PSARC 2006/077 zpool clear

PSARC 2006/139 FMA for ZFS
6284889 arc should replace the znode cache
6333006 DMU & DSL should not panic upon I/O error
6333092 concurrent reads to a file not scaling with number of readers
6338081 ZFS/FMA phase 1
6338386 need persistent error log
6341326 i/o error causes arc buf hash table corruption
6341639 zfs backup/restore should compute/verify checksum of backup stream
6348002 out of space due to changing properties
6354724 inaccurate error message from zfs restore
6354872 dmu_sync() blows predictive accounting
6355416 zpool scrubbing consumes all memory, system hung
6363995 df should only load libzfs when it encounters a ZFS filesystem
6366320 zfs backup/restore doesn't like signals
6368892 mount -m support needed for legacy mounts
6368902 boot archive fstat support needed for ZFS Mountroot
6369424 BFU complains when bfu'ing a ZFS root filesystem
6374062 mountroot support needed for ZFS
6376356 dirtying dbuf obj=43 lvl=0 blkid=0 but not tx_held
6378391 unused members of dmu_objset_stats_t
6378392 clean up zfs_cmd_t structure
6378685 buf_init should allocate its hash table more carefully
6378976 ziltest should be a first class citizen
6381086 zdb segfaults if there is a spa deferred-free bplist
6381203 deadlock due to i/o while assigning (tc_lock held)
6381209 freed space is not immediately available
6381344 'zpool clear'
6381345 FAULTED devices should really be UNAVAIL
6381346 import should mark devices as persistently unavailable
6383272 recursive mutex_enter() during log replay with zfs root
6386326 origin property is not displayed
6386354 libzfs does too much in its _init section, calls exit(1)
6386624 zpool should not complain about non-existent devices from libdiskmgt
6386910 spa needs to be i/o error hardened
6387735 need a mechanism to inject faults into ZFS
6387736 internal ZFS utilities should be placed in an ON-private package
6389928 libzfs should ship a lint library
6390609 malformed vdev config panics on zpool_create()
6390677 version number checking makes upgrades challenging
6390713 ztest hangs in zil_suspend()
6391873 metadata compression should be turned back on
6392113 ztest sometimes reports leaked blocks because ZIL isn't resilvered
6393004 minor memory leak in unique_insert()

8 files changed, 1462 insertions, 0 deletions

diff --git a/usr/src/cmd/zinject/Makefile b/usr/src/cmd/zinject/Makefile
new file mode 100644
index 0000000000..f646689967
--- /dev/null
+++ b/usr/src/cmd/zinject/Makefile
@@ -0,0 +1,54 @@
+#
+# CDDL HEADER START
+#
+# The contents of this file are subject to the terms of the
+# Common Development and Distribution License (the "License").
+# You may not use this file except in compliance with the License.
+#
+# You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
+# or http://www.opensolaris.org/os/licensing.
+# See the License for the specific language governing permissions
+# and limitations under the License.
+#
+# When distributing Covered Code, include this CDDL HEADER in each
+# file and include the License file at usr/src/OPENSOLARIS.LICENSE.
+# If applicable, add the following below this CDDL HEADER, with the
+# fields enclosed by brackets "[]" replaced with your own identifying
+# information: Portions Copyright [yyyy] [name of copyright owner]
+#
+# CDDL HEADER END
+#
+#
+# Copyright 2006 Sun Microsystems, Inc.  All rights reserved.
+# Use is subject to license terms.
+#
+# ident	"%Z%%M%	%I%	%E% SMI"
+#
+
+PROG:sh=	basename `pwd`
+
+include ../Makefile.cmd
+
+$(INTEL_BLD)SUBDIRS	= $(MACH)
+$(BUILD64)SUBDIRS	+= $(MACH64)
+
+all	:=	TARGET = all
+install	:=	TARGET = install
+clean	:=	TARGET = clean
+clobber	:=	TARGET = clobber
+lint	:=	TARGET = lint
+
+.KEEP_STATE:
+
+all clean clobber lint:	$(SUBDIRS)
+
+install:	$(SUBDIRS)
+	-$(RM) $(ROOTUSRSBINPROG)
+	-$(LN) $(ISAEXEC) $(ROOTUSRSBINPROG)
+
+$(SUBDIRS):	FRC
+	@cd $@; pwd; $(MAKE) $(TARGET)
+
+FRC:
+
+include ../Makefile.targ
diff --git a/usr/src/cmd/zinject/Makefile.com b/usr/src/cmd/zinject/Makefile.com
new file mode 100644
index 0000000000..40f1914729
--- /dev/null
+++ b/usr/src/cmd/zinject/Makefile.com
@@ -0,0 +1,55 @@
+#
+# CDDL HEADER START
+#
+# The contents of this file are subject to the terms of the
+# Common Development and Distribution License (the "License").
+# You may not use this file except in compliance with the License.
+#
+# You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
+# or http://www.opensolaris.org/os/licensing.
+# See the License for the specific language governing permissions
+# and limitations under the License.
+#
+# When distributing Covered Code, include this CDDL HEADER in each
+# file and include the License file at usr/src/OPENSOLARIS.LICENSE.
+# If applicable, add the following below this CDDL HEADER, with the
+# fields enclosed by brackets "[]" replaced with your own identifying
+# information: Portions Copyright [yyyy] [name of copyright owner]
+#
+# CDDL HEADER END
+#
+#
+# Copyright 2006 Sun Microsystems, Inc.  All rights reserved.
+# Use is subject to license terms.
+#
+# ident	"%Z%%M%	%I%	%E% SMI"
+#
+
+PROG:sh=	cd ..; basename `pwd`
+SRCS= ../$(PROG).c ../translate.c
+
+include ../../Makefile.cmd
+
+INCS +=	-I../../../lib/libzpool/common
+INCS +=	-I../../../uts/common/fs/zfs
+
+LDLIBS += -lzpool -lzfs
+
+C99MODE=	-xc99=%all
+C99LMODE=	-Xc99=%all
+
+CPPFLAGS += -D_LARGEFILE64_SOURCE=1 -D_REENTRANT $(INCS)
+
+.KEEP_STATE:
+
+all: $(PROG)
+
+$(PROG): $(SRCS)
+	$(LINK.c) -o $(PROG) $(SRCS) $(LDLIBS)
+	$(POST_PROCESS)
+
+clean:
+
+lint:	lint_SRCS
+
+include ../../Makefile.targ
diff --git a/usr/src/cmd/zinject/amd64/Makefile b/usr/src/cmd/zinject/amd64/Makefile
new file mode 100644
index 0000000000..8740a9f3ac
--- /dev/null
+++ b/usr/src/cmd/zinject/amd64/Makefile
@@ -0,0 +1,31 @@
+#
+# CDDL HEADER START
+#
+# The contents of this file are subject to the terms of the
+# Common Development and Distribution License (the "License").
+# You may not use this file except in compliance with the License.
+#
+# You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
+# or http://www.opensolaris.org/os/licensing.
+# See the License for the specific language governing permissions
+# and limitations under the License.
+#
+# When distributing Covered Code, include this CDDL HEADER in each
+# file and include the License file at usr/src/OPENSOLARIS.LICENSE.
+# If applicable, add the following below this CDDL HEADER, with the
+# fields enclosed by brackets "[]" replaced with your own identifying
+# information: Portions Copyright [yyyy] [name of copyright owner]
+#
+# CDDL HEADER END
+#
+#
+# Copyright 2006 Sun Microsystems, Inc.  All rights reserved.
+# Use is subject to license terms.
+#
+# ident	"%Z%%M%	%I%	%E% SMI"
+#
+
+include ../Makefile.com
+include ../../Makefile.cmd.64
+
+install: all $(ROOTUSRSBINPROG64)
diff --git a/usr/src/cmd/zinject/i386/Makefile b/usr/src/cmd/zinject/i386/Makefile
new file mode 100644
index 0000000000..d2cb13dcd1
--- /dev/null
+++ b/usr/src/cmd/zinject/i386/Makefile
@@ -0,0 +1,30 @@
+#
+# CDDL HEADER START
+#
+# The contents of this file are subject to the terms of the
+# Common Development and Distribution License (the "License").
+# You may not use this file except in compliance with the License.
+#
+# You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
+# or http://www.opensolaris.org/os/licensing.
+# See the License for the specific language governing permissions
+# and limitations under the License.
+#
+# When distributing Covered Code, include this CDDL HEADER in each
+# file and include the License file at usr/src/OPENSOLARIS.LICENSE.
+# If applicable, add the following below this CDDL HEADER, with the
+# fields enclosed by brackets "[]" replaced with your own identifying
+# information: Portions Copyright [yyyy] [name of copyright owner]
+#
+# CDDL HEADER END
+#
+#
+# Copyright 2006 Sun Microsystems, Inc.  All rights reserved.
+# Use is subject to license terms.
+#
+# ident	"%Z%%M%	%I%	%E% SMI"
+#
+
+include ../Makefile.com
+
+install: all $(ROOTUSRSBINPROG32)
diff --git a/usr/src/cmd/zinject/sparcv9/Makefile b/usr/src/cmd/zinject/sparcv9/Makefile
new file mode 100644
index 0000000000..8740a9f3ac
--- /dev/null
+++ b/usr/src/cmd/zinject/sparcv9/Makefile
@@ -0,0 +1,31 @@
+#
+# CDDL HEADER START
+#
+# The contents of this file are subject to the terms of the
+# Common Development and Distribution License (the "License").
+# You may not use this file except in compliance with the License.
+#
+# You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
+# or http://www.opensolaris.org/os/licensing.
+# See the License for the specific language governing permissions
+# and limitations under the License.
+#
+# When distributing Covered Code, include this CDDL HEADER in each
+# file and include the License file at usr/src/OPENSOLARIS.LICENSE.
+# If applicable, add the following below this CDDL HEADER, with the
+# fields enclosed by brackets "[]" replaced with your own identifying
+# information: Portions Copyright [yyyy] [name of copyright owner]
+#
+# CDDL HEADER END
+#
+#
+# Copyright 2006 Sun Microsystems, Inc.  All rights reserved.
+# Use is subject to license terms.
+#
+# ident	"%Z%%M%	%I%	%E% SMI"
+#
+
+include ../Makefile.com
+include ../../Makefile.cmd.64
+
+install: all $(ROOTUSRSBINPROG64)
diff --git a/usr/src/cmd/zinject/translate.c b/usr/src/cmd/zinject/translate.c
new file mode 100644
index 0000000000..882b230930
--- /dev/null
+++ b/usr/src/cmd/zinject/translate.c
@@ -0,0 +1,458 @@
+/*
+ * CDDL HEADER START
+ *
+ * The contents of this file are subject to the terms of the
+ * Common Development and Distribution License (the "License").
+ * You may not use this file except in compliance with the License.
+ *
+ * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
+ * or http://www.opensolaris.org/os/licensing.
+ * See the License for the specific language governing permissions
+ * and limitations under the License.
+ *
+ * When distributing Covered Code, include this CDDL HEADER in each
+ * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
+ * If applicable, add the following below this CDDL HEADER, with the
+ * fields enclosed by brackets "[]" replaced with your own identifying
+ * information: Portions Copyright [yyyy] [name of copyright owner]
+ *
+ * CDDL HEADER END
+ */
+/*
+ * Copyright 2006 Sun Microsystems, Inc.  All rights reserved.
+ * Use is subject to license terms.
+ */
+
+#pragma ident	"%Z%%M%	%I%	%E% SMI"
+
+#include <libzfs.h>
+
+#undef verify	/* both libzfs.h and zfs_context.h want to define this */
+
+#include <sys/zfs_context.h>
+
+#include <errno.h>
+#include <fcntl.h>
+#include <stdarg.h>
+#include <stddef.h>
+#include <stdio.h>
+#include <stdlib.h>
+#include <strings.h>
+#include <sys/file.h>
+#include <sys/mntent.h>
+#include <sys/mnttab.h>
+#include <sys/param.h>
+#include <sys/stat.h>
+
+#include <sys/dmu.h>
+#include <sys/dmu_objset.h>
+#include <sys/dnode.h>
+
+#include <sys/mkdev.h>
+
+#include "zinject.h"
+
+extern void kernel_init(int);
+extern void kernel_fini(void);
+
+static int debug;
+
+static void
+ziprintf(const char *fmt, ...)
+{
+	va_list ap;
+
+	if (!debug)
+		return;
+
+	va_start(ap, fmt);
+	(void) vprintf(fmt, ap);
+	va_end(ap);
+}
+
+/*
+ * Given a full path to a file, translate into a dataset name and a relative
+ * path within the dataset.  'dataset' must be at least MAXNAMELEN characters,
+ * and 'relpath' must be at least MAXPATHLEN characters.  We also pass a stat64
+ * buffer, which we need later to get the object ID.
+ */
+static int
+parse_pathname(const char *fullpath, char *dataset, char *relpath,
+    struct stat64 *statbuf)
+{
+	struct extmnttab mp;
+	FILE *fp;
+	int match;
+	const char *rel;
+
+	if (fullpath[0] != '/') {
+		(void) fprintf(stderr, "invalid object '%s': must be full "
+		    "path\n", fullpath);
+		usage();
+		return (-1);
+	}
+
+	if (strlen(fullpath) >= MAXPATHLEN) {
+		(void) fprintf(stderr, "invalid object; pathname too long\n");
+		return (-1);
+	}
+
+	if (stat64(fullpath, statbuf) != 0) {
+		(void) fprintf(stderr, "cannot open '%s': %s\n",
+		    fullpath, strerror(errno));
+		return (-1);
+	}
+
+	if ((fp = fopen(MNTTAB, "r")) == NULL) {
+		(void) fprintf(stderr, "cannot open /etc/mnttab\n");
+		return (-1);
+	}
+
+	match = 0;
+	while (getextmntent(fp, &mp, sizeof (mp)) == 0) {
+		if (makedev(mp.mnt_major, mp.mnt_minor) == statbuf->st_dev) {
+			match = 1;
+			break;
+		}
+	}
+
+	if (!match) {
+		(void) fprintf(stderr, "cannot find mountpoint for '%s'\n",
+		    fullpath);
+		return (-1);
+	}
+
+	if (strcmp(mp.mnt_fstype, MNTTYPE_ZFS) != 0) {
+		(void) fprintf(stderr, "invalid path '%s': not a ZFS "
+		    "filesystem\n", fullpath);
+		return (-1);
+	}
+
+	if (strncmp(fullpath, mp.mnt_mountp, strlen(mp.mnt_mountp)) != 0) {
+		(void) fprintf(stderr, "invalid path '%s': mountpoint "
+		    "doesn't match path\n", fullpath);
+		return (-1);
+	}
+
+	(void) strcpy(dataset, mp.mnt_special);
+
+	rel = fullpath + strlen(mp.mnt_mountp);
+	if (rel[0] == '/')
+		rel++;
+	(void) strcpy(relpath, rel);
+
+	return (0);
+}
+
+/*
+ * Convert from a (dataset, path) pair into a (objset, object) pair.  Note that
+ * we grab the object number from the inode number, since looking this up via
+ * libzpool is a real pain.
+ */
+/* ARGSUSED */
+static int
+object_from_path(const char *dataset, const char *path, struct stat64 *statbuf,
+    zinject_record_t *record)
+{
+	objset_t *os;
+	int err;
+
+	/*
+	 * Before doing any libzpool operations, call sync() to ensure that the
+	 * on-disk state is consistent with the in-core state.
+	 */
+	sync();
+
+	if ((err = dmu_objset_open(dataset, DMU_OST_ZFS,
+	    DS_MODE_STANDARD | DS_MODE_READONLY, &os)) != 0) {
+		(void) fprintf(stderr, "cannot open dataset '%s': %s\n",
+		    dataset, strerror(err));
+		return (-1);
+	}
+
+	record->zi_objset = dmu_objset_id(os);
+	record->zi_object = statbuf->st_ino;
+
+	dmu_objset_close(os);
+
+	return (0);
+}
+
+/*
+ * Calculate the real range based on the type, level, and range given.
+ */
+static int
+calculate_range(const char *dataset, err_type_t type, int level, char *range,
+    zinject_record_t *record)
+{
+	objset_t *os = NULL;
+	dnode_t *dn = NULL;
+	int err;
+	int ret = -1;
+
+	/*
+	 * Determine the numeric range from the string.
+	 */
+	if (range == NULL) {
+		/*
+		 * If range is unspecified, set the range to [0,-1], which
+		 * indicates that the whole object should be treated as an
+		 * error.
+		 */
+		record->zi_start = 0;
+		record->zi_end = -1ULL;
+	} else {
+		char *end;
+
+		/* XXX add support for suffixes */
+		record->zi_start = strtoull(range, &end, 10);
+
+
+		if (*end == '\0')
+			record->zi_end = record->zi_start + 1;
+		else if (*end == ',')
+			record->zi_end = strtoull(end + 1, &end, 10);
+
+		if (*end != '\0') {
+			(void) fprintf(stderr, "invalid range '%s': must be "
+			    "a numeric range of the form 'start[,end]'\n",
+			    range);
+			goto out;
+		}
+	}
+
+	switch (type) {
+	case TYPE_DATA:
+		break;
+
+	case TYPE_DNODE:
+		/*
+		 * If this is a request to inject faults into the dnode, then we
+		 * must translate the current (objset,object) pair into an
+		 * offset within the metadnode for the objset.  Specifying any
+		 * kind of range with type 'dnode' is illegal.
+		 */
+		if (range != NULL) {
+			(void) fprintf(stderr, "range cannot be specified when "
+			    "type is 'dnode'\n");
+			goto out;
+		}
+
+		record->zi_start = record->zi_object * sizeof (dnode_phys_t);
+		record->zi_end = record->zi_start + sizeof (dnode_phys_t);
+		record->zi_object = 0;
+		break;
+	}
+
+	/*
+	 * Get the dnode associated with object, so we can calculate the block
+	 * size.
+	 */
+	if ((err = dmu_objset_open(dataset, DMU_OST_ANY,
+	    DS_MODE_STANDARD | DS_MODE_READONLY, &os)) != 0) {
+		(void) fprintf(stderr, "cannot open dataset '%s': %s\n",
+		    dataset, strerror(err));
+		goto out;
+	}
+
+	if (record->zi_object == 0) {
+		dn = os->os->os_meta_dnode;
+	} else {
+		err = dnode_hold(os->os, record->zi_object, FTAG, &dn);
+		if (err != 0) {
+			(void) fprintf(stderr, "failed to hold dnode "
+			    "for object %llu\n",
+			    (u_longlong_t)record->zi_object);
+			goto out;
+		}
+	}
+
+
+	ziprintf("data shift: %d\n", (int)dn->dn_datablkshift);
+	ziprintf(" ind shift: %d\n", (int)dn->dn_indblkshift);
+
+	/*
+	 * Translate range into block IDs.
+	 */
+	if (record->zi_start != 0 || record->zi_end != -1ULL) {
+		record->zi_start >>= dn->dn_datablkshift;
+		record->zi_end >>= dn->dn_datablkshift;
+	}
+
+	/*
+	 * Check level, and then translate level 0 blkids into ranges
+	 * appropriate for level of indirection.
+	 */
+	record->zi_level = level;
+	if (level > 0) {
+		ziprintf("level 0 blkid range: [%llu, %llu]\n",
+		    record->zi_start, record->zi_end);
+
+		if (level >= dn->dn_nlevels) {
+			(void) fprintf(stderr, "level %d exceeds max level "
+			    "of object (%d)\n", level, dn->dn_nlevels - 1);
+			goto out;
+		}
+
+		if (record->zi_start != 0 || record->zi_end != 0) {
+			int shift = dn->dn_indblkshift - SPA_BLKPTRSHIFT;
+
+			for (; level > 0; level--) {
+				record->zi_start >>= shift;
+				record->zi_end >>= shift;
+			}
+		}
+	}
+
+	ret = 0;
+out:
+	if (dn) {
+		if (dn != os->os->os_meta_dnode)
+			dnode_rele(dn, FTAG);
+	}
+	if (os)
+		dmu_objset_close(os);
+
+	return (ret);
+}
+
+int
+translate_record(err_type_t type, const char *object, const char *range,
+    int level, zinject_record_t *record, char *poolname, char *dataset)
+{
+	char path[MAXPATHLEN];
+	char *slash;
+	struct stat64 statbuf;
+	int ret = -1;
+
+	kernel_init(FREAD);
+
+	debug = (getenv("ZINJECT_DEBUG") != NULL);
+
+	ziprintf("translating: %s\n", object);
+
+	if (MOS_TYPE(type)) {
+		/*
+		 * MOS objects are treated specially.
+		 */
+		switch (type) {
+		case TYPE_MOS:
+			record->zi_type = 0;
+			break;
+		case TYPE_MOSDIR:
+			record->zi_type = DMU_OT_OBJECT_DIRECTORY;
+			break;
+		case TYPE_METASLAB:
+			record->zi_type = DMU_OT_OBJECT_ARRAY;
+			break;
+		case TYPE_CONFIG:
+			record->zi_type = DMU_OT_PACKED_NVLIST;
+			break;
+		case TYPE_BPLIST:
+			record->zi_type = DMU_OT_BPLIST;
+			break;
+		case TYPE_SPACEMAP:
+			record->zi_type = DMU_OT_SPACE_MAP;
+			break;
+		case TYPE_ERRLOG:
+			record->zi_type = DMU_OT_ERROR_LOG;
+			break;
+		}
+
+		dataset[0] = '\0';
+		(void) strcpy(poolname, object);
+		return (0);
+	}
+
+	/*
+	 * Convert a full path into a (dataset, file) pair.
+	 */
+	if (parse_pathname(object, dataset, path, &statbuf) != 0)
+		goto err;
+
+	ziprintf("   dataset: %s\n", dataset);
+	ziprintf("      path: %s\n", path);
+
+	/*
+	 * Convert (dataset, file) into (objset, object)
+	 */
+	if (object_from_path(dataset, path, &statbuf, record) != 0)
+		goto err;
+
+	ziprintf("raw objset: %llu\n", record->zi_objset);
+	ziprintf("raw object: %llu\n", record->zi_object);
+
+	/*
+	 * For the given object, calculate the real (type, level, range)
+	 */
+	if (calculate_range(dataset, type, level, (char *)range, record) != 0)
+		goto err;
+
+	ziprintf("    objset: %llu\n", record->zi_objset);
+	ziprintf("    object: %llu\n", record->zi_object);
+	if (record->zi_start == 0 &&
+	    record->zi_end == -1ULL)
+		ziprintf("     range: all\n");
+	else
+		ziprintf("     range: [%llu, %llu]\n", record->zi_start,
+		    record->zi_end);
+
+	/*
+	 * Copy the pool name
+	 */
+	(void) strcpy(poolname, dataset);
+	if ((slash = strchr(poolname, '/')) != NULL)
+		*slash = '\0';
+
+	ret = 0;
+
+err:
+	kernel_fini();
+	return (ret);
+}
+
+int
+translate_raw(const char *str, zinject_record_t *record)
+{
+	/*
+	 * A raw bookmark of the form objset:object:level:blkid, where each
+	 * number is a hexidecimal value.
+	 */
+	if (sscanf(str, "%llx:%llx:%x:%llx", (u_longlong_t *)&record->zi_objset,
+	    (u_longlong_t *)&record->zi_object, &record->zi_level,
+	    (u_longlong_t *)&record->zi_start) != 4) {
+		(void) fprintf(stderr, "bad raw spec '%s': must be of the form "
+		    "'objset:object:level:blkid'\n", str);
+		return (-1);
+	}
+
+	record->zi_end = record->zi_start;
+
+	return (0);
+}
+
+int
+translate_device(const char *pool, const char *device, zinject_record_t *record)
+{
+	char *end;
+	zpool_handle_t *zhp;
+
+	/*
+	 * Given a device name or GUID, create an appropriate injection record
+	 * with zi_guid set.
+	 */
+	if ((zhp = zpool_open(pool)) == NULL)
+		return (-1);
+
+	record->zi_guid = strtoull(device, &end, 16);
+	if (record->zi_guid == 0 || *end != '\0')
+		record->zi_guid = zpool_vdev_to_guid(zhp, device);
+
+	if (record->zi_guid == 0) {
+		(void) fprintf(stderr, "cannot find device '%s' in pool '%s'\n",
+		    device, pool);
+		return (-1);
+	}
+
+	return (0);
+}
diff --git a/usr/src/cmd/zinject/zinject.c b/usr/src/cmd/zinject/zinject.c
new file mode 100644
index 0000000000..b584fb0de5
--- /dev/null
+++ b/usr/src/cmd/zinject/zinject.c
@@ -0,0 +1,739 @@
+/*
+ * CDDL HEADER START
+ *
+ * The contents of this file are subject to the terms of the
+ * Common Development and Distribution License (the "License").
+ * You may not use this file except in compliance with the License.
+ *
+ * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
+ * or http://www.opensolaris.org/os/licensing.
+ * See the License for the specific language governing permissions
+ * and limitations under the License.
+ *
+ * When distributing Covered Code, include this CDDL HEADER in each
+ * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
+ * If applicable, add the following below this CDDL HEADER, with the
+ * fields enclosed by brackets "[]" replaced with your own identifying
+ * information: Portions Copyright [yyyy] [name of copyright owner]
+ *
+ * CDDL HEADER END
+ */
+/*
+ * Copyright 2006 Sun Microsystems, Inc.  All rights reserved.
+ * Use is subject to license terms.
+ */
+
+#pragma ident	"%Z%%M%	%I%	%E% SMI"
+
+/*
+ * ZFS Fault Injector
+ *
+ * This userland component takes a set of options and uses libzpool to translate
+ * from a user-visible object type and name to an internal representation.
+ * There are two basic types of faults: device faults and data faults.
+ *
+ *
+ * DEVICE FAULTS
+ *
+ * Errors can be injected into a particular vdev using the '-d' option.  This
+ * option takes a path or vdev GUID to uniquely identify the device within a
+ * pool.  There are two types of errors that can be injected, EIO and ENXIO,
+ * that can be controlled through the '-t' option.  The default is ENXIO.  For
+ * EIO failures, any attempt to read data from the device will return EIO, but
+ * subsequent attempt to reopen the device will succeed.  For ENXIO failures,
+ * any attempt to read from the device will return EIO, but any attempt to
+ * reopen the device will also return ENXIO.
+ *
+ * This form of the command looks like:
+ *
+ * 	zinject -d device [-t type] pool
+ *
+ *
+ * DATA FAULTS
+ *
+ * We begin with a tuple of the form:
+ *
+ * 	<type,level,range,object>
+ *
+ * 	type	A string describing the type of data to target.  Each type
+ * 		implicitly describes how to interpret 'object'. Currently,
+ * 		the following values are supported:
+ *
+ * 		data		User data for a file
+ * 		dnode		Dnode for a file or directory
+ *
+ *		The following MOS objects are special.  Instead of injecting
+ *		errors on a particular object or blkid, we inject errors across
+ *		all objects of the given type.
+ *
+ * 		mos		Any data in the MOS
+ * 		mosdir		object directory
+ * 		config		pool configuration
+ * 		bplist		blkptr list
+ * 		spacemap	spacemap
+ * 		metaslab	metaslab
+ * 		errlog		persistent error log
+ *
+ * 	level	Object level.  Defaults to '0', not applicable to all types.  If
+ * 		a range is given, this corresponds to the indirect block
+ * 		corresponding to the specific range.
+ *
+ *	range	A numerical range [start,end) within the object.  Defaults to
+ *		the full size of the file.
+ *
+ * 	object	A string describing the logical location of the object.  For
+ * 		files and directories (currently the only supported types),
+ * 		this is the path of the object on disk.
+ *
+ * This is translated, via libzpool, into the following internal representation:
+ *
+ * 	<type,objset,object,level,range>
+ *
+ * These types should be self-explanatory.  This tuple is then passed to the
+ * kernel via a special ioctl() to initiate fault injection for the given
+ * object.  Note that 'type' is not strictly necessary for fault injection, but
+ * is used when translating existing faults into a human-readable string.
+ *
+ *
+ * The command itself takes one of the forms:
+ *
+ * 	zinject
+ * 	zinject <-a | -u pool>
+ * 	zinject -c <id|all>
+ * 	zinject [-q] <-t type> [-f freq] [-u] [-a] [-m] [-e errno] [-l level]
+ *	    [-r range] <object>
+ * 	zinject [-f freq] [-a] [-m] [-u] -b objset:object:level:start:end pool
+ *
+ * With no arguments, the command prints all currently registered injection
+ * handlers, with their numeric identifiers.
+ *
+ * The '-c' option will clear the given handler, or all handlers if 'all' is
+ * specified.
+ *
+ * The '-e' option takes a string describing the errno to simulate.  This must
+ * be either 'io' or 'checksum'.  In most cases this will result in the same
+ * behavior, but RAID-Z will produce a different set of ereports for this
+ * situation.
+ *
+ * The '-a', '-u', and '-m' flags toggle internal flush behavior.  If '-a' is
+ * specified, then the ARC cache is flushed appropriately.  If '-u' is
+ * specified, then the underlying SPA is unloaded.  Either of these flags can be
+ * specified independently of any other handlers.  The '-m' flag automatically
+ * does an unmount and remount of the underlying dataset to aid in flushing the
+ * cache.
+ *
+ * The '-f' flag controls the frequency of errors injected, expressed as a
+ * integer percentage between 1 and 100.  The default is 100.
+ *
+ * The this form is responsible for actually injecting the handler into the
+ * framework.  It takes the arguments described above, translates them to the
+ * internal tuple using libzpool, and then issues an ioctl() to register the
+ * handler.
+ *
+ * The final form can target a specific bookmark, regardless of whether a
+ * human-readable interface has been designed.  It allows developers to specify
+ * a particular block by number.
+ */
+
+#include <errno.h>
+#include <fcntl.h>
+#include <stdio.h>
+#include <stdlib.h>
+#include <strings.h>
+#include <unistd.h>
+
+#include <sys/fs/zfs.h>
+#include <sys/mount.h>
+
+#include <libzfs.h>
+
+#undef verify	/* both libzfs.h and zfs_context.h want to define this */
+
+#include "zinject.h"
+
+int zfs_fd;
+
+#define	ECKSUM	EBADE
+
+static const char *errtable[TYPE_INVAL] = {
+	"data",
+	"dnode",
+	"mos",
+	"mosdir",
+	"metaslab",
+	"config",
+	"bplist",
+	"spacemap",
+	"errlog"
+};
+
+static err_type_t
+name_to_type(const char *arg)
+{
+	int i;
+	for (i = 0; i < TYPE_INVAL; i++)
+		if (strcmp(errtable[i], arg) == 0)
+			return (i);
+
+	return (TYPE_INVAL);
+}
+
+static const char *
+type_to_name(uint64_t type)
+{
+	switch (type) {
+	case DMU_OT_OBJECT_DIRECTORY:
+		return ("mosdir");
+	case DMU_OT_OBJECT_ARRAY:
+		return ("metaslab");
+	case DMU_OT_PACKED_NVLIST:
+		return ("config");
+	case DMU_OT_BPLIST:
+		return ("bplist");
+	case DMU_OT_SPACE_MAP:
+		return ("spacemap");
+	case DMU_OT_ERROR_LOG:
+		return ("errlog");
+	default:
+		return ("-");
+	}
+}
+
+
+/*
+ * Print usage message.
+ */
+void
+usage(void)
+{
+	(void) printf(
+	    "usage:\n"
+	    "\n"
+	    "\tzinject\n"
+	    "\n"
+	    "\t\tList all active injection records.\n"
+	    "\n"
+	    "\tzinject -c <id|all>\n"
+	    "\n"
+	    "\t\tClear the particular record (if given a numeric ID), or\n"
+	    "\t\tall records if 'all' is specificed.\n"
+	    "\n"
+	    "\tzinject -d device [-e errno] pool\n"
+	    "\t\tInject a fault into a particular device.  'errno' can either\n"
+	    "\t\tbe 'nxio' (the default) or 'io'.\n"
+	    "\n"
+	    "\tzinject -b objset:object:level:blkid pool\n"
+	    "\n"
+	    "\t\tInject an error into pool 'pool' with the numeric bookmark\n"
+	    "\t\tspecified by the remaining tuple.  Each number is in\n"
+	    "\t\thexidecimal, and only one block can be specified.\n"
+	    "\n"
+	    "\tzinject [-q] <-t type> [-e errno] [-l level] [-r range]\n"
+	    "\t    [-a] [-m] [-u] [-f freq] <object>\n"
+	    "\n"
+	    "\t\tInject an error into the object specified by the '-t' option\n"
+	    "\t\tand the object descriptor.  The 'object' parameter is\n"
+	    "\t\tinterperted depending on the '-t' option.\n"
+	    "\n"
+	    "\t\t-q\tQuiet mode.  Only print out the handler number added.\n"
+	    "\t\t-e\tInject a specific error.  Must be either 'io' or\n"
+	    "\t\t\t'checksum'.  Default is 'io'.\n"
+	    "\t\t-l\tInject error at a particular block level. Default is "
+	    "0.\n"
+	    "\t\t-m\tAutomatically remount underlying filesystem.\n"
+	    "\t\t-r\tInject error over a particular logical range of an\n"
+	    "\t\t\tobject.  Will be translated to the appropriate blkid\n"
+	    "\t\t\trange according to the object's properties.\n"
+	    "\t\t-a\tFlush the ARC cache.  Can be specified without any\n"
+	    "\t\t\tassociated object.\n"
+	    "\t\t-u\tUnload the associated pool.  Can be specified with only\n"
+	    "\t\t\ta pool object.\n"
+	    "\t\t-f\tOnly inject errors a fraction of the time.  Expressed as\n"
+	    "\t\t\ta percentage between 1 and 100.\n"
+	    "\n"
+	    "\t-t data\t\tInject an error into the plain file contents of a\n"
+	    "\t\t\tfile.  The object must be specified as a complete path\n"
+	    "\t\t\tto a file on a ZFS filesystem.\n"
+	    "\n"
+	    "\t-t dnode\tInject an error into the metadnode in the block\n"
+	    "\t\t\tcorresponding to the dnode for a file or directory.  The\n"
+	    "\t\t\t'-r' option is incompatible with this mode.  The object\n"
+	    "\t\t\tis specified as a complete path to a file or directory\n"
+	    "\t\t\ton a ZFS filesystem.\n"
+	    "\n"
+	    "\t-t <mos>\tInject errors into the MOS for objects of the given\n"
+	    "\t\t\ttype.  Valid types are: mos, mosdir, config, bplist,\n"
+	    "\t\t\tspacemap, metaslab, errlog\n");
+}
+
+static int
+iter_handlers(int (*func)(int, const char *, zinject_record_t *, void *),
+    void *data)
+{
+	zfs_cmd_t zc;
+	int ret;
+
+	zc.zc_guid = 0;
+
+	while (ioctl(zfs_fd, ZFS_IOC_INJECT_LIST_NEXT, &zc) == 0)
+		if ((ret = func((int)zc.zc_guid, zc.zc_name,
+		    &zc.zc_inject_record, data)) != 0)
+			return (ret);
+
+	return (0);
+}
+
+static int
+print_data_handler(int id, const char *pool, zinject_record_t *record,
+    void *data)
+{
+	int *count = data;
+
+	if (record->zi_guid != 0)
+		return (0);
+
+	if (*count == 0) {
+		(void) printf("%3s  %-15s  %-6s  %-6s  %-8s  %3s  %-15s\n",
+		    "ID", "POOL", "OBJSET", "OBJECT", "TYPE", "LVL",  "RANGE");
+		(void) printf("---  ---------------  ------  "
+		    "------  --------  ---  ---------------\n");
+	}
+
+	*count += 1;
+
+	(void) printf("%3d  %-15s  %-6llu  %-6llu  %-8s  %3d  ", id, pool,
+	    (u_longlong_t)record->zi_objset, (u_longlong_t)record->zi_object,
+	    type_to_name(record->zi_type), record->zi_level);
+
+	if (record->zi_start == 0 &&
+	    record->zi_end == -1ULL)
+		(void) printf("all\n");
+	else
+		(void) printf("[%llu, %llu]\n", (u_longlong_t)record->zi_start,
+		    (u_longlong_t)record->zi_end);
+
+	return (0);
+}
+
+static int
+print_device_handler(int id, const char *pool, zinject_record_t *record,
+    void *data)
+{
+	int *count = data;
+
+	if (record->zi_guid == 0)
+		return (0);
+
+	if (*count == 0) {
+		(void) printf("%3s  %-15s  %s\n", "ID", "POOL", "GUID");
+		(void) printf("---  ---------------  ----------------\n");
+	}
+
+	*count += 1;
+
+	(void) printf("%3d  %-15s  %llx\n", id, pool,
+	    (u_longlong_t)record->zi_guid);
+
+	return (0);
+}
+
+/*
+ * Print all registered error handlers.  Returns the number of handlers
+ * registered.
+ */
+static int
+print_all_handlers(void)
+{
+	int count = 0;
+
+	(void) iter_handlers(print_device_handler, &count);
+	(void) printf("\n");
+	count = 0;
+	(void) iter_handlers(print_data_handler, &count);
+
+	return (count);
+}
+
+/* ARGSUSED */
+static int
+cancel_one_handler(int id, const char *pool, zinject_record_t *record,
+    void *data)
+{
+	zfs_cmd_t zc;
+
+	zc.zc_guid = (uint64_t)id;
+
+	if (ioctl(zfs_fd, ZFS_IOC_CLEAR_FAULT, &zc) != 0) {
+		(void) fprintf(stderr, "failed to remove handler %d: %s\n",
+		    id, strerror(errno));
+		return (1);
+	}
+
+	return (0);
+}
+
+/*
+ * Remove all fault injection handlers.
+ */
+static int
+cancel_all_handlers(void)
+{
+	int ret = iter_handlers(cancel_one_handler, NULL);
+
+	(void) printf("removed all registered handlers\n");
+
+	return (ret);
+}
+
+/*
+ * Remove a specific fault injection handler.
+ */
+static int
+cancel_handler(int id)
+{
+	zfs_cmd_t zc;
+
+	zc.zc_guid = (uint64_t)id;
+
+	if (ioctl(zfs_fd, ZFS_IOC_CLEAR_FAULT, &zc) != 0) {
+		(void) fprintf(stderr, "failed to remove handler %d: %s\n",
+		    id, strerror(errno));
+		return (1);
+	}
+
+	(void) printf("removed handler %d\n", id);
+
+	return (0);
+}
+
+/*
+ * Register a new fault injection handler.
+ */
+static int
+register_handler(const char *pool, int flags, zinject_record_t *record,
+    int quiet)
+{
+	zfs_cmd_t zc;
+
+	(void) strcpy(zc.zc_name, pool);
+	zc.zc_inject_record = *record;
+	zc.zc_guid = flags;
+
+	if (ioctl(zfs_fd, ZFS_IOC_INJECT_FAULT, &zc) != 0) {
+		(void) fprintf(stderr, "failed to add handler: %s\n",
+		    strerror(errno));
+		return (1);
+	}
+
+	if (flags & ZINJECT_NULL)
+		return (0);
+
+	if (quiet) {
+		(void) printf("%llu\n", (u_longlong_t)zc.zc_guid);
+	} else {
+		(void) printf("Added handler %llu with the following "
+		    "properties:\n", (u_longlong_t)zc.zc_guid);
+		(void) printf("  pool: %s\n", pool);
+		if (record->zi_guid) {
+			(void) printf("  vdev: %llx\n",
+			    (u_longlong_t)record->zi_guid);
+		} else {
+			(void) printf("objset: %llu\n",
+			    (u_longlong_t)record->zi_objset);
+			(void) printf("object: %llu\n",
+			    (u_longlong_t)record->zi_object);
+			(void) printf("  type: %llu\n",
+			    (u_longlong_t)record->zi_type);
+			(void) printf(" level: %d\n", record->zi_level);
+			if (record->zi_start == 0 &&
+			    record->zi_end == -1ULL)
+				(void) printf(" range: all\n");
+			else
+				(void) printf(" range: [%llu, %llu)\n",
+				    (u_longlong_t)record->zi_start,
+				    (u_longlong_t)record->zi_end);
+		}
+	}
+
+	return (0);
+}
+
+int
+main(int argc, char **argv)
+{
+	int c;
+	char *range = NULL;
+	char *cancel = NULL;
+	char *end;
+	char *raw = NULL;
+	char *device = NULL;
+	int level = 0;
+	int quiet = 0;
+	int error = 0;
+	int domount = 0;
+	err_type_t type = TYPE_INVAL;
+	zinject_record_t record = { 0 };
+	char pool[MAXNAMELEN];
+	char dataset[MAXNAMELEN];
+	zfs_handle_t *zhp;
+	int ret;
+	int flags = 0;
+
+	if ((zfs_fd = open(ZFS_DEV, O_RDWR)) < 0) {
+		(void) fprintf(stderr, "failed to open ZFS device\n");
+		return (1);
+	}
+
+	if (argc == 1) {
+		/*
+		 * No arguments.  Print the available handlers.  If there are no
+		 * available handlers, direct the user to '-h' for help
+		 * information.
+		 */
+		if (print_all_handlers() == 0) {
+			(void) printf("No handlers registered.\n");
+			(void) printf("Run 'zinject -h' for usage "
+			    "information.\n");
+		}
+
+		return (0);
+	}
+
+	while ((c = getopt(argc, argv, ":ab:d:f:qhc:t:l:mr:e:u")) != -1) {
+		switch (c) {
+		case 'a':
+			flags |= ZINJECT_FLUSH_ARC;
+			break;
+		case 'b':
+			raw = optarg;
+			break;
+		case 'c':
+			cancel = optarg;
+			break;
+		case 'd':
+			device = optarg;
+			break;
+		case 'e':
+			if (strcasecmp(optarg, "io") == 0) {
+				error = EIO;
+			} else if (strcasecmp(optarg, "checksum") == 0) {
+				error = ECKSUM;
+			} else if (strcasecmp(optarg, "nxio") == 0) {
+				error = ENXIO;
+			} else {
+				(void) fprintf(stderr, "invalid error type "
+				    "'%s': must be 'io', 'checksum' or "
+				    "'nxio'\n", optarg);
+				usage();
+				return (1);
+			}
+			break;
+		case 'f':
+			record.zi_freq = atoi(optarg);
+			if (record.zi_freq < 1 || record.zi_freq > 100) {
+				(void) fprintf(stderr, "frequency range must "
+				    "be in the range (0, 100]\n");
+				return (1);
+			}
+			break;
+		case 'h':
+			usage();
+			return (0);
+		case 'l':
+			level = (int)strtol(optarg, &end, 10);
+			if (*end != '\0') {
+				(void) fprintf(stderr, "invalid level '%s': "
+				    "must be an integer\n", optarg);
+				usage();
+				return (1);
+			}
+			break;
+		case 'm':
+			domount = 1;
+			break;
+		case 'q':
+			quiet = 1;
+			break;
+		case 'r':
+			range = optarg;
+			break;
+		case 't':
+			if ((type = name_to_type(optarg)) == TYPE_INVAL) {
+				(void) fprintf(stderr, "invalid type '%s'\n",
+				    optarg);
+				usage();
+				return (1);
+			}
+			break;
+		case 'u':
+			flags |= ZINJECT_UNLOAD_SPA;
+			break;
+		case ':':
+			(void) fprintf(stderr, "option -%c requires an "
+			    "operand\n", optopt);
+			usage();
+			return (1);
+		case '?':
+			(void) fprintf(stderr, "invalid option '%c'\n",
+			    optopt);
+			usage();
+			return (2);
+		}
+	}
+
+	argc -= optind;
+	argv += optind;
+
+	if (cancel != NULL) {
+		/*
+		 * '-c' is invalid with any other options.
+		 */
+		if (raw != NULL || range != NULL || type != TYPE_INVAL ||
+		    level != 0) {
+			(void) fprintf(stderr, "cancel (-c) incompatible with "
+			    "any other options\n");
+			usage();
+			return (2);
+		}
+		if (argc != 0) {
+			(void) fprintf(stderr, "extraneous argument to '-c'\n");
+			usage();
+			return (2);
+		}
+
+		if (strcmp(cancel, "all") == 0) {
+			return (cancel_all_handlers());
+		} else {
+			int id = (int)strtol(cancel, &end, 10);
+			if (*end != '\0') {
+				(void) fprintf(stderr, "invalid handle id '%s':"
+				    " must be an integer or 'all'\n", cancel);
+				usage();
+				return (1);
+			}
+			return (cancel_handler(id));
+		}
+	}
+
+	if (device != NULL) {
+		/*
+		 * Device (-d) injection uses a completely different mechanism
+		 * for doing injection, so handle it separately here.
+		 */
+		if (raw != NULL || range != NULL || type != TYPE_INVAL ||
+		    level != 0) {
+			(void) fprintf(stderr, "device (-d) incompatible with "
+			    "data error injection\n");
+			usage();
+			return (2);
+		}
+
+		if (argc != 1) {
+			(void) fprintf(stderr, "device (-d) injection requires "
+			    "a single pool name\n");
+			usage();
+			return (2);
+		}
+
+		(void) strcpy(pool, argv[0]);
+		dataset[0] = '\0';
+
+		if (error == ECKSUM) {
+			(void) fprintf(stderr, "device error type must be "
+			    "'io' or 'nxio'\n");
+			return (1);
+		}
+
+		if (translate_device(pool, device, &record) != 0)
+			return (1);
+		if (!error)
+			error = ENXIO;
+	} else if (raw != NULL) {
+		if (range != NULL || type != TYPE_INVAL || level != 0) {
+			(void) fprintf(stderr, "raw (-b) format with "
+			    "any other options\n");
+			usage();
+			return (2);
+		}
+
+		if (argc != 1) {
+			(void) fprintf(stderr, "raw (-b) format expects a "
+			    "single pool name\n");
+			usage();
+			return (2);
+		}
+
+		(void) strcpy(pool, argv[0]);
+		dataset[0] = '\0';
+
+		if (error == ENXIO) {
+			(void) fprintf(stderr, "data error type must be "
+			    "'checksum' or 'io'\n");
+			return (1);
+		}
+
+		if (translate_raw(raw, &record) != 0)
+			return (1);
+		if (!error)
+			error = EIO;
+	} else if (type == TYPE_INVAL) {
+		if (flags == 0) {
+			(void) fprintf(stderr, "at least one of '-b', '-d', "
+			    "'-t', '-a', or '-u' must be specified\n");
+			usage();
+			return (2);
+		}
+
+		if (argc == 1 && (flags & ZINJECT_UNLOAD_SPA)) {
+			(void) strcpy(pool, argv[0]);
+			dataset[0] = '\0';
+		} else if (argc != 0) {
+			(void) fprintf(stderr, "extraneous argument for "
+			    "'-f'\n");
+			usage();
+			return (2);
+		}
+
+		flags |= ZINJECT_NULL;
+	} else {
+		if (argc != 1) {
+			(void) fprintf(stderr, "missing object\n");
+			usage();
+			return (2);
+		}
+
+		if (error == ENXIO) {
+			(void) fprintf(stderr, "data error type must be "
+			    "'checksum' or 'io'\n");
+			return (1);
+		}
+
+		if (translate_record(type, argv[0], range, level, &record, pool,
+		    dataset) != 0)
+			return (1);
+		if (!error)
+			error = EIO;
+	}
+
+	/*
+	 * If this is pool-wide metadata, unmount everything.  The ioctl() will
+	 * unload the pool, so that we trigger spa-wide reopen of metadata next
+	 * time we access the pool.
+	 */
+	if (dataset[0] != '\0' && domount) {
+		if ((zhp = zfs_open(dataset, ZFS_TYPE_ANY)) == NULL)
+			return (1);
+
+		if (zfs_unmount(zhp, NULL, 0) != 0)
+			return (1);
+	}
+
+	record.zi_error = error;
+
+	ret = register_handler(pool, flags, &record, quiet);
+
+	if (dataset[0] != '\0' && domount)
+		ret = (zfs_mount(zhp, NULL, 0) != 0);
+
+	return (ret);
+}
diff --git a/usr/src/cmd/zinject/zinject.h b/usr/src/cmd/zinject/zinject.h
new file mode 100644
index 0000000000..bdbc2454c4
--- /dev/null
+++ b/usr/src/cmd/zinject/zinject.h
@@ -0,0 +1,64 @@
+/*
+ * CDDL HEADER START
+ *
+ * The contents of this file are subject to the terms of the
+ * Common Development and Distribution License (the "License").
+ * You may not use this file except in compliance with the License.
+ *
+ * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
+ * or http://www.opensolaris.org/os/licensing.
+ * See the License for the specific language governing permissions
+ * and limitations under the License.
+ *
+ * When distributing Covered Code, include this CDDL HEADER in each
+ * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
+ * If applicable, add the following below this CDDL HEADER, with the
+ * fields enclosed by brackets "[]" replaced with your own identifying
+ * information: Portions Copyright [yyyy] [name of copyright owner]
+ *
+ * CDDL HEADER END
+ */
+/*
+ * Copyright 2006 Sun Microsystems, Inc.  All rights reserved.
+ * Use is subject to license terms.
+ */
+
+#ifndef	_ZINJECT_H
+#define	_ZINJECT_H
+
+#pragma ident	"%Z%%M%	%I%	%E% SMI"
+
+#include <sys/zfs_ioctl.h>
+
+#ifdef	__cplusplus
+extern "C" {
+#endif
+
+typedef enum {
+	TYPE_DATA,		/* plain file contents		*/
+	TYPE_DNODE,		/* metadnode contents		*/
+	TYPE_MOS,		/* all MOS data			*/
+	TYPE_MOSDIR,		/* MOS object directory		*/
+	TYPE_METASLAB,		/* metaslab objects		*/
+	TYPE_CONFIG,		/* MOS config			*/
+	TYPE_BPLIST,		/* block pointer list		*/
+	TYPE_SPACEMAP,		/* space map objects		*/
+	TYPE_ERRLOG,		/* persistent error log		*/
+	TYPE_INVAL
+} err_type_t;
+
+#define	MOS_TYPE(t)	\
+	((t) >= TYPE_MOS && (t) < TYPE_INVAL)
+
+int translate_record(err_type_t type, const char *object, const char *range,
+    int level, zinject_record_t *record, char *poolname, char *dataset);
+int translate_raw(const char *raw, zinject_record_t *record);
+int translate_device(const char *pool, const char *device,
+    zinject_record_t *record);
+void usage(void);
+
+#ifdef	__cplusplus
+}
+#endif
+
+#endif	/* _ZINJECT_H */