/*
* Copyright 2015 Nexenta Systems, Inc. All rights reserved.
* Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved.
- * Copyright (c) 2012 by Delphix. All rights reserved.
+ * Copyright (c) 2012, 2018 by Delphix. All rights reserved.
* Copyright 2015 RackTop Systems.
* Copyright (c) 2016, Intel Corporation.
*/
#include <blkid/blkid.h>
#include "libzfs.h"
#include "libzfs_impl.h"
+#include <libzfs.h>
/*
* Intermediate structures used to gather configuration information.
/*
* Obtain the persistent device id string (describes what)
*
- * used by ZED auto-{online,expand,replace}
+ * used by ZED vdev matching for auto-{online,expand,replace}
*/
-static int
-udev_device_get_devid(struct udev_device *dev, char *bufptr, size_t buflen)
+int
+zfs_device_get_devid(struct udev_device *dev, char *bufptr, size_t buflen)
{
struct udev_list_entry *entry;
const char *bus;
(void) snprintf(bufptr, buflen, "dm-uuid-%s", dm_uuid);
return (0);
}
- return (ENODATA);
+
+ /*
+ * NVME 'by-id' symlinks are similar to bus case
+ */
+ struct udev_device *parent;
+
+ parent = udev_device_get_parent_with_subsystem_devtype(dev,
+ "nvme", NULL);
+ if (parent != NULL)
+ bus = "nvme"; /* continue with bus symlink search */
+ else
+ return (ENODATA);
}
/*
/*
* Obtain the persistent physical location string (describes where)
*
- * used by ZED auto-{online,expand,replace}
+ * used by ZED vdev matching for auto-{online,expand,replace}
*/
-static int
-udev_device_get_physical(struct udev_device *dev, char *bufptr, size_t buflen)
+int
+zfs_device_get_physical(struct udev_device *dev, char *bufptr, size_t buflen)
{
- const char *physpath, *value;
+ const char *physpath = NULL;
/*
- * Skip indirect multipath device nodes
+ * Normal disks use ID_PATH for their physical path. Device mapper
+ * devices are virtual and don't have a physical path. For them we
+ * use ID_VDEV instead, which is setup via the /etc/vdev_id.conf file.
+ * ID_VDEV provides a persistent path to a virtual device. If you
+ * don't have vdev_id.conf setup, you cannot use multipath autoreplace.
*/
- value = udev_device_get_property_value(dev, "DM_MULTIPATH_DEVICE_PATH");
- if (value != NULL && strcmp(value, "1") == 0)
- return (ENODATA); /* skip physical for multipath nodes */
-
- physpath = udev_device_get_property_value(dev, "ID_PATH");
- if (physpath != NULL && physpath[0] != '\0') {
- (void) strlcpy(bufptr, physpath, buflen);
- return (0);
+ if (!((physpath = udev_device_get_property_value(dev, "ID_PATH")) &&
+ physpath[0])) {
+ if (!((physpath =
+ udev_device_get_property_value(dev, "ID_VDEV")) &&
+ physpath[0])) {
+ return (ENODATA);
+ }
}
- return (ENODATA);
+ (void) strlcpy(bufptr, physpath, buflen);
+
+ return (0);
+}
+
+boolean_t
+udev_is_mpath(struct udev_device *dev)
+{
+ return udev_device_get_property_value(dev, "DM_UUID") &&
+ udev_device_get_property_value(dev, "MPATH_SBIN_PATH");
}
/*
static boolean_t
udev_mpath_whole_disk(struct udev_device *dev)
{
- const char *devname, *mapname, *type, *uuid;
+ const char *devname, *type, *uuid;
devname = udev_device_get_property_value(dev, "DEVNAME");
- mapname = udev_device_get_property_value(dev, "DM_NAME");
type = udev_device_get_property_value(dev, "ID_PART_TABLE_TYPE");
uuid = udev_device_get_property_value(dev, "DM_UUID");
if ((devname != NULL && strncmp(devname, "/dev/dm-", 8) == 0) &&
- (mapname != NULL && strncmp(mapname, "mpath", 5) == 0) &&
((type == NULL) || (strcmp(type, "gpt") != 0)) &&
(uuid != NULL)) {
return (B_TRUE);
dev = udev_device_new_from_subsystem_sysname(udev,
"block", sysname);
if ((dev != NULL) && udev_device_is_ready(dev)) {
- struct udev_list_entry *links, *link;
+ struct udev_list_entry *links, *link = NULL;
ret = 0;
links = udev_device_get_devlinks_list_entry(dev);
if (!wholedisk && !udev_mpath_whole_disk(dev))
goto no_dev;
- ret = udev_device_get_devid(dev, ds->vds_devid, sizeof (ds->vds_devid));
+ ret = zfs_device_get_devid(dev, ds->vds_devid, sizeof (ds->vds_devid));
if (ret != 0)
goto no_dev_ref;
/* physical location string (optional) */
- if (udev_device_get_physical(dev, ds->vds_devphys,
+ if (zfs_device_get_physical(dev, ds->vds_devphys,
sizeof (ds->vds_devphys)) != 0) {
ds->vds_devphys[0] = '\0'; /* empty string --> not available */
}
*
* multipath device node example:
* devid: 'dm-uuid-mpath-35000c5006304de3f'
+ *
+ * We also store the enclosure sysfs path for turning on enclosure LEDs
+ * (if applicable):
+ * vdev_enc_sysfs_path: '/sys/class/enclosure/11:0:1:0/SLOT 4'
*/
void
update_vdev_config_dev_strs(nvlist_t *nv)
vdev_dev_strs_t vds;
char *env, *type, *path;
uint64_t wholedisk = 0;
+ char *upath, *spath;
/*
* For the benefit of legacy ZFS implementations, allow
(void) nvlist_add_string(nv, ZPOOL_CONFIG_PHYS_PATH,
vds.vds_devphys);
}
+
+ /* Add enclosure sysfs path (if disk is in an enclosure) */
+ upath = zfs_get_underlying_path(path);
+ spath = zfs_get_enclosure_sysfs_path(upath);
+ if (spath)
+ nvlist_add_string(nv, ZPOOL_CONFIG_VDEV_ENC_SYSFS_PATH,
+ spath);
+ else
+ nvlist_remove_all(nv, ZPOOL_CONFIG_VDEV_ENC_SYSFS_PATH);
+
+ free(upath);
+ free(spath);
} else {
/* clear out any stale entries */
(void) nvlist_remove_all(nv, ZPOOL_CONFIG_DEVID);
(void) nvlist_remove_all(nv, ZPOOL_CONFIG_PHYS_PATH);
+ (void) nvlist_remove_all(nv, ZPOOL_CONFIG_VDEV_ENC_SYSFS_PATH);
}
}
#else
&state) == 0 &&
(state == POOL_STATE_SPARE || state == POOL_STATE_L2CACHE) &&
nvlist_lookup_uint64(config, ZPOOL_CONFIG_GUID, &vdev_guid) == 0) {
- if ((ne = zfs_alloc(hdl, sizeof (name_entry_t))) == NULL)
+ if ((ne = zfs_alloc(hdl, sizeof (name_entry_t))) == NULL) {
+ nvlist_free(config);
return (-1);
+ }
if ((ne->ne_name = zfs_strdup(hdl, path)) == NULL) {
free(ne);
+ nvlist_free(config);
return (-1);
}
ne->ne_guid = vdev_guid;
ne->ne_num_labels = num_labels;
ne->ne_next = pl->names;
pl->names = ne;
+ nvlist_free(config);
return (0);
}
{
nvlist_t *nvl;
zfs_cmd_t zc = {"\0"};
- int err;
+ int err, dstbuf_size;
if (zcmd_write_conf_nvlist(hdl, &zc, config) != 0)
return (NULL);
- if (zcmd_alloc_dst_nvlist(hdl, &zc,
- zc.zc_nvlist_conf_size * 2) != 0) {
+ dstbuf_size = MAX(CONFIG_BUF_MINSIZE, zc.zc_nvlist_conf_size * 4);
+
+ if (zcmd_alloc_dst_nvlist(hdl, &zc, dstbuf_size) != 0) {
zcmd_free_nvlists(&zc);
return (NULL);
}
vdev_label_t *label;
nvlist_t *expected_config = NULL;
uint64_t expected_guid = 0, size;
+ int error;
*config = NULL;
return (0);
size = P2ALIGN_TYPED(statbuf.st_size, sizeof (vdev_label_t), uint64_t);
- if ((label = malloc(sizeof (vdev_label_t))) == NULL)
+ error = posix_memalign((void **)&label, PAGESIZE, sizeof (*label));
+ if (error)
return (-1);
for (l = 0; l < VDEV_LABELS; l++) {
char *rn_name; /* Full path to device */
int rn_order; /* Preferred order (low to high) */
int rn_num_labels; /* Number of valid labels */
+ uint64_t rn_vdev_guid; /* Expected vdev guid when set */
libzfs_handle_t *rn_hdl;
nvlist_t *rn_config; /* Label config */
avl_tree_t *rn_avl;
boolean_t rn_labelpaths;
} rdsk_node_t;
+/*
+ * Sorted by vdev guid and full path to allow for multiple entries with
+ * the same full path name. This is required because it's possible to
+ * have multiple block devices with labels that refer to the same
+ * ZPOOL_CONFIG_PATH yet have different vdev guids. In this case both
+ * entries need to be added to the cache. Scenarios where this can occur
+ * include overwritten pool labels, devices which are visible from multiple
+ * hosts and multipath devices.
+ */
static int
slice_cache_compare(const void *arg1, const void *arg2)
{
const char *nm1 = ((rdsk_node_t *)arg1)->rn_name;
const char *nm2 = ((rdsk_node_t *)arg2)->rn_name;
- char *nm1slice, *nm2slice;
+ uint64_t guid1 = ((rdsk_node_t *)arg1)->rn_vdev_guid;
+ uint64_t guid2 = ((rdsk_node_t *)arg2)->rn_vdev_guid;
int rv;
- /*
- * partitions one and three (slices zero and two) are the most
- * likely to provide results, so put those first
- */
- nm1slice = strstr(nm1, "part1");
- nm2slice = strstr(nm2, "part1");
- if (nm1slice && !nm2slice) {
- return (-1);
- }
- if (!nm1slice && nm2slice) {
- return (1);
- }
- nm1slice = strstr(nm1, "part3");
- nm2slice = strstr(nm2, "part3");
- if (nm1slice && !nm2slice) {
- return (-1);
- }
- if (!nm1slice && nm2slice) {
- return (1);
- }
+ rv = AVL_CMP(guid1, guid2);
+ if (rv)
+ return (rv);
- rv = strcmp(nm1, nm2);
- if (rv == 0)
- return (0);
- return (rv > 0 ? 1 : -1);
+ return (AVL_ISIGN(strcmp(nm1, nm2)));
}
static boolean_t
struct stat64 statbuf;
nvlist_t *config;
char *bname, *dupname;
+ uint64_t vdev_guid = 0;
int error;
int num_labels;
int fd;
(!S_ISREG(statbuf.st_mode) && !S_ISBLK(statbuf.st_mode)))
return;
- if ((fd = open(rn->rn_name, O_RDONLY)) < 0)
+ /*
+ * Preferentially open using O_DIRECT to bypass the block device
+ * cache which may be stale for multipath devices. An EINVAL errno
+ * indicates O_DIRECT is unsupported so fallback to just O_RDONLY.
+ */
+ fd = open(rn->rn_name, O_RDONLY | O_DIRECT);
+ if ((fd < 0) && (errno == EINVAL))
+ fd = open(rn->rn_name, O_RDONLY);
+
+ if (fd < 0)
return;
/*
* This file is too small to hold a zpool
*/
- if (S_ISREG(statbuf.st_mode) &&
- statbuf.st_size < SPA_MINDEVSIZE) {
+ if (S_ISREG(statbuf.st_mode) && statbuf.st_size < SPA_MINDEVSIZE) {
(void) close(fd);
return;
}
- if ((zpool_read_label(fd, &config, &num_labels)) != 0) {
+ error = zpool_read_label(fd, &config, &num_labels);
+ if (error != 0) {
(void) close(fd);
return;
}
return;
}
+ /*
+ * Check that the vdev is for the expected guid. Additional entries
+ * are speculatively added based on the paths stored in the labels.
+ * Entries with valid paths but incorrect guids must be removed.
+ */
+ error = nvlist_lookup_uint64(config, ZPOOL_CONFIG_GUID, &vdev_guid);
+ if (error || (rn->rn_vdev_guid && rn->rn_vdev_guid != vdev_guid)) {
+ (void) close(fd);
+ nvlist_free(config);
+ return;
+ }
+
(void) close(fd);
rn->rn_config = config;
if (path != NULL) {
slice = zfs_alloc(hdl, sizeof (rdsk_node_t));
slice->rn_name = zfs_strdup(hdl, path);
+ slice->rn_vdev_guid = vdev_guid;
slice->rn_avl = rn->rn_avl;
slice->rn_hdl = hdl;
- slice->rn_order = 1;
+ slice->rn_order = IMPORT_ORDER_PREFERRED_1;
slice->rn_labelpaths = B_FALSE;
mutex_enter(rn->rn_lock);
if (avl_find(rn->rn_avl, slice, &where)) {
return;
}
+ slice->rn_vdev_guid = vdev_guid;
slice->rn_avl = rn->rn_avl;
slice->rn_hdl = hdl;
- slice->rn_order = 2;
+ slice->rn_order = IMPORT_ORDER_PREFERRED_2;
slice->rn_labelpaths = B_FALSE;
mutex_enter(rn->rn_lock);
if (avl_find(rn->rn_avl, slice, &where)) {
free(slice);
continue;
}
+ slice->rn_vdev_guid = 0;
slice->rn_lock = lock;
slice->rn_avl = cache;
slice->rn_hdl = hdl;
- slice->rn_order = i+1;
+ slice->rn_order = i + IMPORT_ORDER_SCAN_OFFSET;
slice->rn_labelpaths = B_FALSE;
mutex_enter(lock);
avl_add(cache, slice);
blkid_cache cache;
blkid_dev_iterate iter;
blkid_dev dev;
+ avl_index_t where;
int error;
*slice_cache = NULL;
while (blkid_dev_next(iter, &dev) == 0) {
slice = zfs_alloc(hdl, sizeof (rdsk_node_t));
slice->rn_name = zfs_strdup(hdl, blkid_dev_devname(dev));
+ slice->rn_vdev_guid = 0;
slice->rn_lock = lock;
slice->rn_avl = *slice_cache;
slice->rn_hdl = hdl;
- slice->rn_order = 100;
slice->rn_labelpaths = B_TRUE;
+
+ error = zfs_path_order(slice->rn_name, &slice->rn_order);
+ if (error == 0)
+ slice->rn_order += IMPORT_ORDER_SCAN_OFFSET;
+ else
+ slice->rn_order = IMPORT_ORDER_DEFAULT;
+
mutex_enter(lock);
- avl_add(*slice_cache, slice);
+ if (avl_find(*slice_cache, slice, &where)) {
+ free(slice->rn_name);
+ free(slice);
+ } else {
+ avl_insert(*slice_cache, slice, where);
+ }
mutex_exit(lock);
}
if (slice->rn_config != NULL) {
nvlist_t *config = slice->rn_config;
boolean_t matched = B_TRUE;
+ boolean_t aux = B_FALSE;
+ int fd;
+
+ /*
+ * Check if it's a spare or l2cache device. If it is,
+ * we need to skip the name and guid check since they
+ * don't exist on aux device label.
+ */
+ if (iarg->poolname != NULL || iarg->guid != 0) {
+ uint64_t state;
+ aux = nvlist_lookup_uint64(config,
+ ZPOOL_CONFIG_POOL_STATE, &state) == 0 &&
+ (state == POOL_STATE_SPARE ||
+ state == POOL_STATE_L2CACHE);
+ }
- if (iarg->poolname != NULL) {
+ if (iarg->poolname != NULL && !aux) {
char *pname;
matched = nvlist_lookup_string(config,
ZPOOL_CONFIG_POOL_NAME, &pname) == 0 &&
strcmp(iarg->poolname, pname) == 0;
- } else if (iarg->guid != 0) {
+ } else if (iarg->guid != 0 && !aux) {
uint64_t this_guid;
matched = nvlist_lookup_uint64(config,
if (!matched) {
nvlist_free(config);
} else {
- add_config(hdl, &pools,
- slice->rn_name, slice->rn_order,
- slice->rn_num_labels, config);
+ /*
+ * Verify all remaining entries can be opened
+ * exclusively. This will prune all underlying
+ * multipath devices which otherwise could
+ * result in the vdev appearing as UNAVAIL.
+ *
+ * Under zdb, this step isn't required and
+ * would prevent a zdb -e of active pools with
+ * no cachefile.
+ */
+ fd = open(slice->rn_name, O_RDONLY | O_EXCL);
+ if (fd >= 0 || iarg->can_be_active) {
+ if (fd >= 0)
+ close(fd);
+ add_config(hdl, &pools,
+ slice->rn_name, slice->rn_order,
+ slice->rn_num_labels, config);
+ } else {
+ nvlist_free(config);
+ }
}
}
free(slice->rn_name);
return (zpool_find_import_impl(hdl, import));
}
+static boolean_t
+pool_match(nvlist_t *cfg, char *tgt)
+{
+ uint64_t v, guid = strtoull(tgt, NULL, 0);
+ char *s;
+
+ if (guid != 0) {
+ if (nvlist_lookup_uint64(cfg, ZPOOL_CONFIG_POOL_GUID, &v) == 0)
+ return (v == guid);
+ } else {
+ if (nvlist_lookup_string(cfg, ZPOOL_CONFIG_POOL_NAME, &s) == 0)
+ return (strcmp(s, tgt) == 0);
+ }
+ return (B_FALSE);
+}
+
+int
+zpool_tryimport(libzfs_handle_t *hdl, char *target, nvlist_t **configp,
+ importargs_t *args)
+{
+ nvlist_t *pools;
+ nvlist_t *match = NULL;
+ nvlist_t *config = NULL;
+ char *name = NULL, *sepp = NULL;
+ char sep = '\0';
+ int count = 0;
+ char *targetdup = strdup(target);
+
+ *configp = NULL;
+
+ if ((sepp = strpbrk(targetdup, "/@")) != NULL) {
+ sep = *sepp;
+ *sepp = '\0';
+ }
+
+ pools = zpool_search_import(hdl, args);
+
+ if (pools != NULL) {
+ nvpair_t *elem = NULL;
+ while ((elem = nvlist_next_nvpair(pools, elem)) != NULL) {
+ VERIFY0(nvpair_value_nvlist(elem, &config));
+ if (pool_match(config, targetdup)) {
+ count++;
+ if (match != NULL) {
+ /* multiple matches found */
+ continue;
+ } else {
+ match = config;
+ name = nvpair_name(elem);
+ }
+ }
+ }
+ }
+
+ if (count == 0) {
+ (void) zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
+ "no pools found"));
+ free(targetdup);
+ return (ENOENT);
+ }
+
+ if (count > 1) {
+ (void) zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
+ "%d pools found, use pool GUID\n"), count);
+ free(targetdup);
+ return (EINVAL);
+ }
+
+ *configp = match;
+ free(targetdup);
+
+ return (0);
+}
+
boolean_t
find_guid(nvlist_t *nv, uint64_t guid)
{
/*
* Determines if the pool is in use. If so, it returns true and the state of
- * the pool as well as the name of the pool. Both strings are allocated and
+ * the pool as well as the name of the pool. Name string is allocated and
* must be freed by the caller.
*/
int