[illumos-gate merge]

commit 907c2824088edd487e0b231d3df0654a81aec933
    14202 Need direct callbacks from socket upcalls via ksocket
commit 7939798324de96253652dd33a2a42e25e4ab61f9
    14343 ufs: array subscript 'struct direct[0]' is partly outside array bounds

Conflicts:
    usr/src/uts/common/io/ksocket/ksocket.c
    usr/src/uts/common/sys/ksocket.h

6 files changed, 62 insertions, 33 deletions

diff --git a/usr/src/uts/common/fs/sockfs/sockcommon.c b/usr/src/uts/common/fs/sockfs/sockcommon.c
index e7d69f9896..edcb41951c 100644
--- a/usr/src/uts/common/fs/sockfs/sockcommon.c
+++ b/usr/src/uts/common/fs/sockfs/sockcommon.c
@@ -458,16 +458,16 @@ sonode_constructor(void *buf, void *cdrarg, int kmflags)
 	vp->v_data = so;
 	vn_setops(vp, socket_vnodeops);
 
-	so->so_priv 		= NULL;
+	so->so_priv		= NULL;
 	so->so_oobmsg		= NULL;
 
 	so->so_proto_handle	= NULL;
 
-	so->so_peercred 	= NULL;
+	so->so_peercred		= NULL;
 
 	so->so_rcv_queued	= 0;
-	so->so_rcv_q_head 	= NULL;
-	so->so_rcv_q_last_head 	= NULL;
+	so->so_rcv_q_head	= NULL;
+	so->so_rcv_q_last_head	= NULL;
 	so->so_rcv_head		= NULL;
 	so->so_rcv_last_head	= NULL;
 	so->so_rcv_wanted	= 0;
diff --git a/usr/src/uts/common/fs/ufs/ufs_dir.c b/usr/src/uts/common/fs/ufs/ufs_dir.c
index 8035e76025..02f7e57fcd 100644
--- a/usr/src/uts/common/fs/ufs/ufs_dir.c
+++ b/usr/src/uts/common/fs/ufs/ufs_dir.c
@@ -2870,9 +2870,6 @@ ufs_dirpurgedotdot(
  * Scan the directoy. If clr_dotdot is true clear the ..
  * directory else check to see if the directory is empty.
  *
- * Using a struct dirtemplate here is not precisely
- * what we want, but better than using a struct direct.
- *
  * clr_dotdot is used as a flag to tell us if we need
  * to clear the dotdot entry
  *
@@ -2886,20 +2883,19 @@ ufs_dirscan(
 	int clr_dotdot)
 {
 	offset_t off;
-	struct dirtemplate dbuf;
-	struct direct *dp = (struct direct *)&dbuf;
+	struct tmp_dir dbuf, *dp;
 	int err, count;
 	int empty = 1;	/* Assume it's empty */
-#define	MINDIRSIZ (sizeof (struct dirtemplate) / 2)
 
+	dp = &dbuf;
 	ASSERT(RW_LOCK_HELD(&ip->i_contents));
 
 	ASSERT(ip->i_size <= (offset_t)MAXOFF_T);
 	for (off = 0; off < ip->i_size; off += dp->d_reclen) {
 		err = ufs_rdwri(UIO_READ, FREAD, ip, (caddr_t)dp,
-		    (ssize_t)MINDIRSIZ, off, UIO_SYSSPACE, &count, cr);
+		    sizeof (struct tmp_dir), off, UIO_SYSSPACE, &count, cr);
 		/*
-		 * Since we read MINDIRSIZ, residual must
+		 * Since we read sizeof (struct tmp_dir), residual must
 		 * be 0 unless we're at end of file.
 		 */
 		if (err || count != 0 || dp->d_reclen == 0) {
@@ -3108,20 +3104,19 @@ int
 ufs_xattrdirempty(struct inode *ip, ino_t parentino, struct cred *cr)
 {
 	offset_t off;
-	struct dirtemplate dbuf;
-	struct direct *dp = (struct direct *)&dbuf;
+	struct tmp_dir dbuf, *dp;
 	int err, count;
 	int empty = 1;	/* Assume it's empty */
-#define	MINDIRSIZ (sizeof (struct dirtemplate) / 2)
 
+	dp = &dbuf;
 	ASSERT(RW_LOCK_HELD(&ip->i_contents));
 
 	ASSERT(ip->i_size <= (offset_t)MAXOFF_T);
 	for (off = 0; off < ip->i_size; off += dp->d_reclen) {
 		err = ufs_rdwri(UIO_READ, FREAD, ip, (caddr_t)dp,
-		    (ssize_t)MINDIRSIZ, off, UIO_SYSSPACE, &count, cr);
+		    sizeof (struct tmp_dir), off, UIO_SYSSPACE, &count, cr);
 		/*
-		 * Since we read MINDIRSIZ, residual must
+		 * Since we read sizeof (struct tmp_dir), residual must
 		 * be 0 unless we're at end of file.
 		 */
 
diff --git a/usr/src/uts/common/io/ksocket/ksocket.c b/usr/src/uts/common/io/ksocket/ksocket.c
index 0a5eec209f..10293941a2 100644
--- a/usr/src/uts/common/io/ksocket/ksocket.c
+++ b/usr/src/uts/common/io/ksocket/ksocket.c
@@ -942,5 +942,5 @@ ksocket_krecv_set(ksocket_t ks, ksocket_krecv_f cb, void *arg)
 void
 ksocket_krecv_unblock(ksocket_t ks)
 {
-	so_krecv_unblock(KSTOSO(ks));
+	return (so_krecv_unblock(KSTOSO(ks)));
 }
diff --git a/usr/src/uts/common/io/ksocket/ksocket_impl.h b/usr/src/uts/common/io/ksocket/ksocket_impl.h
index 516a68d358..2b685008e3 100644
--- a/usr/src/uts/common/io/ksocket/ksocket_impl.h
+++ b/usr/src/uts/common/io/ksocket/ksocket_impl.h
@@ -51,7 +51,7 @@
 #define	__KSOCKET_EV_cantrecvmore	KSOCKET_EV_CANTRECVMORE
 #define	__KSOCKET_EV_error		KSOCKET_EV_ERROR
 
-#define	KSOCKET_CALLBACK(so, cbfn, arg) 				\
+#define	KSOCKET_CALLBACK(so, cbfn, arg)					\
 	if ((so)->so_ksock_callbacks.ksock_cb_##cbfn != NULL) {		\
 		(*(so)->so_ksock_callbacks.ksock_cb_##cbfn)(SOTOKS(so),	\
 		    __KSOCKET_EV_##cbfn, (so)->so_ksock_cb_arg, (arg));	\
diff --git a/usr/src/uts/common/sys/fs/ufs_fsdir.h b/usr/src/uts/common/sys/fs/ufs_fsdir.h
index d0719c6f02..8e1c8a7d06 100644
--- a/usr/src/uts/common/sys/fs/ufs_fsdir.h
+++ b/usr/src/uts/common/sys/fs/ufs_fsdir.h
@@ -105,6 +105,19 @@ struct dirtemplate {
 	short		dotdot_namlen;
 	char		dotdot_name[4];		/* ditto */
 };
+
+/*
+ * Reduced structure for manipulating directories.
+ * Note, we are using __packed here to ensure the size of structure
+ * without changing the alignment.
+ */
+struct tmp_dir {
+	uint32_t	d_ino;		/* inode number of entry */
+	ushort_t	d_reclen;	/* length of this record */
+	ushort_t	d_namlen;	/* length of string in d_name */
+	char		d_name[4];	/* name must be no longer than this */
+} __packed;
+
 #endif
 
 #ifdef	__cplusplus
diff --git a/usr/src/uts/common/sys/ksocket.h b/usr/src/uts/common/sys/ksocket.h
index d720caa631..abf6bccfa5 100644
--- a/usr/src/uts/common/sys/ksocket.h
+++ b/usr/src/uts/common/sys/ksocket.h
@@ -84,47 +84,68 @@ typedef struct ksocket_callbacks {
 #define	KSOCKET_SLEEP	SOCKET_SLEEP
 #define	KSOCKET_NOSLEEP	SOCKET_NOSLEEP
 
-extern int 	ksocket_socket(ksocket_t *, int, int, int, int, struct cred *);
-extern int 	ksocket_bind(ksocket_t, struct sockaddr *, socklen_t,
+extern int	ksocket_socket(ksocket_t *, int, int, int, int, struct cred *);
+extern int	ksocket_bind(ksocket_t, struct sockaddr *, socklen_t,
 		    struct cred *);
-extern int 	ksocket_listen(ksocket_t, int, struct cred *);
-extern int 	ksocket_accept(ksocket_t, struct sockaddr *, socklen_t *,
+extern int	ksocket_listen(ksocket_t, int, struct cred *);
+extern int	ksocket_accept(ksocket_t, struct sockaddr *, socklen_t *,
 		    ksocket_t *, struct cred *);
-extern int 	ksocket_connect(ksocket_t, struct sockaddr *, socklen_t,
+extern int	ksocket_connect(ksocket_t, struct sockaddr *, socklen_t,
 		    struct cred *);
-extern int 	ksocket_send(ksocket_t, void *, size_t, int, size_t *,
+extern int	ksocket_send(ksocket_t, void *, size_t, int, size_t *,
 		    struct cred *);
 extern int	ksocket_sendto(ksocket_t, void *, size_t, int,
 		    struct sockaddr *, socklen_t, size_t *, struct cred *);
-extern int 	ksocket_sendmsg(ksocket_t, struct nmsghdr *, int, size_t *,
+extern int	ksocket_sendmsg(ksocket_t, struct nmsghdr *, int, size_t *,
 		    struct cred *);
 extern int	ksocket_sendmblk(ksocket_t, struct nmsghdr *, int,
 		    struct msgb **, struct cred *);
-extern int 	ksocket_recv(ksocket_t, void *, size_t, int, size_t *,
+extern int	ksocket_recv(ksocket_t, void *, size_t, int, size_t *,
 		    struct cred *);
 extern int	ksocket_recvfrom(ksocket_t, void *, size_t, int,
 		    struct sockaddr *, socklen_t *, size_t *, struct cred *);
 extern int	ksocket_recvmsg(ksocket_t, struct nmsghdr *, int, size_t *,
 		    struct cred *);
 extern int	ksocket_shutdown(ksocket_t, int, struct cred *);
-extern int 	ksocket_setsockopt(ksocket_t, int, int, const void *, int,
+extern int	ksocket_setsockopt(ksocket_t, int, int, const void *, int,
 		    struct cred *);
-extern int 	ksocket_getsockopt(ksocket_t, int, int, void *, int *,
+extern int	ksocket_getsockopt(ksocket_t, int, int, void *, int *,
 		    struct cred *);
-extern int 	ksocket_getpeername(ksocket_t, struct sockaddr *, socklen_t *,
+extern int	ksocket_getpeername(ksocket_t, struct sockaddr *, socklen_t *,
 		    struct cred *);
-extern int 	ksocket_getsockname(ksocket_t, struct sockaddr *, socklen_t *,
+extern int	ksocket_getsockname(ksocket_t, struct sockaddr *, socklen_t *,
 		    struct cred *);
 extern int	ksocket_ioctl(ksocket_t, int, intptr_t, int *, struct cred *);
 extern int	ksocket_spoll(ksocket_t, int, short, short *, struct cred *);
 extern int	ksocket_setcallbacks(ksocket_t, ksocket_callbacks_t *, void *,
 		    struct cred *);
-extern int 	ksocket_close(ksocket_t, struct cred *);
+extern int	ksocket_close(ksocket_t, struct cred *);
 extern void	ksocket_hold(ksocket_t);
 extern void	ksocket_rele(ksocket_t);
 
+/*
+ * These functions allow an alternative way for a ksocket to directly
+ * receive data when it arrives in sockfs rather than having it queued
+ * in a socket buffer that it must separately poll. The use of this
+ * results in no data being queued in sockfs.
+ *
+ * When the receive function receives data, it is responsible for always
+ * consuming all of the data. The return value of the callback function
+ * is used to indicate flow control and backpressure (similar to
+ * mac_tx(9E)). If, after processing the data, additional data can be
+ * received and processed, then the callback function should return
+ * B_TRUE. Otherwise it should return B_FALSE. This will result in the
+ * lower level socket interfaces (e.g. TCP) understanding that
+ * backpressure has been asserted (as though the sockfs buffer is full).
+ *
+ * Once whatever conditions that caused the callback function to assert
+ * that it needed to assert flow control are done, then it must call
+ * ksocket_krecv_unblock() to allow the flow to continue. If the receive
+ * callback ever returns B_FALSE there will generally be no additional
+ * data received until this is called.
+ */
 typedef boolean_t (*ksocket_krecv_f)(ksocket_t, struct msgb *, size_t, int,
-		    void *);
+    void *);
 extern int	ksocket_krecv_set(ksocket_t, ksocket_krecv_f, void *);
 extern void	ksocket_krecv_unblock(ksocket_t);