_len = sizeof(min); + ASSERT_THAT(getsockopt(bind_.get(), SOL_SOCKET, SO_RCVBUF, &min, &min_len), + SyscallSucceeds()); + + // Bind bind_ to loopback. + ASSERT_NO_ERRNO(BindLoopback()); + + { + // Send data of size min and verify that it's received. + std::vector<char> buf(min); + RandomizeBuffer(buf.data(), buf.size()); + ASSERT_THAT( + sendto(sock_.get(), buf.data(), buf.size(), 0, bind_addr_, addrlen_), + SyscallSucceedsWithValue(buf.size())); + std::vector<char> received(buf.size()); + EXPECT_THAT( + recv(bind_.get(), received.data(), received.size(), MSG_DONTWAIT), + SyscallSucceedsWithValue(received.size())); + } + + { + // Send data of size min + 1 and verify that its received. Both linux and + // Netstack accept a dgram that exceeds rcvBuf limits if the receive buffer + // is currently empty. + std::vector<char> buf(min + 1); + RandomizeBuffer(buf.data(), buf.size()); + ASSERT_THAT( + sendto(sock_.get(), buf.data(), buf.size(), 0, bind_addr_, addrlen_), + SyscallSucceedsWithValue(buf.size())); + + std::vector<char> received(buf.size()); + EXPECT_THAT( + recv(bind_.get(), received.data(), received.size(), MSG_DONTWAIT), + SyscallSucceedsWithValue(received.size())); + } +} + +// Test that receive buffer limits are enforced. +TEST_P(UdpSocketTest, RecvBufLimits) { + // Bind s_ to loopback. + ASSERT_NO_ERRNO(BindLoopback()); + + int min = 0; + { + // Discover minimum buffer size by trying to set it to zero. + constexpr int kRcvBufSz = 0; + ASSERT_THAT(setsockopt(bind_.get(), SOL_SOCKET, SO_RCVBUF, &kRcvBufSz, + sizeof(kRcvBufSz)), + SyscallSucceeds()); + + socklen_t min_len = sizeof(min); + ASSERT_THAT(getsockopt(bind_.get(), SOL_SOCKET, SO_RCVBUF, &min, &min_len), + SyscallSucceeds()); + } + + // Now set the limit to min * 4. + int new_rcv_buf_sz = min * 4; + if (!IsRunningOnGvisor() || IsRunningWithHostinet()) { + // Linux doubles the value specified so just set to min * 2. + new_rcv_buf_sz = min * 2; + } + + ASSERT_THAT(setsockopt(bind_.get(), SOL_SOCKET, SO_RCVBUF, &new_rcv_buf_sz, + sizeof(new_rcv_buf_sz)), + SyscallSucceeds()); + int rcv_buf_sz = 0; + { + socklen_t rcv_buf_len = sizeof(rcv_buf_sz); + ASSERT_THAT(getsockopt(bind_.get(), SOL_SOCKET, SO_RCVBUF, &rcv_buf_sz, + &rcv_buf_len), + SyscallSucceeds()); + } + + { + std::vector<char> buf(min); + RandomizeBuffer(buf.data(), buf.size()); + + ASSERT_THAT( + sendto(sock_.get(), buf.data(), buf.size(), 0, bind_addr_, addrlen_), + SyscallSucceedsWithValue(buf.size())); + ASSERT_THAT( + sendto(sock_.get(), buf.data(), buf.size(), 0, bind_addr_, addrlen_), + SyscallSucceedsWithValue(buf.size())); + ASSERT_THAT( + sendto(sock_.get(), buf.data(), buf.size(), 0, bind_addr_, addrlen_), + SyscallSucceedsWithValue(buf.size())); + ASSERT_THAT( + sendto(sock_.get(), buf.data(), buf.size(), 0, bind_addr_, addrlen_), + SyscallSucceedsWithValue(buf.size())); + int sent = 4; + if (IsRunningOnGvisor() && !IsRunningWithHostinet()) { + // Linux seems to drop the 4th packet even though technically it should + // fit in the receive buffer. + ASSERT_THAT( + sendto(sock_.get(), buf.data(), buf.size(), 0, bind_addr_, addrlen_), + SyscallSucceedsWithValue(buf.size())); + sent++; + } + + for (int i = 0; i < sent - 1; i++) { + // Receive the data. + std::vector<char> received(buf.size()); + EXPECT_THAT( + recv(bind_.get(), received.data(), received.size(), MSG_DONTWAIT), + SyscallSucceedsWithValue(received.size())); + EXPECT_EQ(memcmp(buf.data(), received.data(), buf.size()), 0); + } + + // The last receive should fail with EAGAIN as the last packet should have + // been dropped due to lack of space in the receive buffer. + std::vector<char> received(buf.size()); + EXPECT_THAT( + recv(bind_.get(), received.data(), received.size(), MSG_DONTWAIT), + SyscallFailsWithErrno(EAGAIN)); + } +} + +#ifndef __fuchsia__ + +// TODO(gvisor.dev/2746): Support SO_ATTACH_FILTER/SO_DETACH_FILTER. +// gVisor currently silently ignores attaching a filter. +TEST_P(UdpSocketTest, SetSocketDetachFilter) { + // Program generated using sudo tcpdump -i lo udp and port 1234 -dd + struct sock_filter code[] = { + {0x28, 0, 0, 0x0000000c}, {0x15, 0, 6, 0x000086dd}, + {0x30, 0, 0, 0x00000014}, {0x15, 0, 15, 0x00000011}, + {0x28, 0, 0, 0x00000036}, {0x15, 12, 0, 0x000004d2}, + {0x28, 0, 0, 0x00000038}, {0x15, 10, 11, 0x000004d2}, + {0x15, 0, 10, 0x00000800}, {0x30, 0, 0, 0x00000017}, + {0x15, 0, 8, 0x00000011}, {0x28, 0, 0, 0x00000014}, + {0x45, 6, 0, 0x00001fff}, {0xb1, 0, 0, 0x0000000e}, + {0x48, 0, 0, 0x0000000e}, {0x15, 2, 0, 0x000004d2}, + {0x48, 0, 0, 0x00000010}, {0x15, 0, 1, 0x000004d2}, + {0x6, 0, 0, 0x00040000}, {0x6, 0, 0, 0x00000000}, + }; + struct sock_fprog bpf = { + .len = ABSL_ARRAYSIZE(code), + .filter = code, + }; + ASSERT_THAT( + setsockopt(sock_.get(), SOL_SOCKET, SO_ATTACH_FILTER, &bpf, sizeof(bpf)), + SyscallSucceeds()); + + constexpr int val = 0; + ASSERT_THAT( + setsockopt(sock_.get(), SOL_SOCKET, SO_DETACH_FILTER, &val, sizeof(val)), + SyscallSucceeds()); +} + +TEST_P(UdpSocketTest, SetSocketDetachFilterNoInstalledFilter) { + // TODO(gvisor.dev/2746): Support SO_ATTACH_FILTER/SO_DETACH_FILTER. + SKIP_IF(IsRunningOnGvisor()); + constexpr int val = 0; + ASSERT_THAT( + setsockopt(sock_.get(), SOL_SOCKET, SO_DETACH_FILTER, &val, sizeof(val)), + SyscallFailsWithErrno(ENOENT)); +} + +TEST_P(UdpSocketTest, GetSocketDetachFilter) { + int val = 0; + socklen_t val_len = sizeof(val); + ASSERT_THAT( + getsockopt(sock_.get(), SOL_SOCKET, SO_DETACH_FILTER, &val, &val_len), + SyscallFailsWithErrno(ENOPROTOOPT)); +} + +#endif // __fuchsia__ + } // namespace testing } // namespace gvisor diff --git a/test/syscalls/linux/udp_socket_test_cases.h b/test/syscalls/linux/udp_socket_test_cases.h index 2fd79d99e..f7e25c805 100644 --- a/test/syscalls/linux/udp_socket_test_cases.h +++ b/test/syscalls/linux/udp_socket_test_cases.h @@ -15,8 +15,12 @@ #ifndef THIRD_PARTY_GOLANG_GVISOR_TEST_SYSCALLS_LINUX_SOCKET_IPV4_UDP_UNBOUND_H_ #define THIRD_PARTY_GOLANG_GVISOR_TEST_SYSCALLS_LINUX_SOCKET_IPV4_UDP_UNBOUND_H_ +#include <sys/socket.h> + #include "gtest/gtest.h" #include "test/syscalls/linux/socket_test_util.h" +#include "test/util/file_descriptor.h" +#include "test/util/posix_error.h" namespace gvisor { namespace testing { @@ -32,42 +36,46 @@ class UdpSocketTest // Creates two sockets that will be used by test cases. void SetUp() override; - // Closes the sockets created by SetUp(). - void TearDown() override { - EXPECT_THAT(close(s_), SyscallSucceeds()); - EXPECT_THAT(close(t_), SyscallSucceeds()); + // Binds the socket bind_ to the loopback and updates bind_addr_. + PosixError BindLoopback(); - for (size_t i = 0; i < ABSL_ARRAYSIZE(ports_); ++i) { - ASSERT_NO_ERRNO(FreeAvailablePort(ports_[i])); - } - } + // Binds the socket bind_ to Any and updates bind_addr_. + PosixError BindAny(); - // First UDP socket. - int s_; + // Binds given socket to address addr and updates. + PosixError BindSocket(int socket, struct sockaddr* addr); - // Second UDP socket. - int t_; + // Return initialized Any address to port 0. + struct sockaddr_storage InetAnyAddr(); - // The length of the socket address. - socklen_t addrlen_; + // Return initialized Loopback address to port 0. + struct sockaddr_storage InetLoopbackAddr(); - // Initialized address pointing to loopback and port TestPort+i. - struct sockaddr* addr_[3]; + // Disconnects socket sockfd. + void Disconnect(int sockfd); - // Initialize "any" address. - struct sockaddr* anyaddr_; + // Get family for the test. + int GetFamily(); - // Used ports. - int ports_[3]; + // Socket used by Bind methods + FileDescriptor bind_; - private: - // Storage for the loopback addresses. - struct sockaddr_storage addr_storage_[3]; + // Second socket used for tests. + FileDescriptor sock_; - // Storage for the "any" address. - struct sockaddr_storage anyaddr_storage_; -}; + // Address for bind_ socket. + struct sockaddr* bind_addr_; + // Initialized to the length based on GetFamily(). + socklen_t addrlen_; + + // Storage for bind_addr_. + struct sockaddr_storage bind_addr_storage_; + + private: + // Helper to initialize addrlen_ for the test case. + socklen_t GetAddrLength(); +}; } // namespace testing } // namespace gvisor diff --git a/test/syscalls/linux/xattr.cc b/test/syscalls/linux/xattr.cc index 3231732ec..cbcf08451 100644 --- a/test/syscalls/linux/xattr.cc +++ b/test/syscalls/linux/xattr.cc @@ -73,9 +73,10 @@ TEST_F(XattrTest, XattrLargeName) { std::string name = "user."; name += std::string(XATTR_NAME_MAX - name.length(), 'a'); - // An xattr should be whitelisted before it can be accessed--do not allow - // arbitrary xattrs to be read/written in gVisor. if (!IsRunningOnGvisor()) { + // In gVisor, access to xattrs is controlled with an explicit list of + // allowed names. This name isn't going to be configured to allow access, so + // don't test it. EXPECT_THAT(setxattr(path, name.c_str(), nullptr, 0, /*flags=*/0), SyscallSucceeds()); EXPECT_THAT(getxattr(path, name.c_str(), nullptr, 0), |