OS-2893 add support for epoll

Reviewed by: Robert Mustacchi <rm@joyent.com>
Reviewed by: Jerry Jelinek <jerry.jelinek@joyent.com>

32 files changed, 1421 insertions, 99 deletions

diff --git a/usr/src/cmd/truss/codes.c b/usr/src/cmd/truss/codes.c
index 069268dc05..4e453da0c1 100644
--- a/usr/src/cmd/truss/codes.c
+++ b/usr/src/cmd/truss/codes.c
@@ -23,7 +23,7 @@
  * Copyright (c) 1989, 2010, Oracle and/or its affiliates. All rights reserved.
  * Copyright (c) 2013 by Delphix. All rights reserved.
  * Copyright 2011 Nexenta Systems, Inc. All rights reserved.
- * Copyright (c) 2012, Joyent, Inc. All rights reserved.
+ * Copyright (c) 2014, Joyent, Inc. All rights reserved.
  * Copyright (c) 2014, OmniTI Computer Consulting, Inc. All rights reserved.
  */
 
@@ -703,6 +703,8 @@ const struct ioc {
 	/* /dev/poll ioctl() control codes */
 	{ (uint_t)DP_POLL,	"DP_POLL",	NULL },
 	{ (uint_t)DP_ISPOLLED,	"DP_ISPOLLED",	NULL },
+	{ (uint_t)DP_PPOLL,	"DP_PPOLL",	NULL },
+	{ (uint_t)DP_EPOLLCOMPAT, "DP_EPOLLCOMPAT",	NULL },
 	/* the old /proc ioctl() control codes */
 #define	PIOC	('q'<<8)
 	{ (uint_t)(PIOC|1),	"PIOCSTATUS",	NULL },
diff --git a/usr/src/lib/libc/amd64/Makefile b/usr/src/lib/libc/amd64/Makefile
index 873c2ded87..cec30a4fbd 100644
--- a/usr/src/lib/libc/amd64/Makefile
+++ b/usr/src/lib/libc/amd64/Makefile
@@ -20,7 +20,7 @@
 #
 #
 # Copyright (c) 2004, 2010, Oracle and/or its affiliates. All rights reserved.
-# Copyright (c) 2012, Joyent, Inc. All rights reserved.
+# Copyright (c) 2014, Joyent, Inc. All rights reserved.
 #
 # Copyright (c) 2013, OmniTI Computer Consulting, Inc. All rights reserved.
 # Copyright 2011 Nexenta Systems, Inc.  All rights reserved.
@@ -851,6 +851,7 @@ PORTSYS=			\
 	chmod.o			\
 	chown.o			\
 	corectl.o		\
+	epoll.o			\
 	exacctsys.o		\
 	execl.o			\
 	execle.o		\
diff --git a/usr/src/lib/libc/i386/Makefile.com b/usr/src/lib/libc/i386/Makefile.com
index d91540d0ae..ddbed44735 100644
--- a/usr/src/lib/libc/i386/Makefile.com
+++ b/usr/src/lib/libc/i386/Makefile.com
@@ -20,9 +20,8 @@
 #
 #
 # Copyright (c) 2004, 2010, Oracle and/or its affiliates. All rights reserved.
-# Copyright (c) 2013, Joyent, Inc.  All rights reserved.
+# Copyright (c) 2014, Joyent, Inc.  All rights reserved.
 # Copyright (c) 2013, OmniTI Computer Consulting, Inc. All rights reserved.
-# Copyright (c) 2012, Joyent, Inc. All rights reserved.
 #
 # Copyright 2011 Nexenta Systems, Inc.  All rights reserved.
 # Use is subject to license terms.
@@ -890,6 +889,7 @@ PORTSYS=			\
 	chmod.o			\
 	chown.o			\
 	corectl.o		\
+	epoll.o			\
 	exacctsys.o		\
 	execl.o			\
 	execle.o		\
diff --git a/usr/src/lib/libc/port/mapfile-vers b/usr/src/lib/libc/port/mapfile-vers
index 86adb85d94..c08f234962 100644
--- a/usr/src/lib/libc/port/mapfile-vers
+++ b/usr/src/lib/libc/port/mapfile-vers
@@ -24,9 +24,8 @@
 # Copyright 2010 Nexenta Systems, Inc.  All rights reserved.
 # Use is subject to license terms.
 #
-# Copyright (c) 2012, Joyent, Inc. All rights reserved.
+# Copyright (c) 2014, Joyent, Inc. All rights reserved.
 # Copyright (c) 2012 by Delphix. All rights reserved.
-# Copyright (c) 2012, Joyent, Inc.  All rights reserved.
 # Copyright (c) 2013, OmniTI Computer Consulting, Inc. All rights reserved.
 # Copyright (c) 2013 Gary Mills
 #
@@ -2721,6 +2720,11 @@ $endif
 	_dgettext;
 	_doprnt;
 	_doscan;
+	epoll_create;
+	epoll_create1;
+	epoll_ctl;
+	epoll_wait;
+	epoll_pwait;
 	_errfp;
 	_errxfp;
 	exportfs;
diff --git a/usr/src/lib/libc/port/sys/epoll.c b/usr/src/lib/libc/port/sys/epoll.c
new file mode 100644
index 0000000000..d90b625293
--- /dev/null
+++ b/usr/src/lib/libc/port/sys/epoll.c
@@ -0,0 +1,207 @@
+/*
+ * This file and its contents are supplied under the terms of the
+ * Common Development and Distribution License ("CDDL"), version 1.0.
+ * You may only use this file in accordance with the terms of version
+ * 1.0 of the CDDL.
+ *
+ * A full copy of the text of the CDDL should have accompanied this
+ * source.  A copy of the CDDL is also available via the Internet at
+ * http://www.illumos.org/license/CDDL.
+ */
+
+/*
+ * Copyright (c) 2014, Joyent, Inc.  All rights reserved.
+ */
+
+#include <sys/types.h>
+#include <sys/epoll.h>
+#include <sys/devpoll.h>
+#include <unistd.h>
+#include <errno.h>
+#include <fcntl.h>
+#include <poll.h>
+
+/*
+ * Events that match their epoll(7) equivalents.
+ */
+#if EPOLLIN != POLLIN
+#error value of EPOLLIN does not match value of POLLIN
+#endif
+
+#if EPOLLPRI != POLLPRI
+#error value of EPOLLPRI does not match value of POLLPRI
+#endif
+
+#if EPOLLOUT != POLLOUT
+#error value of EPOLLOUT does not match value of POLLOUT
+#endif
+
+#if EPOLLRDNORM != POLLRDNORM
+#error value of EPOLLRDNORM does not match value of POLLRDNORM
+#endif
+
+#if EPOLLRDBAND != POLLRDBAND
+#error value of EPOLLRDBAND does not match value of POLLRDBAND
+#endif
+
+#if EPOLLERR != POLLERR
+#error value of EPOLLERR does not match value of POLLERR
+#endif
+
+#if EPOLLHUP != POLLHUP
+#error value of EPOLLHUP does not match value of POLLHUP
+#endif
+
+/*
+ * Events that we ignore entirely.  They can be set in events, but they will
+ * never be returned.
+ */
+#define	EPOLLIGNORED 	(EPOLLMSG | EPOLLWAKEUP)
+
+/*
+ * Events that we swizzle into other bit positions.
+ */
+#define	EPOLLSWIZZLED	\
+	(EPOLLRDHUP | EPOLLONESHOT | EPOLLET | EPOLLWRBAND | EPOLLWRNORM)
+
+int
+epoll_create(int size)
+{
+	int fd;
+
+	/*
+	 * From the epoll_create() man page:  "Since Linux 2.6.8, the size
+	 * argument is ignored, but must be greater than zero."  You keep using
+	 * that word "ignored"...
+	 */
+	if (size <= 0) {
+		errno = EINVAL;
+		return (-1);
+	}
+
+	if ((fd = open("/dev/poll", O_RDWR)) == -1)
+		return (-1);
+
+	if (ioctl(fd, DP_EPOLLCOMPAT, 0) == -1) {
+		(void) close(fd);
+		return (-1);
+	}
+
+	return (fd);
+}
+
+int
+epoll_create1(int flags)
+{
+	int fd, oflags = O_RDWR;
+
+	if (flags & EPOLL_CLOEXEC)
+		oflags |= O_CLOEXEC;
+
+	if ((fd = open("/dev/poll", oflags)) == -1)
+		return (-1);
+
+	if (ioctl(fd, DP_EPOLLCOMPAT, 0) == -1) {
+		(void) close(fd);
+		return (-1);
+	}
+
+	return (fd);
+}
+
+int
+epoll_ctl(int epfd, int op, int fd, struct epoll_event *event)
+{
+	dvpoll_epollfd_t epoll[2];
+	uint32_t events = event->events;
+	uint32_t ev;
+	int i = 0;
+
+	epoll[i].dpep_pollfd.fd = fd;
+
+	switch (op) {
+	case EPOLL_CTL_DEL:
+		ev = POLLREMOVE;
+		break;
+
+	case EPOLL_CTL_MOD:
+		/*
+		 * In the modify case, we pass down two events:  one to
+		 * remove the event and another to add it back.
+		 */
+		epoll[i++].dpep_pollfd.events = POLLREMOVE;
+		epoll[i].dpep_pollfd.fd = fd;
+		/* FALLTHROUGH */
+
+	case EPOLL_CTL_ADD:
+		/*
+		 * Mask off the events that we ignore, and then swizzle the
+		 * events for which our values differ from their epoll(7)
+		 * equivalents.
+		 */
+		ev = events & ~(EPOLLIGNORED | EPOLLSWIZZLED);
+
+		if (events & EPOLLRDHUP)
+			ev |= POLLRDHUP;
+
+		if (events & EPOLLET)
+			ev |= POLLET;
+
+		if (events & EPOLLONESHOT)
+			ev |= POLLONESHOT;
+
+		if (events & EPOLLWRNORM)
+			ev |= POLLWRNORM;
+
+		if (events & EPOLLWRBAND)
+			ev |= POLLWRBAND;
+
+		epoll[i].dpep_data = event->data.u64;
+		break;
+
+	default:
+		errno = EOPNOTSUPP;
+		return (-1);
+	}
+
+	epoll[i].dpep_pollfd.events = ev;
+
+	return (write(epfd, epoll, sizeof (epoll[0]) * (i + 1)) == -1 ? -1 : 0);
+}
+
+int
+epoll_wait(int epfd, struct epoll_event *events,
+    int maxevents, int timeout)
+{
+	struct dvpoll arg;
+
+	if (maxevents <= 0) {
+		errno = EINVAL;
+		return (-1);
+	}
+
+	arg.dp_nfds = maxevents;
+	arg.dp_timeout = timeout;
+	arg.dp_fds = (pollfd_t *)events;
+
+	return (ioctl(epfd, DP_POLL, &arg));
+}
+
+int
+epoll_pwait(int epfd, struct epoll_event *events,
+    int maxevents, int timeout, const sigset_t *sigmask)
+{
+	struct dvpoll arg;
+
+	if (maxevents <= 0) {
+		errno = EINVAL;
+		return (-1);
+	}
+
+	arg.dp_nfds = maxevents;
+	arg.dp_timeout = timeout;
+	arg.dp_fds = (pollfd_t *)events;
+	arg.dp_setp = (sigset_t *)sigmask;
+
+	return (ioctl(epfd, DP_PPOLL, &arg));
+}
diff --git a/usr/src/lib/libc/sparc/Makefile.com b/usr/src/lib/libc/sparc/Makefile.com
index c8779479d3..34d35699d0 100644
--- a/usr/src/lib/libc/sparc/Makefile.com
+++ b/usr/src/lib/libc/sparc/Makefile.com
@@ -20,9 +20,8 @@
 #
 #
 # Copyright (c) 1989, 2010, Oracle and/or its affiliates. All rights reserved.
-# Copyright (c) 2012, Joyent, Inc.  All rights reserved.
 # Copyright (c) 2013, OmniTI Computer Consulting, Inc. All rights reserved.
-# Copyright (c) 2012, Joyent, Inc.  All rights reserved.
+# Copyright (c) 2014, Joyent, Inc.  All rights reserved.
 #
 # Copyright 2011 Nexenta Systems, Inc.  All rights reserved.
 # Use is subject to license terms.
@@ -925,6 +924,7 @@ PORTSYS=			\
 	chmod.o			\
 	chown.o			\
 	corectl.o		\
+	epoll.o			\
 	exacctsys.o		\
 	execl.o			\
 	execle.o		\
diff --git a/usr/src/lib/libc/sparcv9/Makefile.com b/usr/src/lib/libc/sparcv9/Makefile.com
index 7f4a6e4e38..65ac2dedc7 100644
--- a/usr/src/lib/libc/sparcv9/Makefile.com
+++ b/usr/src/lib/libc/sparcv9/Makefile.com
@@ -20,9 +20,8 @@
 #
 #
 # Copyright (c) 1989, 2010, Oracle and/or its affiliates. All rights reserved.
-# Copyright (c) 2012, Joyent, Inc.  All rights reserved.
 # Copyright (c) 2013, OmniTI Computer Consulting, Inc. All rights reserved.
-# Copyright (c) 2012, Joyent, Inc.  All rights reserved.
+# Copyright (c) 2014, Joyent, Inc.  All rights reserved.
 #
 # Copyright 2011 Nexenta Systems, Inc.  All rights reserved.
 # Use is subject to license terms.
@@ -871,6 +870,7 @@ PORTSYS=			\
 	chmod.o			\
 	chown.o			\
 	corectl.o		\
+	epoll.o			\
 	exacctsys.o		\
 	execl.o			\
 	execle.o		\
diff --git a/usr/src/man/man3c/Makefile b/usr/src/man/man3c/Makefile
index f3b6f18a7f..3056f2c923 100644
--- a/usr/src/man/man3c/Makefile
+++ b/usr/src/man/man3c/Makefile
@@ -13,6 +13,7 @@
 # Copyright 2011, Richard Lowe
 # Copyright 2013 Nexenta Systems, Inc.  All rights reserved.
 # Copyright 2013, OmniTI Computer Consulting, Inc. All rights reserved.
+# Copyright (c) 2014, Joyent, Inc.  All rights reserved.
 #
 
 include		$(SRC)/Makefile.master
@@ -106,6 +107,9 @@ MANFILES= 	__fbufsize.3c					\
 	 	enable_extended_FILE_stdio.3c			\
 	 	encrypt.3c					\
 	 	end.3c						\
+	 	epoll_create.3c					\
+	 	epoll_ctl.3c					\
+	 	epoll_wait.3c					\
 	 	err.3c						\
 	 	euclen.3c					\
 	 	exit.3c						\
@@ -706,6 +710,8 @@ MANLINKS=	FD_CLR.3c				\
 	 	endusershell.3c				\
 	 	endutent.3c				\
 	 	endutxent.3c				\
+	 	epoll_create1.3c			\
+	 	epoll_pwait.3c				\
 	 	erand48.3c				\
 	 	errno.3c				\
 	 	errx.3c					\
@@ -1435,6 +1441,9 @@ _etext.3c				:= LINKSRC = end.3c
 edata.3c				:= LINKSRC = end.3c
 etext.3c				:= LINKSRC = end.3c
 
+epoll_create1.3c			:= LINKSRC = epoll_create.3c
+epoll_pwait.3c				:= LINKSRC = epoll_wait.3c
+
 errx.3c					:= LINKSRC = err.3c
 verr.3c					:= LINKSRC = err.3c
 verrx.3c				:= LINKSRC = err.3c
diff --git a/usr/src/man/man3c/epoll_create.3c b/usr/src/man/man3c/epoll_create.3c
new file mode 100644
index 0000000000..87d0a85c38
--- /dev/null
+++ b/usr/src/man/man3c/epoll_create.3c
@@ -0,0 +1,106 @@
+'\" te
+.\"  Copyright (c) 2014, Joyent, Inc. All Rights Reserved.
+.\"  This file and its contents are supplied under the terms of the
+.\"  Common Development and Distribution License ("CDDL"), version 1.0.
+.\"  You may only use this file in accordance with the terms of version
+.\"  1.0 of the CDDL.
+.\" 
+.\"  A full copy of the text of the CDDL should have accompanied this
+.\"  source.  A copy of the CDDL is also available via the Internet at
+.\"  http://www.illumos.org/license/CDDL.
+.TH EPOLL_CREATE 3C "Apr 17, 2014"
+.SH NAME
+epoll_create, epoll_create1 \- create an epoll instance
+.SH SYNOPSIS
+
+.LP
+.nf
+#include <sys/epoll.h>
+
+\fBint\fR \fBepoll_create\fR(\fBint\fR \fIsize\fR);
+.fi
+
+.LP
+.nf
+\fBint\fR \fBepoll_create1\fR(\fBint\fR \fIflags\fR);
+.fi
+
+.SH DESCRIPTION
+.sp
+.LP
+The \fBepoll_create()\fR and \fBepoll_create1()\fR functions both create an
+\fBepoll\fR(5) instance that can be operated upon via \fBepoll_ctl\fR(3C),
+\fBepoll_wait\fR(3C) and \fBepoll_pwait\fR(3C).  \fBepoll\fR instances are
+represented as file descriptors, and should be closed via \fBclose\fR(2).
+
+The only difference between the two functions is their signature;
+\fBepoll_create()\fR takes a size argument that
+is vestigal and is only meaningful in as much as it must be greater than
+zero, while \fBepoll_create1()\fR takes a flags argument that can have
+any of the following values:
+
+.sp
+.ne 2
+.na
+\fBEPOLL_CLOEXEC\fR
+.ad
+.RS 12n
+Instance should be closed upon an
+\fBexec\fR(2); see \fBopen\fR(2)'s description of \fBO_CLOEXEC\fR.
+.RE
+
+.SH RETURN VALUES
+.sp
+.LP
+Upon succesful completion, 0 is returned. Otherwise, -1 is returned and errno
+is set to indicate the error.
+.SH ERRORS
+.sp
+.LP
+The \fBepoll_create()\fR and \fBepoll_create1()\fR functions will fail if:
+.sp
+.ne 2
+.na
+\fB\fBEINVAL\fR\fR
+.ad
+.RS 10n
+Either the \fIsize\fR is zero (\fBepoll_create()\fR) or the \fIflags\fR
+are invalid (\fBepoll_create1()\fR).
+.RE
+
+.sp
+.ne 2
+.na
+\fB\fBEMFILE\fR\fR
+.ad
+.RS 10n
+There are currently {\fBOPEN_MAX\fR} file descriptors open in the calling
+process.
+.RE
+
+.sp
+.ne 2
+.na
+\fB\fBENFILE\fR\fR
+.ad
+.RS 10n
+The maximum allowable number of files is currently open in the system.
+.RE
+
+.sp
+.LP
+.SH NOTES
+.sp
+.LP
+
+The \fBepoll\fR(5) facility is implemented for purposes of offering
+compatibility for Linux-borne applications; native
+applications should continue to prefer using event ports via the
+\fBport_create\fR(3C), \fBport_associate\fR(3C) and \fBport_get\fR(3C)
+interfaces.  See \fBepoll\fR(5) for compatibility details and restrictions.
+.RE
+
+.SH SEE ALSO
+.sp
+.LP
+\fBepoll_ctl\fR(3C), \fBepoll_wait\fR(3C), \fBepoll\fR(5)
diff --git a/usr/src/man/man3c/epoll_ctl.3c b/usr/src/man/man3c/epoll_ctl.3c
new file mode 100644
index 0000000000..0cce6affa4
--- /dev/null
+++ b/usr/src/man/man3c/epoll_ctl.3c
@@ -0,0 +1,316 @@
+'\" te
+.\"  Copyright (c) 2014, Joyent, Inc. All Rights Reserved.
+.\"  This file and its contents are supplied under the terms of the
+.\"  Common Development and Distribution License ("CDDL"), version 1.0.
+.\"  You may only use this file in accordance with the terms of version
+.\"  1.0 of the CDDL.
+.\" 
+.\"  A full copy of the text of the CDDL should have accompanied this
+.\"  source.  A copy of the CDDL is also available via the Internet at
+.\"  http://www.illumos.org/license/CDDL.
+.TH EPOLL_CTL 3C "Apr 17, 2014"
+.SH NAME
+epoll_ctl \- control an epoll instance
+.SH SYNOPSIS
+
+.LP
+.nf
+#include <sys/epoll.h>
+
+\fBint\fR \fBepoll_ctl\fR(\fBint\fR \fIepfd\fR, \fBint\fR \fIop\fR, \fBint\fR \fIfd\fR, \fBstruct epoll_event *\fR\fIevent\fR);
+.fi
+
+.SH DESCRIPTION
+.sp
+.LP
+The \fBepoll_ctl()\fR function executes the operation specified by
+\fIop\fR (as parameterized by \fIevent\fR) on the \fIepfd\fR epoll instance.
+Valid values for \fIop\fR:
+
+.sp
+.ne 2
+.na
+\fBEPOLL_CTL_ADD\fR
+.ad
+.RS 12n
+For the \fBepoll\fR(5) instance specified by \fIepfd\fR, 
+associate the file descriptor specified by \fIfd\fR with the event specified
+by \fIevent\fR.
+.RE
+
+.sp
+.ne 2
+.na
+\fBEPOLL_CTL_DEL\fR
+.ad
+.RS 12n
+For the \fBepoll\fR(5) instance specified by \fIepfd\fR, 
+remove all event associations for the file descriptor specified by \fIfd\fR.
+\fIevent\fR is ignored, and may be NULL.
+.RE
+
+.sp
+.ne 2
+.na
+\fBEPOLL_CTL_MOD\fR
+.ad
+.RS 12n
+For the \fBepoll\fR(5) instance specified by \fIepfd\fR, modify the event
+association for the file descriptor specified by \fIfd\fR to be that
+specified by \fIevent\fR.
+
+.RE
+
+The \fIevent\fR parameter has the following structure:
+
+.in +4
+.nf
+typedef union epoll_data {
+	void    *ptr;
+	int      fd;
+	uint32_t u32;
+	uint64_t u64;
+} epoll_data_t;
+
+struct epoll_event {
+	uint32_t     events;
+	epoll_data_t data;
+};
+.fi
+.in -4
+
+The \fIdata\fR field specifies the datum to
+be associated with the event and 
+will be returned via \fBepoll_wait\fR(3C).  
+The \fIevents\fR field denotes both the desired events (when specified via
+\fBepoll_ctl()\fR) and the events that have occurred (when returned via
+\fBepoll_wait\fR(3C)).
+In either case, the
+\fIevents\fR field is a bitmask constructed by a logical \fBOR\fR operation
+of any combination of the following event flags:
+
+.sp
+.ne 2
+.na
+\fBEPOLLIN\fR
+.as
+.RS 14n
+Data other than high priority data may be read without blocking. For streams,
+this flag is set in the returned \fIevents\fR even if the message is of
+zero length.
+.RE
+
+.sp
+.ne 2
+.na
+\fBEPOLLPRI\fR
+.as
+.RS 14n
+Normal data (priority band equals 0) may be read without blocking. For streams,
+this flag is set in the returned \fIevents\fR even if the message is of zero
+length.
+.RE
+
+.sp
+.ne 2
+.na
+\fBEPOLLOUT\fR
+.as
+.RS 14n
+Normal data (priority band equals 0) may be written without blocking.
+.RE
+
+.sp
+.ne 2
+.na
+\fBEPOLLRDNORM\fR
+.as
+.RS 14n
+Normal data (priority band equals 0) may be read without blocking. For streams,
+this flag is set in the returned \fIrevents\fR even if the message is of
+zero length.
+.RE
+
+.sp
+.ne 2
+.na
+\fBEPOLLRDBAND\fR
+.as
+.RS 14n
+Data from a non-zero priority band may be read without blocking. For streams,
+this flag is set in the returned \fIrevents\fR even if the message is of
+zero length.
+.RE
+
+.sp
+.ne 2
+.na
+\fBEPOLLWRNORM\fR
+.as
+.RS 14n
+The same as \fBEPOLLOUT\fR.
+.RE
+
+.sp
+.ne 2
+.na
+\fBEPOLLWRBAND\fR
+.as
+.RS 14n
+Priority data (priority band > 0) may be written.  This event only examines
+bands that have been written to at least once.
+.RE
+
+.sp
+.ne 2
+.na
+\fBEPOLLMSG\fR
+.as
+.RS 14n
+This exists only for backwards binary and source compatibility with Linux;
+it has no meaning and is ignored.
+.RE
+
+.sp
+.ne 2
+.na
+\fBEPOLLERR\fR
+.as
+.RS 14n
+An error has occurred on the device or stream.  This flag is only valid in the
+returned \fIevents\fR field.
+.RE
+
+.sp
+.ne 2
+.na
+\fBEPOLLHUP\fR
+.as
+.RS 14n
+A hangup has occurred on the stream. This event and \fBEPOLLOUT\fR are mutually
+exclusive; a stream can never be writable if a hangup has occurred. However,
+this event and  \fBEPOLLIN\fR, \fBEPOLLRDNORM\fR, \fBEPOLLRDBAND\fR,
+\fBEPOLLRDHUP\fR or
+\fBEPOLLPRI\fR are not mutually exclusive. This flag is only valid in the
+the \fIevents\fR field returned from \fBepoll_wait\fR(3C); it is not used
+in the \fIevents\fR field specified via \fBepoll_ctl()\fR.
+.RE
+
+.sp
+.ne 2
+.na
+\fBEPOLLRDHUP\fR
+.as
+.RS 14n
+The stream socket peer shutdown the writing half of the connection and no
+further data will be readable via the socket.  This event is not mutually
+exclusive with \fBEPOLLIN\fR.
+.RE
+
+.sp
+.ne 2
+.na
+\fBEPOLLWAKEUP\fR
+.as
+.RS 14n
+This exists only for backwards binary and source compatibility with Linux;
+it has no meaning and is ignored.
+.RE
+
+.sp
+.ne 2
+.na
+\fBEPOLLONESHOT\fR
+.as
+.RS 14n
+Sets the specified event to be in one-shot mode, whereby the event association
+with the \fBepoll\fR(5) instance specified by \fIepfd\fR is removed atomically
+as the event is returned via \fBepoll_wait\fR(3C).  Use of this mode allows
+for resolution of some of the
+races inherent in multithreaded use of \fBepoll_wait\fR(3C).
+.RE
+
+.sp
+.ne 2
+.na
+\fBEPOLLET\fR
+.as
+.RS 14n
+Sets the specified event to be edge-triggered mode instead of the default
+mode of level-triggered.  In this mode, events will be induced by
+transitions on an event source rather than the state of the event source.
+While perhaps superficially appealing, this mode introduces several new
+potential failure modes for user-level software and should be used
+with caution.
+.RE
+
+.SH RETURN VALUES
+.sp
+.LP
+Upon succesful completion, \fBepoll_ctl()\fR returns 0.
+If an error occurs, -1 is returned and errno is set to indicate
+the error.
+
+.SH ERRORS
+.sp
+.LP
+\fBepoll_ctl()\fR will fail if:
+.sp
+.ne 2
+.na
+\fB\fBEBADF\fR\fR
+.ad
+.RS 10n
+\fIepfd\fR is not a valid file descriptor.
+.RE
+
+.sp
+.ne 2
+.na
+\fB\fBEFAULT\fR\fR
+.ad
+.RS 10n
+The memory associated with \fIevent\fR was not mapped.
+.RE
+
+.sp
+.ne 2
+.na
+\fB\fBEEXIST\fR\fR
+.ad
+.RS 10n
+The operation specified was \fBEPOLL_CTL_ADD\fR and the specified file
+descriptor is already associated with an event for the specified
+\fBepoll\fR(5) instance.
+.RE
+
+.sp
+.ne 2
+.na
+\fB\fBENOENT\fR\fR
+.ad
+.RS 10n
+The operation specified was \fBEPOLL_CTL_MOD\fR or \fBEPOLL_CTL_DEL\fR and
+the specified file descriptor is not associated with an event for the
+specified \fBepoll\fR(5) instance.
+.RE
+
+.sp
+.LP
+.SH NOTES
+.sp
+.LP
+
+The \fBepoll\fR(5) facility is implemented for purposes of offering
+compatibility for Linux-borne applications; native
+applications should continue to prefer using event ports via the
+\fBport_create\fR(3C), \fBport_associate\fR(3C) and \fBport_get\fR(3C)
+interfaces.  See \fBepoll\fR(5) for compatibility details and restrictions.
+.RE
+
+.SH SEE ALSO
+.sp
+.LP
+\fBepoll_create\fR(3C), \fBepoll_wait\fR(3C),
+\fBport_create\fR(3C), \fBport_associate\fR(3C), \fBport_get\fR(3C),
+\fBepoll\fR(5)
diff --git a/usr/src/man/man3c/epoll_wait.3c b/usr/src/man/man3c/epoll_wait.3c
new file mode 100644
index 0000000000..6405c30b5d
--- /dev/null
+++ b/usr/src/man/man3c/epoll_wait.3c
@@ -0,0 +1,115 @@
+'\" te
+.\"  Copyright (c) 2014, Joyent, Inc. All Rights Reserved.
+.\"  This file and its contents are supplied under the terms of the
+.\"  Common Development and Distribution License ("CDDL"), version 1.0.
+.\"  You may only use this file in accordance with the terms of version
+.\"  1.0 of the CDDL.
+.\" 
+.\"  A full copy of the text of the CDDL should have accompanied this
+.\"  source.  A copy of the CDDL is also available via the Internet at
+.\"  http://www.illumos.org/license/CDDL.
+.TH EPOLL_WAIT 3C "Apr 17, 2014"
+.SH NAME
+epoll_wait, epoll_pwait \- wait for epoll events
+.SH SYNOPSIS
+
+.LP
+.nf
+#include <sys/epoll.h>
+
+\fBint\fR \fBepoll_wait\fR(\fBint\fR \fIepfd\fR, \fBstruct epoll_event *\fR\fIevents\fR, 
+     \fBint\fR \fImaxevents\fR, \fBint\fR \fItimeout\fR);
+.fi
+
+.LP
+.nf
+\fBint\fR \fBepoll_pwait\fR(\fBint\fR \fIepfd\fR, \fBstruct epoll_event *\fR\fIevents\fR, 
+     \fBint\fR \fImaxevents\fR, \fBint\fR \fItimeout\fR,
+     \fBconst sigset_t *\fR\fIsigmask\fR);
+.fi
+
+.SH DESCRIPTION
+.sp
+.LP
+The \fBepoll_wait()\fR function waits for events on the \fBepoll\fR(5)
+instance specified by \fIepfd\fR.  The \fIevents\fR parameter must point to
+an array of \fImaxevents\fR \fIepoll_event\fR structures to be
+filled in with pending events.  The \fItimeout\fR argument specifies the
+number of milliseconds to wait for an event if none is pending.  A
+\fItimeout\fR of -1 denotes an infinite timeout.
+
+The \fBepoll_pwait()\fR is similar to \fBepoll_wait()\fR, but takes an
+additional \fIsigmask\fR argument that specifies the desired signal mask 
+when \fBepoll_pwait()\fR is blocked.  It is equivalent to atomically
+setting the signal mask, calling \fBepoll_wait()\fR, and restoring the
+signal mask upon return, and is therefore similar to the relationship
+between \fBselect\fR(3C) and \fBpselect\fR(3C).
+
+.SH RETURN VALUES
+.sp
+.LP
+Upon successful completion, \fBepoll_wait()\fR and \fBepoll_pwait()\fR return
+the number of events, or 0 if none was pending and \fItimeout\fR milliseconds
+elapsed.  If an error occurs, -1 is returned and errno is set to indicate
+the error.
+
+.SH ERRORS
+.sp
+.LP
+The \fBepoll_wait()\fR and \fBepoll_pwait()\fR functions will fail if:
+.sp
+.ne 2
+.na
+\fB\fBEBADF\fR\fR
+.ad
+.RS 10n
+\fIepfd\fR is not a valid file descriptor.
+.RE
+
+.sp
+.ne 2
+.na
+\fB\fBEFAULT\fR\fR
+.ad
+.RS 10n
+The memory associated with \fIevents\fR was not mapped or was not writable.
+.RE
+
+.sp
+.ne 2
+.na
+\fB\fBEINTR\fR\fR
+.ad
+.RS 10n
+A signal was received during the \fBepoll_wait()\fR or \fBepoll_pwait()\fR.
+.RE
+
+.sp
+.ne 2
+.na
+\fB\fBEINVAL\fR\fR
+.ad
+.RS 10n
+Either \fIepfd\fR is not a valid \fBepoll\fR(5) instance or \fImaxevents\fR
+is not greater than zero.
+.RE
+
+.sp
+.LP
+.SH NOTES
+.sp
+.LP
+
+The \fBepoll\fR(5) facility is implemented for purposes of offering
+compatibility for Linux-borne applications; native
+applications should continue to prefer using event ports via the
+\fBport_create\fR(3C), \fBport_associate\fR(3C) and \fBport_get\fR(3C)
+interfaces.  See \fBepoll\fR(5) for compatibility details and restrictions.
+.RE
+
+.SH SEE ALSO
+.sp
+.LP
+\fBepoll_create\fR(3C), \fBepoll_ctl\fR(3C),
+\fBport_create\fR(3C), \fBport_associate\fR(3C), \fBport_get\fR(3C), 
+\fBpselect\fR(3C), \fBepoll\fR(5)
diff --git a/usr/src/man/man5/Makefile b/usr/src/man/man5/Makefile
index eff831c3d4..8e805da008 100644
--- a/usr/src/man/man5/Makefile
+++ b/usr/src/man/man5/Makefile
@@ -13,6 +13,7 @@
 # Copyright 2011, Richard Lowe
 # Copyright (c) 2012 by Delphix. All rights reserved.
 # Copyright 2013 Nexenta Systems, Inc.  All rights reserved.
+# Copyright (c) 2014, Joyent, Inc.  All rights reserved.
 #
 
 include		$(SRC)/Makefile.master
@@ -41,6 +42,7 @@ MANFILES=	Intro.5			\
 		dhcp.5			\
 		dhcp_modules.5		\
 		environ.5		\
+		epoll.5			\
 		eqnchar.5		\
 		extendedFILE.5		\
 		filesystem.5		\
diff --git a/usr/src/man/man5/epoll.5 b/usr/src/man/man5/epoll.5
new file mode 100644
index 0000000000..bba13e15d2
--- /dev/null
+++ b/usr/src/man/man5/epoll.5
@@ -0,0 +1,115 @@
+'\" te
+.\"  Copyright (c) 2014, Joyent, Inc. All Rights Reserved.
+.\"  This file and its contents are supplied under the terms of the
+.\"  Common Development and Distribution License ("CDDL"), version 1.0.
+.\"  You may only use this file in accordance with the terms of version
+.\"  1.0 of the CDDL.
+.\" 
+.\"  A full copy of the text of the CDDL should have accompanied this
+.\"  source.  A copy of the CDDL is also available via the Internet at
+.\"  http://www.illumos.org/license/CDDL.
+.TH EPOLL 5 "Apr 17, 2014"
+.SH NAME
+epoll \- Linux-compatible I/O event notification facility 
+.SH SYNOPSIS
+
+.LP
+.nf
+#include <sys/epoll.h>
+.fi
+
+.SH DESCRIPTION
+.sp
+.LP
+
+\fBepoll\fR is a facility for efficient event-oriented I/O that has a 
+similar model to \fBpoll\fR(2), but does not necessitate rescanning a
+set of file descriptors to wait for an event.  \fBepoll\fR is of Linux
+origins, and this facility is designed to be binary-compatible with
+the Linux facility, including the following interfaces:
+
+.RS +4
+.TP
+.ie t \(bu
+.el o
+\fBepoll_create\fR(3C) creates an \fBepoll\fR instance, returning a file
+descriptor. It contains a size arugment which is meaningful only in as
+much as it cannot be 0.
+.RE
+.RS +4
+.TP
+.ie t \(bu
+.el o
+\fBepoll_create1\fR(3C) also creates an \fBepoll\fR instance, but eliminates
+the meaningless size argument -- replacing it instead with a flags
+argument.
+.RE
+.RS +4
+.TP
+.ie t \(bu
+.el o
+\fBepoll_ctl\fR(3C) allows file descriptors to be added
+(via \fBEPOLL_CTL_ADD\fR), deleted (via \fBEPOLL_CTL_DEL\fR) or
+modified (via \fBEPOLL_CTL_MOD\fR) with respect to the \fBepoll\fR'd set
+of file descriptors.
+.RE
+.RS +4
+.TP
+.ie t \(bu
+.el o
+\fBepoll_wait\fR(3C) fetches pending events for file descriptors added
+via \fBepoll_ctl\fR(3C), blocking the caller if no such events are pending.
+.RE
+.RS +4
+.TP
+.ie t \(bu
+.el o
+\fBepoll_pwait\fR(3C) opeates in a similar manner to \fBepoll_wait\fR(3C), but
+allows the caller to specify a signal mask to be set atomically with respect
+to waiting for events.
+.RE
+
+.sp
+.LP
+.SH NOTES
+.sp
+.LP
+
+The \fBepoll\fR facility is implemented
+for purposes of offering compatibility to and portability of Linux-borne
+applications; native applications should continue to prefer using event ports
+via the \fBport_create\fR(3C),
+\fBport_associate\fR(3C) and \fBport_getn\fR(3C) interfaces. 
+In particular, use of \fBepoll\fR in a multithreaded environment is fraught
+with peril; even when using \fBEPOLLONESHOT\fR for one-shot events,
+there are race conditions with respect to \fBclose\fR(2) that are unresolvable.
+(For more details, see the aborted effort in Linux to resolve this via the
+proposed
+\fBEPOLL_CTL_DISABLE\fR operation.)
+The event port facility -- like the BSD kqueue facility that inspired it -- 
+is designed to deal with such issues via explicit event source dissociation.
+
+While a best effort has been made to mimic the Linux semantics, there
+are some semantics that are too peculiar or ill-conceived to merit
+accommodation.  In particular, the Linux \fBepoll\fR facility will -- by
+design -- continue to generate events for closed file descriptors where/when
+the underlying file description remains open.  For example, if one were
+to \fBfork\fR(2) and subsequently close an actively \fBepoll\fR'd file
+descriptor in the parent,
+any events generated in the child on the implicitly duplicated file descriptor
+will continue to be delivered to the parent -- despite the fact that the
+parent itself no longer has any notion of the file description!
+This \fBepoll\fR facility refuses to honor
+these semantics; closing the \fBEPOLL_CTL_ADD\fR'd file descriptor 
+will always result in no further
+events being generated for that event description.
+
+.RE
+.SH SEE ALSO
+.sp
+.LP
+\fBepoll_create\fR(3C), \fBepoll_create1\fR(3C), \fBepoll_ctl\fR(3C),
+\fBepoll_wait\fR(3C), \fBepoll_pwait\fR(3C),
+\fBport_create\fR(3C), \fBport_associate\fR(3C), \fBport_dissociate\fR(3C),
+\fBport_get\fR(3C), 
+\fBpselect\fR(3C)
diff --git a/usr/src/man/man7d/poll.7d b/usr/src/man/man7d/poll.7d
index cd3db77de9..7a3292eb97 100644
--- a/usr/src/man/man7d/poll.7d
+++ b/usr/src/man/man7d/poll.7d
@@ -73,15 +73,6 @@ Pointer to \fBpollfd\fR structure.
 
 .SH DESCRIPTION
 .LP
-Note -
-.sp
-.RS 2
-The \fB/dev/poll\fR device, associated driver and corresponding manpages may be
-removed in a future Solaris release. For similar functionality in the event
-ports framework, see \fBport_create\fR(3C).
-.RE
-.sp
-.LP
 The  \fB/dev/poll\fR driver is a special driver that enables you to monitor
 multiple sets  of polled file descriptors. By using the  \fB/dev/poll\fR
 driver, you can efficiently poll large numbers of file descriptors. Access to
diff --git a/usr/src/man/man9e/chpoll.9e b/usr/src/man/man9e/chpoll.9e
index a2adaf7a9c..24e3379861 100644
--- a/usr/src/man/man9e/chpoll.9e
+++ b/usr/src/man/man9e/chpoll.9e
@@ -123,6 +123,17 @@ The same as \fBPOLLOUT\fR.
 Priority data (priority band > 0) may be written.
 .RE
 
+.sp
+.ne 2
+.na
+\fB\fBPOLLET\fR\fR
+.ad
+.RS 14n
+The desired event is to be edge-triggered; calls to \fBpollwakeup\fR(9F)
+should not be suppressed, even if the event is pending at the time of
+call to the \fBchpoll()\fR function.
+.RE
+
 .RE
 
 .sp
@@ -200,7 +211,11 @@ be called with multiple events at one time. The \fBpollwakup()\fR can be called
 regardless of whether or not the \fBchpoll()\fR entry is called; it should be
 called every time the driver detects the pollable event. The driver must not
 hold any mutex across the call to \fBpollwakeup\fR(9F) that is acquired in its
-\fBchpoll()\fR entry point, or a deadlock may result.
+\fBchpoll()\fR entry point, or a deadlock may result.  Note that if
+\fBPOLLET\fR is set in the specified events, the driver must call
+\fBpollwakeup\fR(9F) on subsequent events, even if events are pending at
+the time of the call to \fBchpoll()\fR.
+
 .RE
 .SH RETURN VALUES
 .sp
diff --git a/usr/src/uts/common/fs/fifofs/fifovnops.c b/usr/src/uts/common/fs/fifofs/fifovnops.c
index ac89e430c7..fee2924093 100644
--- a/usr/src/uts/common/fs/fifofs/fifovnops.c
+++ b/usr/src/uts/common/fs/fifofs/fifovnops.c
@@ -27,7 +27,9 @@
  * Use is subject to license terms.
  */
 
-#pragma ident	"%Z%%M%	%I%	%E% SMI"
+/*
+ * Copyright (c) 2014, Joyent, Inc.  All rights reserved.
+ */
 
 /*
  * FIFOFS file system vnode operations.  This file system
@@ -1832,17 +1834,16 @@ fifo_poll(vnode_t *vp, short events, int anyyet, short *reventsp,
 	}
 
 	/*
-	 * if we happened to get something, return
+	 * if we happened to get something and we're not edge-triggered, return
 	 */
-
-	if ((*reventsp = (short)retevents) != 0) {
+	if ((*reventsp = (short)retevents) != 0 && !(events & POLLET)) {
 		mutex_exit(&fnp->fn_lock->flk_lock);
 		return (0);
 	}
 
 	/*
-	 * If poll() has not found any events yet, set up event cell
-	 * to wake up the poll if a requested event occurs on this
+	 * If poll() has not found any events yet or we're edge-triggered, set
+	 * up event cell to wake up the poll if a requested event occurs on this
 	 * pipe/fifo.
 	 */
 	if (!anyyet) {
diff --git a/usr/src/uts/common/fs/portfs/port_vnops.c b/usr/src/uts/common/fs/portfs/port_vnops.c
index b2f5088e06..ab95c0a1f8 100644
--- a/usr/src/uts/common/fs/portfs/port_vnops.c
+++ b/usr/src/uts/common/fs/portfs/port_vnops.c
@@ -24,6 +24,10 @@
  * Use is subject to license terms.
  */
 
+/*
+ * Copyright (c) 2014, Joyent, Inc.  All rights reserved.
+ */
+
 #include <sys/types.h>
 #include <sys/vnode.h>
 #include <sys/vfs_opreg.h>
@@ -294,14 +298,10 @@ port_poll(vnode_t *vp, short events, int anyyet, short *reventsp,
 		levents |= POLLOUT;
 	levents &= events;
 	*reventsp = levents;
-	if (levents == 0) {
-		if (!anyyet) {
-			*phpp = &pp->port_pollhd;
-			portq->portq_flags |=
-			    events & POLLIN ? PORTQ_POLLIN : 0;
-			portq->portq_flags |=
-			    events & POLLOUT ? PORTQ_POLLOUT : 0;
-		}
+	if ((levents == 0 && !anyyet) || (events & POLLET)) {
+		*phpp = &pp->port_pollhd;
+		portq->portq_flags |= events & POLLIN ? PORTQ_POLLIN : 0;
+		portq->portq_flags |= events & POLLOUT ? PORTQ_POLLOUT : 0;
 	}
 	mutex_exit(&portq->portq_mutex);
 	return (0);
diff --git a/usr/src/uts/common/fs/proc/prvnops.c b/usr/src/uts/common/fs/proc/prvnops.c
index c84b9d3726..84a8ae4d31 100644
--- a/usr/src/uts/common/fs/proc/prvnops.c
+++ b/usr/src/uts/common/fs/proc/prvnops.c
@@ -21,7 +21,7 @@
 
 /*
  * Copyright (c) 1989, 2010, Oracle and/or its affiliates. All rights reserved.
- * Copyright (c) 2013, Joyent, Inc. All rights reserved.
+ * Copyright (c) 2014, Joyent, Inc. All rights reserved.
  */
 
 /*	Copyright (c) 1984,	 1986, 1987, 1988, 1989 AT&T	*/
@@ -6010,7 +6010,7 @@ prpoll(vnode_t *vp, short events, int anyyet, short *reventsp,
 	}
 
 	*reventsp = revents;
-	if (!anyyet && revents == 0) {
+	if ((!anyyet && revents == 0) || (events & POLLET)) {
 		/*
 		 * Arrange to wake up the polling lwp when
 		 * the target process/lwp stops or terminates
diff --git a/usr/src/uts/common/fs/sockfs/sockcommon_sops.c b/usr/src/uts/common/fs/sockfs/sockcommon_sops.c
index 0be628f329..d3ff264eef 100644
--- a/usr/src/uts/common/fs/sockfs/sockcommon_sops.c
+++ b/usr/src/uts/common/fs/sockfs/sockcommon_sops.c
@@ -953,6 +953,13 @@ so_poll(struct sonode *so, short events, int anyyet, short *reventsp,
 	if (!list_is_empty(&so->so_acceptq_list))
 		*reventsp |= (POLLIN|POLLRDNORM) & events;
 
+	/*
+	 * If we're looking for POLLRDHUP, indicate it if we have sent the
+	 * last rx signal for the socket.
+	 */
+	if ((events & POLLRDHUP) && (state & SS_SENTLASTREADSIG))
+		*reventsp |= POLLRDHUP;
+
 	/* Data */
 	/* so_downcalls is null for sctp */
 	if (so->so_downcalls != NULL && so->so_downcalls->sd_poll != NULL) {
@@ -988,14 +995,18 @@ so_poll(struct sonode *so, short events, int anyyet, short *reventsp,
 			*reventsp |= POLLHUP;
 	}
 
-	if (!*reventsp && !anyyet) {
+	if ((!*reventsp && !anyyet) || (events & POLLET)) {
 		/* Check for read events again, but this time under lock */
 		if (events & (POLLIN|POLLRDNORM)) {
 			mutex_enter(&so->so_lock);
 			if (SO_HAVE_DATA(so) ||
 			    !list_is_empty(&so->so_acceptq_list)) {
+				if (events & POLLET)
+					so->so_pollev |= SO_POLLEV_IN;
+
 				mutex_exit(&so->so_lock);
 				*reventsp |= (POLLIN|POLLRDNORM) & events;
+
 				return (0);
 			} else {
 				so->so_pollev |= SO_POLLEV_IN;
diff --git a/usr/src/uts/common/fs/sockfs/socknotify.c b/usr/src/uts/common/fs/sockfs/socknotify.c
index 3d5ba2a7e8..3f858afecc 100644
--- a/usr/src/uts/common/fs/sockfs/socknotify.c
+++ b/usr/src/uts/common/fs/sockfs/socknotify.c
@@ -377,7 +377,7 @@ i_so_notify_last_rx(struct sonode *so, int *pollev, int *sigev)
 		so->so_state |= SS_SENTLASTREADSIG;
 		so->so_pollev &= ~SO_POLLEV_IN;
 
-		*pollev |= POLLIN|POLLRDNORM;
+		*pollev |= POLLIN|POLLRDNORM|POLLRDHUP;
 		*sigev |= SOCKETSIG_READ;
 
 		return (1);
diff --git a/usr/src/uts/common/io/1394/targets/av1394/av1394_async.c b/usr/src/uts/common/io/1394/targets/av1394/av1394_async.c
index 94323582d6..c2c16e848b 100644
--- a/usr/src/uts/common/io/1394/targets/av1394/av1394_async.c
+++ b/usr/src/uts/common/io/1394/targets/av1394/av1394_async.c
@@ -24,7 +24,9 @@
  * Use is subject to license terms.
  */
 
-#pragma ident	"%Z%%M%	%I%	%E% SMI"
+/*
+ * Copyright (c) 2014, Joyent, Inc. All rights reserved.
+ */
 
 /*
  * av1394 asynchronous module
@@ -359,9 +361,10 @@ av1394_async_poll(av1394_inst_t *avp, short events, int anyyet, short *reventsp,
 	AV1394_TNF_ENTER(av1394_async_poll);
 
 	if (events & POLLIN) {
-		if (av1394_peekq(rq)) {
+		if (av1394_peekq(rq))
 			*reventsp |= POLLIN;
-		} else if (!anyyet) {
+
+		if ((!*reventsp && !anyyet) || (events & POLLET)) {
 			mutex_enter(&ap->a_mutex);
 			ap->a_pollevents |= POLLIN;
 			*phpp = &ap->a_pollhead;
diff --git a/usr/src/uts/common/io/devpoll.c b/usr/src/uts/common/io/devpoll.c
index 7d311aaa10..14c2f47fac 100644
--- a/usr/src/uts/common/io/devpoll.c
+++ b/usr/src/uts/common/io/devpoll.c
@@ -25,6 +25,7 @@
 
 /*
  * Copyright (c) 2012 by Delphix. All rights reserved.
+ * Copyright (c) 2014, Joyent, Inc. All rights reserved.
  */
 
 #include <sys/types.h>
@@ -45,6 +46,8 @@
 #include <sys/devpoll.h>
 #include <sys/rctl.h>
 #include <sys/resource.h>
+#include <sys/schedctl.h>
+#include <sys/epoll.h>
 
 #define	RESERVED	1
 
@@ -237,7 +240,8 @@ dpinfo(dev_info_t *dip, ddi_info_cmd_t infocmd, void *arg, void **result)
  *	 stale entries!
  */
 static int
-dp_pcache_poll(pollfd_t *pfdp, pollcache_t *pcp, nfds_t nfds, int *fdcntp)
+dp_pcache_poll(dp_entry_t *dpep, void *dpbuf,
+    pollcache_t *pcp, nfds_t nfds, int *fdcntp)
 {
 	int		start, ostart, end;
 	int		fdcnt, fd;
@@ -247,7 +251,10 @@ dp_pcache_poll(pollfd_t *pfdp, pollcache_t *pcp, nfds_t nfds, int *fdcntp)
 	boolean_t	no_wrap;
 	pollhead_t	*php;
 	polldat_t	*pdp;
+	pollfd_t	*pfdp;
+	epoll_event_t	*epoll;
 	int		error = 0;
+	short		mask = POLLRDHUP | POLLWRNORM | POLLWRBAND;
 
 	ASSERT(MUTEX_HELD(&pcp->pc_lock));
 	if (pcp->pc_bitmap == NULL) {
@@ -257,6 +264,14 @@ dp_pcache_poll(pollfd_t *pfdp, pollcache_t *pcp, nfds_t nfds, int *fdcntp)
 		 */
 		return (error);
 	}
+
+	if (dpep->dpe_flag & DP_ISEPOLLCOMPAT) {
+		pfdp = NULL;
+		epoll = (epoll_event_t *)dpbuf;
+	} else {
+		pfdp = (pollfd_t *)dpbuf;
+		epoll = NULL;
+	}
 retry:
 	start = ostart = pcp->pc_mapstart;
 	end = pcp->pc_mapend;
@@ -316,11 +331,32 @@ repoll:
 				 * polling a closed fd. Hope this will remind
 				 * user to do a POLLREMOVE.
 				 */
-				pfdp[fdcnt].fd = fd;
-				pfdp[fdcnt].revents = POLLNVAL;
-				fdcnt++;
+				if (pfdp != NULL) {
+					pfdp[fdcnt].fd = fd;
+					pfdp[fdcnt].revents = POLLNVAL;
+					fdcnt++;
+					continue;
+				}
+
+				/*
+				 * In the epoll compatibility case, we actually
+				 * perform the implicit removal to remain
+				 * closer to the epoll semantics.
+				 */
+				ASSERT(epoll != NULL);
+
+				pdp->pd_fp = NULL;
+				pdp->pd_events = 0;
+
+				if (php != NULL) {
+					pollhead_delete(php, pdp);
+					pdp->pd_php = NULL;
+				}
+
+				BT_CLEAR(pcp->pc_bitmap, fd);
 				continue;
 			}
+
 			if (fp != pdp->pd_fp) {
 				/*
 				 * user is polling on a cached fd which was
@@ -376,9 +412,59 @@ repoll:
 			}
 
 			if (revent != 0) {
-				pfdp[fdcnt].fd = fd;
-				pfdp[fdcnt].events = pdp->pd_events;
-				pfdp[fdcnt].revents = revent;
+				if (pfdp != NULL) {
+					pfdp[fdcnt].fd = fd;
+					pfdp[fdcnt].events = pdp->pd_events;
+					pfdp[fdcnt].revents = revent;
+				} else {
+					epoll_event_t *ep = &epoll[fdcnt];
+
+					ASSERT(epoll != NULL);
+					ep->data.u64 = pdp->pd_epolldata;
+
+					/*
+					 * If any of the event bits are set for
+					 * which poll and epoll representations
+					 * differ, swizzle in the native epoll
+					 * values.
+					 */
+					if (revent & mask) {
+						ep->events = (revent & ~mask) |
+						    ((revent & POLLRDHUP) ?
+						    EPOLLRDHUP : 0) |
+						    ((revent & POLLWRNORM) ?
+						    EPOLLWRNORM : 0) |
+						    ((revent & POLLWRBAND) ?
+						    EPOLLWRBAND : 0);
+					} else {
+						ep->events = revent;
+					}
+				}
+
+				/*
+				 * If POLLET is set, clear the bit in the
+				 * bitmap -- which effectively latches the
+				 * edge on a pollwakeup() from the driver.
+				 */
+				if (pdp->pd_events & POLLET)
+					BT_CLEAR(pcp->pc_bitmap, fd);
+
+				/*
+				 * If POLLONESHOT is set, perform the implicit
+				 * POLLREMOVE.
+				 */
+				if (pdp->pd_events & POLLONESHOT) {
+					pdp->pd_fp = NULL;
+					pdp->pd_events = 0;
+
+					if (php != NULL) {
+						pollhead_delete(php, pdp);
+						pdp->pd_php = NULL;
+					}
+
+					BT_CLEAR(pcp->pc_bitmap, fd);
+				}
+
 				fdcnt++;
 			} else if (php != NULL) {
 				/*
@@ -392,7 +478,7 @@ repoll:
 				 * in bitmap.
 				 */
 				if ((pdp->pd_php != NULL) &&
-				    ((pcp->pc_flag & T_POLLWAKE) == 0)) {
+				    ((pcp->pc_flag & PC_POLLWAKE) == 0)) {
 					BT_CLEAR(pcp->pc_bitmap, fd);
 				}
 				if (pdp->pd_php == NULL) {
@@ -499,7 +585,9 @@ dpwrite(dev_t dev, struct uio *uiop, cred_t *credp)
 	dp_entry_t	*dpep;
 	pollcache_t	*pcp;
 	pollfd_t	*pollfdp, *pfdp;
-	int		error;
+	dvpoll_epollfd_t *epfdp;
+	uintptr_t	limit;
+	int		error, size;
 	ssize_t		uiosize;
 	nfds_t		pollfdnum;
 	struct pollhead	*php = NULL;
@@ -515,11 +603,18 @@ dpwrite(dev_t dev, struct uio *uiop, cred_t *credp)
 	ASSERT(dpep != NULL);
 	mutex_exit(&devpoll_lock);
 	pcp = dpep->dpe_pcache;
-	if (curproc->p_pid != pcp->pc_pid) {
+
+	if (curproc->p_pid != pcp->pc_pid)
 		return (EACCES);
+
+	if (dpep->dpe_flag & DP_ISEPOLLCOMPAT) {
+		size = sizeof (dvpoll_epollfd_t);
+	} else {
+		size = sizeof (pollfd_t);
 	}
+
 	uiosize = uiop->uio_resid;
-	pollfdnum = uiosize / sizeof (pollfd_t);
+	pollfdnum = uiosize / size;
 	mutex_enter(&curproc->p_lock);
 	if (pollfdnum > (uint_t)rctl_enforced_value(
 	    rctlproc_legacy[RLIMIT_NOFILE], curproc->p_rctls, curproc)) {
@@ -534,6 +629,7 @@ dpwrite(dev_t dev, struct uio *uiop, cred_t *credp)
 	 * each polled fd to the cached set.
 	 */
 	pollfdp = kmem_alloc(uiosize, KM_SLEEP);
+	limit = (uintptr_t)pollfdp + (pollfdnum * size);
 
 	/*
 	 * Although /dev/poll uses the write(2) interface to cache fds, it's
@@ -555,9 +651,27 @@ dpwrite(dev_t dev, struct uio *uiop, cred_t *credp)
 	mutex_enter(&dpep->dpe_lock);
 	dpep->dpe_writerwait++;
 	while (dpep->dpe_refcnt != 0) {
+		/*
+		 * We need to do a bit of a dance here:  we need to drop
+		 * our dpe_lock and grab the pc_lock to broadcast the pc_cv to
+		 * kick any DP_POLL/DP_PPOLL sleepers.
+		 */
+		mutex_exit(&dpep->dpe_lock);
+		mutex_enter(&pcp->pc_lock);
+		pcp->pc_flag |= PC_WRITEWANTED;
+		cv_broadcast(&pcp->pc_cv);
+		mutex_exit(&pcp->pc_lock);
+		mutex_enter(&dpep->dpe_lock);
+
+		if (dpep->dpe_refcnt == 0)
+			break;
+
 		if (!cv_wait_sig_swap(&dpep->dpe_cv, &dpep->dpe_lock)) {
 			dpep->dpe_writerwait--;
 			mutex_exit(&dpep->dpe_lock);
+			mutex_enter(&pcp->pc_lock);
+			pcp->pc_flag &= ~PC_WRITEWANTED;
+			mutex_exit(&pcp->pc_lock);
 			kmem_free(pollfdp, uiosize);
 			return (set_errno(EINTR));
 		}
@@ -565,13 +679,17 @@ dpwrite(dev_t dev, struct uio *uiop, cred_t *credp)
 	dpep->dpe_writerwait--;
 	dpep->dpe_flag |= DP_WRITER_PRESENT;
 	dpep->dpe_refcnt++;
+
 	mutex_exit(&dpep->dpe_lock);
 
 	mutex_enter(&pcp->pc_lock);
+	pcp->pc_flag &= ~PC_WRITEWANTED;
+
 	if (pcp->pc_bitmap == NULL) {
 		pcache_create(pcp, pollfdnum);
 	}
-	for (pfdp = pollfdp; pfdp < pollfdp + pollfdnum; pfdp++) {
+	for (pfdp = pollfdp; (uintptr_t)pfdp < limit;
+	    pfdp = (pollfd_t *)((uintptr_t)pfdp + size)) {
 		fd = pfdp->fd;
 		if ((uint_t)fd >= P_FINFO(curproc)->fi_nfiles)
 			continue;
@@ -580,9 +698,27 @@ dpwrite(dev_t dev, struct uio *uiop, cred_t *credp)
 			if (pdp == NULL) {
 				pdp = pcache_alloc_fd(0);
 				pdp->pd_fd = fd;
+
+				if (dpep->dpe_flag & DP_ISEPOLLCOMPAT) {
+					/* LINTED pointer alignment */
+					epfdp = (dvpoll_epollfd_t *)pfdp;
+					pdp->pd_epolldata = epfdp->dpep_data;
+				}
+
 				pdp->pd_pcache = pcp;
 				pcache_insert_fd(pcp, pdp, pollfdnum);
+			} else {
+				if ((dpep->dpe_flag & DP_ISEPOLLCOMPAT) &&
+				    pdp->pd_fp != NULL) {
+					/*
+					 * epoll semantics demand that we error
+					 * out in this case.
+					 */
+					error = EEXIST;
+					break;
+				}
 			}
+
 			ASSERT(pdp->pd_fd == fd);
 			ASSERT(pdp->pd_pcache == pcp);
 			if (fd >= pcp->pc_mapsize) {
@@ -665,7 +801,17 @@ dpwrite(dev_t dev, struct uio *uiop, cred_t *credp)
 			}
 			releasef(fd);
 		} else {
-			if (pdp == NULL) {
+			if (pdp == NULL || pdp->pd_fp == NULL) {
+				if (dpep->dpe_flag & DP_ISEPOLLCOMPAT) {
+					/*
+					 * As with the add case (above), epoll
+					 * semantics demand that we error out
+					 * in this case.
+					 */
+					error = ENOENT;
+					break;
+				}
+
 				continue;
 			}
 			ASSERT(pdp->pd_fd == fd);
@@ -690,6 +836,17 @@ dpwrite(dev_t dev, struct uio *uiop, cred_t *credp)
 	return (error);
 }
 
+#define	DP_SIGMASK_RESTORE(ksetp) {					\
+	if (ksetp != NULL) {						\
+		mutex_enter(&p->p_lock);				\
+		if (lwp->lwp_cursig == 0) {				\
+			t->t_hold = lwp->lwp_sigoldmask;		\
+			t->t_flag &= ~T_TOMASK;				\
+		}							\
+		mutex_exit(&p->p_lock);					\
+	}								\
+}
+
 /*ARGSUSED*/
 static int
 dpioctl(dev_t dev, int cmd, intptr_t arg, int mode, cred_t *credp, int *rvalp)
@@ -701,7 +858,7 @@ dpioctl(dev_t dev, int cmd, intptr_t arg, int mode, cred_t *credp, int *rvalp)
 	int		error = 0;
 	STRUCT_DECL(dvpoll, dvpoll);
 
-	if (cmd == DP_POLL) {
+	if (cmd == DP_POLL || cmd == DP_PPOLL) {
 		/* do this now, before we sleep on DP_WRITER_PRESENT */
 		now = gethrtime();
 	}
@@ -717,6 +874,27 @@ dpioctl(dev_t dev, int cmd, intptr_t arg, int mode, cred_t *credp, int *rvalp)
 		return (EACCES);
 
 	mutex_enter(&dpep->dpe_lock);
+
+	if (cmd == DP_EPOLLCOMPAT) {
+		if (dpep->dpe_refcnt != 0) {
+			/*
+			 * We can't turn on epoll compatibility while there
+			 * are outstanding operations.
+			 */
+			mutex_exit(&dpep->dpe_lock);
+			return (EBUSY);
+		}
+
+		/*
+		 * epoll compatibility is a one-way street: there's no way
+		 * to turn it off for a particular open.
+		 */
+		dpep->dpe_flag |= DP_ISEPOLLCOMPAT;
+		mutex_exit(&dpep->dpe_lock);
+
+		return (0);
+	}
+
 	while ((dpep->dpe_flag & DP_WRITER_PRESENT) ||
 	    (dpep->dpe_writerwait != 0)) {
 		if (!cv_wait_sig_swap(&dpep->dpe_cv, &dpep->dpe_lock)) {
@@ -729,15 +907,36 @@ dpioctl(dev_t dev, int cmd, intptr_t arg, int mode, cred_t *credp, int *rvalp)
 
 	switch (cmd) {
 	case	DP_POLL:
+	case	DP_PPOLL:
 	{
 		pollstate_t	*ps;
 		nfds_t		nfds;
 		int		fdcnt = 0;
+		size_t		size, fdsize, dpsize;
 		hrtime_t	deadline = 0;
+		k_sigset_t	*ksetp = NULL;
+		k_sigset_t	kset;
+		sigset_t	set;
+		kthread_t	*t = curthread;
+		klwp_t		*lwp = ttolwp(t);
+		struct proc	*p = ttoproc(curthread);
 
 		STRUCT_INIT(dvpoll, mode);
-		error = copyin((caddr_t)arg, STRUCT_BUF(dvpoll),
-		    STRUCT_SIZE(dvpoll));
+
+		/*
+		 * The dp_setp member is only required/consumed for DP_PPOLL,
+		 * which otherwise uses the same structure as DP_POLL.
+		 */
+		if (cmd == DP_POLL) {
+			dpsize = (uintptr_t)STRUCT_FADDR(dvpoll, dp_setp) -
+			    (uintptr_t)STRUCT_FADDR(dvpoll, dp_fds);
+		} else {
+			ASSERT(cmd == DP_PPOLL);
+			dpsize = STRUCT_SIZE(dvpoll);
+		}
+
+		error = copyin((caddr_t)arg, STRUCT_BUF(dvpoll), dpsize);
+
 		if (error) {
 			DP_REFRELE(dpep);
 			return (EFAULT);
@@ -755,6 +954,52 @@ dpioctl(dev_t dev, int cmd, intptr_t arg, int mode, cred_t *credp, int *rvalp)
 			deadline += now;
 		}
 
+		if (cmd == DP_PPOLL) {
+			void *setp = STRUCT_FGETP(dvpoll, dp_setp);
+
+			if (setp != NULL) {
+				if (copyin(setp, &set, sizeof (set))) {
+					DP_REFRELE(dpep);
+					return (EFAULT);
+				}
+
+				sigutok(&set, &kset);
+				ksetp = &kset;
+
+				mutex_enter(&p->p_lock);
+				schedctl_finish_sigblock(t);
+				lwp->lwp_sigoldmask = t->t_hold;
+				t->t_hold = *ksetp;
+				t->t_flag |= T_TOMASK;
+
+				/*
+				 * Like ppoll() with a non-NULL sigset, we'll
+				 * call cv_reltimedwait_sig() just to check for
+				 * signals.  This call will return immediately
+				 * with either 0 (signalled) or -1 (no signal).
+				 * There are some conditions whereby we can
+				 * get 0 from cv_reltimedwait_sig() without
+				 * a true signal (e.g., a directed stop), so
+				 * we restore our signal mask in the unlikely
+				 * event that lwp_cursig is 0.
+				 */
+				if (!cv_reltimedwait_sig(&t->t_delay_cv,
+				    &p->p_lock, 0, TR_CLOCK_TICK)) {
+					if (lwp->lwp_cursig == 0) {
+						t->t_hold = lwp->lwp_sigoldmask;
+						t->t_flag &= ~T_TOMASK;
+					}
+
+					mutex_exit(&p->p_lock);
+
+					DP_REFRELE(dpep);
+					return (EINTR);
+				}
+
+				mutex_exit(&p->p_lock);
+			}
+		}
+
 		if ((nfds = STRUCT_FGET(dvpoll, dp_nfds)) == 0) {
 			/*
 			 * We are just using DP_POLL to sleep, so
@@ -762,17 +1007,29 @@ dpioctl(dev_t dev, int cmd, intptr_t arg, int mode, cred_t *credp, int *rvalp)
 			 * Do not check for signals if we have a zero timeout.
 			 */
 			DP_REFRELE(dpep);
-			if (deadline == 0)
+			if (deadline == 0) {
+				DP_SIGMASK_RESTORE(ksetp);
 				return (0);
+			}
+
 			mutex_enter(&curthread->t_delay_lock);
 			while ((error =
 			    cv_timedwait_sig_hrtime(&curthread->t_delay_cv,
 			    &curthread->t_delay_lock, deadline)) > 0)
 				continue;
 			mutex_exit(&curthread->t_delay_lock);
+
+			DP_SIGMASK_RESTORE(ksetp);
+
 			return (error == 0 ? EINTR : 0);
 		}
 
+		if (dpep->dpe_flag & DP_ISEPOLLCOMPAT) {
+			size = nfds * (fdsize = sizeof (epoll_event_t));
+		} else {
+			size = nfds * (fdsize = sizeof (pollfd_t));
+		}
+
 		/*
 		 * XXX It would be nice not to have to alloc each time, but it
 		 * requires another per thread structure hook. This can be
@@ -782,8 +1039,7 @@ dpioctl(dev_t dev, int cmd, intptr_t arg, int mode, cred_t *credp, int *rvalp)
 			curthread->t_pollstate = pollstate_create();
 			ps = curthread->t_pollstate;
 		}
-		if (ps->ps_dpbufsize < nfds) {
-			struct proc *p = ttoproc(curthread);
+		if (ps->ps_dpbufsize < size) {
 			/*
 			 * The maximum size should be no large than
 			 * current maximum open file count.
@@ -792,27 +1048,28 @@ dpioctl(dev_t dev, int cmd, intptr_t arg, int mode, cred_t *credp, int *rvalp)
 			if (nfds > p->p_fno_ctl) {
 				mutex_exit(&p->p_lock);
 				DP_REFRELE(dpep);
+				DP_SIGMASK_RESTORE(ksetp);
 				return (EINVAL);
 			}
 			mutex_exit(&p->p_lock);
-			kmem_free(ps->ps_dpbuf, sizeof (pollfd_t) *
-			    ps->ps_dpbufsize);
-			ps->ps_dpbuf = kmem_zalloc(sizeof (pollfd_t) *
-			    nfds, KM_SLEEP);
-			ps->ps_dpbufsize = nfds;
+			kmem_free(ps->ps_dpbuf, ps->ps_dpbufsize);
+			ps->ps_dpbuf = kmem_zalloc(size, KM_SLEEP);
+			ps->ps_dpbufsize = size;
 		}
 
 		mutex_enter(&pcp->pc_lock);
 		for (;;) {
-			pcp->pc_flag = 0;
-			error = dp_pcache_poll(ps->ps_dpbuf, pcp, nfds, &fdcnt);
+			pcp->pc_flag &= ~PC_POLLWAKE;
+
+			error = dp_pcache_poll(dpep, ps->ps_dpbuf,
+			    pcp, nfds, &fdcnt);
 			if (fdcnt > 0 || error != 0)
 				break;
 
 			/*
 			 * A pollwake has happened since we polled cache.
 			 */
-			if (pcp->pc_flag & T_POLLWAKE)
+			if (pcp->pc_flag & PC_POLLWAKE)
 				continue;
 
 			/*
@@ -822,8 +1079,40 @@ dpioctl(dev_t dev, int cmd, intptr_t arg, int mode, cred_t *credp, int *rvalp)
 				/* immediate timeout; do not check signals */
 				break;
 			}
-			error = cv_timedwait_sig_hrtime(&pcp->pc_cv,
-			    &pcp->pc_lock, deadline);
+
+			if (!(pcp->pc_flag & PC_WRITEWANTED)) {
+				error = cv_timedwait_sig_hrtime(&pcp->pc_cv,
+				    &pcp->pc_lock, deadline);
+			} else {
+				error = 1;
+			}
+
+			if (error > 0 && (pcp->pc_flag & PC_WRITEWANTED)) {
+				/*
+				 * We've been kicked off of our cv because a
+				 * writer wants in.  We're going to drop our
+				 * reference count and then wait until the
+				 * writer is gone -- at which point we'll
+				 * reacquire the pc_lock and call into
+				 * dp_pcache_poll() to get the updated state.
+				 */
+				mutex_exit(&pcp->pc_lock);
+
+				mutex_enter(&dpep->dpe_lock);
+				dpep->dpe_refcnt--;
+				cv_broadcast(&dpep->dpe_cv);
+
+				while ((dpep->dpe_flag & DP_WRITER_PRESENT) ||
+				    (dpep->dpe_writerwait != 0)) {
+					error = cv_wait_sig_swap(&dpep->dpe_cv,
+					    &dpep->dpe_lock);
+				}
+
+				dpep->dpe_refcnt++;
+				mutex_exit(&dpep->dpe_lock);
+				mutex_enter(&pcp->pc_lock);
+			}
+
 			/*
 			 * If we were awakened by a signal or timeout
 			 * then break the loop, else poll again.
@@ -837,9 +1126,11 @@ dpioctl(dev_t dev, int cmd, intptr_t arg, int mode, cred_t *credp, int *rvalp)
 		}
 		mutex_exit(&pcp->pc_lock);
 
+		DP_SIGMASK_RESTORE(ksetp);
+
 		if (error == 0 && fdcnt > 0) {
-			if (copyout(ps->ps_dpbuf, STRUCT_FGETP(dvpoll,
-			    dp_fds), sizeof (pollfd_t) * fdcnt)) {
+			if (copyout(ps->ps_dpbuf,
+			    STRUCT_FGETP(dvpoll, dp_fds), fdcnt * fdsize)) {
 				DP_REFRELE(dpep);
 				return (EFAULT);
 			}
diff --git a/usr/src/uts/common/io/ksocket/ksocket.c b/usr/src/uts/common/io/ksocket/ksocket.c
index 49ca6f0475..8944fcbff3 100644
--- a/usr/src/uts/common/io/ksocket/ksocket.c
+++ b/usr/src/uts/common/io/ksocket/ksocket.c
@@ -22,6 +22,7 @@
 /*
  * Copyright 2011 Nexenta Systems, Inc.  All rights reserved.
  * Copyright (c) 2008, 2010, Oracle and/or its affiliates. All rights reserved.
+ * Copyright (c) 2014, Joyent, Inc. All rights reserved.
  */
 
 #include <sys/file.h>
@@ -820,7 +821,7 @@ ksocket_spoll(ksocket_t ks, int timo, short events, short *revents,
 		if (error != 0 || *revents != 0)
 			break;
 
-		if (pcp->pc_flag & T_POLLWAKE)
+		if (pcp->pc_flag & PC_POLLWAKE)
 			continue;
 
 		if (timo == -1) {
diff --git a/usr/src/uts/common/io/usb/usba/usba_ugen.c b/usr/src/uts/common/io/usb/usba/usba_ugen.c
index cb20c24270..5852e40799 100644
--- a/usr/src/uts/common/io/usb/usba/usba_ugen.c
+++ b/usr/src/uts/common/io/usb/usba/usba_ugen.c
@@ -23,6 +23,10 @@
  */
 
 /*
+ * Copyright (c) 2014, Joyent, Inc. All rights reserved.
+ */
+
+/*
  * UGEN: USB Generic Driver support code
  *
  * This code provides entry points called by the ugen driver or other
@@ -1082,7 +1086,10 @@ usb_ugen_poll(usb_ugen_hdl_t usb_ugen_hdl, dev_t dev, short events,
 				    ((epp->ep_state &
 				    UGEN_EP_STATE_INTR_IN_POLLING_ON) == 0)) {
 					*reventsp |= POLLIN;
-				} else if (!anyyet) {
+				}
+
+				if ((!*reventsp && !anyyet) ||
+				    (events & POLLET)) {
 					*phpp = &epp->ep_pollhead;
 					epp->ep_state |=
 					    UGEN_EP_STATE_INTR_IN_POLL_PENDING;
@@ -1101,7 +1108,10 @@ usb_ugen_poll(usb_ugen_hdl_t usb_ugen_hdl, dev_t dev, short events,
 				    ((epp->ep_state &
 				    UGEN_EP_STATE_ISOC_IN_POLLING_ON) == 0)) {
 					*reventsp |= POLLIN;
-				} else if (!anyyet) {
+				}
+
+				if ((!*reventsp && !anyyet) ||
+				    (events & POLLET)) {
 					*phpp = &epp->ep_pollhead;
 					epp->ep_state |=
 					    UGEN_EP_STATE_ISOC_IN_POLL_PENDING;
@@ -1115,9 +1125,10 @@ usb_ugen_poll(usb_ugen_hdl_t usb_ugen_hdl, dev_t dev, short events,
 
 			break;
 		case UGEN_MINOR_DEV_STAT_NODE:
-			if (ugenp->ug_ds.dev_stat & UGEN_DEV_STATUS_CHANGED) {
+			if (ugenp->ug_ds.dev_stat & UGEN_DEV_STATUS_CHANGED)
 				*reventsp |= POLLIN;
-			} else if (!anyyet) {
+
+			if ((!*reventsp && !anyyet) || (events & POLLET)) {
 				*phpp = &ugenp->ug_ds.dev_pollhead;
 				ugenp->ug_ds.dev_stat |=
 				    UGEN_DEV_STATUS_POLL_PENDING;
@@ -1131,9 +1142,10 @@ usb_ugen_poll(usb_ugen_hdl_t usb_ugen_hdl, dev_t dev, short events,
 			break;
 		}
 	} else {
-		if (ugenp->ug_ds.dev_stat & UGEN_DEV_STATUS_CHANGED) {
+		if (ugenp->ug_ds.dev_stat & UGEN_DEV_STATUS_CHANGED)
 			*reventsp |= POLLHUP|POLLIN;
-		} else if (!anyyet) {
+
+		if ((!*reventsp && !anyyet) || (events & POLLET)) {
 			*phpp = &ugenp->ug_ds.dev_pollhead;
 			ugenp->ug_ds.dev_stat |=
 			    UGEN_DEV_STATUS_POLL_PENDING;
diff --git a/usr/src/uts/common/os/streamio.c b/usr/src/uts/common/os/streamio.c
index f9df89923f..4c99f92411 100644
--- a/usr/src/uts/common/os/streamio.c
+++ b/usr/src/uts/common/os/streamio.c
@@ -24,6 +24,7 @@
 
 /*
  * Copyright (c) 1988, 2010, Oracle and/or its affiliates. All rights reserved.
+ * Copyright (c) 2014, Joyent, Inc. All rights reserved.
  */
 
 #include <sys/types.h>
@@ -8312,7 +8313,7 @@ chkrd:
 	}
 
 	*reventsp = (short)retevents;
-	if (retevents) {
+	if (retevents && !(events & POLLET)) {
 		if (headlocked)
 			mutex_exit(&stp->sd_lock);
 		return (0);
diff --git a/usr/src/uts/common/sys/Makefile b/usr/src/uts/common/sys/Makefile
index 0a72d3d882..08b2488b97 100644
--- a/usr/src/uts/common/sys/Makefile
+++ b/usr/src/uts/common/sys/Makefile
@@ -22,6 +22,7 @@
 # Copyright (c) 1989, 2010, Oracle and/or its affiliates. All rights reserved.
 # Copyright 2013, Joyent, Inc. All rights reserved.
 # Copyright 2013 Garrett D'Amore <garrett@damore.org>
+# Copyright (c) 2014, Joyent, Inc. All rights reserved.
 #
 
 include $(SRC)/uts/Makefile.uts
@@ -212,6 +213,7 @@ CHKHDRS=			\
 	emul64cmd.h		\
 	emul64var.h		\
 	epm.h			\
+	epoll.h			\
 	errno.h			\
 	errorq.h		\
 	errorq_impl.h		\
diff --git a/usr/src/uts/common/sys/devpoll.h b/usr/src/uts/common/sys/devpoll.h
index 36c815c69f..4e4c76d9b0 100644
--- a/usr/src/uts/common/sys/devpoll.h
+++ b/usr/src/uts/common/sys/devpoll.h
@@ -24,11 +24,13 @@
  * All rights reserved.
  */
 
+/*
+ * Copyright (c) 2014, Joyent, Inc. All rights reserved.
+ */
+
 #ifndef	_SYS_DEVPOLL_H
 #define	_SYS_DEVPOLL_H
 
-#pragma ident	"%Z%%M%	%I%	%E% SMI"
-
 #include <sys/poll_impl.h>
 #include <sys/types32.h>
 
@@ -39,8 +41,10 @@ extern "C" {
 
 /* /dev/poll ioctl */
 #define		DPIOC	(0xD0 << 8)
-#define	DP_POLL		(DPIOC | 1)	/* poll on fds in cached in /dev/poll */
+#define	DP_POLL		(DPIOC | 1)	/* poll on fds cached via /dev/poll */
 #define	DP_ISPOLLED	(DPIOC | 2)	/* is this fd cached in /dev/poll */
+#define	DP_PPOLL	(DPIOC | 3)	/* ppoll on fds cached via /dev/poll */
+#define	DP_EPOLLCOMPAT	(DPIOC | 4)	/* turn on epoll compatibility */
 
 #define	DEVPOLLSIZE	1000		/* /dev/poll table size increment */
 
@@ -51,14 +55,21 @@ typedef struct dvpoll {
 	pollfd_t	*dp_fds;	/* pollfd array */
 	nfds_t		dp_nfds;	/* num of pollfd's in dp_fds[] */
 	int		dp_timeout;	/* time out in milisec */
+	sigset_t	*dp_setp;	/* sigset, if any */
 } dvpoll_t;
 
 typedef struct dvpoll32 {
 	caddr32_t	dp_fds;		/* pollfd array */
 	uint32_t	dp_nfds;	/* num of pollfd's in dp_fds[] */
 	int32_t		dp_timeout;	/* time out in milisec */
+	caddr32_t	dp_setp;	/* sigset, if any */
 } dvpoll32_t;
 
+typedef struct dvpoll_epollfd {
+	pollfd_t	dpep_pollfd;	/* must be first member */
+	uint64_t	dpep_data;	/* data payload */
+} dvpoll_epollfd_t;
+
 #ifdef _KERNEL
 
 typedef struct dp_entry {
@@ -71,6 +82,7 @@ typedef struct dp_entry {
 } dp_entry_t;
 
 #define	DP_WRITER_PRESENT	0x1	/* a write is in progress */
+#define	DP_ISEPOLLCOMPAT	0x2	/* epoll compatibility mode */
 
 #define	DP_REFRELE(dpep) {			\
 	mutex_enter(&(dpep)->dpe_lock);		\
diff --git a/usr/src/uts/common/sys/epoll.h b/usr/src/uts/common/sys/epoll.h
new file mode 100644
index 0000000000..9b1c37f0dd
--- /dev/null
+++ b/usr/src/uts/common/sys/epoll.h
@@ -0,0 +1,88 @@
+/*
+ * This file and its contents are supplied under the terms of the
+ * Common Development and Distribution License ("CDDL"), version 1.0.
+ * You may only use this file in accordance with the terms of version
+ * 1.0 of the CDDL.
+ *
+ * A full copy of the text of the CDDL should have accompanied this
+ * source.  A copy of the CDDL is also available via the Internet at
+ * http://www.illumos.org/license/CDDL.
+ */
+
+/*
+ * Copyright (c) 2014, Joyent, Inc.  All rights reserved.
+ */
+
+#ifndef _SYS_EPOLL_H
+#define	_SYS_EPOLL_H
+
+#include <sys/poll.h>
+
+#ifdef	__cplusplus
+extern "C" {
+#endif
+
+typedef union epoll_data {
+	void		*ptr;
+	int		fd;
+	uint32_t	u32;
+	uint64_t	u64;
+} epoll_data_t;
+
+#if _LONG_LONG_ALIGNMENT == 8 && _LONG_LONG_ALIGNMENT_32 == 4
+#pragma pack(4)
+#endif
+
+typedef struct epoll_event {
+	uint32_t	events;		/* events */
+	epoll_data_t	data;		/* user-specified data */
+} epoll_event_t;
+
+#if _LONG_LONG_ALIGNMENT == 8 && _LONG_LONG_ALIGNMENT_32 == 4
+#pragma pack()
+#endif
+
+/*
+ * Define the EPOLL* constants in terms of their poll(2)/poll(7) equivalents.
+ * Note that the values match the equivalents in Linux to allow for any binary
+ * compatibility layers to not need to translate them.
+ */
+#define	EPOLLIN		0x0001
+#define	EPOLLPRI	0x0002
+#define	EPOLLOUT	0x0004
+#define	EPOLLRDNORM	0x0040
+#define	EPOLLRDBAND	0x0080
+#define	EPOLLWRNORM	0x0100
+#define	EPOLLWRBAND	0x0200
+#define	EPOLLMSG	0x0400		/* not used */
+#define	EPOLLERR	0x0008
+#define	EPOLLHUP	0x0010
+#define	EPOLLRDHUP	0x2000
+
+#define	EPOLLWAKEUP	(1UL << 29)	/* no meaning; silently ignored */
+#define	EPOLLONESHOT	(1UL << 30)	/* translated to POLLONESHOT */
+#define	EPOLLET		(1UL << 31)	/* translated to POLLET */
+
+#define	EPOLL_CTL_ADD	1
+#define	EPOLL_CTL_DEL	2
+#define	EPOLL_CTL_MOD	3
+
+#define	EPOLL_CLOEXEC	02000000
+
+#if !defined(_KERNEL)
+
+extern int epoll_create(int size);
+extern int epoll_create1(int flags);
+extern int epoll_ctl(int epfd, int op, int fd, struct epoll_event *event);
+extern int epoll_wait(int epfd, struct epoll_event *events,
+    int maxevents, int timeout);
+extern int epoll_pwait(int epfd, struct epoll_event *events,
+    int maxevents, int timeout, const sigset_t *sigmask);
+
+#endif /* !_KERNEL */
+
+#ifdef	__cplusplus
+}
+#endif
+
+#endif	/* _SYS_EPOLL_H */
diff --git a/usr/src/uts/common/sys/poll.h b/usr/src/uts/common/sys/poll.h
index b1e222a3a3..4e3673753e 100644
--- a/usr/src/uts/common/sys/poll.h
+++ b/usr/src/uts/common/sys/poll.h
@@ -28,11 +28,13 @@
  * All rights reserved.
  */
 
+/*
+ * Copyright (c) 2014, Joyent, Inc. All rights reserved.
+ */
+
 #ifndef _SYS_POLL_H
 #define	_SYS_POLL_H
 
-#pragma ident	"%Z%%M%	%I%	%E% SMI"	/* SVr4.0 11.9 */
-
 #ifdef	__cplusplus
 extern "C" {
 #endif
@@ -59,6 +61,7 @@ typedef unsigned long	nfds_t;
 #define	POLLWRNORM	POLLOUT
 #define	POLLRDBAND	0x0080		/* out-of-band data is readable */
 #define	POLLWRBAND	0x0100		/* out-of-band data is writeable */
+#define	POLLRDHUP	0x4000		/* read-side hangup */
 
 #define	POLLNORM	POLLRDNORM
 
@@ -70,7 +73,13 @@ typedef unsigned long	nfds_t;
 #define	POLLHUP		0x0010		/* fd has been hung up on */
 #define	POLLNVAL	0x0020		/* invalid pollfd entry */
 
-#define	POLLREMOVE	0x0800	/* remove a cached poll fd from /dev/poll */
+/*
+ * These events will never be specified in revents, but may be specified in
+ * events to control /dev/poll behavior.
+ */
+#define	POLLREMOVE	0x0800		/* remove cached /dev/poll fd */
+#define	POLLONESHOT	0x1000		/* /dev/poll should one-shot this fd */
+#define	POLLET		0x2000		/* edge-triggered /dev/poll fd */
 
 #ifdef _KERNEL
 
diff --git a/usr/src/uts/common/sys/poll_impl.h b/usr/src/uts/common/sys/poll_impl.h
index ede99d0df2..2e866ec4d4 100644
--- a/usr/src/uts/common/sys/poll_impl.h
+++ b/usr/src/uts/common/sys/poll_impl.h
@@ -24,11 +24,13 @@
  * Use is subject to license terms.
  */
 
+/*
+ * Copyright (c) 2014, Joyent, Inc. All rights reserved.
+ */
+
 #ifndef _SYS_POLL_IMPL_H
 #define	_SYS_POLL_IMPL_H
 
-#pragma ident	"%Z%%M%	%I%	%E% SMI"
-
 /*
  * Caching Poll Subsystem:
  *
@@ -160,6 +162,7 @@ typedef struct polldat {
 	int		pd_nsets;	/* num of xref sets, used by poll(2) */
 	xref_t		*pd_ref;	/* ptr to xref info, 1 for each set */
 	struct port_kevent *pd_portev;	/* associated port event struct */
+	uint64_t	pd_epolldata;	/* epoll data, if any */
 } polldat_t;
 
 /*
@@ -187,7 +190,8 @@ typedef struct pollcache {
 } pollcache_t;
 
 /* pc_flag */
-#define	T_POLLWAKE	0x02	/* pollwakeup() occurred */
+#define	PC_POLLWAKE	0x02	/* pollwakeup() occurred */
+#define	PC_WRITEWANTED	0x04	/* writer wishes to modify the pollcache_t */
 
 #if defined(_KERNEL)
 /*
diff --git a/usr/src/uts/common/syscall/poll.c b/usr/src/uts/common/syscall/poll.c
index 7f37529941..c33156a4fc 100644
--- a/usr/src/uts/common/syscall/poll.c
+++ b/usr/src/uts/common/syscall/poll.c
@@ -29,6 +29,7 @@
 
 /*
  * Copyright (c) 2012 by Delphix. All rights reserved.
+ * Copyright (c) 2014, Joyent, Inc. All rights reserved.
  */
 
 /*
@@ -525,13 +526,13 @@ poll_common(pollfd_t *fds, nfds_t nfds, timespec_t *tsp, k_sigset_t *ksetp)
 		}
 
 		/*
-		 * If T_POLLWAKE is set, a pollwakeup() was performed on
+		 * If PC_POLLWAKE is set, a pollwakeup() was performed on
 		 * one of the file descriptors.  This can happen only if
 		 * one of the VOP_POLL() functions dropped pcp->pc_lock.
 		 * The only current cases of this is in procfs (prpoll())
 		 * and STREAMS (strpoll()).
 		 */
-		if (pcp->pc_flag & T_POLLWAKE)
+		if (pcp->pc_flag & PC_POLLWAKE)
 			continue;
 
 		/*
@@ -886,9 +887,9 @@ retry:
 }
 
 /*
- * This function is called to inform a thread that
- * an event being polled for has occurred.
- * The pollstate lock on the thread should be held on entry.
+ * This function is called to inform a thread (or threads) that an event being
+ * polled on has occurred.  The pollstate lock on the thread should be held
+ * on entry.
  */
 void
 pollnotify(pollcache_t *pcp, int fd)
@@ -896,8 +897,8 @@ pollnotify(pollcache_t *pcp, int fd)
 	ASSERT(fd < pcp->pc_mapsize);
 	ASSERT(MUTEX_HELD(&pcp->pc_lock));
 	BT_SET(pcp->pc_bitmap, fd);
-	pcp->pc_flag |= T_POLLWAKE;
-	cv_signal(&pcp->pc_cv);
+	pcp->pc_flag |= PC_POLLWAKE;
+	cv_broadcast(&pcp->pc_cv);
 }
 
 /*
@@ -2024,7 +2025,7 @@ retry:
 				 */
 				if ((pdp->pd_php != NULL) &&
 				    (pollfdp[entry].events == pdp->pd_events) &&
-				    ((pcp->pc_flag & T_POLLWAKE) == 0)) {
+				    ((pcp->pc_flag & PC_POLLWAKE) == 0)) {
 					BT_CLEAR(pcp->pc_bitmap, fd);
 				}
 				/*
@@ -2251,7 +2252,7 @@ pollstate_destroy(pollstate_t *ps)
 	pcacheset_destroy(ps->ps_pcacheset, ps->ps_nsets);
 	ps->ps_pcacheset = NULL;
 	if (ps->ps_dpbuf != NULL) {
-		kmem_free(ps->ps_dpbuf, ps->ps_dpbufsize * sizeof (pollfd_t));
+		kmem_free(ps->ps_dpbuf, ps->ps_dpbufsize);
 		ps->ps_dpbuf = NULL;
 	}
 	mutex_destroy(&ps->ps_lock);
diff --git a/usr/src/uts/sun4v/io/vcc.c b/usr/src/uts/sun4v/io/vcc.c
index feeaf03e8f..85f722e467 100644
--- a/usr/src/uts/sun4v/io/vcc.c
+++ b/usr/src/uts/sun4v/io/vcc.c
@@ -24,6 +24,9 @@
  * Use is subject to license terms.
  */
 
+/*
+ * Copyright (c) 2014, Joyent, Inc. All rights reserved.
+ */
 
 #include <sys/types.h>
 #include <sys/file.h>
@@ -2456,7 +2459,7 @@ vcc_chpoll(dev_t dev, short events, int anyyet,  short *reventsp,
 		*reventsp |= (events & POLLIN);
 	}
 
-	if (((*reventsp) == 0) && (!anyyet)) {
+	if ((((*reventsp) == 0) && (!anyyet)) || (events & POLLET)) {
 		*phpp = &vport->poll;
 		if (events & POLLIN) {
 			mutex_enter(&vport->lock);