Merge branch 'xattr-devel'

14 files changed, 2163 insertions, 41 deletions

diff --git a/fs/nfs/Makefile b/fs/nfs/Makefile
index 2433c3e03cfa..22d11fdc6deb 100644
--- a/fs/nfs/Makefile
+++ b/fs/nfs/Makefile
@@ -30,7 +30,7 @@ nfsv4-y := nfs4proc.o nfs4xdr.o nfs4state.o nfs4renewd.o nfs4super.o nfs4file.o
 nfsv4-$(CONFIG_NFS_USE_LEGACY_DNS) += cache_lib.o
 nfsv4-$(CONFIG_SYSCTL)	+= nfs4sysctl.o
 nfsv4-$(CONFIG_NFS_V4_1)	+= pnfs.o pnfs_dev.o pnfs_nfs.o
-nfsv4-$(CONFIG_NFS_V4_2)	+= nfs42proc.o
+nfsv4-$(CONFIG_NFS_V4_2)	+= nfs42proc.o nfs42xattr.o
 
 obj-$(CONFIG_PNFS_FILE_LAYOUT) += filelayout/
 obj-$(CONFIG_PNFS_BLOCK) += blocklayout/
diff --git a/fs/nfs/client.c b/fs/nfs/client.c
index f1ff3076e4a4..4b8cc93913f7 100644
--- a/fs/nfs/client.c
+++ b/fs/nfs/client.c
@@ -50,6 +50,7 @@
 #include "nfs.h"
 #include "netns.h"
 #include "sysfs.h"
+#include "nfs42.h"
 
 #define NFSDBG_FACILITY		NFSDBG_CLIENT
 
@@ -749,7 +750,7 @@ error:
 static void nfs_server_set_fsinfo(struct nfs_server *server,
 				  struct nfs_fsinfo *fsinfo)
 {
-	unsigned long max_rpc_payload;
+	unsigned long max_rpc_payload, raw_max_rpc_payload;
 
 	/* Work out a lot of parameters */
 	if (server->rsize == 0)
@@ -762,7 +763,9 @@ static void nfs_server_set_fsinfo(struct nfs_server *server,
 	if (fsinfo->wtmax >= 512 && server->wsize > fsinfo->wtmax)
 		server->wsize = nfs_block_size(fsinfo->wtmax, NULL);
 
-	max_rpc_payload = nfs_block_size(rpc_max_payload(server->client), NULL);
+	raw_max_rpc_payload = rpc_max_payload(server->client);
+	max_rpc_payload = nfs_block_size(raw_max_rpc_payload, NULL);
+
 	if (server->rsize > max_rpc_payload)
 		server->rsize = max_rpc_payload;
 	if (server->rsize > NFS_MAX_FILE_IO_SIZE)
@@ -795,6 +798,21 @@ static void nfs_server_set_fsinfo(struct nfs_server *server,
 	server->clone_blksize = fsinfo->clone_blksize;
 	/* We're airborne Set socket buffersize */
 	rpc_setbufsize(server->client, server->wsize + 100, server->rsize + 100);
+
+#ifdef CONFIG_NFS_V4_2
+	/*
+	 * Defaults until limited by the session parameters.
+	 */
+	server->gxasize = min_t(unsigned int, raw_max_rpc_payload,
+				XATTR_SIZE_MAX);
+	server->sxasize = min_t(unsigned int, raw_max_rpc_payload,
+				XATTR_SIZE_MAX);
+	server->lxasize = min_t(unsigned int, raw_max_rpc_payload,
+				nfs42_listxattr_xdrsize(XATTR_LIST_MAX));
+
+	if (fsinfo->xattr_support)
+		server->caps |= NFS_CAP_XATTR;
+#endif
 }
 
 /*
diff --git a/fs/nfs/dir.c b/fs/nfs/dir.c
index 5a331da5f55a..a12f42e7d8c7 100644
--- a/fs/nfs/dir.c
+++ b/fs/nfs/dir.c
@@ -2460,7 +2460,7 @@ static struct nfs_access_entry *nfs_access_search_rbtree(struct inode *inode, co
 	return NULL;
 }
 
-static int nfs_access_get_cached(struct inode *inode, const struct cred *cred, struct nfs_access_entry *res, bool may_block)
+static int nfs_access_get_cached_locked(struct inode *inode, const struct cred *cred, struct nfs_access_entry *res, bool may_block)
 {
 	struct nfs_inode *nfsi = NFS_I(inode);
 	struct nfs_access_entry *cache;
@@ -2533,6 +2533,20 @@ out:
 	return err;
 }
 
+int nfs_access_get_cached(struct inode *inode, const struct cred *cred, struct
+nfs_access_entry *res, bool may_block)
+{
+	int status;
+
+	status = nfs_access_get_cached_rcu(inode, cred, res);
+	if (status != 0)
+		status = nfs_access_get_cached_locked(inode, cred, res,
+		    may_block);
+
+	return status;
+}
+EXPORT_SYMBOL_GPL(nfs_access_get_cached);
+
 static void nfs_access_add_rbtree(struct inode *inode, struct nfs_access_entry *set)
 {
 	struct nfs_inode *nfsi = NFS_I(inode);
@@ -2647,9 +2661,7 @@ static int nfs_do_access(struct inode *inode, const struct cred *cred, int mask)
 
 	trace_nfs_access_enter(inode);
 
-	status = nfs_access_get_cached_rcu(inode, cred, &cache);
-	if (status != 0)
-		status = nfs_access_get_cached(inode, cred, &cache, may_block);
+	status = nfs_access_get_cached(inode, cred, &cache, may_block);
 	if (status == 0)
 		goto out_cached;
 
@@ -2661,6 +2673,10 @@ static int nfs_do_access(struct inode *inode, const struct cred *cred, int mask)
 	 * Determine which access bits we want to ask for...
 	 */
 	cache.mask = NFS_ACCESS_READ | NFS_ACCESS_MODIFY | NFS_ACCESS_EXTEND;
+	if (nfs_server_capable(inode, NFS_CAP_XATTR)) {
+		cache.mask |= NFS_ACCESS_XAREAD | NFS_ACCESS_XAWRITE |
+		    NFS_ACCESS_XALIST;
+	}
 	if (S_ISDIR(inode->i_mode))
 		cache.mask |= NFS_ACCESS_DELETE | NFS_ACCESS_LOOKUP;
 	else
diff --git a/fs/nfs/inode.c b/fs/nfs/inode.c
index 90f77fd0d1ca..aa6493905bbe 100644
--- a/fs/nfs/inode.c
+++ b/fs/nfs/inode.c
@@ -193,6 +193,7 @@ bool nfs_check_cache_invalid(struct inode *inode, unsigned long flags)
 
 	return nfs_check_cache_invalid_not_delegated(inode, flags);
 }
+EXPORT_SYMBOL_GPL(nfs_check_cache_invalid);
 
 static void nfs_set_cache_invalid(struct inode *inode, unsigned long flags)
 {
@@ -204,7 +205,8 @@ static void nfs_set_cache_invalid(struct inode *inode, unsigned long flags)
 			flags &= ~NFS_INO_INVALID_OTHER;
 		flags &= ~(NFS_INO_INVALID_CHANGE
 				| NFS_INO_INVALID_SIZE
-				| NFS_INO_REVAL_PAGECACHE);
+				| NFS_INO_REVAL_PAGECACHE
+				| NFS_INO_INVALID_XATTR);
 	}
 
 	if (inode->i_mapping->nrpages == 0)
@@ -233,11 +235,13 @@ static void nfs_zap_caches_locked(struct inode *inode)
 					| NFS_INO_INVALID_DATA
 					| NFS_INO_INVALID_ACCESS
 					| NFS_INO_INVALID_ACL
+					| NFS_INO_INVALID_XATTR
 					| NFS_INO_REVAL_PAGECACHE);
 	} else
 		nfs_set_cache_invalid(inode, NFS_INO_INVALID_ATTR
 					| NFS_INO_INVALID_ACCESS
 					| NFS_INO_INVALID_ACL
+					| NFS_INO_INVALID_XATTR
 					| NFS_INO_REVAL_PAGECACHE);
 	nfs_zap_label_cache_locked(nfsi);
 }
@@ -542,6 +546,8 @@ nfs_fhget(struct super_block *sb, struct nfs_fh *fh, struct nfs_fattr *fattr, st
 			inode->i_gid = fattr->gid;
 		else if (nfs_server_capable(inode, NFS_CAP_OWNER_GROUP))
 			nfs_set_cache_invalid(inode, NFS_INO_INVALID_OTHER);
+		if (nfs_server_capable(inode, NFS_CAP_XATTR))
+			nfs_set_cache_invalid(inode, NFS_INO_INVALID_XATTR);
 		if (fattr->valid & NFS_ATTR_FATTR_BLOCKS_USED)
 			inode->i_blocks = fattr->du.nfs2.blocks;
 		if (fattr->valid & NFS_ATTR_FATTR_SPACE_USED) {
@@ -1377,6 +1383,8 @@ static void nfs_wcc_update_inode(struct inode *inode, struct nfs_fattr *fattr)
 		inode_set_iversion_raw(inode, fattr->change_attr);
 		if (S_ISDIR(inode->i_mode))
 			nfs_set_cache_invalid(inode, NFS_INO_INVALID_DATA);
+		else if (nfs_server_capable(inode, NFS_CAP_XATTR))
+			nfs_set_cache_invalid(inode, NFS_INO_INVALID_XATTR);
 	}
 	/* If we have atomic WCC data, we may update some attributes */
 	ts = inode->i_ctime;
@@ -1894,7 +1902,8 @@ static int nfs_update_inode(struct inode *inode, struct nfs_fattr *fattr)
 			if (!(have_writers || have_delegation)) {
 				invalid |= NFS_INO_INVALID_DATA
 					| NFS_INO_INVALID_ACCESS
-					| NFS_INO_INVALID_ACL;
+					| NFS_INO_INVALID_ACL
+					| NFS_INO_INVALID_XATTR;
 				/* Force revalidate of all attributes */
 				save_cache_validity |= NFS_INO_INVALID_CTIME
 					| NFS_INO_INVALID_MTIME
@@ -2097,6 +2106,9 @@ struct inode *nfs_alloc_inode(struct super_block *sb)
 #if IS_ENABLED(CONFIG_NFS_V4)
 	nfsi->nfs4_acl = NULL;
 #endif /* CONFIG_NFS_V4 */
+#ifdef CONFIG_NFS_V4_2
+	nfsi->xattr_cache = NULL;
+#endif
 	return &nfsi->vfs_inode;
 }
 EXPORT_SYMBOL_GPL(nfs_alloc_inode);
diff --git a/fs/nfs/nfs42.h b/fs/nfs/nfs42.h
index c891af949886..0fe5aacbcfdf 100644
--- a/fs/nfs/nfs42.h
+++ b/fs/nfs/nfs42.h
@@ -6,6 +6,8 @@
 #ifndef __LINUX_FS_NFS_NFS4_2_H
 #define __LINUX_FS_NFS_NFS4_2_H
 
+#include <linux/xattr.h>
+
 /*
  * FIXME:  four LAYOUTSTATS calls per compound at most! Do we need to support
  * more? Need to consider not to pre-alloc too much for a compound.
@@ -36,5 +38,27 @@ static inline bool nfs42_files_from_same_server(struct file *in,
 	return nfs4_check_serverowner_major_id(c_in->cl_serverowner,
 					       c_out->cl_serverowner);
 }
+
+ssize_t nfs42_proc_getxattr(struct inode *inode, const char *name,
+			    void *buf, size_t buflen);
+int nfs42_proc_setxattr(struct inode *inode, const char *name,
+			const void *buf, size_t buflen, int flags);
+ssize_t nfs42_proc_listxattrs(struct inode *inode, void *buf,
+			       size_t buflen, u64 *cookiep, bool *eofp);
+int nfs42_proc_removexattr(struct inode *inode, const char *name);
+
+/*
+ * Maximum XDR buffer size needed for a listxattr buffer of buflen size.
+ *
+ * The upper boundary is a buffer with all 1-byte sized attribute names.
+ * They would be 7 bytes long in the eventual buffer ("user.x\0"), and
+ * 8 bytes long XDR-encoded.
+ *
+ * Include the trailing eof word as well.
+ */
+static inline u32 nfs42_listxattr_xdrsize(u32 buflen)
+{
+	return ((buflen / (XATTR_USER_PREFIX_LEN + 2)) * 8) + 4;
+}
 #endif /* CONFIG_NFS_V4_2 */
 #endif /* __LINUX_FS_NFS_NFS4_2_H */
diff --git a/fs/nfs/nfs42proc.c b/fs/nfs/nfs42proc.c
index e2ae54b35dfe..e200522469af 100644
--- a/fs/nfs/nfs42proc.c
+++ b/fs/nfs/nfs42proc.c
@@ -1088,3 +1088,251 @@ out_put_src_lock:
 	nfs_put_lock_context(src_lock);
 	return err;
 }
+
+#define NFS4XATTR_MAXPAGES DIV_ROUND_UP(XATTR_SIZE_MAX, PAGE_SIZE)
+
+static int _nfs42_proc_removexattr(struct inode *inode, const char *name)
+{
+	struct nfs_server *server = NFS_SERVER(inode);
+	struct nfs42_removexattrargs args = {
+		.fh = NFS_FH(inode),
+		.xattr_name = name,
+	};
+	struct nfs42_removexattrres res;
+	struct rpc_message msg = {
+		.rpc_proc = &nfs4_procedures[NFSPROC4_CLNT_REMOVEXATTR],
+		.rpc_argp = &args,
+		.rpc_resp = &res,
+	};
+	int ret;
+	unsigned long timestamp = jiffies;
+
+	ret = nfs4_call_sync(server->client, server, &msg, &args.seq_args,
+	    &res.seq_res, 1);
+	if (!ret)
+		nfs4_update_changeattr(inode, &res.cinfo, timestamp, 0);
+
+	return ret;
+}
+
+static int _nfs42_proc_setxattr(struct inode *inode, const char *name,
+				const void *buf, size_t buflen, int flags)
+{
+	struct nfs_server *server = NFS_SERVER(inode);
+	struct page *pages[NFS4XATTR_MAXPAGES];
+	struct nfs42_setxattrargs arg = {
+		.fh		= NFS_FH(inode),
+		.xattr_pages	= pages,
+		.xattr_len	= buflen,
+		.xattr_name	= name,
+		.xattr_flags	= flags,
+	};
+	struct nfs42_setxattrres res;
+	struct rpc_message msg = {
+		.rpc_proc	= &nfs4_procedures[NFSPROC4_CLNT_SETXATTR],
+		.rpc_argp	= &arg,
+		.rpc_resp	= &res,
+	};
+	int ret, np;
+	unsigned long timestamp = jiffies;
+
+	if (buflen > server->sxasize)
+		return -ERANGE;
+
+	if (buflen > 0) {
+		np = nfs4_buf_to_pages_noslab(buf, buflen, arg.xattr_pages);
+		if (np < 0)
+			return np;
+	} else
+		np = 0;
+
+	ret = nfs4_call_sync(server->client, server, &msg, &arg.seq_args,
+	    &res.seq_res, 1);
+
+	for (; np > 0; np--)
+		put_page(pages[np - 1]);
+
+	if (!ret)
+		nfs4_update_changeattr(inode, &res.cinfo, timestamp, 0);
+
+	return ret;
+}
+
+static ssize_t _nfs42_proc_getxattr(struct inode *inode, const char *name,
+				void *buf, size_t buflen)
+{
+	struct nfs_server *server = NFS_SERVER(inode);
+	struct page *pages[NFS4XATTR_MAXPAGES] = {};
+	struct nfs42_getxattrargs arg = {
+		.fh		= NFS_FH(inode),
+		.xattr_pages	= pages,
+		.xattr_len	= buflen,
+		.xattr_name	= name,
+	};
+	struct nfs42_getxattrres res;
+	struct rpc_message msg = {
+		.rpc_proc	= &nfs4_procedures[NFSPROC4_CLNT_GETXATTR],
+		.rpc_argp	= &arg,
+		.rpc_resp	= &res,
+	};
+	int ret, np;
+
+	ret = nfs4_call_sync(server->client, server, &msg, &arg.seq_args,
+	    &res.seq_res, 0);
+	if (ret < 0)
+		return ret;
+
+	/*
+	 * Normally, the caching is done one layer up, but for successful
+	 * RPCS, always cache the result here, even if the caller was
+	 * just querying the length, or if the reply was too big for
+	 * the caller. This avoids a second RPC in the case of the
+	 * common query-alloc-retrieve cycle for xattrs.
+	 *
+	 * Note that xattr_len is always capped to XATTR_SIZE_MAX.
+	 */
+
+	nfs4_xattr_cache_add(inode, name, NULL, pages, res.xattr_len);
+
+	if (buflen) {
+		if (res.xattr_len > buflen)
+			return -ERANGE;
+		_copy_from_pages(buf, pages, 0, res.xattr_len);
+	}
+
+	np = DIV_ROUND_UP(res.xattr_len, PAGE_SIZE);
+	while (--np >= 0)
+		__free_page(pages[np]);
+
+	return res.xattr_len;
+}
+
+static ssize_t _nfs42_proc_listxattrs(struct inode *inode, void *buf,
+				 size_t buflen, u64 *cookiep, bool *eofp)
+{
+	struct nfs_server *server = NFS_SERVER(inode);
+	struct page **pages;
+	struct nfs42_listxattrsargs arg = {
+		.fh		= NFS_FH(inode),
+		.cookie		= *cookiep,
+	};
+	struct nfs42_listxattrsres res = {
+		.eof = false,
+		.xattr_buf = buf,
+		.xattr_len = buflen,
+	};
+	struct rpc_message msg = {
+		.rpc_proc	= &nfs4_procedures[NFSPROC4_CLNT_LISTXATTRS],
+		.rpc_argp	= &arg,
+		.rpc_resp	= &res,
+	};
+	u32 xdrlen;
+	int ret, np;
+
+
+	res.scratch = alloc_page(GFP_KERNEL);
+	if (!res.scratch)
+		return -ENOMEM;
+
+	xdrlen = nfs42_listxattr_xdrsize(buflen);
+	if (xdrlen > server->lxasize)
+		xdrlen = server->lxasize;
+	np = xdrlen / PAGE_SIZE + 1;
+
+	pages = kcalloc(np, sizeof(struct page *), GFP_KERNEL);
+	if (pages == NULL) {
+		__free_page(res.scratch);
+		return -ENOMEM;
+	}
+
+	arg.xattr_pages = pages;
+	arg.count = xdrlen;
+
+	ret = nfs4_call_sync(server->client, server, &msg, &arg.seq_args,
+	    &res.seq_res, 0);
+
+	if (ret >= 0) {
+		ret = res.copied;
+		*cookiep = res.cookie;
+		*eofp = res.eof;
+	}
+
+	while (--np >= 0) {
+		if (pages[np])
+			__free_page(pages[np]);
+	}
+
+	__free_page(res.scratch);
+	kfree(pages);
+
+	return ret;
+
+}
+
+ssize_t nfs42_proc_getxattr(struct inode *inode, const char *name,
+			      void *buf, size_t buflen)
+{
+	struct nfs4_exception exception = { };
+	ssize_t err;
+
+	do {
+		err = _nfs42_proc_getxattr(inode, name, buf, buflen);
+		if (err >= 0)
+			break;
+		err = nfs4_handle_exception(NFS_SERVER(inode), err,
+				&exception);
+	} while (exception.retry);
+
+	return err;
+}
+
+int nfs42_proc_setxattr(struct inode *inode, const char *name,
+			      const void *buf, size_t buflen, int flags)
+{
+	struct nfs4_exception exception = { };
+	int err;
+
+	do {
+		err = _nfs42_proc_setxattr(inode, name, buf, buflen, flags);
+		if (!err)
+			break;
+		err = nfs4_handle_exception(NFS_SERVER(inode), err,
+				&exception);
+	} while (exception.retry);
+
+	return err;
+}
+
+ssize_t nfs42_proc_listxattrs(struct inode *inode, void *buf,
+			      size_t buflen, u64 *cookiep, bool *eofp)
+{
+	struct nfs4_exception exception = { };
+	ssize_t err;
+
+	do {
+		err = _nfs42_proc_listxattrs(inode, buf, buflen,
+		    cookiep, eofp);
+		if (err >= 0)
+			break;
+		err = nfs4_handle_exception(NFS_SERVER(inode), err,
+				&exception);
+	} while (exception.retry);
+
+	return err;
+}
+
+int nfs42_proc_removexattr(struct inode *inode, const char *name)
+{
+	struct nfs4_exception exception = { };
+	int err;
+
+	do {
+		err = _nfs42_proc_removexattr(inode, name);
+		if (!err)
+			break;
+		err = nfs4_handle_exception(NFS_SERVER(inode), err,
+				&exception);
+	} while (exception.retry);
+
+	return err;
+}
diff --git a/fs/nfs/nfs42xattr.c b/fs/nfs/nfs42xattr.c
new file mode 100644
index 000000000000..23fdab977a2a
--- /dev/null
+++ b/fs/nfs/nfs42xattr.c
@@ -0,0 +1,1083 @@
+// SPDX-License-Identifier: GPL-2.0
+
+/*
+ * Copyright 2019, 2020 Amazon.com, Inc. or its affiliates. All rights reserved.
+ *
+ * User extended attribute client side cache functions.
+ *
+ * Author: Frank van der Linden <fllinden@amazon.com>
+ */
+#include <linux/errno.h>
+#include <linux/nfs_fs.h>
+#include <linux/hashtable.h>
+#include <linux/refcount.h>
+#include <uapi/linux/xattr.h>
+
+#include "nfs4_fs.h"
+#include "internal.h"
+
+/*
+ * User extended attributes client side caching is implemented by having
+ * a cache structure attached to NFS inodes. This structure is allocated
+ * when needed, and freed when the cache is zapped.
+ *
+ * The cache structure contains as hash table of entries, and a pointer
+ * to a special-cased entry for the listxattr cache.
+ *
+ * Accessing and allocating / freeing the caches is done via reference
+ * counting. The cache entries use a similar refcounting scheme.
+ *
+ * This makes freeing a cache, both from the shrinker and from the
+ * zap cache path, easy. It also means that, in current use cases,
+ * the large majority of inodes will not waste any memory, as they
+ * will never have any user extended attributes assigned to them.
+ *
+ * Attribute entries are hashed in to a simple hash table. They are
+ * also part of an LRU.
+ *
+ * There are three shrinkers.
+ *
+ * Two shrinkers deal with the cache entries themselves: one for
+ * large entries (> PAGE_SIZE), and one for smaller entries. The
+ * shrinker for the larger entries works more aggressively than
+ * those for the smaller entries.
+ *
+ * The other shrinker frees the cache structures themselves.
+ */
+
+/*
+ * 64 buckets is a good default. There is likely no reasonable
+ * workload that uses more than even 64 user extended attributes.
+ * You can certainly add a lot more - but you get what you ask for
+ * in those circumstances.
+ */
+#define NFS4_XATTR_HASH_SIZE	64
+
+#define NFSDBG_FACILITY	NFSDBG_XATTRCACHE
+
+struct nfs4_xattr_cache;
+struct nfs4_xattr_entry;
+
+struct nfs4_xattr_bucket {
+	spinlock_t lock;
+	struct hlist_head hlist;
+	struct nfs4_xattr_cache *cache;
+	bool draining;
+};
+
+struct nfs4_xattr_cache {
+	struct kref ref;
+	spinlock_t hash_lock;	/* protects hashtable and lru */
+	struct nfs4_xattr_bucket buckets[NFS4_XATTR_HASH_SIZE];
+	struct list_head lru;
+	struct list_head dispose;
+	atomic_long_t nent;
+	spinlock_t listxattr_lock;
+	struct inode *inode;
+	struct nfs4_xattr_entry *listxattr;
+	struct work_struct work;
+};
+
+struct nfs4_xattr_entry {
+	struct kref ref;
+	struct hlist_node hnode;
+	struct list_head lru;
+	struct list_head dispose;
+	char *xattr_name;
+	void *xattr_value;
+	size_t xattr_size;
+	struct nfs4_xattr_bucket *bucket;
+	uint32_t flags;
+};
+
+#define	NFS4_XATTR_ENTRY_EXTVAL	0x0001
+
+/*
+ * LRU list of NFS inodes that have xattr caches.
+ */
+static struct list_lru nfs4_xattr_cache_lru;
+static struct list_lru nfs4_xattr_entry_lru;
+static struct list_lru nfs4_xattr_large_entry_lru;
+
+static struct kmem_cache *nfs4_xattr_cache_cachep;
+
+static struct workqueue_struct *nfs4_xattr_cache_wq;
+
+/*
+ * Hashing helper functions.
+ */
+static void
+nfs4_xattr_hash_init(struct nfs4_xattr_cache *cache)
+{
+	unsigned int i;
+
+	for (i = 0; i < NFS4_XATTR_HASH_SIZE; i++) {
+		INIT_HLIST_HEAD(&cache->buckets[i].hlist);
+		spin_lock_init(&cache->buckets[i].lock);
+		cache->buckets[i].cache = cache;
+		cache->buckets[i].draining = false;
+	}
+}
+
+/*
+ * Locking order:
+ * 1. inode i_lock or bucket lock
+ * 2. list_lru lock (taken by list_lru_* functions)
+ */
+
+/*
+ * Wrapper functions to add a cache entry to the right LRU.
+ */
+static bool
+nfs4_xattr_entry_lru_add(struct nfs4_xattr_entry *entry)
+{
+	struct list_lru *lru;
+
+	lru = (entry->flags & NFS4_XATTR_ENTRY_EXTVAL) ?
+	    &nfs4_xattr_large_entry_lru : &nfs4_xattr_entry_lru;
+
+	return list_lru_add(lru, &entry->lru);
+}
+
+static bool
+nfs4_xattr_entry_lru_del(struct nfs4_xattr_entry *entry)
+{
+	struct list_lru *lru;
+
+	lru = (entry->flags & NFS4_XATTR_ENTRY_EXTVAL) ?
+	    &nfs4_xattr_large_entry_lru : &nfs4_xattr_entry_lru;
+
+	return list_lru_del(lru, &entry->lru);
+}
+
+/*
+ * This function allocates cache entries. They are the normal
+ * extended attribute name/value pairs, but may also be a listxattr
+ * cache. Those allocations use the same entry so that they can be
+ * treated as one by the memory shrinker.
+ *
+ * xattr cache entries are allocated together with names. If the
+ * value fits in to one page with the entry structure and the name,
+ * it will also be part of the same allocation (kmalloc). This is
+ * expected to be the vast majority of cases. Larger allocations
+ * have a value pointer that is allocated separately by kvmalloc.
+ *
+ * Parameters:
+ *
+ * @name:  Name of the extended attribute. NULL for listxattr cache
+ *         entry.
+ * @value: Value of attribute, or listxattr cache. NULL if the
+ *         value is to be copied from pages instead.
+ * @pages: Pages to copy the value from, if not NULL. Passed in to
+ *	   make it easier to copy the value after an RPC, even if
+ *	   the value will not be passed up to application (e.g.
+ *	   for a 'query' getxattr with NULL buffer).
+ * @len:   Length of the value. Can be 0 for zero-length attribues.
+ *         @value and @pages will be NULL if @len is 0.
+ */
+static struct nfs4_xattr_entry *
+nfs4_xattr_alloc_entry(const char *name, const void *value,
+		       struct page **pages, size_t len)
+{
+	struct nfs4_xattr_entry *entry;
+	void *valp;
+	char *namep;
+	size_t alloclen, slen;
+	char *buf;
+	uint32_t flags;
+
+	BUILD_BUG_ON(sizeof(struct nfs4_xattr_entry) +
+	    XATTR_NAME_MAX + 1 > PAGE_SIZE);
+
+	alloclen = sizeof(struct nfs4_xattr_entry);
+	if (name != NULL) {
+		slen = strlen(name) + 1;
+		alloclen += slen;
+	} else
+		slen = 0;
+
+	if (alloclen + len <= PAGE_SIZE) {
+		alloclen += len;
+		flags = 0;
+	} else {
+		flags = NFS4_XATTR_ENTRY_EXTVAL;
+	}
+
+	buf = kmalloc(alloclen, GFP_KERNEL_ACCOUNT | GFP_NOFS);
+	if (buf == NULL)
+		return NULL;
+	entry = (struct nfs4_xattr_entry *)buf;
+
+	if (name != NULL) {
+		namep = buf + sizeof(struct nfs4_xattr_entry);
+		memcpy(namep, name, slen);
+	} else {
+		namep = NULL;
+	}
+
+
+	if (flags & NFS4_XATTR_ENTRY_EXTVAL) {
+		valp = kvmalloc(len, GFP_KERNEL_ACCOUNT | GFP_NOFS);
+		if (valp == NULL) {
+			kfree(buf);
+			return NULL;
+		}
+	} else if (len != 0) {
+		valp = buf + sizeof(struct nfs4_xattr_entry) + slen;
+	} else
+		valp = NULL;
+
+	if (valp != NULL) {
+		if (value != NULL)
+			memcpy(valp, value, len);
+		else
+			_copy_from_pages(valp, pages, 0, len);
+	}
+
+	entry->flags = flags;
+	entry->xattr_value = valp;
+	kref_init(&entry->ref);
+	entry->xattr_name = namep;
+	entry->xattr_size = len;
+	entry->bucket = NULL;
+	INIT_LIST_HEAD(&entry->lru);
+	INIT_LIST_HEAD(&entry->dispose);
+	INIT_HLIST_NODE(&entry->hnode);
+
+	return entry;
+}
+
+static void
+nfs4_xattr_free_entry(struct nfs4_xattr_entry *entry)
+{
+	if (entry->flags & NFS4_XATTR_ENTRY_EXTVAL)
+		kvfree(entry->xattr_value);
+	kfree(entry);
+}
+
+static void
+nfs4_xattr_free_entry_cb(struct kref *kref)
+{
+	struct nfs4_xattr_entry *entry;
+
+	entry = container_of(kref, struct nfs4_xattr_entry, ref);
+
+	if (WARN_ON(!list_empty(&entry->lru)))
+		return;
+
+	nfs4_xattr_free_entry(entry);
+}
+
+static void
+nfs4_xattr_free_cache_cb(struct kref *kref)
+{
+	struct nfs4_xattr_cache *cache;
+	int i;
+
+	cache = container_of(kref, struct nfs4_xattr_cache, ref);
+
+	for (i = 0; i < NFS4_XATTR_HASH_SIZE; i++) {
+		if (WARN_ON(!hlist_empty(&cache->buckets[i].hlist)))
+			return;
+		cache->buckets[i].draining = false;
+	}
+
+	cache->listxattr = NULL;
+
+	kmem_cache_free(nfs4_xattr_cache_cachep, cache);
+
+}
+
+static struct nfs4_xattr_cache *
+nfs4_xattr_alloc_cache(void)
+{
+	struct nfs4_xattr_cache *cache;
+
+	cache = kmem_cache_alloc(nfs4_xattr_cache_cachep,
+	    GFP_KERNEL_ACCOUNT | GFP_NOFS);
+	if (cache == NULL)
+		return NULL;
+
+	kref_init(&cache->ref);
+	atomic_long_set(&cache->nent, 0);
+
+	return cache;
+}
+
+/*
+ * Set the listxattr cache, which is a special-cased cache entry.
+ * The special value ERR_PTR(-ESTALE) is used to indicate that
+ * the cache is being drained - this prevents a new listxattr
+ * cache from being added to what is now a stale cache.
+ */
+static int
+nfs4_xattr_set_listcache(struct nfs4_xattr_cache *cache,
+			 struct nfs4_xattr_entry *new)
+{
+	struct nfs4_xattr_entry *old;
+	int ret = 1;
+
+	spin_lock(&cache->listxattr_lock);
+
+	old = cache->listxattr;
+
+	if (old == ERR_PTR(-ESTALE)) {
+		ret = 0;
+		goto out;
+	}
+
+	cache->listxattr = new;
+	if (new != NULL && new != ERR_PTR(-ESTALE))
+		nfs4_xattr_entry_lru_add(new);
+
+	if (old != NULL) {
+		nfs4_xattr_entry_lru_del(old);
+		kref_put(&old->ref, nfs4_xattr_free_entry_cb);
+	}
+out:
+	spin_unlock(&cache->listxattr_lock);
+
+	return ret;
+}
+
+/*
+ * Unlink a cache from its parent inode, clearing out an invalid
+ * cache. Must be called with i_lock held.
+ */
+static struct nfs4_xattr_cache *
+nfs4_xattr_cache_unlink(struct inode *inode)
+{
+	struct nfs_inode *nfsi;
+	struct nfs4_xattr_cache *oldcache;
+
+	nfsi = NFS_I(inode);
+
+	oldcache = nfsi->xattr_cache;
+	if (oldcache != NULL) {
+		list_lru_del(&nfs4_xattr_cache_lru, &oldcache->lru);
+		oldcache->inode = NULL;
+	}
+	nfsi->xattr_cache = NULL;
+	nfsi->cache_validity &= ~NFS_INO_INVALID_XATTR;
+
+	return oldcache;
+
+}
+
+/*
+ * Discard a cache. Usually called by a worker, since walking all
+ * the entries can take up some cycles that we don't want to waste
+ * in the I/O path. Can also be called from the shrinker callback.
+ *
+ * The cache is dead, it has already been unlinked from its inode,
+ * and no longer appears on the cache LRU list.
+ *
+ * Mark all buckets as draining, so that no new entries are added. This
+ * could still happen in the unlikely, but possible case that another
+ * thread had grabbed a reference before it was unlinked from the inode,
+ * and is still holding it for an add operation.
+ *
+ * Remove all entries from the LRU lists, so that there is no longer
+ * any way to 'find' this cache. Then, remove the entries from the hash
+ * table.
+ *
+ * At that point, the cache will remain empty and can be freed when the final
+ * reference drops, which is very likely the kref_put at the end of
+ * this function, or the one called immediately afterwards in the
+ * shrinker callback.
+ */
+static void
+nfs4_xattr_discard_cache(struct nfs4_xattr_cache *cache)
+{
+	unsigned int i;
+	struct nfs4_xattr_entry *entry;
+	struct nfs4_xattr_bucket *bucket;
+	struct hlist_node *n;
+
+	nfs4_xattr_set_listcache(cache, ERR_PTR(-ESTALE));
+
+	for (i = 0; i < NFS4_XATTR_HASH_SIZE; i++) {
+		bucket = &cache->buckets[i];
+
+		spin_lock(&bucket->lock);
+		bucket->draining = true;
+		hlist_for_each_entry_safe(entry, n, &bucket->hlist, hnode) {
+			nfs4_xattr_entry_lru_del(entry);
+			hlist_del_init(&entry->hnode);
+			kref_put(&entry->ref, nfs4_xattr_free_entry_cb);
+		}
+		spin_unlock(&bucket->lock);
+	}
+
+	atomic_long_set(&cache->nent, 0);
+
+	kref_put(&cache->ref, nfs4_xattr_free_cache_cb);
+}
+
+static void
+nfs4_xattr_discard_cache_worker(struct work_struct *work)
+{
+	struct nfs4_xattr_cache *cache = container_of(work,
+	    struct nfs4_xattr_cache, work);
+
+	nfs4_xattr_discard_cache(cache);
+}
+
+static void
+nfs4_xattr_reap_cache(struct nfs4_xattr_cache *cache)
+{
+	queue_work(nfs4_xattr_cache_wq, &cache->work);
+}
+
+/*
+ * Get a referenced copy of the cache structure. Avoid doing allocs
+ * while holding i_lock. Which means that we do some optimistic allocation,
+ * and might have to free the result in rare cases.
+ *
+ * This function only checks the NFS_INO_INVALID_XATTR cache validity bit
+ * and acts accordingly, replacing the cache when needed. For the read case
+ * (!add), this means that the caller must make sure that the cache
+ * is valid before caling this function. getxattr and listxattr call
+ * revalidate_inode to do this. The attribute cache timeout (for the
+ * non-delegated case) is expected to be dealt with in the revalidate
+ * call.
+ */
+
+static struct nfs4_xattr_cache *
+nfs4_xattr_get_cache(struct inode *inode, int add)
+{
+	struct nfs_inode *nfsi;
+	struct nfs4_xattr_cache *cache, *oldcache, *newcache;
+
+	nfsi = NFS_I(inode);
+
+	cache = oldcache = NULL;
+
+	spin_lock(&inode->i_lock);
+
+	if (nfsi->cache_validity & NFS_INO_INVALID_XATTR)
+		oldcache = nfs4_xattr_cache_unlink(inode);
+	else
+		cache = nfsi->xattr_cache;
+
+	if (cache != NULL)
+		kref_get(&cache->ref);
+
+	spin_unlock(&inode->i_lock);
+
+	if (add && cache == NULL) {
+		newcache = NULL;
+
+		cache = nfs4_xattr_alloc_cache();
+		if (cache == NULL)
+			goto out;
+
+		spin_lock(&inode->i_lock);
+		if (nfsi->cache_validity & NFS_INO_INVALID_XATTR) {
+			/*
+			 * The cache was invalidated again. Give up,
+			 * since what we want to enter is now likely
+			 * outdated anyway.
+			 */
+			spin_unlock(&inode->i_lock);
+			kref_put(&cache->ref, nfs4_xattr_free_cache_cb);
+			cache = NULL;
+			goto out;
+		}
+
+		/*
+		 * Check if someone beat us to it.
+		 */
+		if (nfsi->xattr_cache != NULL) {
+			newcache = nfsi->xattr_cache;
+			kref_get(&newcache->ref);
+		} else {
+			kref_get(&cache->ref);
+			nfsi->xattr_cache = cache;
+			cache->inode = inode;
+			list_lru_add(&nfs4_xattr_cache_lru, &cache->lru);
+		}
+
+		spin_unlock(&inode->i_lock);
+
+		/*
+		 * If there was a race, throw away the cache we just
+		 * allocated, and use the new one allocated by someone
+		 * else.
+		 */
+		if (newcache != NULL) {
+			kref_put(&cache->ref, nfs4_xattr_free_cache_cb);
+			cache = newcache;
+		}
+	}
+
+out:
+	/*
+	 * Discarding an old cache is done via a workqueue.
+	 */
+	if (oldcache != NULL)
+		nfs4_xattr_reap_cache(oldcache);
+
+	return cache;
+}
+
+static inline struct nfs4_xattr_bucket *
+nfs4_xattr_hash_bucket(struct nfs4_xattr_cache *cache, const char *name)
+{
+	return &cache->buckets[jhash(name, strlen(name), 0) &
+	    (ARRAY_SIZE(cache->buckets) - 1)];
+}
+
+static struct nfs4_xattr_entry *
+nfs4_xattr_get_entry(struct nfs4_xattr_bucket *bucket, const char *name)
+{
+	struct nfs4_xattr_entry *entry;
+
+	entry = NULL;
+
+	hlist_for_each_entry(entry, &bucket->hlist, hnode) {
+		if (!strcmp(entry->xattr_name, name))
+			break;
+	}
+
+	return entry;
+}
+
+static int
+nfs4_xattr_hash_add(struct nfs4_xattr_cache *cache,