This series is quite a bit different from v2. First, I cleaned things up along the lines of Chuck's review of v2, but then Olga suggested that we should just give up on trying to do new rpcbind registrations when one fails.
I've added a patches to do that and this now seems to fix the reproducer that syzbot provided recently. I have high hopes that it will fix the other syzbot hangs that have been reported. This also sets a foundation for some other work I have that will move the rpcbind registration into userland. I'll submit that series separately after this one is resolved. Please consider these for v7.4. To: Chuck Lever <[email protected]> To: NeilBrown <[email protected]> To: Olga Kornievskaia <[email protected]> To: Dai Ngo <[email protected]> To: Tom Talpey <[email protected]> To: Trond Myklebust <[email protected]> To: Anna Schumaker <[email protected]> To: David S. Miller <[email protected]> To: Eric Dumazet <[email protected]> To: Jakub Kicinski <[email protected]> To: Paolo Abeni <[email protected]> To: Simon Horman <[email protected]> To: Shuah Khan <[email protected]> Cc: Slawomir Stepien <[email protected]> Cc: [email protected] Cc: [email protected] Cc: [email protected] Cc: Trond Myklebust <[email protected]> Cc: [email protected] Signed-off-by: Jeff Layton <[email protected]> --- Changes in v2: - New patch 4: when a later registration failed, svc_register() left the entries that it had already set in rpcbind. Those entries pointed at a port that the caller then closed. - Tests: behavioural fixes for several tests. Several assertions checked only an errno that both the fixed kernel and the broken kernel return. val_bad_transport() now also requires that the rpcbind stub saw no traffic. val_second_entry_bad() requires that no listener started. func_empty_destroys requires that the kernel dropped the local rpcbind client and then connected again. The two -EBUSY tests require that the listener set does not change. find_listener() now matches the address too. - Tests: the stub read the revents of a newly accepted pollfd that poll() had not written. The stub could therefore start a blocking read with no readiness event. - Tests: the config fragment now includes NAMESPACES, SHMEM, TMPFS and UNIX. Without them, every test skipped. - Link to v1: https://lore.kernel.org/r/[email protected] --- Changes in v3: - Dropped v2's patch 4, the svc_register() unwind. rpcbind matches an UNSET on [program, version, netid], and the v2 fallback ignores the protocol completely. The unwind could therefore clear entries that belong to other live listeners. Chuck Lever found that. - A failed registration stays fatal. Patch 3 makes CONFIG_NFS_LOCALIO=y agree with CONFIG_NFS_LOCALIO=n on that point. - Tests: sem_register_refused now reads the listener set back, because -EACCES alone does not show that no listener was created. sem_create_failure_extack is new. The netlink socket now asks for NETLINK_EXT_ACK and NETLINK_CAP_ACK, so that a test can read the extack. - Tests: FIXTURE_TEARDOWN removes any listener or thread that a test left behind. Those pinned the netns. - Stop attempting to register any more listeners when the first svc_register() call fails. - Bound the unregistration side the same way. svc_register(), svc_unregister() and the per-listener teardown in svc_delete_xprt() all give up once the local rpcbind stops answering, so one listener_set pays one timeout in each direction rather than one per listener. - Link to v2: https://lore.kernel.org/r/[email protected] --- Jeff Layton (14): NFSD: cap the number of listeners accepted in listener_set NFSD: validate transport name in listener_set before serv creation SUNRPC: keep the first error in svc_register() SUNRPC: bound the local rpcbind client timeout to 1s NFSD: report listener creation failures through extack SUNRPC: report local rpcbind calls that get no answer SUNRPC: stop svc_register() once rpcbind stops answering SUNRPC: stop the svc_unregister() sweep once rpcbind stops answering SUNRPC: stop unregistering listeners once rpcbind stops answering NFSD: stop registering with rpcbind after a failure in listener_set selftests/nfsd: exercise listener_set request validation selftests/nfsd: add a per-netns rpcbind stub and the listener round-trips selftests/nfsd: check that listener_set asks rpcbind once selftests/nfsd: check that listener removal asks rpcbind once MAINTAINERS | 1 + fs/nfsd/nfsctl.c | 87 +- include/linux/sunrpc/clnt.h | 3 +- include/linux/sunrpc/svc.h | 7 +- net/sunrpc/rpcb_clnt.c | 22 +- net/sunrpc/svc.c | 65 +- net/sunrpc/svc_xprt.c | 20 + tools/testing/selftests/Makefile | 1 + tools/testing/selftests/nfsd/.gitignore | 1 + tools/testing/selftests/nfsd/Makefile | 6 + tools/testing/selftests/nfsd/config | 8 + .../testing/selftests/nfsd/nfsd_netlink_listener.c | 1328 ++++++++++++++++++++ tools/testing/selftests/nfsd/settings | 1 + 13 files changed, 1529 insertions(+), 21 deletions(-) --- base-commit: e247236b7ffa1e0940f834787feade8570c5db91 change-id: 20260717-nfsd-nl-hang-10a3b3e93f2a Best regards, -- Jeff Layton <[email protected]>
