Created attachment 107401 [details] NFS-Kconfig-select-DNS_RESOLVER.patch Upon building from defconfig and only selecting CONFIG_NFS_V4 as modular, people get the following error when building: fs/built-in.o: In function `nfs_dns_resolve_name': (.text+0x1170cc): undefined reference to `dns_query' make: *** [vmlinux] Error 1 Attached you will find a patch to fix this by Ben Kohler whom explains: > What seems to be going on is: marking CONFIG_NFS_V4=m also selects > CONFIG_DNS_RESOLVER=m, but it actually also causes the built-in > CONFIG_NFS_FS=y (nfs.o) to use some functions (like dns_query) from the > modular dns_resolv.o. This was discovered by Jaime Martin. Downstream bug at https://bugs.gentoo.org/show_bug.cgi?id=473724
Comment on attachment 107401 [details] NFS-Kconfig-select-DNS_RESOLVER.patch commit c2e8139c9f797baa46515af6d350c51823736cbb (NFS: Use kernel DNS resolver [ver #2]) already selects DNS_RESOLVER as part of the config NFS_USE_KERNEL_DNS definition. If the first select failed, how will selecting it again help?
OK, hang on... You're also apparently missing commit c8d74d9b68b655e85ee4603f8918c3233a74f085 (NFSv4: Move the DNS resolver into the NFSv4 module)...
(In reply to Trond Myklebust from comment #1) > Comment on attachment 107401 [details] > NFS-Kconfig-select-DNS_RESOLVER.patch > > commit c2e8139c9f797baa46515af6d350c51823736cbb (NFS: Use kernel DNS > resolver [ver #2]) already selects DNS_RESOLVER as part of the config > NFS_USE_KERNEL_DNS definition. > If the first select failed, how will selecting it again help? Hmm, odd, then perhaps the reporting user did select another option so NFS_USE_KERNEL_DNS was not satisfied; so under the condition that some other option that it depends on was changed it was possible to put it in that state. Anyhow... (In reply to Trond Myklebust from comment #2) > OK, hang on... > > You're also apparently missing commit > c8d74d9b68b655e85ee4603f8918c3233a74f085 (NFSv4: Move the DNS resolver into > the NFSv4 module)... Ah, oops; sorry, I only checked whether Kconfig changed. So yeah, that indeed sounds like the commit that would deal with this and no longer require the Kconfig relation anymore; thank you very much for pointing this out.