Implement stateful open() for NFSv4 as per RFC3010-bis.

The resulting state is saved in the NFS-specific part of the
struct inode.

Initially we just start with 3 possible states:
  - open for read
  - open for write
  - open for read/write

fs/nfs/inode.c

@@ -140,6 +140,10 @@ nfs_clear_inode(struct inode *inode)
 	cred = nfsi->cache_access.cred;
 	if (cred)
 		put_rpccred(cred);
+	/* Clean up the V4 state */
+	nfs4_put_shareowner(inode, nfsi->wo_owner);
+	nfs4_put_shareowner(inode, nfsi->ro_owner);
+	nfs4_put_shareowner(inode, nfsi->rw_owner);
 }
 
 void
@@ -1492,9 +1496,18 @@ static struct file_system_type nfs4_fs_type = {
 	.kill_sb	= nfs_kill_super,
 	.fs_flags	= FS_ODD_RENAME,
 };
+
+#define nfs4_zero_state(nfsi) \
+	do { \
+		(nfsi)->wo_owner = NULL; \
+		(nfsi)->ro_owner = NULL; \
+		(nfsi)->rw_owner = NULL; \
+	} while(0)
 #define register_nfs4fs() register_filesystem(&nfs4_fs_type)
 #define unregister_nfs4fs() unregister_filesystem(&nfs4_fs_type)
 #else
+#define nfs4_zero_state(nfsi) \
+	do { } while (0)
 #define register_nfs4fs() (0)
 #define unregister_nfs4fs()
 #endif
@@ -1516,6 +1529,7 @@ static struct inode *nfs_alloc_inode(struct super_block *sb)
 		return NULL;
 	nfsi->flags = 0;
 	nfsi->mm_cred = NULL;
+	nfs4_zero_state(nfsi);
 	return &nfsi->vfs_inode;
 }
 

fs/nfs/nfs4state.c

@@ -42,6 +42,16 @@
 #include <linux/slab.h>
 #include <linux/nfs_fs.h>
 
+/* This protects most of the client-side state. */
+static spinlock_t               state_spinlock = SPIN_LOCK_UNLOCKED;
+
+nfs4_stateid zero_stateid =
+	{ 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0 };
+
+nfs4_stateid one_stateid =
+	{ 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff };
+
+
 /*
  * nfs4_get_client(): returns an empty client structure
  * nfs4_put_client(): drops reference to client structure
@@ -54,11 +64,8 @@ nfs4_get_client(void)
 {
 	struct nfs4_client *clp;
 
-        if ((clp = kmalloc(sizeof(*clp), GFP_KERNEL))) {
-                atomic_set(&clp->cl_count, 1);
-                clp->cl_clientid = 0;
-                INIT_LIST_HEAD(&clp->cl_lockowners);
-        }
+	if ((clp = kmalloc(sizeof(*clp), GFP_KERNEL)))
+		memset(clp, 0, sizeof(nfs4_verifier));
 	return clp;
 }
 
@@ -66,12 +73,153 @@ void
 nfs4_put_client(struct nfs4_client *clp)
 {
 	BUG_ON(!clp);
-        BUG_ON(!atomic_read(&clp->cl_count));
-        
-        if (atomic_dec_and_test(&clp->cl_count)) {
-                BUG_ON(!list_empty(&clp->cl_lockowners));
 	kfree(clp);
+}
+
+static inline u32
+nfs4_alloc_lockowner_id(struct nfs4_client *clp)
+{
+	u32 res;
+
+	spin_lock(&state_spinlock);
+	res = clp->cl_lockowner_id ++;
+	spin_unlock(&state_spinlock);
+	return res;
+}
+
+/*
+ * nfs4_get_shareowner(): this is called on the OPEN or CREATE path to
+ * obtain a new shareowner.
+ *
+ * There are three shareowners (open_owner4 in rfc3010) per inode,
+ * one for each possible combination of share lock access. Since
+ * Linux does not support the deny access type, there are
+ * three (not 9) referenced by the nfs_inode:
+ *
+ * O_WRONLY: inode->wo_owner
+ * O_RDONLY: inode->ro_owner
+ * O_RDWR:   inode->rw_owner
+ *
+ * We create a new shareowner the first time a file is OPENed with
+ * one of the above shares. All other OPENs with a similar
+ * share use the single stateid associated with the inode.
+ *
+ */
+struct nfs4_shareowner *
+nfs4_get_shareowner(struct inode *dir)
+{
+	struct nfs4_client *clp;
+	struct nfs4_shareowner *sp;
+
+	sp = kmalloc(sizeof(*sp),GFP_KERNEL);
+	if (!sp)
+		return NULL;
+	clp = (NFS_SB(dir->i_sb))->nfs4_state;
+	BUG_ON(!clp);
+	init_MUTEX(&sp->so_sema);
+	sp->so_seqid = 0;                 /* arbitrary */
+	memset(sp->so_stateid, 0, sizeof(nfs4_stateid));
+	sp->so_id = nfs4_alloc_lockowner_id(clp);
+	return sp;
+}
+
+/*
+ * Called for each inode shareowner in nfs_clear_inode, 
+ * or if nfs4_do_open fails.
+ */
+void
+nfs4_put_shareowner(struct inode *inode, struct nfs4_shareowner *sp)
+{
+	if (!sp)
+		return;
+        kfree(sp);
+}
+
+/*
+* Called with sp->so_sema held.
+*
+* Increment the seqid if the OPEN/OPEN_DOWNGRADE/CLOSE succeeded, or
+* failed with a seqid incrementing error -
+* see comments nfs_fs.h:seqid_mutating_error()
+*/
+void
+nfs4_increment_seqid(u32 status, struct nfs4_shareowner *sp)
+{
+	if (status == NFS_OK || seqid_mutating_err(status))
+		sp->so_seqid++;
+}
+
+/*
+* Called by nfs4_proc_open to set the appropriate stateid
+*/
+int
+nfs4_set_inode_share(struct inode * inode, struct nfs4_shareowner *sp, unsigned int open_flags)
+{
+	struct nfs_inode *nfsi = NFS_I(inode);
+
+	switch (open_flags & O_ACCMODE) {
+		case O_RDONLY:
+			if (!nfsi->ro_owner) {
+				nfsi->ro_owner = sp;
+				return 0;
+			}
+			break;
+		case O_WRONLY:
+			if (!nfsi->wo_owner) {
+				nfsi->wo_owner = sp;
+				return 0;
+			}
+			break;
+		case O_RDWR:
+			if (!nfsi->rw_owner) {
+				nfsi->rw_owner = sp;
+				return 0;
+			}
+	}
+	return -EBUSY;
+}
+
+/*
+* Boolean test to determine if an OPEN call goes on the wire.
+*
+* Called by nfs4_proc_open.
+*/
+int
+nfs4_test_shareowner(struct inode *inode, unsigned int open_flags)
+{
+	struct nfs_inode *nfsi = NFS_I(inode);
+
+	switch (open_flags & O_ACCMODE) {
+		case O_RDONLY:
+			if(nfsi->ro_owner)
+				return 0;
+			break;
+		case O_WRONLY:
+			if(nfsi->wo_owner)
+				return 0;
+			break;
+		case O_RDWR:
+			if(nfsi->rw_owner)
+				return 0;
+        }
+        return 1;
+}
+
+struct nfs4_shareowner *
+nfs4_get_inode_share(struct inode * inode, unsigned int open_flags)
+{
+	struct nfs_inode *nfsi = NFS_I(inode);
+
+	switch (open_flags & O_ACCMODE) {
+		case O_RDONLY:
+			return nfsi->ro_owner;
+		case O_WRONLY:
+			return nfsi->wo_owner;
+		case O_RDWR:
+			return nfsi->rw_owner;
 	}
+	/* Duh gcc warning if we don't... */
+	return NULL;
 }
 
 /*

include/linux/nfs4.h

@@ -206,6 +206,8 @@ enum {
 	NFSPROC4_CLNT_READ,
 	NFSPROC4_CLNT_WRITE,
 	NFSPROC4_CLNT_COMMIT,
+	NFSPROC4_CLNT_OPEN,
+	NFSPROC4_CLNT_OPEN_CONFIRM,
 };
 
 #endif

include/linux/nfs_fs.h

@@ -155,6 +155,13 @@ struct nfs_inode {
 
 	wait_queue_head_t	nfs_i_wait;
 
+#ifdef CONFIG_NFS_V4
+        /* NFSv4 state */
+	struct nfs4_shareowner   *ro_owner;
+	struct nfs4_shareowner   *wo_owner;
+	struct nfs4_shareowner   *rw_owner;
+#endif /* CONFIG_NFS_V4*/
+
 	struct inode		vfs_inode;
 };
 
@@ -435,28 +442,74 @@ extern void * nfs_root_data(void);
 #define NFS_JUKEBOX_RETRY_TIME (5 * HZ)
 
 #ifdef CONFIG_NFS_V4
+
+/*
+ * In a seqid-mutating op, this macro controls which error return
+ * values trigger incrementation of the seqid.
+ *
+ * from rfc 3010:
+ * The client MUST monotonically increment the sequence number for the
+ * CLOSE, LOCK, LOCKU, OPEN, OPEN_CONFIRM, and OPEN_DOWNGRADE
+ * operations.  This is true even in the event that the previous
+ * operation that used the sequence number received an error.  The only
+ * exception to this rule is if the previous operation received one of
+ * the following errors: NFSERR_STALE_CLIENTID, NFSERR_STALE_STATEID,
+ * NFSERR_BAD_STATEID, NFSERR_BAD_SEQID, NFSERR_BADXDR,
+ * NFSERR_RESOURCE, NFSERR_NOFILEHANDLE.
+ *
+ */
+#define seqid_mutating_err(err)       \
+(((err) != NFSERR_STALE_CLIENTID) &&  \
+ ((err) != NFSERR_STALE_STATEID)  &&  \
+ ((err) != NFSERR_BAD_STATEID)    &&  \
+ ((err) != NFSERR_BAD_SEQID)      &&  \
+ ((err) != NFSERR_BAD_XDR)        &&  \
+ ((err) != NFSERR_RESOURCE)       &&  \
+ ((err) != NFSERR_NOFILEHANDLE))
+
 struct nfs4_client {
-        atomic_t                cl_count;       /* refcount */
         u64                     cl_clientid;    /* constant */
 	nfs4_verifier           cl_confirm;     
 
-        /*
-         * Starts a list of lockowners, linked through lo_list.
-	 */
-        struct list_head        cl_lockowners;  /* protected by state_spinlock */
+	u32			cl_lockowner_id;
 };
 
+/*
+* The ->so_sema is held during all shareowner seqid-mutating operations:
+* OPEN, OPEN_DOWNGRADE, and CLOSE.
+* Its purpose is to properly serialize so_seqid, as mandated by
+* the protocol.
+*/
+struct nfs4_shareowner {
+	u32                  so_id;      /* 32-bit identifier, unique */
+	struct semaphore     so_sema;
+	u32                  so_seqid;   /* protected by so_sema */
+	nfs4_stateid         so_stateid; /* protected by so_sema */
+	unsigned int         so_flags;   /* protected by so_sema */
+};
+
+
 /* nfs4proc.c */
 extern int nfs4_proc_renew(struct nfs_server *server);
 
 /* nfs4renewd.c */
 extern int nfs4_init_renewd(struct nfs_server *server);
-#endif /* CONFIG_NFS_V4 */
-
-#ifdef CONFIG_NFS_V4
 
+/* nfs4state.c */
 extern struct nfs4_client *nfs4_get_client(void);
 extern void nfs4_put_client(struct nfs4_client *clp);
+extern struct nfs4_shareowner * nfs4_get_shareowner(struct inode *inode);
+void nfs4_put_shareowner(struct inode *inode, struct nfs4_shareowner *sp);
+extern int nfs4_set_inode_share(struct inode * inode,
+                     struct nfs4_shareowner *sp, unsigned int flags);
+extern void nfs4_increment_seqid(u32 status, struct nfs4_shareowner *sp);
+extern int nfs4_test_shareowner(struct inode *inode, unsigned int open_flags);
+struct nfs4_shareowner * nfs4_get_inode_share(struct inode * inode, unsigned int open_flags);
+
+
+
+
+
 
 struct nfs4_mount_data;
 static inline int
@@ -481,6 +534,7 @@ destroy_nfsv4_state(struct nfs_server *server)
 #else
 #define create_nfsv4_state(server, data)  0
 #define destroy_nfsv4_state(server)       do { } while (0)
+#define nfs4_put_shareowner(inode, owner) do { } while (0)
 #endif
 
 #endif /* __KERNEL__ */

include/linux/nfs_xdr.h

@@ -87,6 +87,51 @@ struct nfs_pathconf {
 	__u32			max_namelen; /* max name length */
 };
 
+/*
+ * Arguments to the open call.
+ */
+struct nfs_openargs {
+	struct nfs_fh *         fh;
+	__u32                   seqid;
+	__u32                   share_access;
+	__u64                   clientid;
+	__u32                   id;
+	__u32                   opentype;
+	__u32                   createmode;
+	union {
+		struct iattr *  attrs;    /* UNCHECKED, GUARDED */
+		nfs4_verifier   verifier; /* EXCLUSIVE */
+	} u;
+	struct qstr *           name;
+	struct nfs4_getattr *   f_getattr;
+	struct nfs4_getattr *   d_getattr;
+};
+
+struct nfs_openres {
+	__u32                   status;
+	nfs4_stateid            stateid;
+	struct nfs_fh           fh;
+	struct nfs4_change_info * cinfo;
+	__u32                   rflags;
+	struct nfs4_getattr *   f_getattr;
+	struct nfs4_getattr *   d_getattr;
+};
+
+/*
+ * Arguments to the open_confirm call.
+ */
+struct nfs_open_confirmargs {
+	struct nfs_fh *         fh;
+	nfs4_stateid            stateid;
+	__u32                   seqid;
+};
+
+struct nfs_open_confirmres {
+	__u32                   status;
+	nfs4_stateid            stateid;
+};
+
+
 /*
  * Arguments to the read call.
  */