The comment in nfs_get_root() basically describes the patch:
Some authentication types (gss/krb5, most notably)
are such that root won't be able to present a
credential for GETATTR (ie, getroot()).
An easy way (ie, without this patch) to have unattended mounts is to
have a root/host@REALM (or similar) principal stashed in a keytab, which
root (rather, gssd) can use. However, this might not be desirable for
many sites. In any case, RFC2623 specifically describes the problem
addressed here.
Notes:
- Root inode gets inum of 1. This doesn't seem to matter, but may be
aesthetically unpleasing. I wanted to choose an inum unlikely to
conflict with an existing inum (although NFS has specific support
for that). It looks like more work than it's worth to change the
inum after the info is available. AFAICT it's not critical info.
- Solaris has this "wierd" (but understandable) behavior that after
mounting without a credential, the mount point is not visible at all
until an access is attempted with a credential. This now-you-see-it-
now-you-don't behavior doesn't seem worthwhile to reproduce here.
- Unfortunately, MOUNT_VERSION must go to 5. Some kernels with ver 4
do not understand pseudoflavor. Keeping it at 4 means that the
userland mount can't know for sure whether the kernel accepted the
option or not. (Unless I'm missing some hack that could be done.)
- This doesn't actually work as-is. A separate patch is needed for
rpc_setbufsize() which I will provide separately, after getting
feedback for this patch.
Does this patch look reasonable? It works in my environment, against
a netapp server (with the rpcsec_gss patch I provided earlier).
If interested, I can supply the userland changes as well, just email me
for the patch. (I will send it to the util-linux folks if this change
makes it into the kernel.)
/fc
diff -uNrp linux-2.6.0-test1.0/fs/nfs/inode.c linux-2.6.0-test1.1/fs/nfs/inode.c
--- linux-2.6.0-test1.0/fs/nfs/inode.c Tue Jul 15 21:19:59 2003
+++ linux-2.6.0-test1.1/fs/nfs/inode.c Tue Jul 15 22:56:05 2003
@@ -151,15 +151,16 @@ void
nfs_put_super(struct super_block *sb)
{
struct nfs_server *server = NFS_SB(sb);
- struct rpc_clnt *rpc;
#ifdef CONFIG_NFS_V4
if (server->idmap != NULL)
nfs_idmap_delete(server);
#endif /* CONFIG_NFS_V4 */
- if ((rpc = server->client) != NULL)
- rpc_shutdown_client(rpc);
+ if (server->client != NULL)
+ rpc_shutdown_client(server->client);
+ if (server->client_sys != NULL)
+ rpc_shutdown_client(server->client_sys);
if (!(server->flags & NFS_MOUNT_NONLM))
lockd_down(); /* release rpc.lockd */
@@ -226,27 +227,57 @@ nfs_block_size(unsigned long bsize, unsi
/*
* Obtain the root inode of the file system.
*/
-static struct inode *
-nfs_get_root(struct super_block *sb, struct nfs_fh *rootfh)
+static int
+nfs_get_root(struct inode **rooti, rpc_authflavor_t authflavor, struct super_block *sb, struct nfs_fh *rootfh)
{
struct nfs_server *server = NFS_SB(sb);
- struct nfs_fattr fattr;
- struct inode *inode;
+ struct nfs_fattr fattr = { };
int error;
- if ((error = server->rpc_ops->getroot(server, rootfh, &fattr)) < 0) {
+ error = server->rpc_ops->getroot(server, rootfh, &fattr);
+ if (error == -EACCES && authflavor > RPC_AUTH_MAXFLAVOR) {
+ /*
+ * Some authentication types (gss/krb5, most notably)
+ * are such that root won't be able to present a
+ * credential for GETATTR (ie, getroot()).
+ *
+ * We still want the mount to succeed.
+ *
+ * So we fake the attr values and mark the inode as such.
+ * On the first succesful traversal, we fix everything.
+ * The auth type test isn't quite correct, but whatever.
+ */
+ dfprintk(VFS, "NFS: faking root inode\n");
+
+ fattr.fileid = 1;
+ fattr.nlink = 2; /* minimum for a dir */
+ fattr.type = NFDIR;
+ fattr.mode = S_IFDIR|S_IRUGO|S_IXUGO;
+ fattr.size = 4096;
+ fattr.du.nfs3.used = 1;
+ fattr.valid = NFS_ATTR_FATTR|NFS_ATTR_FATTR_V3;
+ } else if (error < 0) {
printk(KERN_NOTICE "nfs_get_root: getattr error = %d\n", -error);
- return NULL;
+ *rooti = NULL; /* superfluous ... but safe */
+ return error;
}
- inode = __nfs_fhget(sb, rootfh, &fattr);
- return inode;
+ *rooti = __nfs_fhget(sb, rootfh, &fattr);
+ if (error == -EACCES && authflavor > RPC_AUTH_MAXFLAVOR) {
+ if (*rooti) {
+ NFS_FLAGS(*rooti) |= NFS_INO_FAKE_ROOT;
+ NFS_CACHEINV((*rooti));
+ error = 0;
+ }
+ }
+ return error;
}
/*
* Do NFS version-independent mount processing, and sanity checking
*/
-int nfs_sb_init(struct super_block *sb)
+static int
+nfs_sb_init(struct super_block *sb, rpc_authflavor_t authflavor)
{
struct nfs_server *server;
struct inode *root_inode = NULL;
@@ -267,8 +298,7 @@ int nfs_sb_init(struct super_block *sb)
sb->s_op = &nfs_sops;
/* Did getting the root inode fail? */
- root_inode = nfs_get_root(sb, &server->fh);
- if (!root_inode)
+ if (nfs_get_root(&root_inode, authflavor, sb, &server->fh) < 0)
goto out_no_root;
sb->s_root = d_alloc_root(root_inode);
if (!sb->s_root)
@@ -346,19 +376,66 @@ out_no_root:
}
/*
+ * Create an RPC client handle.
+ */
+static struct rpc_clnt *
+nfs_create_client(struct nfs_server *server, const struct nfs_mount_data *data)
+{
+ struct rpc_timeout timeparms;
+ struct rpc_xprt *xprt = NULL;
+ struct rpc_clnt *clnt = NULL;
+ int tcp = (data->flags & NFS_MOUNT_TCP);
+
+ /* Initialize timeout values */
+ timeparms.to_initval = data->timeo * HZ / 10;
+ timeparms.to_retries = data->retrans;
+ timeparms.to_maxval = tcp ? RPC_MAX_TCP_TIMEOUT : RPC_MAX_UDP_TIMEOUT;
+ timeparms.to_exponential = 1;
+
+ if (!timeparms.to_initval)
+ timeparms.to_initval = (tcp ? 600 : 11) * HZ / 10;
+ if (!timeparms.to_retries)
+ timeparms.to_retries = 5;
+
+ /* create transport and client */
+ xprt = xprt_create_proto(tcp ? IPPROTO_TCP : IPPROTO_UDP,
+ &server->addr, &timeparms);
+ if (xprt == NULL) {
+ printk(KERN_WARNING "NFS: cannot create RPC transport.\n");
+ goto out_fail;
+ }
+ clnt = rpc_create_client(xprt, server->hostname, &nfs_program,
+ server->rpc_ops->version, data->pseudoflavor);
+ if (clnt == NULL) {
+ printk(KERN_WARNING "NFS: cannot create RPC client.\n");
+ goto out_fail;
+ }
+
+ clnt->cl_intr = (server->flags & NFS_MOUNT_INTR) ? 1 : 0;
+ clnt->cl_softrtry = (server->flags & NFS_MOUNT_SOFT) ? 1 : 0;
+ clnt->cl_droppriv = (server->flags & NFS_MOUNT_BROKEN_SUID) ? 1 : 0;
+ clnt->cl_chatty = 1;
+
+ return clnt;
+
+out_fail:
+ if (xprt)
+ xprt_destroy(xprt);
+ return NULL;
+}
+
+/*
* The way this works is that the mount process passes a structure
* in the data argument which contains the server's IP address
* and the root file handle obtained from the server's mount
* daemon. We stash these away in the private superblock fields.
*/
-int nfs_fill_super(struct super_block *sb, struct nfs_mount_data *data, int silent)
+static int
+nfs_fill_super(struct super_block *sb, struct nfs_mount_data *data, int silent)
{
struct nfs_server *server;
- struct rpc_xprt *xprt = NULL;
- struct rpc_clnt *clnt = NULL;
- struct rpc_timeout timeparms;
- int tcp, err = -EIO;
- u32 authflavor;
+ int err = -EIO;
+ rpc_authflavor_t authflavor;
server = NFS_SB(sb);
sb->s_blocksize_bits = 0;
@@ -400,46 +477,20 @@ int nfs_fill_super(struct super_block *s
server->rpc_ops = &nfs_v2_clientops;
}
- /* Which protocol do we use? */
- tcp = (data->flags & NFS_MOUNT_TCP);
-
- /* Initialize timeout values */
- timeparms.to_initval = data->timeo * HZ / 10;
- timeparms.to_retries = data->retrans;
- timeparms.to_maxval = tcp? RPC_MAX_TCP_TIMEOUT : RPC_MAX_UDP_TIMEOUT;
- timeparms.to_exponential = 1;
-
- if (!timeparms.to_initval)
- timeparms.to_initval = (tcp ? 600 : 11) * HZ / 10;
- if (!timeparms.to_retries)
- timeparms.to_retries = 5;
-
- /* Now create transport and client */
- xprt = xprt_create_proto(tcp? IPPROTO_TCP : IPPROTO_UDP,
- &server->addr, &timeparms);
- if (xprt == NULL) {
- printk(KERN_WARNING "NFS: cannot create RPC transport.\n");
- goto out_fail;
- }
-
- if (data->flags & NFS_MOUNT_SECFLAVOUR)
- authflavor = data->pseudoflavor;
- else
- authflavor = RPC_AUTH_UNIX;
-
- clnt = rpc_create_client(xprt, server->hostname, &nfs_program,
- server->rpc_ops->version, authflavor);
- if (clnt == NULL) {
- printk(KERN_WARNING "NFS: cannot create RPC client.\n");
- xprt_destroy(xprt);
+ /* Fill in pseudoflavor for mount version < 5 */
+ if (!(data->flags & NFS_MOUNT_SECFLAVOUR))
+ data->pseudoflavor = RPC_AUTH_UNIX;
+ authflavor = data->pseudoflavor; /* save for sb_init() */
+ /* XXX maybe we want to add a server->pseudoflavor field */
+
+ /* Create RPC client handles */
+ server->client = nfs_create_client(server, data);
+ if (server->client == NULL)
goto out_fail;
- }
-
- clnt->cl_intr = (server->flags & NFS_MOUNT_INTR) ? 1 : 0;
- clnt->cl_softrtry = (server->flags & NFS_MOUNT_SOFT) ? 1 : 0;
- clnt->cl_droppriv = (server->flags & NFS_MOUNT_BROKEN_SUID) ? 1 : 0;
- clnt->cl_chatty = 1;
- server->client = clnt;
+ data->pseudoflavor = RPC_AUTH_UNIX; /* RFC 2623, sec 2.3.2 */
+ server->client_sys = nfs_create_client(server, data);
+ if (server->client_sys == NULL)
+ goto out_shutdown;
/* Fire up rpciod if not yet running */
if (rpciod_up() != 0) {
@@ -447,7 +498,7 @@ int nfs_fill_super(struct super_block *s
goto out_shutdown;
}
- err = nfs_sb_init(sb);
+ err = nfs_sb_init(sb, authflavor);
if (err != 0)
goto out_noinit;
@@ -466,7 +517,10 @@ int nfs_fill_super(struct super_block *s
out_noinit:
rpciod_down();
out_shutdown:
- rpc_shutdown_client(server->client);
+ if (server->client)
+ rpc_shutdown_client(server->client);
+ if (server->client_sys)
+ rpc_shutdown_client(server->client_sys);
out_fail:
if (server->hostname)
kfree(server->hostname);
@@ -904,6 +958,11 @@ __nfs_revalidate_inode(struct nfs_server
goto out_nowait;
if (NFS_STALE(inode) && inode != inode->i_sb->s_root->d_inode)
goto out_nowait;
+ if (NFS_FAKE_ROOT(inode)) {
+ dfprintk(VFS, "NFS: not revalidating fake root\n");
+ status = 0;
+ goto out_nowait;
+ }
while (NFS_REVALIDATING(inode)) {
status = nfs_wait_on_inode(inode, NFS_INO_REVALIDATING);
@@ -1007,6 +1066,13 @@ __nfs_refresh_inode(struct inode *inode,
inode->i_sb->s_id, inode->i_ino,
atomic_read(&inode->i_count), fattr->valid);
+ /* First successful call after mount, fill real data. */
+ if (NFS_FAKE_ROOT(inode)) {
+ dfprintk(VFS, "NFS: updating fake root\n");
+ nfsi->fileid = fattr->fileid;
+ NFS_FLAGS(inode) &= ~NFS_INO_FAKE_ROOT;
+ }
+
if (nfsi->fileid != fattr->fileid) {
printk(KERN_ERR "nfs_refresh_inode: inode number mismatch\n"
"expected (%s/0x%Lx), got (%s/0x%Lx)\n",
@@ -1229,6 +1295,8 @@ static struct super_block *nfs_get_sb(st
root->size = NFS2_FHSIZE;
memcpy(root->data, data->old_root.data, NFS2_FHSIZE);
}
+ if (data->version < 5)
+ data->flags &= ~NFS_MOUNT_SECFLAVOUR;
}
if (root->size > sizeof(root->data)) {
diff -uNrp linux-2.6.0-test1.0/fs/nfs/nfs3proc.c linux-2.6.0-test1.1/fs/nfs/nfs3proc.c
--- linux-2.6.0-test1.0/fs/nfs/nfs3proc.c Tue Jul 15 21:10:30 2003
+++ linux-2.6.0-test1.1/fs/nfs/nfs3proc.c Tue Jul 15 21:45:39 2003
@@ -681,7 +681,7 @@ nfs3_proc_fsinfo(struct nfs_server *serv
dprintk("NFS call fsinfo\n");
info->fattr->valid = 0;
- status = rpc_call(server->client, NFS3PROC_FSINFO, fhandle, info, 0);
+ status = rpc_call(server->client_sys, NFS3PROC_FSINFO, fhandle, info, 0);
dprintk("NFS reply fsinfo: %d\n", status);
return status;
}
diff -uNrp linux-2.6.0-test1.0/include/linux/nfs_fs.h linux-2.6.0-test1.1/include/linux/nfs_fs.h
--- linux-2.6.0-test1.0/include/linux/nfs_fs.h Tue Jul 15 21:10:08 2003
+++ linux-2.6.0-test1.1/include/linux/nfs_fs.h Tue Jul 15 21:47:23 2003
@@ -172,6 +172,7 @@ struct nfs_inode {
#define NFS_INO_ADVISE_RDPLUS 0x0002 /* advise readdirplus */
#define NFS_INO_REVALIDATING 0x0004 /* revalidating attrs */
#define NFS_INO_FLUSH 0x0008 /* inode is due for flushing */
+#define NFS_INO_FAKE_ROOT 0x0080 /* root inode placeholder */
static inline struct nfs_inode *NFS_I(struct inode *inode)
{
@@ -207,6 +208,7 @@ do { \
#define NFS_FLAGS(inode) (NFS_I(inode)->flags)
#define NFS_REVALIDATING(inode) (NFS_FLAGS(inode) & NFS_INO_REVALIDATING)
#define NFS_STALE(inode) (NFS_FLAGS(inode) & NFS_INO_STALE)
+#define NFS_FAKE_ROOT(inode) (NFS_FLAGS(inode) & NFS_INO_FAKE_ROOT)
#define NFS_FILEID(inode) (NFS_I(inode)->fileid)
diff -uNrp linux-2.6.0-test1.0/include/linux/nfs_fs_sb.h linux-2.6.0-test1.1/include/linux/nfs_fs_sb.h
--- linux-2.6.0-test1.0/include/linux/nfs_fs_sb.h Tue Jul 15 21:10:08 2003
+++ linux-2.6.0-test1.1/include/linux/nfs_fs_sb.h Tue Jul 15 21:45:39 2003
@@ -9,6 +9,7 @@
*/
struct nfs_server {
struct rpc_clnt * client; /* RPC client handle */
+ struct rpc_clnt * client_sys; /* 2nd handle for FSINFO */
struct nfs_rpc_ops * rpc_ops; /* NFS protocol vector */
struct backing_dev_info backing_dev_info;
int flags; /* various flags */
diff -uNrp linux-2.6.0-test1.0/include/linux/nfs_mount.h linux-2.6.0-test1.1/include/linux/nfs_mount.h
--- linux-2.6.0-test1.0/include/linux/nfs_mount.h Tue Jul 15 21:10:08 2003
+++ linux-2.6.0-test1.1/include/linux/nfs_mount.h Tue Jul 15 21:45:39 2003
@@ -20,7 +20,7 @@
* mount-to-kernel version compatibility. Some of these aren't used yet
* but here they are anyway.
*/
-#define NFS_MOUNT_VERSION 4
+#define NFS_MOUNT_VERSION 5
struct nfs_mount_data {
int version; /* 1 */
@@ -40,7 +40,7 @@ struct nfs_mount_data {
int namlen; /* 2 */
unsigned int bsize; /* 3 */
struct nfs3_fh root; /* 4 */
- int pseudoflavor; /* 4 */
+ int pseudoflavor; /* 5 */
};
/* bits in the flags field */
@@ -57,7 +57,7 @@ struct nfs_mount_data {
#define NFS_MOUNT_NONLM 0x0200 /* 3 */
#define NFS_MOUNT_BROKEN_SUID 0x0400 /* 4 */
#define NFS_MOUNT_STRICTLOCK 0x1000 /* reserved for NFSv4 */
-#define NFS_MOUNT_SECFLAVOUR 0x2000 /* reserved */
+#define NFS_MOUNT_SECFLAVOUR 0x2000 /* 5 */
#define NFS_MOUNT_FLAGMASK 0xFFFF
#endif
On Tue, Jul 15, 2003 at 11:26:05PM -0700, Frank Cusack wrote:
> - This doesn't actually work as-is. A separate patch is needed for
> rpc_setbufsize() which I will provide separately, after getting
> feedback for this patch.
Since I don't expect a known-broken patch to be applied ... here's
the bufsize patch.
- fix null dereference on xprt->inet if !connected,
which happens if a rpc cred wasn't available (root+auth_gss case)
- set bufsize on reconnect
diff -uNrp linux-2.6.0-test1.2/net/sunrpc/clnt.c linux-2.6.0-test1.3/net/sunrpc/clnt.c
--- linux-2.6.0-test1.2/net/sunrpc/clnt.c Tue Jul 15 23:29:15 2003
+++ linux-2.6.0-test1.3/net/sunrpc/clnt.c Wed Jul 16 00:02:31 2003
@@ -385,7 +385,8 @@ rpc_setbufsize(struct rpc_clnt *clnt, un
xprt->rcvsize = 0;
if (rcvsize)
xprt->rcvsize = rcvsize + RPC_SLACK_SPACE;
- xprt_sock_setbufsize(xprt);
+ if (xprt_connected(xprt))
+ xprt_sock_setbufsize(xprt);
}
/*
diff -uNrp linux-2.6.0-test1.2/net/sunrpc/xprt.c linux-2.6.0-test1.3/net/sunrpc/xprt.c
--- linux-2.6.0-test1.2/net/sunrpc/xprt.c Tue Jul 15 23:29:15 2003
+++ linux-2.6.0-test1.3/net/sunrpc/xprt.c Wed Jul 16 00:02:37 2003
@@ -436,6 +436,7 @@ xprt_connect(struct rpc_task *task)
goto out_write;
}
xprt_bind_socket(xprt, sock);
+ xprt_sock_setbufsize(xprt);
if (!xprt->stream)
goto out_write;
On Tue, Jul 15, 2003 at 11:26:05PM -0700, Frank Cusack wrote:
> The comment in nfs_get_root() basically describes the patch:
>
> Some authentication types (gss/krb5, most notably)
> are such that root won't be able to present a
> credential for GETATTR (ie, getroot()).
[..]
> Does this patch look reasonable? It works in my environment, against
> a netapp server (with the rpcsec_gss patch I provided earlier).
The way this patch was imported in bk-cvs causes a compile-time failure
in fs/nfs/inode.c;
Below is the -what seems to me- obvious fix.
One comment though wrt this code: it seems both the spellings 'flavor' and
'flavour' are used in this piece of code which is somewhat confusing. Should
that be fixed?
Index: fs/nfs/inode.c
===================================================================
RCS file: /home/cvs/linux-2.5/fs/nfs/inode.c,v
retrieving revision 1.79
diff -u -r1.79 inode.c
--- CVS/fs/nfs/inode.c 17 Jul 2003 17:28:19 -0000 1.79
+++ linux-2.5/fs/nfs/inode.c 17 Jul 2003 17:40:29 -0000
@@ -1441,7 +1441,7 @@
if ((server->idmap = nfs_idmap_new(server)) == NULL)
printk(KERN_WARNING "NFS: couldn't start IDmap\n");
- err = nfs_sb_init(sb);
+ err = nfs_sb_init(sb, authflavour);
if (err == 0)
return 0;
rpciod_down();
VrGr,
--
Jasper Spaans http://jsp.vs19.net/contact/
Linus,
Could we please back off the merging of these particular patches,
in order to give time for some critical debate?
I'm glad Frank has been working on this, but if there has been a
proper debate on the merits of these patches, then I must have missed
it. I have indeed seen the patches pass through my inbox once, but I
have not had time yet to review them properly, and feed back my
comments to Frank.
I have clear doubts about this patch.
The 'fake root' inode does not appear to have a well defined
behaviour when it tries to circumvent revalidation rules. AFAICS
things like 'stat()', etc will at times tend to give odd results even
if the user does have the correct credentials.
On a similar vein, I'm also unconvinced that there exist no races
between the lookup of new files, and the first setting of the
NFS_FILEID() on the root directory. fileid=1 is *not* a reserved inode
number under NFS v2 or v3, so it is perfectly legitimate for the
server to have other inodes with that number. In this case,
iget5_locked() may end up assigning the same inode to the root inode +
some other file.
I also have questions about smaller issues, such as why we need a
full second RPC client in the superblock just in order to (very
temporarily) add a second authentication flavour.
Cheers,
Trond