Re: [PATCH 0/3] *at syscalls: Intro

[Jeff Garzik <jgarzik@pobox.com>]

Ulrich Drepper wrote:
> Here is a series of patches which introduce in total 11 new system calls
> which take a file descriptor/filename pair instead of a single file name.
> These functions, openat etc, have been discussed on numerous occasions.
> They are needed to implement race-free filesystem traversal, they are
> necessary to implement a virtual per-thread current working directory
> (think multi-threaded backup software), etc.
> 
> We have in glibc today implementations of the interfaces which use the
> /proc/self/fd magic.  But this code is rather expensive.  Here are some
> results (similar to what Jim Meyering posted before):
> 
> The test creates a deep directory hierarchy on a tmpfs filesystem.
> Then rm -fr is used to remove all directories.  Without syscall support
> I get this:
> 
> real    0m31.921s
> user    0m0.688s
> sys     0m31.234s
> 
> With syscall support the results are much better:
> 
> real    0m20.699s
> user    0m0.536s
> sys     0m20.149s
> 
> 
> The implemenation is really small:
> 
>  arch/i386/kernel/syscall_table.S |   11 ++
>  arch/x86_64/ia32/ia32entry.S     |   11 ++
>  fs/compat.c                      |   48 +++++++++--
>  fs/exec.c                        |    2 
>  fs/namei.c                       |  167 ++++++++++++++++++++++++++++++---------
>  fs/open.c                        |   60 +++++++++++---
>  fs/stat.c                        |   54 ++++++++++--
>  include/asm-i386/unistd.h        |   13 ++-
>  include/asm-x86_64/ia32_unistd.h |   13 ++-
>  include/asm-x86_64/unistd.h      |   24 +++++
>  include/linux/fcntl.h            |    7 +
>  include/linux/fs.h               |    7 +
>  include/linux/namei.h            |    7 -
>  include/linux/time.h             |    2 
>  14 files changed, 355 insertions(+), 71 deletions(-)
> 
> I've split the patch in three parts.  The first part is the actual
> code change.  It mostly consists of passing down an additional parameter
> with a file descriptor and add wrapper functions which pass down the
> default parameter AT_FDCWD.  Three new constants are defined in
> <linux/fcntl.h> which must correspond to the values already in use
> in glibc.  In a few cases I've modified some code which would not
> necesarily need changing but the change makes it a bit more efficient
> in presence of the wrapper functions.
> 
> The real change needed is the additional else-clause in what is now
> do_path_lookup.  That's it.

Definitely gets my ACK, for my own motivations:  I want to create 
race-free cp(1)/mv(1)/rm(1) utilities for my posixutils project.

It would be nice to add the additional argument to path_lookup(), rather 
than calling do_path_lookup() everywhere.

	Jeff