static int
zpool_find_load_time(zpool_handle_t *zhp, void *arg)
{
struct load_time_arg *lta = arg;
uint64_t pool_guid;
uint64_t *tod;
nvlist_t *config;
uint_t nelem;
if (lta->lt_found)
return (0);
pool_guid = zpool_get_prop_int(zhp, ZPOOL_PROP_GUID, NULL);
if (pool_guid != lta->lt_guid)
return (0);
if ((config = zpool_get_config(zhp, NULL)) == NULL) {
zpool_close(zhp);
return (-1);
}
if (nvlist_lookup_uint64_array(config, ZPOOL_CONFIG_LOADED_TIME,
&tod, &nelem) == 0 && nelem == 2) {
lta->lt_found = B_TRUE;
lta->lt_time->ertv_sec = tod[0];
lta->lt_time->ertv_nsec = tod[1];
}
return (0);
}
/*ARGSUSED*/
static int
zfs_mark_pool(zpool_handle_t *zhp, void *unused)
{
zfs_case_t *zcp;
uint64_t pool_guid;
nvlist_t *config, *vd;
int ret;
pool_guid = zpool_get_prop_int(zhp, ZPOOL_PROP_GUID, NULL);
/*
* Mark any cases associated with just this pool.
*/
for (zcp = uu_list_first(zfs_cases); zcp != NULL;
zcp = uu_list_next(zfs_cases, zcp)) {
if (zcp->zc_data.zc_pool_guid == pool_guid &&
zcp->zc_data.zc_vdev_guid == 0)
zcp->zc_present = B_TRUE;
}
if ((config = zpool_get_config(zhp, NULL)) == NULL) {
zpool_close(zhp);
return (-1);
}
ret = nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, &vd);
assert(ret == 0);
zfs_mark_vdev(pool_guid, vd);
zpool_close(zhp);
return (0);
}
static int
zfsdle_vdev_online(zpool_handle_t *zhp, void *data)
{
char *devname = data;
boolean_t avail_spare, l2cache;
vdev_state_t newstate;
nvlist_t *tgt;
zed_log_msg(LOG_INFO, "zfsdle_vdev_online: searching for '%s' in '%s'",
devname, zpool_get_name(zhp));
if ((tgt = zpool_find_vdev_by_physpath(zhp, devname,
&avail_spare, &l2cache, NULL)) != NULL) {
char *path, fullpath[MAXPATHLEN];
uint64_t wholedisk = 0ULL;
verify(nvlist_lookup_string(tgt, ZPOOL_CONFIG_PATH,
&path) == 0);
verify(nvlist_lookup_uint64(tgt, ZPOOL_CONFIG_WHOLE_DISK,
&wholedisk) == 0);
(void) strlcpy(fullpath, path, sizeof (fullpath));
if (wholedisk) {
char *spath = zfs_strip_partition(fullpath);
if (!spath) {
zed_log_msg(LOG_INFO, "%s: Can't alloc",
__func__);
return (0);
}
(void) strlcpy(fullpath, spath, sizeof (fullpath));
free(spath);
/*
* We need to reopen the pool associated with this
* device so that the kernel can update the size
* of the expanded device.
*/
(void) zpool_reopen(zhp);
}
if (zpool_get_prop_int(zhp, ZPOOL_PROP_AUTOEXPAND, NULL)) {
zed_log_msg(LOG_INFO, "zfsdle_vdev_online: setting "
"device '%s' to ONLINE state in pool '%s'",
fullpath, zpool_get_name(zhp));
if (zpool_get_state(zhp) != POOL_STATE_UNAVAIL)
(void) zpool_vdev_online(zhp, fullpath, 0,
&newstate);
}
zpool_close(zhp);
return (1);
}
zpool_close(zhp);
return (0);
}
/*
* Include snaps if they were requested or if this a zfs list where types
* were not specified and the "listsnapshots" property is set on this pool.
*/
static int
zfs_include_snapshots(zfs_handle_t *zhp, callback_data_t *cb)
{
zpool_handle_t *zph;
if ((cb->cb_flags & ZFS_ITER_PROP_LISTSNAPS) == 0)
return (cb->cb_types & ZFS_TYPE_SNAPSHOT);
zph = zfs_get_pool_handle(zhp);
return (zpool_get_prop_int(zph, ZPOOL_PROP_LISTSNAPS, NULL));
}
/*ARGSUSED*/
static int
zfs_mark_pool(zpool_handle_t *zhp, void *unused)
{
zfs_case_t *zcp;
uint64_t pool_guid;
uint64_t *tod;
er_timeval_t loaded = { 0 };
nvlist_t *config, *vd;
uint_t nelem = 0;
int ret;
pool_guid = zpool_get_prop_int(zhp, ZPOOL_PROP_GUID, NULL);
/*
* Mark any cases associated with just this pool.
*/
for (zcp = uu_list_first(zfs_cases); zcp != NULL;
zcp = uu_list_next(zfs_cases, zcp)) {
if (zcp->zc_data.zc_pool_guid == pool_guid &&
zcp->zc_data.zc_vdev_guid == 0)
zcp->zc_present = B_TRUE;
}
if ((config = zpool_get_config(zhp, NULL)) == NULL) {
zpool_close(zhp);
return (-1);
}
(void) nvlist_lookup_uint64_array(config, ZPOOL_CONFIG_LOADED_TIME,
&tod, &nelem);
if (nelem == 2) {
loaded.ertv_sec = tod[0];
loaded.ertv_nsec = tod[1];
for (zcp = uu_list_first(zfs_cases); zcp != NULL;
zcp = uu_list_next(zfs_cases, zcp)) {
if (zcp->zc_data.zc_pool_guid == pool_guid &&
zcp->zc_data.zc_vdev_guid == 0) {
zcp->zc_when = loaded;
}
}
}
ret = nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, &vd);
if (ret) {
zpool_close(zhp);
return (-1);
}
zfs_mark_vdev(pool_guid, vd, &loaded);
zpool_close(zhp);
return (0);
}
static int
find_pool(zpool_handle_t *zhp, void *data)
{
cbdata_t *cbp = data;
if (zpool_get_prop_int(zhp, ZPOOL_PROP_GUID, NULL) == cbp->cb_guid) {
cbp->cb_pool = zhp;
return (1);
}
zpool_close(zhp);
return (0);
}
static int
zfsdle_vdev_online(zpool_handle_t *zhp, void *data)
{
char *devname = data;
boolean_t avail_spare, l2cache;
vdev_state_t newstate;
nvlist_t *tgt;
syseventd_print(9, "zfsdle_vdev_online: searching for %s in pool %s\n",
devname, zpool_get_name(zhp));
if ((tgt = zpool_find_vdev_by_physpath(zhp, devname,
&avail_spare, &l2cache, NULL)) != NULL) {
char *path, fullpath[MAXPATHLEN];
uint64_t wholedisk = 0ULL;
verify(nvlist_lookup_string(tgt, ZPOOL_CONFIG_PATH,
&path) == 0);
verify(nvlist_lookup_uint64(tgt, ZPOOL_CONFIG_WHOLE_DISK,
&wholedisk) == 0);
(void) strlcpy(fullpath, path, sizeof (fullpath));
if (wholedisk) {
fullpath[strlen(fullpath) - 2] = '\0';
/*
* We need to reopen the pool associated with this
* device so that the kernel can update the size
* of the expanded device.
*/
(void) zpool_reopen(zhp);
}
if (zpool_get_prop_int(zhp, ZPOOL_PROP_AUTOEXPAND, NULL)) {
syseventd_print(9, "zfsdle_vdev_online: setting device"
" device %s to ONLINE state in pool %s.\n",
fullpath, zpool_get_name(zhp));
if (zpool_get_state(zhp) != POOL_STATE_UNAVAIL)
(void) zpool_vdev_online(zhp, fullpath, 0,
&newstate);
}
zpool_close(zhp);
return (1);
}
zpool_close(zhp);
return (0);
}
static boolean_t
encryption_feature_is_enabled(zpool_handle_t *zph)
{
nvlist_t *features;
uint64_t feat_refcount;
/* check that features can be enabled */
if (zpool_get_prop_int(zph, ZPOOL_PROP_VERSION, NULL)
< SPA_VERSION_FEATURES)
return (B_FALSE);
/* check for crypto feature */
features = zpool_get_features(zph);
if (!features || nvlist_lookup_uint64(features,
spa_feature_table[SPA_FEATURE_ENCRYPTION].fi_guid,
&feat_refcount) != 0)
return (B_FALSE);
return (B_TRUE);
}
static int
zfsdle_vdev_online(zpool_handle_t *zhp, void *data)
{
char *devname = data;
boolean_t avail_spare, l2cache;
nvlist_t *tgt;
int error;
zed_log_msg(LOG_INFO, "zfsdle_vdev_online: searching for '%s' in '%s'",
devname, zpool_get_name(zhp));
if ((tgt = zpool_find_vdev_by_physpath(zhp, devname,
&avail_spare, &l2cache, NULL)) != NULL) {
char *path, fullpath[MAXPATHLEN];
uint64_t wholedisk;
error = nvlist_lookup_string(tgt, ZPOOL_CONFIG_PATH, &path);
if (error) {
zpool_close(zhp);
return (0);
}
error = nvlist_lookup_uint64(tgt, ZPOOL_CONFIG_WHOLE_DISK,
&wholedisk);
if (error)
wholedisk = 0;
if (wholedisk) {
path = strrchr(path, '/');
if (path != NULL) {
path = zfs_strip_partition(path + 1);
if (path == NULL) {
zpool_close(zhp);
return (0);
}
} else {
zpool_close(zhp);
return (0);
}
(void) strlcpy(fullpath, path, sizeof (fullpath));
free(path);
/*
* We need to reopen the pool associated with this
* device so that the kernel can update the size of
* the expanded device. When expanding there is no
* need to restart the scrub from the beginning.
*/
boolean_t scrub_restart = B_FALSE;
(void) zpool_reopen_one(zhp, &scrub_restart);
} else {
(void) strlcpy(fullpath, path, sizeof (fullpath));
}
if (zpool_get_prop_int(zhp, ZPOOL_PROP_AUTOEXPAND, NULL)) {
vdev_state_t newstate;
if (zpool_get_state(zhp) != POOL_STATE_UNAVAIL) {
error = zpool_vdev_online(zhp, fullpath, 0,
&newstate);
zed_log_msg(LOG_INFO, "zfsdle_vdev_online: "
"setting device '%s' to ONLINE state "
"in pool '%s': %d", fullpath,
zpool_get_name(zhp), error);
}
}
zpool_close(zhp);
return (1);
}
zpool_close(zhp);
return (0);
}
/*
* The device associated with the given vdev (either by devid or physical path)
* has been added to the system. If 'isdisk' is set, then we only attempt a
* replacement if it's a whole disk. This also implies that we should label the
* disk first.
*
* First, we attempt to online the device (making sure to undo any spare
* operation when finished). If this succeeds, then we're done. If it fails,
* and the new state is VDEV_CANT_OPEN, it indicates that the device was opened,
* but that the label was not what we expected. If the 'autoreplace' property
* is enabled, then we relabel the disk (if specified), and attempt a 'zpool
* replace'. If the online is successful, but the new state is something else
* (REMOVED or FAULTED), it indicates that we're out of sync or in some sort of
* race, and we should avoid attempting to relabel the disk.
*
* Also can arrive here from a ESC_ZFS_VDEV_CHECK event
*/
static void
zfs_process_add(zpool_handle_t *zhp, nvlist_t *vdev, boolean_t labeled)
{
char *path;
vdev_state_t newstate;
nvlist_t *nvroot, *newvd;
pendingdev_t *device;
uint64_t wholedisk = 0ULL;
uint64_t offline = 0ULL;
uint64_t guid = 0ULL;
char *physpath = NULL, *new_devid = NULL, *enc_sysfs_path = NULL;
char rawpath[PATH_MAX], fullpath[PATH_MAX];
char devpath[PATH_MAX];
int ret;
int is_dm = 0;
int is_sd = 0;
uint_t c;
vdev_stat_t *vs;
if (nvlist_lookup_string(vdev, ZPOOL_CONFIG_PATH, &path) != 0)
return;
/* Skip healthy disks */
verify(nvlist_lookup_uint64_array(vdev, ZPOOL_CONFIG_VDEV_STATS,
(uint64_t **)&vs, &c) == 0);
if (vs->vs_state == VDEV_STATE_HEALTHY) {
zed_log_msg(LOG_INFO, "%s: %s is already healthy, skip it.",
__func__, path);
return;
}
(void) nvlist_lookup_string(vdev, ZPOOL_CONFIG_PHYS_PATH, &physpath);
(void) nvlist_lookup_string(vdev, ZPOOL_CONFIG_VDEV_ENC_SYSFS_PATH,
&enc_sysfs_path);
(void) nvlist_lookup_uint64(vdev, ZPOOL_CONFIG_WHOLE_DISK, &wholedisk);
(void) nvlist_lookup_uint64(vdev, ZPOOL_CONFIG_OFFLINE, &offline);
(void) nvlist_lookup_uint64(vdev, ZPOOL_CONFIG_GUID, &guid);
if (offline)
return; /* don't intervene if it was taken offline */
is_dm = zfs_dev_is_dm(path);
zed_log_msg(LOG_INFO, "zfs_process_add: pool '%s' vdev '%s', phys '%s'"
" wholedisk %d, dm %d (%llu)", zpool_get_name(zhp), path,
physpath ? physpath : "NULL", wholedisk, is_dm,
(long long unsigned int)guid);
/*
* The VDEV guid is preferred for identification (gets passed in path)
*/
if (guid != 0) {
(void) snprintf(fullpath, sizeof (fullpath), "%llu",
(long long unsigned int)guid);
} else {
/*
* otherwise use path sans partition suffix for whole disks
*/
(void) strlcpy(fullpath, path, sizeof (fullpath));
if (wholedisk) {
char *spath = zfs_strip_partition(fullpath);
if (!spath) {
zed_log_msg(LOG_INFO, "%s: Can't alloc",
__func__);
return;
}
(void) strlcpy(fullpath, spath, sizeof (fullpath));
free(spath);
}
}
/*
* Attempt to online the device.
*/
if (zpool_vdev_online(zhp, fullpath,
ZFS_ONLINE_CHECKREMOVE | ZFS_ONLINE_UNSPARE, &newstate) == 0 &&
(newstate == VDEV_STATE_HEALTHY ||
newstate == VDEV_STATE_DEGRADED)) {
zed_log_msg(LOG_INFO, " zpool_vdev_online: vdev %s is %s",
fullpath, (newstate == VDEV_STATE_HEALTHY) ?
"HEALTHY" : "DEGRADED");
return;
}
/*
* vdev_id alias rule for using scsi_debug devices (FMA automated
* testing)
*/
if (physpath != NULL && strcmp("scsidebug", physpath) == 0)
is_sd = 1;
/*
* If the pool doesn't have the autoreplace property set, then use
* vdev online to trigger a FMA fault by posting an ereport.
*/
if (!zpool_get_prop_int(zhp, ZPOOL_PROP_AUTOREPLACE, NULL) ||
!(wholedisk || is_dm) || (physpath == NULL)) {
(void) zpool_vdev_online(zhp, fullpath, ZFS_ONLINE_FORCEFAULT,
&newstate);
zed_log_msg(LOG_INFO, "Pool's autoreplace is not enabled or "
"not a whole disk for '%s'", fullpath);
return;
}
/*
* Convert physical path into its current device node. Rawpath
* needs to be /dev/disk/by-vdev for a scsi_debug device since
* /dev/disk/by-path will not be present.
*/
(void) snprintf(rawpath, sizeof (rawpath), "%s%s",
is_sd ? DEV_BYVDEV_PATH : DEV_BYPATH_PATH, physpath);
if (realpath(rawpath, devpath) == NULL && !is_dm) {
zed_log_msg(LOG_INFO, " realpath: %s failed (%s)",
rawpath, strerror(errno));
(void) zpool_vdev_online(zhp, fullpath, ZFS_ONLINE_FORCEFAULT,
&newstate);
zed_log_msg(LOG_INFO, " zpool_vdev_online: %s FORCEFAULT (%s)",
fullpath, libzfs_error_description(g_zfshdl));
return;
}
/* Only autoreplace bad disks */
if ((vs->vs_state != VDEV_STATE_DEGRADED) &&
(vs->vs_state != VDEV_STATE_FAULTED) &&
(vs->vs_state != VDEV_STATE_CANT_OPEN)) {
return;
}
nvlist_lookup_string(vdev, "new_devid", &new_devid);
if (is_dm) {
/* Don't label device mapper or multipath disks. */
} else if (!labeled) {
/*
* we're auto-replacing a raw disk, so label it first
*/
char *leafname;
/*
* If this is a request to label a whole disk, then attempt to
* write out the label. Before we can label the disk, we need
* to map the physical string that was matched on to the under
* lying device node.
*
* If any part of this process fails, then do a force online
* to trigger a ZFS fault for the device (and any hot spare
* replacement).
*/
leafname = strrchr(devpath, '/') + 1;
/*
* If this is a request to label a whole disk, then attempt to
* write out the label.
*/
if (zpool_label_disk(g_zfshdl, zhp, leafname) != 0) {
zed_log_msg(LOG_INFO, " zpool_label_disk: could not "
"label '%s' (%s)", leafname,
libzfs_error_description(g_zfshdl));
(void) zpool_vdev_online(zhp, fullpath,
ZFS_ONLINE_FORCEFAULT, &newstate);
return;
}
/*
* The disk labeling is asynchronous on Linux. Just record
* this label request and return as there will be another
* disk add event for the partition after the labeling is
* completed.
*/
device = malloc(sizeof (pendingdev_t));
(void) strlcpy(device->pd_physpath, physpath,
sizeof (device->pd_physpath));
list_insert_tail(&g_device_list, device);
zed_log_msg(LOG_INFO, " zpool_label_disk: async '%s' (%llu)",
leafname, (u_longlong_t)guid);
return; /* resumes at EC_DEV_ADD.ESC_DISK for partition */
} else /* labeled */ {
boolean_t found = B_FALSE;
/*
* match up with request above to label the disk
*/
for (device = list_head(&g_device_list); device != NULL;
device = list_next(&g_device_list, device)) {
if (strcmp(physpath, device->pd_physpath) == 0) {
list_remove(&g_device_list, device);
free(device);
found = B_TRUE;
break;
}
zed_log_msg(LOG_INFO, "zpool_label_disk: %s != %s",
physpath, device->pd_physpath);
}
if (!found) {
/* unexpected partition slice encountered */
zed_log_msg(LOG_INFO, "labeled disk %s unexpected here",
fullpath);
(void) zpool_vdev_online(zhp, fullpath,
ZFS_ONLINE_FORCEFAULT, &newstate);
return;
}
zed_log_msg(LOG_INFO, " zpool_label_disk: resume '%s' (%llu)",
physpath, (u_longlong_t)guid);
(void) snprintf(devpath, sizeof (devpath), "%s%s",
DEV_BYID_PATH, new_devid);
}
/*
* Construct the root vdev to pass to zpool_vdev_attach(). While adding
* the entire vdev structure is harmless, we construct a reduced set of
* path/physpath/wholedisk to keep it simple.
*/
if (nvlist_alloc(&nvroot, NV_UNIQUE_NAME, 0) != 0) {
zed_log_msg(LOG_WARNING, "zfs_mod: nvlist_alloc out of memory");
return;
}
if (nvlist_alloc(&newvd, NV_UNIQUE_NAME, 0) != 0) {
zed_log_msg(LOG_WARNING, "zfs_mod: nvlist_alloc out of memory");
nvlist_free(nvroot);
return;
}
if (nvlist_add_string(newvd, ZPOOL_CONFIG_TYPE, VDEV_TYPE_DISK) != 0 ||
nvlist_add_string(newvd, ZPOOL_CONFIG_PATH, path) != 0 ||
nvlist_add_string(newvd, ZPOOL_CONFIG_DEVID, new_devid) != 0 ||
(physpath != NULL && nvlist_add_string(newvd,
ZPOOL_CONFIG_PHYS_PATH, physpath) != 0) ||
(enc_sysfs_path != NULL && nvlist_add_string(newvd,
ZPOOL_CONFIG_VDEV_ENC_SYSFS_PATH, enc_sysfs_path) != 0) ||
nvlist_add_uint64(newvd, ZPOOL_CONFIG_WHOLE_DISK, wholedisk) != 0 ||
nvlist_add_string(nvroot, ZPOOL_CONFIG_TYPE, VDEV_TYPE_ROOT) != 0 ||
nvlist_add_nvlist_array(nvroot, ZPOOL_CONFIG_CHILDREN, &newvd,
1) != 0) {
zed_log_msg(LOG_WARNING, "zfs_mod: unable to add nvlist pairs");
nvlist_free(newvd);
nvlist_free(nvroot);
return;
}
nvlist_free(newvd);
/*
* Wait for udev to verify the links exist, then auto-replace
* the leaf disk at same physical location.
*/
if (zpool_label_disk_wait(path, 3000) != 0) {
zed_log_msg(LOG_WARNING, "zfs_mod: expected replacement "
"disk %s is missing", path);
nvlist_free(nvroot);
return;
}
ret = zpool_vdev_attach(zhp, fullpath, path, nvroot, B_TRUE);
zed_log_msg(LOG_INFO, " zpool_vdev_replace: %s with %s (%s)",
fullpath, path, (ret == 0) ? "no errors" :
libzfs_error_description(g_zfshdl));
nvlist_free(nvroot);
}
/*
* 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
* must be freed by the caller.
*/
int
zpool_in_use(libzfs_handle_t *hdl, int fd, pool_state_t *state, char **namestr,
boolean_t *inuse)
{
nvlist_t *config;
char *name;
boolean_t ret;
uint64_t guid, vdev_guid;
zpool_handle_t *zhp;
nvlist_t *pool_config;
uint64_t stateval, isspare;
aux_cbdata_t cb = { 0 };
boolean_t isactive;
*inuse = B_FALSE;
if (zpool_read_label(fd, &config, NULL) != 0 && errno == ENOMEM) {
(void) no_memory(hdl);
return (-1);
}
if (config == NULL)
return (0);
verify(nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_STATE,
&stateval) == 0);
verify(nvlist_lookup_uint64(config, ZPOOL_CONFIG_GUID,
&vdev_guid) == 0);
if (stateval != POOL_STATE_SPARE && stateval != POOL_STATE_L2CACHE) {
verify(nvlist_lookup_string(config, ZPOOL_CONFIG_POOL_NAME,
&name) == 0);
verify(nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_GUID,
&guid) == 0);
}
switch (stateval) {
case POOL_STATE_EXPORTED:
/*
* A pool with an exported state may in fact be imported
* read-only, so check the in-core state to see if it's
* active and imported read-only. If it is, set
* its state to active.
*/
if (pool_active(hdl, name, guid, &isactive) == 0 && isactive &&
(zhp = zpool_open_canfail(hdl, name)) != NULL) {
if (zpool_get_prop_int(zhp, ZPOOL_PROP_READONLY, NULL))
stateval = POOL_STATE_ACTIVE;
/*
* All we needed the zpool handle for is the
* readonly prop check.
*/
zpool_close(zhp);
}
ret = B_TRUE;
break;
case POOL_STATE_ACTIVE:
/*
* For an active pool, we have to determine if it's really part
* of a currently active pool (in which case the pool will exist
* and the guid will be the same), or whether it's part of an
* active pool that was disconnected without being explicitly
* exported.
*/
if (pool_active(hdl, name, guid, &isactive) != 0) {
nvlist_free(config);
return (-1);
}
if (isactive) {
/*
* Because the device may have been removed while
* offlined, we only report it as active if the vdev is
* still present in the config. Otherwise, pretend like
* it's not in use.
*/
if ((zhp = zpool_open_canfail(hdl, name)) != NULL &&
(pool_config = zpool_get_config(zhp, NULL))
!= NULL) {
nvlist_t *nvroot;
verify(nvlist_lookup_nvlist(pool_config,
ZPOOL_CONFIG_VDEV_TREE, &nvroot) == 0);
ret = find_guid(nvroot, vdev_guid);
} else {
ret = B_FALSE;
}
/*
* If this is an active spare within another pool, we
* treat it like an unused hot spare. This allows the
* user to create a pool with a hot spare that currently
* in use within another pool. Since we return B_TRUE,
* libdiskmgt will continue to prevent generic consumers
* from using the device.
*/
if (ret && nvlist_lookup_uint64(config,
ZPOOL_CONFIG_IS_SPARE, &isspare) == 0 && isspare)
stateval = POOL_STATE_SPARE;
if (zhp != NULL)
zpool_close(zhp);
} else {
stateval = POOL_STATE_POTENTIALLY_ACTIVE;
ret = B_TRUE;
}
break;
case POOL_STATE_SPARE:
/*
* For a hot spare, it can be either definitively in use, or
* potentially active. To determine if it's in use, we iterate
* over all pools in the system and search for one with a spare
* with a matching guid.
*
* Due to the shared nature of spares, we don't actually report
* the potentially active case as in use. This means the user
* can freely create pools on the hot spares of exported pools,
* but to do otherwise makes the resulting code complicated, and
* we end up having to deal with this case anyway.
*/
cb.cb_zhp = NULL;
cb.cb_guid = vdev_guid;
cb.cb_type = ZPOOL_CONFIG_SPARES;
if (zpool_iter(hdl, find_aux, &cb) == 1) {
name = (char *)zpool_get_name(cb.cb_zhp);
ret = B_TRUE;
} else {
ret = B_FALSE;
}
break;
case POOL_STATE_L2CACHE:
/*
* Check if any pool is currently using this l2cache device.
*/
cb.cb_zhp = NULL;
cb.cb_guid = vdev_guid;
cb.cb_type = ZPOOL_CONFIG_L2CACHE;
if (zpool_iter(hdl, find_aux, &cb) == 1) {
name = (char *)zpool_get_name(cb.cb_zhp);
ret = B_TRUE;
} else {
ret = B_FALSE;
}
break;
default:
ret = B_FALSE;
}
if (ret) {
if ((*namestr = zfs_strdup(hdl, name)) == NULL) {
if (cb.cb_zhp)
zpool_close(cb.cb_zhp);
nvlist_free(config);
return (-1);
}
*state = (pool_state_t)stateval;
}
if (cb.cb_zhp)
zpool_close(cb.cb_zhp);
nvlist_free(config);
*inuse = ret;
return (0);
}
/*
* Mount the given filesystem.
*/
int
zfs_mount(zfs_handle_t *zhp, const char *options, int flags)
{
struct stat buf;
char mountpoint[ZFS_MAXPROPLEN];
char mntopts[MNT_LINE_MAX];
libzfs_handle_t *hdl = zhp->zfs_hdl;
int remount = 0, rc;
if (options == NULL) {
(void) strlcpy(mntopts, MNTOPT_DEFAULTS, sizeof (mntopts));
} else {
(void) strlcpy(mntopts, options, sizeof (mntopts));
}
if (strstr(mntopts, MNTOPT_REMOUNT) != NULL)
remount = 1;
/*
* If the pool is imported read-only then all mounts must be read-only
*/
if (zpool_get_prop_int(zhp->zpool_hdl, ZPOOL_PROP_READONLY, NULL))
(void) strlcat(mntopts, "," MNTOPT_RO, sizeof (mntopts));
/*
* Load encryption key if required and not already present.
* Don't need to check ZFS_PROP_ENCRYPTION because encrypted
* datasets have keystatus of ZFS_CRYPT_KEY_NONE.
*/
fprintf(stderr, "zfs_mount: mount, keystatus is %d\r\n",
zfs_prop_get_int(zhp, ZFS_PROP_KEYSTATUS));
if (zfs_prop_get_int(zhp, ZFS_PROP_KEYSTATUS) ==
ZFS_CRYPT_KEY_UNAVAILABLE) {
fprintf(stderr, "loading KEY\r\n");
(void )zfs_key_load(zhp, B_FALSE, B_FALSE, B_FALSE);
}
/*
* Append default mount options which apply to the mount point.
* This is done because under Linux (unlike Solaris) multiple mount
* points may reference a single super block. This means that just
* given a super block there is no back reference to update the per
* mount point options.
*/
rc = zfs_add_options(zhp, mntopts, sizeof (mntopts));
if (rc) {
zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
"default options unavailable"));
return (zfs_error_fmt(hdl, EZFS_MOUNTFAILED,
dgettext(TEXT_DOMAIN, "cannot mount '%s'"),
mountpoint));
}
/*
* Append zfsutil option so the mount helper allow the mount
*/
strlcat(mntopts, "," MNTOPT_ZFSUTIL, sizeof (mntopts));
if (!zfs_is_mountable(zhp, mountpoint, sizeof (mountpoint), NULL))
return (0);
/* Create the directory if it doesn't already exist */
if (lstat(mountpoint, &buf) != 0) {
if (mkdirp(mountpoint, 0755) != 0) {
zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
"failed to create mountpoint"));
return (zfs_error_fmt(hdl, EZFS_MOUNTFAILED,
dgettext(TEXT_DOMAIN, "cannot mount '%s'"),
mountpoint));
}
}
/*
* Determine if the mountpoint is empty. If so, refuse to perform the
* mount. We don't perform this check if 'remount' is
* specified or if overlay option(-O) is given
*/
if ((flags & MS_OVERLAY) == 0 && !remount &&
!dir_is_empty(mountpoint)) {
zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
"directory is not empty"));
return (zfs_error_fmt(hdl, EZFS_MOUNTFAILED,
dgettext(TEXT_DOMAIN, "cannot mount '%s'"), mountpoint));
}
/* perform the mount */
rc = do_mount(zfs_get_name(zhp), mountpoint, mntopts);
if (rc) {
/*
* Generic errors are nasty, but there are just way too many
* from mount(), and they're well-understood. We pick a few
* common ones to improve upon.
*/
if (rc == EBUSY) {
zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
"mountpoint or dataset is busy"));
} else if (rc == EPERM) {
zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
"Insufficient privileges"));
} else if (rc == ENOTSUP) {
char buf[256];
int spa_version;
VERIFY(zfs_spa_version(zhp, &spa_version) == 0);
(void) snprintf(buf, sizeof (buf),
dgettext(TEXT_DOMAIN, "Can't mount a version %lld "
"file system on a version %d pool. Pool must be"
" upgraded to mount this file system."),
(u_longlong_t)zfs_prop_get_int(zhp,
ZFS_PROP_VERSION), spa_version);
zfs_error_aux(hdl, dgettext(TEXT_DOMAIN, buf));
} else {
zfs_error_aux(hdl, strerror(rc));
}
return (zfs_error_fmt(hdl, EZFS_MOUNTFAILED,
dgettext(TEXT_DOMAIN, "cannot mount '%s'"),
zhp->zfs_name));
}
/* remove the mounted entry before re-adding on remount */
if (remount)
libzfs_mnttab_remove(hdl, zhp->zfs_name);
/* add the mounted entry into our cache */
libzfs_mnttab_add(hdl, zfs_get_name(zhp), mountpoint, mntopts);
return (0);
}
/*
* Mount the given filesystem.
*/
int
zfs_mount(zfs_handle_t *zhp, const char *options, int flags)
{
struct stat buf;
char mountpoint[ZFS_MAXPROPLEN];
char mntopts[MNT_LINE_MAX];
libzfs_handle_t *hdl = zhp->zfs_hdl;
if (options == NULL)
mntopts[0] = '\0';
else
(void) strlcpy(mntopts, options, sizeof (mntopts));
/*
* If the pool is imported read-only then all mounts must be read-only
*/
if (zpool_get_prop_int(zhp->zpool_hdl, ZPOOL_PROP_READONLY, NULL))
flags |= MS_RDONLY;
if (!zfs_is_mountable(zhp, mountpoint, sizeof (mountpoint), NULL))
return (0);
/* Create the directory if it doesn't already exist */
if (lstat(mountpoint, &buf) != 0) {
if (mkdirp(mountpoint, 0755) != 0) {
zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
"failed to create mountpoint"));
return (zfs_error_fmt(hdl, EZFS_MOUNTFAILED,
dgettext(TEXT_DOMAIN, "cannot mount '%s'"),
mountpoint));
}
}
/*
* Determine if the mountpoint is empty. If so, refuse to perform the
* mount. We don't perform this check if MS_OVERLAY is specified, which
* would defeat the point. We also avoid this check if 'remount' is
* specified.
*/
if ((flags & MS_OVERLAY) == 0 &&
strstr(mntopts, MNTOPT_REMOUNT) == NULL &&
!dir_is_empty(mountpoint)) {
zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
"directory is not empty"));
return (zfs_error_fmt(hdl, EZFS_MOUNTFAILED,
dgettext(TEXT_DOMAIN, "cannot mount '%s'"), mountpoint));
}
/* perform the mount */
if (mount(zfs_get_name(zhp), mountpoint, MS_OPTIONSTR | flags,
MNTTYPE_ZFS, NULL, 0, mntopts, sizeof (mntopts)) != 0) {
/*
* Generic errors are nasty, but there are just way too many
* from mount(), and they're well-understood. We pick a few
* common ones to improve upon.
*/
if (errno == EBUSY) {
zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
"mountpoint or dataset is busy"));
} else if (errno == EPERM) {
zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
"Insufficient privileges"));
} else if (errno == ENOTSUP) {
char buf[256];
int spa_version;
VERIFY(zfs_spa_version(zhp, &spa_version) == 0);
(void) snprintf(buf, sizeof (buf),
dgettext(TEXT_DOMAIN, "Can't mount a version %lld "
"file system on a version %d pool. Pool must be"
" upgraded to mount this file system."),
(u_longlong_t)zfs_prop_get_int(zhp,
ZFS_PROP_VERSION), spa_version);
zfs_error_aux(hdl, dgettext(TEXT_DOMAIN, buf));
} else {
zfs_error_aux(hdl, strerror(errno));
}
return (zfs_error_fmt(hdl, EZFS_MOUNTFAILED,
dgettext(TEXT_DOMAIN, "cannot mount '%s'"),
zhp->zfs_name));
}
/* add the mounted entry into our cache */
libzfs_mnttab_add(hdl, zfs_get_name(zhp), mountpoint,
mntopts);
return (0);
}
/*
* The device associated with the given vdev (either by devid or physical path)
* has been added to the system. If 'isdisk' is set, then we only attempt a
* replacement if it's a whole disk. This also implies that we should label the
* disk first.
*
* First, we attempt to online the device (making sure to undo any spare
* operation when finished). If this succeeds, then we're done. If it fails,
* and the new state is VDEV_CANT_OPEN, it indicates that the device was opened,
* but that the label was not what we expected. If the 'autoreplace' property
* is not set, then we relabel the disk (if specified), and attempt a 'zpool
* replace'. If the online is successful, but the new state is something else
* (REMOVED or FAULTED), it indicates that we're out of sync or in some sort of
* race, and we should avoid attempting to relabel the disk.
*/
static void
zfs_process_add(zpool_handle_t *zhp, nvlist_t *vdev, boolean_t isdisk)
{
char *path;
vdev_state_t newstate;
nvlist_t *nvroot, *newvd;
uint64_t wholedisk = 0ULL;
uint64_t offline = 0ULL;
char *physpath = NULL;
char rawpath[PATH_MAX], fullpath[PATH_MAX];
size_t len;
if (nvlist_lookup_string(vdev, ZPOOL_CONFIG_PATH, &path) != 0)
return;
(void) nvlist_lookup_string(vdev, ZPOOL_CONFIG_PHYS_PATH, &physpath);
(void) nvlist_lookup_uint64(vdev, ZPOOL_CONFIG_WHOLE_DISK, &wholedisk);
(void) nvlist_lookup_uint64(vdev, ZPOOL_CONFIG_OFFLINE, &offline);
/*
* We should have a way to online a device by guid. With the current
* interface, we are forced to chop off the 's0' for whole disks.
*/
(void) strlcpy(fullpath, path, sizeof (fullpath));
if (wholedisk)
fullpath[strlen(fullpath) - 2] = '\0';
/*
* Attempt to online the device. It would be nice to online this by
* GUID, but the current interface only supports lookup by path.
*/
if (offline ||
(zpool_vdev_online(zhp, fullpath,
ZFS_ONLINE_CHECKREMOVE | ZFS_ONLINE_UNSPARE, &newstate) == 0 &&
(newstate == VDEV_STATE_HEALTHY ||
newstate == VDEV_STATE_DEGRADED)))
return;
/*
* If the pool doesn't have the autoreplace property set, then attempt a
* true online (without the unspare flag), which will trigger a FMA
* fault.
*/
if (!zpool_get_prop_int(zhp, ZPOOL_PROP_AUTOREPLACE, NULL) ||
(isdisk && !wholedisk)) {
(void) zpool_vdev_online(zhp, fullpath, ZFS_ONLINE_FORCEFAULT,
&newstate);
return;
}
if (isdisk) {
/*
* If this is a request to label a whole disk, then attempt to
* write out the label. Before we can label the disk, we need
* access to a raw node. Ideally, we'd like to walk the devinfo
* tree and find a raw node from the corresponding parent node.
* This is overly complicated, and since we know how we labeled
* this device in the first place, we know it's save to switch
* from /dev/dsk to /dev/rdsk and append the backup slice.
*
* If any part of this process fails, then do a force online to
* trigger a ZFS fault for the device (and any hot spare
* replacement).
*/
if (strncmp(path, "/dev/dsk/", 9) != 0) {
(void) zpool_vdev_online(zhp, fullpath,
ZFS_ONLINE_FORCEFAULT, &newstate);
return;
}
(void) strlcpy(rawpath, path + 9, sizeof (rawpath));
len = strlen(rawpath);
rawpath[len - 2] = '\0';
if (zpool_label_disk(g_zfshdl, zhp, rawpath) != 0) {
(void) zpool_vdev_online(zhp, fullpath,
ZFS_ONLINE_FORCEFAULT, &newstate);
return;
}
}
/*
* Cosntruct the root vdev to pass to zpool_vdev_attach(). While adding
* the entire vdev structure is harmless, we construct a reduced set of
* path/physpath/wholedisk to keep it simple.
*/
if (nvlist_alloc(&nvroot, NV_UNIQUE_NAME, 0) != 0)
return;
if (nvlist_alloc(&newvd, NV_UNIQUE_NAME, 0) != 0) {
nvlist_free(nvroot);
return;
}
if (nvlist_add_string(newvd, ZPOOL_CONFIG_TYPE, VDEV_TYPE_DISK) != 0 ||
nvlist_add_string(newvd, ZPOOL_CONFIG_PATH, path) != 0 ||
(physpath != NULL && nvlist_add_string(newvd,
ZPOOL_CONFIG_PHYS_PATH, physpath) != 0) ||
nvlist_add_uint64(newvd, ZPOOL_CONFIG_WHOLE_DISK, wholedisk) != 0 ||
nvlist_add_string(nvroot, ZPOOL_CONFIG_TYPE, VDEV_TYPE_ROOT) != 0 ||
nvlist_add_nvlist_array(nvroot, ZPOOL_CONFIG_CHILDREN, &newvd,
1) != 0) {
nvlist_free(newvd);
nvlist_free(nvroot);
return;
}
nvlist_free(newvd);
(void) zpool_vdev_attach(zhp, fullpath, path, nvroot, B_TRUE);
nvlist_free(nvroot);
}
/*
* The device associated with the given vdev (either by devid or physical path)
* has been added to the system. If 'isdisk' is set, then we only attempt a
* replacement if it's a whole disk. This also implies that we should label the
* disk first.
*
* First, we attempt to online the device (making sure to undo any spare
* operation when finished). If this succeeds, then we're done. If it fails,
* and the new state is VDEV_CANT_OPEN, it indicates that the device was opened,
* but that the label was not what we expected. If the 'autoreplace' property
* is not set, then we relabel the disk (if specified), and attempt a 'zpool
* replace'. If the online is successful, but the new state is something else
* (REMOVED or FAULTED), it indicates that we're out of sync or in some sort of
* race, and we should avoid attempting to relabel the disk.
*
* Also can arrive here from a ESC_ZFS_VDEV_CHECK event
*/
static void
zfs_process_add(zpool_handle_t *zhp, nvlist_t *vdev, boolean_t labeled)
{
char *path;
vdev_state_t newstate;
nvlist_t *nvroot, *newvd;
pendingdev_t *device;
uint64_t wholedisk = 0ULL;
uint64_t offline = 0ULL;
uint64_t guid = 0ULL;
char *physpath = NULL, *new_devid = NULL;
char rawpath[PATH_MAX], fullpath[PATH_MAX];
char devpath[PATH_MAX];
int ret;
if (nvlist_lookup_string(vdev, ZPOOL_CONFIG_PATH, &path) != 0)
return;
(void) nvlist_lookup_string(vdev, ZPOOL_CONFIG_PHYS_PATH, &physpath);
(void) nvlist_lookup_uint64(vdev, ZPOOL_CONFIG_WHOLE_DISK, &wholedisk);
(void) nvlist_lookup_uint64(vdev, ZPOOL_CONFIG_OFFLINE, &offline);
(void) nvlist_lookup_uint64(vdev, ZPOOL_CONFIG_GUID, &guid);
if (offline)
return; /* don't intervene if it was taken offline */
zed_log_msg(LOG_INFO, "zfs_process_add: pool '%s' vdev '%s' (%llu)",
zpool_get_name(zhp), path, (long long unsigned int)guid);
/*
* The VDEV guid is preferred for identification (gets passed in path)
*/
if (guid != 0) {
(void) snprintf(fullpath, sizeof (fullpath), "%llu",
(long long unsigned int)guid);
} else {
/*
* otherwise use path sans partition suffix for whole disks
*/
(void) strlcpy(fullpath, path, sizeof (fullpath));
if (wholedisk) {
char *spath = zfs_strip_partition(g_zfshdl, fullpath);
(void) strlcpy(fullpath, spath, sizeof (fullpath));
free(spath);
}
}
/*
* Attempt to online the device.
*/
if (zpool_vdev_online(zhp, fullpath,
ZFS_ONLINE_CHECKREMOVE | ZFS_ONLINE_UNSPARE, &newstate) == 0 &&
(newstate == VDEV_STATE_HEALTHY ||
newstate == VDEV_STATE_DEGRADED)) {
zed_log_msg(LOG_INFO, " zpool_vdev_online: vdev %s is %s",
fullpath, (newstate == VDEV_STATE_HEALTHY) ?
"HEALTHY" : "DEGRADED");
return;
}
/*
* If the pool doesn't have the autoreplace property set, then attempt
* a true online (without the unspare flag), which will trigger a FMA
* fault.
*/
if (!zpool_get_prop_int(zhp, ZPOOL_PROP_AUTOREPLACE, NULL) ||
!wholedisk || physpath == NULL) {
(void) zpool_vdev_online(zhp, fullpath, ZFS_ONLINE_FORCEFAULT,
&newstate);
zed_log_msg(LOG_INFO, " zpool_vdev_online: %s FORCEFAULT (%s)",
fullpath, libzfs_error_description(g_zfshdl));
return;
}
/*
* convert physical path into its current device node
*/
(void) snprintf(rawpath, sizeof (rawpath), "%s%s", DEV_BYPATH_PATH,
physpath);
if (realpath(rawpath, devpath) == NULL) {
zed_log_msg(LOG_INFO, " realpath: %s failed (%s)",
rawpath, strerror(errno));
(void) zpool_vdev_online(zhp, fullpath, ZFS_ONLINE_FORCEFAULT,
&newstate);
zed_log_msg(LOG_INFO, " zpool_vdev_online: %s FORCEFAULT (%s)",
fullpath, libzfs_error_description(g_zfshdl));
return;
}
/*
* we're auto-replacing a raw disk, so label it first
*/
if (!labeled) {
char *leafname;
/*
* If this is a request to label a whole disk, then attempt to
* write out the label. Before we can label the disk, we need
* to map the physical string that was matched on to the under
* lying device node.
*
* If any part of this process fails, then do a force online
* to trigger a ZFS fault for the device (and any hot spare
* replacement).
*/
leafname = strrchr(devpath, '/') + 1;
/*
* If this is a request to label a whole disk, then attempt to
* write out the label.
*/
if (zpool_label_disk(g_zfshdl, zhp, leafname) != 0) {
zed_log_msg(LOG_INFO, " zpool_label_disk: could not "
"label '%s' (%s)", leafname,
libzfs_error_description(g_zfshdl));
(void) zpool_vdev_online(zhp, fullpath,
ZFS_ONLINE_FORCEFAULT, &newstate);
return;
}
/*
* The disk labeling is asynchronous on Linux. Just record
* this label request and return as there will be another
* disk add event for the partition after the labeling is
* completed.
*/
device = malloc(sizeof (pendingdev_t));
(void) strlcpy(device->pd_physpath, physpath,
sizeof (device->pd_physpath));
list_insert_tail(&g_device_list, device);
zed_log_msg(LOG_INFO, " zpool_label_disk: async '%s' (%llu)",
leafname, (long long unsigned int)guid);
return; /* resumes at EC_DEV_ADD.ESC_DISK for partition */
} else { /* labeled */
boolean_t found = B_FALSE;
/*
* match up with request above to label the disk
*/
for (device = list_head(&g_device_list); device != NULL;
device = list_next(&g_device_list, device)) {
if (strcmp(physpath, device->pd_physpath) == 0) {
list_remove(&g_device_list, device);
free(device);
found = B_TRUE;
break;
}
}
if (!found) {
/* unexpected partition slice encountered */
(void) zpool_vdev_online(zhp, fullpath,
ZFS_ONLINE_FORCEFAULT, &newstate);
return;
}
zed_log_msg(LOG_INFO, " zpool_label_disk: resume '%s' (%llu)",
physpath, (long long unsigned int)guid);
if (nvlist_lookup_string(vdev, "new_devid", &new_devid) != 0) {
zed_log_msg(LOG_INFO, " auto replace: missing devid!");
return;
}
(void) snprintf(devpath, sizeof (devpath), "%s%s",
DEV_BYID_PATH, new_devid);
path = devpath;
}
/*
* Construct the root vdev to pass to zpool_vdev_attach(). While adding
* the entire vdev structure is harmless, we construct a reduced set of
* path/physpath/wholedisk to keep it simple.
*/
if (nvlist_alloc(&nvroot, NV_UNIQUE_NAME, 0) != 0) {
zed_log_msg(LOG_WARNING, "zfs_mod: nvlist_alloc out of memory");
return;
}
if (nvlist_alloc(&newvd, NV_UNIQUE_NAME, 0) != 0) {
zed_log_msg(LOG_WARNING, "zfs_mod: nvlist_alloc out of memory");
nvlist_free(nvroot);
return;
}
if (nvlist_add_string(newvd, ZPOOL_CONFIG_TYPE, VDEV_TYPE_DISK) != 0 ||
nvlist_add_string(newvd, ZPOOL_CONFIG_PATH, path) != 0 ||
nvlist_add_string(newvd, ZPOOL_CONFIG_DEVID, new_devid) != 0 ||
(physpath != NULL && nvlist_add_string(newvd,
ZPOOL_CONFIG_PHYS_PATH, physpath) != 0) ||
nvlist_add_uint64(newvd, ZPOOL_CONFIG_WHOLE_DISK, wholedisk) != 0 ||
nvlist_add_string(nvroot, ZPOOL_CONFIG_TYPE, VDEV_TYPE_ROOT) != 0 ||
nvlist_add_nvlist_array(nvroot, ZPOOL_CONFIG_CHILDREN, &newvd,
1) != 0) {
zed_log_msg(LOG_WARNING, "zfs_mod: unable to add nvlist pairs");
nvlist_free(newvd);
nvlist_free(nvroot);
return;
}
nvlist_free(newvd);
/*
* auto replace a leaf disk at same physical location
*/
ret = zpool_vdev_attach(zhp, fullpath, path, nvroot, B_TRUE);
zed_log_msg(LOG_INFO, " zpool_vdev_replace: %s with %s (%s)",
fullpath, path, (ret == 0) ? "no errors" :
libzfs_error_description(g_zfshdl));
nvlist_free(nvroot);
}
/*
* Mount the given filesystem.
*
* 'flags' appears pretty much always 0 here.
*/
int
zfs_mount(zfs_handle_t *zhp, const char *options, int flags)
{
struct stat buf;
char mountpoint[ZFS_MAXPROPLEN];
char mntopts[MNT_LINE_MAX];
libzfs_handle_t *hdl = zhp->zfs_hdl;
int remount;
if (options == NULL) {
mntopts[0] = '\0';
} else {
(void) strlcpy(mntopts, options, sizeof (mntopts));
}
if (strstr(mntopts, MNTOPT_REMOUNT) != NULL)
remount = 1;
/*
* If the pool is imported read-only then all mounts must be read-only
*/
#ifdef __LINUX__
if (zpool_get_prop_int(zhp->zpool_hdl, ZPOOL_PROP_READONLY, NULL))
(void) strlcat(mntopts, "," MNTOPT_RO, sizeof (mntopts));
#else
if (zpool_get_prop_int(zhp->zpool_hdl, ZPOOL_PROP_READONLY, NULL))
flags |= MS_RDONLY;
#endif /* __LINUX__ */
if (!zfs_is_mountable(zhp, mountpoint, sizeof (mountpoint), NULL))
return (0);
#ifdef __LINUX__
/*
* Append default mount options which apply to the mount point.
* This is done because under Linux (unlike Solaris) multiple mount
* points may reference a single super block. This means that just
* given a super block there is no back reference to update the per
* mount point options.
*/
rc = zfs_add_options(zhp, &flags);
if (rc) {
zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
"default options unavailable"));
return (zfs_error_fmt(hdl, EZFS_MOUNTFAILED,
dgettext(TEXT_DOMAIN, "cannot mount '%s'"),
mountpoint));
}
/*
* Append zfsutil option so the mount helper allow the mount
*/
strlcat(mntopts, "," MNTOPT_ZFSUTIL, sizeof (mntopts));
#endif /* __LINUX__ */
/* Create the directory if it doesn't already exist */
#ifdef __APPLE__
if (zfs_get_type(zhp) != ZFS_TYPE_SNAPSHOT &&
lstat(mountpoint, &buf) != 0) {
#else
if (lstat(mountpoint, &buf) != 0) {
#endif
if (mkdirp(mountpoint, 0755) != 0) {
zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
"failed to create mountpoint"));
return (zfs_error_fmt(hdl, EZFS_MOUNTFAILED,
dgettext(TEXT_DOMAIN, "cannot mount '%s'"),
mountpoint));
}
}
/*
* Determine if the mountpoint is empty. If so, refuse to perform the
* mount. We don't perform this check if 'remount' is
* specified or if overlay option(-O) is given
*/
if ((flags & MS_OVERLAY) == 0 && !remount &&
!dir_is_empty(mountpoint)) {
zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
"directory is not empty"));
return (zfs_error_fmt(hdl, EZFS_MOUNTFAILED,
dgettext(TEXT_DOMAIN, "cannot mount '%s'"), mountpoint));
}
/* perform the mount */
#ifdef __LINUX__
rc = do_mount(zfs_get_name(zhp), mountpoint, mntopts);
#elif defined(__APPLE__) || defined (__FREEBSD__)
if (zmount(zfs_get_name(zhp), mountpoint, MS_OPTIONSTR | flags,
MNTTYPE_ZFS, NULL, 0, mntopts, sizeof (mntopts)) != 0) {
#elif defined(__illumos__)
if (mount(zfs_get_name(zhp), mountpoint, MS_OPTIONSTR | flags,
MNTTYPE_ZFS, NULL, 0, mntopts, sizeof (mntopts)) != 0) {
#endif /* __LINUX__*/
/*
* Generic errors are nasty, but there are just way too many
* from mount(), and they're well-understood. We pick a few
* common ones to improve upon.
*/
if (errno == EBUSY) {
zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
"mountpoint or dataset is busy"));
} else if (errno == EPERM) {
zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
"Insufficient privileges"));
} else if (errno == ENOTSUP) {
char buf[256];
int spa_version;
VERIFY(zfs_spa_version(zhp, &spa_version) == 0);
(void) snprintf(buf, sizeof (buf),
dgettext(TEXT_DOMAIN, "Can't mount a version %lld "
"file system on a version %d pool. Pool must be"
" upgraded to mount this file system."),
(u_longlong_t)zfs_prop_get_int(zhp,
ZFS_PROP_VERSION), spa_version);
zfs_error_aux(hdl, dgettext(TEXT_DOMAIN, buf));
#ifdef __APPLE__
} else if (((errno == ESRCH) || (errno == EINVAL) ||
(errno == ENOENT && lstat(mountpoint, &buf) != 0)) &&
zfs_get_type(zhp) == ZFS_TYPE_SNAPSHOT) {
zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
"The parent file system must be mounted first."));
#endif
} else {
zfs_error_aux(hdl, strerror(errno));
}
return (zfs_error_fmt(hdl, EZFS_MOUNTFAILED,
dgettext(TEXT_DOMAIN, "cannot mount '%s'"),
zhp->zfs_name));
}
#ifdef __APPLE__
if (zhp->zfs_type == ZFS_TYPE_SNAPSHOT)
fprintf(stderr, "ZFS: snapshot mountpoint '%s'\n", mountpoint);
if (!(flags & MS_RDONLY))
zfs_mount_seticon(mountpoint);
#endif
/* remove the mounted entry before re-adding on remount */
if (remount)
libzfs_mnttab_remove(hdl, zhp->zfs_name);
/* add the mounted entry into our cache */
libzfs_mnttab_add(hdl, zfs_get_name(zhp), mountpoint, mntopts);
return (0);
}
/*
* Unmount a single filesystem.
*/
static int
unmount_one(libzfs_handle_t *hdl, const char *mountpoint, int flags)
{
int error;
#if 0
error = unmount(mountpoint, flags);
if (unmount(mountpoint, flags) != 0) {
return (zfs_error_fmt(hdl, EZFS_UMOUNTFAILED,
dgettext(TEXT_DOMAIN, "cannot unmount '%s'"),
mountpoint));
}
#else
error = do_unmount(mountpoint, flags);
if (error != 0) {
return (zfs_error_fmt(hdl, EZFS_UMOUNTFAILED,
dgettext(TEXT_DOMAIN, "cannot unmount '%s'"),
mountpoint));
}
#endif
return (0);
}
/*
* Unmount the given filesystem.
*/
int
zfs_unmount(zfs_handle_t *zhp, const char *mountpoint, int flags)
{
libzfs_handle_t *hdl = zhp->zfs_hdl;
#ifdef __LINUX__
struct mnttab search = { 0 }, entry;
#else
struct mnttab entry;
#endif /* __LINUX__ */
char *mntpt = NULL;
/* check to see if need to unmount the filesystem */
if (mountpoint != NULL ||
(((zfs_get_type(zhp) == ZFS_TYPE_FILESYSTEM) ||
(zfs_get_type(zhp) == ZFS_TYPE_SNAPSHOT)) &&
libzfs_mnttab_find(hdl, zhp->zfs_name, &entry) == 0)) {
/*
* mountpoint may have come from a call to
* getmnt/getmntany if it isn't NULL. If it is NULL,
* we know it comes from getmntany which can then get
* overwritten later. We strdup it to play it safe.
*/
if (mountpoint == NULL)
mntpt = zfs_strdup(zhp->zfs_hdl, entry.mnt_mountp);
else
mntpt = zfs_strdup(zhp->zfs_hdl, mountpoint);
/*
* Unshare and unmount the filesystem
*/
#ifdef __illumos__
if (zfs_unshare_proto(zhp, mntpt, share_all_proto) != 0)
#else
if (zfs_unshare_nfs(zhp, mntpt) != 0)
#endif
return (-1);
if (unmount_one(hdl, mntpt, flags) != 0) {
free(mntpt);
#ifdef __illumos__
(void) zfs_shareall(zhp);
#else
(void) zfs_share_nfs(zhp);
#endif
return (-1);
}
libzfs_mnttab_remove(hdl, zhp->zfs_name);
free(mntpt);
}
return (0);
}