From 670147be53c9200bb842b6109125ac007bbccc23 Mon Sep 17 00:00:00 2001 From: Rob Norris Date: Thu, 18 Jul 2024 13:24:05 +1000 Subject: [PATCH] zvol: ensure device minors are properly cleaned up Currently, if a minor is in use when we try to remove it, we'll skip it and never come back to it again. Since the zvol state is hung off the minor in the kernel, this can get us into weird situations if something tries to use it after the removal fails. It's even worse at pool export, as there's now a vestigial zvol state with no pool under it. It's weirder again if the pool is subsequently reimported, as the zvol code (reasonably) assumes the zvol state has been properly setup, when it's actually left over from the previous import of the pool. This commit attempts to tackle that by setting a flag on the zvol if its minor can't be removed, and then checking that flag when a request is made and rejecting it, thus stopping new work coming in. The flag also causes a condvar to be signaled when the last client finishes. For the case where a single minor is being removed (eg changing volmode), it will wait for this signal before proceeding. Meanwhile, when removing all minors, a background task is created for each minor that couldn't be removed on the spot, and those tasks then wake and clean up. Since any new tasks are queued on to the pool's spa_zvol_taskq, spa_export_common() will continue to wait at export until all minors are removed. Reviewed-by: Tony Hutter Reviewed-by: Brian Behlendorf Sponsored-by: Klara, Inc. Sponsored-by: Wasabi Technology, Inc. Signed-off-by: Rob Norris Closes #14872 Closes #16364 --- include/sys/zvol_impl.h | 5 ++ module/os/freebsd/zfs/zvol_os.c | 10 ++- module/os/linux/zfs/zvol_os.c | 15 ++++ module/zfs/zvol.c | 118 +++++++++++++++++++++++++------- 4 files changed, 122 insertions(+), 26 deletions(-) diff --git a/include/sys/zvol_impl.h b/include/sys/zvol_impl.h index 777b1b93ee..3cd0d78c35 100644 --- a/include/sys/zvol_impl.h +++ b/include/sys/zvol_impl.h @@ -18,6 +18,9 @@ * * CDDL HEADER END */ +/* + * Copyright (c) 2024, Klara, Inc. + */ #ifndef _SYS_ZVOL_IMPL_H #define _SYS_ZVOL_IMPL_H @@ -27,6 +30,7 @@ #define ZVOL_RDONLY (1<<0) /* zvol is readonly (writes rejected) */ #define ZVOL_WRITTEN_TO (1<<1) /* zvol has been written to (needs flush) */ #define ZVOL_EXCL (1<<2) /* zvol has O_EXCL client right now */ +#define ZVOL_REMOVING (1<<3) /* zvol waiting to remove minor */ /* * The in-core state of each volume. @@ -50,6 +54,7 @@ typedef struct zvol_state { kmutex_t zv_state_lock; /* protects zvol_state_t */ atomic_t zv_suspend_ref; /* refcount for suspend */ krwlock_t zv_suspend_lock; /* suspend lock */ + kcondvar_t zv_removing_cv; /* ready to remove minor */ struct zvol_state_os *zv_zso; /* private platform state */ boolean_t zv_threading; /* volthreading property */ } zvol_state_t; diff --git a/module/os/freebsd/zfs/zvol_os.c b/module/os/freebsd/zfs/zvol_os.c index a53a541414..ddb20b0314 100644 --- a/module/os/freebsd/zfs/zvol_os.c +++ b/module/os/freebsd/zfs/zvol_os.c @@ -30,6 +30,7 @@ * Copyright (c) 2012, 2017 by Delphix. All rights reserved. * Copyright (c) 2013, Joyent, Inc. All rights reserved. * Copyright (c) 2014 Integros [integros.com] + * Copyright (c) 2024, Klara, Inc. */ /* Portions Copyright 2011 Martin Matuska */ @@ -250,7 +251,7 @@ retry: } mutex_enter(&zv->zv_state_lock); - if (zv->zv_zso->zso_dying) { + if (zv->zv_zso->zso_dying || zv->zv_flags & ZVOL_REMOVING) { rw_exit(&zvol_state_lock); err = SET_ERROR(ENXIO); goto out_zv_locked; @@ -683,6 +684,11 @@ zvol_geom_bio_strategy(struct bio *bp) rw_enter(&zv->zv_suspend_lock, ZVOL_RW_READER); + if (zv->zv_flags & ZVOL_REMOVING) { + error = SET_ERROR(ENXIO); + goto resume; + } + switch (bp->bio_cmd) { case BIO_READ: doread = B_TRUE; @@ -1358,6 +1364,7 @@ zvol_os_free(zvol_state_t *zv) } mutex_destroy(&zv->zv_state_lock); + cv_destroy(&zv->zv_removing_cv); dataset_kstats_destroy(&zv->zv_kstat); kmem_free(zv->zv_zso, sizeof (struct zvol_state_os)); kmem_free(zv, sizeof (zvol_state_t)); @@ -1415,6 +1422,7 @@ zvol_os_create_minor(const char *name) zv = kmem_zalloc(sizeof (*zv), KM_SLEEP); zv->zv_hash = hash; mutex_init(&zv->zv_state_lock, NULL, MUTEX_DEFAULT, NULL); + cv_init(&zv->zv_removing_cv, NULL, CV_DEFAULT, NULL); zv->zv_zso = kmem_zalloc(sizeof (struct zvol_state_os), KM_SLEEP); zv->zv_volmode = volmode; if (zv->zv_volmode == ZFS_VOLMODE_GEOM) { diff --git a/module/os/linux/zfs/zvol_os.c b/module/os/linux/zfs/zvol_os.c index ba6a24f312..83f80f62ae 100644 --- a/module/os/linux/zfs/zvol_os.c +++ b/module/os/linux/zfs/zvol_os.c @@ -20,6 +20,7 @@ */ /* * Copyright (c) 2012, 2020 by Delphix. All rights reserved. + * Copyright (c) 2024, Klara, Inc. */ #include @@ -526,6 +527,11 @@ zvol_request_impl(zvol_state_t *zv, struct bio *bio, struct request *rq, uint64_t size = io_size(bio, rq); int rw = io_data_dir(bio, rq); + if (unlikely(zv->zv_flags & ZVOL_REMOVING)) { + END_IO(zv, bio, rq, -SET_ERROR(ENXIO)); + goto out; + } + if (zvol_request_sync || zv->zv_threading == B_FALSE) force_sync = 1; @@ -734,6 +740,13 @@ retry: } mutex_enter(&zv->zv_state_lock); + + if (unlikely(zv->zv_flags & ZVOL_REMOVING)) { + mutex_exit(&zv->zv_state_lock); + rw_exit(&zvol_state_lock); + return (-SET_ERROR(ENXIO)); + } + /* * Make sure zvol is not suspended during first open * (hold zv_suspend_lock) and respect proper lock acquisition @@ -1313,6 +1326,7 @@ zvol_alloc(dev_t dev, const char *name) list_link_init(&zv->zv_next); mutex_init(&zv->zv_state_lock, NULL, MUTEX_DEFAULT, NULL); + cv_init(&zv->zv_removing_cv, NULL, CV_DEFAULT, NULL); #ifdef HAVE_BLK_MQ zv->zv_zso->use_blk_mq = zvol_use_blk_mq; @@ -1438,6 +1452,7 @@ zvol_os_free(zvol_state_t *zv) ida_simple_remove(&zvol_ida, MINOR(zv->zv_zso->zvo_dev) >> ZVOL_MINOR_BITS); + cv_destroy(&zv->zv_removing_cv); mutex_destroy(&zv->zv_state_lock); dataset_kstats_destroy(&zv->zv_kstat); diff --git a/module/zfs/zvol.c b/module/zfs/zvol.c index 5b6a3f5cb4..001f774a6d 100644 --- a/module/zfs/zvol.c +++ b/module/zfs/zvol.c @@ -37,6 +37,7 @@ * Copyright 2014 Nexenta Systems, Inc. All rights reserved. * Copyright (c) 2016 Actifio, Inc. All rights reserved. * Copyright (c) 2012, 2019 by Delphix. All rights reserved. + * Copyright (c) 2024, Klara, Inc. */ /* @@ -894,6 +895,9 @@ zvol_resume(zvol_state_t *zv) */ atomic_dec(&zv->zv_suspend_ref); + if (zv->zv_flags & ZVOL_REMOVING) + cv_broadcast(&zv->zv_removing_cv); + return (SET_ERROR(error)); } @@ -929,6 +933,9 @@ zvol_last_close(zvol_state_t *zv) ASSERT(RW_READ_HELD(&zv->zv_suspend_lock)); ASSERT(MUTEX_HELD(&zv->zv_state_lock)); + if (zv->zv_flags & ZVOL_REMOVING) + cv_broadcast(&zv->zv_removing_cv); + zvol_shutdown_zv(zv); dmu_objset_disown(zv->zv_objset, 1, zv); @@ -1221,6 +1228,41 @@ zvol_create_minor(const char *name) * Remove minors for specified dataset including children and snapshots. */ +/* + * Remove the minor for a given zvol. This will do it all: + * - flag the zvol for removal, so new requests are rejected + * - wait until outstanding requests are completed + * - remove it from lists + * - free it + * It's also usable as a taskq task, and smells nice too. + */ +static void +zvol_remove_minor_task(void *arg) +{ + zvol_state_t *zv = (zvol_state_t *)arg; + + ASSERT(!RW_LOCK_HELD(&zvol_state_lock)); + ASSERT(!MUTEX_HELD(&zv->zv_state_lock)); + + mutex_enter(&zv->zv_state_lock); + while (zv->zv_open_count > 0 || atomic_read(&zv->zv_suspend_ref)) { + zv->zv_flags |= ZVOL_REMOVING; + cv_wait(&zv->zv_removing_cv, &zv->zv_state_lock); + } + mutex_exit(&zv->zv_state_lock); + + rw_enter(&zvol_state_lock, RW_WRITER); + mutex_enter(&zv->zv_state_lock); + + zvol_remove(zv); + zvol_os_clear_private(zv); + + mutex_exit(&zv->zv_state_lock); + rw_exit(&zvol_state_lock); + + zvol_os_free(zv); +} + static void zvol_free_task(void *arg) { @@ -1233,11 +1275,13 @@ zvol_remove_minors_impl(const char *name) zvol_state_t *zv, *zv_next; int namelen = ((name) ? strlen(name) : 0); taskqid_t t; - list_t free_list; + list_t delay_list, free_list; if (zvol_inhibit_dev) return; + list_create(&delay_list, sizeof (zvol_state_t), + offsetof(zvol_state_t, zv_next)); list_create(&free_list, sizeof (zvol_state_t), offsetof(zvol_state_t, zv_next)); @@ -1256,9 +1300,24 @@ zvol_remove_minors_impl(const char *name) * one is currently using this zv */ - /* If in use, leave alone */ + /* + * If in use, try to throw everyone off and try again + * later. + */ if (zv->zv_open_count > 0 || atomic_read(&zv->zv_suspend_ref)) { + zv->zv_flags |= ZVOL_REMOVING; + t = taskq_dispatch( + zv->zv_objset->os_spa->spa_zvol_taskq, + zvol_remove_minor_task, zv, TQ_SLEEP); + if (t == TASKQID_INVALID) { + /* + * Couldn't create the task, so we'll + * do it in place once the loop is + * finished. + */ + list_insert_head(&delay_list, zv); + } mutex_exit(&zv->zv_state_lock); continue; } @@ -1285,7 +1344,11 @@ zvol_remove_minors_impl(const char *name) } rw_exit(&zvol_state_lock); - /* Drop zvol_state_lock before calling zvol_free() */ + /* Wait for zvols that we couldn't create a remove task for */ + while ((zv = list_remove_head(&delay_list)) != NULL) + zvol_remove_minor_task(zv); + + /* Free any that we couldn't free in parallel earlier */ while ((zv = list_remove_head(&free_list)) != NULL) zvol_os_free(zv); } @@ -1305,33 +1368,38 @@ zvol_remove_minor_impl(const char *name) zv_next = list_next(&zvol_state_list, zv); mutex_enter(&zv->zv_state_lock); - if (strcmp(zv->zv_name, name) == 0) { - /* - * By holding zv_state_lock here, we guarantee that no - * one is currently using this zv - */ - - /* If in use, leave alone */ - if (zv->zv_open_count > 0 || - atomic_read(&zv->zv_suspend_ref)) { - mutex_exit(&zv->zv_state_lock); - continue; - } - zvol_remove(zv); - - zvol_os_clear_private(zv); - mutex_exit(&zv->zv_state_lock); + if (strcmp(zv->zv_name, name) == 0) + /* Found, leave the the loop with zv_lock held */ break; - } else { - mutex_exit(&zv->zv_state_lock); - } + mutex_exit(&zv->zv_state_lock); } - /* Drop zvol_state_lock before calling zvol_free() */ + if (zv == NULL) { + rw_exit(&zvol_state_lock); + return; + } + + ASSERT(MUTEX_HELD(&zv->zv_state_lock)); + + if (zv->zv_open_count > 0 || atomic_read(&zv->zv_suspend_ref)) { + /* + * In use, so try to throw everyone off, then wait + * until finished. + */ + zv->zv_flags |= ZVOL_REMOVING; + mutex_exit(&zv->zv_state_lock); + rw_exit(&zvol_state_lock); + zvol_remove_minor_task(zv); + return; + } + + zvol_remove(zv); + zvol_os_clear_private(zv); + + mutex_exit(&zv->zv_state_lock); rw_exit(&zvol_state_lock); - if (zv != NULL) - zvol_os_free(zv); + zvol_os_free(zv); } /*