diff options
Diffstat (limited to 'sys/contrib/openzfs/module/os/freebsd')
9 files changed, 215 insertions, 158 deletions
diff --git a/sys/contrib/openzfs/module/os/freebsd/zfs/kmod_core.c b/sys/contrib/openzfs/module/os/freebsd/zfs/kmod_core.c index c114db14a916..b218c0da8125 100644 --- a/sys/contrib/openzfs/module/os/freebsd/zfs/kmod_core.c +++ b/sys/contrib/openzfs/module/os/freebsd/zfs/kmod_core.c @@ -112,7 +112,6 @@ static int zfs__fini(void); static void zfs_shutdown(void *, int); static eventhandler_tag zfs_shutdown_event_tag; -static eventhandler_tag zfs_mountroot_event_tag; #define ZFS_MIN_KSTACK_PAGES 4 @@ -311,9 +310,6 @@ zfs_modevent(module_t mod, int type, void *unused __unused) zfs_shutdown_event_tag = EVENTHANDLER_REGISTER( shutdown_post_sync, zfs_shutdown, NULL, SHUTDOWN_PRI_FIRST); - zfs_mountroot_event_tag = EVENTHANDLER_REGISTER( - mountroot, spa_boot_init, NULL, - SI_ORDER_ANY); } return (err); case MOD_UNLOAD: @@ -322,9 +318,6 @@ zfs_modevent(module_t mod, int type, void *unused __unused) if (zfs_shutdown_event_tag != NULL) EVENTHANDLER_DEREGISTER(shutdown_post_sync, zfs_shutdown_event_tag); - if (zfs_mountroot_event_tag != NULL) - EVENTHANDLER_DEREGISTER(mountroot, - zfs_mountroot_event_tag); } return (err); case MOD_SHUTDOWN: diff --git a/sys/contrib/openzfs/module/os/freebsd/zfs/sysctl_os.c b/sys/contrib/openzfs/module/os/freebsd/zfs/sysctl_os.c index ace2360c032d..ebc2c0eeb6d2 100644 --- a/sys/contrib/openzfs/module/os/freebsd/zfs/sysctl_os.c +++ b/sys/contrib/openzfs/module/os/freebsd/zfs/sysctl_os.c @@ -163,6 +163,13 @@ param_set_arc_int(SYSCTL_HANDLER_ARGS) return (0); } +static void +warn_deprecated_sysctl(const char *old, const char *new) +{ + printf("WARNING: sysctl vfs.zfs.%s is deprecated. Use vfs.zfs.%s instead.\n", + old, new); +} + int param_set_arc_max(SYSCTL_HANDLER_ARGS) { @@ -185,12 +192,15 @@ param_set_arc_max(SYSCTL_HANDLER_ARGS) if (val != 0) zfs_arc_max = arc_c_max; + if (arg2 != 0) + warn_deprecated_sysctl("arc_max", "arc.max"); + return (0); } SYSCTL_PROC(_vfs_zfs, OID_AUTO, arc_max, CTLTYPE_ULONG | CTLFLAG_RWTUN | CTLFLAG_MPSAFE, - NULL, 0, param_set_arc_max, "LU", + NULL, 1, param_set_arc_max, "LU", "Maximum ARC size in bytes (LEGACY)"); int @@ -214,12 +224,15 @@ param_set_arc_min(SYSCTL_HANDLER_ARGS) if (val != 0) zfs_arc_min = arc_c_min; + if (arg2 != 0) + warn_deprecated_sysctl("arc_min", "arc.min"); + return (0); } SYSCTL_PROC(_vfs_zfs, OID_AUTO, arc_min, CTLTYPE_ULONG | CTLFLAG_RWTUN | CTLFLAG_MPSAFE, - NULL, 0, param_set_arc_min, "LU", + NULL, 1, param_set_arc_min, "LU", "Minimum ARC size in bytes (LEGACY)"); extern uint_t zfs_arc_free_target; @@ -242,6 +255,9 @@ param_set_arc_free_target(SYSCTL_HANDLER_ARGS) zfs_arc_free_target = val; + if (arg2 != 0) + warn_deprecated_sysctl("arc_free_target", "arc.free_target"); + return (0); } @@ -251,7 +267,7 @@ param_set_arc_free_target(SYSCTL_HANDLER_ARGS) */ SYSCTL_PROC(_vfs_zfs, OID_AUTO, arc_free_target, CTLTYPE_UINT | CTLFLAG_RW | CTLFLAG_MPSAFE, - NULL, 0, param_set_arc_free_target, "IU", + NULL, 1, param_set_arc_free_target, "IU", "Desired number of free pages below which ARC triggers reclaim" " (LEGACY)"); @@ -270,12 +286,15 @@ param_set_arc_no_grow_shift(SYSCTL_HANDLER_ARGS) arc_no_grow_shift = val; + if (arg2 != 0) + warn_deprecated_sysctl("arc_no_grow_shift", "arc.no_grow_shift"); + return (0); } SYSCTL_PROC(_vfs_zfs, OID_AUTO, arc_no_grow_shift, CTLTYPE_INT | CTLFLAG_RWTUN | CTLFLAG_MPSAFE, - NULL, 0, param_set_arc_no_grow_shift, "I", + NULL, 1, param_set_arc_no_grow_shift, "I", "log2(fraction of ARC which must be free to allow growing) (LEGACY)"); extern uint64_t l2arc_write_max; @@ -746,12 +765,15 @@ param_set_min_auto_ashift(SYSCTL_HANDLER_ARGS) zfs_vdev_min_auto_ashift = val; + if (arg2 != 0) + warn_deprecated_sysctl("min_auto_ashift", + "vdev.min_auto_ashift"); + return (0); } SYSCTL_PROC(_vfs_zfs, OID_AUTO, min_auto_ashift, - CTLTYPE_UINT | CTLFLAG_RWTUN | CTLFLAG_MPSAFE, - &zfs_vdev_min_auto_ashift, sizeof (zfs_vdev_min_auto_ashift), + CTLTYPE_UINT | CTLFLAG_RWTUN | CTLFLAG_MPSAFE, NULL, 1, param_set_min_auto_ashift, "IU", "Min ashift used when creating new top-level vdev. (LEGACY)"); @@ -771,12 +793,15 @@ param_set_max_auto_ashift(SYSCTL_HANDLER_ARGS) zfs_vdev_max_auto_ashift = val; + if (arg2 != 0) + warn_deprecated_sysctl("max_auto_ashift", + "vdev.max_auto_ashift"); + return (0); } SYSCTL_PROC(_vfs_zfs, OID_AUTO, max_auto_ashift, - CTLTYPE_UINT | CTLFLAG_RWTUN | CTLFLAG_MPSAFE, - &zfs_vdev_max_auto_ashift, sizeof (zfs_vdev_max_auto_ashift), + CTLTYPE_UINT | CTLFLAG_RWTUN | CTLFLAG_MPSAFE, NULL, 1, param_set_max_auto_ashift, "IU", "Max ashift used when optimizing for logical -> physical sector size on" " new top-level vdevs. (LEGACY)"); diff --git a/sys/contrib/openzfs/module/os/freebsd/zfs/zfs_acl.c b/sys/contrib/openzfs/module/os/freebsd/zfs/zfs_acl.c index b15a3e6e38c0..cb5787269db2 100644 --- a/sys/contrib/openzfs/module/os/freebsd/zfs/zfs_acl.c +++ b/sys/contrib/openzfs/module/os/freebsd/zfs/zfs_acl.c @@ -1175,7 +1175,7 @@ zfs_aclset_common(znode_t *zp, zfs_acl_t *aclp, cred_t *cr, dmu_tx_t *tx) int count = 0; zfs_acl_phys_t acl_phys; - if (zp->z_zfsvfs->z_replay == B_FALSE) { + if (ZTOV(zp) != NULL && zp->z_zfsvfs->z_replay == B_FALSE) { ASSERT_VOP_IN_SEQC(ZTOV(zp)); } diff --git a/sys/contrib/openzfs/module/os/freebsd/zfs/zfs_ctldir.c b/sys/contrib/openzfs/module/os/freebsd/zfs/zfs_ctldir.c index a222c5de4a2a..d0a9c662e6f0 100644 --- a/sys/contrib/openzfs/module/os/freebsd/zfs/zfs_ctldir.c +++ b/sys/contrib/openzfs/module/os/freebsd/zfs/zfs_ctldir.c @@ -674,7 +674,6 @@ zfsctl_root_readdir(struct vop_readdir_args *ap) zfs_uio_t uio; int *eofp = ap->a_eofflag; off_t dots_offset; - ssize_t orig_resid; int error; zfs_uio_init(&uio, ap->a_uio); @@ -694,11 +693,13 @@ zfsctl_root_readdir(struct vop_readdir_args *ap) return (0); } - orig_resid = zfs_uio_resid(&uio); error = sfs_readdir_common(zfsvfs->z_root, ZFSCTL_INO_ROOT, ap, &uio, &dots_offset); - if (error != 0) - goto err; + if (error != 0) { + if (error == ENAMETOOLONG) /* ran out of destination space */ + error = 0; + return (error); + } if (zfs_uio_offset(&uio) != dots_offset) return (SET_ERROR(EINVAL)); @@ -711,11 +712,8 @@ zfsctl_root_readdir(struct vop_readdir_args *ap) entry.d_reclen = sizeof (entry); error = vfs_read_dirent(ap, &entry, zfs_uio_offset(&uio)); if (error != 0) { -err: - if (error == ENAMETOOLONG) { - error = orig_resid == zfs_uio_resid(&uio) ? - EINVAL : 0; - } + if (error == ENAMETOOLONG) + error = 0; return (SET_ERROR(error)); } if (eofp != NULL) @@ -764,8 +762,7 @@ zfsctl_common_pathconf(struct vop_pathconf_args *ap) return (0); case _PC_MIN_HOLE_SIZE: - *ap->a_retval = (int)SPA_MINBLOCKSIZE; - return (0); + return (EINVAL); case _PC_ACL_EXTENDED: *ap->a_retval = 0; @@ -1060,21 +1057,17 @@ zfsctl_snapdir_readdir(struct vop_readdir_args *ap) zfs_uio_t uio; int *eofp = ap->a_eofflag; off_t dots_offset; - ssize_t orig_resid; int error; zfs_uio_init(&uio, ap->a_uio); - orig_resid = zfs_uio_resid(&uio); ASSERT3S(vp->v_type, ==, VDIR); error = sfs_readdir_common(ZFSCTL_INO_ROOT, ZFSCTL_INO_SNAPDIR, ap, &uio, &dots_offset); if (error != 0) { - if (error == ENAMETOOLONG) { /* ran out of destination space */ - error = orig_resid == zfs_uio_resid(&uio) ? - EINVAL : 0; - } + if (error == ENAMETOOLONG) /* ran out of destination space */ + error = 0; return (error); } @@ -1092,13 +1085,9 @@ zfsctl_snapdir_readdir(struct vop_readdir_args *ap) dsl_pool_config_exit(dmu_objset_pool(zfsvfs->z_os), FTAG); if (error != 0) { if (error == ENOENT) { - if (orig_resid == zfs_uio_resid(&uio)) { - error = EINVAL; - } else { - error = 0; - if (eofp != NULL) - *eofp = 1; - } + if (eofp != NULL) + *eofp = 1; + error = 0; } zfs_exit(zfsvfs, FTAG); return (error); @@ -1111,10 +1100,8 @@ zfsctl_snapdir_readdir(struct vop_readdir_args *ap) entry.d_reclen = sizeof (entry); error = vfs_read_dirent(ap, &entry, zfs_uio_offset(&uio)); if (error != 0) { - if (error == ENAMETOOLONG) { - error = orig_resid == zfs_uio_resid(&uio) ? - EINVAL : 0; - } + if (error == ENAMETOOLONG) + error = 0; zfs_exit(zfsvfs, FTAG); return (SET_ERROR(error)); } diff --git a/sys/contrib/openzfs/module/os/freebsd/zfs/zfs_file_os.c b/sys/contrib/openzfs/module/os/freebsd/zfs/zfs_file_os.c index 21e5f7938f9f..ca13569a1235 100644 --- a/sys/contrib/openzfs/module/os/freebsd/zfs/zfs_file_os.c +++ b/sys/contrib/openzfs/module/os/freebsd/zfs/zfs_file_os.c @@ -164,8 +164,9 @@ zfs_file_write(zfs_file_t *fp, const void *buf, size_t count, ssize_t *resid) int zfs_file_pwrite(zfs_file_t *fp, const void *buf, size_t count, loff_t off, - ssize_t *resid) + uint8_t ashift, ssize_t *resid) { + (void) ashift; return (zfs_file_write_impl(fp, buf, count, &off, resid)); } diff --git a/sys/contrib/openzfs/module/os/freebsd/zfs/zfs_vnops_os.c b/sys/contrib/openzfs/module/os/freebsd/zfs/zfs_vnops_os.c index 174141a5deab..f34a2fd37a77 100644 --- a/sys/contrib/openzfs/module/os/freebsd/zfs/zfs_vnops_os.c +++ b/sys/contrib/openzfs/module/os/freebsd/zfs/zfs_vnops_os.c @@ -61,6 +61,7 @@ #include <sys/fs/zfs.h> #include <sys/dmu.h> #include <sys/dmu_objset.h> +#include <sys/dsl_dataset.h> #include <sys/spa.h> #include <sys/txg.h> #include <sys/dbuf.h> @@ -388,7 +389,9 @@ zfs_ioctl(vnode_t *vp, ulong_t com, intptr_t data, int flag, cred_t *cred, error = vn_lock(vp, LK_EXCLUSIVE); if (error) return (error); + vn_seqc_write_begin(vp); error = zfs_ioctl_setxattr(vp, fsx, cred); + vn_seqc_write_end(vp); VOP_UNLOCK(vp); return (error); } @@ -1695,7 +1698,6 @@ zfs_readdir(vnode_t *vp, zfs_uio_t *uio, cred_t *cr, int *eofp, objset_t *os; caddr_t outbuf; size_t bufsize; - ssize_t orig_resid; zap_cursor_t zc; zap_attribute_t *zap; uint_t bytes_wanted; @@ -1744,7 +1746,6 @@ zfs_readdir(vnode_t *vp, zfs_uio_t *uio, cred_t *cr, int *eofp, error = 0; os = zfsvfs->z_os; offset = zfs_uio_offset(uio); - orig_resid = zfs_uio_resid(uio); prefetch = zp->z_zn_prefetch; zap = zap_attribute_long_alloc(); @@ -1924,7 +1925,7 @@ update: kmem_free(outbuf, bufsize); if (error == ENOENT) - error = orig_resid == zfs_uio_resid(uio) ? EINVAL : 0; + error = 0; ZFS_ACCESSTIME_STAMP(zfsvfs, zp); @@ -2205,6 +2206,7 @@ zfs_setattr_dir(znode_t *dzp) if (err) break; + vn_seqc_write_begin(ZTOV(zp)); mutex_enter(&dzp->z_lock); if (zp->z_uid != dzp->z_uid) { @@ -2254,6 +2256,7 @@ sa_add_projid_err: dmu_tx_abort(tx); } tx = NULL; + vn_seqc_write_end(ZTOV(zp)); if (err != 0 && err != ENOENT) break; @@ -4113,6 +4116,7 @@ zfs_pathconf(vnode_t *vp, int cmd, ulong_t *valp, cred_t *cr, { znode_t *zp; zfsvfs_t *zfsvfs; + uint_t blksize, iosize; int error; switch (cmd) { @@ -4124,8 +4128,20 @@ zfs_pathconf(vnode_t *vp, int cmd, ulong_t *valp, cred_t *cr, *valp = 64; return (0); case _PC_MIN_HOLE_SIZE: - *valp = (int)SPA_MINBLOCKSIZE; - return (0); + iosize = vp->v_mount->mnt_stat.f_iosize; + if (vp->v_type == VREG) { + zp = VTOZ(vp); + blksize = zp->z_blksz; + if (zp->z_size <= blksize) + blksize = MAX(blksize, iosize); + *valp = (int)blksize; + return (0); + } + if (vp->v_type == VDIR) { + *valp = (int)iosize; + return (0); + } + return (EINVAL); case _PC_ACL_EXTENDED: #if 0 /* POSIX ACLs are not implemented for ZFS on FreeBSD yet. */ zp = VTOZ(vp); @@ -4207,8 +4223,20 @@ zfs_getpages(struct vnode *vp, vm_page_t *ma, int count, int *rbehind, zfs_vmobject_wlock(object); (void) vm_page_grab_pages(object, OFF_TO_IDX(start), - VM_ALLOC_NORMAL | VM_ALLOC_WAITOK | VM_ALLOC_ZERO, + VM_ALLOC_NORMAL | VM_ALLOC_WAITOK, ma, count); + if (!vm_page_all_valid(ma[count - 1])) { + /* + * Later in this function, we copy DMU data to + * invalid pages only. The last page may not be + * entirely filled though, if the file does not + * end on a page boundary. Therefore, we zero + * that last page here to make sure it does not + * contain garbage after the end of file. + */ + ASSERT(vm_page_none_valid(ma[count - 1])); + vm_page_zero_invalid(ma[count - 1], FALSE); + } zfs_vmobject_wunlock(object); } if (blksz == zp->z_blksz) @@ -5729,6 +5757,9 @@ zfs_freebsd_pathconf(struct vop_pathconf_args *ap) { ulong_t val; int error; +#ifdef _PC_CLONE_BLKSIZE + zfsvfs_t *zfsvfs; +#endif error = zfs_pathconf(ap->a_vp, ap->a_name, &val, curthread->td_ucred, NULL); @@ -5775,6 +5806,21 @@ zfs_freebsd_pathconf(struct vop_pathconf_args *ap) *ap->a_retval = 1; return (0); #endif +#ifdef _PC_CLONE_BLKSIZE + case _PC_CLONE_BLKSIZE: + zfsvfs = (zfsvfs_t *)ap->a_vp->v_mount->mnt_data; + if (zfs_bclone_enabled && + spa_feature_is_enabled(dmu_objset_spa(zfsvfs->z_os), + SPA_FEATURE_BLOCK_CLONING)) + *ap->a_retval = dsl_dataset_feature_is_active( + zfsvfs->z_os->os_dsl_dataset, + SPA_FEATURE_LARGE_BLOCKS) ? + SPA_MAXBLOCKSIZE : + SPA_OLD_MAXBLOCKSIZE; + else + *ap->a_retval = 0; + return (0); +#endif default: return (vop_stdpathconf(ap)); } diff --git a/sys/contrib/openzfs/module/os/freebsd/zfs/zfs_znode_os.c b/sys/contrib/openzfs/module/os/freebsd/zfs/zfs_znode_os.c index 7cd0a153577c..649022ab5bcb 100644 --- a/sys/contrib/openzfs/module/os/freebsd/zfs/zfs_znode_os.c +++ b/sys/contrib/openzfs/module/os/freebsd/zfs/zfs_znode_os.c @@ -817,6 +817,10 @@ zfs_mknode(znode_t *dzp, vattr_t *vap, dmu_tx_t *tx, cred_t *cr, (*zpp)->z_dnodesize = dnodesize; (*zpp)->z_projid = projid; + vnode_t *vp = ZTOV(*zpp); + if (!(flag & IS_ROOT_NODE)) + vn_seqc_write_begin(vp); + if (vap->va_mask & AT_XVATTR) zfs_xvattr_set(*zpp, (xvattr_t *)vap, tx); @@ -825,7 +829,7 @@ zfs_mknode(znode_t *dzp, vattr_t *vap, dmu_tx_t *tx, cred_t *cr, VERIFY0(zfs_aclset_common(*zpp, acl_ids->z_aclp, cr, tx)); } if (!(flag & IS_ROOT_NODE)) { - vnode_t *vp = ZTOV(*zpp); + vn_seqc_write_end(vp); vp->v_vflag |= VV_FORCEINSMQ; int err = insmntque(vp, zfsvfs->z_vfs); vp->v_vflag &= ~VV_FORCEINSMQ; diff --git a/sys/contrib/openzfs/module/os/freebsd/zfs/zio_crypt.c b/sys/contrib/openzfs/module/os/freebsd/zfs/zio_crypt.c index 91cf38016e00..8562c42b3220 100644 --- a/sys/contrib/openzfs/module/os/freebsd/zfs/zio_crypt.c +++ b/sys/contrib/openzfs/module/os/freebsd/zfs/zio_crypt.c @@ -437,6 +437,7 @@ zio_crypt_key_wrap(crypto_key_t *cwkey, zio_crypt_key_t *key, uint8_t *iv, ASSERT3U(crypt, <, ZIO_CRYPT_FUNCTIONS); + memset(&cuio_s, 0, sizeof (cuio_s)); zfs_uio_init(&cuio, &cuio_s); keydata_len = zio_crypt_table[crypt].ci_keylen; @@ -519,6 +520,7 @@ zio_crypt_key_unwrap(crypto_key_t *cwkey, uint64_t crypt, uint64_t version, keydata_len = zio_crypt_table[crypt].ci_keylen; rw_init(&key->zk_salt_lock, NULL, RW_DEFAULT, NULL); + memset(&cuio_s, 0, sizeof (cuio_s)); zfs_uio_init(&cuio, &cuio_s); /* diff --git a/sys/contrib/openzfs/module/os/freebsd/zfs/zvol_os.c b/sys/contrib/openzfs/module/os/freebsd/zfs/zvol_os.c index 265dfd55fc4d..3ddbfcb97184 100644 --- a/sys/contrib/openzfs/module/os/freebsd/zfs/zvol_os.c +++ b/sys/contrib/openzfs/module/os/freebsd/zfs/zvol_os.c @@ -31,7 +31,7 @@ * 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. + * Copyright (c) 2024, 2025, Klara, Inc. */ /* Portions Copyright 2011 Martin Matuska <mm@FreeBSD.org> */ @@ -183,6 +183,7 @@ static struct filterops zvol_filterops_vnode = { .f_isfd = 1, .f_detach = zvol_filter_detach, .f_event = zvol_filter_vnode, + .f_copy = knote_triv_copy, }; extern uint_t zfs_geom_probe_vdev_key; @@ -196,7 +197,6 @@ DECLARE_GEOM_CLASS(zfs_zvol_class, zfs_zvol); static int zvol_geom_open(struct g_provider *pp, int flag, int count); static int zvol_geom_close(struct g_provider *pp, int flag, int count); -static void zvol_geom_destroy(zvol_state_t *zv); static int zvol_geom_access(struct g_provider *pp, int acr, int acw, int ace); static void zvol_geom_bio_start(struct bio *bp); static int zvol_geom_bio_getattr(struct bio *bp); @@ -226,25 +226,14 @@ zvol_geom_open(struct g_provider *pp, int flag, int count) } retry: - rw_enter(&zvol_state_lock, ZVOL_RW_READER); - /* - * Obtain a copy of private under zvol_state_lock to make sure either - * the result of zvol free code setting private to NULL is observed, - * or the zv is protected from being freed because of the positive - * zv_open_count. - */ - zv = pp->private; - if (zv == NULL) { - rw_exit(&zvol_state_lock); - err = SET_ERROR(ENXIO); - goto out_locked; - } + zv = atomic_load_ptr(&pp->private); + if (zv == NULL) + return (SET_ERROR(ENXIO)); mutex_enter(&zv->zv_state_lock); if (zv->zv_zso->zso_dying || zv->zv_flags & ZVOL_REMOVING) { - rw_exit(&zvol_state_lock); err = SET_ERROR(ENXIO); - goto out_zv_locked; + goto out_locked; } ASSERT3S(zv->zv_volmode, ==, ZFS_VOLMODE_GEOM); @@ -257,8 +246,24 @@ retry: drop_suspend = B_TRUE; if (!rw_tryenter(&zv->zv_suspend_lock, ZVOL_RW_READER)) { mutex_exit(&zv->zv_state_lock); + + /* + * Removal may happen while the locks are down, so + * we can't trust zv any longer; we have to start over. + */ + zv = atomic_load_ptr(&pp->private); + if (zv == NULL) + return (SET_ERROR(ENXIO)); + rw_enter(&zv->zv_suspend_lock, ZVOL_RW_READER); mutex_enter(&zv->zv_state_lock); + + if (zv->zv_zso->zso_dying || + zv->zv_flags & ZVOL_REMOVING) { + err = SET_ERROR(ENXIO); + goto out_locked; + } + /* Check to see if zv_suspend_lock is needed. */ if (zv->zv_open_count != 0) { rw_exit(&zv->zv_suspend_lock); @@ -266,7 +271,6 @@ retry: } } } - rw_exit(&zvol_state_lock); ASSERT(MUTEX_HELD(&zv->zv_state_lock)); @@ -294,7 +298,7 @@ retry: if (drop_namespace) mutex_exit(&spa_namespace_lock); if (err) - goto out_zv_locked; + goto out_locked; pp->mediasize = zv->zv_volsize; pp->stripeoffset = 0; pp->stripesize = zv->zv_volblocksize; @@ -329,9 +333,8 @@ out_opened: zvol_last_close(zv); wakeup(zv); } -out_zv_locked: - mutex_exit(&zv->zv_state_lock); out_locked: + mutex_exit(&zv->zv_state_lock); if (drop_suspend) rw_exit(&zv->zv_suspend_lock); return (err); @@ -345,12 +348,9 @@ zvol_geom_close(struct g_provider *pp, int flag, int count) boolean_t drop_suspend = B_TRUE; int new_open_count; - rw_enter(&zvol_state_lock, ZVOL_RW_READER); - zv = pp->private; - if (zv == NULL) { - rw_exit(&zvol_state_lock); + zv = atomic_load_ptr(&pp->private); + if (zv == NULL) return (SET_ERROR(ENXIO)); - } mutex_enter(&zv->zv_state_lock); if (zv->zv_flags & ZVOL_EXCL) { @@ -377,6 +377,15 @@ zvol_geom_close(struct g_provider *pp, int flag, int count) mutex_exit(&zv->zv_state_lock); rw_enter(&zv->zv_suspend_lock, ZVOL_RW_READER); mutex_enter(&zv->zv_state_lock); + + /* + * Unlike in zvol_geom_open(), we don't check if + * removal started here, because we might be one of the + * openers that needs to be thrown out! If we're the + * last, we need to call zvol_last_close() below to + * finish cleanup. So, no special treatment for us. + */ + /* Check to see if zv_suspend_lock is needed. */ new_open_count = zv->zv_open_count - count; if (new_open_count != 0) { @@ -387,7 +396,6 @@ zvol_geom_close(struct g_provider *pp, int flag, int count) } else { drop_suspend = B_FALSE; } - rw_exit(&zvol_state_lock); ASSERT(MUTEX_HELD(&zv->zv_state_lock)); @@ -408,20 +416,6 @@ zvol_geom_close(struct g_provider *pp, int flag, int count) return (0); } -static void -zvol_geom_destroy(zvol_state_t *zv) -{ - struct zvol_state_geom *zsg = &zv->zv_zso->zso_geom; - struct g_provider *pp = zsg->zsg_provider; - - ASSERT3S(zv->zv_volmode, ==, ZFS_VOLMODE_GEOM); - - g_topology_assert(); - - zsg->zsg_provider = NULL; - g_wither_geom(pp->geom, ENXIO); -} - void zvol_wait_close(zvol_state_t *zv) { @@ -454,7 +448,7 @@ zvol_geom_access(struct g_provider *pp, int acr, int acw, int ace) ("Unsupported access request to %s (acr=%d, acw=%d, ace=%d).", pp->name, acr, acw, ace)); - if (pp->private == NULL) { + if (atomic_load_ptr(&pp->private) == NULL) { if (acr <= 0 && acw <= 0 && ace <= 0) return (0); return (pp->error); @@ -921,25 +915,14 @@ zvol_cdev_open(struct cdev *dev, int flags, int fmt, struct thread *td) boolean_t drop_suspend = B_FALSE; retry: - rw_enter(&zvol_state_lock, ZVOL_RW_READER); - /* - * Obtain a copy of si_drv2 under zvol_state_lock to make sure either - * the result of zvol free code setting si_drv2 to NULL is observed, - * or the zv is protected from being freed because of the positive - * zv_open_count. - */ - zv = dev->si_drv2; - if (zv == NULL) { - rw_exit(&zvol_state_lock); - err = SET_ERROR(ENXIO); - goto out_locked; - } + zv = atomic_load_ptr(&dev->si_drv2); + if (zv == NULL) + return (SET_ERROR(ENXIO)); mutex_enter(&zv->zv_state_lock); - if (zv->zv_zso->zso_dying) { - rw_exit(&zvol_state_lock); + if (zv->zv_zso->zso_dying || zv->zv_flags & ZVOL_REMOVING) { err = SET_ERROR(ENXIO); - goto out_zv_locked; + goto out_locked; } ASSERT3S(zv->zv_volmode, ==, ZFS_VOLMODE_DEV); @@ -954,6 +937,13 @@ retry: mutex_exit(&zv->zv_state_lock); rw_enter(&zv->zv_suspend_lock, ZVOL_RW_READER); mutex_enter(&zv->zv_state_lock); + + if (unlikely(zv->zv_flags & ZVOL_REMOVING)) { + /* Removal started while locks were down. */ + err = SET_ERROR(ENXIO); + goto out_locked; + } + /* Check to see if zv_suspend_lock is needed. */ if (zv->zv_open_count != 0) { rw_exit(&zv->zv_suspend_lock); @@ -961,7 +951,6 @@ retry: } } } - rw_exit(&zvol_state_lock); ASSERT(MUTEX_HELD(&zv->zv_state_lock)); @@ -989,7 +978,7 @@ retry: if (drop_namespace) mutex_exit(&spa_namespace_lock); if (err) - goto out_zv_locked; + goto out_locked; } ASSERT(MUTEX_HELD(&zv->zv_state_lock)); @@ -1016,9 +1005,8 @@ out_opened: zvol_last_close(zv); wakeup(zv); } -out_zv_locked: - mutex_exit(&zv->zv_state_lock); out_locked: + mutex_exit(&zv->zv_state_lock); if (drop_suspend) rw_exit(&zv->zv_suspend_lock); return (err); @@ -1030,12 +1018,9 @@ zvol_cdev_close(struct cdev *dev, int flags, int fmt, struct thread *td) zvol_state_t *zv; boolean_t drop_suspend = B_TRUE; - rw_enter(&zvol_state_lock, ZVOL_RW_READER); - zv = dev->si_drv2; - if (zv == NULL) { - rw_exit(&zvol_state_lock); + zv = atomic_load_ptr(&dev->si_drv2); + if (zv == NULL) return (SET_ERROR(ENXIO)); - } mutex_enter(&zv->zv_state_lock); if (zv->zv_flags & ZVOL_EXCL) { @@ -1060,6 +1045,15 @@ zvol_cdev_close(struct cdev *dev, int flags, int fmt, struct thread *td) mutex_exit(&zv->zv_state_lock); rw_enter(&zv->zv_suspend_lock, ZVOL_RW_READER); mutex_enter(&zv->zv_state_lock); + + /* + * Unlike in zvol_cdev_open(), we don't check if + * removal started here, because we might be one of the + * openers that needs to be thrown out! If we're the + * last, we need to call zvol_last_close() below to + * finish cleanup. So, no special treatment for us. + */ + /* Check to see if zv_suspend_lock is needed. */ if (zv->zv_open_count != 1) { rw_exit(&zv->zv_suspend_lock); @@ -1069,7 +1063,6 @@ zvol_cdev_close(struct cdev *dev, int flags, int fmt, struct thread *td) } else { drop_suspend = B_FALSE; } - rw_exit(&zvol_state_lock); ASSERT(MUTEX_HELD(&zv->zv_state_lock)); @@ -1101,7 +1094,8 @@ zvol_cdev_ioctl(struct cdev *dev, ulong_t cmd, caddr_t data, int error; boolean_t sync; - zv = dev->si_drv2; + zv = atomic_load_ptr(&dev->si_drv2); + ASSERT3P(zv, !=, NULL); error = 0; KASSERT(zv->zv_open_count > 0, @@ -1162,6 +1156,7 @@ zvol_cdev_ioctl(struct cdev *dev, ulong_t cmd, caddr_t data, *(off_t *)data = 0; break; case DIOCGATTR: { + rw_enter(&zv->zv_suspend_lock, ZVOL_RW_READER); spa_t *spa = dmu_objset_spa(zv->zv_objset); struct diocgattr_arg *arg = (struct diocgattr_arg *)data; uint64_t refd, avail, usedobjs, availobjs; @@ -1186,6 +1181,7 @@ zvol_cdev_ioctl(struct cdev *dev, ulong_t cmd, caddr_t data, arg->value.off = refd / DEV_BSIZE; } else error = SET_ERROR(ENOIOCTL); + rw_exit(&zv->zv_suspend_lock); break; } case FIOSEEKHOLE: @@ -1196,10 +1192,12 @@ zvol_cdev_ioctl(struct cdev *dev, ulong_t cmd, caddr_t data, hole = (cmd == FIOSEEKHOLE); noff = *off; + rw_enter(&zv->zv_suspend_lock, ZVOL_RW_READER); lr = zfs_rangelock_enter(&zv->zv_rangelock, 0, UINT64_MAX, RL_READER); error = dmu_offset_next(zv->zv_objset, ZVOL_OBJ, hole, &noff); zfs_rangelock_exit(lr); + rw_exit(&zv->zv_suspend_lock); *off = noff; break; } @@ -1400,42 +1398,65 @@ zvol_alloc(const char *name, uint64_t volsize, uint64_t volblocksize, * Remove minor node for the specified volume. */ void -zvol_os_free(zvol_state_t *zv) +zvol_os_remove_minor(zvol_state_t *zv) { - ASSERT(!RW_LOCK_HELD(&zv->zv_suspend_lock)); - ASSERT(!MUTEX_HELD(&zv->zv_state_lock)); + ASSERT(MUTEX_HELD(&zv->zv_state_lock)); ASSERT0(zv->zv_open_count); + ASSERT0(atomic_read(&zv->zv_suspend_ref)); + ASSERT(zv->zv_flags & ZVOL_REMOVING); - ZFS_LOG(1, "ZVOL %s destroyed.", zv->zv_name); - - rw_destroy(&zv->zv_suspend_lock); - zfs_rangelock_fini(&zv->zv_rangelock); + struct zvol_state_os *zso = zv->zv_zso; + zv->zv_zso = NULL; if (zv->zv_volmode == ZFS_VOLMODE_GEOM) { - struct zvol_state_geom *zsg = &zv->zv_zso->zso_geom; - struct g_provider *pp __maybe_unused = zsg->zsg_provider; - - ASSERT0P(pp->private); + struct zvol_state_geom *zsg = &zso->zso_geom; + struct g_provider *pp = zsg->zsg_provider; + atomic_store_ptr(&pp->private, NULL); + mutex_exit(&zv->zv_state_lock); g_topology_lock(); - zvol_geom_destroy(zv); + g_wither_geom(pp->geom, ENXIO); g_topology_unlock(); } else if (zv->zv_volmode == ZFS_VOLMODE_DEV) { - struct zvol_state_dev *zsd = &zv->zv_zso->zso_dev; + struct zvol_state_dev *zsd = &zso->zso_dev; struct cdev *dev = zsd->zsd_cdev; + if (dev != NULL) + atomic_store_ptr(&dev->si_drv2, NULL); + mutex_exit(&zv->zv_state_lock); + if (dev != NULL) { - ASSERT0P(dev->si_drv2); destroy_dev(dev); knlist_clear(&zsd->zsd_selinfo.si_note, 0); knlist_destroy(&zsd->zsd_selinfo.si_note); } } + kmem_free(zso, sizeof (struct zvol_state_os)); + + mutex_enter(&zv->zv_state_lock); +} + +void +zvol_os_free(zvol_state_t *zv) +{ + ASSERT(!RW_LOCK_HELD(&zv->zv_suspend_lock)); + ASSERT(!MUTEX_HELD(&zv->zv_state_lock)); + ASSERT0(zv->zv_open_count); + ASSERT0P(zv->zv_zso); + + ASSERT0P(zv->zv_objset); + ASSERT0P(zv->zv_zilog); + ASSERT0P(zv->zv_dn); + + ZFS_LOG(1, "ZVOL %s destroyed.", zv->zv_name); + + rw_destroy(&zv->zv_suspend_lock); + zfs_rangelock_fini(&zv->zv_rangelock); + 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)); zvol_minors--; } @@ -1538,28 +1559,6 @@ out_doi: return (error); } -void -zvol_os_clear_private(zvol_state_t *zv) -{ - ASSERT(RW_LOCK_HELD(&zvol_state_lock)); - if (zv->zv_volmode == ZFS_VOLMODE_GEOM) { - struct zvol_state_geom *zsg = &zv->zv_zso->zso_geom; - struct g_provider *pp = zsg->zsg_provider; - - if (pp->private == NULL) /* already cleared */ - return; - - pp->private = NULL; - ASSERT(!RW_LOCK_HELD(&zv->zv_suspend_lock)); - } else if (zv->zv_volmode == ZFS_VOLMODE_DEV) { - struct zvol_state_dev *zsd = &zv->zv_zso->zso_dev; - struct cdev *dev = zsd->zsd_cdev; - - if (dev != NULL) - dev->si_drv2 = NULL; - } -} - int zvol_os_update_volsize(zvol_state_t *zv, uint64_t volsize) { |