4 * The contents of this file are subject to the terms of the
5 * Common Development and Distribution License (the "License").
6 * You may not use this file except in compliance with the License.
8 * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
9 * or http://www.opensolaris.org/os/licensing.
10 * See the License for the specific language governing permissions
11 * and limitations under the License.
13 * When distributing Covered Code, include this CDDL HEADER in each
14 * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
15 * If applicable, add the following below this CDDL HEADER, with the
16 * fields enclosed by brackets "[]" replaced with your own identifying
17 * information: Portions Copyright [yyyy] [name of copyright owner]
22 * Copyright (c) 2007, 2010, Oracle and/or its affiliates. All rights reserved.
23 * Copyright (c) 2012 by Delphix. All rights reserved.
24 * Copyright 2014 Nexenta Systems, Inc. All rights reserved.
25 * Copyright (c) 2016, Intel Corporation.
29 * ZFS syseventd module.
31 * file origin: openzfs/usr/src/cmd/syseventd/modules/zfs_mod/zfs_mod.c
33 * The purpose of this module is to identify when devices are added to the
34 * system, and appropriately online or replace the affected vdevs.
36 * When a device is added to the system:
38 * 1. Search for any vdevs whose devid matches that of the newly added
41 * 2. If no vdevs are found, then search for any vdevs whose udev path
42 * matches that of the new device.
44 * 3. If no vdevs match by either method, then ignore the event.
46 * 4. Attempt to online the device with a flag to indicate that it should
47 * be unspared when resilvering completes. If this succeeds, then the
48 * same device was inserted and we should continue normally.
50 * 5. If the pool does not have the 'autoreplace' property set, attempt to
51 * online the device again without the unspare flag, which will
52 * generate a FMA fault.
54 * 6. If the pool has the 'autoreplace' property set, and the matching vdev
55 * is a whole disk, then label the new disk and attempt a 'zpool
58 * The module responds to EC_DEV_ADD events. The special ESC_ZFS_VDEV_CHECK
59 * event indicates that a device failed to open during pool load, but the
60 * autoreplace property was set. In this case, we deferred the associated
61 * FMA fault until our module had a chance to process the autoreplace logic.
62 * If the device could not be replaced, then the second online attempt will
63 * trigger the FMA fault that we skipped earlier.
65 * ZFS on Linux porting notes:
66 * In lieu of a thread pool, just spawn a thread on demmand.
67 * Linux udev provides a disk insert for both the disk and the partition
74 #include <libnvpair.h>
82 #include <sys/sunddi.h>
83 #include <sys/sysevent/eventdefs.h>
84 #include <sys/sysevent/dev.h>
87 #include "zfs_agents.h"
88 #include "../zed_log.h"
90 #define DEV_BYID_PATH "/dev/disk/by-id/"
91 #define DEV_BYPATH_PATH "/dev/disk/by-path/"
93 typedef void (*zfs_process_func_t)(zpool_handle_t *, nvlist_t *, boolean_t);
95 libzfs_handle_t *g_zfshdl;
96 list_t g_pool_list; /* list of unavailable pools at initialization */
97 list_t g_device_list; /* list of disks with asynchronous label request */
98 boolean_t g_enumeration_done;
101 typedef struct unavailpool {
102 zpool_handle_t *uap_zhp;
103 pthread_t uap_enable_tid; /* dataset enable thread if activated */
104 list_node_t uap_node;
107 typedef struct pendingdev {
108 char pd_physpath[128];
113 zfs_toplevel_state(zpool_handle_t *zhp)
119 verify(nvlist_lookup_nvlist(zpool_get_config(zhp, NULL),
120 ZPOOL_CONFIG_VDEV_TREE, &nvroot) == 0);
121 verify(nvlist_lookup_uint64_array(nvroot, ZPOOL_CONFIG_VDEV_STATS,
122 (uint64_t **)&vs, &c) == 0);
123 return (vs->vs_state);
127 zfs_unavail_pool(zpool_handle_t *zhp, void *data)
129 zed_log_msg(LOG_INFO, "zfs_unavail_pool: examining '%s' (state %d)",
130 zpool_get_name(zhp), (int)zfs_toplevel_state(zhp));
132 if (zfs_toplevel_state(zhp) < VDEV_STATE_DEGRADED) {
134 uap = malloc(sizeof (unavailpool_t));
136 uap->uap_enable_tid = 0;
137 list_insert_tail((list_t *)data, uap);
145 * Two stage replace on Linux
146 * since we get disk notifications
147 * we can wait for partitioned disk slice to show up!
149 * First stage tags the disk, initiates async partitioning, and returns
150 * Second stage finds the tag and proceeds to ZFS labeling/replace
152 * disk-add --> label-disk + tag-disk --> partition-add --> zpool_vdev_attach
154 * 1. physical match with no fs, no partition
155 * tag it top, partition disk
157 * 2. physical match again, see partion and tag
162 * The device associated with the given vdev (either by devid or physical path)
163 * has been added to the system. If 'isdisk' is set, then we only attempt a
164 * replacement if it's a whole disk. This also implies that we should label the
167 * First, we attempt to online the device (making sure to undo any spare
168 * operation when finished). If this succeeds, then we're done. If it fails,
169 * and the new state is VDEV_CANT_OPEN, it indicates that the device was opened,
170 * but that the label was not what we expected. If the 'autoreplace' property
171 * is not set, then we relabel the disk (if specified), and attempt a 'zpool
172 * replace'. If the online is successful, but the new state is something else
173 * (REMOVED or FAULTED), it indicates that we're out of sync or in some sort of
174 * race, and we should avoid attempting to relabel the disk.
176 * Also can arrive here from a ESC_ZFS_VDEV_CHECK event
179 zfs_process_add(zpool_handle_t *zhp, nvlist_t *vdev, boolean_t labeled)
182 vdev_state_t newstate;
183 nvlist_t *nvroot, *newvd;
184 pendingdev_t *device;
185 uint64_t wholedisk = 0ULL;
186 uint64_t offline = 0ULL;
187 uint64_t guid = 0ULL;
188 char *physpath = NULL, *new_devid = NULL, *enc_sysfs_path = NULL;
189 char rawpath[PATH_MAX], fullpath[PATH_MAX];
190 char devpath[PATH_MAX];
196 if (nvlist_lookup_string(vdev, ZPOOL_CONFIG_PATH, &path) != 0)
199 /* Skip healthy disks */
200 verify(nvlist_lookup_uint64_array(vdev, ZPOOL_CONFIG_VDEV_STATS,
201 (uint64_t **)&vs, &c) == 0);
202 if (vs->vs_state == VDEV_STATE_HEALTHY) {
203 zed_log_msg(LOG_INFO, "%s: %s is already healthy, skip it.",
208 (void) nvlist_lookup_string(vdev, ZPOOL_CONFIG_PHYS_PATH, &physpath);
209 (void) nvlist_lookup_string(vdev, ZPOOL_CONFIG_VDEV_ENC_SYSFS_PATH,
211 (void) nvlist_lookup_uint64(vdev, ZPOOL_CONFIG_WHOLE_DISK, &wholedisk);
212 (void) nvlist_lookup_uint64(vdev, ZPOOL_CONFIG_OFFLINE, &offline);
213 (void) nvlist_lookup_uint64(vdev, ZPOOL_CONFIG_GUID, &guid);
216 return; /* don't intervene if it was taken offline */
218 #ifdef HAVE_LIBDEVMAPPER
219 is_dm = zfs_dev_is_dm(path);
221 zed_log_msg(LOG_INFO, "zfs_process_add: pool '%s' vdev '%s', phys '%s'"
222 " wholedisk %d, dm %d (%llu)", zpool_get_name(zhp), path,
223 physpath ? physpath : "NULL", wholedisk, is_dm,
224 (long long unsigned int)guid);
227 * The VDEV guid is preferred for identification (gets passed in path)
230 (void) snprintf(fullpath, sizeof (fullpath), "%llu",
231 (long long unsigned int)guid);
234 * otherwise use path sans partition suffix for whole disks
236 (void) strlcpy(fullpath, path, sizeof (fullpath));
238 char *spath = zfs_strip_partition(fullpath);
240 zed_log_msg(LOG_INFO, "%s: Can't alloc",
245 (void) strlcpy(fullpath, spath, sizeof (fullpath));
251 * Attempt to online the device.
253 if (zpool_vdev_online(zhp, fullpath,
254 ZFS_ONLINE_CHECKREMOVE | ZFS_ONLINE_UNSPARE, &newstate) == 0 &&
255 (newstate == VDEV_STATE_HEALTHY ||
256 newstate == VDEV_STATE_DEGRADED)) {
257 zed_log_msg(LOG_INFO, " zpool_vdev_online: vdev %s is %s",
258 fullpath, (newstate == VDEV_STATE_HEALTHY) ?
259 "HEALTHY" : "DEGRADED");
264 * If the pool doesn't have the autoreplace property set, then attempt
265 * a true online (without the unspare flag), which will trigger a FMA
268 if (!is_dm && (!zpool_get_prop_int(zhp, ZPOOL_PROP_AUTOREPLACE, NULL) ||
269 !wholedisk || physpath == NULL)) {
270 (void) zpool_vdev_online(zhp, fullpath, ZFS_ONLINE_FORCEFAULT,
272 zed_log_msg(LOG_INFO, " zpool_vdev_online: %s FORCEFAULT (%s)",
273 fullpath, libzfs_error_description(g_zfshdl));
278 * convert physical path into its current device node
280 (void) snprintf(rawpath, sizeof (rawpath), "%s%s", DEV_BYPATH_PATH,
282 if (realpath(rawpath, devpath) == NULL && !is_dm) {
283 zed_log_msg(LOG_INFO, " realpath: %s failed (%s)",
284 rawpath, strerror(errno));
286 (void) zpool_vdev_online(zhp, fullpath, ZFS_ONLINE_FORCEFAULT,
289 zed_log_msg(LOG_INFO, " zpool_vdev_online: %s FORCEFAULT (%s)",
290 fullpath, libzfs_error_description(g_zfshdl));
294 if (!zpool_get_prop_int(zhp, ZPOOL_PROP_AUTOREPLACE, NULL)) {
295 zed_log_msg(LOG_INFO, "%s: Autoreplace is not enabled on this"
296 " pool, ignore disk.", __func__);
300 /* Only autoreplace bad disks */
301 if ((vs->vs_state != VDEV_STATE_DEGRADED) &&
302 (vs->vs_state != VDEV_STATE_FAULTED) &&
303 (vs->vs_state != VDEV_STATE_CANT_OPEN)) {
307 nvlist_lookup_string(vdev, "new_devid", &new_devid);
310 /* Don't label device mapper or multipath disks. */
311 } else if (!labeled) {
313 * we're auto-replacing a raw disk, so label it first
318 * If this is a request to label a whole disk, then attempt to
319 * write out the label. Before we can label the disk, we need
320 * to map the physical string that was matched on to the under
323 * If any part of this process fails, then do a force online
324 * to trigger a ZFS fault for the device (and any hot spare
327 leafname = strrchr(devpath, '/') + 1;
330 * If this is a request to label a whole disk, then attempt to
331 * write out the label.
333 if (zpool_label_disk(g_zfshdl, zhp, leafname) != 0) {
334 zed_log_msg(LOG_INFO, " zpool_label_disk: could not "
335 "label '%s' (%s)", leafname,
336 libzfs_error_description(g_zfshdl));
338 (void) zpool_vdev_online(zhp, fullpath,
339 ZFS_ONLINE_FORCEFAULT, &newstate);
344 * The disk labeling is asynchronous on Linux. Just record
345 * this label request and return as there will be another
346 * disk add event for the partition after the labeling is
349 device = malloc(sizeof (pendingdev_t));
350 (void) strlcpy(device->pd_physpath, physpath,
351 sizeof (device->pd_physpath));
352 list_insert_tail(&g_device_list, device);
354 zed_log_msg(LOG_INFO, " zpool_label_disk: async '%s' (%llu)",
355 leafname, (u_longlong_t) guid);
357 return; /* resumes at EC_DEV_ADD.ESC_DISK for partition */
359 } else /* labeled */ {
360 boolean_t found = B_FALSE;
362 * match up with request above to label the disk
364 for (device = list_head(&g_device_list); device != NULL;
365 device = list_next(&g_device_list, device)) {
366 if (strcmp(physpath, device->pd_physpath) == 0) {
367 list_remove(&g_device_list, device);
374 /* unexpected partition slice encountered */
375 (void) zpool_vdev_online(zhp, fullpath,
376 ZFS_ONLINE_FORCEFAULT, &newstate);
380 zed_log_msg(LOG_INFO, " zpool_label_disk: resume '%s' (%llu)",
381 physpath, (u_longlong_t) guid);
383 (void) snprintf(devpath, sizeof (devpath), "%s%s",
384 DEV_BYID_PATH, new_devid);
388 * Construct the root vdev to pass to zpool_vdev_attach(). While adding
389 * the entire vdev structure is harmless, we construct a reduced set of
390 * path/physpath/wholedisk to keep it simple.
392 if (nvlist_alloc(&nvroot, NV_UNIQUE_NAME, 0) != 0) {
393 zed_log_msg(LOG_WARNING, "zfs_mod: nvlist_alloc out of memory");
396 if (nvlist_alloc(&newvd, NV_UNIQUE_NAME, 0) != 0) {
397 zed_log_msg(LOG_WARNING, "zfs_mod: nvlist_alloc out of memory");
402 if (nvlist_add_string(newvd, ZPOOL_CONFIG_TYPE, VDEV_TYPE_DISK) != 0 ||
403 nvlist_add_string(newvd, ZPOOL_CONFIG_PATH, path) != 0 ||
404 nvlist_add_string(newvd, ZPOOL_CONFIG_DEVID, new_devid) != 0 ||
405 (physpath != NULL && nvlist_add_string(newvd,
406 ZPOOL_CONFIG_PHYS_PATH, physpath) != 0) ||
407 nvlist_add_string(newvd, ZPOOL_CONFIG_VDEV_ENC_SYSFS_PATH,
408 enc_sysfs_path) != 0 ||
409 nvlist_add_uint64(newvd, ZPOOL_CONFIG_WHOLE_DISK, wholedisk) != 0 ||
410 nvlist_add_string(nvroot, ZPOOL_CONFIG_TYPE, VDEV_TYPE_ROOT) != 0 ||
411 nvlist_add_nvlist_array(nvroot, ZPOOL_CONFIG_CHILDREN, &newvd,
413 zed_log_msg(LOG_WARNING, "zfs_mod: unable to add nvlist pairs");
422 * auto replace a leaf disk at same physical location
424 ret = zpool_vdev_attach(zhp, fullpath, path, nvroot, B_TRUE);
426 zed_log_msg(LOG_INFO, " zpool_vdev_replace: %s with %s (%s)",
427 fullpath, path, (ret == 0) ? "no errors" :
428 libzfs_error_description(g_zfshdl));
434 * Utility functions to find a vdev matching given criteria.
436 typedef struct dev_data {
437 const char *dd_compare;
439 zfs_process_func_t dd_func;
441 boolean_t dd_islabeled;
442 uint64_t dd_pool_guid;
443 uint64_t dd_vdev_guid;
444 const char *dd_new_devid;
448 zfs_iter_vdev(zpool_handle_t *zhp, nvlist_t *nvl, void *data)
450 dev_data_t *dp = data;
456 * First iterate over any children.
458 if (nvlist_lookup_nvlist_array(nvl, ZPOOL_CONFIG_CHILDREN,
459 &child, &children) == 0) {
460 for (c = 0; c < children; c++)
461 zfs_iter_vdev(zhp, child[c], data);
465 /* once a vdev was matched and processed there is nothing left to do */
470 * Match by GUID if available otherwise fallback to devid or physical
472 if (dp->dd_vdev_guid != 0) {
475 if (nvlist_lookup_uint64(nvl, ZPOOL_CONFIG_GUID,
476 &guid) != 0 || guid != dp->dd_vdev_guid) {
479 zed_log_msg(LOG_INFO, " zfs_iter_vdev: matched on %llu", guid);
480 dp->dd_found = B_TRUE;
482 } else if (dp->dd_compare != NULL) {
484 * NOTE: On Linux there is an event for partition, so unlike
485 * illumos, substring matching is not required to accomodate
486 * the partition suffix. An exact match will be present in
487 * the dp->dd_compare value.
489 if (nvlist_lookup_string(nvl, dp->dd_prop, &path) != 0 ||
490 strcmp(dp->dd_compare, path) != 0)
493 zed_log_msg(LOG_INFO, " zfs_iter_vdev: matched %s on %s",
495 dp->dd_found = B_TRUE;
497 /* pass the new devid for use by replacing code */
498 if (dp->dd_new_devid != NULL) {
499 (void) nvlist_add_string(nvl, "new_devid",
504 (dp->dd_func)(zhp, nvl, dp->dd_islabeled);
508 zfs_enable_ds(void *arg)
510 unavailpool_t *pool = (unavailpool_t *)arg;
512 assert(pool->uap_enable_tid = pthread_self());
514 (void) zpool_enable_datasets(pool->uap_zhp, NULL, 0);
515 zpool_close(pool->uap_zhp);
516 pool->uap_zhp = NULL;
518 /* Note: zfs_slm_fini() will cleanup this pool entry on exit */
523 zfs_iter_pool(zpool_handle_t *zhp, void *data)
525 nvlist_t *config, *nvl;
526 dev_data_t *dp = data;
530 zed_log_msg(LOG_INFO, "zfs_iter_pool: evaluating vdevs on %s (by %s)",
531 zpool_get_name(zhp), dp->dd_vdev_guid ? "GUID" : dp->dd_prop);
534 * For each vdev in this pool, look for a match to apply dd_func
536 if ((config = zpool_get_config(zhp, NULL)) != NULL) {
537 if (dp->dd_pool_guid == 0 ||
538 (nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_GUID,
539 &pool_guid) == 0 && pool_guid == dp->dd_pool_guid)) {
540 (void) nvlist_lookup_nvlist(config,
541 ZPOOL_CONFIG_VDEV_TREE, &nvl);
542 zfs_iter_vdev(zhp, nvl, data);
547 * if this pool was originally unavailable,
548 * then enable its datasets asynchronously
550 if (g_enumeration_done) {
551 for (pool = list_head(&g_pool_list); pool != NULL;
552 pool = list_next(&g_pool_list, pool)) {
554 if (pool->uap_enable_tid != 0)
555 continue; /* entry already processed */
556 if (strcmp(zpool_get_name(zhp),
557 zpool_get_name(pool->uap_zhp)))
559 if (zfs_toplevel_state(zhp) >= VDEV_STATE_DEGRADED) {
560 /* send to a background thread; keep on list */
561 (void) pthread_create(&pool->uap_enable_tid,
562 NULL, zfs_enable_ds, pool);
569 return (dp->dd_found); /* cease iteration after a match */
573 * Given a physical device location, iterate over all
574 * (pool, vdev) pairs which correspond to that location.
577 devphys_iter(const char *physical, const char *devid, zfs_process_func_t func,
580 dev_data_t data = { 0 };
582 data.dd_compare = physical;
584 data.dd_prop = ZPOOL_CONFIG_PHYS_PATH;
585 data.dd_found = B_FALSE;
586 data.dd_islabeled = is_slice;
587 data.dd_new_devid = devid; /* used by auto replace code */
589 (void) zpool_iter(g_zfshdl, zfs_iter_pool, &data);
591 return (data.dd_found);
595 * Given a device identifier, find any vdevs with a matching devid.
596 * On Linux we can match devid directly which is always a whole disk.
599 devid_iter(const char *devid, zfs_process_func_t func, boolean_t is_slice)
601 dev_data_t data = { 0 };
603 data.dd_compare = devid;
605 data.dd_prop = ZPOOL_CONFIG_DEVID;
606 data.dd_found = B_FALSE;
607 data.dd_islabeled = is_slice;
608 data.dd_new_devid = devid;
610 (void) zpool_iter(g_zfshdl, zfs_iter_pool, &data);
612 return (data.dd_found);
616 * Handle a EC_DEV_ADD.ESC_DISK event.
619 * Expects: DEV_PHYS_PATH string in schema
620 * Matches: vdev's ZPOOL_CONFIG_PHYS_PATH or ZPOOL_CONFIG_DEVID
622 * path: '/dev/dsk/c0t1d0s0' (persistent)
623 * devid: 'id1,sd@SATA_____Hitachi_HDS72101______JP2940HZ3H74MC/a'
624 * phys_path: '/pci@0,0/pci103c,1609@11/disk@1,0:a'
627 * provides: DEV_PHYS_PATH and DEV_IDENTIFIER strings in schema
628 * Matches: vdev's ZPOOL_CONFIG_PHYS_PATH or ZPOOL_CONFIG_DEVID
630 * path: '/dev/sdc1' (not persistent)
631 * devid: 'ata-SAMSUNG_HD204UI_S2HGJD2Z805891-part1'
632 * phys_path: 'pci-0000:04:00.0-sas-0x4433221106000000-lun-0'
635 zfs_deliver_add(nvlist_t *nvl, boolean_t is_lofi)
637 char *devpath = NULL, *devid;
641 * Expecting a devid string and an optional physical location
643 if (nvlist_lookup_string(nvl, DEV_IDENTIFIER, &devid) != 0)
646 (void) nvlist_lookup_string(nvl, DEV_PHYS_PATH, &devpath);
648 is_slice = (nvlist_lookup_boolean(nvl, DEV_IS_PART) == 0);
650 zed_log_msg(LOG_INFO, "zfs_deliver_add: adding %s (%s) (is_slice %d)",
651 devid, devpath ? devpath : "NULL", is_slice);
654 * Iterate over all vdevs looking for a match in the folllowing order:
655 * 1. ZPOOL_CONFIG_DEVID (identifies the unique disk)
656 * 2. ZPOOL_CONFIG_PHYS_PATH (identifies disk physical location).
658 * For disks, we only want to pay attention to vdevs marked as whole
659 * disks. For multipath devices does whole disk apply? (TBD).
661 if (!devid_iter(devid, zfs_process_add, is_slice) && devpath != NULL) {
663 (void) devphys_iter(devpath, devid, zfs_process_add,
672 * Called when we receive a VDEV_CHECK event, which indicates a device could not
673 * be opened during initial pool open, but the autoreplace property was set on
674 * the pool. In this case, we treat it as if it were an add event.
677 zfs_deliver_check(nvlist_t *nvl)
679 dev_data_t data = { 0 };
681 if (nvlist_lookup_uint64(nvl, ZFS_EV_POOL_GUID,
682 &data.dd_pool_guid) != 0 ||
683 nvlist_lookup_uint64(nvl, ZFS_EV_VDEV_GUID,
684 &data.dd_vdev_guid) != 0 ||
685 data.dd_vdev_guid == 0)
688 zed_log_msg(LOG_INFO, "zfs_deliver_check: pool '%llu', vdev %llu",
689 data.dd_pool_guid, data.dd_vdev_guid);
691 data.dd_func = zfs_process_add;
693 (void) zpool_iter(g_zfshdl, zfs_iter_pool, &data);
699 zfsdle_vdev_online(zpool_handle_t *zhp, void *data)
701 char *devname = data;
702 boolean_t avail_spare, l2cache;
703 vdev_state_t newstate;
706 zed_log_msg(LOG_INFO, "zfsdle_vdev_online: searching for '%s' in '%s'",
707 devname, zpool_get_name(zhp));
709 if ((tgt = zpool_find_vdev_by_physpath(zhp, devname,
710 &avail_spare, &l2cache, NULL)) != NULL) {
711 char *path, fullpath[MAXPATHLEN];
712 uint64_t wholedisk = 0ULL;
714 verify(nvlist_lookup_string(tgt, ZPOOL_CONFIG_PATH,
716 verify(nvlist_lookup_uint64(tgt, ZPOOL_CONFIG_WHOLE_DISK,
719 (void) strlcpy(fullpath, path, sizeof (fullpath));
721 char *spath = zfs_strip_partition(fullpath);
723 zed_log_msg(LOG_INFO, "%s: Can't alloc",
728 (void) strlcpy(fullpath, spath, sizeof (fullpath));
732 * We need to reopen the pool associated with this
733 * device so that the kernel can update the size
734 * of the expanded device.
736 (void) zpool_reopen(zhp);
739 if (zpool_get_prop_int(zhp, ZPOOL_PROP_AUTOEXPAND, NULL)) {
740 zed_log_msg(LOG_INFO, "zfsdle_vdev_online: setting "
741 "device '%s' to ONLINE state in pool '%s'",
742 fullpath, zpool_get_name(zhp));
743 if (zpool_get_state(zhp) != POOL_STATE_UNAVAIL)
744 (void) zpool_vdev_online(zhp, fullpath, 0,
755 * This function handles the ESC_DEV_DLE event.
758 zfs_deliver_dle(nvlist_t *nvl)
762 if (nvlist_lookup_string(nvl, DEV_PHYS_PATH, &devname) != 0) {
763 zed_log_msg(LOG_INFO, "zfs_deliver_event: no physpath");
767 if (zpool_iter(g_zfshdl, zfsdle_vdev_online, devname) != 1) {
768 zed_log_msg(LOG_INFO, "zfs_deliver_event: device '%s' not "
776 * syseventd daemon module event handler
778 * Handles syseventd daemon zfs device related events:
780 * EC_DEV_ADD.ESC_DISK
781 * EC_DEV_STATUS.ESC_DEV_DLE
782 * EC_ZFS.ESC_ZFS_VDEV_CHECK
784 * Note: assumes only one thread active at a time (not thread safe)
787 zfs_slm_deliver_event(const char *class, const char *subclass, nvlist_t *nvl)
790 boolean_t is_lofi = B_FALSE, is_check = B_FALSE, is_dle = B_FALSE;
792 if (strcmp(class, EC_DEV_ADD) == 0) {
794 * We're mainly interested in disk additions, but we also listen
795 * for new loop devices, to allow for simplified testing.
797 if (strcmp(subclass, ESC_DISK) == 0)
799 else if (strcmp(subclass, ESC_LOFI) == 0)
805 } else if (strcmp(class, EC_ZFS) == 0 &&
806 strcmp(subclass, ESC_ZFS_VDEV_CHECK) == 0) {
808 * This event signifies that a device failed to open
809 * during pool load, but the 'autoreplace' property was
810 * set, so we should pretend it's just been added.
813 } else if (strcmp(class, EC_DEV_STATUS) == 0 &&
814 strcmp(subclass, ESC_DEV_DLE) == 0) {
821 ret = zfs_deliver_dle(nvl);
823 ret = zfs_deliver_check(nvl);
825 ret = zfs_deliver_add(nvl, is_lofi);
832 zfs_enum_pools(void *arg)
834 (void) zpool_iter(g_zfshdl, zfs_unavail_pool, (void *)&g_pool_list);
836 * Linux - instead of using a thread pool, each list entry
837 * will spawn a thread when an unavailable pool transitions
838 * to available. zfs_slm_fini will wait for these threads.
840 g_enumeration_done = B_TRUE;
845 * called from zed daemon at startup
847 * sent messages from zevents or udev monitor
849 * For now, each agent has it's own libzfs instance
852 zfs_slm_init(libzfs_handle_t *zfs_hdl)
854 if ((g_zfshdl = libzfs_init()) == NULL)
858 * collect a list of unavailable pools (asynchronously,
859 * since this can take a while)
861 list_create(&g_pool_list, sizeof (struct unavailpool),
862 offsetof(struct unavailpool, uap_node));
864 if (pthread_create(&g_zfs_tid, NULL, zfs_enum_pools, NULL) != 0) {
865 list_destroy(&g_pool_list);
869 list_create(&g_device_list, sizeof (struct pendingdev),
870 offsetof(struct pendingdev, pd_node));
879 pendingdev_t *device;
881 /* wait for zfs_enum_pools thread to complete */
882 (void) pthread_join(g_zfs_tid, NULL);
884 while ((pool = (list_head(&g_pool_list))) != NULL) {
886 * each pool entry has two possibilities
887 * 1. was made available (so wait for zfs_enable_ds thread)
888 * 2. still unavailable (just close the pool)
890 if (pool->uap_enable_tid)
891 (void) pthread_join(pool->uap_enable_tid, NULL);
892 else if (pool->uap_zhp != NULL)
893 zpool_close(pool->uap_zhp);
895 list_remove(&g_pool_list, pool);
898 list_destroy(&g_pool_list);
900 while ((device = (list_head(&g_device_list))) != NULL) {
901 list_remove(&g_device_list, device);
904 list_destroy(&g_device_list);
906 libzfs_fini(g_zfshdl);
910 zfs_slm_event(const char *class, const char *subclass, nvlist_t *nvl)
912 static pthread_mutex_t serialize = PTHREAD_MUTEX_INITIALIZER;
915 * Serialize incoming events from zfs or libudev sources
917 (void) pthread_mutex_lock(&serialize);
918 zed_log_msg(LOG_INFO, "zfs_slm_event: %s.%s", class, subclass);
919 (void) zfs_slm_deliver_event(class, subclass, nvl);
920 (void) pthread_mutex_unlock(&serialize);