[PATCH v2 01/11] xattr: break delegations in {set,remove}xattr and add _locked versions

[PATCH v2 01/11] xattr: break delegations in {set,remove}xattr and add _locked versions

[Frank van der Linden]

set/removexattr on an exported filesystem should break NFS delegations.
This is true in general, but also for the upcoming support for
RFC 8726 (NFSv4 extended attribute support). Make sure that they do.

Additonally, they need to grow a _locked variant, since the NFS server code
will call this with i_rwsem held. It needs to do that in fh_lock to be
able to atomically provide the before and after change attributes.

Cc: stable@vger.kernel.org
Cc: linux-fsdevel@vger.kernel.org
Cc: Al Viro <viro@zeniv.linux.org.uk>
Signed-off-by: Frank van der Linden <fllinden@amazon.com>
---
 fs/xattr.c            | 84 +++++++++++++++++++++++++++++++++++++++----
 include/linux/xattr.h |  2 ++
 2 files changed, 79 insertions(+), 7 deletions(-)

diff --git a/fs/xattr.c b/fs/xattr.c
index 90dd78f0eb27..f2854570d411 100644
--- a/fs/xattr.c
+++ b/fs/xattr.c
@@ -204,10 +204,22 @@ int __vfs_setxattr_noperm(struct dentry *dentry, const char *name,
 	return error;
 }
 
-
+/**
+ * __vfs_setxattr_locked: set an extended attribute while holding the inode
+ * lock
+ *
+ *  @dentry - object to perform setxattr on
+ *  @name - xattr name to set
+ *  @value - value to set @name to
+ *  @size - size of @value
+ *  @flags - flags to pass into filesystem operations
+ *  @delegated_inode - on return, will contain an inode pointer that
+ *  a delegation was broken on, NULL if none.
+ */
 int
-vfs_setxattr(struct dentry *dentry, const char *name, const void *value,
-		size_t size, int flags)
+__vfs_setxattr_locked(struct dentry *dentry, const char *name,
+		const void *value, size_t size, int flags,
+		struct inode **delegated_inode)
 {
 	struct inode *inode = dentry->d_inode;
 	int error;
@@ -216,15 +228,40 @@ vfs_setxattr(struct dentry *dentry, const char *name, const void *value,
 	if (error)
 		return error;
 
-	inode_lock(inode);
 	error = security_inode_setxattr(dentry, name, value, size, flags);
 	if (error)
 		goto out;
 
+	error = try_break_deleg(inode, delegated_inode);
+	if (error)
+		goto out;
+
 	error = __vfs_setxattr_noperm(dentry, name, value, size, flags);
 
 out:
+	return error;
+}
+EXPORT_SYMBOL_GPL(__vfs_setxattr_locked);
+
+int
+vfs_setxattr(struct dentry *dentry, const char *name, const void *value,
+		size_t size, int flags)
+{
+	struct inode *inode = dentry->d_inode;
+	struct inode *delegated_inode = NULL;
+	int error;
+
+retry_deleg:
+	inode_lock(inode);
+	error = __vfs_setxattr_locked(dentry, name, value, size, flags,
+	    &delegated_inode);
 	inode_unlock(inode);
+
+	if (delegated_inode) {
+		error = break_deleg_wait(&delegated_inode);
+		if (!error)
+			goto retry_deleg;
+	}
 	return error;
 }
 EXPORT_SYMBOL_GPL(vfs_setxattr);
@@ -378,8 +415,18 @@ __vfs_removexattr(struct dentry *dentry, const char *name)
 }
 EXPORT_SYMBOL(__vfs_removexattr);
 
+/**
+ * __vfs_removexattr_locked: set an extended attribute while holding the inode
+ * lock
+ *
+ *  @dentry - object to perform setxattr on
+ *  @name - name of xattr to remove
+ *  @delegated_inode - on return, will contain an inode pointer that
+ *  a delegation was broken on, NULL if none.
+ */
 int
-vfs_removexattr(struct dentry *dentry, const char *name)
+__vfs_removexattr_locked(struct dentry *dentry, const char *name,
+		struct inode **delegated_inode)
 {
 	struct inode *inode = dentry->d_inode;
 	int error;
@@ -388,11 +435,14 @@ vfs_removexattr(struct dentry *dentry, const char *name)
 	if (error)
 		return error;
 
-	inode_lock(inode);
 	error = security_inode_removexattr(dentry, name);
 	if (error)
 		goto out;
 
+	error = try_break_deleg(inode, delegated_inode);
+	if (error)
+		goto out;
+
 	error = __vfs_removexattr(dentry, name);
 
 	if (!error) {
@@ -401,12 +451,32 @@ vfs_removexattr(struct dentry *dentry, const char *name)
 	}
 
 out:
+	return error;
+}
+EXPORT_SYMBOL_GPL(__vfs_removexattr_locked);
+
+int
+vfs_removexattr(struct dentry *dentry, const char *name)
+{
+	struct inode *inode = dentry->d_inode;
+	struct inode *delegated_inode = NULL;
+	int error;
+
+retry_deleg:
+	inode_lock(inode);
+	error = __vfs_removexattr_locked(dentry, name, &delegated_inode);
 	inode_unlock(inode);
+
+	if (delegated_inode) {
+		error = break_deleg_wait(&delegated_inode);
+		if (!error)
+			goto retry_deleg;
+	}
+
 	return error;
 }
 EXPORT_SYMBOL_GPL(vfs_removexattr);
 
-
 /*
  * Extended attribute SET operations
  */
diff --git a/include/linux/xattr.h b/include/linux/xattr.h
index 6dad031be3c2..3a71ad716da5 100644
--- a/include/linux/xattr.h
+++ b/include/linux/xattr.h
@@ -51,8 +51,10 @@ ssize_t vfs_getxattr(struct dentry *, const char *, void *, size_t);
 ssize_t vfs_listxattr(struct dentry *d, char *list, size_t size);
 int __vfs_setxattr(struct dentry *, struct inode *, const char *, const void *, size_t, int);
 int __vfs_setxattr_noperm(struct dentry *, const char *, const void *, size_t, int);
+int __vfs_setxattr_locked(struct dentry *, const char *, const void *, size_t, int, struct inode **);
 int vfs_setxattr(struct dentry *, const char *, const void *, size_t, int);
 int __vfs_removexattr(struct dentry *, const char *);
+int __vfs_removexattr_locked(struct dentry *, const char *, struct inode **);
 int vfs_removexattr(struct dentry *, const char *);
 
 ssize_t generic_listxattr(struct dentry *dentry, char *buffer, size_t buffer_size);
-- 
2.17.2

[PATCH v2 02/11] xattr: add a function to check if a namespace is supported

[Frank van der Linden]

Add a function that checks is an extended attribute namespace is
supported for an inode.

To be used by the nfs server code when being queried for extended
attributes support.

Cc: linux-fsdevel@vger.kernel.org
Cc: Al Viro <viro@zeniv.linux.org.uk>
Signed-off-by: Frank van der Linden <fllinden@amazon.com>
---
 fs/xattr.c            | 27 +++++++++++++++++++++++++++
 include/linux/xattr.h |  2 ++
 2 files changed, 29 insertions(+)

diff --git a/fs/xattr.c b/fs/xattr.c
index f2854570d411..1b079550a96d 100644
--- a/fs/xattr.c
+++ b/fs/xattr.c
@@ -134,6 +134,33 @@ xattr_permission(struct inode *inode, const char *name, int mask)
 	return inode_permission(inode, mask);
 }
 
+/*
+ * Look for any handler that deals with the specified namespace.
+ */
+int
+xattr_supported_namespace(struct inode *inode, const char *prefix)
+{
+	const struct xattr_handler **handlers = inode->i_sb->s_xattr;
+	const struct xattr_handler *handler;
+	size_t preflen;
+
+	if (!(inode->i_opflags & IOP_XATTR)) {
+		if (unlikely(is_bad_inode(inode)))
+			return -EIO;
+		return -EOPNOTSUPP;
+	}
+
+	preflen = strlen(prefix);
+
+	for_each_xattr_handler(handlers, handler) {
+		if (!strncmp(xattr_prefix(handler), prefix, preflen))
+			return 0;
+	}
+
+	return -EOPNOTSUPP;
+}
+EXPORT_SYMBOL(xattr_supported_namespace);
+
 int
 __vfs_setxattr(struct dentry *dentry, struct inode *inode, const char *name,
 	       const void *value, size_t size, int flags)
diff --git a/include/linux/xattr.h b/include/linux/xattr.h
index 3a71ad716da5..32e377602068 100644
--- a/include/linux/xattr.h
+++ b/include/linux/xattr.h
@@ -61,6 +61,8 @@ ssize_t generic_listxattr(struct dentry *dentry, char *buffer, size_t buffer_siz
 ssize_t vfs_getxattr_alloc(struct dentry *dentry, const char *name,
 			   char **xattr_value, size_t size, gfp_t flags);
 
+int xattr_supported_namespace(struct inode *inode, const char *prefix);
+
 static inline const char *xattr_prefix(const struct xattr_handler *handler)
 {
 	return handler->prefix ?: handler->name;
-- 
2.17.2

[PATCH v2 10/11] nfsd: implement the xattr functions and en/decode logic

[Frank van der Linden]

Implement the main entry points for the *XATTR operations.

Add functions to calculate the reply size for the user extended attribute
operations, and implement the XDR encode / decode logic for these
operations.

Add the user extended attributes operations to nfsd4_ops.

Cc: linux-fsdevel@vger.kernel.org
Signed-off-by: Frank van der Linden <fllinden@amazon.com>
---
 fs/nfsd/nfs4proc.c   | 132 +++++++++++++
 fs/nfsd/nfs4xdr.c    | 450 +++++++++++++++++++++++++++++++++++++++++++
 include/linux/nfs4.h |   2 +-
 3 files changed, 583 insertions(+), 1 deletion(-)

diff --git a/fs/nfsd/nfs4proc.c b/fs/nfsd/nfs4proc.c
index ee317ae0b609..bb00982f5c6a 100644
--- a/fs/nfsd/nfs4proc.c
+++ b/fs/nfsd/nfs4proc.c
@@ -2098,6 +2098,80 @@ nfsd4_layoutreturn(struct svc_rqst *rqstp,
 }
 #endif /* CONFIG_NFSD_PNFS */
 
+static __be32
+nfsd4_getxattr(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
+	       union nfsd4_op_u *u)
+{
+	struct nfsd4_getxattr *getxattr = &u->getxattr;
+
+	return nfsd_getxattr(rqstp, &cstate->current_fh,
+			     getxattr->getxa_name, getxattr->getxa_buf,
+			     &getxattr->getxa_len);
+}
+
+static __be32
+nfsd4_setxattr(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
+	   union nfsd4_op_u *u)
+{
+	struct nfsd4_setxattr *setxattr = &u->setxattr;
+	__be32 ret;
+
+	if (opens_in_grace(SVC_NET(rqstp)))
+		return nfserr_grace;
+
+	ret = nfsd_setxattr(rqstp, &cstate->current_fh, setxattr->setxa_name,
+			    setxattr->setxa_buf, setxattr->setxa_len,
+			    setxattr->setxa_flags);
+
+	if (!ret)
+		set_change_info(&setxattr->setxa_cinfo, &cstate->current_fh);
+
+	return ret;
+}
+
+static __be32
+nfsd4_listxattrs(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
+	   union nfsd4_op_u *u)
+{
+	int len;
+	__be32 ret;
+
+	/*
+	 * Get the entire list, then copy out only the user attributes
+	 * in the encode function. lsxa_buf was previously allocated as
+	 * tmp svc space, and will be automatically freed later.
+	 */
+	len = XATTR_LIST_MAX;
+
+	ret = nfsd_listxattr(rqstp, &cstate->current_fh, u->listxattrs.lsxa_buf,
+			     &len);
+	if (ret)
+		return ret;
+
+	u->listxattrs.lsxa_len = len;
+
+	return nfs_ok;
+}
+
+static __be32
+nfsd4_removexattr(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
+	   union nfsd4_op_u *u)
+{
+	struct nfsd4_removexattr *removexattr = &u->removexattr;
+	__be32 ret;
+
+	if (opens_in_grace(SVC_NET(rqstp)))
+		return nfserr_grace;
+
+	ret = nfsd_removexattr(rqstp, &cstate->current_fh,
+	    removexattr->rmxa_name);
+
+	if (!ret)
+		set_change_info(&removexattr->rmxa_cinfo, &cstate->current_fh);
+
+	return ret;
+}
+
 /*
  * NULL call.
  */
@@ -2705,6 +2779,42 @@ static inline u32 nfsd4_seek_rsize(struct svc_rqst *rqstp, struct nfsd4_op *op)
 	return (op_encode_hdr_size + 3) * sizeof(__be32);
 }
 
+static inline u32 nfsd4_getxattr_rsize(struct svc_rqst *rqstp,
+				       struct nfsd4_op *op)
+{
+	u32 maxcount, rlen;
+
+	maxcount = svc_max_payload(rqstp);
+	rlen = min_t(u32, XATTR_SIZE_MAX, maxcount);
+
+	return (op_encode_hdr_size + 1 + XDR_QUADLEN(rlen)) * sizeof(__be32);
+}
+
+static inline u32 nfsd4_setxattr_rsize(struct svc_rqst *rqstp,
+				       struct nfsd4_op *op)
+{
+	return (op_encode_hdr_size + op_encode_change_info_maxsz)
+		* sizeof(__be32);
+}
+static inline u32 nfsd4_listxattrs_rsize(struct svc_rqst *rqstp,
+					 struct nfsd4_op *op)
+{
+	u32 maxcount, rlen;
+
+	maxcount = svc_max_payload(rqstp);
+	rlen = min(op->u.listxattrs.lsxa_maxcount, maxcount);
+
+	return (op_encode_hdr_size + 4 + XDR_QUADLEN(rlen)) * sizeof(__be32);
+}
+
+static inline u32 nfsd4_removexattr_rsize(struct svc_rqst *rqstp,
+					  struct nfsd4_op *op)
+{
+	return (op_encode_hdr_size + op_encode_change_info_maxsz)
+		* sizeof(__be32);
+}
+
+
 static const struct nfsd4_operation nfsd4_ops[] = {
 	[OP_ACCESS] = {
 		.op_func = nfsd4_access,
@@ -3086,6 +3196,28 @@ static const struct nfsd4_operation nfsd4_ops[] = {
 		.op_name = "OP_COPY_NOTIFY",
 		.op_rsize_bop = nfsd4_copy_notify_rsize,
 	},
+	[OP_GETXATTR] = {
+		.op_func = nfsd4_getxattr,
+		.op_name = "OP_GETXATTR",
+		.op_rsize_bop = nfsd4_getxattr_rsize,
+	},
+	[OP_SETXATTR] = {
+		.op_func = nfsd4_setxattr,
+		.op_flags = OP_MODIFIES_SOMETHING | OP_CACHEME,
+		.op_name = "OP_SETXATTR",
+		.op_rsize_bop = nfsd4_setxattr_rsize,
+	},
+	[OP_LISTXATTRS] = {
+		.op_func = nfsd4_listxattrs,
+		.op_name = "OP_LISTXATTRS",
+		.op_rsize_bop = nfsd4_listxattrs_rsize,
+	},
+	[OP_REMOVEXATTR] = {
+		.op_func = nfsd4_removexattr,
+		.op_flags = OP_MODIFIES_SOMETHING | OP_CACHEME,
+		.op_name = "OP_REMOVEXATTR",
+		.op_rsize_bop = nfsd4_removexattr_rsize,
+	},
 };
 
 /**
diff --git a/fs/nfsd/nfs4xdr.c b/fs/nfsd/nfs4xdr.c
index f6322add2992..4c4508f1bb8e 100644
--- a/fs/nfsd/nfs4xdr.c
+++ b/fs/nfsd/nfs4xdr.c
@@ -41,6 +41,8 @@
 #include <linux/pagemap.h>
 #include <linux/sunrpc/svcauth_gss.h>
 #include <linux/sunrpc/addr.h>
+#include <linux/xattr.h>
+#include <uapi/linux/xattr.h>
 
 #include "idmap.h"
 #include "acl.h"
@@ -1877,6 +1879,224 @@ nfsd4_decode_seek(struct nfsd4_compoundargs *argp, struct nfsd4_seek *seek)
 	DECODE_TAIL;
 }
 
+/*
+ * XDR data that is more than PAGE_SIZE in size is normally part of a
+ * read or write. However, the size of extended attributes is limited
+ * by the maximum request size, and then further limited by the underlying
+ * filesystem limits. This can exceed PAGE_SIZE (currently, XATTR_SIZE_MAX
+ * is 64k). Since there is no kvec- or page-based interface to xattrs,
+ * and we're not dealing with contiguous pages, we need to do some copying.
+ */
+
+/*
+ * Decode data into buffer. Uses head and pages constructed by
+ * svcxdr_construct_vector.
+ */
+static __be32
+nfsd4_vbuf_from_vector(struct nfsd4_compoundargs *argp, struct kvec *head,
+		       struct page **pages, char **bufp, u32 buflen)
+{
+	char *tmp, *dp;
+	u32 len;
+
+	if (buflen <= head->iov_len) {
+		/*
+		 * We're in luck, the head has enough space. Just return
+		 * the head, no need for copying.
+		 */
+		*bufp = head->iov_base;
+		return 0;
+	}
+
+	tmp = svcxdr_tmpalloc(argp, buflen);
+	if (tmp == NULL)
+		return nfserr_jukebox;
+
+	dp = tmp;
+	memcpy(dp, head->iov_base, head->iov_len);
+	buflen -= head->iov_len;
+	dp += head->iov_len;
+
+	while (buflen > 0) {
+		len = min_t(u32, buflen, PAGE_SIZE);
+		memcpy(dp, page_address(*pages), len);
+
+		buflen -= len;
+		dp += len;
+		pages++;
+	}
+
+	*bufp = tmp;
+	return 0;
+}
+
+/*
+ * Get a user extended attribute name from the XDR buffer.
+ * It will not have the "user." prefix, so prepend it.
+ * Lastly, check for nul characters in the name.
+ */
+static __be32
+nfsd4_decode_xattr_name(struct nfsd4_compoundargs *argp, char **namep)
+{
+	DECODE_HEAD;
+	char *name, *sp, *dp;
+	u32 namelen, cnt;
+
+	READ_BUF(4);
+	namelen = be32_to_cpup(p++);
+
+	if (namelen > (XATTR_NAME_MAX - XATTR_USER_PREFIX_LEN))
+		return nfserr_nametoolong;
+
+	if (namelen == 0)
+		goto xdr_error;
+
+	READ_BUF(namelen);
+
+	name = svcxdr_tmpalloc(argp, namelen + XATTR_USER_PREFIX_LEN + 1);
+	if (!name)
+		return nfserr_jukebox;
+
+	memcpy(name, XATTR_USER_PREFIX, XATTR_USER_PREFIX_LEN);
+
+	/*
+	 * Copy the extended attribute name over while checking for 0
+	 * characters.
+	 */
+	sp = (char *)p;
+	dp = name + XATTR_USER_PREFIX_LEN;
+	cnt = namelen;
+
+	while (cnt-- > 0) {
+		if (*sp == '\0')
+			goto xdr_error;
+		*dp++ = *sp++;
+	}
+	*dp = '\0';
+
+	*namep = name;
+
+	DECODE_TAIL;
+}
+
+/*
+ * A GETXATTR op request comes without a length specifier. We just set the
+ * maximum length for the reply based on XATTR_SIZE_MAX and the maximum
+ * channel reply size, allocate a buffer of that length and pass it to
+ * vfs_getxattr.
+ */
+static __be32
+nfsd4_decode_getxattr(struct nfsd4_compoundargs *argp,
+		      struct nfsd4_getxattr *getxattr)
+{
+	__be32 status;
+	u32 maxcount;
+
+	status = nfsd4_decode_xattr_name(argp, &getxattr->getxa_name);
+	if (status)
+		return status;
+
+	maxcount = svc_max_payload(argp->rqstp);
+	maxcount = min_t(u32, XATTR_SIZE_MAX, maxcount);
+
+	getxattr->getxa_buf = svcxdr_tmpalloc(argp, maxcount);
+	if (!getxattr->getxa_buf)
+		status = nfserr_jukebox;
+	getxattr->getxa_len = maxcount;
+
+	return status;
+}
+
+static __be32
+nfsd4_decode_setxattr(struct nfsd4_compoundargs *argp,
+		      struct nfsd4_setxattr *setxattr)
+{
+	DECODE_HEAD;
+	u32 flags, maxcount, size;
+	struct kvec head;
+	struct page **pagelist;
+
+	READ_BUF(4);
+	flags = be32_to_cpup(p++);
+
+	if (flags > SETXATTR4_REPLACE)
+		return nfserr_inval;
+	setxattr->setxa_flags = flags;
+
+	status = nfsd4_decode_xattr_name(argp, &setxattr->setxa_name);
+	if (status)
+		return status;
+
+	maxcount = svc_max_payload(argp->rqstp);
+	maxcount = min_t(u32, XATTR_SIZE_MAX, maxcount);
+
+	READ_BUF(4);
+	size = be32_to_cpup(p++);
+	if (size > maxcount)
+		return nfserr_xattr2big;
+
+	setxattr->setxa_len = size;
+	if (size > 0) {
+		status = svcxdr_construct_vector(argp, &head, &pagelist, size);
+		if (status)
+			return status;
+
+		status = nfsd4_vbuf_from_vector(argp, &head, pagelist,
+		    &setxattr->setxa_buf, size);
+	}
+
+	DECODE_TAIL;
+}
+
+static __be32
+nfsd4_decode_listxattrs(struct nfsd4_compoundargs *argp,
+			struct nfsd4_listxattrs *listxattrs)
+{
+	DECODE_HEAD;
+	u32 maxcount;
+
+	READ_BUF(12);
+	p = xdr_decode_hyper(p, &listxattrs->lsxa_cookie);
+
+	/*
+	 * If the cookie  is too large to have even one user.x attribute
+	 * plus trailing '\0' left in a maximum size buffer, it's invalid.
+	 */
+	if (listxattrs->lsxa_cookie >=
+	    (XATTR_LIST_MAX / (XATTR_USER_PREFIX_LEN + 2)))
+		return nfserr_badcookie;
+
+	maxcount = be32_to_cpup(p++);
+	if (maxcount < 8)
+		/* Always need at least 2 words (length and one character) */
+		return nfserr_inval;
+
+	maxcount = min(maxcount, svc_max_payload(argp->rqstp));
+	listxattrs->lsxa_maxcount = maxcount;
+
+	/*
+	 * Unfortunately, there is no interface to only list xattrs for
+	 * one prefix. So there is no good way to convert maxcount to
+	 * a maximum value to pass to vfs_listxattr, as we don't know
+	 * how many of the returned attributes will be user attributes.
+	 *
+	 * So, always ask vfs_listxattr for the maximum size, and encode
+	 * as many as possible.
+	 */
+	listxattrs->lsxa_buf = svcxdr_tmpalloc(argp, XATTR_LIST_MAX);
+	if (!listxattrs->lsxa_buf)
+		status = nfserr_jukebox;
+
+	DECODE_TAIL;
+}
+
+static __be32
+nfsd4_decode_removexattr(struct nfsd4_compoundargs *argp,
+			 struct nfsd4_removexattr *removexattr)
+{
+	return nfsd4_decode_xattr_name(argp, &removexattr->rmxa_name);
+}
+
 static __be32
 nfsd4_decode_noop(struct nfsd4_compoundargs *argp, void *p)
 {
@@ -1973,6 +2193,11 @@ static const nfsd4_dec nfsd4_dec_ops[] = {
 	[OP_SEEK]		= (nfsd4_dec)nfsd4_decode_seek,
 	[OP_WRITE_SAME]		= (nfsd4_dec)nfsd4_decode_notsupp,
 	[OP_CLONE]		= (nfsd4_dec)nfsd4_decode_clone,
+	/* RFC 8276 extended atributes operations */
+	[OP_GETXATTR]		= (nfsd4_dec)nfsd4_decode_getxattr,
+	[OP_SETXATTR]		= (nfsd4_dec)nfsd4_decode_setxattr,
+	[OP_LISTXATTRS]		= (nfsd4_dec)nfsd4_decode_listxattrs,
+	[OP_REMOVEXATTR]	= (nfsd4_dec)nfsd4_decode_removexattr,
 };
 
 static inline bool
@@ -4456,6 +4681,225 @@ nfsd4_encode_noop(struct nfsd4_compoundres *resp, __be32 nfserr, void *p)
 	return nfserr;
 }
 
+/*
+ * Encode kmalloc-ed buffer in to XDR stream.
+ */
+static int
+nfsd4_vbuf_to_stream(struct xdr_stream *xdr, char *buf, u32 buflen)
+{
+	u32 cplen;
+	__be32 *p;
+
+	cplen = min_t(unsigned long, buflen,
+		      ((void *)xdr->end - (void *)xdr->p));
+	p = xdr_reserve_space(xdr, cplen);
+	if (!p)
+		return nfserr_resource;
+
+	memcpy(p, buf, cplen);
+	buf += cplen;
+	buflen -= cplen;
+
+	while (buflen) {
+		cplen = min_t(u32, buflen, PAGE_SIZE);
+		p = xdr_reserve_space(xdr, cplen);
+		if (!p)
+			return nfserr_resource;
+
+		memcpy(p, buf, cplen);
+
+		if (cplen < PAGE_SIZE) {
+			/*
+			 * We're done, with a length that wasn't page
+			 * aligned, so possibly not word aligned. Pad
+			 * any trailing bytes with 0.
+			 */
+			xdr_encode_opaque_fixed(p, NULL, cplen);
+			break;
+		}
+
+		buflen -= PAGE_SIZE;
+		buf += PAGE_SIZE;
+	}
+
+	return 0;
+}
+
+static __be32
+nfsd4_encode_getxattr(struct nfsd4_compoundres *resp, __be32 nfserr,
+		      struct nfsd4_getxattr *getxattr)
+{
+	struct xdr_stream *xdr = &resp->xdr;
+	__be32 *p;
+
+	p = xdr_reserve_space(xdr, 4);
+	if (!p)
+		return nfserr_resource;
+
+	*p = cpu_to_be32(getxattr->getxa_len);
+
+	if (getxattr->getxa_len == 0)
+		return 0;
+
+	return nfsd4_vbuf_to_stream(xdr, getxattr->getxa_buf,
+				    getxattr->getxa_len);
+}
+
+static __be32
+nfsd4_encode_setxattr(struct nfsd4_compoundres *resp, __be32 nfserr,
+		      struct nfsd4_setxattr *setxattr)
+{
+	struct xdr_stream *xdr = &resp->xdr;
+	__be32 *p;
+
+	p = xdr_reserve_space(xdr, 20);
+	if (!p)
+		return nfserr_resource;
+
+	encode_cinfo(p, &setxattr->setxa_cinfo);
+
+	return 0;
+}
+
+/*
+ * See if there are cookie values that can be rejected outright.
+ */
+static __be32
+nfsd4_listxattr_validate_cookie(struct nfsd4_listxattrs *listxattrs,
+				u32 *offsetp)
+{
+	u64 cookie = listxattrs->lsxa_cookie;
+
+	/*
+	 * If the cookie is larger than the maximum number we can fit
+	 * in either the buffer we just got back from vfs_listxattr, or,
+	 * XDR-encoded, in the return buffer, it's invalid.
+	 */
+	if (cookie > (listxattrs->lsxa_len) / (XATTR_USER_PREFIX_LEN + 2))
+		return nfserr_badcookie;
+
+	if (cookie > (listxattrs->lsxa_maxcount /
+		      (XDR_QUADLEN(XATTR_USER_PREFIX_LEN + 2) + 4)))
+		return nfserr_badcookie;
+
+	*offsetp = (u32)cookie;
+	return 0;
+}
+
+static __be32
+nfsd4_encode_listxattrs(struct nfsd4_compoundres *resp, __be32 nfserr,
+			struct nfsd4_listxattrs *listxattrs)
+{
+	struct xdr_stream *xdr = &resp->xdr;
+	u32 cookie_offset, count_offset, eof;
+	u32 left, xdrleft, slen, count;
+	u32 xdrlen, offset;
+	u64 cookie;
+	char *sp;
+	__be32 status;
+	__be32 *p;
+	u32 nuser;
+
+	eof = 1;
+
+	status = nfsd4_listxattr_validate_cookie(listxattrs, &offset);
+	if (status)
+		return status;
+
+	/*
+	 * Reserve space for the cookie and the name array count. Record
+	 * the offsets to save them later.
+	 */
+	cookie_offset = xdr->buf->len;
+	count_offset = cookie_offset + 8;
+	p = xdr_reserve_space(xdr, 12);
+	if (!p)
+		return nfserr_resource;
+
+	count = 0;
+	left = listxattrs->lsxa_len;
+	sp = listxattrs->lsxa_buf;
+	nuser = 0;
+
+	xdrleft = listxattrs->lsxa_maxcount;
+
+	while (left > 0 && xdrleft > 0) {
+		slen = strlen(sp);
+
+		/*
+		 * Check if this a user. attribute, skip it if not.
+		 */
+		if (strncmp(sp, XATTR_USER_PREFIX, XATTR_USER_PREFIX_LEN))
+			goto contloop;
+
+		slen -= XATTR_USER_PREFIX_LEN;
+		xdrlen = 4 + ((slen + 3) & ~3);
+		if (xdrlen > xdrleft) {
+			if (count == 0) {
+				/*
+				 * Can't even fit the first attribute name.
+				 */
+				return nfserr_toosmall;
+			}
+			eof = 0;
+			goto wreof;
+		}
+
+		left -= XATTR_USER_PREFIX_LEN;
+		sp += XATTR_USER_PREFIX_LEN;
+		if (nuser++ < offset)
+			goto contloop;
+
+
+		p = xdr_reserve_space(xdr, xdrlen);
+		if (!p)
+			return nfserr_resource;
+
+		p = xdr_encode_opaque(p, sp, slen);
+
+		xdrleft -= xdrlen;
+		count++;
+contloop:
+		sp += slen + 1;
+		left -= slen + 1;
+	}
+
+	/*
+	 * If there were user attributes to copy, but we didn't copy
+	 * any, the offset was too large (e.g. the cookie was invalid).
+	 */
+	if (nuser > 0 && count == 0)
+		return nfserr_badcookie;
+
+wreof:
+	p = xdr_reserve_space(xdr, 4);
+	if (!p)
+		return nfserr_resource;
+	*p = cpu_to_be32(eof);
+
+	cookie = offset + count;
+
+	write_bytes_to_xdr_buf(xdr->buf, cookie_offset, &cookie, 8);
+	count = htonl(count);
+	write_bytes_to_xdr_buf(xdr->buf, count_offset, &count, 4);
+	return 0;
+}
+
+static __be32
+nfsd4_encode_removexattr(struct nfsd4_compoundres *resp, __be32 nfserr,
+			 struct nfsd4_removexattr *removexattr)
+{
+	struct xdr_stream *xdr = &resp->xdr;
+	__be32 *p;
+
+	p = xdr_reserve_space(xdr, 20);
+	if (!p)
+		return nfserr_resource;
+
+	p = encode_cinfo(p, &removexattr->rmxa_cinfo);
+	return 0;
+}
+
 typedef __be32(* nfsd4_enc)(struct nfsd4_compoundres *, __be32, void *);
 
 /*
@@ -4545,6 +4989,12 @@ static const nfsd4_enc nfsd4_enc_ops[] = {
 	[OP_SEEK]		= (nfsd4_enc)nfsd4_encode_seek,
 	[OP_WRITE_SAME]		= (nfsd4_enc)nfsd4_encode_noop,
 	[OP_CLONE]		= (nfsd4_enc)nfsd4_encode_noop,
+
+	/* RFC 8276 extended atributes operations */
+	[OP_GETXATTR]		= (nfsd4_enc)nfsd4_encode_getxattr,
+	[OP_SETXATTR]		= (nfsd4_enc)nfsd4_encode_setxattr,
+	[OP_LISTXATTRS]		= (nfsd4_enc)nfsd4_encode_listxattrs,
+	[OP_REMOVEXATTR]	= (nfsd4_enc)nfsd4_encode_removexattr,
 };
 
 /*
diff --git a/include/linux/nfs4.h b/include/linux/nfs4.h
index 350aeda0c48c..cbe50e452f19 100644
--- a/include/linux/nfs4.h
+++ b/include/linux/nfs4.h
@@ -165,7 +165,7 @@ Needs to be updated if more operations are defined in future.*/
 #define FIRST_NFS4_OP	OP_ACCESS
 #define LAST_NFS40_OP	OP_RELEASE_LOCKOWNER
 #define LAST_NFS41_OP	OP_RECLAIM_COMPLETE
-#define LAST_NFS42_OP	OP_CLONE
+#define LAST_NFS42_OP	OP_REMOVEXATTR
 #define LAST_NFS4_OP	LAST_NFS42_OP
 
 enum nfsstat4 {
-- 
2.17.2

vfs_listxattr and the NFS server: namespaces

[Frank van der Linden]

On Fri, Mar 27, 2020 at 11:27:16PM +0000, Frank van der Linden wrote:
> Implement the main entry points for the *XATTR operations.
> 
> Add functions to calculate the reply size for the user extended attribute
> operations, and implement the XDR encode / decode logic for these
> operations.
> 
> Add the user extended attributes operations to nfsd4_ops.

The reason I Cc-ed this one to linux-fsdevel, is the following piece of
code:
> +	/*
> +	 * Unfortunately, there is no interface to only list xattrs for
> +	 * one prefix. So there is no good way to convert maxcount to
> +	 * a maximum value to pass to vfs_listxattr, as we don't know
> +	 * how many of the returned attributes will be user attributes.
> +	 *
> +	 * So, always ask vfs_listxattr for the maximum size, and encode
> +	 * as many as possible.
> +	 */
> +	listxattrs->lsxa_buf = svcxdr_tmpalloc(argp, XATTR_LIST_MAX);
> +	if (!listxattrs->lsxa_buf)
> +		status = nfserr_jukebox;
> +
> +	DECODE_TAIL;
> +}

Naturally, it is a waste to always (temporarily) allocate XATTR_LIST_MAX
(currently 64k) bytes for every listxattr request, when in reality you
probably won't need anywhere near that.

The reason for doing that is that, while the NFS request comes with a 
maximum size, you can't translate that in to a maximum size to pass
to vfs_listxattr, since you are specifying the max size for "user."
extended attributes only, since that is the namespace that the NFS
implementation is restricted to. But vfs_listxattr always retrieves
all extended attributes.

The question is then: how to avoid doing that? It might be a good
idea to be able to specify a namespace to vfs_listxattr. This isn't
terribly hard to implement, and can be phased in. E.g:

* Add a "const char *namespace" argument to the listxattr inode op
* For the normal use case it will be NULL, meaning all xattrs. This
  includes use in the system call path (a new system call that
  includes a namespace argument could be added).
* A filesystem that does not support only returning xattrs for one
  namespace will return a specified error if @namespace != NULL, after
  which the fs-indepdent code in fs/xattr.c falls back to pre-allocating
  an XATTR_LIST_MAX buffer and then filtering out all attributes that
  are not in the specified namespace.
* Since the initial use is from the NFS code, converting XFS and ext4
  to support the @namespace argument should catch most use cases of
  the @namespace argument, the rest can be converted over time.

Does this sound reasonable? Or is it overkill?

Another way to do it is to add a vfs_listxattr_alloc variant, and
have the NFS server code filter out the user. xattrs as now. There is
one problem with that: it's possible for the size of listxattr result
to increase between the first null (length probe) call and the second call,
in which case the NFS server would return a spurious error. A small chance,
but it's there nonetheless. And it would be hard to distinguish from
other errors. So I think I still would prefer using an extra namespace
argument.

Alternatively, we could decide that the temporarily allocation of 64k
is no big deal.

Thoughts?

- Frank

Re: vfs_listxattr and the NFS server: namespaces

[Chuck Lever]

> On Mar 29, 2020, at 4:25 PM, Frank van der Linden <fllinden@amazon.com> wrote:
> 
> On Fri, Mar 27, 2020 at 11:27:16PM +0000, Frank van der Linden wrote:
>> Implement the main entry points for the *XATTR operations.
>> 
>> Add functions to calculate the reply size for the user extended attribute
>> operations, and implement the XDR encode / decode logic for these
>> operations.
>> 
>> Add the user extended attributes operations to nfsd4_ops.
> 
> The reason I Cc-ed this one to linux-fsdevel, is the following piece of
> code:
>> +	/*
>> +	 * Unfortunately, there is no interface to only list xattrs for
>> +	 * one prefix. So there is no good way to convert maxcount to
>> +	 * a maximum value to pass to vfs_listxattr, as we don't know
>> +	 * how many of the returned attributes will be user attributes.
>> +	 *
>> +	 * So, always ask vfs_listxattr for the maximum size, and encode
>> +	 * as many as possible.
>> +	 */
>> +	listxattrs->lsxa_buf = svcxdr_tmpalloc(argp, XATTR_LIST_MAX);
>> +	if (!listxattrs->lsxa_buf)
>> +		status = nfserr_jukebox;
>> +
>> +	DECODE_TAIL;
>> +}
> 
> Naturally, it is a waste to always (temporarily) allocate XATTR_LIST_MAX
> (currently 64k) bytes for every listxattr request, when in reality you
> probably won't need anywhere near that.
> 
> The reason for doing that is that, while the NFS request comes with a 
> maximum size, you can't translate that in to a maximum size to pass
> to vfs_listxattr, since you are specifying the max size for "user."
> extended attributes only, since that is the namespace that the NFS
> implementation is restricted to. But vfs_listxattr always retrieves
> all extended attributes.
> 
> The question is then: how to avoid doing that? It might be a good
> idea to be able to specify a namespace to vfs_listxattr. This isn't
> terribly hard to implement, and can be phased in. E.g:
> 
> * Add a "const char *namespace" argument to the listxattr inode op
> * For the normal use case it will be NULL, meaning all xattrs. This
>  includes use in the system call path (a new system call that
>  includes a namespace argument could be added).
> * A filesystem that does not support only returning xattrs for one
>  namespace will return a specified error if @namespace != NULL, after
>  which the fs-indepdent code in fs/xattr.c falls back to pre-allocating
>  an XATTR_LIST_MAX buffer and then filtering out all attributes that
>  are not in the specified namespace.
> * Since the initial use is from the NFS code, converting XFS and ext4
>  to support the @namespace argument should catch most use cases of
>  the @namespace argument, the rest can be converted over time.
> 
> Does this sound reasonable? Or is it overkill?
> 
> Another way to do it is to add a vfs_listxattr_alloc variant, and
> have the NFS server code filter out the user. xattrs as now. There is
> one problem with that: it's possible for the size of listxattr result
> to increase between the first null (length probe) call and the second call,
> in which case the NFS server would return a spurious error. A small chance,
> but it's there nonetheless. And it would be hard to distinguish from
> other errors. So I think I still would prefer using an extra namespace
> argument.
> 
> Alternatively, we could decide that the temporarily allocation of 64k
> is no big deal.

An order-4 allocation is not something that can be relied upon. I would
rather find a way to avoid the need for it if at all possible.


--
Chuck Lever

Re: vfs_listxattr and the NFS server: namespaces

[Frank van der Linden]

On Sun, Mar 29, 2020 at 05:54:38PM -0400, Chuck Lever wrote:
> > Alternatively, we could decide that the temporarily allocation of 64k
> > is no big deal.
> 
> An order-4 allocation is not something that can be relied upon. I would
> rather find a way to avoid the need for it if at all possible.

It's not strictly an order-4 allocation anymore, as I changed
svcxdr_tmpalloc to use kvmalloc.

But sure, it's best to avoid it, it's a waste.

- Frank

Re: [PATCH v2 01/11] xattr: break delegations in {set,remove}xattr and add _locked versions

[Sasha Levin]

Hi

[This is an automated email]

This commit has been processed because it contains a -stable tag.
The stable tag indicates that it's relevant for the following trees: all

The bot has tested the following trees: v5.5.13, v5.4.28, v4.19.113, v4.14.174, v4.9.217, v4.4.217.

v5.5.13: Build OK!
v5.4.28: Build OK!
v4.19.113: Build OK!
v4.14.174: Build OK!
v4.9.217: Build OK!
v4.4.217: Failed to apply! Possible dependencies:
    5d6c31910bc0 ("xattr: Add __vfs_{get,set,remove}xattr helpers")
    6b2553918d8b ("replace ->follow_link() with new method that could stay in RCU mode")
    aa80deab33a8 ("namei: page_getlink() and page_follow_link_light() are the same thing")
    cd3417c8fc95 ("kill free_page_put_link()")
    ce23e6401334 ("->getxattr(): pass dentry and inode as separate arguments")
    fceef393a538 ("switch ->get_link() to delayed_call, kill ->put_link()")


NOTE: The patch will not be queued to stable trees until it is upstream.

How should we proceed with this patch?

-- 
Thanks
Sasha