On Sun, 2025-09-07 at 13:56 -0400, rtm@xxxxxxxxxxxxx wrote: > [You don't often get email from rtm@xxxxxxxxxxxxx. Learn why this is > important at https://aka.ms/LearnAboutSenderIdentification ;] > > In this code in fs/nfs/nfs4proc.c _nfs_open_and_get_state(): > > dentry = opendata->dentry; > if (d_really_is_negative(dentry)) { > struct dentry *alias; > d_drop(dentry); > alias = d_splice_alias(igrab(state->inode), dentry); > /* d_splice_alias() can't fail here - it's a non- > directory */ > if (alias) { > dput(ctx->dentry); > ctx->dentry = dentry = alias; > } > > d_splice_alias() can fail, returning ERR_PTR(-ELOOP). Then this call > later on causes a crash: No. It can't return ELOOP. The reason why it can't is that OP_OPEN is not allowed to return NFS4_OK for anything other than a regular file. Hence the server cannot return a directory, or anything which could otherwise be an ancestor to 'dentry'. Hence the comment above... IOW: Your implementation of OP_OPEN is broken because it only returns either NFS4ERR_NOENT or NFS4_OK, whereas in this case it should clearly have return NFS4ERR_ISDIR. -- Trond Myklebust Linux NFS client maintainer, Hammerspace trondmy@xxxxxxxxxx, trond.myklebust@xxxxxxxxxxxxxxx
