PATCH: Change in-kernel afs client filesystem name to 'kafs'

PATCH: Change in-kernel afs client filesystem name to 'kafs'

[Troy Benjegerdes]

[-- Attachment #1: Type: text/plain, Size: 414 bytes --]

This patch changes the in-kernel AFS client filesystem name to 'kafs',
as well as allowing the AFS cache manager port to be set as a module
parameter. This is usefull for having a system boot with the root
filesystem on afs with the kernel AFS client, while still having the
option of loading the OpenAFS kernel module for use as a read-write
filesystem later.

Signed-off-by: Troy Benjegerdes <hozer@hozed.org>



[-- Attachment #2: kafs.patch --]
[-- Type: text/plain, Size: 2480 bytes --]

# HG changeset patch
# User hozer@minbar-g5.scl.ameslab.gov
# Node ID f2b66f75ac858c7eacf681c73ddad3f0058df480
# Parent  28e3d8204fdbfc0f2bef6dd513ab79184047a1ab
Change mount name to 'kafs' and allow setting the cache manager port

diff -r 28e3d8204fdb -r f2b66f75ac85 fs/afs/main.c
--- a/fs/afs/main.c	Fri Jun 23 22:47:27 2006 +0700
+++ b/fs/afs/main.c	Fri Jun 23 18:11:35 2006 -0500
@@ -42,6 +42,11 @@ module_param(rootcell, charp, 0);
 module_param(rootcell, charp, 0);
 MODULE_PARM_DESC(rootcell, "root AFS cell name and VL server IP addr list");
 
+static uint rxrpc_port = 7001;
+
+module_param(rxrpc_port, uint, 0);
+MODULE_PARM_DESC(rxrpc_port, "kAFS cache manager port (default 7001)");
+
 
 static struct rxrpc_peer_ops afs_peer_ops = {
 	.adding		= afs_adding_peer,
@@ -57,7 +62,7 @@ static struct cachefs_netfs_operations a
 };
 
 struct cachefs_netfs afs_cache_netfs = {
-	.name			= "afs",
+	.name			= "kafs",
 	.version		= 0,
 	.ops			= &afs_cache_ops,
 };
@@ -71,7 +76,8 @@ static int __init afs_init(void)
 {
 	int loop, ret;
 
-	printk(KERN_INFO "kAFS: Red Hat AFS client v0.1 registering.\n");
+	printk(KERN_INFO "kAFS: Red Hat AFS client v0.1 registering, using port %d.\n",
+			rxrpc_port);
 
 	/* initialise the callback hash table */
 	spin_lock_init(&afs_cb_hash_lock);
@@ -113,7 +119,7 @@ static int __init afs_init(void)
 		goto error_kafstimod;
 
 	/* create the RxRPC transport */
-	ret = rxrpc_create_transport(7001, &afs_transport);
+	ret = rxrpc_create_transport(rxrpc_port, &afs_transport);
 	if (ret < 0)
 		goto error_kafsasyncd;
 
diff -r 28e3d8204fdb -r f2b66f75ac85 fs/afs/mntpt.c
--- a/fs/afs/mntpt.c	Fri Jun 23 22:47:27 2006 +0700
+++ b/fs/afs/mntpt.c	Fri Jun 23 18:11:35 2006 -0500
@@ -203,7 +203,7 @@ static struct vfsmount *afs_mntpt_do_aut
 
 	/* try and do the mount */
 	kdebug("--- attempting mount %s -o %s ---", devname, options);
-	mnt = do_kern_mount("afs", 0, devname, options);
+	mnt = do_kern_mount("kafs", 0, devname, options);
 	kdebug("--- mount result %p ---", mnt);
 
 	free_page((unsigned long) devname);
diff -r 28e3d8204fdb -r f2b66f75ac85 fs/afs/super.c
--- a/fs/afs/super.c	Fri Jun 23 22:47:27 2006 +0700
+++ b/fs/afs/super.c	Fri Jun 23 18:11:35 2006 -0500
@@ -50,7 +50,7 @@ static void afs_destroy_inode(struct ino
 
 static struct file_system_type afs_fs_type = {
 	.owner		= THIS_MODULE,
-	.name		= "afs",
+	.name		= "kafs",
 	.get_sb		= afs_get_sb,
 	.kill_sb	= kill_anon_super,
 	.fs_flags	= FS_BINARY_MOUNTDATA,

Re: PATCH: Change in-kernel afs client filesystem name to 'kafs'

[Arjan van de Ven]

On Fri, 2006-06-23 at 19:41 -0500, Troy Benjegerdes wrote:
> This patch changes the in-kernel AFS client filesystem name to 'kafs',
> as well as allowing the AFS cache manager port to be set as a module
> parameter. This is usefull for having a system boot with the root
> filesystem on afs with the kernel AFS client, while still having the
> option of loading the OpenAFS kernel module for use as a read-write
> filesystem later.

sounds weird... the filesystem it implements is afs.
your change also breaks userspace, since the fs type is a mount option
so your change is userspace visible and means people need to fix their
scripts...

maybe openafs should start using "openafs" as type; they're not in the
kernel so they aren't yet bound by the userspace ABI....

Re: PATCH: Change in-kernel afs client filesystem name to 'kafs'

[Troy Benjegerdes]

On Sat, Jun 24, 2006 at 02:16:00PM +0200, Arjan van de Ven wrote:
> On Fri, 2006-06-23 at 19:41 -0500, Troy Benjegerdes wrote:
> > This patch changes the in-kernel AFS client filesystem name to 'kafs',
> > as well as allowing the AFS cache manager port to be set as a module
> > parameter. This is usefull for having a system boot with the root
> > filesystem on afs with the kernel AFS client, while still having the
> > option of loading the OpenAFS kernel module for use as a read-write
> > filesystem later.
> 
> sounds weird... the filesystem it implements is afs.
> your change also breaks userspace, since the fs type is a mount option
> so your change is userspace visible and means people need to fix their
> scripts...
> 
> maybe openafs should start using "openafs" as type; they're not in the
> kernel so they aren't yet bound by the userspace ABI....

This is the most idiotic argument I have ever heard, and completely
ignores reality. The kernel module is named already named "kafs", not "afs"
and *NOBODY* is using it because it does not have authentication or
read-write support.

If you want to talk about breaking userspace and existing scripts, the
AFS filesystem (of which the OpenAFS code is derived from) has existed
and been in production far longer than Linux.

( see http://portal.acm.org/citation.cfm?doid=35037.35059 )

At Iowa State University alone there are around 20K people that depend
on OpenAFS running on Linux for webmail service.

The only scripts I could think of that would be impacted by changing the
kernel afs mount name to 'kafs' would be the ones I am writing, and anything
David Howells has. That amounts to *maybe* 10 users.

If you really care about breaking long-standing userspace ABI's, will
you accept a patch allowing kAFS, OpenAFS, and Arla to register to
recieve a callback when the AFS syscall number is called?

Now, if you could suggest another way to have both the *experimental* 
in-kernel AFS client and OpenAFS client with mounted filesystems, please
let me know.

Re: PATCH: Change in-kernel afs client filesystem name to 'kafs'

[Christoph Hellwig]

On Sat, Jun 24, 2006 at 01:01:41PM -0500, Troy Benjegerdes wrote:
> Now, if you could suggest another way to have both the *experimental* 
> in-kernel AFS client and OpenAFS client with mounted filesystems, please
> let me know.

Don't do that.  Or at least stay far away from the vger lists with any
bugreports or suggestions that come up when running OpenAFS.  It's some
of the worst code around, and it has a license that doesn't event permit 
the blatantly stupid poking into kernel internals it does.

Re: PATCH: Change in-kernel afs client filesystem name to 'kafs'

[Troy Benjegerdes]

On Sat, Jun 24, 2006 at 07:09:53PM +0100, Christoph Hellwig wrote:
> On Sat, Jun 24, 2006 at 01:01:41PM -0500, Troy Benjegerdes wrote:
> > Now, if you could suggest another way to have both the *experimental* 
> > in-kernel AFS client and OpenAFS client with mounted filesystems, please
> > let me know.
> 
> Don't do that.  Or at least stay far away from the vger lists with any
> bugreports or suggestions that come up when running OpenAFS.  It's some
> of the worst code around, and it has a license that doesn't event permit 
> the blatantly stupid poking into kernel internals it does.

I'm going to do it, but I know who to flame when OpenAFS breaks, and
it's not the kernel list. Believe me I've done it enough already.

Besides the syscall table modification, which is just plain stupid, is
there any other poking around in kernel internals the openafs code does
that shouldn't be there?

Re: PATCH: Change in-kernel afs client filesystem name to 'kafs'

[Alan Cox]

Ar Sad, 2006-06-24 am 13:27 -0500, ysgrifennodd Troy Benjegerdes:
> Besides the syscall table modification, which is just plain stupid, is
> there any other poking around in kernel internals the openafs code does
> that shouldn't be there?

It certainly used to pull interesting tricks to hide its auth info and
had some "novel" views of the kernel locking. Given the cachefs and FUSE
work though we ought to be at the point where almost all of OpenAFS can
be done in user space and the rest via the fuse/cachefs work.

Re: PATCH: Change in-kernel afs client filesystem name to 'kafs'

[Troy Benjegerdes]

[-- Attachment #1: Type: text/plain, Size: 918 bytes --]

On Sat, Jun 24, 2006 at 02:16:00PM +0200, Arjan van de Ven wrote:
> On Fri, 2006-06-23 at 19:41 -0500, Troy Benjegerdes wrote:
> > This patch changes the in-kernel AFS client filesystem name to 'kafs',
> > as well as allowing the AFS cache manager port to be set as a module
> > parameter. This is usefull for having a system boot with the root
> > filesystem on afs with the kernel AFS client, while still having the
> > option of loading the OpenAFS kernel module for use as a read-write
> > filesystem later.
> 
> sounds weird... the filesystem it implements is afs.
> your change also breaks userspace, since the fs type is a mount option
> so your change is userspace visible and means people need to fix their
> scripts...

Is something like this any better? (unfortunately I have not been able
to test it yet since ppc32 with CONFIG_KEXEC is currently broken because 
'reserve_crashkernel()' is not defined..



[-- Attachment #2: kafs2.patch --]
[-- Type: text/plain, Size: 3993 bytes --]

diff -r 58a4bf8bafa2 fs/Kconfig
--- a/fs/Kconfig	Sat Jun 24 05:58:44 2006 +0700
+++ b/fs/Kconfig	Mon Jun 26 01:20:35 2006 -0500
@@ -1851,6 +1851,15 @@ config AFS_FS
 
 	  If unsure, say N.
 
+config AFS_DEFAULT_ROOTCELL
+        string "Kernel AFS client default root cell and VL server IP addr list"
+        depends on AFS_FS
+        default "grand.central.org:18.7.14.88:128.2.191.224"
+        help
+          This is the default root cell for the in-kernel AFS client. This
+	  can be changed for your local configureation either at compile
+	  time or with the 'rootcell' module parameter
+
 config RXRPC
 	tristate
 
diff -r 58a4bf8bafa2 fs/afs/internal.h
--- a/fs/afs/internal.h	Sat Jun 24 05:58:44 2006 +0700
+++ b/fs/afs/internal.h	Mon Jun 26 01:20:35 2006 -0500
@@ -100,6 +100,7 @@ extern void afs_key_unregister(void);
 #ifdef AFS_CACHING_SUPPORT
 extern struct cachefs_netfs afs_cache_netfs;
 #endif
+extern char *kafs_name;
 
 /*
  * mntpt.c
diff -r 58a4bf8bafa2 fs/afs/main.c
--- a/fs/afs/main.c	Sat Jun 24 05:58:44 2006 +0700
+++ b/fs/afs/main.c	Mon Jun 26 01:20:35 2006 -0500
@@ -37,11 +37,18 @@ MODULE_AUTHOR("Red Hat, Inc.");
 MODULE_AUTHOR("Red Hat, Inc.");
 MODULE_LICENSE("GPL");
 
-static char *rootcell;
+static char *rootcell = CONFIG_AFS_DEFAULT_ROOTCELL;
+char *kafs_name = "afs";
+static uint rxrpc_port = 7001;
 
 module_param(rootcell, charp, 0);
 MODULE_PARM_DESC(rootcell, "root AFS cell name and VL server IP addr list");
 
+module_param(rxrpc_port, uint, 0);
+MODULE_PARM_DESC(rxrpc_port, "kAFS cache manager port (default 7001)");
+
+module_param(kafs_name, charp, 0);
+MODULE_PARM_DESC(kafs_name, "kAFS filesystem name (suggest using 'kafs')");
 
 static struct rxrpc_peer_ops afs_peer_ops = {
 	.adding		= afs_adding_peer,
@@ -57,7 +64,7 @@ static struct cachefs_netfs_operations a
 };
 
 struct cachefs_netfs afs_cache_netfs = {
-	.name			= "afs",
+	.name			= kafs_name,
 	.version		= 0,
 	.ops			= &afs_cache_ops,
 };
@@ -71,7 +78,10 @@ static int __init afs_init(void)
 {
 	int loop, ret;
 
-	printk(KERN_INFO "kAFS: Red Hat AFS client v0.1 registering.\n");
+	printk(KERN_INFO "kAFS: Red Hat AFS client v0.1\n");
+	printk(KERN_INFO "registering filesystem %s using callback port %d.\n",
+			kafs_name, rxrpc_port);
+	printk(KERN_INFO "using root cell/vldb servers: %s", rootcell);
 
 	/* initialise the callback hash table */
 	spin_lock_init(&afs_cb_hash_lock);
@@ -113,7 +123,7 @@ static int __init afs_init(void)
 		goto error_kafstimod;
 
 	/* create the RxRPC transport */
-	ret = rxrpc_create_transport(7001, &afs_transport);
+	ret = rxrpc_create_transport(rxrpc_port, &afs_transport);
 	if (ret < 0)
 		goto error_kafsasyncd;
 
diff -r 58a4bf8bafa2 fs/afs/mntpt.c
--- a/fs/afs/mntpt.c	Sat Jun 24 05:58:44 2006 +0700
+++ b/fs/afs/mntpt.c	Mon Jun 26 01:20:35 2006 -0500
@@ -25,7 +25,6 @@
 #include "vnode.h"
 #include "internal.h"
 
-
 static struct dentry *afs_mntpt_lookup(struct inode *dir,
 				       struct dentry *dentry,
 				       struct nameidata *nd);
@@ -203,7 +202,7 @@ static struct vfsmount *afs_mntpt_do_aut
 
 	/* try and do the mount */
 	kdebug("--- attempting mount %s -o %s ---", devname, options);
-	mnt = do_kern_mount("afs", 0, devname, options);
+	mnt = do_kern_mount(kafs_name, 0, devname, options);
 	kdebug("--- mount result %p ---", mnt);
 
 	free_page((unsigned long) devname);
diff -r 58a4bf8bafa2 fs/afs/super.c
--- a/fs/afs/super.c	Sat Jun 24 05:58:44 2006 +0700
+++ b/fs/afs/super.c	Mon Jun 26 01:20:35 2006 -0500
@@ -50,7 +50,7 @@ static void afs_destroy_inode(struct ino
 
 static struct file_system_type afs_fs_type = {
 	.owner		= THIS_MODULE,
-	.name		= "afs",
+	.name		= NULL,
 	.get_sb		= afs_get_sb,
 	.kill_sb	= kill_anon_super,
 	.fs_flags	= FS_BINARY_MOUNTDATA,
@@ -96,6 +96,7 @@ int __init afs_fs_init(void)
 	}
 
 	/* now export our filesystem to lesser mortals */
+	afs_fs_type.name = kafs_name;
 	ret = register_filesystem(&afs_fs_type);
 	if (ret < 0) {
 		kmem_cache_destroy(afs_inode_cachep);

Re: PATCH: Change in-kernel afs client filesystem name to 'kafs'

[Christoph Hellwig]

On Fri, Jun 23, 2006 at 07:41:59PM -0500, Troy Benjegerdes wrote:
> This patch changes the in-kernel AFS client filesystem name to 'kafs',
> as well as allowing the AFS cache manager port to be set as a module
> parameter. This is usefull for having a system boot with the root
> filesystem on afs with the kernel AFS client, while still having the
> option of loading the OpenAFS kernel module for use as a read-write
> filesystem later.

NACK.  OpenAFS isn't even legal to load into the kernel, we shouldn't
support it.  Better help making the kernel afs client fully features
than wasting your time on this.

Re: PATCH: Change in-kernel afs client filesystem name to 'kafs'

[Troy Benjegerdes]

On Sat, Jun 24, 2006 at 02:37:03PM +0100, Christoph Hellwig wrote:
> On Fri, Jun 23, 2006 at 07:41:59PM -0500, Troy Benjegerdes wrote:
> > This patch changes the in-kernel AFS client filesystem name to 'kafs',
> > as well as allowing the AFS cache manager port to be set as a module
> > parameter. This is usefull for having a system boot with the root
> > filesystem on afs with the kernel AFS client, while still having the
> > option of loading the OpenAFS kernel module for use as a read-write
> > filesystem later.
> 
> NACK.  OpenAFS isn't even legal to load into the kernel, we shouldn't
> support it.  Better help making the kernel afs client fully features
> than wasting your time on this.

So, you are telling me that even though OpenAFS has a license
substationally similiar in intent to the GPLv2, that it's not legal to
load into the kernel, even though the OpenAFS code predates the Linux
kernel by about 5 years? Are you going to personally sue me if I put up
an initramfs image with the OpenAFS kernel module in it?

This is not a binary-only linux-only module we are talking about here.
This is a cross-platform code that loads into multiple different
operating system kernels and has the same sort of protection of user's
right to modify the code that Linux does. It only differs in some of the
details.

My home directory is in AFS. And if you want me to help make the kernel
AFS client full-featured, I need to have both kAFS and OpenAFS loaded
and mounted at the same time. So quit bitching about OpenAFS and get out of
my way so I can work on kAFS.

Re: PATCH: Change in-kernel afs client filesystem name to 'kafs'

[Josh Boyer]

On 6/24/06, Troy Benjegerdes <hozer@hozed.org> wrote:
> On Sat, Jun 24, 2006 at 02:37:03PM +0100, Christoph Hellwig wrote:
> > On Fri, Jun 23, 2006 at 07:41:59PM -0500, Troy Benjegerdes wrote:
> > > This patch changes the in-kernel AFS client filesystem name to 'kafs',
> > > as well as allowing the AFS cache manager port to be set as a module
> > > parameter. This is usefull for having a system boot with the root
> > > filesystem on afs with the kernel AFS client, while still having the
> > > option of loading the OpenAFS kernel module for use as a read-write
> > > filesystem later.
> >
> > NACK.  OpenAFS isn't even legal to load into the kernel, we shouldn't
> > support it.  Better help making the kernel afs client fully features
> > than wasting your time on this.
>
> So, you are telling me that even though OpenAFS has a license
> substationally similiar in intent to the GPLv2, that it's not legal to
> load into the kernel, even though the OpenAFS code predates the Linux
> kernel by about 5 years? Are you going to personally sue me if I put up
> an initramfs image with the OpenAFS kernel module in it?

I doubt anyone would waste their time.

Btw, release date has nothing to do with licensing issues.  The
original BSD license was around before the GPL, and yet they were
incompatible.

>
> My home directory is in AFS. And if you want me to help make the kernel
> AFS client full-featured, I need to have both kAFS and OpenAFS loaded
> and mounted at the same time. So quit bitching about OpenAFS and get out of
> my way so I can work on kAFS.

Nobody is in your way.  If you truly want to improve kAFS to the point
where it is usable, then by all means do so.  There are many people
that would thank you.

However, I hope that you would agree that the end goal is to have kAFS
be a drop in replacement such that it can be used inter-changeably
with other AFS implementations.  Given that, integrating a patch the
changes the filesystem type which will later be reverted just is
silly.

There is nothing preventing you from doing this in your local tree
while you work on kAFS.  Pushing it to the mainline kernel is just
wrong though.  The fact your local setup requires you have both kAFS
and OpenAFS loaded at the same time is not reason enough to change it
in mainline.

josh