| Message ID | cover.1742919341.git.trond.myklebust@hammerspace.com (mailing list archive) |
|---|---|
| Headers | show |
| Series | Ensure that ENETUNREACH terminates state recovery | expand |
On Tue, 25 Mar 2025 at 17:19, <trondmy@kernel.org> wrote: > > From: Trond Myklebust <trond.myklebust@hammerspace.com> > > With the recent patch series that caused containerised mounts which > return ENETUNREACH or ENETDOWN errors to report fatal errors, we also > want to ensure that the state manager thread also triggers fatal errors > in the processes or threads that are waiting for recovery to complete. > > --- > v2: > - Return EIO instead of ENETUNREACH in nfs4_wait_clnt_recover() > > Trond Myklebust (4): > SUNRPC: rpcbind should never reset the port to the value '0' > SUNRPC: rpc_clnt_set_transport() must not change the autobind setting > NFSv4: clp->cl_cons_state < 0 signifies an invalid nfs_client > NFSv4: Treat ENETUNREACH errors as fatal for state recovery > > fs/nfs/nfs4state.c | 14 +++++++++++--- > net/sunrpc/clnt.c | 3 --- > net/sunrpc/rpcb_clnt.c | 5 +++-- > 3 files changed, 14 insertions(+), 8 deletions(-) 1. Can this "ENETUNREACH or ENETDOWN are fatal" feature be turned off? 2. We have concerns about this feature - what will happen if a switch or router gets rebooted? What will happen if you unplug your laptop? What will happen if you enable/disable your VPN software on a machine or container? What will happen if you switch WIFIs on your laptop? All these scenarios will trigger a temporary ENETUNREACH or ENETDOWN, and should NOT be fatal for mounts or containers. Lionel
On Tue, 2025-03-25 at 20:40 +0100, Lionel Cons wrote: > On Tue, 25 Mar 2025 at 17:19, <trondmy@kernel.org> wrote: > > > > From: Trond Myklebust <trond.myklebust@hammerspace.com> > > > > With the recent patch series that caused containerised mounts which > > return ENETUNREACH or ENETDOWN errors to report fatal errors, we > > also > > want to ensure that the state manager thread also triggers fatal > > errors > > in the processes or threads that are waiting for recovery to > > complete. > > > > --- > > v2: > > - Return EIO instead of ENETUNREACH in nfs4_wait_clnt_recover() > > > > Trond Myklebust (4): > > SUNRPC: rpcbind should never reset the port to the value '0' > > SUNRPC: rpc_clnt_set_transport() must not change the autobind > > setting > > NFSv4: clp->cl_cons_state < 0 signifies an invalid nfs_client > > NFSv4: Treat ENETUNREACH errors as fatal for state recovery > > > > fs/nfs/nfs4state.c | 14 +++++++++++--- > > net/sunrpc/clnt.c | 3 --- > > net/sunrpc/rpcb_clnt.c | 5 +++-- > > 3 files changed, 14 insertions(+), 8 deletions(-) > > 1. Can this "ENETUNREACH or ENETDOWN are fatal" feature be turned > off? > > 2. We have concerns about this feature - what will happen if a switch > or router gets rebooted? What will happen if you unplug your laptop? > What will happen if you enable/disable your VPN software on a machine > or container? What will happen if you switch WIFIs on your laptop? > > All these scenarios will trigger a temporary ENETUNREACH or ENETDOWN, > and should NOT be fatal for mounts or containers. TL;DR: Yes, you can turn it off using a mount option, but we preserve the existing hard mount behaviour unless you are actually mounting in a container. See the description in the first series of patches: https://lore.kernel.org/linux-nfs/2a37ba5a-f212-4456-a7c1-3f96b1148b3b@oracle.com/T/#mf3df4516c5fcb0ef22c1c1c6f5433535f4d4805a
From: Trond Myklebust <trond.myklebust@hammerspace.com> With the recent patch series that caused containerised mounts which return ENETUNREACH or ENETDOWN errors to report fatal errors, we also want to ensure that the state manager thread also triggers fatal errors in the processes or threads that are waiting for recovery to complete. --- v2: - Return EIO instead of ENETUNREACH in nfs4_wait_clnt_recover() Trond Myklebust (4): SUNRPC: rpcbind should never reset the port to the value '0' SUNRPC: rpc_clnt_set_transport() must not change the autobind setting NFSv4: clp->cl_cons_state < 0 signifies an invalid nfs_client NFSv4: Treat ENETUNREACH errors as fatal for state recovery fs/nfs/nfs4state.c | 14 +++++++++++--- net/sunrpc/clnt.c | 3 --- net/sunrpc/rpcb_clnt.c | 5 +++-- 3 files changed, 14 insertions(+), 8 deletions(-)