Messages in this thread | | | Date | Mon, 24 Nov 1997 19:33:22 -0500 (EST) | From | "Steven N. Hirsch" <> | Subject | Re: updated smbfs patch |
| |
On Sun, 23 Nov 1997, Bill Hawes wrote:
> Hi Steve, > I've done a substantial reworking of the timestamp-related code and have > things working quite a bit better. There are so many bugs in NT and Win > 95 to work around that I'm now maintaining mtime and atime locally and > posting updates to the server when a file is closed. > > One nice side effect of these changes is that writes are about 3X faster > on the low end, by virtue of eliminating the getattr call. Bonnie and > iozone now return more reasonable write speeds. > > One thing to watch out for -- do not enable the Win 95 bug CONFIG option > unless you intend to talk to a Win 95 server. The CONFIG option avoids a > Win 95 bug, but exposes a bug in NT in that core protocol timestamps > aren't correct. Eventually I'll have a runtime test for this. > > Please test out the attached patch and let me know of any remaining > anomalies.
Bill,
I tested your code against NT 4.0, and the time/date stuff is working correctly. The write speed is improved by ~4x also - not too shabby!
However, the problem persists with Linux not detecting new files created on the server. Once the file exists on the server, no incantation will cause it to show up in a directory listing (though it can be directly stat-ed) until there exists no Linux process with the server as current working directory. Then, after an ls is done for the full pathname to the server directory it appears in listings from that point onward.
Not sure if I'm making sense (long day), but it's 100% repeatable here.
I think we really need to automagically detect the type of server being mounted, as many users will have a conglomeration of Win95/WfW/NT boxes.
Steve
| |