Bug 1097802

Description of problem:
This problem part of a patch that has been applied in the RHEL 6.5 release set.
There is some information about the problem here https://access.redhat.com/site/solutions/634593 however I have no access to this document.

The spec file indicates that this is part of bugzilla bug 10006221 (possibly 1006221) but I have no access to this bug either.

Finally the changelog states that in release 8.4-27
"tail -F now disables inotify when encountering a symlink. Polling is used instead."

I understand the need for this patch, as a means to detect renames and unlinks of the watch file and that inotify currently is unable to do this properly when the file being watched is a symlink however, the patch is incorrectly written and generates too much scope.

In the RPM sources, the patch labelled "coreutils-8.4-tail-symlink-polling.patch" achieves this purpose using the following logic:

    if (!lstat(F[i].name, &st) && S_ISLNK(st.st_mode))

However, this is not what I believe ooprala had in mind, as this will apply to any symlink, regardless of follow mode.

This has the effect of producing a warning using a simple tail command to print out any file that is a symlink, irrespective of if you intend to follow it or not.

Please alter the patch so that it uses this logic instead:

    if (!lstat(F[i].name, &st) && S_ISLNK(st.st_mode) && follow_mode == Follow_name)

This ensure that we only produce the warning if we pass tail -F (since tail -f is the implicit default) and prevents warnings being printed when our intention is to merely tail without following.

We also avoid disabling inotify where we actually are interested in following the file descriptor itself where it would be more beneficial and efficient to do so.

Version-Release number of selected component (if applicable):
8.4-27 ? and above.


How reproducible:
100%


Steps to Reproduce:
1. Create a symlink.
2. Run "tail path-to-symlink"

Actual results:
Tail prints to stderr a warning about not using inotify and to stdout the typical tail output.

Expected results:
Just the typical tail output.
If we want to name follow then print the warning if the user passes tail -F path-to-symlink.

Additional Info:
I did not supply a patch as it would end up ultimately being a patch to a patch, which just didn't feel appropriate to me.