Note that I added process_path to the revised patch, it's not the same
code that was in 1.10.6. In 1.10.6 pathconf still failed for exactly the
reason you described, but defaulting to NAME_MAX always saved the day. I
feel quite stupid for staring at it for so long and not seeing the
missing process_path :)
On 07/15/2012 03:41 PM, Marcelo E. Magallon wrote:
Hi Martin,
thanks for the patches!
About the readdir one, I saw that change in the newer releases,
and even if I can see why it works, I'm not convinced it's
correct.
As you already know, FUSE sends something like "/src" in the
path argument. When passing that to pathconf, if "/src" doesn't
happen to exist in the global hierarchy, pathconf will return -1
because it can't determine the corresponding filesystem's
limits. If by chance it _does_ happen to exist, it won't
necessarily return a correct value, because it's comparing
"/src" vs "/something/else/src". If, say, "/something" is on a
different filesystem, then the information for "/src" and
"/something" doesn't necessarily match. By using process_path,
you are turning the absolute path into a relative one wrt to the
original filesystem, and everything works as expected.
So, unless I'm missing something, the call to pathconf will fail
most of the time and the code will be using the fallback value
most of the time, which is obviously not your intention.
Again, many thanks!
Marcelo
--
To UNSUBSCRIBE, email to debian-bugs-rc-requ...@lists.debian.org
with a subject of "unsubscribe". Trouble? Contact listmas...@lists.debian.org
