There's no need to dynamically allocate this memory, and doing so may
create the possibility of races on shutdown of the rpc client. (We've
witnessed it only after adding rpcsec_gss support to the server, after
which the rpc code can send destroys calls that expect to still be able
to access the rpc_stats structure after it has been destroyed.)
Such races are in theory possible if the module containing this "static"
memory is removed very quickly after an rpc client is destroyed, but
we haven't seen that happen.
Signed-off-by: J. Bruce Fields <bfields@citi.umich.edu>
Move the code that actually parses the filehandle and looks up the
dentry and export to a separate function. This simplifies the reference
counting a little and moves fh_verify() a little closer to the kernel
ideal of small, minimally-indentended functions. Clean up a few other
minor style sins along the way.
Signed-off-by: J. Bruce Fields <bfields@citi.umich.edu>
Cc: Neil Brown <neilb@suse.de>
While lease is correctly checked by supplying the type argument to
vfs_setlease(), it's stored with fl_type uninitialized. This breaks the
logic when checking the type of the lease. The fix is to initialize
fl_type.
The old code still happened to function correctly since F_RDLCK is zero,
and we only implement read delegations currently (nor write
delegations). But that's no excuse for not fixing this.
Signed-off-by: Felix Blyakher <felixb@sgi.com>
Signed-off-by: J. Bruce Fields <bfields@citi.umich.edu>
fs/nfsd/vfs.c:991:27: warning: Using plain integer as NULL pointer
Signed-off-by: Harvey Harrison <harvey.harrison@gmail.com>
Signed-off-by: J. Bruce Fields <bfields@citi.umich.edu>
Add extern to nfsd/nfsd.h
fs/nfsd/nfssvc.c:146:5: warning: symbol 'nfsd_nrthreads' was not declared. Should it be static?
fs/nfsd/nfssvc.c:261:5: warning: symbol 'nfsd_nrpools' was not declared. Should it be static?
fs/nfsd/nfssvc.c:269:5: warning: symbol 'nfsd_get_nrthreads' was not declared. Should it be static?
fs/nfsd/nfssvc.c:281:5: warning: symbol 'nfsd_set_nrthreads' was not declared. Should it be static?
fs/nfsd/export.c:1534:23: warning: symbol 'nfs_exports_op' was not declared. Should it be static?
Add include of auth.h
fs/nfsd/auth.c:27:5: warning: symbol 'nfsd_setuser' was not declared. Should it be static?
Make static, move forward declaration closer to where it's needed.
fs/nfsd/nfs4state.c:1877:1: warning: symbol 'laundromat_main' was not declared. Should it be static?
Make static, forward declaration was already marked static.
fs/nfsd/nfs4idmap.c:206:1: warning: symbol 'idtoname_parse' was not declared. Should it be static?
fs/nfsd/vfs.c:1156:1: warning: symbol 'nfsd_create_setattr' was not declared. Should it be static?
Signed-off-by: Harvey Harrison <harvey.harrison@gmail.com>
Signed-off-by: J. Bruce Fields <bfields@citi.umich.edu>
This patch makes the needlessly global nfsd_create_setattr() static.
Signed-off-by: Adrian Bunk <bunk@kernel.org>
Signed-off-by: J. Bruce Fields <bfields@citi.umich.edu>
Likewise, distros usually leave CONFIG_NFSD_TCP enabled.
TCP support in the Linux NFS server is stable enough that we can leave it
on always. CONFIG_NFSD_TCP adds about 10 lines of code, and defaults to
"Y" anyway.
Signed-off-by: Chuck Lever <chuck.lever@oracle.com>
Signed-off-by: J. Bruce Fields <bfields@citi.umich.edu>
The code here is difficult to understand; attempt to clarify somewhat by
pulling out one of the more mystifying conditionals into a separate
function.
While we're here, also add lease_time to the list of attributes that we
don't really need to cross a mountpoint to fetch.
Signed-off-by: J. Bruce Fields <bfields@citi.umich.edu>
Cc: Peter Staubach <staubach@redhat.com>
This adds IPv6 support to the interfaces that are used to express nfsd
exports. All addressed are stored internally as IPv6; backwards
compatibility is maintained using mapped addresses.
Thanks to Bruce Fields, Brian Haley, Neil Brown and Hideaki Joshifuji
for comments
Signed-off-by: Aurelien Charbon <aurelien.charbon@bull.net>
Cc: Neil Brown <neilb@suse.de>
Cc: Brian Haley <brian.haley@hp.com>
Cc: YOSHIFUJI Hideaki / 吉藤英明 <yoshfuji@linux-ipv6.org>
Signed-off-by: J. Bruce Fields <bfields@citi.umich.edu>
If we depend on the inodes for writeability, we will not catch the r/o mounts
when implemented.
This patches uses __mnt_want_write(). It does not guarantee that the mount
will stay writeable after the check. But, this is OK for one of the checks
because it is just for a printk().
The other two are probably unnecessary and duplicate existing checks in the
VFS. This won't make them better checks than before, but it will make them
detect r/o mounts.
Acked-by: Al Viro <viro@ZenIV.linux.org.uk>
Signed-off-by: Christoph Hellwig <hch@infradead.org>
Signed-off-by: Dave Hansen <haveblue@us.ibm.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
This basically audits the callers of xattr_permission(), which calls
permission() and can perform writes to the filesystem.
[AV: add missing parts - removexattr() and nfsd posix acls, plug for a leak
spotted by Miklos]
Acked-by: Al Viro <viro@ZenIV.linux.org.uk>
Signed-off-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Dave Hansen <haveblue@us.ibm.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
This also uses the little helper in the NFS code to make an if() a little bit
less ugly. We introduced the helper at the beginning of the series.
Acked-by: Al Viro <viro@ZenIV.linux.org.uk>
Signed-off-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Dave Hansen <haveblue@us.ibm.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
This takes care of all of the direct callers of vfs_mknod().
Since a few of these cases also handle normal file creation
as well, this also covers some calls to vfs_create().
So that we don't have to make three mnt_want/drop_write()
calls inside of the switch statement, we move some of its
logic outside of the switch and into a helper function
suggested by Christoph.
This also encapsulates a fix for mknod(S_IFREG) that Miklos
found.
[AV: merged mkdir handling, added missing nfsd pieces]
Acked-by: Al Viro <viro@ZenIV.linux.org.uk>
Signed-off-by: Christoph Hellwig <hch@infradead.org>
Signed-off-by: Dave Hansen <haveblue@us.ibm.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
Elevate the write count during the vfs_rmdir() and vfs_unlink().
[AV: merged rmdir and unlink parts, added missing pieces in nfsd]
Acked-by: Serge Hallyn <serue@us.ibm.com>
Acked-by: Al Viro <viro@ZenIV.linux.org.uk>
Signed-off-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Dave Hansen <haveblue@us.ibm.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
If someone decides to demote a file from r/w to just
r/o, they can use this same code as __fput().
NFS does just that, and will use this in the next
patch.
AV: drop write access in __fput() only after we evict from file list.
Signed-off-by: Dave Hansen <haveblue@us.ibm.com>
Cc: Erez Zadok <ezk@cs.sunysb.edu>
Cc: Trond Myklebust <trond.myklebust@fys.uio.no>
Cc: "J Bruce Fields" <bfields@fieldses.org>
Acked-by: Al Viro <viro@ZenIV.linux.org.uk>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
This bug was always here, but before my commit 6fa02839bf
("recheck for secure ports in fh_verify"), it could only be triggered by
failure of a kmalloc(). After that commit it could be triggered by a
client making a request from a non-reserved port for access to an export
marked "secure". (Exports are "secure" by default.)
The result is a struct svc_export with a reference count one too low,
resulting in likely oopses next time the export is accessed.
The reference counting here is not straightforward; a later patch will
clean up fh_verify().
Thanks to Lukas Hejtmanek for the bug report and followup.
Signed-off-by: J. Bruce Fields <bfields@citi.umich.edu>
Cc: Lukas Hejtmanek <xhejtman@ics.muni.cz>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
Sorry for the noise, but here's the v3 of this compilation fix :)
There are some places, which declare the char buf[...] on the stack
to push it later into dprintk(). Since the dprintk sometimes (if the
CONFIG_SYSCTL=n) becomes an empty do { } while (0) stub, these buffers
cause gcc to produce appropriate warnings.
Wrap these buffers with RPC_IFDEBUG macro, as Trond proposed, to
compile them out when not needed.
Signed-off-by: Pavel Emelyanov <xemul@openvz.org>
Acked-by: J. Bruce Fields <bfields@citi.umich.edu>
Signed-off-by: Trond Myklebust <Trond.Myklebust@netapp.com>
d_path() is used on a <dentry,vfsmount> pair. Lets use a struct path to
reflect this.
[akpm@linux-foundation.org: fix build in mm/memory.c]
Signed-off-by: Jan Blunck <jblunck@suse.de>
Acked-by: Bryan Wu <bryan.wu@analog.com>
Acked-by: Christoph Hellwig <hch@infradead.org>
Cc: Al Viro <viro@zeniv.linux.org.uk>
Cc: "J. Bruce Fields" <bfields@fieldses.org>
Cc: Neil Brown <neilb@suse.de>
Cc: Michael Halcrow <mhalcrow@us.ibm.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
seq_path() is always called with a dentry and a vfsmount from a struct path.
Make seq_path() take it directly as an argument.
Signed-off-by: Jan Blunck <jblunck@suse.de>
Cc: Christoph Hellwig <hch@lst.de>
Cc: Al Viro <viro@zeniv.linux.org.uk>
Cc: "J. Bruce Fields" <bfields@fieldses.org>
Cc: Neil Brown <neilb@suse.de>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
I'm embedding struct path into struct svc_expkey.
Signed-off-by: Jan Blunck <jblunck@suse.de>
Cc: Al Viro <viro@zeniv.linux.org.uk>
Acked-by: "J. Bruce Fields" <bfields@fieldses.org>
Cc: Neil Brown <neilb@suse.de>
Cc: Christoph Hellwig <hch@lst.de>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
* Add path_put() functions for releasing a reference to the dentry and
vfsmount of a struct path in the right order
* Switch from path_release(nd) to path_put(&nd->path)
* Rename dput_path() to path_put_conditional()
[akpm@linux-foundation.org: fix cifs]
Signed-off-by: Jan Blunck <jblunck@suse.de>
Signed-off-by: Andreas Gruenbacher <agruen@suse.de>
Acked-by: Christoph Hellwig <hch@lst.de>
Cc: <linux-fsdevel@vger.kernel.org>
Cc: Al Viro <viro@zeniv.linux.org.uk>
Cc: Steven French <sfrench@us.ibm.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
This is the central patch of a cleanup series. In most cases there is no good
reason why someone would want to use a dentry for itself. This series reflects
that fact and embeds a struct path into nameidata.
Together with the other patches of this series
- it enforced the correct order of getting/releasing the reference count on
<dentry,vfsmount> pairs
- it prepares the VFS for stacking support since it is essential to have a
struct path in every place where the stack can be traversed
- it reduces the overall code size:
without patch series:
text data bss dec hex filename
5321639 858418 715768 6895825 6938d1 vmlinux
with patch series:
text data bss dec hex filename
5320026 858418 715768 6894212 693284 vmlinux
This patch:
Switch from nd->{dentry,mnt} to nd->path.{dentry,mnt} everywhere.
[akpm@linux-foundation.org: coding-style fixes]
[akpm@linux-foundation.org: fix cifs]
[akpm@linux-foundation.org: fix smack]
Signed-off-by: Jan Blunck <jblunck@suse.de>
Signed-off-by: Andreas Gruenbacher <agruen@suse.de>
Acked-by: Christoph Hellwig <hch@lst.de>
Cc: Al Viro <viro@zeniv.linux.org.uk>
Cc: Casey Schaufler <casey@schaufler-ca.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
The patch supports legacy (32-bit) capability userspace, and where possible
translates 32-bit capabilities to/from userspace and the VFS to 64-bit
kernel space capabilities. If a capability set cannot be compressed into
32-bits for consumption by user space, the system call fails, with -ERANGE.
FWIW libcap-2.00 supports this change (and earlier capability formats)
http://www.kernel.org/pub/linux/libs/security/linux-privs/kernel-2.6/
[akpm@linux-foundation.org: coding-syle fixes]
[akpm@linux-foundation.org: use get_task_comm()]
[ezk@cs.sunysb.edu: build fix]
[akpm@linux-foundation.org: do not initialise statics to 0 or NULL]
[akpm@linux-foundation.org: unused var]
[serue@us.ibm.com: export __cap_ symbols]
Signed-off-by: Andrew G. Morgan <morgan@kernel.org>
Cc: Stephen Smalley <sds@tycho.nsa.gov>
Acked-by: Serge Hallyn <serue@us.ibm.com>
Cc: Chris Wright <chrisw@sous-sol.org>
Cc: James Morris <jmorris@namei.org>
Cc: Casey Schaufler <casey@schaufler-ca.com>
Signed-off-by: Erez Zadok <ezk@cs.sunysb.edu>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
Neil Brown points out that we're checking buf[size-1] in a couple places
without first checking whether size is zero.
Actually, given the implementation of simple_transaction_get(), buf[-1]
is zero, so in both of these cases the subsequent check of the value of
buf[size-1] will catch this case.
But it seems fragile to depend on that, so add explicit checks for this
case.
Signed-off-by: J. Bruce Fields <bfields@citi.umich.edu>
Acked-by: NeilBrown <neilb@suse.de>
Neither EPERM and ENOENT map to valid errors for PUTROOTFH according to
rfc 3530, and, if anything, ENOENT is likely to be slightly more
informative; so don't bother mapping ENOENT to EPERM. (Probably this
was originally done because one likely cause was that there is an fsid=0
export but that it isn't permitted to this particular client. Now that
we allow WRONGSEC returns, this is somewhat less likely.)
In the long term we should work to make this situation less likely,
perhaps by turning off nfsv4 service entirely in the absence of the
pseudofs root, or constructing a pseudofilesystem root ourselves in the
kernel as necessary.
Thanks to Benny Halevy <bhalevy@panasas.com> for pointing out this
problem.
Signed-off-by: J. Bruce Fields <bfields@citi.umich.edu>
Cc: Benny Halevy <bhalevy@panasas.com>
Create a transport independent version of the svc_sock_names function.
The toclose capability of the svc_sock_names service can be implemented
using the svc_xprt_find and svc_xprt_close services.
Signed-off-by: Tom Tucker <tom@opengridcomputing.com>
Acked-by: Neil Brown <neilb@suse.de>
Reviewed-by: Chuck Lever <chuck.lever@oracle.com>
Reviewed-by: Greg Banks <gnb@sgi.com>
Signed-off-by: J. Bruce Fields <bfields@citi.umich.edu>
Update the write handler for the portlist file to allow creating new
listening endpoints on a transport. The general form of the string is:
<transport_name><space><port number>
For example:
echo "tcp 2049" > /proc/fs/nfsd/portlist
This is intended to support the creation of a listening endpoint for
RDMA transports without adding #ifdef code to the nfssvc.c file.
Transports can also be removed as follows:
'-'<transport_name><space><port number>
For example:
echo "-tcp 2049" > /proc/fs/nfsd/portlist
Attempting to add a listener with an invalid transport string results
in EPROTONOSUPPORT and a perror string of "Protocol not supported".
Attempting to remove an non-existent listener (.e.g. bad proto or port)
results in ENOTCONN and a perror string of
"Transport endpoint is not connected"
Signed-off-by: Tom Tucker <tom@opengridcomputing.com>
Acked-by: Neil Brown <neilb@suse.de>
Reviewed-by: Chuck Lever <chuck.lever@oracle.com>
Reviewed-by: Greg Banks <gnb@sgi.com>
Signed-off-by: J. Bruce Fields <bfields@citi.umich.edu>
Move sk_list and sk_ready to svc_xprt. This involves close because these
lists are walked by svcs when closing all their transports. So I combined
the moving of these lists to svc_xprt with making close transport independent.
The svc_force_sock_close has been changed to svc_close_all and takes a list
as an argument. This removes some svc internals knowledge from the svcs.
This code races with module removal and transport addition.
Thanks to Simon Holm Thøgersen for a compile fix.
Signed-off-by: Tom Tucker <tom@opengridcomputing.com>
Acked-by: Neil Brown <neilb@suse.de>
Reviewed-by: Chuck Lever <chuck.lever@oracle.com>
Reviewed-by: Greg Banks <gnb@sgi.com>
Signed-off-by: J. Bruce Fields <bfields@citi.umich.edu>
Cc: Simon Holm Thøgersen <odie@cs.aau.dk>
Modify the various kernel RPC svcs to use the svc_create_xprt service.
Signed-off-by: Tom Tucker <tom@opengridcomputing.com>
Acked-by: Neil Brown <neilb@suse.de>
Reviewed-by: Chuck Lever <chuck.lever@oracle.com>
Reviewed-by: Greg Banks <gnb@sgi.com>
Signed-off-by: J. Bruce Fields <bfields@citi.umich.edu>
Document these checks a little better and inline, as suggested by Neil
Brown (note both functions have two callers). Remove an obviously bogus
check while we're there (checking whether unsigned value is negative).
Signed-off-by: J. Bruce Fields <bfields@citi.umich.edu>
Cc: Neil Brown <neilb@suse.de>
The server silently ignores attempts to set the uid and gid on create.
Based on the comment, this appears to have been done to prevent some
overly-clever IRIX client from causing itself problems.
Perhaps we should remove that hack completely. For now, at least, it
makes sense to allow root (when no_root_squash is set) to set uid and
gid.
While we're there, since nfsd_create and nfsd_create_v3 share the same
logic, pull that out into a separate function. And spell out the
individual modifications of ia_valid instead of doing them both at once
inside a conditional.
Thanks to Roger Willcocks <roger@filmlight.ltd.uk> for the bug report
and original patch on which this is based.
Signed-off-by: J. Bruce Fields <bfields@citi.umich.edu>
This patch addresses a compatibility issue with a Linux NFS server and
AIX NFS client.
I have exported /export as fsid=0 with sec=krb5:krb5i
I have mount --bind /home onto /export/home
I have exported /export/home with sec=krb5i
The AIX client mounts / -o sec=krb5:krb5i onto /mnt
If I do an ls /mnt, the AIX client gets a permission error. Looking at
the network traceIwe see a READDIR looking for attributes
FATTR4_RDATTR_ERROR and FATTR4_MOUNTED_ON_FILEID. The response gives a
NFS4ERR_WRONGSEC which the AIX client is not expecting.
Since the AIX client is only asking for an attribute that is an
attribute of the parent file system (pseudo root in my example), it
seems reasonable that there should not be an error.
In discussing this issue with Bruce Fields, I initially proposed
ignoring the error in nfsd4_encode_dirent_fattr() if all that was being
asked for was FATTR4_RDATTR_ERROR and FATTR4_MOUNTED_ON_FILEID, however,
Bruce suggested that we avoid calling cross_mnt() if only these
attributes are requested.
The following patch implements bypassing cross_mnt() if only
FATTR4_RDATTR_ERROR and FATTR4_MOUNTED_ON_FILEID are called. Since there
is some complexity in the code in nfsd4_encode_fattr(), I didn't want to
duplicate code (and introduce a maintenance nightmare), so I added a
parameter to nfsd4_encode_fattr() that indicates whether it should
ignore cross mounts and simply fill in the attribute using the passed in
dentry as opposed to it's parent.
Signed-off-by: Frank Filz <ffilzlnx@us.ibm.com>
Signed-off-by: J. Bruce Fields <bfields@citi.umich.edu>
The failure to return a stateowner from nfs4_preprocess_seqid_op() means
in the case where a lock request is of a type incompatible with an open
(due to, e.g., an application attempting a write lock on a file open for
read), means that fs/nfsd/nfs4xdr.c:ENCODE_SEQID_OP_TAIL() never bumps
the seqid as it should. The client, attempting to close the file
afterwards, then gets an (incorrect) bad sequence id error. Worse, this
prevents the open file from ever being closed, so we leak state.
Thanks to Benny Halevy and Trond Myklebust for analysis, and to Steven
Wilton for the report and extensive data-gathering.
Cc: Benny Halevy <bhalevy@panasas.com>
Cc: Steven Wilton <steven.wilton@team.eftel.com.au>
Cc: Trond Myklebust <trond.myklebust@fys.uio.no>
Signed-off-by: J. Bruce Fields <bfields@citi.umich.edu>
When the callback channel fails, we inform the client of that by
returning a cb_path_down error the next time it tries to renew its
lease.
If we wait most of a lease period before deciding that a callback has
failed and that the callback channel is down, then we decrease the
chances that the client will find out in time to do anything about it.
So, mark the channel down as soon as we recognize that an rpc has
failed. However, continue trying to recall delegations anyway, in hopes
it will come back up. This will prevent more delegations from being
given out, and ensure cb_path_down is returned to renew calls earlier,
while still making the best effort to deliver recalls of existing
delegations.
Also fix a couple comments and remove a dprink that doesn't seem likely
to be useful.
Signed-off-by: J. Bruce Fields <bfields@citi.umich.edu>
Declare this variable in the one function where it's used, and clean up
some minor style problems.
Signed-off-by: J. Bruce Fields <bfields@citi.umich.edu>
We generate a unique cl_confirm for every new client; so if we've
already checked that this cl_confirm agrees with the cl_confirm of
unconf, then we already know that it does not agree with the cl_confirm
of conf.
Signed-off-by: J. Bruce Fields <bfields@citi.umich.edu>
Again, the only way conf and unconf can have the same clientid is if
they were created in the "probable callback update" case of setclientid,
in which case we already know that the cl_verifier fields must agree.
Signed-off-by: J. Bruce Fields <bfields@citi.umich.edu>
If conf and unconf are both found in the lookup by cl_clientid, then
they share the same cl_clientid. We always create a unique new
cl_clientid field when creating a new client--the only exception is the
"probable callback update" case in setclientid, where we copy the old
cl_clientid from another clientid with the same name.
Therefore two clients with the same cl_client field also always share
the same cl_name field, and a couple of the checks here are redundant.
Thanks to Simon Holm Thøgersen for a compile fix.
Signed-off-by: J. Bruce Fields <bfields@citi.umich.edu>
Cc: Simon Holm Thøgersen <odie@cs.aau.dk>
Using a counter instead of the nanoseconds value seems more likely to
produce a unique cl_confirm.
Signed-off-by: J. Bruce Fields <bfields@citi.umich.edu>
We're supposed to generate a different cl_confirm verifier for each new
client, so these to cl_confirm values should never be the same.
Signed-off-by: J. Bruce Fields <bfields@citi.umich.edu>
Most of these comments just summarize the code.
The matching of code to the cases described in the RFC may still be
useful, though; add specific section references to make that easier to
follow. Also update references to the outdated RFC 3010.
Signed-off-by: J. Bruce Fields <bfields@citi.umich.edu>