dsl_dir.c

Go to the documentation of this file.

/*
 * CDDL HEADER START
 *
 * The contents of this file are subject to the terms of the
 * Common Development and Distribution License (the "License").
 * You may not use this file except in compliance with the License.
 *
 * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
 * or http://www.opensolaris.org/os/licensing.
 * See the License for the specific language governing permissions
 * and limitations under the License.
 *
 * When distributing Covered Code, include this CDDL HEADER in each
 * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
 * If applicable, add the following below this CDDL HEADER, with the
 * fields enclosed by brackets "[]" replaced with your own identifying
 * information: Portions Copyright [yyyy] [name of copyright owner]
 *
 * CDDL HEADER END
 */
/*
 * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved.
 * Copyright (c) 2011 Pawel Jakub Dawidek <pawel@dawidek.net>.
 * All rights reserved.
 */

#include <sys/dmu.h>
#include <sys/dmu_objset.h>
#include <sys/dmu_tx.h>
#include <sys/dsl_dataset.h>
#include <sys/dsl_dir.h>
#include <sys/dsl_prop.h>
#include <sys/dsl_synctask.h>
#include <sys/dsl_deleg.h>
#include <sys/spa.h>
#include <sys/metaslab.h>
#include <sys/zap.h>
#include <sys/zio.h>
#include <sys/arc.h>
#include <sys/sunddi.h>
#include <sys/zvol.h>
#ifdef _KERNEL
#include <sys/zfs_vfsops.h>
#endif
#include "zfs_namecheck.h"

static uint64_t dsl_dir_space_towrite(dsl_dir_t *dd);
static void dsl_dir_set_reservation_sync(void *arg1, void *arg2, dmu_tx_t *tx);


/* ARGSUSED */
static void
dsl_dir_evict(dmu_buf_t *db, void *arg)
{
        dsl_dir_t *dd = arg;
        dsl_pool_t *dp = dd->dd_pool;
        int t;

        for (t = 0; t < TXG_SIZE; t++) {
                ASSERT(!txg_list_member(&dp->dp_dirty_dirs, dd, t));
                ASSERT(dd->dd_tempreserved[t] == 0);
                ASSERT(dd->dd_space_towrite[t] == 0);
        }

        if (dd->dd_parent)
                dsl_dir_close(dd->dd_parent, dd);

        spa_close(dd->dd_pool->dp_spa, dd);

        /*
         * The props callback list should have been cleaned up by
         * objset_evict().
         */
        list_destroy(&dd->dd_prop_cbs);
        mutex_destroy(&dd->dd_lock);
        kmem_free(dd, sizeof (dsl_dir_t));
}

int
dsl_dir_open_obj(dsl_pool_t *dp, uint64_t ddobj,
    const char *tail, void *tag, dsl_dir_t **ddp)
{
        dmu_buf_t *dbuf;
        dsl_dir_t *dd;
        int err;

        ASSERT(RW_LOCK_HELD(&dp->dp_config_rwlock) ||
            dsl_pool_sync_context(dp));

        err = dmu_bonus_hold(dp->dp_meta_objset, ddobj, tag, &dbuf);
        if (err)
                return (err);
        dd = dmu_buf_get_user(dbuf);
#ifdef ZFS_DEBUG
        {
                dmu_object_info_t doi;
                dmu_object_info_from_db(dbuf, &doi);
                ASSERT3U(doi.doi_type, ==, DMU_OT_DSL_DIR);
                ASSERT3U(doi.doi_bonus_size, >=, sizeof (dsl_dir_phys_t));
        }
#endif
        if (dd == NULL) {
                dsl_dir_t *winner;

                dd = kmem_zalloc(sizeof (dsl_dir_t), KM_SLEEP);
                dd->dd_object = ddobj;
                dd->dd_dbuf = dbuf;
                dd->dd_pool = dp;
                dd->dd_phys = dbuf->db_data;
                mutex_init(&dd->dd_lock, NULL, MUTEX_DEFAULT, NULL);

                list_create(&dd->dd_prop_cbs, sizeof (dsl_prop_cb_record_t),
                    offsetof(dsl_prop_cb_record_t, cbr_node));

                dsl_dir_snap_cmtime_update(dd);

                if (dd->dd_phys->dd_parent_obj) {
                        err = dsl_dir_open_obj(dp, dd->dd_phys->dd_parent_obj,
                            NULL, dd, &dd->dd_parent);
                        if (err)
                                goto errout;
                        if (tail) {
#ifdef ZFS_DEBUG
                                uint64_t foundobj;

                                err = zap_lookup(dp->dp_meta_objset,
                                    dd->dd_parent->dd_phys->dd_child_dir_zapobj,
                                    tail, sizeof (foundobj), 1, &foundobj);
                                ASSERT(err || foundobj == ddobj);
#endif
                                (void) strcpy(dd->dd_myname, tail);
                        } else {
                                err = zap_value_search(dp->dp_meta_objset,
                                    dd->dd_parent->dd_phys->dd_child_dir_zapobj,
                                    ddobj, 0, dd->dd_myname);
                        }
                        if (err)
                                goto errout;
                } else {
                        (void) strcpy(dd->dd_myname, spa_name(dp->dp_spa));
                }

                if (dsl_dir_is_clone(dd)) {
                        dmu_buf_t *origin_bonus;
                        dsl_dataset_phys_t *origin_phys;

                        /*
                         * We can't open the origin dataset, because
                         * that would require opening this dsl_dir.
                         * Just look at its phys directly instead.
                         */
                        err = dmu_bonus_hold(dp->dp_meta_objset,
                            dd->dd_phys->dd_origin_obj, FTAG, &origin_bonus);
                        if (err)
                                goto errout;
                        origin_phys = origin_bonus->db_data;
                        dd->dd_origin_txg =
                            origin_phys->ds_creation_txg;
                        dmu_buf_rele(origin_bonus, FTAG);
                }

                winner = dmu_buf_set_user_ie(dbuf, dd, &dd->dd_phys,
                    dsl_dir_evict);
                if (winner) {
                        if (dd->dd_parent)
                                dsl_dir_close(dd->dd_parent, dd);
                        mutex_destroy(&dd->dd_lock);
                        kmem_free(dd, sizeof (dsl_dir_t));
                        dd = winner;
                } else {
                        spa_open_ref(dp->dp_spa, dd);
                }
        }

        /*
         * The dsl_dir_t has both open-to-close and instantiate-to-evict
         * holds on the spa.  We need the open-to-close holds because
         * otherwise the spa_refcnt wouldn't change when we open a
         * dir which the spa also has open, so we could incorrectly
         * think it was OK to unload/export/destroy the pool.  We need
         * the instantiate-to-evict hold because the dsl_dir_t has a
         * pointer to the dd_pool, which has a pointer to the spa_t.
         */
        spa_open_ref(dp->dp_spa, tag);
        ASSERT3P(dd->dd_pool, ==, dp);
        ASSERT3U(dd->dd_object, ==, ddobj);
        ASSERT3P(dd->dd_dbuf, ==, dbuf);
        *ddp = dd;
        return (0);

errout:
        if (dd->dd_parent)
                dsl_dir_close(dd->dd_parent, dd);
        mutex_destroy(&dd->dd_lock);
        kmem_free(dd, sizeof (dsl_dir_t));
        dmu_buf_rele(dbuf, tag);
        return (err);
}

void
dsl_dir_close(dsl_dir_t *dd, void *tag)
{
        dprintf_dd(dd, "%s\n", "");
        spa_close(dd->dd_pool->dp_spa, tag);
        dmu_buf_rele(dd->dd_dbuf, tag);
}

void
dsl_dir_name(dsl_dir_t *dd, char *buf)
{
        if (dd->dd_parent) {
                dsl_dir_name(dd->dd_parent, buf);
                (void) strcat(buf, "/");
        } else {
                buf[0] = '\0';
        }
        if (!MUTEX_HELD(&dd->dd_lock)) {
                /*
                 * recursive mutex so that we can use
                 * dprintf_dd() with dd_lock held
                 */
                mutex_enter(&dd->dd_lock);
                (void) strcat(buf, dd->dd_myname);
                mutex_exit(&dd->dd_lock);
        } else {
                (void) strcat(buf, dd->dd_myname);
        }
}

int
dsl_dir_namelen(dsl_dir_t *dd)
{
        int result = 0;

        if (dd->dd_parent) {
                /* parent's name + 1 for the "/" */
                result = dsl_dir_namelen(dd->dd_parent) + 1;
        }

        if (!MUTEX_HELD(&dd->dd_lock)) {
                /* see dsl_dir_name */
                mutex_enter(&dd->dd_lock);
                result += strlen(dd->dd_myname);
                mutex_exit(&dd->dd_lock);
        } else {
                result += strlen(dd->dd_myname);
        }

        return (result);
}

static int
getcomponent(const char *path, char *component, const char **nextp)
{
        char *p;
        if ((path == NULL) || (path[0] == '\0'))
                return (ENOENT);
        /* This would be a good place to reserve some namespace... */
        p = strpbrk(path, "/@");
        if (p && (p[1] == '/' || p[1] == '@')) {
                /* two separators in a row */
                return (EINVAL);
        }
        if (p == NULL || p == path) {
                /*
                 * if the first thing is an @ or /, it had better be an
                 * @ and it had better not have any more ats or slashes,
                 * and it had better have something after the @.
                 */
                if (p != NULL &&
                    (p[0] != '@' || strpbrk(path+1, "/@") || p[1] == '\0'))
                        return (EINVAL);
                if (strlen(path) >= MAXNAMELEN)
                        return (ENAMETOOLONG);
                (void) strcpy(component, path);
                p = NULL;
        } else if (p[0] == '/') {
                if (p-path >= MAXNAMELEN)
                        return (ENAMETOOLONG);
                (void) strncpy(component, path, p - path);
                component[p-path] = '\0';
                p++;
        } else if (p[0] == '@') {
                /*
                 * if the next separator is an @, there better not be
                 * any more slashes.
                 */
                if (strchr(path, '/'))
                        return (EINVAL);
                if (p-path >= MAXNAMELEN)
                        return (ENAMETOOLONG);
                (void) strncpy(component, path, p - path);
                component[p-path] = '\0';
        } else {
                ASSERT(!"invalid p");
        }
        *nextp = p;
        return (0);
}

int
dsl_dir_open_spa(spa_t *spa, const char *name, void *tag,
    dsl_dir_t **ddp, const char **tailp)
{
        char buf[MAXNAMELEN];
        const char *next, *nextnext = NULL;
        int err;
        dsl_dir_t *dd;
        dsl_pool_t *dp;
        uint64_t ddobj;
        int openedspa = FALSE;

        dprintf("%s\n", name);

        err = getcomponent(name, buf, &next);
        if (err)
                return (err);
        if (spa == NULL) {
                err = spa_open(buf, &spa, FTAG);
                if (err) {
                        dprintf("spa_open(%s) failed\n", buf);
                        return (err);
                }
                openedspa = TRUE;

                /* XXX this assertion belongs in spa_open */
                ASSERT(!dsl_pool_sync_context(spa_get_dsl(spa)));
        }

        dp = spa_get_dsl(spa);

        rw_enter(&dp->dp_config_rwlock, RW_READER);
        err = dsl_dir_open_obj(dp, dp->dp_root_dir_obj, NULL, tag, &dd);
        if (err) {
                rw_exit(&dp->dp_config_rwlock);
                if (openedspa)
                        spa_close(spa, FTAG);
                return (err);
        }

        while (next != NULL) {
                dsl_dir_t *child_ds;
                err = getcomponent(next, buf, &nextnext);
                if (err)
                        break;
                ASSERT(next[0] != '\0');
                if (next[0] == '@')
                        break;
                dprintf("looking up %s in obj%lld\n",
                    buf, dd->dd_phys->dd_child_dir_zapobj);

                err = zap_lookup(dp->dp_meta_objset,
                    dd->dd_phys->dd_child_dir_zapobj,
                    buf, sizeof (ddobj), 1, &ddobj);
                if (err) {
                        if (err == ENOENT)
                                err = 0;
                        break;
                }

                err = dsl_dir_open_obj(dp, ddobj, buf, tag, &child_ds);
                if (err)
                        break;
                dsl_dir_close(dd, tag);
                dd = child_ds;
                next = nextnext;
        }
        rw_exit(&dp->dp_config_rwlock);

        if (err) {
                dsl_dir_close(dd, tag);
                if (openedspa)
                        spa_close(spa, FTAG);
                return (err);
        }

        /*
         * It's an error if there's more than one component left, or
         * tailp==NULL and there's any component left.
         */
        if (next != NULL &&
            (tailp == NULL || (nextnext && nextnext[0] != '\0'))) {
                /* bad path name */
                dsl_dir_close(dd, tag);
                dprintf("next=%p (%s) tail=%p\n", next, next?next:"", tailp);
                err = ENOENT;
        }
        if (tailp)
                *tailp = next;
        if (openedspa)
                spa_close(spa, FTAG);
        *ddp = dd;
        return (err);
}

int
dsl_dir_open(const char *name, void *tag, dsl_dir_t **ddp, const char **tailp)
{
        return (dsl_dir_open_spa(NULL, name, tag, ddp, tailp));
}

uint64_t
dsl_dir_create_sync(dsl_pool_t *dp, dsl_dir_t *pds, const char *name,
    dmu_tx_t *tx)
{
        objset_t *mos = dp->dp_meta_objset;
        uint64_t ddobj;
        dsl_dir_phys_t *ddphys;
        dmu_buf_t *dbuf;

        ddobj = dmu_object_alloc(mos, DMU_OT_DSL_DIR, 0,
            DMU_OT_DSL_DIR, sizeof (dsl_dir_phys_t), tx);
        if (pds) {
                VERIFY(0 == zap_add(mos, pds->dd_phys->dd_child_dir_zapobj,
                    name, sizeof (uint64_t), 1, &ddobj, tx));
        } else {
                /* it's the root dir */
                VERIFY(0 == zap_add(mos, DMU_POOL_DIRECTORY_OBJECT,
                    DMU_POOL_ROOT_DATASET, sizeof (uint64_t), 1, &ddobj, tx));
        }
        VERIFY(0 == dmu_bonus_hold(mos, ddobj, FTAG, &dbuf));
        dmu_buf_will_dirty(dbuf, tx);
        ddphys = dbuf->db_data;

        ddphys->dd_creation_time = gethrestime_sec();
        if (pds)
                ddphys->dd_parent_obj = pds->dd_object;
        ddphys->dd_props_zapobj = zap_create(mos,
            DMU_OT_DSL_PROPS, DMU_OT_NONE, 0, tx);
        ddphys->dd_child_dir_zapobj = zap_create(mos,
            DMU_OT_DSL_DIR_CHILD_MAP, DMU_OT_NONE, 0, tx);
        if (spa_version(dp->dp_spa) >= SPA_VERSION_USED_BREAKDOWN)
                ddphys->dd_flags |= DD_FLAG_USED_BREAKDOWN;
        dmu_buf_rele(dbuf, FTAG);

        return (ddobj);
}

/* ARGSUSED */
int
dsl_dir_destroy_check(void *arg1, void *arg2, dmu_tx_t *tx)
{
        dsl_dataset_t *ds = arg1;
        dsl_dir_t *dd = ds->ds_dir;
        dsl_pool_t *dp = dd->dd_pool;
        objset_t *mos = dp->dp_meta_objset;
        int err;
        uint64_t count;

        /*
         * There should be exactly two holds, both from
         * dsl_dataset_destroy: one on the dd directory, and one on its
         * head ds.  If there are more holds, then a concurrent thread is
         * performing a lookup inside this dir while we're trying to destroy
         * it.  To minimize this possibility, we perform this check only
         * in syncing context and fail the operation if we encounter
         * additional holds.  The dp_config_rwlock ensures that nobody else
         * opens it after we check.
         */
        if (dmu_tx_is_syncing(tx) && dmu_buf_refcount(dd->dd_dbuf) > 2)
                return (EBUSY);

        err = zap_count(mos, dd->dd_phys->dd_child_dir_zapobj, &count);
        if (err)
                return (err);
        if (count != 0)
                return (EEXIST);

        return (0);
}

void
dsl_dir_destroy_sync(void *arg1, void *tag, dmu_tx_t *tx)
{
        dsl_dataset_t *ds = arg1;
        dsl_dir_t *dd = ds->ds_dir;
        objset_t *mos = dd->dd_pool->dp_meta_objset;
        dsl_prop_setarg_t psa;
        uint64_t value = 0;
        uint64_t obj;
        dd_used_t t;

        ASSERT(RW_WRITE_HELD(&dd->dd_pool->dp_config_rwlock));
        ASSERT(dd->dd_phys->dd_head_dataset_obj == 0);

        /* Remove our reservation. */
        dsl_prop_setarg_init_uint64(&psa, "reservation",
            (ZPROP_SRC_NONE | ZPROP_SRC_LOCAL | ZPROP_SRC_RECEIVED),
            &value);
        psa.psa_effective_value = 0;    /* predict default value */

        dsl_dir_set_reservation_sync(ds, &psa, tx);

        ASSERT0(dd->dd_phys->dd_used_bytes);
        ASSERT0(dd->dd_phys->dd_reserved);
        for (t = 0; t < DD_USED_NUM; t++)
                ASSERT0(dd->dd_phys->dd_used_breakdown[t]);

        VERIFY(0 == zap_destroy(mos, dd->dd_phys->dd_child_dir_zapobj, tx));
        VERIFY(0 == zap_destroy(mos, dd->dd_phys->dd_props_zapobj, tx));
        VERIFY(0 == dsl_deleg_destroy(mos, dd->dd_phys->dd_deleg_zapobj, tx));
        VERIFY(0 == zap_remove(mos,
            dd->dd_parent->dd_phys->dd_child_dir_zapobj, dd->dd_myname, tx));

        obj = dd->dd_object;
        dsl_dir_close(dd, tag);
        VERIFY(0 == dmu_object_free(mos, obj, tx));
}

boolean_t
dsl_dir_is_clone(dsl_dir_t *dd)
{
        return (dd->dd_phys->dd_origin_obj &&
            (dd->dd_pool->dp_origin_snap == NULL ||
            dd->dd_phys->dd_origin_obj !=
            dd->dd_pool->dp_origin_snap->ds_object));
}

void
dsl_dir_stats(dsl_dir_t *dd, nvlist_t *nv)
{
        mutex_enter(&dd->dd_lock);
        dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_USED,
            dd->dd_phys->dd_used_bytes);
        dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_QUOTA, dd->dd_phys->dd_quota);
        dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_RESERVATION,
            dd->dd_phys->dd_reserved);
        dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_COMPRESSRATIO,
            dd->dd_phys->dd_compressed_bytes == 0 ? 100 :
            (dd->dd_phys->dd_uncompressed_bytes * 100 /
            dd->dd_phys->dd_compressed_bytes));
        if (dd->dd_phys->dd_flags & DD_FLAG_USED_BREAKDOWN) {
                dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_USEDSNAP,
                    dd->dd_phys->dd_used_breakdown[DD_USED_SNAP]);
                dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_USEDDS,
                    dd->dd_phys->dd_used_breakdown[DD_USED_HEAD]);
                dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_USEDREFRESERV,
                    dd->dd_phys->dd_used_breakdown[DD_USED_REFRSRV]);
                dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_USEDCHILD,
                    dd->dd_phys->dd_used_breakdown[DD_USED_CHILD] +
                    dd->dd_phys->dd_used_breakdown[DD_USED_CHILD_RSRV]);
        }
        mutex_exit(&dd->dd_lock);

        rw_enter(&dd->dd_pool->dp_config_rwlock, RW_READER);
        if (dsl_dir_is_clone(dd)) {
                dsl_dataset_t *ds;
                char buf[MAXNAMELEN];

                VERIFY(0 == dsl_dataset_hold_obj(dd->dd_pool,
                    dd->dd_phys->dd_origin_obj, FTAG, &ds));
                dsl_dataset_name(ds, buf);
                dsl_dataset_rele(ds, FTAG);
                dsl_prop_nvlist_add_string(nv, ZFS_PROP_ORIGIN, buf);
        }
        rw_exit(&dd->dd_pool->dp_config_rwlock);
}

void
dsl_dir_dirty(dsl_dir_t *dd, dmu_tx_t *tx)
{
        dsl_pool_t *dp = dd->dd_pool;

        ASSERT(dd->dd_phys);

        if (txg_list_add(&dp->dp_dirty_dirs, dd, tx->tx_txg) == 0) {
                /* up the hold count until we can be written out */
                dmu_buf_add_ref(dd->dd_dbuf, dd);
        }
}

static int64_t
parent_delta(dsl_dir_t *dd, uint64_t used, int64_t delta)
{
        uint64_t old_accounted = MAX(used, dd->dd_phys->dd_reserved);
        uint64_t new_accounted = MAX(used + delta, dd->dd_phys->dd_reserved);
        return (new_accounted - old_accounted);
}

void
dsl_dir_sync(dsl_dir_t *dd, dmu_tx_t *tx)
{
        ASSERT(dmu_tx_is_syncing(tx));

        mutex_enter(&dd->dd_lock);
        ASSERT0(dd->dd_tempreserved[tx->tx_txg&TXG_MASK]);
        dprintf_dd(dd, "txg=%llu towrite=%lluK\n", tx->tx_txg,
            dd->dd_space_towrite[tx->tx_txg&TXG_MASK] / 1024);
        dd->dd_space_towrite[tx->tx_txg&TXG_MASK] = 0;
        mutex_exit(&dd->dd_lock);

        /* release the hold from dsl_dir_dirty */
        dmu_buf_rele(dd->dd_dbuf, dd);
}

static uint64_t
dsl_dir_space_towrite(dsl_dir_t *dd)
{
        uint64_t space = 0;
        int i;

        ASSERT(MUTEX_HELD(&dd->dd_lock));

        for (i = 0; i < TXG_SIZE; i++) {
                space += dd->dd_space_towrite[i&TXG_MASK];
                ASSERT3U(dd->dd_space_towrite[i&TXG_MASK], >=, 0);
        }
        return (space);
}

uint64_t
dsl_dir_space_available(dsl_dir_t *dd,
    dsl_dir_t *ancestor, int64_t delta, int ondiskonly)
{
        uint64_t parentspace, myspace, quota, used;

        /*
         * If there are no restrictions otherwise, assume we have
         * unlimited space available.
         */
        quota = UINT64_MAX;
        parentspace = UINT64_MAX;

        if (dd->dd_parent != NULL) {
                parentspace = dsl_dir_space_available(dd->dd_parent,
                    ancestor, delta, ondiskonly);
        }

        mutex_enter(&dd->dd_lock);
        if (dd->dd_phys->dd_quota != 0)
                quota = dd->dd_phys->dd_quota;
        used = dd->dd_phys->dd_used_bytes;
        if (!ondiskonly)
                used += dsl_dir_space_towrite(dd);

        if (dd->dd_parent == NULL) {
                uint64_t poolsize = dsl_pool_adjustedsize(dd->dd_pool, FALSE);
                quota = MIN(quota, poolsize);
        }

        if (dd->dd_phys->dd_reserved > used && parentspace != UINT64_MAX) {
                /*
                 * We have some space reserved, in addition to what our
                 * parent gave us.
                 */
                parentspace += dd->dd_phys->dd_reserved - used;
        }

        if (dd == ancestor) {
                ASSERT(delta <= 0);
                ASSERT(used >= -delta);
                used += delta;
                if (parentspace != UINT64_MAX)
                        parentspace -= delta;
        }

        if (used > quota) {
                /* over quota */
                myspace = 0;
        } else {
                /*
                 * the lesser of the space provided by our parent and
                 * the space left in our quota
                 */
                myspace = MIN(parentspace, quota - used);
        }

        mutex_exit(&dd->dd_lock);

        return (myspace);
}

struct tempreserve {
        list_node_t tr_node;
        dsl_pool_t *tr_dp;
        dsl_dir_t *tr_ds;
        uint64_t tr_size;
};

static int
dsl_dir_tempreserve_impl(dsl_dir_t *dd, uint64_t asize, boolean_t netfree,
    boolean_t ignorequota, boolean_t checkrefquota, list_t *tr_list,
    dmu_tx_t *tx, boolean_t first)
{
        uint64_t txg = tx->tx_txg;
        uint64_t est_inflight, used_on_disk, quota, parent_rsrv;
        uint64_t deferred = 0;
        struct tempreserve *tr;
        int retval = EDQUOT;
        int txgidx = txg & TXG_MASK;
        int i;
        uint64_t ref_rsrv = 0;

        ASSERT3U(txg, !=, 0);
        ASSERT3S(asize, >, 0);

        mutex_enter(&dd->dd_lock);

        /*
         * Check against the dsl_dir's quota.  We don't add in the delta
         * when checking for over-quota because they get one free hit.
         */
        est_inflight = dsl_dir_space_towrite(dd);
        for (i = 0; i < TXG_SIZE; i++)
                est_inflight += dd->dd_tempreserved[i];
        used_on_disk = dd->dd_phys->dd_used_bytes;

        /*
         * On the first iteration, fetch the dataset's used-on-disk and
         * refreservation values. Also, if checkrefquota is set, test if
         * allocating this space would exceed the dataset's refquota.
         */
        if (first && tx->tx_objset) {
                int error;
                dsl_dataset_t *ds = tx->tx_objset->os_dsl_dataset;

                error = dsl_dataset_check_quota(ds, checkrefquota,
                    asize, est_inflight, &used_on_disk, &ref_rsrv);
                if (error) {
                        mutex_exit(&dd->dd_lock);
                        return (error);
                }
        }

        /*
         * If this transaction will result in a net free of space,
         * we want to let it through.
         */
        if (ignorequota || netfree || dd->dd_phys->dd_quota == 0)
                quota = UINT64_MAX;
        else
                quota = dd->dd_phys->dd_quota;

        /*
         * Adjust the quota against the actual pool size at the root
         * minus any outstanding deferred frees.
         * To ensure that it's possible to remove files from a full
         * pool without inducing transient overcommits, we throttle
         * netfree transactions against a quota that is slightly larger,
         * but still within the pool's allocation slop.  In cases where
         * we're very close to full, this will allow a steady trickle of
         * removes to get through.
         */
        if (dd->dd_parent == NULL) {
                spa_t *spa = dd->dd_pool->dp_spa;
                uint64_t poolsize = dsl_pool_adjustedsize(dd->dd_pool, netfree);
                deferred = metaslab_class_get_deferred(spa_normal_class(spa));
                if (poolsize - deferred < quota) {
                        quota = poolsize - deferred;
                        retval = ENOSPC;
                }
        }

        /*
         * If they are requesting more space, and our current estimate
         * is over quota, they get to try again unless the actual
         * on-disk is over quota and there are no pending changes (which
         * may free up space for us).
         */
        if (used_on_disk + est_inflight >= quota) {
                if (est_inflight > 0 || used_on_disk < quota ||
                    (retval == ENOSPC && used_on_disk < quota + deferred))
                        retval = ERESTART;
                dprintf_dd(dd, "failing: used=%lluK inflight = %lluK "
                    "quota=%lluK tr=%lluK err=%d\n",
                    used_on_disk>>10, est_inflight>>10,
                    quota>>10, asize>>10, retval);
                mutex_exit(&dd->dd_lock);
                return (retval);
        }

        /* We need to up our estimated delta before dropping dd_lock */
        dd->dd_tempreserved[txgidx] += asize;

        parent_rsrv = parent_delta(dd, used_on_disk + est_inflight,
            asize - ref_rsrv);
        mutex_exit(&dd->dd_lock);

        tr = kmem_zalloc(sizeof (struct tempreserve), KM_SLEEP);
        tr->tr_ds = dd;
        tr->tr_size = asize;
        list_insert_tail(tr_list, tr);

        /* see if it's OK with our parent */
        if (dd->dd_parent && parent_rsrv) {
                boolean_t ismos = (dd->dd_phys->dd_head_dataset_obj == 0);

                return (dsl_dir_tempreserve_impl(dd->dd_parent,
                    parent_rsrv, netfree, ismos, TRUE, tr_list, tx, FALSE));
        } else {
                return (0);
        }
}

int
dsl_dir_tempreserve_space(dsl_dir_t *dd, uint64_t lsize, uint64_t asize,
    uint64_t fsize, uint64_t usize, void **tr_cookiep, dmu_tx_t *tx)
{
        int err;
        list_t *tr_list;

        if (asize == 0) {
                *tr_cookiep = NULL;
                return (0);
        }

        tr_list = kmem_alloc(sizeof (list_t), KM_SLEEP);
        list_create(tr_list, sizeof (struct tempreserve),
            offsetof(struct tempreserve, tr_node));
        ASSERT3S(asize, >, 0);
        ASSERT3S(fsize, >=, 0);

        err = arc_tempreserve_space(lsize, tx->tx_txg);
        if (err == 0) {
                struct tempreserve *tr;

                tr = kmem_zalloc(sizeof (struct tempreserve), KM_SLEEP);
                tr->tr_size = lsize;
                list_insert_tail(tr_list, tr);

                err = dsl_pool_tempreserve_space(dd->dd_pool, asize, tx);
        } else {
                if (err == EAGAIN) {
                        txg_delay(dd->dd_pool, tx->tx_txg, 1);
                        err = ERESTART;
                }
                dsl_pool_memory_pressure(dd->dd_pool);
        }

        if (err == 0) {
                struct tempreserve *tr;

                tr = kmem_zalloc(sizeof (struct tempreserve), KM_SLEEP);
                tr->tr_dp = dd->dd_pool;
                tr->tr_size = asize;
                list_insert_tail(tr_list, tr);

                err = dsl_dir_tempreserve_impl(dd, asize, fsize >= asize,
                    FALSE, asize > usize, tr_list, tx, TRUE);
        }

        if (err)
                dsl_dir_tempreserve_clear(tr_list, tx);
        else
                *tr_cookiep = tr_list;

        return (err);
}

void
dsl_dir_tempreserve_clear(void *tr_cookie, dmu_tx_t *tx)
{
        int txgidx = tx->tx_txg & TXG_MASK;
        list_t *tr_list = tr_cookie;
        struct tempreserve *tr;

        ASSERT3U(tx->tx_txg, !=, 0);

        if (tr_cookie == NULL)
                return;

        while (tr = list_head(tr_list)) {
                if (tr->tr_dp) {
                        dsl_pool_tempreserve_clear(tr->tr_dp, tr->tr_size, tx);
                } else if (tr->tr_ds) {
                        mutex_enter(&tr->tr_ds->dd_lock);
                        ASSERT3U(tr->tr_ds->dd_tempreserved[txgidx], >=,
                            tr->tr_size);
                        tr->tr_ds->dd_tempreserved[txgidx] -= tr->tr_size;
                        mutex_exit(&tr->tr_ds->dd_lock);
                } else {
                        arc_tempreserve_clear(tr->tr_size);
                }
                list_remove(tr_list, tr);
                kmem_free(tr, sizeof (struct tempreserve));
        }

        kmem_free(tr_list, sizeof (list_t));
}

static void
dsl_dir_willuse_space_impl(dsl_dir_t *dd, int64_t space, dmu_tx_t *tx)
{
        int64_t parent_space;
        uint64_t est_used;

        mutex_enter(&dd->dd_lock);
        if (space > 0)
                dd->dd_space_towrite[tx->tx_txg & TXG_MASK] += space;

        est_used = dsl_dir_space_towrite(dd) + dd->dd_phys->dd_used_bytes;
        parent_space = parent_delta(dd, est_used, space);
        mutex_exit(&dd->dd_lock);

        /* Make sure that we clean up dd_space_to* */
        dsl_dir_dirty(dd, tx);

        /* XXX this is potentially expensive and unnecessary... */
        if (parent_space && dd->dd_parent)
                dsl_dir_willuse_space_impl(dd->dd_parent, parent_space, tx);
}

void
dsl_dir_willuse_space(dsl_dir_t *dd, int64_t space, dmu_tx_t *tx)
{
        dsl_pool_willuse_space(dd->dd_pool, space, tx);
        dsl_dir_willuse_space_impl(dd, space, tx);
}

void
dsl_dir_diduse_space(dsl_dir_t *dd, dd_used_t type,
    int64_t used, int64_t compressed, int64_t uncompressed, dmu_tx_t *tx)
{
        int64_t accounted_delta;
        boolean_t needlock = !MUTEX_HELD(&dd->dd_lock);

        ASSERT(dmu_tx_is_syncing(tx));
        ASSERT(type < DD_USED_NUM);

        if (needlock)
                mutex_enter(&dd->dd_lock);
        accounted_delta = parent_delta(dd, dd->dd_phys->dd_used_bytes, used);
        ASSERT(used >= 0 || dd->dd_phys->dd_used_bytes >= -used);
        ASSERT(compressed >= 0 ||
            dd->dd_phys->dd_compressed_bytes >= -compressed);
        ASSERT(uncompressed >= 0 ||
            dd->dd_phys->dd_uncompressed_bytes >= -uncompressed);
        dmu_buf_will_dirty(dd->dd_dbuf, tx);
        dd->dd_phys->dd_used_bytes += used;
        dd->dd_phys->dd_uncompressed_bytes += uncompressed;
        dd->dd_phys->dd_compressed_bytes += compressed;

        if (dd->dd_phys->dd_flags & DD_FLAG_USED_BREAKDOWN) {
                ASSERT(used > 0 ||
                    dd->dd_phys->dd_used_breakdown[type] >= -used);
                dd->dd_phys->dd_used_breakdown[type] += used;
#ifdef DEBUG
                dd_used_t t;
                uint64_t u = 0;
                for (t = 0; t < DD_USED_NUM; t++)
                        u += dd->dd_phys->dd_used_breakdown[t];
                ASSERT3U(u, ==, dd->dd_phys->dd_used_bytes);
#endif
        }
        if (needlock)
                mutex_exit(&dd->dd_lock);

        if (dd->dd_parent != NULL) {
                dsl_dir_diduse_space(dd->dd_parent, DD_USED_CHILD,
                    accounted_delta, compressed, uncompressed, tx);
                dsl_dir_transfer_space(dd->dd_parent,
                    used - accounted_delta,
                    DD_USED_CHILD_RSRV, DD_USED_CHILD, tx);
        }
}

void
dsl_dir_transfer_space(dsl_dir_t *dd, int64_t delta,
    dd_used_t oldtype, dd_used_t newtype, dmu_tx_t *tx)
{
        boolean_t needlock = !MUTEX_HELD(&dd->dd_lock);

        ASSERT(dmu_tx_is_syncing(tx));
        ASSERT(oldtype < DD_USED_NUM);
        ASSERT(newtype < DD_USED_NUM);

        if (delta == 0 || !(dd->dd_phys->dd_flags & DD_FLAG_USED_BREAKDOWN))
                return;

        if (needlock)
                mutex_enter(&dd->dd_lock);
        ASSERT(delta > 0 ?
            dd->dd_phys->dd_used_breakdown[oldtype] >= delta :
            dd->dd_phys->dd_used_breakdown[newtype] >= -delta);
        ASSERT(dd->dd_phys->dd_used_bytes >= ABS(delta));
        dmu_buf_will_dirty(dd->dd_dbuf, tx);
        dd->dd_phys->dd_used_breakdown[oldtype] -= delta;
        dd->dd_phys->dd_used_breakdown[newtype] += delta;
        if (needlock)
                mutex_exit(&dd->dd_lock);
}

static int
dsl_dir_set_quota_check(void *arg1, void *arg2, dmu_tx_t *tx)
{
        dsl_dataset_t *ds = arg1;
        dsl_dir_t *dd = ds->ds_dir;
        dsl_prop_setarg_t *psa = arg2;
        int err;
        uint64_t towrite;

        if ((err = dsl_prop_predict_sync(ds->ds_dir, psa)) != 0)
                return (err);

        if (psa->psa_effective_value == 0)
                return (0);

        mutex_enter(&dd->dd_lock);
        /*
         * If we are doing the preliminary check in open context, and
         * there are pending changes, then don't fail it, since the
         * pending changes could under-estimate the amount of space to be
         * freed up.
         */
        towrite = dsl_dir_space_towrite(dd);
        if ((dmu_tx_is_syncing(tx) || towrite == 0) &&
            (psa->psa_effective_value < dd->dd_phys->dd_reserved ||
            psa->psa_effective_value < dd->dd_phys->dd_used_bytes + towrite)) {
                err = ENOSPC;
        }
        mutex_exit(&dd->dd_lock);
        return (err);
}

extern dsl_syncfunc_t dsl_prop_set_sync;

static void
dsl_dir_set_quota_sync(void *arg1, void *arg2, dmu_tx_t *tx)
{
        dsl_dataset_t *ds = arg1;
        dsl_dir_t *dd = ds->ds_dir;
        dsl_prop_setarg_t *psa = arg2;
        uint64_t effective_value = psa->psa_effective_value;

        dsl_prop_set_sync(ds, psa, tx);
        DSL_PROP_CHECK_PREDICTION(dd, psa);

        dmu_buf_will_dirty(dd->dd_dbuf, tx);

        mutex_enter(&dd->dd_lock);
        dd->dd_phys->dd_quota = effective_value;
        mutex_exit(&dd->dd_lock);
}

int
dsl_dir_set_quota(const char *ddname, zprop_source_t source, uint64_t quota)
{
        dsl_dir_t *dd;
        dsl_dataset_t *ds;
        dsl_prop_setarg_t psa;
        int err;

        dsl_prop_setarg_init_uint64(&psa, "quota", source, &quota);

        err = dsl_dataset_hold(ddname, FTAG, &ds);
        if (err)
                return (err);

        err = dsl_dir_open(ddname, FTAG, &dd, NULL);
        if (err) {
                dsl_dataset_rele(ds, FTAG);
                return (err);
        }

        ASSERT(ds->ds_dir == dd);

        /*
         * If someone removes a file, then tries to set the quota, we want to
         * make sure the file freeing takes effect.
         */
        txg_wait_open(dd->dd_pool, 0);

        err = dsl_sync_task_do(dd->dd_pool, dsl_dir_set_quota_check,
            dsl_dir_set_quota_sync, ds, &psa, 0);

        dsl_dir_close(dd, FTAG);
        dsl_dataset_rele(ds, FTAG);
        return (err);
}

int
dsl_dir_set_reservation_check(void *arg1, void *arg2, dmu_tx_t *tx)
{
        dsl_dataset_t *ds = arg1;
        dsl_dir_t *dd = ds->ds_dir;
        dsl_prop_setarg_t *psa = arg2;
        uint64_t effective_value;
        uint64_t used, avail;
        int err;

        if ((err = dsl_prop_predict_sync(ds->ds_dir, psa)) != 0)
                return (err);

        effective_value = psa->psa_effective_value;

        /*
         * If we are doing the preliminary check in open context, the
         * space estimates may be inaccurate.
         */
        if (!dmu_tx_is_syncing(tx))
                return (0);

        mutex_enter(&dd->dd_lock);
        used = dd->dd_phys->dd_used_bytes;
        mutex_exit(&dd->dd_lock);

        if (dd->dd_parent) {
                avail = dsl_dir_space_available(dd->dd_parent,
                    NULL, 0, FALSE);
        } else {
                avail = dsl_pool_adjustedsize(dd->dd_pool, B_FALSE) - used;
        }

        if (MAX(used, effective_value) > MAX(used, dd->dd_phys->dd_reserved)) {
                uint64_t delta = MAX(used, effective_value) -
                    MAX(used, dd->dd_phys->dd_reserved);

                if (delta > avail)
                        return (ENOSPC);
                if (dd->dd_phys->dd_quota > 0 &&
                    effective_value > dd->dd_phys->dd_quota)
                        return (ENOSPC);
        }

        return (0);
}

static void
dsl_dir_set_reservation_sync(void *arg1, void *arg2, dmu_tx_t *tx)
{
        dsl_dataset_t *ds = arg1;
        dsl_dir_t *dd = ds->ds_dir;
        dsl_prop_setarg_t *psa = arg2;
        uint64_t effective_value = psa->psa_effective_value;
        uint64_t used;
        int64_t delta;

        dsl_prop_set_sync(ds, psa, tx);
        DSL_PROP_CHECK_PREDICTION(dd, psa);

        dmu_buf_will_dirty(dd->dd_dbuf, tx);

        mutex_enter(&dd->dd_lock);
        used = dd->dd_phys->dd_used_bytes;
        delta = MAX(used, effective_value) -
            MAX(used, dd->dd_phys->dd_reserved);
        dd->dd_phys->dd_reserved = effective_value;

        if (dd->dd_parent != NULL) {
                /* Roll up this additional usage into our ancestors */
                dsl_dir_diduse_space(dd->dd_parent, DD_USED_CHILD_RSRV,
                    delta, 0, 0, tx);
        }
        mutex_exit(&dd->dd_lock);
}

int
dsl_dir_set_reservation(const char *ddname, zprop_source_t source,
    uint64_t reservation)
{
        dsl_dir_t *dd;
        dsl_dataset_t *ds;
        dsl_prop_setarg_t psa;
        int err;

        dsl_prop_setarg_init_uint64(&psa, "reservation", source, &reservation);

        err = dsl_dataset_hold(ddname, FTAG, &ds);
        if (err)
                return (err);

        err = dsl_dir_open(ddname, FTAG, &dd, NULL);
        if (err) {
                dsl_dataset_rele(ds, FTAG);
                return (err);
        }

        ASSERT(ds->ds_dir == dd);

        err = dsl_sync_task_do(dd->dd_pool, dsl_dir_set_reservation_check,
            dsl_dir_set_reservation_sync, ds, &psa, 0);

        dsl_dir_close(dd, FTAG);
        dsl_dataset_rele(ds, FTAG);
        return (err);
}

static dsl_dir_t *
closest_common_ancestor(dsl_dir_t *ds1, dsl_dir_t *ds2)
{
        for (; ds1; ds1 = ds1->dd_parent) {
                dsl_dir_t *dd;
                for (dd = ds2; dd; dd = dd->dd_parent) {
                        if (ds1 == dd)
                                return (dd);
                }
        }
        return (NULL);
}

static int64_t
would_change(dsl_dir_t *dd, int64_t delta, dsl_dir_t *ancestor)
{
        if (dd == ancestor)
                return (delta);

        mutex_enter(&dd->dd_lock);
        delta = parent_delta(dd, dd->dd_phys->dd_used_bytes, delta);
        mutex_exit(&dd->dd_lock);
        return (would_change(dd->dd_parent, delta, ancestor));
}

struct renamearg {
        dsl_dir_t *newparent;
        const char *mynewname;
        boolean_t allowmounted;
};

static int
dsl_dir_rename_check(void *arg1, void *arg2, dmu_tx_t *tx)
{
        dsl_dir_t *dd = arg1;
        struct renamearg *ra = arg2;
        dsl_pool_t *dp = dd->dd_pool;
        objset_t *mos = dp->dp_meta_objset;
        int err;
        uint64_t val;

        /*
         * There should only be one reference, from dmu_objset_rename().
         * Fleeting holds are also possible (eg, from "zfs list" getting
         * stats), but any that are present in open context will likely
         * be gone by syncing context, so only fail from syncing
         * context.
         * Don't check if we allow renaming of busy (mounted) dataset.
         */
        if (!ra->allowmounted && dmu_tx_is_syncing(tx) &&
            dmu_buf_refcount(dd->dd_dbuf) > 1) {
                return (EBUSY);
        }

        /* check for existing name */
        err = zap_lookup(mos, ra->newparent->dd_phys->dd_child_dir_zapobj,
            ra->mynewname, 8, 1, &val);
        if (err == 0)
                return (EEXIST);
        if (err != ENOENT)
                return (err);

        if (ra->newparent != dd->dd_parent) {
                /* is there enough space? */
                uint64_t myspace =
                    MAX(dd->dd_phys->dd_used_bytes, dd->dd_phys->dd_reserved);

                /* no rename into our descendant */
                if (closest_common_ancestor(dd, ra->newparent) == dd)
                        return (EINVAL);

                if (err = dsl_dir_transfer_possible(dd->dd_parent,
                    ra->newparent, myspace))
                        return (err);
        }

        return (0);
}

static void
dsl_dir_rename_sync(void *arg1, void *arg2, dmu_tx_t *tx)
{
        char oldname[MAXPATHLEN], newname[MAXPATHLEN];
        dsl_dir_t *dd = arg1;
        struct renamearg *ra = arg2;
        dsl_pool_t *dp = dd->dd_pool;
        objset_t *mos = dp->dp_meta_objset;
        int err;

        ASSERT(ra->allowmounted || dmu_buf_refcount(dd->dd_dbuf) <= 2);

        if (ra->newparent != dd->dd_parent) {
                dsl_dir_diduse_space(dd->dd_parent, DD_USED_CHILD,
                    -dd->dd_phys->dd_used_bytes,
                    -dd->dd_phys->dd_compressed_bytes,
                    -dd->dd_phys->dd_uncompressed_bytes, tx);
                dsl_dir_diduse_space(ra->newparent, DD_USED_CHILD,
                    dd->dd_phys->dd_used_bytes,
                    dd->dd_phys->dd_compressed_bytes,
                    dd->dd_phys->dd_uncompressed_bytes, tx);

                if (dd->dd_phys->dd_reserved > dd->dd_phys->dd_used_bytes) {
                        uint64_t unused_rsrv = dd->dd_phys->dd_reserved -
                            dd->dd_phys->dd_used_bytes;

                        dsl_dir_diduse_space(dd->dd_parent, DD_USED_CHILD_RSRV,
                            -unused_rsrv, 0, 0, tx);
                        dsl_dir_diduse_space(ra->newparent, DD_USED_CHILD_RSRV,
                            unused_rsrv, 0, 0, tx);
                }
        }

        dmu_buf_will_dirty(dd->dd_dbuf, tx);

        /* remove from old parent zapobj */
        dsl_dir_name(dd, oldname);
        err = zap_remove(mos, dd->dd_parent->dd_phys->dd_child_dir_zapobj,
            dd->dd_myname, tx);
        ASSERT0(err);

        (void) strcpy(dd->dd_myname, ra->mynewname);
        dsl_dir_close(dd->dd_parent, dd);
        dd->dd_phys->dd_parent_obj = ra->newparent->dd_object;
        VERIFY(0 == dsl_dir_open_obj(dd->dd_pool,
            ra->newparent->dd_object, NULL, dd, &dd->dd_parent));

        /* add to new parent zapobj */
        err = zap_add(mos, ra->newparent->dd_phys->dd_child_dir_zapobj,
            dd->dd_myname, 8, 1, &dd->dd_object, tx);
        ASSERT0(err);
        dsl_dir_name(dd, newname);
#ifdef _KERNEL
        zfsvfs_update_fromname(oldname, newname);
        zvol_rename_minors(oldname, newname);
#endif

        spa_history_log_internal(LOG_DS_RENAME, dd->dd_pool->dp_spa,
            tx, "dataset = %llu", dd->dd_phys->dd_head_dataset_obj);
}

int
dsl_dir_rename(dsl_dir_t *dd, const char *newname, int flags)
{
        struct renamearg ra;
        int err;

        /* new parent should exist */
        err = dsl_dir_open(newname, FTAG, &ra.newparent, &ra.mynewname);
        if (err)
                return (err);

        /* can't rename to different pool */
        if (dd->dd_pool != ra.newparent->dd_pool) {
                err = ENXIO;
                goto out;
        }

        /* new name should not already exist */
        if (ra.mynewname == NULL) {
                err = EEXIST;
                goto out;
        }

        ra.allowmounted = !!(flags & ZFS_RENAME_ALLOW_MOUNTED);

        err = dsl_sync_task_do(dd->dd_pool,
            dsl_dir_rename_check, dsl_dir_rename_sync, dd, &ra, 3);

out:
        dsl_dir_close(ra.newparent, FTAG);
        return (err);
}

int
dsl_dir_transfer_possible(dsl_dir_t *sdd, dsl_dir_t *tdd, uint64_t space)
{
        dsl_dir_t *ancestor;
        int64_t adelta;
        uint64_t avail;

        ancestor = closest_common_ancestor(sdd, tdd);
        adelta = would_change(sdd, -space, ancestor);
        avail = dsl_dir_space_available(tdd, ancestor, adelta, FALSE);
        if (avail < space)
                return (ENOSPC);

        return (0);
}

timestruc_t
dsl_dir_snap_cmtime(dsl_dir_t *dd)
{
        timestruc_t t;

        mutex_enter(&dd->dd_lock);
        t = dd->dd_snap_cmtime;
        mutex_exit(&dd->dd_lock);

        return (t);
}

void
dsl_dir_snap_cmtime_update(dsl_dir_t *dd)
{
        timestruc_t t;

        gethrestime(&t);
        mutex_enter(&dd->dd_lock);
        dd->dd_snap_cmtime = t;
        mutex_exit(&dd->dd_lock);
}