069f55e237020c4a4907b235fc38fafc6442ce94Eric Schrock * CDDL HEADER START
069f55e237020c4a4907b235fc38fafc6442ce94Eric Schrock * The contents of this file are subject to the terms of the
069f55e237020c4a4907b235fc38fafc6442ce94Eric Schrock * Common Development and Distribution License (the "License").
069f55e237020c4a4907b235fc38fafc6442ce94Eric Schrock * You may not use this file except in compliance with the License.
069f55e237020c4a4907b235fc38fafc6442ce94Eric Schrock * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
069f55e237020c4a4907b235fc38fafc6442ce94Eric Schrock * See the License for the specific language governing permissions
069f55e237020c4a4907b235fc38fafc6442ce94Eric Schrock * and limitations under the License.
069f55e237020c4a4907b235fc38fafc6442ce94Eric Schrock * When distributing Covered Code, include this CDDL HEADER in each
069f55e237020c4a4907b235fc38fafc6442ce94Eric Schrock * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
069f55e237020c4a4907b235fc38fafc6442ce94Eric Schrock * If applicable, add the following below this CDDL HEADER, with the
069f55e237020c4a4907b235fc38fafc6442ce94Eric Schrock * fields enclosed by brackets "[]" replaced with your own identifying
069f55e237020c4a4907b235fc38fafc6442ce94Eric Schrock * information: Portions Copyright [yyyy] [name of copyright owner]
069f55e237020c4a4907b235fc38fafc6442ce94Eric Schrock * CDDL HEADER END
069f55e237020c4a4907b235fc38fafc6442ce94Eric Schrock * Copyright 2009 Sun Microsystems, Inc. All rights reserved.
069f55e237020c4a4907b235fc38fafc6442ce94Eric Schrock * Use is subject to license terms.
069f55e237020c4a4907b235fc38fafc6442ce94Eric Schrock * This file is responsible for determining the relationship between I/O
069f55e237020c4a4907b235fc38fafc6442ce94Eric Schrock * devices paths and physical locations. In the world of MPxIO and external
069f55e237020c4a4907b235fc38fafc6442ce94Eric Schrock * enclosures, the device path is not synonymous with the physical location.
069f55e237020c4a4907b235fc38fafc6442ce94Eric Schrock * If you remove a drive and insert it into a different slot, it will end up
069f55e237020c4a4907b235fc38fafc6442ce94Eric Schrock * with the same path under MPxIO. If you recable storage enclosures, the
069f55e237020c4a4907b235fc38fafc6442ce94Eric Schrock * device paths may change. All of this makes it difficult to implement the
069f55e237020c4a4907b235fc38fafc6442ce94Eric Schrock * 'autoreplace' property, which is supposed to automatically manage disk
069f55e237020c4a4907b235fc38fafc6442ce94Eric Schrock * replacement based on physical slot.
069f55e237020c4a4907b235fc38fafc6442ce94Eric Schrock * In order to work around these limitations, we have a per-vdev FRU property
069f55e237020c4a4907b235fc38fafc6442ce94Eric Schrock * that is the libtopo path (minus disk-specific authority information) to the
069f55e237020c4a4907b235fc38fafc6442ce94Eric Schrock * physical location of the device on the system. This is an optional
069f55e237020c4a4907b235fc38fafc6442ce94Eric Schrock * property, and is only needed when using the 'autoreplace' property or when
069f55e237020c4a4907b235fc38fafc6442ce94Eric Schrock * generating FMA faults against vdevs.
069f55e237020c4a4907b235fc38fafc6442ce94Eric Schrock * Because the FMA packages depend on ZFS, we have to dlopen() libtopo in case
069f55e237020c4a4907b235fc38fafc6442ce94Eric Schrock * it is not present. We only need this once per library instance, so it is
069f55e237020c4a4907b235fc38fafc6442ce94Eric Schrock * not part of the libzfs handle.
069f55e237020c4a4907b235fc38fafc6442ce94Eric Schrockstatic topo_hdl_t *(*_topo_open)(int, const char *, int *);
069f55e237020c4a4907b235fc38fafc6442ce94Eric Schrockstatic char *(*_topo_snap_hold)(topo_hdl_t *, const char *, int *);
069f55e237020c4a4907b235fc38fafc6442ce94Eric Schrockstatic void (*_topo_snap_release)(topo_hdl_t *);
069f55e237020c4a4907b235fc38fafc6442ce94Eric Schrockstatic topo_walk_t *(*_topo_walk_init)(topo_hdl_t *, const char *,
069f55e237020c4a4907b235fc38fafc6442ce94Eric Schrock topo_walk_cb_t, void *, int *);
069f55e237020c4a4907b235fc38fafc6442ce94Eric Schrockstatic int (*_topo_walk_step)(topo_walk_t *, int);
069f55e237020c4a4907b235fc38fafc6442ce94Eric Schrockstatic void (*_topo_hdl_strfree)(topo_hdl_t *, char *);
069f55e237020c4a4907b235fc38fafc6442ce94Eric Schrockstatic int (*_topo_prop_get_string)(tnode_t *, const char *, const char *,
069f55e237020c4a4907b235fc38fafc6442ce94Eric Schrock char **, int *);
069f55e237020c4a4907b235fc38fafc6442ce94Eric Schrockstatic int (*_topo_node_fru)(tnode_t *, nvlist_t **, nvlist_t *, int *);
069f55e237020c4a4907b235fc38fafc6442ce94Eric Schrockstatic int (*_topo_fmri_nvl2str)(topo_hdl_t *, nvlist_t *, char **, int *);
069f55e237020c4a4907b235fc38fafc6442ce94Eric Schrockstatic int (*_topo_fmri_strcmp_noauth)(topo_hdl_t *, const char *,
069f55e237020c4a4907b235fc38fafc6442ce94Eric Schrock const char *);
069f55e237020c4a4907b235fc38fafc6442ce94Eric Schrock const char *p;
069f55e237020c4a4907b235fc38fafc6442ce94Eric Schrock h = (h << 4) + *p;
069f55e237020c4a4907b235fc38fafc6442ce94Eric Schrock if ((g = (h & 0xf0000000)) != 0) {
069f55e237020c4a4907b235fc38fafc6442ce94Eric Schrock h ^= (g >> 24);
069f55e237020c4a4907b235fc38fafc6442ce94Eric Schrocklibzfs_fru_gather(topo_hdl_t *thp, tnode_t *tn, void *arg)
069f55e237020c4a4907b235fc38fafc6442ce94Eric Schrock * If this is the chassis node, and we don't yet have the system
069f55e237020c4a4907b235fc38fafc6442ce94Eric Schrock * chassis ID, then fill in this value now.
069f55e237020c4a4907b235fc38fafc6442ce94Eric Schrock if (_topo_prop_get_string(tn, FM_FMRI_AUTHORITY,
069f55e237020c4a4907b235fc38fafc6442ce94Eric Schrock (void) strlcpy(hdl->libzfs_chassis_id, devpath,
069f55e237020c4a4907b235fc38fafc6442ce94Eric Schrock * Skip non-disk nodes.
069f55e237020c4a4907b235fc38fafc6442ce94Eric Schrock * Get the devfs path and FRU.
069f55e237020c4a4907b235fc38fafc6442ce94Eric Schrock if (_topo_prop_get_string(tn, "io", "devfs-path", &devpath, &err) != 0)
069f55e237020c4a4907b235fc38fafc6442ce94Eric Schrock if (_topo_node_fru(tn, &fru, NULL, &err) != 0) {
069f55e237020c4a4907b235fc38fafc6442ce94Eric Schrock * Convert the FRU into a string.
069f55e237020c4a4907b235fc38fafc6442ce94Eric Schrock if (_topo_fmri_nvl2str(thp, fru, &frustr, &err) != 0) {
069f55e237020c4a4907b235fc38fafc6442ce94Eric Schrock * Finally, we have a FRU string and device path. Add it to the hash.
069f55e237020c4a4907b235fc38fafc6442ce94Eric Schrock if ((frup = calloc(sizeof (libzfs_fru_t), 1)) == NULL) {
069f55e237020c4a4907b235fc38fafc6442ce94Eric Schrock if ((frup->zf_device = strdup(devpath)) == NULL ||
069f55e237020c4a4907b235fc38fafc6442ce94Eric Schrock * Called during initialization to setup the dynamic libtopo connection.
069f55e237020c4a4907b235fc38fafc6442ce94Eric Schrock if (sysinfo(SI_ARCHITECTURE_64, isa, sizeof (isa)) < 0)
069f55e237020c4a4907b235fc38fafc6442ce94Eric Schrock if ((_topo_dlhandle = dlopen(path, RTLD_LAZY)) == NULL)
069f55e237020c4a4907b235fc38fafc6442ce94Eric Schrock _topo_close = (void (*)())
069f55e237020c4a4907b235fc38fafc6442ce94Eric Schrock _topo_snap_hold = (char *(*)())
069f55e237020c4a4907b235fc38fafc6442ce94Eric Schrock _topo_node_name = (char *(*)())
069f55e237020c4a4907b235fc38fafc6442ce94Eric Schrock dlsym(_topo_dlhandle, "topo_fmri_strcmp_noauth");
069f55e237020c4a4907b235fc38fafc6442ce94Eric Schrock if (_topo_open == NULL || _topo_close == NULL ||
069f55e237020c4a4907b235fc38fafc6442ce94Eric Schrock _topo_snap_hold == NULL || _topo_snap_release == NULL ||
069f55e237020c4a4907b235fc38fafc6442ce94Eric Schrock _topo_walk_init == NULL || _topo_walk_step == NULL ||
069f55e237020c4a4907b235fc38fafc6442ce94Eric Schrock _topo_walk_fini == NULL || _topo_hdl_strfree == NULL ||
069f55e237020c4a4907b235fc38fafc6442ce94Eric Schrock _topo_node_name == NULL || _topo_prop_get_string == NULL ||
069f55e237020c4a4907b235fc38fafc6442ce94Eric Schrock _topo_node_fru == NULL || _topo_fmri_nvl2str == NULL ||
069f55e237020c4a4907b235fc38fafc6442ce94Eric Schrock * Refresh the mappings from device path -> FMRI. We do this by walking the
069f55e237020c4a4907b235fc38fafc6442ce94Eric Schrock * hc topology looking for disk nodes, and recording the io/devfs-path and FRU.
069f55e237020c4a4907b235fc38fafc6442ce94Eric Schrock * Note that we strip out the disk-specific authority information (serial,
069f55e237020c4a4907b235fc38fafc6442ce94Eric Schrock * part, revision, etc) so that we are left with only the identifying
069f55e237020c4a4907b235fc38fafc6442ce94Eric Schrock * characteristics of the slot (hc path and chassis-id).
069f55e237020c4a4907b235fc38fafc6442ce94Eric Schrock * Clear the FRU hash and initialize our basic structures.
069f55e237020c4a4907b235fc38fafc6442ce94Eric Schrock if ((hdl->libzfs_topo_hdl = _topo_open(TOPO_VERSION,
069f55e237020c4a4907b235fc38fafc6442ce94Eric Schrock if ((uuid = _topo_snap_hold(thp, NULL, &err)) == NULL)
069f55e237020c4a4907b235fc38fafc6442ce94Eric Schrock calloc(ZFS_FRU_HASH_SIZE * sizeof (void *), 1)) == NULL)
069f55e237020c4a4907b235fc38fafc6442ce94Eric Schrock * We now have a topo snapshot, so iterate over the hc topology looking
069f55e237020c4a4907b235fc38fafc6442ce94Eric Schrock * for disks to add to the hash.
069f55e237020c4a4907b235fc38fafc6442ce94Eric Schrock * Given a devfs path, return the FRU for the device, if known. This will
069f55e237020c4a4907b235fc38fafc6442ce94Eric Schrock * automatically call libzfs_fru_refresh() if it hasn't already been called by
069f55e237020c4a4907b235fc38fafc6442ce94Eric Schrock * the consumer. The string returned is valid until the next call to
069f55e237020c4a4907b235fc38fafc6442ce94Eric Schrock * libzfs_fru_refresh().
069f55e237020c4a4907b235fc38fafc6442ce94Eric Schrocklibzfs_fru_lookup(libzfs_handle_t *hdl, const char *devpath)
069f55e237020c4a4907b235fc38fafc6442ce94Eric Schrock for (frup = hdl->libzfs_fru_hash[idx]; frup != NULL;
069f55e237020c4a4907b235fc38fafc6442ce94Eric Schrock * Given a fru path, return the device path. This will automatically call
069f55e237020c4a4907b235fc38fafc6442ce94Eric Schrock * libzfs_fru_refresh() if it hasn't already been called by the consumer. The
069f55e237020c4a4907b235fc38fafc6442ce94Eric Schrock * string returned is valid until the next call to libzfs_fru_refresh().
069f55e237020c4a4907b235fc38fafc6442ce94Eric Schrocklibzfs_fru_devpath(libzfs_handle_t *hdl, const char *fru)
069f55e237020c4a4907b235fc38fafc6442ce94Eric Schrock for (idx = 0; idx < ZFS_FRU_HASH_SIZE; idx++) {
069f55e237020c4a4907b235fc38fafc6442ce94Eric Schrock for (frup = hdl->libzfs_fru_hash[idx]; frup != NULL;
069f55e237020c4a4907b235fc38fafc6442ce94Eric Schrock if (_topo_fmri_strcmp_noauth(hdl->libzfs_topo_hdl,
069f55e237020c4a4907b235fc38fafc6442ce94Eric Schrock * Change the stored FRU for the given vdev.
069f55e237020c4a4907b235fc38fafc6442ce94Eric Schrockzpool_fru_set(zpool_handle_t *zhp, uint64_t vdev_guid, const char *fru)
069f55e237020c4a4907b235fc38fafc6442ce94Eric Schrock (void) strncpy(zc.zc_name, zhp->zpool_name, sizeof (zc.zc_name));
069f55e237020c4a4907b235fc38fafc6442ce94Eric Schrock (void) strncpy(zc.zc_value, fru, sizeof (zc.zc_value));
069f55e237020c4a4907b235fc38fafc6442ce94Eric Schrock if (zfs_ioctl(zhp->zpool_hdl, ZFS_IOC_VDEV_SETFRU, &zc) != 0)
069f55e237020c4a4907b235fc38fafc6442ce94Eric Schrock return (zpool_standard_error_fmt(zhp->zpool_hdl, errno,
069f55e237020c4a4907b235fc38fafc6442ce94Eric Schrock * Compare to two FRUs, ignoring any authority information.
069f55e237020c4a4907b235fc38fafc6442ce94Eric Schrocklibzfs_fru_compare(libzfs_handle_t *hdl, const char *a, const char *b)
069f55e237020c4a4907b235fc38fafc6442ce94Eric Schrock return (strcmp(a, b) == 0);
069f55e237020c4a4907b235fc38fafc6442ce94Eric Schrock return (_topo_fmri_strcmp_noauth(hdl->libzfs_topo_hdl, a, b));
069f55e237020c4a4907b235fc38fafc6442ce94Eric Schrock * This special function checks to see whether the FRU indicates it's supposed
069f55e237020c4a4907b235fc38fafc6442ce94Eric Schrock * to be in the system chassis, but the chassis-id doesn't match. This can
069f55e237020c4a4907b235fc38fafc6442ce94Eric Schrock * happen in a clustered case, where both head nodes have the same logical
069f55e237020c4a4907b235fc38fafc6442ce94Eric Schrock * disk, but opening the device on the other head node is meaningless.
069f55e237020c4a4907b235fc38fafc6442ce94Eric Schrocklibzfs_fru_notself(libzfs_handle_t *hdl, const char *fru)
069f55e237020c4a4907b235fc38fafc6442ce94Eric Schrock if ((chassisid = strstr(fru, ":chassis-id=")) == NULL)
069f55e237020c4a4907b235fc38fafc6442ce94Eric Schrock if (strncmp(chassisid, hdl->libzfs_chassis_id, len) == 0 &&
069f55e237020c4a4907b235fc38fafc6442ce94Eric Schrock (chassisid[len] == '/' || chassisid[len] == ':'))
069f55e237020c4a4907b235fc38fafc6442ce94Eric Schrock * Clear memory associated with the FRU hash.
069f55e237020c4a4907b235fc38fafc6442ce94Eric Schrocklibzfs_fru_clear(libzfs_handle_t *hdl, boolean_t final)
069f55e237020c4a4907b235fc38fafc6442ce94Eric Schrock while ((frup = hdl->libzfs_fru_list) != NULL) {
069f55e237020c4a4907b235fc38fafc6442ce94Eric Schrock ZFS_FRU_HASH_SIZE * sizeof (void *));