[
https://issues.apache.org/jira/browse/HADOOP-10014?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=13786302#comment-13786302
]
Daryn Sharp commented on HADOOP-10014:
--------------------------------------
I think it's going to be dicey and complicated to pass context, especially w/o
incompatible fs subclass changes, but it'd be interesting to see a POC.
Basically the path resolution needs to occur from the top down. I think it's
ok for individual filesystem to provide primitives like a getFileLinkStatus,
getLinkTarget, resolveLink, listLinkStatus, etc. But transparent resolution
for the existing API methods needs to occur from the top-down which is why it
may simpler to apply a "SymlinkFileSystem" derived from {{FilterFileSystem}}
atop an arbitrary fs. At that rate, maybe the existing APIs should return
symlinks and the symlink fs should handle resolving them... FileSystem.get
would then return a wrapped fs if symlinks are enabled.
> Symlink resolution is fundamentally broken for multi-layered filesystems
> ------------------------------------------------------------------------
>
> Key: HADOOP-10014
> URL: https://issues.apache.org/jira/browse/HADOOP-10014
> Project: Hadoop Common
> Issue Type: Sub-task
> Components: fs
> Affects Versions: 2.0.0-alpha, 3.0.0
> Reporter: Daryn Sharp
> Priority: Critical
>
> Symlink resolution is performed on a per-filesystem basis. In a
> multi-layered filesystem, the symlinks need to be resolved relative to the
> highest level filesystem in the stack. Otherwise, fs implementations like
> viewfs and chroot fs behave incorrectly. Absolute symlinks may violate the
> base of the chroot. Links that should have crossed viewfs mount points are
> again incorrectly resolved relative to the base filesystem.
> Symlink resolution has occur above the level of any individual fs to allow a
> multi-layered fs stack to work correctly, such as via a symlink-aware
> {{FilteredFileSystem}} that wraps any arbitrary fs to ensure links are
> resolved from the top-down of the stack.
--
This message was sent by Atlassian JIRA
(v6.1#6144)
