My wife and I use a couple of desktops at home for most of our computing needs. Most of our user files (e.g., /usr/home) are stored on a separate server, shared over NFS and amounted via autofs. Said server also provides an NFS export that we use for our shared documents - stuff that we both need read-write access to. For the sake of simplicity I take advantage of the -mapall parameter in the exports file on the server. This nifty little option ensures that whomever has access to mount the NFS export read-write will read, write and delete files as a single user on the server. This effectively solves the problem of setting up more complicated permissions for file sharing between a small group of people. The /etc/exports file looks like this:

/usr/home       -alldirs                client1 client2
/srv/fileshare  -mapall=fileshare       client1 client2

The problem with this setup rears its ugly head when you're interacting with the NFS mounts on the client through Thunar, the default file manager for XFCE (I've come to appreciate a lot of what XFCE brings to the table over the past 6 or 7 years since we escaped the insanity that is Gnome3). Thunar uses some glib magic for interacting with the file system, and when a user hits the "Delete" key or picks to "Move to Trash" from the context menu, a poorly-documented gvfs binary named gvfs-trash is invoked and in both cases it is supposed to move the offending file to the user's waste bin. On the NFS mount this function failed with the following unhelpful message:

Error trashing file: Unable to find or create trash directory.

After digging around with truss to figure out what the heck gvfs-trash was actually doing and scouring the Trash specification, I discovered that gvfs-trash is supposed to look for a directory named .Trash in the root of the NFS mount and that this directory must have the sticky bit set.

Okay. Makes sense. I guess.

# mkdir /srv/fileshare/.Trash
# chmod 1777 /srv/fileshare/.Trash

At first glance this seems to work fine. Running gvfs-trash FILE-TO-TRASH now executes without error and gvfs (and in turn glib) goes ahead and creates /srv/fileshare/.Trash/1001, puts a couple of sub-directories under the one named after my effective UID, moves the offending file and Bob's your uncle. Because we're using the -mapall option, everything is created using the user we specified on the server in the exports file (in this case the user name is fileshare).

Unfortunately the problem also occurs the second time gvfs-trash is run to trash a file. It fails again with the same damned error message. Okay, thanks gvfs.

It turns out that not only must .Trash exist with the sticky bit, but the UID directory immediately below must be owned by the user who is trashing the file, never mind that this didn't matter when we were trashing the first file. To "fix" this problem I ran the following commands:

# chown matt:matt /srv/fileshare/.Trash/1001
# chmod 1777 /srv/fileshare/.Trash/1001

Since I have a limited number of user accounts this ungainly fix appears to work, at least for the short-term.

I think the moral of the story here is that when you are implementing anything that tries to be halfway "smart" about what it is doing (in this case, gvfs-trash and the ungodly mess of glib below it), you need to be VERY careful that your process a) provides a more helpful error message when it fails, and b) doesn't work in weird and unexpected ways (in this case, working the first time but failing the second time). Generally speaking, your users shouldn't have to run system-level debugging tools and refer to the spec when easily returned and friendly error messages would otherwise save them a lot of time.