OS-2432 tail -f/-F missing some truncation events

OS-2433 tail -f/-F can produce duplicate output after a truncation

13 files changed, 154 insertions, 25 deletions

diff --git a/usr/src/cmd/tail/forward.c b/usr/src/cmd/tail/forward.c
index 54ac882ec8..8b87dfff06 100644
--- a/usr/src/cmd/tail/forward.c
+++ b/usr/src/cmd/tail/forward.c
@@ -328,25 +328,31 @@ associate(file_info_t *file, boolean_t assoc, port_event_t *ev)
 		/*
 		 * To assure that we cannot possibly drop a FILE_TRUNC event,
 		 * we have two different PORT_SOURCE_FILE associations with the
-		 * port:  one to get either FILE_MODIFIED or FILE_TRUNC events,
-		 * and another to get only FILE_TRUNC events.  This assures that
-		 * we always have an active association for FILE_TRUNC events.
-		 * (Without the second association, there is a window where a
-		 * file truncation could occur after a port_get() but before
-		 * the port_associate() call to re-associate the object --
-		 * resulting in loss of the truncation event.)
+		 * port:  one to get only FILE_MODIFIED events and another to
+		 * get only FILE_TRUNC events.  This assures that we always
+		 * have an active association for FILE_TRUNC events when the
+		 * seek offset is non-zero.  Note that the association order
+		 * _must_ be FILE_TRUNC followed by FILE_MODIFIED:  if a single
+		 * event induces both a FILE_TRUNC and a FILE_MODIFIED (as
+		 * a VE_CREATE vnode event does), we must process the
+		 * FILE_TRUNC before FILE_MODIFIED -- and the order in which
+		 * these are processed will be the association order.  So
+		 * if we see a FILE_TRUNC, we dissociate/reassociate the
+		 * FILE_MODIFIED association.
 		 */
-		if (ev == NULL || (ev->portev_events & FILE_MODIFIED) ||
+		if (ev == NULL || (ev->portev_events & FILE_TRUNC) ||
 		    !(ev->portev_events & (FILE_MODIFIED | FILE_TRUNC))) {
 			(void) port_associate(port, PORT_SOURCE_FILE,
-			    (uintptr_t)&file->fobj[0],
-			    FILE_MODIFIED | FILE_TRUNC, file);
+			    (uintptr_t)&file->fobj[0], FILE_TRUNC, file);
+			(void) port_dissociate(port, PORT_SOURCE_FILE,
+			    (uintptr_t)&file->fobj[1]);
+			ev = NULL;
 		}
 
-		if (ev == NULL || (ev->portev_events & FILE_TRUNC) ||
+		if (ev == NULL || (ev->portev_events & FILE_MODIFIED) ||
 		    !(ev->portev_events & (FILE_MODIFIED | FILE_TRUNC))) {
 			(void) port_associate(port, PORT_SOURCE_FILE,
-			    (uintptr_t)&file->fobj[1], FILE_TRUNC, file);
+			    (uintptr_t)&file->fobj[1], FILE_MODIFIED, file);
 		}
 	} else {
 		for (i = 0; i <= 1; i++) {
diff --git a/usr/src/cmd/tail/tests/tailtests.sh b/usr/src/cmd/tail/tests/tailtests.sh
index c0ae3afcc4..d3a8bb1cf2 100644
--- a/usr/src/cmd/tail/tests/tailtests.sh
+++ b/usr/src/cmd/tail/tests/tailtests.sh
@@ -24,8 +24,8 @@ checktest()
 
 	if [[ "$actual" != "$output" ]]; then
 		echo "$CMD: test $test: FAIL"
-		echo -e "$CMD: test $test: expected output:\n$o"
-		echo -e "$CMD: test $test: actual output:\n$a"
+		echo -e "$CMD: test $test: expected output:\n$output"
+		echo -e "$CMD: test $test: actual output:\n$actual"
 	else
 		echo "$CMD: test $test: pass"
 	fi
@@ -177,6 +177,7 @@ checktest "$a" "$o" 19
 # Now we want to do a series of follow tests.
 #
 if [[ $DIR == "" ]]; then
+	export TMPDIR=/var/tmp
 	tdir=$(mktemp -d -t tailtest.XXXXXXXX || exit 1)
 else
 	tdir=$(mktemp -d $DIR/tailtest.XXXXXXXX || exit 1)
@@ -434,8 +435,45 @@ EOF
 	a=`cat $out`
 	checktest "$a" "$o" 27b
 	rm $moved
+
+	#
+	# Verify that -F will deal properly with the file being truncated
+	# not by truncation, but rather via an ftruncate() from logadm.
+	#
+	cat /dev/null > $follow
+	( $PROG -F $follow > $out ) &
+	child=$!
+	echo -e "a\nb\nc\nd\ne\nf" >> $follow
+	logadm -c $follow
+	sleep 2
+	echo -e "g\nh\ni" >> $follow
+	sleep 2
+	kill $child
+	sleep 1
+
+	o=`echo -e "a\nb\nc\nd\ne\nf\ng\nh\ni\n"`
+	a=`cat $out`
+	checktest "$a" "$o" 27c
 fi
 
+#
+# We're now going to test that while we may miss output due to truncations
+# occurring faster than tail can read, we don't ever repeat output.
+#
+cat /dev/null > $follow
+( $PROG -f $follow > $out ) &
+tchild=$!
+( let i=0 ; while true; do echo $i > $follow ; sleep 0.1; let i=i+1 ; done ) &
+child=$!
+sleep 10
+kill $tchild
+kill $child
+
+a=`sort $out | uniq -c | sort -n | tail -1 | awk '{ print $1 }'`
+o=1
+
+checktest "$a" "$o" 28
+
 echo "$CMD: completed"
 
 exit $errs
diff --git a/usr/src/uts/common/fs/nfs/nfs3_vnops.c b/usr/src/uts/common/fs/nfs/nfs3_vnops.c
index 4e5882ad7c..fd1595b9f5 100644
--- a/usr/src/uts/common/fs/nfs/nfs3_vnops.c
+++ b/usr/src/uts/common/fs/nfs/nfs3_vnops.c
@@ -29,7 +29,7 @@
  */
 
 /*
- * Copyright (c) 2012, Joyent, Inc. All rights reserved.
+ * Copyright (c) 2013, Joyent, Inc. All rights reserved.
  */
 
 #include <sys/param.h>
@@ -1331,7 +1331,12 @@ nfs3_setattr(vnode_t *vp, struct vattr *vap, int flags, cred_t *cr,
 	if (error)
 		return (error);
 
-	return (nfs3setattr(vp, vap, flags, cr));
+	error = nfs3setattr(vp, vap, flags, cr);
+
+	if (error == 0 && (vap->va_mask & AT_SIZE) && vap->va_size == 0)
+		vnevent_truncate(vp, ct);
+
+	return (error);
 }
 
 static int
@@ -5521,6 +5526,9 @@ nfs3_space(vnode_t *vp, int cmd, struct flock64 *bfp, int flag,
 			va.va_mask = AT_SIZE;
 			va.va_size = bfp->l_start;
 			error = nfs3setattr(vp, &va, 0, cr);
+
+			if (error == 0 && bfp->l_start == 0)
+				vnevent_truncate(vp, ct);
 		} else
 			error = EINVAL;
 	}
diff --git a/usr/src/uts/common/fs/nfs/nfs4_vnops.c b/usr/src/uts/common/fs/nfs/nfs4_vnops.c
index f05a0717d9..0c97fcc176 100644
--- a/usr/src/uts/common/fs/nfs/nfs4_vnops.c
+++ b/usr/src/uts/common/fs/nfs/nfs4_vnops.c
@@ -32,7 +32,7 @@
  */
 
 /*
- * Copyright (c) 2012, Joyent, Inc. All rights reserved.
+ * Copyright (c) 2013, Joyent, Inc. All rights reserved.
  */
 
 #include <sys/param.h>
@@ -3716,6 +3716,8 @@ static int
 nfs4_setattr(vnode_t *vp, struct vattr *vap, int flags, cred_t *cr,
     caller_context_t *ct)
 {
+	int error;
+
 	if (vap->va_mask & AT_NOSET)
 		return (EINVAL);
 
@@ -3731,8 +3733,12 @@ nfs4_setattr(vnode_t *vp, struct vattr *vap, int flags, cred_t *cr,
 	 * to setattr (e.g. basic without chmod) then we will
 	 * need to add a check here before calling the server.
 	 */
+	error = nfs4setattr(vp, vap, flags, cr, NULL);
+
+	if (error == 0 && (vap->va_mask & AT_SIZE) && vap->va_size == 0)
+		vnevent_truncate(vp, ct);
 
-	return (nfs4setattr(vp, vap, flags, cr, NULL));
+	return (error);
 }
 
 /*
@@ -10999,6 +11005,9 @@ nfs4_space(vnode_t *vp, int cmd, struct flock64 *bfp, int flag,
 			va.va_mask = AT_SIZE;
 			va.va_size = bfp->l_start;
 			error = nfs4setattr(vp, &va, 0, cr, NULL);
+
+			if (error == 0 && bfp->l_start == 0)
+				vnevent_truncate(vp, ct);
 		} else
 			error = EINVAL;
 	}
diff --git a/usr/src/uts/common/fs/nfs/nfs_vnops.c b/usr/src/uts/common/fs/nfs/nfs_vnops.c
index fa31e3693f..df128f4bc6 100644
--- a/usr/src/uts/common/fs/nfs/nfs_vnops.c
+++ b/usr/src/uts/common/fs/nfs/nfs_vnops.c
@@ -26,7 +26,7 @@
  */
 
 /*
- * Copyright (c) 2012, Joyent, Inc. All rights reserved.
+ * Copyright (c) 2013, Joyent, Inc. All rights reserved.
  */
 
 #include <sys/param.h>
@@ -1171,7 +1171,12 @@ nfs_setattr(vnode_t *vp, struct vattr *vap, int flags, cred_t *cr,
 	if (error)
 		return (error);
 
-	return (nfssetattr(vp, vap, flags, cr));
+	error = nfssetattr(vp, vap, flags, cr);
+
+	if (error == 0 && (mask & AT_SIZE) && vap->va_size == 0)
+		vnevent_truncate(vp, ct);
+
+	return (error);
 }
 
 static int
@@ -4618,6 +4623,9 @@ nfs_space(vnode_t *vp, int cmd, struct flock64 *bfp, int flag,
 			va.va_mask = AT_SIZE;
 			va.va_size = bfp->l_start;
 			error = nfssetattr(vp, &va, 0, cr);
+
+			if (error == 0 && bfp->l_start == 0)
+				vnevent_truncate(vp, ct);
 		} else
 			error = EINVAL;
 	}
diff --git a/usr/src/uts/common/fs/pcfs/pc_vnops.c b/usr/src/uts/common/fs/pcfs/pc_vnops.c
index ce9711140b..e2c6f3a371 100644
--- a/usr/src/uts/common/fs/pcfs/pc_vnops.c
+++ b/usr/src/uts/common/fs/pcfs/pc_vnops.c
@@ -24,6 +24,10 @@
  * Use is subject to license terms.
  */
 
+/*
+ * Copyright (c) 2013, Joyent, Inc. All rights reserved.
+ */
+
 #include <sys/param.h>
 #include <sys/t_lock.h>
 #include <sys/systm.h>
@@ -772,8 +776,12 @@ pcfs_setattr(
 			goto out;
 		}
 		error = pc_truncate(pcp, (uint_t)vap->va_size);
+
 		if (error)
 			goto out;
+
+		if (vap->va_size == 0)
+			vnevent_truncate(vp, ct);
 	}
 	/*
 	 * Change file modified times.
diff --git a/usr/src/uts/common/fs/portfs/port_fop.c b/usr/src/uts/common/fs/portfs/port_fop.c
index 1094dc160c..03a03dd2a6 100644
--- a/usr/src/uts/common/fs/portfs/port_fop.c
+++ b/usr/src/uts/common/fs/portfs/port_fop.c
@@ -1828,12 +1828,19 @@ port_fop_sendevent(vnode_t *vp, int events, vnode_t *dvp, char *cname)
 	if (!removeall) {
 		/*
 		 * All the active ones are in the beginning of the list.
+		 * Note that we process this list in reverse order to assure
+		 * that events are delivered in the order that they were
+		 * associated.
 		 */
-		for (pfp = (portfop_t *)list_head(&pvp->pvp_pfoplist);
-		    pfp && pfp->pfop_flags & PORT_FOP_ACTIVE; pfp = npfp) {
+		for (pfp = (portfop_t *)list_tail(&pvp->pvp_pfoplist);
+		    pfp && !(pfp->pfop_flags & PORT_FOP_ACTIVE); pfp = npfp) {
+			npfp = list_prev(&pvp->pvp_pfoplist, pfp);
+		}
+
+		for (; pfp != NULL; pfp = npfp) {
 			int levents = events;
 
-			npfp = list_next(&pvp->pvp_pfoplist, pfp);
+			npfp = list_prev(&pvp->pvp_pfoplist, pfp);
 			/*
 			 * Hard links case - If the file is being
 			 * removed/renamed, and the name matches
@@ -2348,6 +2355,9 @@ port_fop_vnevent(femarg_t *vf, vnevent_t vnevent, vnode_t *dvp, char *name,
 	case	VE_MOUNTEDOVER:
 			port_fop_sendevent(vp, MOUNTEDOVER, NULL, NULL);
 		break;
+	case	VE_TRUNCATE:
+			port_fop_sendevent(vp, FILE_TRUNC, NULL, NULL);
+		break;
 	default:
 		break;
 	}
diff --git a/usr/src/uts/common/fs/tmpfs/tmp_vnops.c b/usr/src/uts/common/fs/tmpfs/tmp_vnops.c
index a2074f284f..602c61602c 100644
--- a/usr/src/uts/common/fs/tmpfs/tmp_vnops.c
+++ b/usr/src/uts/common/fs/tmpfs/tmp_vnops.c
@@ -792,6 +792,10 @@ tmp_setattr(
 		error = tmpnode_trunc(tm, tp, (ulong_t)vap->va_size);
 		rw_exit(&tp->tn_contents);
 		rw_exit(&tp->tn_rwlock);
+
+		if (error == 0 && vap->va_size == 0)
+			vnevent_truncate(vp, ct);
+
 		goto out1;
 	}
 out:
@@ -2368,6 +2372,9 @@ tmp_space(
 		if ((bfp->l_start > MAXOFF_T) || (bfp->l_len > MAXOFF_T))
 			return (EFBIG);
 		error = tmp_freesp(vp, bfp, flag);
+
+		if (error == 0 && bfp->l_start == 0)
+			vnevent_truncate(vp, ct);
 	}
 	return (error);
 }
diff --git a/usr/src/uts/common/fs/udfs/udf_vnops.c b/usr/src/uts/common/fs/udfs/udf_vnops.c
index 90ad1b8d48..307d3987ed 100644
--- a/usr/src/uts/common/fs/udfs/udf_vnops.c
+++ b/usr/src/uts/common/fs/udfs/udf_vnops.c
@@ -23,6 +23,10 @@
  * Copyright (c) 1998, 2010, Oracle and/or its affiliates. All rights reserved.
  */
 
+/*
+ * Copyright (c) 2013, Joyent, Inc. All rights reserved.
+ */
+
 #include <sys/types.h>
 #include <sys/t_lock.h>
 #include <sys/param.h>
@@ -564,6 +568,9 @@ udf_setattr(
 		if (error = ud_itrunc(ip, vap->va_size, 0, cr)) {
 			goto update_inode;
 		}
+
+		if (vap->va_size == 0)
+			vnevent_truncate(vp, ct);
 	}
 	/*
 	 * Change file access or modified times.
@@ -1621,6 +1628,9 @@ udf_space(
 		error =  EINVAL;
 	} else if ((error = convoff(vp, bfp, 0, offset)) == 0) {
 		error = ud_freesp(vp, bfp, flag, cr);
+
+		if (error == 0 && bfp->l_start == 0)
+			vnevent_truncate(vp, ct);
 	}
 
 	return (error);
diff --git a/usr/src/uts/common/fs/ufs/ufs_vnops.c b/usr/src/uts/common/fs/ufs/ufs_vnops.c
index e210339dfb..92bdb0d0a7 100644
--- a/usr/src/uts/common/fs/ufs/ufs_vnops.c
+++ b/usr/src/uts/common/fs/ufs/ufs_vnops.c
@@ -21,6 +21,7 @@
 
 /*
  * Copyright (c) 1984, 2010, Oracle and/or its affiliates. All rights reserved.
+ * Copyright (c) 2013, Joyent, Inc. All rights reserved.
  */
 
 /*	Copyright (c) 1983, 1984, 1985, 1986, 1987, 1988, 1989 AT&T	*/
@@ -2191,6 +2192,9 @@ again:
 			rw_enter(&ip->i_contents, RW_WRITER);
 			goto update_inode;
 		}
+
+		if (error == 0 && vap->va_size)
+			vnevent_truncate(vp, ct);
 	}
 
 	if (ulp) {
@@ -4460,6 +4464,9 @@ ufs_space(struct vnode *vp, int cmd, struct flock64 *bfp, int flag,
 			if (error)
 				return (error);
 			error = ufs_freesp(vp, bfp, flag, cr);
+
+			if (error == 0 && bfp->l_start == 0)
+				vnevent_truncate(vp, ct);
 		} else if (cmd == F_ALLOCSP) {
 			error = ufs_lockfs_begin(ufsvfsp, &ulp,
 			    ULOCKFS_FALLOCATE_MASK);
diff --git a/usr/src/uts/common/fs/vnode.c b/usr/src/uts/common/fs/vnode.c
index 67f21866ec..679940cf5e 100644
--- a/usr/src/uts/common/fs/vnode.c
+++ b/usr/src/uts/common/fs/vnode.c
@@ -21,7 +21,7 @@
 
 /*
  * Copyright (c) 1988, 2010, Oracle and/or its affiliates. All rights reserved.
- * Copyright (c) 2012, Joyent Inc. All rights reserved.
+ * Copyright (c) 2013, Joyent, Inc. All rights reserved.
  */
 
 /*	Copyright (c) 1983, 1984, 1985, 1986, 1987, 1988, 1989 AT&T	*/
@@ -2589,6 +2589,15 @@ vnevent_mountedover(vnode_t *vp, caller_context_t *ct)
 	(void) VOP_VNEVENT(vp, VE_MOUNTEDOVER, NULL, NULL, ct);
 }
 
+void
+vnevent_truncate(vnode_t *vp, caller_context_t *ct)
+{
+	if (vp == NULL || vp->v_femhead == NULL) {
+		return;
+	}
+	(void) VOP_VNEVENT(vp, VE_TRUNCATE, NULL, NULL, ct);
+}
+
 /*
  * Vnode accessors.
  */
diff --git a/usr/src/uts/common/fs/zfs/zfs_vnops.c b/usr/src/uts/common/fs/zfs/zfs_vnops.c
index 84960c9a13..f1b5abc1f1 100644
--- a/usr/src/uts/common/fs/zfs/zfs_vnops.c
+++ b/usr/src/uts/common/fs/zfs/zfs_vnops.c
@@ -2757,6 +2757,9 @@ top:
 			ZFS_EXIT(zfsvfs);
 			return (err);
 		}
+
+		if (vap->va_size == 0)
+			vnevent_truncate(ZTOV(zp), ct);
 	}
 
 	if (mask & (AT_ATIME|AT_MTIME) ||
@@ -4748,6 +4751,9 @@ zfs_space(vnode_t *vp, int cmd, flock64_t *bfp, int flag,
 
 	error = zfs_freesp(zp, off, len, flag, TRUE);
 
+	if (error == 0 && off == 0 && len == 0)
+		vnevent_truncate(ZTOV(zp), ct);
+
 	ZFS_EXIT(zfsvfs);
 	return (error);
 }
diff --git a/usr/src/uts/common/sys/vnode.h b/usr/src/uts/common/sys/vnode.h
index d29152346e..af9516fe52 100644
--- a/usr/src/uts/common/sys/vnode.h
+++ b/usr/src/uts/common/sys/vnode.h
@@ -21,6 +21,7 @@
 
 /*
  * Copyright (c) 1988, 2010, Oracle and/or its affiliates. All rights reserved.
+ * Copyright (c) 2013, Joyent, Inc. All rights reserved.
  */
 
 /*	Copyright (c) 1983, 1984, 1985, 1986, 1987, 1988, 1989 AT&T	*/
@@ -733,7 +734,8 @@ typedef enum vnevent	{
 	VE_CREATE	= 5,	/* Create with vnode's name which exists */
 	VE_LINK		= 6, 	/* Link with vnode's name as source */
 	VE_RENAME_DEST_DIR	= 7, 	/* Rename with vnode as target dir */
-	VE_MOUNTEDOVER	= 8 	/* File or Filesystem got mounted over vnode */
+	VE_MOUNTEDOVER	= 8, 	/* File or Filesystem got mounted over vnode */
+	VE_TRUNCATE = 9		/* Truncate */
 } vnevent_t;
 
 /*
@@ -1290,6 +1292,7 @@ void	vnevent_create(vnode_t *, caller_context_t *);
 void	vnevent_link(vnode_t *, caller_context_t *);
 void	vnevent_rename_dest_dir(vnode_t *, caller_context_t *ct);
 void	vnevent_mountedover(vnode_t *, caller_context_t *);
+void	vnevent_truncate(vnode_t *, caller_context_t *);
 int	vnevent_support(vnode_t *, caller_context_t *);
 
 /* Vnode specific data */