/*
* Add the environment variable specified by the format string [fmt].
*/
static void
_zed_event_add_var(uint64_t eid, zed_strings_t *zsp, const char *fmt, ...)
{
char buf[4096];
va_list vargs;
int n;
const char *p;
size_t namelen;
assert(zsp != NULL);
assert(fmt != NULL);
va_start(vargs, fmt);
n = vsnprintf(buf, sizeof (buf), fmt, vargs);
va_end(vargs);
p = strchr(buf, '=');
namelen = (p) ? p - buf : strlen(buf);
if ((n < 0) || (n >= sizeof (buf))) {
zed_log_msg(LOG_WARNING, "Failed to add %.*s for eid=%llu: %s",
namelen, buf, eid, "Exceeded buffer size");
} else if (!p) {
zed_log_msg(LOG_WARNING, "Failed to add %.*s for eid=%llu: %s",
namelen, buf, eid, "Missing assignment");
} else if (zed_strings_add(zsp, buf) < 0) {
zed_log_msg(LOG_WARNING, "Failed to add %.*s for eid=%llu: %s",
namelen, buf, eid, strerror(ENOMEM));
}
}
static void
zfs_iter_vdev(zpool_handle_t *zhp, nvlist_t *nvl, void *data)
{
dev_data_t *dp = data;
char *path;
uint_t c, children;
nvlist_t **child;
/*
* First iterate over any children.
*/
if (nvlist_lookup_nvlist_array(nvl, ZPOOL_CONFIG_CHILDREN,
&child, &children) == 0) {
for (c = 0; c < children; c++)
zfs_iter_vdev(zhp, child[c], data);
return;
}
/* once a vdev was matched and processed there is nothing left to do */
if (dp->dd_found)
return;
/*
* Match by GUID if available otherwise fallback to devid or physical
*/
if (dp->dd_vdev_guid != 0) {
uint64_t guid;
if (nvlist_lookup_uint64(nvl, ZPOOL_CONFIG_GUID,
&guid) != 0 || guid != dp->dd_vdev_guid) {
return;
}
zed_log_msg(LOG_INFO, " zfs_iter_vdev: matched on %llu", guid);
dp->dd_found = B_TRUE;
} else if (dp->dd_compare != NULL) {
/*
* NOTE: On Linux there is an event for partition, so unlike
* illumos, substring matching is not required to accomodate
* the partition suffix. An exact match will be present in
* the dp->dd_compare value.
*/
if (nvlist_lookup_string(nvl, dp->dd_prop, &path) != 0 ||
strcmp(dp->dd_compare, path) != 0) {
return;
}
zed_log_msg(LOG_INFO, " zfs_iter_vdev: matched %s on %s",
dp->dd_prop, path);
dp->dd_found = B_TRUE;
/* pass the new devid for use by replacing code */
if (dp->dd_islabeled && dp->dd_new_devid != NULL) {
(void) nvlist_add_string(nvl, "new_devid",
dp->dd_new_devid);
}
}
(dp->dd_func)(zhp, nvl, dp->dd_islabeled);
}
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);
}
/*
* Write the [eid] & [etime] of the last processed event to the opened
* [zcp] state_file. Note that etime[] is an array of size 2.
* Return 0 on success, -1 on error.
*/
int
zed_conf_write_state(struct zed_conf *zcp, uint64_t eid, int64_t etime[])
{
ssize_t len;
struct iovec iov[3];
ssize_t n;
if (!zcp) {
errno = EINVAL;
zed_log_msg(LOG_ERR,
"Failed to write state file: %s", strerror(errno));
return (-1);
}
if (lseek(zcp->state_fd, 0, SEEK_SET) == (off_t) -1) {
zed_log_msg(LOG_WARNING,
"Failed to reposition state file offset: %s",
strerror(errno));
return (-1);
}
len = 0;
iov[0].iov_base = &eid;
len += iov[0].iov_len = sizeof (eid);
iov[1].iov_base = &etime[0];
len += iov[1].iov_len = sizeof (etime[0]);
iov[2].iov_base = &etime[1];
len += iov[2].iov_len = sizeof (etime[1]);
n = writev(zcp->state_fd, iov, 3);
if (n < 0) {
zed_log_msg(LOG_WARNING,
"Failed to write state file \"%s\": %s",
zcp->state_file, strerror(errno));
return (-1);
}
if (n != len) {
errno = EIO;
zed_log_msg(LOG_WARNING,
"Failed to write state file \"%s\": Wrote %d of %d bytes",
zcp->state_file, n, len);
return (-1);
}
if (fdatasync(zcp->state_fd) < 0) {
zed_log_msg(LOG_WARNING,
"Failed to sync state file \"%s\": %s",
zcp->state_file, strerror(errno));
return (-1);
}
return (0);
}
/*
* Destroy the configuration [zcp].
*
* Note: zfs_hdl & zevent_fd are destroyed via zed_event_fini().
*/
void
zed_conf_destroy(struct zed_conf *zcp)
{
if (!zcp)
return;
if (zcp->state_fd >= 0) {
if (close(zcp->state_fd) < 0)
zed_log_msg(LOG_WARNING,
"Failed to close state file \"%s\": %s",
zcp->state_file, strerror(errno));
zcp->state_fd = -1;
}
if (zcp->pid_file) {
if ((unlink(zcp->pid_file) < 0) && (errno != ENOENT))
zed_log_msg(LOG_WARNING,
"Failed to remove PID file \"%s\": %s",
zcp->pid_file, strerror(errno));
}
if (zcp->pid_fd >= 0) {
if (close(zcp->pid_fd) < 0)
zed_log_msg(LOG_WARNING,
"Failed to close PID file \"%s\": %s",
zcp->pid_file, strerror(errno));
zcp->pid_fd = -1;
}
if (zcp->conf_file) {
free(zcp->conf_file);
zcp->conf_file = NULL;
}
if (zcp->pid_file) {
free(zcp->pid_file);
zcp->pid_file = NULL;
}
if (zcp->zedlet_dir) {
free(zcp->zedlet_dir);
zcp->zedlet_dir = NULL;
}
if (zcp->state_file) {
free(zcp->state_file);
zcp->state_file = NULL;
}
if (zcp->zedlets) {
zed_strings_destroy(zcp->zedlets);
zcp->zedlets = NULL;
}
free(zcp);
}
/*
* Handle a EC_DEV_ADD.ESC_DISK event.
*
* illumos
* Expects: DEV_PHYS_PATH string in schema
* Matches: vdev's ZPOOL_CONFIG_PHYS_PATH or ZPOOL_CONFIG_DEVID
*
* path: '/dev/dsk/c0t1d0s0' (persistent)
* devid: 'id1,sd@SATA_____Hitachi_HDS72101______JP2940HZ3H74MC/a'
* phys_path: '/pci@0,0/pci103c,1609@11/disk@1,0:a'
*
* linux
* provides: DEV_PHYS_PATH and DEV_IDENTIFIER strings in schema
* Matches: vdev's ZPOOL_CONFIG_PHYS_PATH or ZPOOL_CONFIG_DEVID
*
* path: '/dev/sdc1' (not persistent)
* devid: 'ata-SAMSUNG_HD204UI_S2HGJD2Z805891-part1'
* phys_path: 'pci-0000:04:00.0-sas-0x4433221106000000-lun-0'
*/
static int
zfs_deliver_add(nvlist_t *nvl, boolean_t is_lofi)
{
char *devpath = NULL, *devid;
boolean_t is_slice;
/*
* Expecting a devid string and an optional physical location
*/
if (nvlist_lookup_string(nvl, DEV_IDENTIFIER, &devid) != 0)
return (-1);
(void) nvlist_lookup_string(nvl, DEV_PHYS_PATH, &devpath);
is_slice = (nvlist_lookup_boolean(nvl, DEV_IS_PART) == 0);
zed_log_msg(LOG_INFO, "zfs_deliver_add: adding %s (%s) (is_slice %d)",
devid, devpath ? devpath : "NULL", is_slice);
/*
* Iterate over all vdevs looking for a match in the folllowing order:
* 1. ZPOOL_CONFIG_DEVID (identifies the unique disk)
* 2. ZPOOL_CONFIG_PHYS_PATH (identifies disk physical location).
*
* For disks, we only want to pay attention to vdevs marked as whole
* disks or are a multipath device.
*/
if (!devid_iter(devid, zfs_process_add, is_slice) && devpath != NULL)
(void) devphys_iter(devpath, devid, zfs_process_add, is_slice);
return (0);
}
/*
* This function handles the ESC_DEV_DLE event.
*/
static int
zfs_deliver_dle(nvlist_t *nvl)
{
char *devname;
if (nvlist_lookup_string(nvl, DEV_PHYS_PATH, &devname) != 0) {
zed_log_msg(LOG_INFO, "zfs_deliver_event: no physpath");
return (-1);
}
if (zpool_iter(g_zfshdl, zfsdle_vdev_online, devname) != 1) {
zed_log_msg(LOG_INFO, "zfs_deliver_event: device '%s' not "
"found", devname);
return (1);
}
return (0);
}
static int
_zed_event_add_array_err(uint64_t eid, const char *name)
{
errno = EMSGSIZE;
zed_log_msg(LOG_WARNING,
"Failed to convert nvpair \"%s\" for eid=%llu: "
"Exceeded buffer size", name, eid);
return (-1);
}
/*
* Read the opened [zcp] state_file to obtain the eid & etime of the last event
* processed. Write the state from the last event to the [eidp] & [etime] args
* passed by reference. Note that etime[] is an array of size 2.
* Return 0 on success, -1 on error.
*/
int
zed_conf_read_state(struct zed_conf *zcp, uint64_t *eidp, int64_t etime[])
{
ssize_t len;
struct iovec iov[3];
ssize_t n;
if (!zcp || !eidp || !etime) {
errno = EINVAL;
zed_log_msg(LOG_ERR,
"Failed to read state file: %s", strerror(errno));
return (-1);
}
if (lseek(zcp->state_fd, 0, SEEK_SET) == (off_t) -1) {
zed_log_msg(LOG_WARNING,
"Failed to reposition state file offset: %s",
strerror(errno));
return (-1);
}
len = 0;
iov[0].iov_base = eidp;
len += iov[0].iov_len = sizeof (*eidp);
iov[1].iov_base = &etime[0];
len += iov[1].iov_len = sizeof (etime[0]);
iov[2].iov_base = &etime[1];
len += iov[2].iov_len = sizeof (etime[1]);
n = readv(zcp->state_fd, iov, 3);
if (n == 0) {
*eidp = 0;
} else if (n < 0) {
zed_log_msg(LOG_WARNING,
"Failed to read state file \"%s\": %s",
zcp->state_file, strerror(errno));
return (-1);
} else if (n != len) {
errno = EIO;
zed_log_msg(LOG_WARNING,
"Failed to read state file \"%s\": Read %d of %d bytes",
zcp->state_file, n, len);
return (-1);
}
return (0);
}
/*
* This function handles the ESC_DEV_DLE device change event. Use the
* provided vdev guid when looking up a disk or partition, when the guid
* is not present assume the entire disk is owned by ZFS and append the
* expected -part1 partition information then lookup by physical path.
*/
static int
zfs_deliver_dle(nvlist_t *nvl)
{
char *devname, name[MAXPATHLEN];
uint64_t guid;
if (nvlist_lookup_uint64(nvl, ZFS_EV_VDEV_GUID, &guid) == 0) {
sprintf(name, "%llu", (u_longlong_t)guid);
} else if (nvlist_lookup_string(nvl, DEV_PHYS_PATH, &devname) == 0) {
strlcpy(name, devname, MAXPATHLEN);
zfs_append_partition(name, MAXPATHLEN);
} else {
zed_log_msg(LOG_INFO, "zfs_deliver_dle: no guid or physpath");
}
if (zpool_iter(g_zfshdl, zfsdle_vdev_online, name) != 1) {
zed_log_msg(LOG_INFO, "zfs_deliver_dle: device '%s' not "
"found", name);
return (1);
}
return (0);
}
static int
zfs_iter_pool(zpool_handle_t *zhp, void *data)
{
nvlist_t *config, *nvl;
dev_data_t *dp = data;
uint64_t pool_guid;
unavailpool_t *pool;
zed_log_msg(LOG_INFO, "zfs_iter_pool: evaluating vdevs on %s (by %s)",
zpool_get_name(zhp), dp->dd_vdev_guid ? "GUID" : dp->dd_prop);
/*
* For each vdev in this pool, look for a match to apply dd_func
*/
if ((config = zpool_get_config(zhp, NULL)) != NULL) {
if (dp->dd_pool_guid == 0 ||
(nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_GUID,
&pool_guid) == 0 && pool_guid == dp->dd_pool_guid)) {
(void) nvlist_lookup_nvlist(config,
ZPOOL_CONFIG_VDEV_TREE, &nvl);
zfs_iter_vdev(zhp, nvl, data);
}
}
/*
* if this pool was originally unavailable,
* then enable its datasets asynchronously
*/
if (g_enumeration_done) {
for (pool = list_head(&g_pool_list); pool != NULL;
pool = list_next(&g_pool_list, pool)) {
if (pool->uap_enable_tid != 0)
continue; /* entry already processed */
if (strcmp(zpool_get_name(zhp),
zpool_get_name(pool->uap_zhp)))
continue;
if (zfs_toplevel_state(zhp) >= VDEV_STATE_DEGRADED) {
/* send to a background thread; keep on list */
(void) pthread_create(&pool->uap_enable_tid,
NULL, zfs_enable_ds, pool);
break;
}
}
}
zpool_close(zhp);
return (dp->dd_found); /* cease iteration after a match */
}
static int
zfs_unavail_pool(zpool_handle_t *zhp, void *data)
{
zed_log_msg(LOG_INFO, "zfs_unavail_pool: examining '%s' (state %d)",
zpool_get_name(zhp), (int)zfs_toplevel_state(zhp));
if (zfs_toplevel_state(zhp) < VDEV_STATE_DEGRADED) {
unavailpool_t *uap;
uap = malloc(sizeof (unavailpool_t));
uap->uap_zhp = zhp;
list_insert_tail((list_t *)data, uap);
} else {
zpool_close(zhp);
}
return (0);
}
/*
* Close the libzfs interface.
*/
void
zed_event_fini(struct zed_conf *zcp)
{
if (!zcp)
zed_log_die("Failed zed_event_fini: %s", strerror(EINVAL));
if (zcp->zevent_fd >= 0) {
if (close(zcp->zevent_fd) < 0)
zed_log_msg(LOG_WARNING, "Failed to close \"%s\": %s",
ZFS_DEV, strerror(errno));
zcp->zevent_fd = -1;
}
if (zcp->zfs_hdl) {
libzfs_fini(zcp->zfs_hdl);
zcp->zfs_hdl = NULL;
}
}
/*
* Called when we receive a VDEV_CHECK event, which indicates a device could not
* be opened during initial pool open, but the autoreplace property was set on
* the pool. In this case, we treat it as if it were an add event.
*/
static int
zfs_deliver_check(nvlist_t *nvl)
{
dev_data_t data = { 0 };
if (nvlist_lookup_uint64(nvl, ZFS_EV_POOL_GUID,
&data.dd_pool_guid) != 0 ||
nvlist_lookup_uint64(nvl, ZFS_EV_VDEV_GUID,
&data.dd_vdev_guid) != 0 ||
data.dd_vdev_guid == 0)
return (0);
zed_log_msg(LOG_INFO, "zfs_deliver_check: pool '%llu', vdev %llu",
data.dd_pool_guid, data.dd_vdev_guid);
data.dd_func = zfs_process_add;
(void) zpool_iter(g_zfshdl, zfs_iter_pool, &data);
return (0);
}
/*
* Lock all current and future pages in the virtual memory address space.
* Access to locked pages will never be delayed by a page fault.
* EAGAIN is tested up to max_tries in case this is a transient error.
*/
static void
_lock_memory(void)
{
#if HAVE_MLOCKALL
int i = 0;
const int max_tries = 10;
for (i = 0; i < max_tries; i++) {
if (mlockall(MCL_CURRENT | MCL_FUTURE) == 0) {
zed_log_msg(LOG_INFO, "Locked all pages in memory");
return;
}
if (errno != EAGAIN)
break;
}
zed_log_die("Failed to lock memory pages: %s", strerror(errno));
#else /* HAVE_MLOCKALL */
zed_log_die("Failed to lock memory pages: mlockall() not supported");
#endif /* HAVE_MLOCKALL */
}
/*
* 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);
}
/*
* ZFS Event Daemon (ZED).
*/
int
main(int argc, char *argv[])
{
struct zed_conf *zcp;
uint64_t saved_eid;
int64_t saved_etime[2];
zed_log_init(argv[0]);
zed_log_stderr_open(LOG_NOTICE);
zcp = zed_conf_create();
zed_conf_parse_opts(zcp, argc, argv);
if (zcp->do_verbose)
zed_log_stderr_open(LOG_INFO);
if (geteuid() != 0)
zed_log_die("Must be run as root");
(void) umask(0);
_setup_sig_handlers();
zed_conf_parse_file(zcp);
zed_file_close_from(STDERR_FILENO + 1);
if (chdir("/") < 0)
zed_log_die("Failed to change to root directory");
if (zed_conf_scan_dir(zcp) < 0)
exit(EXIT_FAILURE);
if (zcp->do_memlock)
_lock_memory();
if (!zcp->do_foreground) {
_become_daemon();
zed_log_syslog_open(LOG_DAEMON);
zed_log_stderr_close();
}
zed_log_msg(LOG_NOTICE,
"ZFS Event Daemon %s-%s", ZFS_META_VERSION, ZFS_META_RELEASE);
(void) zed_conf_write_pid(zcp);
if (zed_conf_open_state(zcp) < 0)
exit(EXIT_FAILURE);
if (zed_conf_read_state(zcp, &saved_eid, saved_etime) < 0)
exit(EXIT_FAILURE);
retry:
if (zed_event_init(zcp)) {
/*
* If we failed to open /dev/zfs, but force was requested, we
* sleep waiting for it to come alive. This lets zed sit around
* waiting for the kernel module to load.
*/
if (zcp->do_force) {
sleep(30);
if (!_got_exit) goto retry;
}
zed_log_die("Failed to initialize libzfs");
}
zed_event_seek(zcp, saved_eid, saved_etime);
while (!_got_exit) {
if (_got_hup) {
_got_hup = 0;
(void) zed_conf_scan_dir(zcp);
}
if (zed_event_service(zcp)) break;
}
zed_log_msg(LOG_NOTICE, "Exiting");
zed_event_fini(zcp);
if (zcp->do_force && !_got_exit) {
goto retry;
}
zed_conf_destroy(zcp);
zed_log_fini();
exit(EXIT_SUCCESS);
}
/*
* ZFS Event Daemon (ZED).
*/
int
main(int argc, char *argv[])
{
struct zed_conf *zcp;
uint64_t saved_eid;
int64_t saved_etime[2];
zed_log_init(argv[0]);
zed_log_stderr_open(LOG_NOTICE);
zcp = zed_conf_create();
zed_conf_parse_opts(zcp, argc, argv);
if (zcp->do_verbose)
zed_log_stderr_open(LOG_INFO);
if (geteuid() != 0)
zed_log_die("Must be run as root");
(void) umask(0);
_setup_sig_handlers();
zed_conf_parse_file(zcp);
zed_file_close_from(STDERR_FILENO + 1);
if (chdir("/") < 0)
zed_log_die("Failed to change to root directory");
if (zed_conf_scan_dir(zcp) < 0)
exit(EXIT_FAILURE);
if (zcp->do_memlock)
_lock_memory();
if (!zcp->do_foreground) {
_become_daemon();
zed_log_syslog_open(LOG_DAEMON);
zed_log_stderr_close();
}
zed_log_msg(LOG_NOTICE,
"ZFS Event Daemon %s-%s", ZFS_META_VERSION, ZFS_META_RELEASE);
(void) zed_conf_write_pid(zcp);
if (zed_conf_open_state(zcp) < 0)
exit(EXIT_FAILURE);
if (zed_conf_read_state(zcp, &saved_eid, saved_etime) < 0)
exit(EXIT_FAILURE);
zed_event_init(zcp);
zed_event_seek(zcp, saved_eid, saved_etime);
while (!_got_exit) {
if (_got_hup) {
_got_hup = 0;
(void) zed_conf_scan_dir(zcp);
}
zed_event_service(zcp);
}
zed_log_msg(LOG_NOTICE, "Exiting");
zed_event_fini(zcp);
zed_conf_destroy(zcp);
zed_log_fini();
exit(EXIT_SUCCESS);
}
/*
* Add an environment variable for [eid] to the container [zsp].
*
* The variable name is the concatenation of [prefix] and [name] converted to
* uppercase with non-alphanumeric characters converted to underscores;
* [prefix] is optional, and [name] must begin with an alphabetic character.
* If the converted variable name already exists within the container [zsp],
* its existing value will be replaced with the new value.
*
* The variable value is specified by the format string [fmt].
*
* Returns 0 on success, and -1 on error (with errno set).
*
* All environment variables in [zsp] should be added through this function.
*/
static int
_zed_event_add_var(uint64_t eid, zed_strings_t *zsp,
const char *prefix, const char *name, const char *fmt, ...)
{
char keybuf[MAXBUF];
char valbuf[MAXBUF];
char *dstp;
const char *srcp;
const char *lastp;
int n;
int buflen;
va_list vargs;
assert(zsp != NULL);
assert(fmt != NULL);
if (!name) {
errno = EINVAL;
zed_log_msg(LOG_WARNING,
"Failed to add variable for eid=%llu: Name is empty", eid);
return (-1);
} else if (!isalpha(name[0])) {
errno = EINVAL;
zed_log_msg(LOG_WARNING,
"Failed to add variable for eid=%llu: "
"Name \"%s\" is invalid", eid, name);
return (-1);
}
/*
* Construct the string key by converting PREFIX (if present) and NAME.
*/
dstp = keybuf;
lastp = keybuf + sizeof (keybuf);
if (prefix) {
for (srcp = prefix; *srcp && (dstp < lastp); srcp++)
*dstp++ = isalnum(*srcp) ? toupper(*srcp) : '_';
}
for (srcp = name; *srcp && (dstp < lastp); srcp++)
*dstp++ = isalnum(*srcp) ? toupper(*srcp) : '_';
if (dstp == lastp) {
errno = ENAMETOOLONG;
zed_log_msg(LOG_WARNING,
"Failed to add variable for eid=%llu: Name too long", eid);
return (-1);
}
*dstp = '\0';
/*
* Construct the string specified by "[PREFIX][NAME]=[FMT]".
*/
dstp = valbuf;
buflen = sizeof (valbuf);
n = strlcpy(dstp, keybuf, buflen);
if (n >= sizeof (valbuf)) {
errno = EMSGSIZE;
zed_log_msg(LOG_WARNING, "Failed to add %s for eid=%llu: %s",
keybuf, eid, "Exceeded buffer size");
return (-1);
}
dstp += n;
buflen -= n;
*dstp++ = '=';
buflen--;
va_start(vargs, fmt);
n = vsnprintf(dstp, buflen, fmt, vargs);
va_end(vargs);
if ((n < 0) || (n >= buflen)) {
errno = EMSGSIZE;
zed_log_msg(LOG_WARNING, "Failed to add %s for eid=%llu: %s",
keybuf, eid, "Exceeded buffer size");
return (-1);
} else if (zed_strings_add(zsp, keybuf, valbuf) < 0) {
zed_log_msg(LOG_WARNING, "Failed to add %s for eid=%llu: %s",
keybuf, eid, strerror(errno));
return (-1);
}
return (0);
}
/*
* Convert the nvpair [nvp] to a string which is added to the environment
* of the child process.
* Return 0 on success, -1 on error.
*
* FIXME: Refactor with cmd/zpool/zpool_main.c:zpool_do_events_nvprint()?
*/
static void
_zed_event_add_nvpair(uint64_t eid, zed_strings_t *zsp, nvpair_t *nvp)
{
const char *name;
data_type_t type;
const char *prefix = ZEVENT_VAR_PREFIX;
boolean_t b;
double d;
uint8_t i8;
uint16_t i16;
uint32_t i32;
uint64_t i64;
char *str;
assert(zsp != NULL);
assert(nvp != NULL);
name = nvpair_name(nvp);
type = nvpair_type(nvp);
switch (type) {
case DATA_TYPE_BOOLEAN:
_zed_event_add_var(eid, zsp, prefix, name, "%s", "1");
break;
case DATA_TYPE_BOOLEAN_VALUE:
(void) nvpair_value_boolean_value(nvp, &b);
_zed_event_add_var(eid, zsp, prefix, name, "%s", b ? "1" : "0");
break;
case DATA_TYPE_BYTE:
(void) nvpair_value_byte(nvp, &i8);
_zed_event_add_var(eid, zsp, prefix, name, "%d", i8);
break;
case DATA_TYPE_INT8:
(void) nvpair_value_int8(nvp, (int8_t *) &i8);
_zed_event_add_var(eid, zsp, prefix, name, "%d", i8);
break;
case DATA_TYPE_UINT8:
(void) nvpair_value_uint8(nvp, &i8);
_zed_event_add_var(eid, zsp, prefix, name, "%u", i8);
break;
case DATA_TYPE_INT16:
(void) nvpair_value_int16(nvp, (int16_t *) &i16);
_zed_event_add_var(eid, zsp, prefix, name, "%d", i16);
break;
case DATA_TYPE_UINT16:
(void) nvpair_value_uint16(nvp, &i16);
_zed_event_add_var(eid, zsp, prefix, name, "%u", i16);
break;
case DATA_TYPE_INT32:
(void) nvpair_value_int32(nvp, (int32_t *) &i32);
_zed_event_add_var(eid, zsp, prefix, name, "%d", i32);
break;
case DATA_TYPE_UINT32:
(void) nvpair_value_uint32(nvp, &i32);
_zed_event_add_var(eid, zsp, prefix, name, "%u", i32);
break;
case DATA_TYPE_INT64:
(void) nvpair_value_int64(nvp, (int64_t *) &i64);
_zed_event_add_var(eid, zsp, prefix, name,
"%lld", (longlong_t) i64);
break;
case DATA_TYPE_UINT64:
(void) nvpair_value_uint64(nvp, &i64);
_zed_event_add_var(eid, zsp, prefix, name,
(_zed_event_value_is_hex(name) ? "0x%.16llX" : "%llu"),
(u_longlong_t) i64);
/*
* shadow readable strings for vdev state pairs
*/
if (strcmp(name, FM_EREPORT_PAYLOAD_ZFS_VDEV_STATE) == 0 ||
strcmp(name, FM_EREPORT_PAYLOAD_ZFS_VDEV_LASTSTATE) == 0) {
char alt[32];
(void) snprintf(alt, sizeof (alt), "%s_str", name);
_zed_event_add_var(eid, zsp, prefix, alt, "%s",
zpool_state_to_name(i64, VDEV_AUX_NONE));
}
break;
case DATA_TYPE_DOUBLE:
(void) nvpair_value_double(nvp, &d);
_zed_event_add_var(eid, zsp, prefix, name, "%g", d);
break;
case DATA_TYPE_HRTIME:
(void) nvpair_value_hrtime(nvp, (hrtime_t *) &i64);
_zed_event_add_var(eid, zsp, prefix, name,
"%llu", (u_longlong_t) i64);
break;
case DATA_TYPE_NVLIST:
_zed_event_add_var(eid, zsp, prefix, name,
"%s", "_NOT_IMPLEMENTED_"); /* FIXME */
break;
case DATA_TYPE_STRING:
(void) nvpair_value_string(nvp, &str);
_zed_event_add_var(eid, zsp, prefix, name,
"%s", (str ? str : "<NULL>"));
break;
case DATA_TYPE_BOOLEAN_ARRAY:
_zed_event_add_var(eid, zsp, prefix, name,
"%s", "_NOT_IMPLEMENTED_"); /* FIXME */
break;
case DATA_TYPE_BYTE_ARRAY:
_zed_event_add_var(eid, zsp, prefix, name,
"%s", "_NOT_IMPLEMENTED_"); /* FIXME */
break;
case DATA_TYPE_INT8_ARRAY:
_zed_event_add_int8_array(eid, zsp, prefix, nvp);
break;
case DATA_TYPE_UINT8_ARRAY:
_zed_event_add_uint8_array(eid, zsp, prefix, nvp);
break;
case DATA_TYPE_INT16_ARRAY:
_zed_event_add_int16_array(eid, zsp, prefix, nvp);
break;
case DATA_TYPE_UINT16_ARRAY:
_zed_event_add_uint16_array(eid, zsp, prefix, nvp);
break;
case DATA_TYPE_INT32_ARRAY:
_zed_event_add_int32_array(eid, zsp, prefix, nvp);
break;
case DATA_TYPE_UINT32_ARRAY:
_zed_event_add_uint32_array(eid, zsp, prefix, nvp);
break;
case DATA_TYPE_INT64_ARRAY:
_zed_event_add_int64_array(eid, zsp, prefix, nvp);
break;
case DATA_TYPE_UINT64_ARRAY:
_zed_event_add_uint64_array(eid, zsp, prefix, nvp);
break;
case DATA_TYPE_STRING_ARRAY:
_zed_event_add_string_array(eid, zsp, prefix, nvp);
break;
case DATA_TYPE_NVLIST_ARRAY:
_zed_event_add_var(eid, zsp, prefix, name,
"%s", "_NOT_IMPLEMENTED_"); /* FIXME */
break;
default:
errno = EINVAL;
zed_log_msg(LOG_WARNING,
"Failed to convert nvpair \"%s\" for eid=%llu: "
"Unrecognized type=%u", name, eid, (unsigned int) type);
break;
}
}
/*
* Open and lock the [zcp] state_file.
* Return 0 on success, -1 on error.
*
* FIXME: Move state information into kernel.
*/
int
zed_conf_open_state(struct zed_conf *zcp)
{
char dirbuf[PATH_MAX];
mode_t dirmode = S_IRWXU | S_IRGRP | S_IXGRP | S_IROTH | S_IXOTH;
int n;
char *p;
int rv;
if (!zcp || !zcp->state_file) {
errno = EINVAL;
zed_log_msg(LOG_ERR, "Failed to open state file: %s",
strerror(errno));
return (-1);
}
n = strlcpy(dirbuf, zcp->state_file, sizeof (dirbuf));
if (n >= sizeof (dirbuf)) {
errno = ENAMETOOLONG;
zed_log_msg(LOG_WARNING, "Failed to open state file: %s",
strerror(errno));
return (-1);
}
p = strrchr(dirbuf, '/');
if (p)
*p = '\0';
if ((mkdirp(dirbuf, dirmode) < 0) && (errno != EEXIST)) {
zed_log_msg(LOG_WARNING,
"Failed to create directory \"%s\": %s",
dirbuf, strerror(errno));
return (-1);
}
if (zcp->state_fd >= 0) {
if (close(zcp->state_fd) < 0) {
zed_log_msg(LOG_WARNING,
"Failed to close state file \"%s\": %s",
zcp->state_file, strerror(errno));
return (-1);
}
}
if (zcp->do_zero)
(void) unlink(zcp->state_file);
zcp->state_fd = open(zcp->state_file,
(O_RDWR | O_CREAT), (S_IRUSR | S_IWUSR | S_IRGRP | S_IROTH));
if (zcp->state_fd < 0) {
zed_log_msg(LOG_WARNING, "Failed to open state file \"%s\": %s",
zcp->state_file, strerror(errno));
return (-1);
}
rv = zed_file_lock(zcp->state_fd);
if (rv < 0) {
zed_log_msg(LOG_WARNING, "Failed to lock state file \"%s\": %s",
zcp->state_file, strerror(errno));
return (-1);
}
if (rv > 0) {
pid_t pid = zed_file_is_locked(zcp->state_fd);
if (pid < 0) {
zed_log_msg(LOG_WARNING,
"Failed to test lock on state file \"%s\"",
zcp->state_file);
} else if (pid > 0) {
zed_log_msg(LOG_WARNING,
"Found PID %d bound to state file \"%s\"",
pid, zcp->state_file);
} else {
zed_log_msg(LOG_WARNING,
"Inconsistent lock state on state file \"%s\"",
zcp->state_file);
}
return (-1);
}
return (0);
}
/*
* Write the PID file specified in [zcp].
* Return 0 on success, -1 on error.
*
* This must be called after fork()ing to become a daemon (so the correct PID
* is recorded), but before daemonization is complete and the parent process
* exits (for synchronization with systemd).
*/
int
zed_conf_write_pid(struct zed_conf *zcp)
{
const mode_t dirmode = S_IRWXU | S_IRGRP | S_IXGRP | S_IROTH | S_IXOTH;
const mode_t filemode = S_IRUSR | S_IWUSR | S_IRGRP | S_IROTH;
char buf[PATH_MAX];
int n;
char *p;
mode_t mask;
int rv;
if (!zcp || !zcp->pid_file) {
errno = EINVAL;
zed_log_msg(LOG_ERR, "Failed to create PID file: %s",
strerror(errno));
return (-1);
}
assert(zcp->pid_fd == -1);
/*
* Create PID file directory if needed.
*/
n = strlcpy(buf, zcp->pid_file, sizeof (buf));
if (n >= sizeof (buf)) {
errno = ENAMETOOLONG;
zed_log_msg(LOG_ERR, "Failed to create PID file: %s",
strerror(errno));
goto err;
}
p = strrchr(buf, '/');
if (p)
*p = '\0';
if ((mkdirp(buf, dirmode) < 0) && (errno != EEXIST)) {
zed_log_msg(LOG_ERR, "Failed to create directory \"%s\": %s",
buf, strerror(errno));
goto err;
}
/*
* Obtain PID file lock.
*/
mask = umask(0);
umask(mask | 022);
zcp->pid_fd = open(zcp->pid_file, (O_RDWR | O_CREAT), filemode);
umask(mask);
if (zcp->pid_fd < 0) {
zed_log_msg(LOG_ERR, "Failed to open PID file \"%s\": %s",
zcp->pid_file, strerror(errno));
goto err;
}
rv = zed_file_lock(zcp->pid_fd);
if (rv < 0) {
zed_log_msg(LOG_ERR, "Failed to lock PID file \"%s\": %s",
zcp->pid_file, strerror(errno));
goto err;
} else if (rv > 0) {
pid_t pid = zed_file_is_locked(zcp->pid_fd);
if (pid < 0) {
zed_log_msg(LOG_ERR,
"Failed to test lock on PID file \"%s\"",
zcp->pid_file);
} else if (pid > 0) {
zed_log_msg(LOG_ERR,
"Found PID %d bound to PID file \"%s\"",
pid, zcp->pid_file);
} else {
zed_log_msg(LOG_ERR,
"Inconsistent lock state on PID file \"%s\"",
zcp->pid_file);
}
goto err;
}
/*
* Write PID file.
*/
n = snprintf(buf, sizeof (buf), "%d\n", (int) getpid());
if ((n < 0) || (n >= sizeof (buf))) {
errno = ERANGE;
zed_log_msg(LOG_ERR, "Failed to write PID file \"%s\": %s",
zcp->pid_file, strerror(errno));
} else if (zed_file_write_n(zcp->pid_fd, buf, n) != n) {
zed_log_msg(LOG_ERR, "Failed to write PID file \"%s\": %s",
zcp->pid_file, strerror(errno));
} else if (fdatasync(zcp->pid_fd) < 0) {
zed_log_msg(LOG_ERR, "Failed to sync PID file \"%s\": %s",
zcp->pid_file, strerror(errno));
} else {
return (0);
}
err:
if (zcp->pid_fd >= 0) {
(void) close(zcp->pid_fd);
zcp->pid_fd = -1;
}
return (-1);
}
/*
* Convert the nvpair [nvp] to a string which is added to the environment
* of the child process.
* Return 0 on success, -1 on error.
*
* FIXME: Refactor with cmd/zpool/zpool_main.c:zpool_do_events_nvprint()?
*/
static void
_zed_event_add_nvpair(uint64_t eid, zed_strings_t *zsp, nvpair_t *nvp)
{
const char *name;
data_type_t type;
char buf[4096];
int buflen;
int n;
char *p;
const char *q;
const char *fmt;
boolean_t b;
double d;
uint8_t i8;
uint16_t i16;
uint32_t i32;
uint64_t i64;
char *str;
assert(zsp != NULL);
assert(nvp != NULL);
name = nvpair_name(nvp);
type = nvpair_type(nvp);
buflen = sizeof (buf);
/* Copy NAME prefix for ZED zevent namespace. */
n = strlcpy(buf, ZEVENT_VAR_PREFIX, sizeof (buf));
if (n >= sizeof (buf)) {
zed_log_msg(LOG_WARNING,
"Failed to convert nvpair \"%s\" for eid=%llu: %s",
name, eid, "Exceeded buffer size");
return;
}
buflen -= n;
p = buf + n;
/* Convert NAME to alphanumeric uppercase. */
for (q = name; *q && (buflen > 0); q++) {
*p++ = isalnum(*q) ? toupper(*q) : '_';
buflen--;
}
/* Separate NAME from VALUE. */
if (buflen > 0) {
*p++ = '=';
buflen--;
}
*p = '\0';
/* Convert VALUE. */
switch (type) {
case DATA_TYPE_BOOLEAN:
n = snprintf(p, buflen, "%s", "1");
break;
case DATA_TYPE_BOOLEAN_VALUE:
(void) nvpair_value_boolean_value(nvp, &b);
n = snprintf(p, buflen, "%s", b ? "1" : "0");
break;
case DATA_TYPE_BYTE:
(void) nvpair_value_byte(nvp, &i8);
n = snprintf(p, buflen, "%d", i8);
break;
case DATA_TYPE_INT8:
(void) nvpair_value_int8(nvp, (int8_t *) &i8);
n = snprintf(p, buflen, "%d", i8);
break;
case DATA_TYPE_UINT8:
(void) nvpair_value_uint8(nvp, &i8);
n = snprintf(p, buflen, "%u", i8);
break;
case DATA_TYPE_INT16:
(void) nvpair_value_int16(nvp, (int16_t *) &i16);
n = snprintf(p, buflen, "%d", i16);
break;
case DATA_TYPE_UINT16:
(void) nvpair_value_uint16(nvp, &i16);
n = snprintf(p, buflen, "%u", i16);
break;
case DATA_TYPE_INT32:
(void) nvpair_value_int32(nvp, (int32_t *) &i32);
n = snprintf(p, buflen, "%d", i32);
break;
case DATA_TYPE_UINT32:
(void) nvpair_value_uint32(nvp, &i32);
n = snprintf(p, buflen, "%u", i32);
break;
case DATA_TYPE_INT64:
(void) nvpair_value_int64(nvp, (int64_t *) &i64);
n = snprintf(p, buflen, "%lld", (longlong_t) i64);
break;
case DATA_TYPE_UINT64:
(void) nvpair_value_uint64(nvp, &i64);
fmt = _zed_event_value_is_hex(name) ? "0x%.16llX" : "%llu";
n = snprintf(p, buflen, fmt, (u_longlong_t) i64);
break;
case DATA_TYPE_DOUBLE:
(void) nvpair_value_double(nvp, &d);
n = snprintf(p, buflen, "%g", d);
break;
case DATA_TYPE_HRTIME:
(void) nvpair_value_hrtime(nvp, (hrtime_t *) &i64);
n = snprintf(p, buflen, "%llu", (u_longlong_t) i64);
break;
case DATA_TYPE_NVLIST:
/* FIXME */
n = snprintf(p, buflen, "%s", "_NOT_IMPLEMENTED_");
break;
case DATA_TYPE_STRING:
(void) nvpair_value_string(nvp, &str);
n = snprintf(p, buflen, "%s", (str ? str : "<NULL>"));
break;
case DATA_TYPE_BOOLEAN_ARRAY:
/* FIXME */
n = snprintf(p, buflen, "%s", "_NOT_IMPLEMENTED_");
break;
case DATA_TYPE_BYTE_ARRAY:
/* FIXME */
n = snprintf(p, buflen, "%s", "_NOT_IMPLEMENTED_");
break;
case DATA_TYPE_INT8_ARRAY:
n = _zed_event_convert_int8_array(p, buflen, nvp);
break;
case DATA_TYPE_UINT8_ARRAY:
n = _zed_event_convert_uint8_array(p, buflen, nvp);
break;
case DATA_TYPE_INT16_ARRAY:
n = _zed_event_convert_int16_array(p, buflen, nvp);
break;
case DATA_TYPE_UINT16_ARRAY:
n = _zed_event_convert_uint16_array(p, buflen, nvp);
break;
case DATA_TYPE_INT32_ARRAY:
n = _zed_event_convert_int32_array(p, buflen, nvp);
break;
case DATA_TYPE_UINT32_ARRAY:
n = _zed_event_convert_uint32_array(p, buflen, nvp);
break;
case DATA_TYPE_INT64_ARRAY:
n = _zed_event_convert_int64_array(p, buflen, nvp);
break;
case DATA_TYPE_UINT64_ARRAY:
fmt = _zed_event_value_is_hex(name) ? "0x%.16llX " : "%llu ";
n = _zed_event_convert_uint64_array(p, buflen, nvp, fmt);
break;
case DATA_TYPE_STRING_ARRAY:
n = _zed_event_convert_string_array(p, buflen, nvp);
break;
case DATA_TYPE_NVLIST_ARRAY:
/* FIXME */
n = snprintf(p, buflen, "%s", "_NOT_IMPLEMENTED_");
break;
default:
zed_log_msg(LOG_WARNING,
"Failed to convert nvpair \"%s\" for eid=%llu: "
"Unrecognized type=%u", name, eid, (unsigned int) type);
return;
}
if ((n < 0) || (n >= sizeof (buf))) {
zed_log_msg(LOG_WARNING,
"Failed to convert nvpair \"%s\" for eid=%llu: %s",
name, eid, "Exceeded buffer size");
return;
}
if (zed_strings_add(zsp, buf) < 0) {
zed_log_msg(LOG_WARNING,
"Failed to convert nvpair \"%s\" for eid=%llu: %s",
name, eid, strerror(ENOMEM));
return;
}
}
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);
}
/*
* Seek to the event specified by [saved_eid] and [saved_etime].
* This protects against processing a given event more than once.
* Return 0 upon a successful seek to the specified event, or -1 otherwise.
*
* A zevent is considered to be uniquely specified by its (eid,time) tuple.
* The unsigned 64b eid is set to 1 when the kernel module is loaded, and
* incremented by 1 for each new event. Since the state file can persist
* across a kernel module reload, the time must be checked to ensure a match.
*/
int
zed_event_seek(struct zed_conf *zcp, uint64_t saved_eid, int64_t saved_etime[])
{
uint64_t eid;
int found;
nvlist_t *nvl;
int n_dropped;
int64_t *etime;
uint_t nelem;
int rv;
if (!zcp) {
errno = EINVAL;
zed_log_msg(LOG_ERR, "Failed to seek zevent: %s",
strerror(errno));
return (-1);
}
eid = 0;
found = 0;
while ((eid < saved_eid) && !found) {
rv = zpool_events_next(zcp->zfs_hdl, &nvl, &n_dropped,
ZEVENT_NONBLOCK, zcp->zevent_fd);
if ((rv != 0) || !nvl)
break;
if (n_dropped > 0) {
zed_log_msg(LOG_WARNING, "Missed %d events", n_dropped);
/*
* FIXME: Increase max size of event nvlist in
* /sys/module/zfs/parameters/zfs_zevent_len_max ?
*/
}
if (nvlist_lookup_uint64(nvl, "eid", &eid) != 0) {
zed_log_msg(LOG_WARNING, "Failed to lookup zevent eid");
} else if (nvlist_lookup_int64_array(nvl, "time",
&etime, &nelem) != 0) {
zed_log_msg(LOG_WARNING,
"Failed to lookup zevent time (eid=%llu)", eid);
} else if (nelem != 2) {
zed_log_msg(LOG_WARNING,
"Failed to lookup zevent time (eid=%llu, nelem=%u)",
eid, nelem);
} else if ((eid != saved_eid) ||
(etime[0] != saved_etime[0]) ||
(etime[1] != saved_etime[1])) {
/* no-op */
} else {
found = 1;
}
free(nvl);
}
if (!found && (saved_eid > 0)) {
if (zpool_events_seek(zcp->zfs_hdl, ZEVENT_SEEK_START,
zcp->zevent_fd) < 0)
zed_log_msg(LOG_WARNING, "Failed to seek to eid=0");
else
eid = 0;
}
zed_log_msg(LOG_NOTICE, "Processing events since eid=%llu", eid);
return (found ? 0 : -1);
}
/*
* 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);
}
/*
* Scan the [zcp] zedlet_dir for files to exec based on the event class.
* Files must be executable by user, but not writable by group or other.
* Dotfiles are ignored.
*
* Return 0 on success with an updated set of zedlets,
* or -1 on error with errno set.
*
* FIXME: Check if zedlet_dir and all parent dirs are secure.
*/
int
zed_conf_scan_dir(struct zed_conf *zcp)
{
zed_strings_t *zedlets;
DIR *dirp;
struct dirent *direntp;
char pathname[PATH_MAX];
struct stat st;
int n;
if (!zcp) {
errno = EINVAL;
zed_log_msg(LOG_ERR, "Failed to scan zedlet dir: %s",
strerror(errno));
return (-1);
}
zedlets = zed_strings_create();
if (!zedlets) {
errno = ENOMEM;
zed_log_msg(LOG_WARNING, "Failed to scan dir \"%s\": %s",
zcp->zedlet_dir, strerror(errno));
return (-1);
}
dirp = opendir(zcp->zedlet_dir);
if (!dirp) {
int errno_bak = errno;
zed_log_msg(LOG_WARNING, "Failed to open dir \"%s\": %s",
zcp->zedlet_dir, strerror(errno));
zed_strings_destroy(zedlets);
errno = errno_bak;
return (-1);
}
while ((direntp = readdir(dirp))) {
if (direntp->d_name[0] == '.')
continue;
n = snprintf(pathname, sizeof (pathname),
"%s/%s", zcp->zedlet_dir, direntp->d_name);
if ((n < 0) || (n >= sizeof (pathname))) {
zed_log_msg(LOG_WARNING, "Failed to stat \"%s\": %s",
direntp->d_name, strerror(ENAMETOOLONG));
continue;
}
if (stat(pathname, &st) < 0) {
zed_log_msg(LOG_WARNING, "Failed to stat \"%s\": %s",
pathname, strerror(errno));
continue;
}
if (!S_ISREG(st.st_mode)) {
zed_log_msg(LOG_INFO,
"Ignoring \"%s\": not a regular file",
direntp->d_name);
continue;
}
if ((st.st_uid != 0) && !zcp->do_force) {
zed_log_msg(LOG_NOTICE,
"Ignoring \"%s\": not owned by root",
direntp->d_name);
continue;
}
if (!(st.st_mode & S_IXUSR)) {
zed_log_msg(LOG_INFO,
"Ignoring \"%s\": not executable by user",
direntp->d_name);
continue;
}
if ((st.st_mode & S_IWGRP) & !zcp->do_force) {
zed_log_msg(LOG_NOTICE,
"Ignoring \"%s\": writable by group",
direntp->d_name);
continue;
}
if ((st.st_mode & S_IWOTH) & !zcp->do_force) {
zed_log_msg(LOG_NOTICE,
"Ignoring \"%s\": writable by other",
direntp->d_name);
continue;
}
if (zed_strings_add(zedlets, direntp->d_name) < 0) {
zed_log_msg(LOG_WARNING,
"Failed to register \"%s\": %s",
direntp->d_name, strerror(errno));
continue;
}
if (zcp->do_verbose)
zed_log_msg(LOG_INFO,
"Registered zedlet \"%s\"", direntp->d_name);
}
if (closedir(dirp) < 0) {
int errno_bak = errno;
zed_log_msg(LOG_WARNING, "Failed to close dir \"%s\": %s",
zcp->zedlet_dir, strerror(errno));
zed_strings_destroy(zedlets);
errno = errno_bak;
return (-1);
}
if (zcp->zedlets)
zed_strings_destroy(zcp->zedlets);
zcp->zedlets = zedlets;
return (0);
}
/*
* Fork a child process to handle event [eid]. The program [prog]
* in directory [dir] is executed with the envionment [env].
*
* The file descriptor [zfd] is the zevent_fd used to track the
* current cursor location within the zevent nvlist.
*/
static void
_zed_exec_fork_child(uint64_t eid, const char *dir, const char *prog,
char *env[], int zfd)
{
char path[PATH_MAX];
int n;
pid_t pid;
int fd;
pid_t wpid;
int status;
assert(dir != NULL);
assert(prog != NULL);
assert(env != NULL);
assert(zfd >= 0);
n = snprintf(path, sizeof (path), "%s/%s", dir, prog);
if ((n < 0) || (n >= sizeof (path))) {
zed_log_msg(LOG_WARNING,
"Failed to fork \"%s\" for eid=%llu: %s",
prog, eid, strerror(ENAMETOOLONG));
return;
}
pid = fork();
if (pid < 0) {
zed_log_msg(LOG_WARNING,
"Failed to fork \"%s\" for eid=%llu: %s",
prog, eid, strerror(errno));
return;
} else if (pid == 0) {
(void) umask(022);
if ((fd = open("/dev/null", O_RDWR)) != -1) {
(void) dup2(fd, STDIN_FILENO);
(void) dup2(fd, STDOUT_FILENO);
(void) dup2(fd, STDERR_FILENO);
}
(void) dup2(zfd, ZEVENT_FILENO);
zed_file_close_from(ZEVENT_FILENO + 1);
execle(path, prog, NULL, env);
_exit(127);
} else {
zed_log_msg(LOG_INFO, "Invoking \"%s\" eid=%llu pid=%d",
prog, eid, pid);
/* FIXME: Timeout rogue child processes with sigalarm? */
restart:
wpid = waitpid(pid, &status, 0);
if (wpid == (pid_t) -1) {
if (errno == EINTR)
goto restart;
zed_log_msg(LOG_WARNING,
"Failed to wait for \"%s\" eid=%llu pid=%d",
prog, eid, pid);
} else if (WIFEXITED(status)) {
zed_log_msg(LOG_INFO,
"Finished \"%s\" eid=%llu pid=%d exit=%d",
prog, eid, pid, WEXITSTATUS(status));
} else if (WIFSIGNALED(status)) {
zed_log_msg(LOG_INFO,
"Finished \"%s\" eid=%llu pid=%d sig=%d/%s",
prog, eid, pid, WTERMSIG(status),
strsignal(WTERMSIG(status)));
} else {
zed_log_msg(LOG_INFO,
"Finished \"%s\" eid=%llu pid=%d status=0x%X",
prog, eid, (unsigned int) status);
}
}
}