14469 nvme could raise dynamic lun expansion sysevents

Reviewed by: Robert Mustacchi <rm+illumos@fingolfin.org>
Reviewed by: Jason King <jason.brian.king+illumos@gmail.com>
Approved by: Dan McDonald <danmcd@joyent.com>

3 files changed, 176 insertions, 25 deletions

diff --git a/usr/src/cmd/syseventd/modules/zfs_mod/zfs_mod.c b/usr/src/cmd/syseventd/modules/zfs_mod/zfs_mod.c
index 4697128c90..82c296a669 100644
--- a/usr/src/cmd/syseventd/modules/zfs_mod/zfs_mod.c
+++ b/usr/src/cmd/syseventd/modules/zfs_mod/zfs_mod.c
@@ -22,6 +22,7 @@
  * Copyright (c) 2007, 2010, Oracle and/or its affiliates. All rights reserved.
  * Copyright (c) 2012 by Delphix. All rights reserved.
  * Copyright 2016 Nexenta Systems, Inc. All rights reserved.
+ * Copyright 2022 OmniOS Community Edition (OmniOSce) Association.
  */
 
 /*
@@ -32,15 +33,15 @@
  *
  * When a device is added to the system:
  *
- * 	1. Search for any vdevs whose devid matches that of the newly added
+ *	1. Search for any vdevs whose devid matches that of the newly added
  *	   device.
  *
- * 	2. If no vdevs are found, then search for any vdevs whose devfs path
+ *	2. If no vdevs are found, then search for any vdevs whose devfs path
  *	   matches that of the new device.
  *
  *	3. If no vdevs match by either method, then ignore the event.
  *
- * 	4. Attempt to online the device with a flag to indicate that it should
+ *	4. Attempt to online the device with a flag to indicate that it should
  *	   be unspared when resilvering completes.  If this succeeds, then the
  *	   same device was inserted and we should continue normally.
  *
@@ -319,11 +320,11 @@ zfs_iter_vdev(zpool_handle_t *zhp, nvlist_t *nvl, void *data)
 		 * string.  However, we allow substring matches in the following
 		 * cases:
 		 *
-		 * 	<path>:		This is a devpath, and the target is one
-		 * 			of its children.
+		 *	<path>:		This is a devpath, and the target is one
+		 *			of its children.
 		 *
-		 * 	<path/>		This is a devid for a whole disk, and
-		 * 			the target is one of its children.
+		 *	<path/>		This is a devid for a whole disk, and
+		 *			the target is one of its children.
 		 */
 		if (path[len] != '\0' && path[len] != ':' &&
 		    path[len - 1] != '/')
@@ -555,7 +556,7 @@ zfsdle_vdev_online(zpool_handle_t *zhp, void *data)
 	vdev_state_t newstate;
 	nvlist_t *tgt;
 
-	syseventd_print(9, "zfsdle_vdev_online: searching for %s in pool %s\n",
+	syseventd_print(9, "%s: searching for %s in pool %s\n", __func__,
 	    devname, zpool_get_name(zhp));
 
 	if ((tgt = zpool_find_vdev_by_physpath(zhp, devname,
@@ -568,6 +569,11 @@ zfsdle_vdev_online(zpool_handle_t *zhp, void *data)
 		verify(nvlist_lookup_uint64(tgt, ZPOOL_CONFIG_WHOLE_DISK,
 		    &wholedisk) == 0);
 
+		syseventd_print(9, "%s: "
+		    "found %s in pool %s (wholedisk: %s)\n", __func__,
+		    path, zpool_get_name(zhp),
+		    wholedisk != 0 ? "true" : "false");
+
 		(void) strlcpy(fullpath, path, sizeof (fullpath));
 		if (wholedisk) {
 			fullpath[strlen(fullpath) - 2] = '\0';
@@ -581,12 +587,13 @@ zfsdle_vdev_online(zpool_handle_t *zhp, void *data)
 		}
 
 		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)
+			syseventd_print(9, "%s: "
+			    "setting device %s to ONLINE state in pool %s.\n",
+			    __func__, 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);
diff --git a/usr/src/uts/common/io/blkdev/blkdev.c b/usr/src/uts/common/io/blkdev/blkdev.c
index c0bdb3dab2..611666b0a1 100644
--- a/usr/src/uts/common/io/blkdev/blkdev.c
+++ b/usr/src/uts/common/io/blkdev/blkdev.c
@@ -26,6 +26,7 @@
  * Copyright 2017 The MathWorks, Inc.  All rights reserved.
  * Copyright 2019 Western Digital Corporation.
  * Copyright 2020 Joyent, Inc.
+ * Copyright 2022 OmniOS Community Edition (OmniOSce) Association.
  */
 
 #include <sys/types.h>
@@ -55,6 +56,11 @@
 #include <sys/note.h>
 #include <sys/blkdev.h>
 #include <sys/scsi/impl/inquiry.h>
+#include <sys/taskq.h>
+#include <sys/taskq_impl.h>
+#include <sys/disp.h>
+#include <sys/sysevent/eventdefs.h>
+#include <sys/sysevent/dev.h>
 
 /*
  * blkdev is a driver which provides a lot of the common functionality
@@ -122,8 +128,8 @@
  *
  * Locks
  * -----
- * There are 4 instance global locks d_ocmutex, d_ksmutex, d_errmutex and
- * d_statemutex. As well a q_iomutex per waitq/runq pair.
+ * There are 5 instance global locks d_ocmutex, d_ksmutex, d_errmutex,
+ * d_statemutex and d_dle_mutex. As well a q_iomutex per waitq/runq pair.
  *
  * Lock Hierarchy
  * --------------
@@ -139,11 +145,16 @@ typedef struct bd bd_t;
 typedef struct bd_xfer_impl bd_xfer_impl_t;
 typedef struct bd_queue bd_queue_t;
 
+typedef enum {
+	BD_DLE_PENDING	= 1 << 0,
+	BD_DLE_RUNNING	= 1 << 1
+} bd_dle_state_t;
+
 struct bd {
 	void		*d_private;
 	dev_info_t	*d_dip;
-	kmutex_t	d_ocmutex;
-	kmutex_t	d_ksmutex;
+	kmutex_t	d_ocmutex;	/* open/close */
+	kmutex_t	d_ksmutex;	/* kstat */
 	kmutex_t	d_errmutex;
 	kmutex_t	d_statemutex;
 	kcondvar_t	d_statecv;
@@ -183,6 +194,10 @@ struct bd {
 	ddi_dma_attr_t	d_dma;
 	bd_ops_t	d_ops;
 	bd_handle_t	d_handle;
+
+	kmutex_t	d_dle_mutex;
+	taskq_ent_t	d_dle_ent;
+	bd_dle_state_t	d_dle_state;
 };
 
 struct bd_handle {
@@ -328,20 +343,34 @@ static struct modlinkage modlinkage = {
 
 static void *bd_state;
 static krwlock_t bd_lock;
+static taskq_t *bd_taskq;
 
 int
 _init(void)
 {
-	int	rv;
+	char taskq_name[TASKQ_NAMELEN];
+	const char *name;
+	int rv;
 
 	rv = ddi_soft_state_init(&bd_state, sizeof (struct bd), 2);
-	if (rv != DDI_SUCCESS) {
+	if (rv != DDI_SUCCESS)
 		return (rv);
+
+	name = mod_modname(&modlinkage);
+	(void) snprintf(taskq_name, sizeof (taskq_name), "%s_taskq", name);
+	bd_taskq = taskq_create(taskq_name, 1, minclsyspri, 0, 0, 0);
+	if (bd_taskq == NULL) {
+		cmn_err(CE_WARN, "%s: unable to create %s", name, taskq_name);
+		ddi_soft_state_fini(&bd_state);
+		return (DDI_FAILURE);
 	}
+
 	rw_init(&bd_lock, NULL, RW_DRIVER, NULL);
+
 	rv = mod_install(&modlinkage);
 	if (rv != DDI_SUCCESS) {
 		rw_destroy(&bd_lock);
+		taskq_destroy(bd_taskq);
 		ddi_soft_state_fini(&bd_state);
 	}
 	return (rv);
@@ -355,6 +384,7 @@ _fini(void)
 	rv = mod_remove(&modlinkage);
 	if (rv == DDI_SUCCESS) {
 		rw_destroy(&bd_lock);
+		taskq_destroy(bd_taskq);
 		ddi_soft_state_fini(&bd_state);
 	}
 	return (rv);
@@ -696,6 +726,8 @@ bd_attach(dev_info_t *dip, ddi_attach_cmd_t cmd)
 	mutex_init(&bd->d_ocmutex, NULL, MUTEX_DRIVER, NULL);
 	mutex_init(&bd->d_statemutex, NULL, MUTEX_DRIVER, NULL);
 	cv_init(&bd->d_statecv, NULL, CV_DRIVER, NULL);
+	mutex_init(&bd->d_dle_mutex, NULL, MUTEX_DRIVER, NULL);
+	bd->d_dle_state = 0;
 
 	bd->d_cache = kmem_cache_create(kcache, sizeof (bd_xfer_impl_t), 8,
 	    bd_xfer_ctor, bd_xfer_dtor, NULL, bd, NULL, 0);
@@ -853,6 +885,7 @@ fail_drive_info:
 	mutex_destroy(&bd->d_statemutex);
 	mutex_destroy(&bd->d_ocmutex);
 	mutex_destroy(&bd->d_ksmutex);
+	mutex_destroy(&bd->d_dle_mutex);
 	ddi_soft_state_free(bd_state, inst);
 	return (DDI_FAILURE);
 }
@@ -891,6 +924,7 @@ bd_detach(dev_info_t *dip, ddi_detach_cmd_t cmd)
 	mutex_destroy(&bd->d_ocmutex);
 	mutex_destroy(&bd->d_statemutex);
 	cv_destroy(&bd->d_statecv);
+	mutex_destroy(&bd->d_dle_mutex);
 	bd_queues_free(bd);
 	ddi_soft_state_free(bd_state, ddi_get_instance(dip));
 	return (DDI_SUCCESS);
@@ -1890,6 +1924,69 @@ bd_runq_exit(bd_xfer_impl_t *xi, int err)
 }
 
 static void
+bd_dle_sysevent_task(void *arg)
+{
+	nvlist_t *attr = NULL;
+	char *path = NULL;
+	bd_t *bd = arg;
+	dev_info_t *dip = bd->d_dip;
+	size_t n;
+
+	mutex_enter(&bd->d_dle_mutex);
+	bd->d_dle_state &= ~BD_DLE_PENDING;
+	bd->d_dle_state |= BD_DLE_RUNNING;
+	mutex_exit(&bd->d_dle_mutex);
+
+	dev_err(dip, CE_NOTE, "!dynamic LUN expansion");
+
+	if (nvlist_alloc(&attr, NV_UNIQUE_NAME_TYPE, KM_SLEEP) != 0) {
+		mutex_enter(&bd->d_dle_mutex);
+		bd->d_dle_state &= ~(BD_DLE_RUNNING|BD_DLE_PENDING);
+		mutex_exit(&bd->d_dle_mutex);
+		return;
+	}
+
+	path = kmem_zalloc(MAXPATHLEN, KM_SLEEP);
+
+	n = snprintf(path, MAXPATHLEN, "/devices");
+	(void) ddi_pathname(dip, path + n);
+	n = strlen(path);
+	n += snprintf(path + n, MAXPATHLEN - n, ":x");
+
+	for (;;) {
+		/*
+		 * On receipt of this event, the ZFS sysevent module will scan
+		 * active zpools for child vdevs matching this physical path.
+		 * In order to catch both whole disk pools and those with an
+		 * EFI boot partition, generate separate sysevents for minor
+		 * node 'a' and 'b'. (By comparison, io/scsi/targets/sd.c sends
+		 * events for just 'a')
+		 */
+		for (char c = 'a'; c < 'c'; c++) {
+			path[n - 1] = c;
+
+			if (nvlist_add_string(attr, DEV_PHYS_PATH, path) != 0)
+				break;
+
+			(void) ddi_log_sysevent(dip, DDI_VENDOR_SUNW,
+			    EC_DEV_STATUS, ESC_DEV_DLE, attr, NULL, DDI_SLEEP);
+		}
+
+		mutex_enter(&bd->d_dle_mutex);
+		if ((bd->d_dle_state & BD_DLE_PENDING) == 0) {
+			bd->d_dle_state &= ~BD_DLE_RUNNING;
+			mutex_exit(&bd->d_dle_mutex);
+			break;
+		}
+		bd->d_dle_state &= ~BD_DLE_PENDING;
+		mutex_exit(&bd->d_dle_mutex);
+	}
+
+	nvlist_free(attr);
+	kmem_free(path, MAXPATHLEN);
+}
+
+static void
 bd_update_state(bd_t *bd)
 {
 	enum	dkio_state	state = DKIO_INSERTED;
@@ -1908,8 +2005,7 @@ bd_update_state(bd_t *bd)
 	if ((media.m_blksize < 512) ||
 	    (!ISP2(media.m_blksize)) ||
 	    (P2PHASE(bd->d_maxxfer, media.m_blksize))) {
-		cmn_err(CE_WARN, "%s%d: Invalid media block size (%d)",
-		    ddi_driver_name(bd->d_dip), ddi_get_instance(bd->d_dip),
+		dev_err(bd->d_dip, CE_WARN, "Invalid media block size (%d)",
 		    media.m_blksize);
 		/*
 		 * We can't use the media, treat it as not present.
@@ -1954,6 +2050,21 @@ done:
 	if (docmlb) {
 		if (state == DKIO_INSERTED) {
 			(void) cmlb_validate(bd->d_cmlbh, 0, 0);
+
+			mutex_enter(&bd->d_dle_mutex);
+			/*
+			 * If there is already an event pending, there's
+			 * nothing to do; we coalesce multiple events.
+			 */
+			if ((bd->d_dle_state & BD_DLE_PENDING) == 0) {
+				if ((bd->d_dle_state & BD_DLE_RUNNING) == 0) {
+					taskq_dispatch_ent(bd_taskq,
+					    bd_dle_sysevent_task, bd, 0,
+					    &bd->d_dle_ent);
+				}
+				bd->d_dle_state |= BD_DLE_PENDING;
+			}
+			mutex_exit(&bd->d_dle_mutex);
 		} else {
 			cmlb_invalidate(bd->d_cmlbh, 0);
 		}
diff --git a/usr/src/uts/common/io/nvme/nvme.c b/usr/src/uts/common/io/nvme/nvme.c
index ad076201e0..e9c779e323 100644
--- a/usr/src/uts/common/io/nvme/nvme.c
+++ b/usr/src/uts/common/io/nvme/nvme.c
@@ -445,6 +445,8 @@ static int nvme_open(dev_t *, int, int, cred_t *);
 static int nvme_close(dev_t, int, int, cred_t *);
 static int nvme_ioctl(dev_t, int, intptr_t, int, cred_t *, int *);
 
+static void nvme_changed_ns(nvme_t *, int);
+
 static ddi_ufm_ops_t nvme_ufm_ops = {
 	NULL,
 	nvme_ufm_fill_image,
@@ -1955,11 +1957,7 @@ nvme_async_event_task(void *arg)
 
 				if (nsid == 0)	/* end of list */
 					break;
-
-				dev_err(nvme->n_dip, CE_CONT,
-				    "namespace %u (%s) has changed.\n",
-				    nsid, nvme->n_ns[nsid - 1].ns_name);
-				/* TODO: handle namespace resize. */
+				nvme_changed_ns(nvme, nsid);
 			}
 
 			break;
@@ -2693,6 +2691,41 @@ nvme_prepare_devid(nvme_t *nvme, uint32_t nsid)
 	    nvme->n_idctl->id_vid, model, serial, nsid);
 }
 
+static void
+nvme_changed_ns(nvme_t *nvme, int nsid)
+{
+	nvme_namespace_t *ns = &nvme->n_ns[nsid - 1];
+	nvme_identify_nsid_t *idns, *oidns;
+
+	dev_err(nvme->n_dip, CE_NOTE, "!namespace %u (%s) has changed.",
+	    nsid, ns->ns_name);
+
+	if (ns->ns_ignore)
+		return;
+
+	/*
+	 * The namespace has changed in some way. At present, we only update
+	 * the device capacity and trigger blkdev to check the device state.
+	 */
+
+	if (nvme_identify(nvme, B_FALSE, nsid, (void **)&idns) != 0) {
+		dev_err(nvme->n_dip, CE_WARN,
+		    "!failed to identify namespace %d", nsid);
+		return;
+	}
+
+	oidns = ns->ns_idns;
+	ns->ns_idns = idns;
+	kmem_free(oidns, sizeof (nvme_identify_nsid_t));
+
+	ns->ns_block_count = idns->id_nsize;
+	ns->ns_block_size =
+	    1 << idns->id_lbaf[idns->id_flbas.lba_format].lbaf_lbads;
+	ns->ns_best_block_size = ns->ns_block_size;
+
+	bd_state_change(ns->ns_bd_hdl);
+}
+
 static int
 nvme_init_ns(nvme_t *nvme, int nsid)
 {