Ok, I have implemented and tested this (tested ext2 only). I did it
without a new system call and it's quite backward/forward compatible. I
put the d_type value at the end of each record from getdents(), in such
a way that it can be recognised unambiguously. Patch available on
request.
These are the results:
- You forgot the S_ISSOCK case in ext2_set_de_type.
e2fsck knows about it, but new unix domain sockets are untyped...
[treescan is equivalent to "find" with no predicates: it prints the
names of all the filesystem entries it can find]
- With nothing in memory, treescan takes about the same
time to scan my disks with the d_type information as it does with
the O_DIRECTORY filetype optimisation. 2% CPU, ~53 seconds.
This is no surprise. Both methods avoid reading non-directory
inodes, and in this case the system is I/O bound.
- With everything in memory, treescan takes 0.41 seconds to scan my
disks with d_type information, and 1.06 seconds with the O_DIRECTORY
optimisation only. 100% CPU.
This saving must be due to fewer system calls and fewer directory
lookups. Note, I do not have any large directories on this system.
-- Jamie
-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@vger.rutgers.edu
Please read the FAQ at http://www.tux.org/lkml/