- on huge files clients can timeout - we lose ACLs and EA information threrfore we should by default return NT_STATUS_NOT_SAME_DEVICE. commits 583de7b582956d3bec7e875d88ef16b3b8ac6e53 0769a1833aff2057e7f6ab05713d7fd6886d6040 2f1fa4f8ca020c5d96bd3ac8706c54cd881aaa03 a4dc50ef1273259c53f96ee4aaaef917e8daeffe address this problem. They make the old limited cross filesystem rename support an optional feature which can be enabled via the crossrename vfs module.
Created attachment 5471 [details] patch for 3.5.0
Comment on attachment 5471 [details] patch for 3.5.0 I think the patch is fine, but I'm not sure we should change the behavior between 3.5.x release would cause trouble in existing setups which just blindly upgrade the package.
I think we should not backport the change
for the records in case someone stumbles upon it: in 3.5.x we stay upstream with correct cross-fs rename behaviour of subdirectories and at the same time broken EA+ACL lossy cross-fs renames of single files. This will only be fixed for the 3.6 series.
Comment on attachment 5471 [details] patch for 3.5.0 +1 from me. I concur on making this 3.6.0 only.