On Mar 3, 1:30 am, "f...@nk" <[email protected]> wrote: > When I provide a stat argument to the filler callback in the readdir() > function, then the getattr() function will not be called for that > directory entry on Linux. With MacFUSE however, the getattr() > function is still called, and the stat buffer that is passed to it is > empty.
How are you checking that the getattr() callback is not called on Linux? Are you just looking at libfuse's debug output (the -d argument)? I'll presume you are. That debug output shows low-level FUSE API operations. If you actually trace your file system's getattr () callback, you should see *that* being called on Linux even if you fill in the stat structure in filler(). The last I checked, Linux FUSE does not do much with the stat structure you populate in filler(). It takes only st_mode and potentially st_ino from it and discards the other stat contents. On Linux, when you do an "ls -l" in a directory, and in your user-space file system you do fill in the stat structure, you will not see a GETATTR for each directory entry in the debug output. Instead, you will see a LOOKUP for each directory entry. Both the GETATTR and LOOKUP low-level operations end up calling the high-level getattr() callback. Now, if you compare this with the libfuse debug output on Mac OS X, you'll indeed see GETATTRs, but then you will not see LOOKUPs like in the case of Linux. Based on what I said above, this means that one way or another, your file system's getattr() callback will still be called. Upon the first "ls -l", you should see a LOOKUP for each directory entry. On subsequent "ls -l"s, you'll see GETATTRs instead. This is because MacFUSE employs vnode name caching in the kernel. (That is, the lookups are cached.) Even though a name is found in the cache, if the attributes aren't valid, the kernel has to call up to user space. You can see the cache in action this way: cd to the mounted file system so that it becomes busy. Then try to unmount it from another shell. This will flush all vnodes except the root vnode, and the file system will fail to unmount becuase it is busy. Now repeat your "ls - l". You'll see that things go back to the first "ls -l" case. Because the name cache has been purged, you'll see LOOKUPs for each directory entry. That said, on both Mac OS X and Linux, there's potential for *marginally* optimizing things by better utilizing the stat information you might supply in filler(). Then again, even now, if you specify a large attr_timeout value, say, 60 seconds, you should not see GETATTR calls for that duration. > Is this normal behavior of MacFUSE? Yes it is. Rule of thumb: The nature of the Mac OS X and Linux file system layers are vastly different. The MacFUSE and Linux FUSE kernel extension implementations are also vastly different. MacFUSE gives you (a superset of) the FUSE API, but because of the aforementioned differences, behavioral differences will abound. Comparing number of callbacks for similar operations, sequence of callbacks received, etc., isn't going to be very useful. Amit --~--~---------~--~----~------------~-------~--~----~ You received this message because you are subscribed to the Google Groups "MacFUSE" group. To post to this group, send email to [email protected] To unsubscribe from this group, send email to [email protected] For more options, visit this group at http://groups.google.com/group/macfuse?hl=en -~----------~----~----~----~------~----~------~--~---
