Skip to contentSkip to conversation

nfsd: add principal to the data being tracked by nfsdcld

RFC PATCH In review Scott Mayhew 2 participants 8 messages Jul 30, 2019 – Jul 31, 2019

Subsystems (2)
lore
Share & tools
lore
#0 lore raw

cover nfsd: add principal to the data being tracked by nfsdcld

At the spring bakeathon, Chuck suggested that we should store the kerberos principal in addition to the client id string in nfsdcld. The idea is to prevent an illegitimate client from reclaiming another client's opens by supplying that client's id string. This is an initial attempt at doing that.

The first patch lays some groundwork for supporting multiple message versions for the nfsdcld upcalls, adding fields for version and message length to the nfsd4_client_tracking_ops (these fields are only used for the nfsdcld upcalls and ignored for the other tracking methods), as well as an upcall to get the maximum version supported by the userspace daemon.

Initially I had played with the idea of hanging a version number off either the cld_net or nfsd_net structure, exposing that via a proc file, and having the userspace daemon write the desired version number to the proc file prior to opening the rpc_pipefs file. That was potentially problematic if nfsd was already trying to queue an upcall using a version that nfscld didn't support... so I like the GetVersion upcall idea better. Another option might be to have different rpc_pipefs files for the different message versions and decide which version to use based on which pipefs file the daemon opens.

The second patch actually adds the v2 message, which adds the principal (actually just the first 1024 bytes) to the Cld_Create upcall and to the Cld_GraceStart downcall (which is what loads the data in the reclaim_str_hashtbl). I couldn't really figure out what the maximum length of a kerberos principal actually is (looking at krb5.h the length field in the struct krb5_data is an unsigned int, so I guess it can be pretty big). I don't think the principal will be that large in practice, and even if it is the first 1024 bytes should be sufficient for our purposes.

Scott Mayhew (2):
nfsd: add a "GetVersion" upcall for nfsdcld
nfsd: add support for upcall version 2

 fs/nfsd/nfs4recover.c         | 332 +++++++++++++++++++++++++++-------
 fs/nfsd/nfs4state.c           |   6 +-
 fs/nfsd/state.h               |   3 +-
 include/uapi/linux/nfsd/cld.h |  37 +++-
 4 files changed, 311 insertions(+), 67 deletions(-)
Signature · 1 line
2.17.2
#1 lore raw

RFC PATCH 1/2 nfsd: add a "GetVersion" upcall for nfsdcld

Add a "GetVersion" upcall to allow nfsd to determine the maximum upcall version that the nfsdcld userspace daemon supports. If the daemon responds with -EOPNOTSUPP, then we know it only supports v1.

---
 fs/nfsd/nfs4recover.c         | 167 ++++++++++++++++++++++++----------
 include/uapi/linux/nfsd/cld.h |  11 ++-
 2 files changed, 127 insertions(+), 51 deletions(-)
fs/nfsd/nfs4recover.c +117 -50
index 87679557d0d6..58a61339d40c 100644
@@ -59,8 +59,12 @@ struct nfsd4_client_tracking_ops {
 	void (*remove)(struct nfs4_client *);
 	int (*check)(struct nfs4_client *);
 	void (*grace_done)(struct nfsd_net *);
+	uint8_t version;
+	size_t msglen;
 };
 
+static const struct nfsd4_client_tracking_ops nfsd4_cld_tracking_ops;
+
 /* Globals */
 static char user_recovery_dirname[PATH_MAX] = "/var/lib/nfs/v4recovery";
 
@@ -718,6 +722,8 @@ static const struct nfsd4_client_tracking_ops nfsd4_legacy_tracking_ops = {
 	.remove		= nfsd4_remove_clid_dir,
 	.check		= nfsd4_check_legacy_client,
 	.grace_done	= nfsd4_recdir_purge_old,
+	.version	= 1,
+	.msglen		= 0,
 };
 
 /* Globals */
@@ -737,19 +743,24 @@ struct cld_upcall {
 	struct list_head	 cu_list;
 	struct cld_net		*cu_net;
 	struct completion	 cu_done;
-	struct cld_msg		 cu_msg;
+	union {
+		struct cld_msg_hdr	 cu_hdr;
+		struct cld_msg		 cu_msg;
+	} cu_u;
 };
 
 static int
-__cld_pipe_upcall(struct rpc_pipe *pipe, struct cld_msg *cmsg)
+__cld_pipe_upcall(struct rpc_pipe *pipe, void *cmsg)
 {
 	int ret;
 	struct rpc_pipe_msg msg;
-	struct cld_upcall *cup = container_of(cmsg, struct cld_upcall, cu_msg);
+	struct cld_upcall *cup = container_of(cmsg, struct cld_upcall, cu_u);
+	struct nfsd_net *nn = net_generic(pipe->dentry->d_sb->s_fs_info,
+					  nfsd_net_id);
 
 	memset(&msg, 0, sizeof(msg));
 	msg.data = cmsg;
-	msg.len = sizeof(*cmsg);
+	msg.len = nn->client_tracking_ops->msglen;
 
 	ret = rpc_queue_upcall(pipe, &msg);
 	if (ret < 0) {
@@ -765,7 +776,7 @@ __cld_pipe_upcall(struct rpc_pipe *pipe, struct cld_msg *cmsg)
 }
 
 static int
-cld_pipe_upcall(struct rpc_pipe *pipe, struct cld_msg *cmsg)
+cld_pipe_upcall(struct rpc_pipe *pipe, void *cmsg)
 {
 	int ret;
 
@@ -809,7 +820,7 @@ __cld_pipe_inprogress_downcall(const struct cld_msg __user *cmsg,
 			kfree(name.data);
 			return -EFAULT;
 		}
-		return sizeof(*cmsg);
+		return nn->client_tracking_ops->msglen;
 	}
 	return -EFAULT;
 }
@@ -818,6 +829,7 @@ static ssize_t
 cld_pipe_downcall(struct file *filp, const char __user *src, size_t mlen)
 {
 	struct cld_upcall *tmp, *cup;
+	struct cld_msg_hdr __user *hdr = (struct cld_msg_hdr __user *)src;
 	struct cld_msg __user *cmsg = (struct cld_msg __user *)src;
 	uint32_t xid;
 	struct nfsd_net *nn = net_generic(file_inode(filp)->i_sb->s_fs_info,
@@ -825,14 +837,14 @@ cld_pipe_downcall(struct file *filp, const char __user *src, size_t mlen)
 	struct cld_net *cn = nn->cld_net;
 	int16_t status;
 
-	if (mlen != sizeof(*cmsg)) {
+	if (mlen != nn->client_tracking_ops->msglen) {
 		dprintk("%s: got %zu bytes, expected %zu\n", __func__, mlen,
-			sizeof(*cmsg));
+			nn->client_tracking_ops->msglen);
 		return -EINVAL;
 	}
 
 	/* copy just the xid so we can try to find that */
-	if (copy_from_user(&xid, &cmsg->cm_xid, sizeof(xid)) != 0) {
+	if (copy_from_user(&xid, &hdr->cm_xid, sizeof(xid)) != 0) {
 		dprintk("%s: error when copying xid from userspace", __func__);
 		return -EFAULT;
 	}
@@ -842,7 +854,7 @@ cld_pipe_downcall(struct file *filp, const char __user *src, size_t mlen)
 	 * list (for -EINPROGRESS, we just want to make sure the xid is
 	 * valid, not remove the upcall from the list)
 	 */
-	if (get_user(status, &cmsg->cm_status)) {
+	if (get_user(status, &hdr->cm_status)) {
 		dprintk("%s: error when copying status from userspace", __func__);
 		return -EFAULT;
 	}
@@ -851,7 +863,7 @@ cld_pipe_downcall(struct file *filp, const char __user *src, size_t mlen)
 	cup = NULL;
 	spin_lock(&cn->cn_lock);
 	list_for_each_entry(tmp, &cn->cn_list, cu_list) {
-		if (get_unaligned(&tmp->cu_msg.cm_xid) == xid) {
+		if (get_unaligned(&tmp->cu_u.cu_hdr.cm_xid) == xid) {
 			cup = tmp;
 			if (status != -EINPROGRESS)
 				list_del_init(&cup->cu_list);
@@ -869,7 +881,7 @@ cld_pipe_downcall(struct file *filp, const char __user *src, size_t mlen)
 	if (status == -EINPROGRESS)
 		return __cld_pipe_inprogress_downcall(cmsg, nn);
 
-	if (copy_from_user(&cup->cu_msg, src, mlen) != 0)
+	if (copy_from_user(&cup->cu_u.cu_msg, src, mlen) != 0)
 		return -EFAULT;
 
 	complete(&cup->cu_done);
@@ -881,7 +893,7 @@ cld_pipe_destroy_msg(struct rpc_pipe_msg *msg)
 {
 	struct cld_msg *cmsg = msg->data;
 	struct cld_upcall *cup = container_of(cmsg, struct cld_upcall,
-						 cu_msg);
+						 cu_u.cu_msg);
 
 	/* errno >= 0 means we got a downcall */
 	if (msg->errno >= 0)
@@ -1012,9 +1024,10 @@ nfsd4_remove_cld_pipe(struct net *net)
 }
 
 static struct cld_upcall *
-alloc_cld_upcall(struct cld_net *cn)
+alloc_cld_upcall(struct nfsd_net *nn)
 {
 	struct cld_upcall *new, *tmp;
+	struct cld_net *cn = nn->cld_net;
 
 	new = kzalloc(sizeof(*new), GFP_KERNEL);
 	if (!new)
@@ -1024,20 +1037,20 @@ alloc_cld_upcall(struct cld_net *cn)
 restart_search:
 	spin_lock(&cn->cn_lock);
 	list_for_each_entry(tmp, &cn->cn_list, cu_list) {
-		if (tmp->cu_msg.cm_xid == cn->cn_xid) {
+		if (tmp->cu_u.cu_msg.cm_xid == cn->cn_xid) {
 			cn->cn_xid++;
 			spin_unlock(&cn->cn_lock);
 			goto restart_search;
 		}
 	}
 	init_completion(&new->cu_done);
-	new->cu_msg.cm_vers = CLD_UPCALL_VERSION;
-	put_unaligned(cn->cn_xid++, &new->cu_msg.cm_xid);
+	new->cu_u.cu_msg.cm_vers = nn->client_tracking_ops->version;
+	put_unaligned(cn->cn_xid++, &new->cu_u.cu_msg.cm_xid);
 	new->cu_net = cn;
 	list_add(&new->cu_list, &cn->cn_list);
 	spin_unlock(&cn->cn_lock);
 
-	dprintk("%s: allocated xid %u\n", __func__, new->cu_msg.cm_xid);
+	dprintk("%s: allocated xid %u\n", __func__, new->cu_u.cu_msg.cm_xid);
 
 	return new;
 }
@@ -1066,20 +1079,20 @@ nfsd4_cld_create(struct nfs4_client *clp)
 	if (test_bit(NFSD4_CLIENT_STABLE, &clp->cl_flags))
 		return;
 
-	cup = alloc_cld_upcall(cn);
+	cup = alloc_cld_upcall(nn);
 	if (!cup) {
 		ret = -ENOMEM;
 		goto out_err;
 	}
 
-	cup->cu_msg.cm_cmd = Cld_Create;
-	cup->cu_msg.cm_u.cm_name.cn_len = clp->cl_name.len;
-	memcpy(cup->cu_msg.cm_u.cm_name.cn_id, clp->cl_name.data,
+	cup->cu_u.cu_msg.cm_cmd = Cld_Create;
+	cup->cu_u.cu_msg.cm_u.cm_name.cn_len = clp->cl_name.len;
+	memcpy(cup->cu_u.cu_msg.cm_u.cm_name.cn_id, clp->cl_name.data,
 			clp->cl_name.len);
 
-	ret = cld_pipe_upcall(cn->cn_pipe, &cup->cu_msg);
+	ret = cld_pipe_upcall(cn->cn_pipe, &cup->cu_u.cu_msg);
 	if (!ret) {
-		ret = cup->cu_msg.cm_status;
+		ret = cup->cu_u.cu_msg.cm_status;
 		set_bit(NFSD4_CLIENT_STABLE, &clp->cl_flags);
 	}
 
@@ -1103,20 +1116,20 @@ nfsd4_cld_remove(struct nfs4_client *clp)
 	if (!test_bit(NFSD4_CLIENT_STABLE, &clp->cl_flags))
 		return;
 
-	cup = alloc_cld_upcall(cn);
+	cup = alloc_cld_upcall(nn);
 	if (!cup) {
 		ret = -ENOMEM;
 		goto out_err;
 	}
 
-	cup->cu_msg.cm_cmd = Cld_Remove;
-	cup->cu_msg.cm_u.cm_name.cn_len = clp->cl_name.len;
-	memcpy(cup->cu_msg.cm_u.cm_name.cn_id, clp->cl_name.data,
+	cup->cu_u.cu_msg.cm_cmd = Cld_Remove;
+	cup->cu_u.cu_msg.cm_u.cm_name.cn_len = clp->cl_name.len;
+	memcpy(cup->cu_u.cu_msg.cm_u.cm_name.cn_id, clp->cl_name.data,
 			clp->cl_name.len);
 
-	ret = cld_pipe_upcall(cn->cn_pipe, &cup->cu_msg);
+	ret = cld_pipe_upcall(cn->cn_pipe, &cup->cu_u.cu_msg);
 	if (!ret) {
-		ret = cup->cu_msg.cm_status;
+		ret = cup->cu_u.cu_msg.cm_status;
 		clear_bit(NFSD4_CLIENT_STABLE, &clp->cl_flags);
 	}
 
@@ -1145,21 +1158,21 @@ nfsd4_cld_check_v0(struct nfs4_client *clp)
 	if (test_bit(NFSD4_CLIENT_STABLE, &clp->cl_flags))
 		return 0;
 
-	cup = alloc_cld_upcall(cn);
+	cup = alloc_cld_upcall(nn);
 	if (!cup) {
 		printk(KERN_ERR "NFSD: Unable to check client record on "
 				"stable storage: %d\n", -ENOMEM);
 		return -ENOMEM;
 	}
 
-	cup->cu_msg.cm_cmd = Cld_Check;
-	cup->cu_msg.cm_u.cm_name.cn_len = clp->cl_name.len;
-	memcpy(cup->cu_msg.cm_u.cm_name.cn_id, clp->cl_name.data,
+	cup->cu_u.cu_msg.cm_cmd = Cld_Check;
+	cup->cu_u.cu_msg.cm_u.cm_name.cn_len = clp->cl_name.len;
+	memcpy(cup->cu_u.cu_msg.cm_u.cm_name.cn_id, clp->cl_name.data,
 			clp->cl_name.len);
 
-	ret = cld_pipe_upcall(cn->cn_pipe, &cup->cu_msg);
+	ret = cld_pipe_upcall(cn->cn_pipe, &cup->cu_u.cu_msg);
 	if (!ret) {
-		ret = cup->cu_msg.cm_status;
+		ret = cup->cu_u.cu_msg.cm_status;
 		set_bit(NFSD4_CLIENT_STABLE, &clp->cl_flags);
 	}
 
@@ -1223,16 +1236,16 @@ nfsd4_cld_grace_start(struct nfsd_net *nn)
 	struct cld_upcall *cup;
 	struct cld_net *cn = nn->cld_net;
 
-	cup = alloc_cld_upcall(cn);
+	cup = alloc_cld_upcall(nn);
 	if (!cup) {
 		ret = -ENOMEM;
 		goto out_err;
 	}
 
-	cup->cu_msg.cm_cmd = Cld_GraceStart;
-	ret = cld_pipe_upcall(cn->cn_pipe, &cup->cu_msg);
+	cup->cu_u.cu_msg.cm_cmd = Cld_GraceStart;
+	ret = cld_pipe_upcall(cn->cn_pipe, &cup->cu_u.cu_msg);
 	if (!ret)
-		ret = cup->cu_msg.cm_status;
+		ret = cup->cu_u.cu_msg.cm_status;
 
 	free_cld_upcall(cup);
 out_err:
@@ -1250,17 +1263,17 @@ nfsd4_cld_grace_done_v0(struct nfsd_net *nn)
 	struct cld_upcall *cup;
 	struct cld_net *cn = nn->cld_net;
 
-	cup = alloc_cld_upcall(cn);
+	cup = alloc_cld_upcall(nn);
 	if (!cup) {
 		ret = -ENOMEM;
 		goto out_err;
 	}
 
-	cup->cu_msg.cm_cmd = Cld_GraceDone;
-	cup->cu_msg.cm_u.cm_gracetime = (int64_t)nn->boot_time;
-	ret = cld_pipe_upcall(cn->cn_pipe, &cup->cu_msg);
+	cup->cu_u.cu_msg.cm_cmd = Cld_GraceDone;
+	cup->cu_u.cu_msg.cm_u.cm_gracetime = (int64_t)nn->boot_time;
+	ret = cld_pipe_upcall(cn->cn_pipe, &cup->cu_u.cu_msg);
 	if (!ret)
-		ret = cup->cu_msg.cm_status;
+		ret = cup->cu_u.cu_msg.cm_status;
 
 	free_cld_upcall(cup);
 out_err:
@@ -1279,16 +1292,16 @@ nfsd4_cld_grace_done(struct nfsd_net *nn)
 	struct cld_upcall *cup;
 	struct cld_net *cn = nn->cld_net;
 
-	cup = alloc_cld_upcall(cn);
+	cup = alloc_cld_upcall(nn);
 	if (!cup) {
 		ret = -ENOMEM;
 		goto out_err;
 	}
 
-	cup->cu_msg.cm_cmd = Cld_GraceDone;
-	ret = cld_pipe_upcall(cn->cn_pipe, &cup->cu_msg);
+	cup->cu_u.cu_msg.cm_cmd = Cld_GraceDone;
+	ret = cld_pipe_upcall(cn->cn_pipe, &cup->cu_u.cu_msg);
 	if (!ret)
-		ret = cup->cu_msg.cm_status;
+		ret = cup->cu_u.cu_msg.cm_status;
 
 	free_cld_upcall(cup);
 out_err:
@@ -1336,6 +1349,50 @@ cld_running(struct nfsd_net *nn)
 	return pipe->nreaders || pipe->nwriters;
 }
 
+static int
+nfsd4_cld_get_version(struct nfsd_net *nn)
+{
+	int ret = 0;
+	struct cld_upcall *cup;
+	struct cld_net *cn = nn->cld_net;
+	uint8_t version;
+
+	cup = alloc_cld_upcall(nn);
+	if (!cup) {
+		ret = -ENOMEM;
+		goto out_err;
+	}
+	cup->cu_u.cu_msg.cm_cmd = Cld_GetVersion;
+	ret = cld_pipe_upcall(cn->cn_pipe, &cup->cu_u.cu_msg);
+	if (!ret) {
+		ret = cup->cu_u.cu_msg.cm_status;
+		if (ret)
+			goto out_free;
+		version = cup->cu_u.cu_msg.cm_u.cm_version;
+		dprintk("%s: userspace returned version %u\n",
+				__func__, version);
+		if (version < 1)
+			version = 1;
+		else if (version > CLD_UPCALL_VERSION)
+			version = CLD_UPCALL_VERSION;
+
+		switch (version) {
+		case 1:
+			nn->client_tracking_ops = &nfsd4_cld_tracking_ops;
+			break;
+		default:
+			break;
+		}
+	}
+out_free:
+	free_cld_upcall(cup);
+out_err:
+	if (ret)
+		dprintk("%s: Unable to get version from userspace: %d\n",
+			__func__, ret);
+	return ret;
+}
+
 static int
 nfsd4_cld_tracking_init(struct net *net)
 {
@@ -1368,10 +1425,14 @@ nfsd4_cld_tracking_init(struct net *net)
 		goto err_remove;
 	}
 
+	status = nfsd4_cld_get_version(nn);
+	if (status == -EOPNOTSUPP)
+		pr_warn("NFSD: nfsdcld GetVersion upcall failed. Please upgrade nfsdcld.\n");
+
 	status = nfsd4_cld_grace_start(nn);
 	if (status) {
 		if (status == -EOPNOTSUPP)
-			printk(KERN_WARNING "NFSD: Please upgrade nfsdcld.\n");
+			pr_warn("NFSD: nfsdcld GraceStart upcall failed. Please upgrade nfsdcld.\n");
 		nfs4_release_reclaim(nn);
 		goto err_remove;
 	} else
@@ -1403,6 +1464,8 @@ static const struct nfsd4_client_tracking_ops nfsd4_cld_tracking_ops_v0 = {
 	.remove		= nfsd4_cld_remove,
 	.check		= nfsd4_cld_check_v0,
 	.grace_done	= nfsd4_cld_grace_done_v0,
+	.version	= 1,
+	.msglen		= sizeof(struct cld_msg),
 };
 
 /* For newer nfsdcld's */
@@ -1413,6 +1476,8 @@ static const struct nfsd4_client_tracking_ops nfsd4_cld_tracking_ops = {
 	.remove		= nfsd4_cld_remove,
 	.check		= nfsd4_cld_check,
 	.grace_done	= nfsd4_cld_grace_done,
+	.version	= 1,
+	.msglen		= sizeof(struct cld_msg),
 };
 
 /* upcall via usermodehelper */
@@ -1760,6 +1825,8 @@ static const struct nfsd4_client_tracking_ops nfsd4_umh_tracking_ops = {
 	.remove		= nfsd4_umh_cltrack_remove,
 	.check		= nfsd4_umh_cltrack_check,
 	.grace_done	= nfsd4_umh_cltrack_grace_done,
+	.version	= 1,
+	.msglen		= 0,
 };
 
 int
include/uapi/linux/nfsd/cld.h +10 -1
index b1e9de4f07d5..c5aad16d10c0 100644
@@ -36,7 +36,8 @@ enum cld_command {
 	Cld_Remove,		/* remove record of this cm_id */
 	Cld_Check,		/* is this cm_id allowed? */
 	Cld_GraceDone,		/* grace period is complete */
-	Cld_GraceStart,
+	Cld_GraceStart,		/* grace start (upload client records) */
+	Cld_GetVersion,		/* query max supported upcall version */
 };
 
 /* representation of long-form NFSv4 client ID */
@@ -54,7 +55,15 @@ struct cld_msg {
 	union {
 		__s64		cm_gracetime;	/* grace period start time */
 		struct cld_name	cm_name;
+		__u8		cm_version;	/* for getting max version */
 	} __attribute__((packed)) cm_u;
 } __attribute__((packed));
 
+struct cld_msg_hdr {
+	__u8		cm_vers;		/* upcall version */
+	__u8		cm_cmd;			/* upcall command */
+	__s16		cm_status;		/* return code */
+	__u32		cm_xid;			/* transaction id */
+} __attribute__((packed));
+
 #endif /* !_NFSD_CLD_H */
Signature · 1 line
2.17.2
#2 lore raw

RFC PATCH 2/2 nfsd: add support for upcall version 2

In reply to Scott Mayhew

Version 2 upcalls will allow the nfsd to include the kerberos principal string (actually the first 1024 bytes of it) in the Cld_Create upcall. If a principal is present in the svc_cred, then it will be included in the Cld_Create upcall. Upon a subsequent restart, the principal will be returned in the Cld_Gracestart downcall and stored in the reclaim_str_hashtbl so it can be used when handling reclaim opens.

---
 fs/nfsd/nfs4recover.c         | 167 +++++++++++++++++++++++++++++++---
 fs/nfsd/nfs4state.c           |   6 +-
 fs/nfsd/state.h               |   3 +-
 include/uapi/linux/nfsd/cld.h |  26 +++++-
 4 files changed, 185 insertions(+), 17 deletions(-)
fs/nfsd/nfs4recover.c +153 -14
index 58a61339d40c..682760457368 100644
@@ -64,6 +64,7 @@ struct nfsd4_client_tracking_ops {
 };
 
 static const struct nfsd4_client_tracking_ops nfsd4_cld_tracking_ops;
+static const struct nfsd4_client_tracking_ops nfsd4_cld_tracking_ops_v2;
 
 /* Globals */
 static char user_recovery_dirname[PATH_MAX] = "/var/lib/nfs/v4recovery";
@@ -177,6 +178,7 @@ __nfsd4_create_reclaim_record_grace(struct nfs4_client *clp,
 		const char *dname, int len, struct nfsd_net *nn)
 {
 	struct xdr_netobj name;
+	struct xdr_netobj princ = { .len = 0, .data = NULL };
 	struct nfs4_client_reclaim *crp;
 
 	name.data = kmemdup(dname, len, GFP_KERNEL);
@@ -186,7 +188,7 @@ __nfsd4_create_reclaim_record_grace(struct nfs4_client *clp,
 		return;
 	}
 	name.len = len;
-	crp = nfs4_client_to_reclaim(name, nn);
+	crp = nfs4_client_to_reclaim(name, princ, nn);
 	if (!crp) {
 		kfree(name.data);
 		return;
@@ -486,6 +488,7 @@ static int
 load_recdir(struct dentry *parent, struct dentry *child, struct nfsd_net *nn)
 {
 	struct xdr_netobj name;
+	struct xdr_netobj princ = { .len = 0, .data = NULL };
 
 	if (child->d_name.len != HEXDIR_LEN - 1) {
 		printk("%s: illegal name %pd in recovery directory\n",
@@ -500,7 +503,7 @@ load_recdir(struct dentry *parent, struct dentry *child, struct nfsd_net *nn)
 		goto out;
 	}
 	name.len = HEXDIR_LEN;
-	if (!nfs4_client_to_reclaim(name, nn))
+	if (!nfs4_client_to_reclaim(name, princ, nn))
 		kfree(name.data);
 out:
 	return 0;
@@ -746,6 +749,7 @@ struct cld_upcall {
 	union {
 		struct cld_msg_hdr	 cu_hdr;
 		struct cld_msg		 cu_msg;
+		struct cld_msg_v2	 cu_msg_v2;
 	} cu_u;
 };
 
@@ -792,12 +796,12 @@ cld_pipe_upcall(struct rpc_pipe *pipe, void *cmsg)
 }
 
 static ssize_t
-__cld_pipe_inprogress_downcall(const struct cld_msg __user *cmsg,
+__cld_pipe_inprogress_downcall(const struct cld_msg_v2 __user *cmsg,
 		struct nfsd_net *nn)
 {
 	uint8_t cmd;
-	struct xdr_netobj name;
-	uint16_t namelen;
+	struct xdr_netobj name, princ = { .len = 0, .data = NULL };
+	uint16_t namelen, princlen;
 	struct cld_net *cn = nn->cld_net;
 
 	if (get_user(cmd, &cmsg->cm_cmd)) {
@@ -805,18 +809,43 @@ __cld_pipe_inprogress_downcall(const struct cld_msg __user *cmsg,
 		return -EFAULT;
 	}
 	if (cmd == Cld_GraceStart) {
-		if (get_user(namelen, &cmsg->cm_u.cm_name.cn_len))
-			return -EFAULT;
-		name.data = memdup_user(&cmsg->cm_u.cm_name.cn_id, namelen);
-		if (IS_ERR_OR_NULL(name.data))
-			return -EFAULT;
-		name.len = namelen;
+		if (nn->client_tracking_ops->version >= 2) {
+			const struct cld_clntinfo __user *ci;
+
+			ci = &cmsg->cm_u.cm_clntinfo;
+			if (get_user(namelen, &ci->cc_name.cn_len))
+				return -EFAULT;
+			name.data = memdup_user(&ci->cc_name.cn_id, namelen);
+			if (IS_ERR_OR_NULL(name.data))
+				return -EFAULT;
+			name.len = namelen;
+			get_user(princlen, &ci->cc_principal.cp_len);
+			if (princlen > 0) {
+				princ.data = memdup_user(
+						&ci->cc_principal.cp_data,
+						princlen);
+				if (IS_ERR_OR_NULL(princ.data))
+					return -EFAULT;
+				princ.len = princlen;
+			} else
+				princ.len = 0;
+		} else {
+			const struct cld_name __user *cnm;
+
+			cnm = &cmsg->cm_u.cm_name;
+			if (get_user(namelen, &cnm->cn_len))
+				return -EFAULT;
+			name.data = memdup_user(&cnm->cn_id, namelen);
+			if (IS_ERR_OR_NULL(name.data))
+				return -EFAULT;
+			name.len = namelen;
+		}
 		if (name.len > 5 && memcmp(name.data, "hash:", 5) == 0) {
 			name.len = name.len - 5;
 			memmove(name.data, name.data + 5, name.len);
 			cn->cn_has_legacy = true;
 		}
-		if (!nfs4_client_to_reclaim(name, nn)) {
+		if (!nfs4_client_to_reclaim(name, princ, nn)) {
 			kfree(name.data);
 			return -EFAULT;
 		}
@@ -830,7 +859,7 @@ cld_pipe_downcall(struct file *filp, const char __user *src, size_t mlen)
 {
 	struct cld_upcall *tmp, *cup;
 	struct cld_msg_hdr __user *hdr = (struct cld_msg_hdr __user *)src;
-	struct cld_msg __user *cmsg = (struct cld_msg __user *)src;
+	struct cld_msg_v2 __user *cmsg = (struct cld_msg_v2 __user *)src;
 	uint32_t xid;
 	struct nfsd_net *nn = net_generic(file_inode(filp)->i_sb->s_fs_info,
 						nfsd_net_id);
@@ -881,7 +910,7 @@ cld_pipe_downcall(struct file *filp, const char __user *src, size_t mlen)
 	if (status == -EINPROGRESS)
 		return __cld_pipe_inprogress_downcall(cmsg, nn);
 
-	if (copy_from_user(&cup->cu_u.cu_msg, src, mlen) != 0)
+	if (copy_from_user(&cup->cu_u.cu_msg_v2, src, mlen) != 0)
 		return -EFAULT;
 
 	complete(&cup->cu_done);
@@ -1103,6 +1132,54 @@ nfsd4_cld_create(struct nfs4_client *clp)
 				"record on stable storage: %d\n", ret);
 }
 
+/* Ask daemon to create a new record */
+static void
+nfsd4_cld_create_v2(struct nfs4_client *clp)
+{
+	int ret, len;
+	struct cld_upcall *cup;
+	struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
+	struct cld_net *cn = nn->cld_net;
+	struct cld_msg_v2 *cmsg;
+
+	/* Don't upcall if it's already stored */
+	if (test_bit(NFSD4_CLIENT_STABLE, &clp->cl_flags))
+		return;
+
+	cup = alloc_cld_upcall(nn);
+	if (!cup) {
+		ret = -ENOMEM;
+		goto out_err;
+	}
+
+	cmsg = &cup->cu_u.cu_msg_v2;
+	cmsg->cm_cmd = Cld_Create;
+	cmsg->cm_u.cm_clntinfo.cc_name.cn_len = clp->cl_name.len;
+	memcpy(cmsg->cm_u.cm_clntinfo.cc_name.cn_id, clp->cl_name.data,
+			clp->cl_name.len);
+	if (clp->cl_cred.cr_principal) {
+		len = strlen(clp->cl_cred.cr_principal);
+		if (len > NFS4_OPAQUE_LIMIT)
+			len = NFS4_OPAQUE_LIMIT;
+		cmsg->cm_u.cm_clntinfo.cc_principal.cp_len = len;
+		memcpy(cmsg->cm_u.cm_clntinfo.cc_principal.cp_data,
+				clp->cl_cred.cr_principal, len);
+	} else
+		cmsg->cm_u.cm_clntinfo.cc_principal.cp_len = 0;
+
+	ret = cld_pipe_upcall(cn->cn_pipe, cmsg);
+	if (!ret) {
+		ret = cmsg->cm_status;
+		set_bit(NFSD4_CLIENT_STABLE, &clp->cl_flags);
+	}
+
+	free_cld_upcall(cup);
+out_err:
+	if (ret)
+		pr_err("NFSD: Unable to create client record on stable storage: %d\n",
+				ret);
+}
+
 /* Ask daemon to create a new record */
 static void
 nfsd4_cld_remove(struct nfs4_client *clp)
@@ -1229,6 +1306,53 @@ nfsd4_cld_check(struct nfs4_client *clp)
 	return 0;
 }
 
+static int
+nfsd4_cld_check_v2(struct nfs4_client *clp)
+{
+	struct nfs4_client_reclaim *crp;
+	struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
+	struct cld_net *cn = nn->cld_net;
+	int status;
+	char dname[HEXDIR_LEN];
+	struct xdr_netobj name;
+
+	/* did we already find that this client is stable? */
+	if (test_bit(NFSD4_CLIENT_STABLE, &clp->cl_flags))
+		return 0;
+
+	/* look for it in the reclaim hashtable otherwise */
+	crp = nfsd4_find_reclaim_client(clp->cl_name, nn);
+	if (crp)
+		goto found;
+
+	if (cn->cn_has_legacy) {
+		status = nfs4_make_rec_clidname(dname, &clp->cl_name);
+		if (status)
+			return -ENOENT;
+
+		name.data = kmemdup(dname, HEXDIR_LEN, GFP_KERNEL);
+		if (!name.data) {
+			dprintk("%s: failed to allocate memory for name.data\n",
+					__func__);
+			return -ENOENT;
+		}
+		name.len = HEXDIR_LEN;
+		crp = nfsd4_find_reclaim_client(name, nn);
+		kfree(name.data);
+		if (crp)
+			goto found;
+
+	}
+	return -ENOENT;
+found:
+	if (crp->cr_principal.len)
+		if (memcmp(crp->cr_principal.data, clp->cl_cred.cr_principal,
+				crp->cr_principal.len))
+			return -ENOENT;
+	crp->cr_clp = clp;
+	return 0;
+}
+
 static int
 nfsd4_cld_grace_start(struct nfsd_net *nn)
 {
@@ -1380,6 +1504,9 @@ nfsd4_cld_get_version(struct nfsd_net *nn)
 		case 1:
 			nn->client_tracking_ops = &nfsd4_cld_tracking_ops;
 			break;
+		case 2:
+			nn->client_tracking_ops = &nfsd4_cld_tracking_ops_v2;
+			break;
 		default:
 			break;
 		}
@@ -1480,6 +1607,18 @@ static const struct nfsd4_client_tracking_ops nfsd4_cld_tracking_ops = {
 	.msglen		= sizeof(struct cld_msg),
 };
 
+/* v2 create/check ops include the principal, if available */
+static const struct nfsd4_client_tracking_ops nfsd4_cld_tracking_ops_v2 = {
+	.init		= nfsd4_cld_tracking_init,
+	.exit		= nfsd4_cld_tracking_exit,
+	.create		= nfsd4_cld_create_v2,
+	.remove		= nfsd4_cld_remove,
+	.check		= nfsd4_cld_check_v2,
+	.grace_done	= nfsd4_cld_grace_done,
+	.version	= 2,
+	.msglen		= sizeof(struct cld_msg_v2),
+};
+
 /* upcall via usermodehelper */
 static char cltrack_prog[PATH_MAX] = "/sbin/nfsdcltrack";
 module_param_string(cltrack_prog, cltrack_prog, sizeof(cltrack_prog),
fs/nfsd/nfs4state.c +5 -1
index 7857942c5ca6..3d2fa7fb079e 100644
@@ -6888,7 +6888,8 @@ nfs4_has_reclaimed_state(struct xdr_netobj name, struct nfsd_net *nn)
  * will be freed in nfs4_remove_reclaim_record in the normal case).
  */
 struct nfs4_client_reclaim *
-nfs4_client_to_reclaim(struct xdr_netobj name, struct nfsd_net *nn)
+nfs4_client_to_reclaim(struct xdr_netobj name, struct xdr_netobj princ,
+		struct nfsd_net *nn)
 {
 	unsigned int strhashval;
 	struct nfs4_client_reclaim *crp;
@@ -6901,6 +6902,8 @@ nfs4_client_to_reclaim(struct xdr_netobj name, struct nfsd_net *nn)
 		list_add(&crp->cr_strhash, &nn->reclaim_str_hashtbl[strhashval]);
 		crp->cr_name.data = name.data;
 		crp->cr_name.len = name.len;
+		crp->cr_principal.data = princ.data;
+		crp->cr_principal.len = princ.len;
 		crp->cr_clp = NULL;
 		nn->reclaim_str_hashtbl_size++;
 	}
@@ -6912,6 +6915,7 @@ nfs4_remove_reclaim_record(struct nfs4_client_reclaim *crp, struct nfsd_net *nn)
 {
 	list_del(&crp->cr_strhash);
 	kfree(crp->cr_name.data);
+	kfree(crp->cr_principal.data);
 	kfree(crp);
 	nn->reclaim_str_hashtbl_size--;
 }
fs/nfsd/state.h +2 -1
index 5dbd16946e8e..aac63470164c 100644
@@ -378,6 +378,7 @@ struct nfs4_client_reclaim {
 	struct list_head	cr_strhash;	/* hash by cr_name */
 	struct nfs4_client	*cr_clp;	/* pointer to associated clp */
 	struct xdr_netobj	cr_name;	/* recovery dir name */
+	struct xdr_netobj	cr_principal;
 };
 
 /* A reasonable value for REPLAY_ISIZE was estimated as follows:  
@@ -645,7 +646,7 @@ extern void nfsd4_shutdown_callback(struct nfs4_client *);
 extern void nfsd4_shutdown_copy(struct nfs4_client *clp);
 extern void nfsd4_prepare_cb_recall(struct nfs4_delegation *dp);
 extern struct nfs4_client_reclaim *nfs4_client_to_reclaim(struct xdr_netobj name,
-							struct nfsd_net *nn);
+				struct xdr_netobj princ, struct nfsd_net *nn);
 extern bool nfs4_has_reclaimed_state(struct xdr_netobj name, struct nfsd_net *nn);
 
 struct nfs4_file *find_file(struct knfsd_fh *fh);
include/uapi/linux/nfsd/cld.h +25 -1
index c5aad16d10c0..8d200f0b07d6 100644
@@ -26,7 +26,7 @@
 #include <linux/types.h>
 
 /* latest upcall version available */
-#define CLD_UPCALL_VERSION 1
+#define CLD_UPCALL_VERSION 2
 
 /* defined by RFC3530 */
 #define NFS4_OPAQUE_LIMIT 1024
@@ -46,6 +46,17 @@ struct cld_name {
 	unsigned char	cn_id[NFS4_OPAQUE_LIMIT];	/* client-provided */
 } __attribute__((packed));
 
+/* principal of the form servicetype@hostname */
+struct cld_principal {
+	__u16		cp_len;				/* length of cp_data */
+	unsigned char	cp_data[NFS4_OPAQUE_LIMIT];	/* princ from cred */
+} __attribute__((packed));
+
+struct cld_clntinfo {
+	struct cld_name		cc_name;
+	struct cld_principal	cc_principal;
+} __attribute__((packed));
+
 /* message struct for communication with userspace */
 struct cld_msg {
 	__u8		cm_vers;		/* upcall version */
@@ -59,6 +70,19 @@ struct cld_msg {
 	} __attribute__((packed)) cm_u;
 } __attribute__((packed));
 
+/* version 2 message includes the principal */
+struct cld_msg_v2 {
+	__u8		cm_vers;		/* upcall version */
+	__u8		cm_cmd;			/* upcall command */
+	__s16		cm_status;		/* return code */
+	__u32		cm_xid;			/* transaction id */
+	union {
+		struct cld_name	cm_name;
+		__u8		cm_version;	/* for getting max version */
+		struct cld_clntinfo cm_clntinfo; /* name & princ */
+	} __attribute__((packed)) cm_u;
+} __attribute__((packed));
+
 struct cld_msg_hdr {
 	__u8		cm_vers;		/* upcall version */
 	__u8		cm_cmd;			/* upcall command */
Signature · 1 line
2.17.2
#3 lore raw
In reply to Scott Mayhew
Quoted 5 lines · Scott Mayhew

At the spring bakeathon, Chuck suggested that we should store the kerberos principal in addition to the client id string in nfsdcld. The idea is to prevent an illegitimate client from reclaiming another client's opens by supplying that client's id string. This is an initial attempt at doing that.

Great to see this, thanks!

Quoted 7 lines

Initially I had played with the idea of hanging a version number off either the cld_net or nfsd_net structure, exposing that via a proc file, and having the userspace daemon write the desired version number to the proc file prior to opening the rpc_pipefs file. That was potentially problematic if nfsd was already trying to queue an upcall using a version that nfscld didn't support... so I like the GetVersion upcall idea better.

Sounds OK to me.

It queries the version once on nfsd startup and sticks with that version. We allow rpc.mountd to be restarted while nfsd is running. So the one case I think it wouldn't handle is the case where somebody downgrades mountd while nfsd is running.

My feeling is that handling that case would be way too much trouble, so actually I'm going to consider that a plus. But it might be worth documenting (if only in a patch changelog).

Quoted 8 lines

The second patch actually adds the v2 message, which adds the principal (actually just the first 1024 bytes) to the Cld_Create upcall and to the Cld_GraceStart downcall (which is what loads the data in the reclaim_str_hashtbl). I couldn't really figure out what the maximum length of a kerberos principal actually is (looking at krb5.h the length field in the struct krb5_data is an unsigned int, so I guess it can be pretty big). I don't think the principal will be that large in practice, and even if it is the first 1024 bytes should be sufficient for our purposes.

How does it fail when principals are longer? Does it error out, or treat two principals as equal if they agree in the first 1024 bytes?

--b.

#4 lore raw
Quoted 2 lines · J. Bruce Fields

How does it fail when principals are longer? Does it error out, or treat two principals as equal if they agree in the first 1024 bytes?

I guess it's being compared against a string passed from gss-proxy? We could also check for limits there.

--b.

#5 lore raw
Quoted 6 lines · J. Bruce Fields
Quoted 2 lines · J. Bruce Fields

How does it fail when principals are longer? Does it error out, or treat two principals as equal if they agree in the first 1024 bytes?

I guess it's being compared against a string passed from gss-proxy? We could also check for limits there.

I'm using cr_principal (servicetype@hostname) since it's set by both gssproxy and rpc.svcgssd. cr_raw_principal (servicetype/hostname@REALM) is only set by gssproxy.

-Scott

Quoted 2 lines

--b.

#6 lore raw
In reply to J. Bruce Fields
Quoted 17 lines · J. Bruce Fields
Quoted 5 lines · Scott Mayhew

At the spring bakeathon, Chuck suggested that we should store the kerberos principal in addition to the client id string in nfsdcld. The idea is to prevent an illegitimate client from reclaiming another client's opens by supplying that client's id string. This is an initial attempt at doing that.

Great to see this, thanks!

Quoted 7 lines

Initially I had played with the idea of hanging a version number off either the cld_net or nfsd_net structure, exposing that via a proc file, and having the userspace daemon write the desired version number to the proc file prior to opening the rpc_pipefs file. That was potentially problematic if nfsd was already trying to queue an upcall using a version that nfscld didn't support... so I like the GetVersion upcall idea better.

Quoted 6 lines

Sounds OK to me.

It queries the version once on nfsd startup and sticks with that version. We allow rpc.mountd to be restarted while nfsd is running. So the one case I think it wouldn't handle is the case where somebody downgrades mountd while nfsd is running.

I'm assuming you meant nfsdcld rather than mountd here... currently if someone were to downgrade nfsdcld while nfsd is running then that case wouldn't be handled, so a restart of nfsd would also be necessary.

Quoted 5 lines

My feeling is that handling that case would be way too much trouble, so actually I'm going to consider that a plus. But it might be worth documenting (if only in a patch changelog).

To handle that scenario, We'd need to change the database schema. I'd really rather do that in an external script than stick downgrade logic into nfsdcld... mainly because I'd want to check if there was any data in the columns being eliminated and warn the user & ask for confirmation.

We'd also need to change how nfsdcld reacts when it gets a message version it doesn't support. Currently it just reopens the pipe file, which causes the upcall to fail with -EAGAIN, which causes nfsd to retry the upcall, and it just goes into a loop. I could implement a "version not supported" downcall. I'm not sure what error number to use... maybe -EPROTONOSUPP? Also even if we implemented a "version not supported" downcall now, that wouldn't handle the problem with existing nfsdcld's. The only thing I can think of there is to add a counter to cld_pipe_upcall() and exit the loop after a certain number of iterations (10? 100? 1000?).

Quoted 11 lines
Quoted 8 lines

The second patch actually adds the v2 message, which adds the principal (actually just the first 1024 bytes) to the Cld_Create upcall and to the Cld_GraceStart downcall (which is what loads the data in the reclaim_str_hashtbl). I couldn't really figure out what the maximum length of a kerberos principal actually is (looking at krb5.h the length field in the struct krb5_data is an unsigned int, so I guess it can be pretty big). I don't think the principal will be that large in practice, and even if it is the first 1024 bytes should be sufficient for our purposes.

How does it fail when principals are longer? Does it error out, or treat two principals as equal if they agree in the first 1024 bytes?

It treats them as equal.

Quoted 2 lines

--b.

#7 lore raw
Quoted 19 lines · Scott Mayhew
Quoted 17 lines · J. Bruce Fields
Quoted 5 lines · Scott Mayhew

At the spring bakeathon, Chuck suggested that we should store the kerberos principal in addition to the client id string in nfsdcld. The idea is to prevent an illegitimate client from reclaiming another client's opens by supplying that client's id string. This is an initial attempt at doing that.

Great to see this, thanks!

Quoted 7 lines

Initially I had played with the idea of hanging a version number off either the cld_net or nfsd_net structure, exposing that via a proc file, and having the userspace daemon write the desired version number to the proc file prior to opening the rpc_pipefs file. That was potentially problematic if nfsd was already trying to queue an upcall using a version that nfscld didn't support... so I like the GetVersion upcall idea better.

Quoted 8 lines
Quoted 6 lines

Sounds OK to me.

It queries the version once on nfsd startup and sticks with that version. We allow rpc.mountd to be restarted while nfsd is running. So the one case I think it wouldn't handle is the case where somebody downgrades mountd while nfsd is running.

I'm assuming you meant nfsdcld rather than mountd here...

Oops, right.

Quoted 3 lines

currently if
someone were to downgrade nfsdcld while nfsd is running then that case wouldn't be handled, so a restart of nfsd would also be necessary.

Right.

Quoted 10 lines
Quoted 3 lines

My feeling is that handling that case would be way too much trouble, so actually I'm going to consider that a plus. But it might be worth documenting (if only in a patch changelog).

To handle that scenario, We'd need to change the database schema. I'd really rather do that in an external script than stick downgrade logic into nfsdcld... mainly because I'd want to check if there was any data in the columns being eliminated and warn the user & ask for confirmation.

Quoted 10 lines

We'd also need to change how nfsdcld reacts when it gets a message version it doesn't support. Currently it just reopens the pipe file, which causes the upcall to fail with -EAGAIN, which causes nfsd to retry the upcall, and it just goes into a loop. I could implement a "version not supported" downcall. I'm not sure what error number to use... maybe -EPROTONOSUPP? Also even if we implemented a "version not supported" downcall now, that wouldn't handle the problem with existing nfsdcld's. The only thing I can think of there is to add a counter to cld_pipe_upcall() and exit the loop after a certain number of iterations (10? 100? 1000?).

This sounds too complicated. Let's just add a note that downgrades require an nfsd restart.

Quoted 9 lines
Quoted 9 lines
Quoted 8 lines

The second patch actually adds the v2 message, which adds the principal (actually just the first 1024 bytes) to the Cld_Create upcall and to the Cld_GraceStart downcall (which is what loads the data in the reclaim_str_hashtbl). I couldn't really figure out what the maximum length of a kerberos principal actually is (looking at krb5.h the length field in the struct krb5_data is an unsigned int, so I guess it can be pretty big). I don't think the principal will be that large in practice, and even if it is the first 1024 bytes should be sufficient for our purposes.

Quoted 4 lines
Quoted 2 lines

How does it fail when principals are longer? Does it error out, or treat two principals as equal if they agree in the first 1024 bytes?

It treats them as equal.

Got it, thanks.

--b.