Comment 43 for bug 406466

I can imagine why mounting with serverino could made a difference. If the server has broken support for returning serverinode numbers (or at least for the specific findfirst level that returns search information + server inode numbers) then things like resume keys could get messed up which would make it hard to parse the response, or could cause us to not find the next entry in the search response if the directory has more than a certain number of files (perhaps 50).

To help the tracing it may be helpful to uncomment the tracing call (in fs/cifs/readdir.c)

/* cFYI(1, ("filldir on %s",pqst->name)); */

or to add a debug entry before the call to filldir (in the same file fs/cifs/readdir.c)

Obviously if we decide that the findfirst or findnext response from the server is messed up, then the next question is how to detect this in the client - and if we can detect it in the client, then we turn off serverinode numbers on the fly (after logging something about the broken server) - but first would be good to get wireshark to decode this level or decode some of it by hand.