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 *dsphys;
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);
dsphys = dbuf->db_data;
dsphys->dd_creation_time = gethrestime_sec();
if (pds)
dsphys->dd_parent_obj = pds->dd_object;
dsphys->dd_props_zapobj = zap_create(mos,
DMU_OT_DSL_PROPS, DMU_OT_NONE, 0, tx);
dsphys->dd_child_dir_zapobj = zap_create(mos,
DMU_OT_DSL_DIR_CHILD_MAP, DMU_OT_NONE, 0, tx);
dmu_buf_rele(dbuf, FTAG);
return (ddobj);
}
void
spa_history_create_obj(spa_t *spa, dmu_tx_t *tx)
{
dmu_buf_t *dbp;
spa_history_phys_t *shpp;
objset_t *mos = spa->spa_meta_objset;
ASSERT(spa->spa_history == 0);
spa->spa_history = dmu_object_alloc(mos, DMU_OT_SPA_HISTORY,
SPA_MAXBLOCKSIZE, DMU_OT_SPA_HISTORY_OFFSETS,
sizeof (spa_history_phys_t), tx);
VERIFY(zap_add(mos, DMU_POOL_DIRECTORY_OBJECT,
DMU_POOL_HISTORY, sizeof (uint64_t), 1,
&spa->spa_history, tx) == 0);
VERIFY(0 == dmu_bonus_hold(mos, spa->spa_history, FTAG, &dbp));
ASSERT(dbp->db_size >= sizeof (spa_history_phys_t));
shpp = dbp->db_data;
dmu_buf_will_dirty(dbp, tx);
/*
* Figure out maximum size of history log. We set it at
* 0.1% of pool size, with a max of 1G and min of 128KB.
*/
shpp->sh_phys_max_off =
metaslab_class_get_dspace(spa_normal_class(spa)) / 1000;
shpp->sh_phys_max_off = MIN(shpp->sh_phys_max_off, 1<<30);
shpp->sh_phys_max_off = MAX(shpp->sh_phys_max_off, 128<<10);
dmu_buf_rele(dbp, FTAG);
}
/*
* load initial fuid domain and idx trees. This function is used by
* both the kernel and zdb.
*/
uint64_t
zfs_fuid_table_load(objset_t *os, uint64_t fuid_obj, avl_tree_t *idx_tree,
avl_tree_t *domain_tree)
{
dmu_buf_t *db;
uint64_t fuid_size;
avl_create(idx_tree, idx_compare,
sizeof (fuid_domain_t), offsetof(fuid_domain_t, f_idxnode));
avl_create(domain_tree, domain_compare,
sizeof (fuid_domain_t), offsetof(fuid_domain_t, f_domnode));
VERIFY(0 == dmu_bonus_hold(os, fuid_obj, FTAG, &db));
fuid_size = *(uint64_t *)db->db_data;
dmu_buf_rele(db, FTAG);
if (fuid_size) {
nvlist_t **fuidnvp;
nvlist_t *nvp = NULL;
uint_t count;
char *packed;
int i;
packed = kmem_alloc(fuid_size, KM_SLEEP);
VERIFY(dmu_read(os, fuid_obj, 0, fuid_size, packed) == 0);
VERIFY(nvlist_unpack(packed, fuid_size,
&nvp, 0) == 0);
VERIFY(nvlist_lookup_nvlist_array(nvp, FUID_NVP_ARRAY,
&fuidnvp, &count) == 0);
for (i = 0; i != count; i++) {
fuid_domain_t *domnode;
char *domain;
uint64_t idx;
VERIFY(nvlist_lookup_string(fuidnvp[i], FUID_DOMAIN,
&domain) == 0);
VERIFY(nvlist_lookup_uint64(fuidnvp[i], FUID_IDX,
&idx) == 0);
domnode = kmem_alloc(sizeof (fuid_domain_t), KM_SLEEP);
domnode->f_idx = idx;
domnode->f_ksid = ksid_lookupdomain(domain);
avl_add(idx_tree, domnode);
avl_add(domain_tree, domnode);
}
nvlist_free(nvp);
kmem_free(packed, fuid_size);
}
return (fuid_size);
}
int
sa_handle_get(objset_t *objset, uint64_t objid, void *userp,
sa_handle_type_t hdl_type, sa_handle_t **handlepp)
{
dmu_buf_t *db;
int error;
if (error = dmu_bonus_hold(objset, objid, NULL, &db))
return (error);
return (sa_handle_get_from_db(objset, db, userp, hdl_type,
handlepp));
}
static int
bplist_hold(bplist_t *bpl)
{
ASSERT(MUTEX_HELD(&bpl->bpl_lock));
if (bpl->bpl_dbuf == NULL) {
int err = dmu_bonus_hold(bpl->bpl_mos,
bpl->bpl_object, bpl, &bpl->bpl_dbuf);
if (err)
return (err);
bpl->bpl_phys = bpl->bpl_dbuf->db_data;
}
return (0);
}
int
vdev_metaslab_init(vdev_t *vd, uint64_t txg)
{
spa_t *spa = vd->vdev_spa;
objset_t *mos = spa->spa_meta_objset;
metaslab_class_t *mc = spa_metaslab_class_select(spa);
uint64_t m;
uint64_t oldc = vd->vdev_ms_count;
uint64_t newc = vd->vdev_asize >> vd->vdev_ms_shift;
metaslab_t **mspp;
int error;
if (vd->vdev_ms_shift == 0) /* not being allocated from yet */
return (0);
dprintf("%s oldc %llu newc %llu\n", vdev_description(vd), oldc, newc);
ASSERT(oldc <= newc);
if (vd->vdev_mg == NULL)
vd->vdev_mg = metaslab_group_create(mc, vd);
mspp = kmem_zalloc(newc * sizeof (*mspp), KM_SLEEP);
if (oldc != 0) {
bcopy(vd->vdev_ms, mspp, oldc * sizeof (*mspp));
kmem_free(vd->vdev_ms, oldc * sizeof (*mspp));
}
vd->vdev_ms = mspp;
vd->vdev_ms_count = newc;
for (m = oldc; m < newc; m++) {
space_map_obj_t smo = { 0, 0, 0 };
if (txg == 0) {
uint64_t object = 0;
error = dmu_read(mos, vd->vdev_ms_array,
m * sizeof (uint64_t), sizeof (uint64_t), &object);
if (error)
return (error);
if (object != 0) {
dmu_buf_t *db;
error = dmu_bonus_hold(mos, object, FTAG, &db);
if (error)
return (error);
ASSERT3U(db->db_size, ==, sizeof (smo));
bcopy(db->db_data, &smo, db->db_size);
ASSERT3U(smo.smo_object, ==, object);
dmu_buf_rele(db, FTAG);
}
}
static int
zvol_first_open(zvol_state_t *zv)
{
objset_t *os;
uint64_t volsize;
int error;
uint64_t ro;
/* lie and say we're read-only */
error = dmu_objset_own(zv->zv_name, DMU_OST_ZVOL, 1, zvol_tag, &os);
if (error)
return (SET_ERROR(-error));
zv->zv_objset = os;
error = dsl_prop_get_integer(zv->zv_name, "readonly", &ro, NULL);
if (error)
goto out_owned;
error = zap_lookup(os, ZVOL_ZAP_OBJ, "size", 8, 1, &volsize);
if (error)
goto out_owned;
error = dmu_bonus_hold(os, ZVOL_OBJ, zvol_tag, &zv->zv_dbuf);
if (error)
goto out_owned;
set_capacity(zv->zv_disk, volsize >> 9);
zv->zv_volsize = volsize;
zv->zv_zilog = zil_open(os, zvol_get_data);
if (ro || dmu_objset_is_snapshot(os) ||
!spa_writeable(dmu_objset_spa(os))) {
set_disk_ro(zv->zv_disk, 1);
zv->zv_flags |= ZVOL_RDONLY;
} else {
set_disk_ro(zv->zv_disk, 0);
zv->zv_flags &= ~ZVOL_RDONLY;
}
out_owned:
if (error) {
dmu_objset_disown(os, zvol_tag);
zv->zv_objset = NULL;
}
return (SET_ERROR(-error));
}
int
dmu_objset_userspace_upgrade(objset_t *os)
{
uint64_t obj;
int err = 0;
if (dmu_objset_userspace_present(os))
return (0);
if (!dmu_objset_userused_enabled(os->os))
return (ENOTSUP);
if (dmu_objset_is_snapshot(os))
return (EINVAL);
/*
* We simply need to mark every object dirty, so that it will be
* synced out and now accounted. If this is called
* concurrently, or if we already did some work before crashing,
* that's fine, since we track each object's accounted state
* independently.
*/
for (obj = 0; err == 0; err = dmu_object_next(os, &obj, FALSE, 0)) {
dmu_tx_t *tx;
dmu_buf_t *db;
int objerr;
if (issig(JUSTLOOKING) && issig(FORREAL))
return (EINTR);
objerr = dmu_bonus_hold(os, obj, FTAG, &db);
if (objerr)
continue;
tx = dmu_tx_create(os);
dmu_tx_hold_bonus(tx, obj);
objerr = dmu_tx_assign(tx, TXG_WAIT);
if (objerr) {
dmu_tx_abort(tx);
continue;
}
dmu_buf_will_dirty(db, tx);
dmu_buf_rele(db, FTAG);
dmu_tx_commit(tx);
}
os->os->os_flags |= OBJSET_FLAG_USERACCOUNTING_COMPLETE;
txg_wait_synced(dmu_objset_pool(os), 0);
return (0);
}
vdev_indirect_births_t *
vdev_indirect_births_open(objset_t *os, uint64_t births_object)
{
vdev_indirect_births_t *vib = kmem_zalloc(sizeof (*vib), KM_SLEEP);
vib->vib_objset = os;
vib->vib_object = births_object;
VERIFY0(dmu_bonus_hold(os, vib->vib_object, vib, &vib->vib_dbuf));
vib->vib_phys = vib->vib_dbuf->db_data;
if (vib->vib_phys->vib_count > 0) {
uint64_t births_size = vdev_indirect_births_size_impl(vib);
vib->vib_entries = kmem_alloc(births_size, KM_SLEEP);
VERIFY0(dmu_read(vib->vib_objset, vib->vib_object, 0,
births_size, vib->vib_entries, DMU_READ_PREFETCH));
}
ASSERT(vdev_indirect_births_verify(vib));
return (vib);
}
/*
* Read out the command history.
*/
int
spa_history_get(spa_t *spa, uint64_t *offp, uint64_t *len, char *buf)
{
objset_t *mos = spa->spa_meta_objset;
dmu_buf_t *dbp;
uint64_t read_len, phys_read_off, phys_eof;
uint64_t leftover = 0;
spa_history_phys_t *shpp;
int err;
/*
* If the command history doesn't exist (older pool),
* that's ok, just return ENOENT.
*/
if (!spa->spa_history)
return (ENOENT);
/*
* The history is logged asynchronously, so when they request
* the first chunk of history, make sure everything has been
* synced to disk so that we get it.
*/
if (*offp == 0 && spa_writeable(spa))
txg_wait_synced(spa_get_dsl(spa), 0);
if ((err = dmu_bonus_hold(mos, spa->spa_history, FTAG, &dbp)) != 0)
return (err);
shpp = dbp->db_data;
#ifdef ZFS_DEBUG
{
dmu_object_info_t doi;
dmu_object_info_from_db(dbp, &doi);
ASSERT3U(doi.doi_bonus_type, ==, DMU_OT_SPA_HISTORY_OFFSETS);
}
#endif
mutex_enter(&spa->spa_history_lock);
phys_eof = spa_history_log_to_phys(shpp->sh_eof, shpp);
if (*offp < shpp->sh_pool_create_len) {
/* read in just the zpool create history */
phys_read_off = *offp;
read_len = MIN(*len, shpp->sh_pool_create_len -
phys_read_off);
} else {
/*
* Need to reset passed in offset to BOF if the passed in
* offset has since been overwritten.
*/
*offp = MAX(*offp, shpp->sh_bof);
phys_read_off = spa_history_log_to_phys(*offp, shpp);
/*
* Read up to the minimum of what the user passed down or
* the EOF (physical or logical). If we hit physical EOF,
* use 'leftover' to read from the physical BOF.
*/
if (phys_read_off <= phys_eof) {
read_len = MIN(*len, phys_eof - phys_read_off);
} else {
read_len = MIN(*len,
shpp->sh_phys_max_off - phys_read_off);
if (phys_read_off + *len > shpp->sh_phys_max_off) {
leftover = MIN(*len - read_len,
phys_eof - shpp->sh_pool_create_len);
}
}
}
/* offset for consumer to use next */
*offp += read_len + leftover;
/* tell the consumer how much you actually read */
*len = read_len + leftover;
if (read_len == 0) {
mutex_exit(&spa->spa_history_lock);
dmu_buf_rele(dbp, FTAG);
return (0);
}
err = dmu_read(mos, spa->spa_history, phys_read_off, read_len, buf,
DMU_READ_PREFETCH);
if (leftover && err == 0) {
err = dmu_read(mos, spa->spa_history, shpp->sh_pool_create_len,
leftover, buf + read_len, DMU_READ_PREFETCH);
}
mutex_exit(&spa->spa_history_lock);
dmu_buf_rele(dbp, FTAG);
return (err);
}
/*ARGSUSED*/
static void
spa_history_log_sync(void *arg1, void *arg2, dmu_tx_t *tx)
{
spa_t *spa = arg1;
history_arg_t *hap = arg2;
const char *history_str = hap->ha_history_str;
objset_t *mos = spa->spa_meta_objset;
dmu_buf_t *dbp;
spa_history_phys_t *shpp;
size_t reclen;
uint64_t le_len;
nvlist_t *nvrecord;
char *record_packed = NULL;
int ret;
/*
* If we have an older pool that doesn't have a command
* history object, create it now.
*/
mutex_enter(&spa->spa_history_lock);
if (!spa->spa_history)
spa_history_create_obj(spa, tx);
mutex_exit(&spa->spa_history_lock);
/*
* Get the offset of where we need to write via the bonus buffer.
* Update the offset when the write completes.
*/
VERIFY(0 == dmu_bonus_hold(mos, spa->spa_history, FTAG, &dbp));
shpp = dbp->db_data;
dmu_buf_will_dirty(dbp, tx);
#ifdef ZFS_DEBUG
{
dmu_object_info_t doi;
dmu_object_info_from_db(dbp, &doi);
ASSERT3U(doi.doi_bonus_type, ==, DMU_OT_SPA_HISTORY_OFFSETS);
}
#endif
VERIFY(nvlist_alloc(&nvrecord, NV_UNIQUE_NAME, KM_SLEEP) == 0);
VERIFY(nvlist_add_uint64(nvrecord, ZPOOL_HIST_TIME,
gethrestime_sec()) == 0);
VERIFY(nvlist_add_uint64(nvrecord, ZPOOL_HIST_WHO, hap->ha_uid) == 0);
if (hap->ha_zone != NULL)
VERIFY(nvlist_add_string(nvrecord, ZPOOL_HIST_ZONE,
hap->ha_zone) == 0);
#ifdef _KERNEL
VERIFY(nvlist_add_string(nvrecord, ZPOOL_HIST_HOST,
utsname.nodename) == 0);
#endif
if (hap->ha_log_type == LOG_CMD_POOL_CREATE ||
hap->ha_log_type == LOG_CMD_NORMAL) {
VERIFY(nvlist_add_string(nvrecord, ZPOOL_HIST_CMD,
history_str) == 0);
zfs_dbgmsg("command: %s", history_str);
} else {
VERIFY(nvlist_add_uint64(nvrecord, ZPOOL_HIST_INT_EVENT,
hap->ha_event) == 0);
VERIFY(nvlist_add_uint64(nvrecord, ZPOOL_HIST_TXG,
tx->tx_txg) == 0);
VERIFY(nvlist_add_string(nvrecord, ZPOOL_HIST_INT_STR,
history_str) == 0);
zfs_dbgmsg("internal %s pool:%s txg:%llu %s",
zfs_history_event_names[hap->ha_event], spa_name(spa),
(longlong_t)tx->tx_txg, history_str);
}
VERIFY(nvlist_size(nvrecord, &reclen, NV_ENCODE_XDR) == 0);
record_packed = kmem_alloc(reclen, KM_SLEEP);
VERIFY(nvlist_pack(nvrecord, &record_packed, &reclen,
NV_ENCODE_XDR, KM_SLEEP) == 0);
mutex_enter(&spa->spa_history_lock);
if (hap->ha_log_type == LOG_CMD_POOL_CREATE)
VERIFY(shpp->sh_eof == shpp->sh_pool_create_len);
/* write out the packed length as little endian */
le_len = LE_64((uint64_t)reclen);
ret = spa_history_write(spa, &le_len, sizeof (le_len), shpp, tx);
if (!ret)
ret = spa_history_write(spa, record_packed, reclen, shpp, tx);
if (!ret && hap->ha_log_type == LOG_CMD_POOL_CREATE) {
shpp->sh_pool_create_len += sizeof (le_len) + reclen;
shpp->sh_bof = shpp->sh_pool_create_len;
}
mutex_exit(&spa->spa_history_lock);
nvlist_free(nvrecord);
kmem_free(record_packed, reclen);
dmu_buf_rele(dbp, FTAG);
strfree(hap->ha_history_str);
if (hap->ha_zone != NULL)
strfree(hap->ha_zone);
kmem_free(hap, sizeof (history_arg_t));
}
/*ARGSUSED*/
static void
spa_history_log_sync(void *arg, dmu_tx_t *tx)
{
nvlist_t *nvl = arg;
spa_t *spa = dmu_tx_pool(tx)->dp_spa;
objset_t *mos = spa->spa_meta_objset;
dmu_buf_t *dbp;
spa_history_phys_t *shpp;
size_t reclen;
uint64_t le_len;
char *record_packed = NULL;
int ret;
/*
* If we have an older pool that doesn't have a command
* history object, create it now.
*/
mutex_enter(&spa->spa_history_lock);
if (!spa->spa_history)
spa_history_create_obj(spa, tx);
mutex_exit(&spa->spa_history_lock);
/*
* Get the offset of where we need to write via the bonus buffer.
* Update the offset when the write completes.
*/
VERIFY0(dmu_bonus_hold(mos, spa->spa_history, FTAG, &dbp));
shpp = dbp->db_data;
dmu_buf_will_dirty(dbp, tx);
#ifdef ZFS_DEBUG
{
dmu_object_info_t doi;
dmu_object_info_from_db(dbp, &doi);
ASSERT3U(doi.doi_bonus_type, ==, DMU_OT_SPA_HISTORY_OFFSETS);
}
#endif
fnvlist_add_uint64(nvl, ZPOOL_HIST_TIME, gethrestime_sec());
#ifdef _KERNEL
fnvlist_add_string(nvl, ZPOOL_HIST_HOST, utsname.nodename);
#endif
if (nvlist_exists(nvl, ZPOOL_HIST_CMD)) {
zfs_dbgmsg("command: %s",
fnvlist_lookup_string(nvl, ZPOOL_HIST_CMD));
} else if (nvlist_exists(nvl, ZPOOL_HIST_INT_NAME)) {
if (nvlist_exists(nvl, ZPOOL_HIST_DSNAME)) {
zfs_dbgmsg("txg %lld %s %s (id %llu) %s",
fnvlist_lookup_uint64(nvl, ZPOOL_HIST_TXG),
fnvlist_lookup_string(nvl, ZPOOL_HIST_INT_NAME),
fnvlist_lookup_string(nvl, ZPOOL_HIST_DSNAME),
fnvlist_lookup_uint64(nvl, ZPOOL_HIST_DSID),
fnvlist_lookup_string(nvl, ZPOOL_HIST_INT_STR));
} else {
zfs_dbgmsg("txg %lld %s %s",
fnvlist_lookup_uint64(nvl, ZPOOL_HIST_TXG),
fnvlist_lookup_string(nvl, ZPOOL_HIST_INT_NAME),
fnvlist_lookup_string(nvl, ZPOOL_HIST_INT_STR));
}
} else if (nvlist_exists(nvl, ZPOOL_HIST_IOCTL)) {
zfs_dbgmsg("ioctl %s",
fnvlist_lookup_string(nvl, ZPOOL_HIST_IOCTL));
}
record_packed = fnvlist_pack(nvl, &reclen);
mutex_enter(&spa->spa_history_lock);
/* write out the packed length as little endian */
le_len = LE_64((uint64_t)reclen);
ret = spa_history_write(spa, &le_len, sizeof (le_len), shpp, tx);
if (!ret)
ret = spa_history_write(spa, record_packed, reclen, shpp, tx);
/* The first command is the create, which we keep forever */
if (ret == 0 && shpp->sh_pool_create_len == 0 &&
nvlist_exists(nvl, ZPOOL_HIST_CMD)) {
shpp->sh_pool_create_len = shpp->sh_bof = shpp->sh_eof;
}
mutex_exit(&spa->spa_history_lock);
fnvlist_pack_free(record_packed, reclen);
dmu_buf_rele(dbp, FTAG);
fnvlist_free(nvl);
}
/*ARGSUSED*/
static void
spa_history_log_sync(void *arg, dmu_tx_t *tx)
{
nvlist_t *nvl = arg;
spa_t *spa = dmu_tx_pool(tx)->dp_spa;
objset_t *mos = spa->spa_meta_objset;
dmu_buf_t *dbp;
spa_history_phys_t *shpp;
size_t reclen;
uint64_t le_len;
char *record_packed = NULL;
int ret;
/*
* If we have an older pool that doesn't have a command
* history object, create it now.
*/
mutex_enter(&spa->spa_history_lock);
if (!spa->spa_history)
spa_history_create_obj(spa, tx);
mutex_exit(&spa->spa_history_lock);
/*
* Get the offset of where we need to write via the bonus buffer.
* Update the offset when the write completes.
*/
VERIFY0(dmu_bonus_hold(mos, spa->spa_history, FTAG, &dbp));
shpp = dbp->db_data;
dmu_buf_will_dirty(dbp, tx);
#ifdef ZFS_DEBUG
{
dmu_object_info_t doi;
dmu_object_info_from_db(dbp, &doi);
ASSERT3U(doi.doi_bonus_type, ==, DMU_OT_SPA_HISTORY_OFFSETS);
}
#endif
fnvlist_add_uint64(nvl, ZPOOL_HIST_TIME, gethrestime_sec());
fnvlist_add_string(nvl, ZPOOL_HIST_HOST, utsname()->nodename);
if (nvlist_exists(nvl, ZPOOL_HIST_CMD)) {
zfs_dbgmsg("command: %s",
fnvlist_lookup_string(nvl, ZPOOL_HIST_CMD));
} else if (nvlist_exists(nvl, ZPOOL_HIST_INT_NAME)) {
if (nvlist_exists(nvl, ZPOOL_HIST_DSNAME)) {
zfs_dbgmsg("txg %lld %s %s (id %llu) %s",
fnvlist_lookup_uint64(nvl, ZPOOL_HIST_TXG),
fnvlist_lookup_string(nvl, ZPOOL_HIST_INT_NAME),
fnvlist_lookup_string(nvl, ZPOOL_HIST_DSNAME),
fnvlist_lookup_uint64(nvl, ZPOOL_HIST_DSID),
fnvlist_lookup_string(nvl, ZPOOL_HIST_INT_STR));
} else {
zfs_dbgmsg("txg %lld %s %s",
fnvlist_lookup_uint64(nvl, ZPOOL_HIST_TXG),
fnvlist_lookup_string(nvl, ZPOOL_HIST_INT_NAME),
fnvlist_lookup_string(nvl, ZPOOL_HIST_INT_STR));
}
/*
* The history sysevent is posted only for internal history
* messages to show what has happened, not how it happened. For
* example, the following command:
*
* # zfs destroy -r tank/foo
*
* will result in one sysevent posted per dataset that is
* destroyed as a result of the command - which could be more
* than one event in total. By contrast, if the sysevent was
* posted as a result of the ZPOOL_HIST_CMD key being present
* it would result in only one sysevent being posted with the
* full command line arguments, requiring the consumer to know
* how to parse and understand zfs(1M) command invocations.
*/
spa_history_log_notify(spa, nvl);
} else if (nvlist_exists(nvl, ZPOOL_HIST_IOCTL)) {
zfs_dbgmsg("ioctl %s",
fnvlist_lookup_string(nvl, ZPOOL_HIST_IOCTL));
}
VERIFY3U(nvlist_pack(nvl, &record_packed, &reclen, NV_ENCODE_NATIVE,
KM_SLEEP), ==, 0);
mutex_enter(&spa->spa_history_lock);
/* write out the packed length as little endian */
le_len = LE_64((uint64_t)reclen);
ret = spa_history_write(spa, &le_len, sizeof (le_len), shpp, tx);
if (!ret)
ret = spa_history_write(spa, record_packed, reclen, shpp, tx);
/* The first command is the create, which we keep forever */
if (ret == 0 && shpp->sh_pool_create_len == 0 &&
nvlist_exists(nvl, ZPOOL_HIST_CMD)) {
shpp->sh_pool_create_len = shpp->sh_bof = shpp->sh_eof;
}
mutex_exit(&spa->spa_history_lock);
fnvlist_pack_free(record_packed, reclen);
dmu_buf_rele(dbp, FTAG);
fnvlist_free(nvl);
}
int
sa_buf_hold(objset_t *objset, uint64_t obj_num, void *tag, dmu_buf_t **db)
{
return (dmu_bonus_hold(objset, obj_num, tag, db));
}
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);
}
#endif
/* XXX assert bonus buffer size is correct */
if (dd == NULL) {
dsl_dir_t *winner;
#ifndef __APPLE__
int err;
#endif
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;
dd->dd_used_bytes = dd->dd_phys->dd_used_bytes;
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));
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) {
mutex_destroy(&dd->dd_lock);
kmem_free(dd, sizeof (dsl_dir_t));
dmu_buf_rele(dbuf, tag);
return (err);
}
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) {
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);
}
} else {
(void) strcpy(dd->dd_myname, spa_name(dp->dp_spa));
}
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);
}
int
dsl_dir_hold_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(dsl_pool_config_held(dp));
err = dmu_bonus_hold(dp->dp_meta_objset, ddobj, tag, &dbuf);
if (err != 0)
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_bonus_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_hold_obj(dp, dd->dd_phys->dd_parent_obj,
NULL, dd, &dd->dd_parent);
if (err != 0)
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 != 0)
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 != 0)
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_rele(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_rele(dd->dd_parent, dd);
mutex_destroy(&dd->dd_lock);
kmem_free(dd, sizeof (dsl_dir_t));
dmu_buf_rele(dbuf, tag);
return (err);
}
static int
zvol_first_open(zvol_state_t *zv)
{
objset_t *os;
uint64_t volsize;
int locked = 0;
int error;
uint64_t ro;
/*
* In all other cases the spa_namespace_lock is taken before the
* bdev->bd_mutex lock. But in this case the Linux __blkdev_get()
* function calls fops->open() with the bdev->bd_mutex lock held.
*
* To avoid a potential lock inversion deadlock we preemptively
* try to take the spa_namespace_lock(). Normally it will not
* be contended and this is safe because spa_open_common() handles
* the case where the caller already holds the spa_namespace_lock.
*
* When it is contended we risk a lock inversion if we were to
* block waiting for the lock. Luckily, the __blkdev_get()
* function allows us to return -ERESTARTSYS which will result in
* bdev->bd_mutex being dropped, reacquired, and fops->open() being
* called again. This process can be repeated safely until both
* locks are acquired.
*/
if (!mutex_owned(&spa_namespace_lock)) {
locked = mutex_tryenter(&spa_namespace_lock);
if (!locked)
return (-ERESTARTSYS);
}
/* lie and say we're read-only */
error = dmu_objset_own(zv->zv_name, DMU_OST_ZVOL, 1, zvol_tag, &os);
if (error)
goto out_mutex;
error = zap_lookup(os, ZVOL_ZAP_OBJ, "size", 8, 1, &volsize);
if (error) {
dmu_objset_disown(os, zvol_tag);
goto out_mutex;
}
zv->zv_objset = os;
error = dmu_bonus_hold(os, ZVOL_OBJ, zvol_tag, &zv->zv_dbuf);
if (error) {
dmu_objset_disown(os, zvol_tag);
goto out_mutex;
}
set_capacity(zv->zv_disk, volsize >> 9);
zv->zv_volsize = volsize;
zv->zv_zilog = zil_open(os, zvol_get_data);
VERIFY(dsl_prop_get_integer(zv->zv_name, "readonly", &ro, NULL) == 0);
if (ro || dmu_objset_is_snapshot(os) ||
!spa_writeable(dmu_objset_spa(os))) {
set_disk_ro(zv->zv_disk, 1);
zv->zv_flags |= ZVOL_RDONLY;
} else {
set_disk_ro(zv->zv_disk, 0);
zv->zv_flags &= ~ZVOL_RDONLY;
}
out_mutex:
if (locked)
mutex_exit(&spa_namespace_lock);
return (-error);
}
static
#endif
int
zfs_fuid_find_by_domain(zfsvfs_t *zfsvfs, const char *domain, char **retdomain,
dmu_tx_t *tx)
{
fuid_domain_t searchnode, *findnode;
avl_index_t loc;
/*
* If the dummy "nobody" domain then return an index of 0
* to cause the created FUID to be a standard POSIX id
* for the user nobody.
*/
if (domain[0] == '\0') {
*retdomain = "";
return (0);
}
searchnode.f_ksid = ksid_lookupdomain(domain);
if (retdomain) {
*retdomain = searchnode.f_ksid->kd_name;
}
if (!zfsvfs->z_fuid_loaded)
zfs_fuid_init(zfsvfs, tx);
rw_enter(&zfsvfs->z_fuid_lock, RW_READER);
findnode = avl_find(&zfsvfs->z_fuid_domain, &searchnode, &loc);
rw_exit(&zfsvfs->z_fuid_lock);
if (findnode) {
ksiddomain_rele(searchnode.f_ksid);
return (findnode->f_idx);
} else {
fuid_domain_t *domnode;
nvlist_t *nvp;
nvlist_t **fuids;
uint64_t retidx;
size_t nvsize = 0;
char *packed;
dmu_buf_t *db;
int i = 0;
domnode = kmem_alloc(sizeof (fuid_domain_t), KM_SLEEP);
domnode->f_ksid = searchnode.f_ksid;
rw_enter(&zfsvfs->z_fuid_lock, RW_WRITER);
retidx = domnode->f_idx = avl_numnodes(&zfsvfs->z_fuid_idx) + 1;
avl_add(&zfsvfs->z_fuid_domain, domnode);
avl_add(&zfsvfs->z_fuid_idx, domnode);
/*
* Now resync the on-disk nvlist.
*/
VERIFY(nvlist_alloc(&nvp, NV_UNIQUE_NAME, KM_SLEEP) == 0);
domnode = avl_first(&zfsvfs->z_fuid_domain);
fuids = kmem_alloc(retidx * sizeof (void *), KM_SLEEP);
while (domnode) {
VERIFY(nvlist_alloc(&fuids[i],
NV_UNIQUE_NAME, KM_SLEEP) == 0);
VERIFY(nvlist_add_uint64(fuids[i], FUID_IDX,
domnode->f_idx) == 0);
VERIFY(nvlist_add_uint64(fuids[i],
FUID_OFFSET, 0) == 0);
VERIFY(nvlist_add_string(fuids[i++], FUID_DOMAIN,
domnode->f_ksid->kd_name) == 0);
domnode = AVL_NEXT(&zfsvfs->z_fuid_domain, domnode);
}
VERIFY(nvlist_add_nvlist_array(nvp, FUID_NVP_ARRAY,
fuids, retidx) == 0);
for (i = 0; i != retidx; i++)
nvlist_free(fuids[i]);
kmem_free(fuids, retidx * sizeof (void *));
VERIFY(nvlist_size(nvp, &nvsize, NV_ENCODE_XDR) == 0);
packed = kmem_alloc(nvsize, KM_SLEEP);
VERIFY(nvlist_pack(nvp, &packed, &nvsize,
NV_ENCODE_XDR, KM_SLEEP) == 0);
nvlist_free(nvp);
zfsvfs->z_fuid_size = nvsize;
dmu_write(zfsvfs->z_os, zfsvfs->z_fuid_obj, 0,
zfsvfs->z_fuid_size, packed, tx);
kmem_free(packed, zfsvfs->z_fuid_size);
VERIFY(0 == dmu_bonus_hold(zfsvfs->z_os, zfsvfs->z_fuid_obj,
FTAG, &db));
dmu_buf_will_dirty(db, tx);
*(uint64_t *)db->db_data = zfsvfs->z_fuid_size;
dmu_buf_rele(db, FTAG);
rw_exit(&zfsvfs->z_fuid_lock);
return (retidx);
}
}
/*
* sync out AVL trees to persistent storage.
*/
void
zfs_fuid_sync(zfsvfs_t *zfsvfs, dmu_tx_t *tx)
{
#ifdef HAVE_ZPL
nvlist_t *nvp;
nvlist_t **fuids;
size_t nvsize = 0;
char *packed;
dmu_buf_t *db;
fuid_domain_t *domnode;
int numnodes;
int i;
if (!zfsvfs->z_fuid_dirty) {
return;
}
rw_enter(&zfsvfs->z_fuid_lock, RW_WRITER);
/*
* First see if table needs to be created?
*/
if (zfsvfs->z_fuid_obj == 0) {
zfsvfs->z_fuid_obj = dmu_object_alloc(zfsvfs->z_os,
DMU_OT_FUID, 1 << 14, DMU_OT_FUID_SIZE,
sizeof (uint64_t), tx);
VERIFY(zap_add(zfsvfs->z_os, MASTER_NODE_OBJ,
ZFS_FUID_TABLES, sizeof (uint64_t), 1,
&zfsvfs->z_fuid_obj, tx) == 0);
}
VERIFY(nvlist_alloc(&nvp, NV_UNIQUE_NAME, KM_SLEEP) == 0);
numnodes = avl_numnodes(&zfsvfs->z_fuid_idx);
fuids = kmem_alloc(numnodes * sizeof (void *), KM_SLEEP);
for (i = 0, domnode = avl_first(&zfsvfs->z_fuid_domain); domnode; i++,
domnode = AVL_NEXT(&zfsvfs->z_fuid_domain, domnode)) {
VERIFY(nvlist_alloc(&fuids[i], NV_UNIQUE_NAME, KM_SLEEP) == 0);
VERIFY(nvlist_add_uint64(fuids[i], FUID_IDX,
domnode->f_idx) == 0);
VERIFY(nvlist_add_uint64(fuids[i], FUID_OFFSET, 0) == 0);
VERIFY(nvlist_add_string(fuids[i], FUID_DOMAIN,
domnode->f_ksid->kd_name) == 0);
}
VERIFY(nvlist_add_nvlist_array(nvp, FUID_NVP_ARRAY,
fuids, numnodes) == 0);
for (i = 0; i != numnodes; i++)
nvlist_free(fuids[i]);
kmem_free(fuids, numnodes * sizeof (void *));
VERIFY(nvlist_size(nvp, &nvsize, NV_ENCODE_XDR) == 0);
packed = kmem_alloc(nvsize, KM_SLEEP);
VERIFY(nvlist_pack(nvp, &packed, &nvsize,
NV_ENCODE_XDR, KM_SLEEP) == 0);
nvlist_free(nvp);
zfsvfs->z_fuid_size = nvsize;
dmu_write(zfsvfs->z_os, zfsvfs->z_fuid_obj, 0,
zfsvfs->z_fuid_size, packed, tx);
kmem_free(packed, zfsvfs->z_fuid_size);
VERIFY(0 == dmu_bonus_hold(zfsvfs->z_os, zfsvfs->z_fuid_obj,
FTAG, &db));
dmu_buf_will_dirty(db, tx);
*(uint64_t *)db->db_data = zfsvfs->z_fuid_size;
dmu_buf_rele(db, FTAG);
zfsvfs->z_fuid_dirty = B_FALSE;
rw_exit(&zfsvfs->z_fuid_lock);
#endif /* HAVE_ZPL */
}