static void
dsl_onexit_hold_cleanup(spa_t *spa, nvlist_t *holds, minor_t minor)
{
zfs_hold_cleanup_arg_t *ca;
if (minor == 0 || nvlist_empty(holds)) {
fnvlist_free(holds);
return;
}
ASSERT(spa != NULL);
ca = kmem_alloc(sizeof (*ca), KM_SLEEP);
(void) strlcpy(ca->zhca_spaname, spa_name(spa),
sizeof (ca->zhca_spaname));
ca->zhca_spa_load_guid = spa_load_guid(spa);
ca->zhca_holds = holds;
VERIFY0(zfs_onexit_add_cb(minor,
dsl_dataset_user_release_onexit, ca, NULL));
}
void
spa_history_log_version(spa_t *spa, history_internal_events_t event)
{
#ifdef _KERNEL
uint64_t current_vers = spa_version(spa);
if (current_vers >= SPA_VERSION_ZPOOL_HISTORY) {
spa_history_log_internal(event, spa, NULL,
"pool spa %llu; zfs spa %llu; zpl %d; uts %s %s %s %s",
(u_longlong_t)current_vers, SPA_VERSION, ZPL_VERSION,
utsname.nodename, utsname.release, utsname.version,
utsname.machine);
}
#ifndef __DARWIN__
cmn_err(CE_CONT, "!%s version %llu pool %s using %llu",
event == LOG_POOL_IMPORT ? "imported" :
event == LOG_POOL_CREATE ? "created" : "accessed",
(u_longlong_t)current_vers, spa_name(spa), SPA_VERSION);
#endif
#endif
}
int
zcp_dataset_hold_error(lua_State *state, dsl_pool_t *dp, const char *dsname,
int error)
{
if (error == ENOENT) {
(void) zcp_argerror(state, 1, "no such dataset '%s'", dsname);
return (0); /* not reached; zcp_argerror will longjmp */
} else if (error == EXDEV) {
(void) zcp_argerror(state, 1,
"dataset '%s' is not in the target pool '%s'",
dsname, spa_name(dp->dp_spa));
return (0); /* not reached; zcp_argerror will longjmp */
} else if (error == EIO) {
(void) luaL_error(state,
"I/O error while accessing dataset '%s'", dsname);
return (0); /* not reached; luaL_error will longjmp */
} else if (error != 0) {
(void) luaL_error(state,
"unexpected error %d while accessing dataset '%s'",
error, dsname);
return (0); /* not reached; luaL_error will longjmp */
}
return (0);
}
/*
* Synchronize pool configuration to disk. This must be called with the
* namespace lock held. Synchronizing the pool cache is typically done after
* the configuration has been synced to the MOS. This exposes a window where
* the MOS config will have been updated but the cache file has not. If
* the system were to crash at that instant then the cached config may not
* contain the correct information to open the pool and an explicit import
* would be required.
*/
void
spa_config_sync(spa_t *target, boolean_t removing, boolean_t postsysevent)
{
spa_config_dirent_t *dp, *tdp;
nvlist_t *nvl;
char *pool_name;
boolean_t ccw_failure;
int error = 0;
ASSERT(MUTEX_HELD(&spa_namespace_lock));
if (rootdir == NULL || !(spa_mode_global & FWRITE))
return;
/*
* Iterate over all cachefiles for the pool, past or present. When the
* cachefile is changed, the new one is pushed onto this list, allowing
* us to update previous cachefiles that no longer contain this pool.
*/
ccw_failure = B_FALSE;
for (dp = list_head(&target->spa_config_list); dp != NULL;
dp = list_next(&target->spa_config_list, dp)) {
spa_t *spa = NULL;
if (dp->scd_path == NULL)
continue;
/*
* Iterate over all pools, adding any matching pools to 'nvl'.
*/
nvl = NULL;
while ((spa = spa_next(spa)) != NULL) {
/*
* Skip over our own pool if we're about to remove
* ourselves from the spa namespace or any pool that
* is readonly. Since we cannot guarantee that a
* readonly pool would successfully import upon reboot,
* we don't allow them to be written to the cache file.
*/
if ((spa == target && removing) ||
!spa_writeable(spa))
continue;
mutex_enter(&spa->spa_props_lock);
tdp = list_head(&spa->spa_config_list);
if (spa->spa_config == NULL ||
tdp == NULL ||
tdp->scd_path == NULL ||
strcmp(tdp->scd_path, dp->scd_path) != 0) {
mutex_exit(&spa->spa_props_lock);
continue;
}
if (nvl == NULL)
nvl = fnvlist_alloc();
if (spa->spa_import_flags & ZFS_IMPORT_TEMP_NAME)
pool_name = fnvlist_lookup_string(
spa->spa_config, ZPOOL_CONFIG_POOL_NAME);
else
pool_name = spa_name(spa);
fnvlist_add_nvlist(nvl, pool_name, spa->spa_config);
mutex_exit(&spa->spa_props_lock);
}
error = spa_config_write(dp, nvl);
if (error != 0)
ccw_failure = B_TRUE;
nvlist_free(nvl);
}
if (ccw_failure) {
/*
* Keep trying so that configuration data is
* written if/when any temporary filesystem
* resource issues are resolved.
*/
if (target->spa_ccw_fail_time == 0) {
zfs_ereport_post(FM_EREPORT_ZFS_CONFIG_CACHE_WRITE,
target, NULL, NULL, 0, 0);
}
target->spa_ccw_fail_time = gethrtime();
spa_async_request(target, SPA_ASYNC_CONFIG_UPDATE);
} else {
/*
* Do not rate limit future attempts to update
* the config cache.
*/
target->spa_ccw_fail_time = 0;
}
/*
* Remove any config entries older than the current one.
*/
dp = list_head(&target->spa_config_list);
while ((tdp = list_next(&target->spa_config_list, dp)) != NULL) {
list_remove(&target->spa_config_list, tdp);
if (tdp->scd_path != NULL)
spa_strfree(tdp->scd_path);
kmem_free(tdp, sizeof (spa_config_dirent_t));
}
spa_config_generation++;
if (postsysevent)
spa_event_notify(target, NULL, ESC_ZFS_CONFIG_SYNC);
}
/*
* Synchronize pool configuration to disk. This must be called with the
* namespace lock held. Synchronizing the pool cache is typically done after
* the configuration has been synced to the MOS. This exposes a window where
* the MOS config will have been updated but the cache file has not. If
* the system were to crash at that instant then the cached config may not
* contain the correct information to open the pool and an explicity import
* would be required.
*/
void
spa_config_sync(spa_t *target, boolean_t removing, boolean_t postsysevent)
{
spa_config_dirent_t *dp, *tdp;
nvlist_t *nvl;
char *pool_name;
ASSERT(MUTEX_HELD(&spa_namespace_lock));
if (rootdir == NULL || !(spa_mode_global & FWRITE))
return;
/*
* Iterate over all cachefiles for the pool, past or present. When the
* cachefile is changed, the new one is pushed onto this list, allowing
* us to update previous cachefiles that no longer contain this pool.
*/
for (dp = list_head(&target->spa_config_list); dp != NULL;
dp = list_next(&target->spa_config_list, dp)) {
spa_t *spa = NULL;
if (dp->scd_path == NULL)
continue;
/*
* Iterate over all pools, adding any matching pools to 'nvl'.
*/
nvl = NULL;
while ((spa = spa_next(spa)) != NULL) {
/*
* Skip over our own pool if we're about to remove
* ourselves from the spa namespace or any pool that
* is readonly. Since we cannot guarantee that a
* readonly pool would successfully import upon reboot,
* we don't allow them to be written to the cache file.
*/
if ((spa == target && removing) ||
!spa_writeable(spa))
continue;
mutex_enter(&spa->spa_props_lock);
tdp = list_head(&spa->spa_config_list);
if (spa->spa_config == NULL ||
tdp->scd_path == NULL ||
strcmp(tdp->scd_path, dp->scd_path) != 0) {
mutex_exit(&spa->spa_props_lock);
continue;
}
if (nvl == NULL)
VERIFY(nvlist_alloc(&nvl, NV_UNIQUE_NAME,
KM_SLEEP) == 0);
if (spa->spa_import_flags & ZFS_IMPORT_TEMP_NAME) {
VERIFY0(nvlist_lookup_string(spa->spa_config,
ZPOOL_CONFIG_POOL_NAME, &pool_name));
} else
pool_name = spa_name(spa);
VERIFY(nvlist_add_nvlist(nvl, pool_name,
spa->spa_config) == 0);
mutex_exit(&spa->spa_props_lock);
}
spa_config_write(dp, nvl);
nvlist_free(nvl);
}
/*
* Remove any config entries older than the current one.
*/
dp = list_head(&target->spa_config_list);
while ((tdp = list_next(&target->spa_config_list, dp)) != NULL) {
list_remove(&target->spa_config_list, tdp);
if (tdp->scd_path != NULL)
spa_strfree(tdp->scd_path);
kmem_free(tdp, sizeof (spa_config_dirent_t));
}
spa_config_generation++;
if (postsysevent)
spa_event_notify(target, NULL, FM_EREPORT_ZFS_CONFIG_SYNC);
}
/*
* Find all objsets under name, call func on each
*/
int
dmu_objset_find_spa(spa_t *spa, const char *name,
int func(spa_t *, uint64_t, const char *, void *), void *arg, int flags)
{
dsl_dir_t *dd;
dsl_pool_t *dp;
dsl_dataset_t *ds;
zap_cursor_t zc;
zap_attribute_t *attr;
char *child;
uint64_t thisobj;
int err;
if (name == NULL)
name = spa_name(spa);
err = dsl_dir_open_spa(spa, name, FTAG, &dd, NULL);
if (err)
return (err);
/* Don't visit hidden ($MOS & $ORIGIN) objsets. */
if (dd->dd_myname[0] == '$') {
dsl_dir_close(dd, FTAG);
return (0);
}
thisobj = dd->dd_phys->dd_head_dataset_obj;
attr = kmem_alloc(sizeof (zap_attribute_t), KM_SLEEP);
dp = dd->dd_pool;
/*
* Iterate over all children.
*/
if (flags & DS_FIND_CHILDREN) {
for (zap_cursor_init(&zc, dp->dp_meta_objset,
dd->dd_phys->dd_child_dir_zapobj);
zap_cursor_retrieve(&zc, attr) == 0;
(void) zap_cursor_advance(&zc)) {
ASSERT(attr->za_integer_length == sizeof (uint64_t));
ASSERT(attr->za_num_integers == 1);
child = kmem_alloc(MAXPATHLEN, KM_SLEEP);
(void) strcpy(child, name);
(void) strcat(child, "/");
(void) strcat(child, attr->za_name);
err = dmu_objset_find_spa(spa, child, func, arg, flags);
kmem_free(child, MAXPATHLEN);
if (err)
break;
}
zap_cursor_fini(&zc);
if (err) {
dsl_dir_close(dd, FTAG);
kmem_free(attr, sizeof (zap_attribute_t));
return (err);
}
}
/*
* Iterate over all snapshots.
*/
if (flags & DS_FIND_SNAPSHOTS) {
if (!dsl_pool_sync_context(dp))
rw_enter(&dp->dp_config_rwlock, RW_READER);
err = dsl_dataset_hold_obj(dp, thisobj, FTAG, &ds);
if (!dsl_pool_sync_context(dp))
rw_exit(&dp->dp_config_rwlock);
if (err == 0) {
uint64_t snapobj = ds->ds_phys->ds_snapnames_zapobj;
dsl_dataset_rele(ds, FTAG);
for (zap_cursor_init(&zc, dp->dp_meta_objset, snapobj);
zap_cursor_retrieve(&zc, attr) == 0;
(void) zap_cursor_advance(&zc)) {
ASSERT(attr->za_integer_length ==
sizeof (uint64_t));
ASSERT(attr->za_num_integers == 1);
child = kmem_alloc(MAXPATHLEN, KM_SLEEP);
(void) strcpy(child, name);
(void) strcat(child, "@");
(void) strcat(child, attr->za_name);
err = func(spa, attr->za_first_integer,
child, arg);
kmem_free(child, MAXPATHLEN);
if (err)
break;
}
zap_cursor_fini(&zc);
}
}
dsl_dir_close(dd, FTAG);
kmem_free(attr, sizeof (zap_attribute_t));
if (err)
return (err);
/*
* Apply to self if appropriate.
*/
err = func(spa, thisobj, name, arg);
return (err);
}
static int
zfs_vfs_getattr(struct mount *mp, struct vfs_attr *fsap, __unused vfs_context_t context)
{
zfsvfs_t *zfsvfs = vfs_fsprivate(mp);
uint64_t refdbytes, availbytes, usedobjs, availobjs;
ZFS_ENTER(zfsvfs);
dmu_objset_space(zfsvfs->z_os,
&refdbytes, &availbytes, &usedobjs, &availobjs);
VFSATTR_RETURN(fsap, f_objcount, usedobjs);
VFSATTR_RETURN(fsap, f_maxobjcount, 0x7fffffffffffffff);
/*
* Carbon depends on f_filecount and f_dircount so
* make up some values based on total objects.
*/
VFSATTR_RETURN(fsap, f_filecount, usedobjs - (usedobjs / 4));
VFSATTR_RETURN(fsap, f_dircount, usedobjs / 4);
/*
* The underlying storage pool actually uses multiple block sizes.
* We report the fragsize as the smallest block size we support,
* and we report our blocksize as the filesystem's maximum blocksize.
*/
VFSATTR_RETURN(fsap, f_bsize, 1UL << SPA_MINBLOCKSHIFT);
VFSATTR_RETURN(fsap, f_iosize, zfsvfs->z_max_blksz);
/*
* The following report "total" blocks of various kinds in the
* file system, but reported in terms of f_frsize - the
* "fragment" size.
*/
VFSATTR_RETURN(fsap, f_blocks,
(u_int64_t)((refdbytes + availbytes) >> SPA_MINBLOCKSHIFT));
VFSATTR_RETURN(fsap, f_bfree, (u_int64_t)(availbytes >> SPA_MINBLOCKSHIFT));
VFSATTR_RETURN(fsap, f_bavail, fsap->f_bfree); /* no root reservation */
VFSATTR_RETURN(fsap, f_bused, fsap->f_blocks - fsap->f_bfree);
/*
* statvfs() should really be called statufs(), because it assumes
* static metadata. ZFS doesn't preallocate files, so the best
* we can do is report the max that could possibly fit in f_files,
* and that minus the number actually used in f_ffree.
* For f_ffree, report the smaller of the number of object available
* and the number of blocks (each object will take at least a block).
*/
VFSATTR_RETURN(fsap, f_ffree, (u_int64_t)MIN(availobjs, fsap->f_bfree));
VFSATTR_RETURN(fsap, f_files, fsap->f_ffree + usedobjs);
#if 0
statp->f_flag = vf_to_stf(vfsp->vfs_flag);
#endif
if (VFSATTR_IS_ACTIVE(fsap, f_fsid)) {
VFSATTR_RETURN(fsap, f_fsid, vfs_statfs(mp)->f_fsid);
}
if (VFSATTR_IS_ACTIVE(fsap, f_capabilities)) {
bcopy(&zfs_capabilities, &fsap->f_capabilities, sizeof (zfs_capabilities));
VFSATTR_SET_SUPPORTED(fsap, f_capabilities);
}
if (VFSATTR_IS_ACTIVE(fsap, f_attributes)) {
bcopy(&zfs_attributes, &fsap->f_attributes.validattr, sizeof (zfs_attributes));
bcopy(&zfs_attributes, &fsap->f_attributes.nativeattr, sizeof (zfs_attributes));
VFSATTR_SET_SUPPORTED(fsap, f_attributes);
}
if (VFSATTR_IS_ACTIVE(fsap, f_create_time)) {
dmu_objset_stats_t dmu_stat;
dmu_objset_fast_stat(zfsvfs->z_os, &dmu_stat);
fsap->f_create_time.tv_sec = dmu_stat.dds_creation_time;
fsap->f_create_time.tv_nsec = 0;
VFSATTR_SET_SUPPORTED(fsap, f_create_time);
}
if (VFSATTR_IS_ACTIVE(fsap, f_modify_time)) {
if (zfsvfs->z_mtime_vp != NULL) {
znode_t *mzp;
mzp = VTOZ(zfsvfs->z_mtime_vp);
ZFS_TIME_DECODE(&fsap->f_modify_time, mzp->z_phys->zp_mtime);
} else {
fsap->f_modify_time.tv_sec = 0;
fsap->f_modify_time.tv_nsec = 0;
}
VFSATTR_SET_SUPPORTED(fsap, f_modify_time);
}
/*
* For Carbon compatibility, pretend to support this legacy/unused
* attribute
*/
if (VFSATTR_IS_ACTIVE(fsap, f_backup_time)) {
fsap->f_backup_time.tv_sec = 0;
fsap->f_backup_time.tv_nsec = 0;
VFSATTR_SET_SUPPORTED(fsap, f_backup_time);
}
if (VFSATTR_IS_ACTIVE(fsap, f_vol_name)) {
spa_t *spa = dmu_objset_spa(zfsvfs->z_os);
spa_config_enter(spa, RW_READER, FTAG);
strlcpy(fsap->f_vol_name, spa_name(spa), MAXPATHLEN);
spa_config_exit(spa, FTAG);
VFSATTR_SET_SUPPORTED(fsap, f_vol_name);
}
VFSATTR_RETURN(fsap, f_fssubtype, 0);
VFSATTR_RETURN(fsap, f_signature, 0x5a21); /* 'Z!' */
VFSATTR_RETURN(fsap, f_carbon_fsid, 0);
ZFS_EXIT(zfsvfs);
return (0);
}
/*
* Generate the pool's configuration based on the current in-core state.
* We infer whether to generate a complete config or just one top-level config
* based on whether vd is the root vdev.
*/
nvlist_t *
spa_config_generate(spa_t *spa, vdev_t *vd, uint64_t txg, int getstats)
{
nvlist_t *config, *nvroot;
vdev_t *rvd = spa->spa_root_vdev;
unsigned long hostid = 0;
#ifdef __APPLE__
boolean_t skipdevpaths = (txg != -1ULL && txg != 0ULL);
#endif
ASSERT(spa_config_held(spa, RW_READER) ||
spa_config_held(spa, RW_WRITER));
if (vd == NULL)
vd = rvd;
/*
* If txg is -1, report the current value of spa->spa_config_txg.
*/
if (txg == -1ULL)
txg = spa->spa_config_txg;
VERIFY(nvlist_alloc(&config, NV_UNIQUE_NAME, KM_SLEEP) == 0);
VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_VERSION,
spa_version(spa)) == 0);
VERIFY(nvlist_add_string(config, ZPOOL_CONFIG_POOL_NAME,
spa_name(spa)) == 0);
VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_POOL_STATE,
spa_state(spa)) == 0);
VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_POOL_TXG,
txg) == 0);
VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_POOL_GUID,
spa_guid(spa)) == 0);
#ifndef __APPLE__
(void) ddi_strtoul(hw_serial, NULL, 10, &hostid);
#endif
if (hostid != 0) {
VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_HOSTID,
hostid) == 0);
}
/*XXX NOEL: If host id identification is a feature we are
* interested in having (primarily for poor man's cluster
* support), then query for the hostid here and add it to the
* nvlist
*/
#ifndef __APPLE__
VERIFY(nvlist_add_string(config, ZPOOL_CONFIG_HOSTNAME,
utsname.nodename) == 0);
#endif
if (vd != rvd) {
VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_TOP_GUID,
vd->vdev_top->vdev_guid) == 0);
VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_GUID,
vd->vdev_guid) == 0);
if (vd->vdev_isspare)
VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_IS_SPARE,
1ULL) == 0);
if (vd->vdev_islog)
VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_IS_LOG,
1ULL) == 0);
vd = vd->vdev_top; /* label contains top config */
}
#ifdef __APPLE__
nvroot = vdev_config_generate(spa, vd, getstats, B_FALSE, B_FALSE,
skipdevpaths);
#else
nvroot = vdev_config_generate(spa, vd, getstats, B_FALSE, B_FALSE);
#endif
VERIFY(nvlist_add_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, nvroot) == 0);
nvlist_free(nvroot);
return (config);
}
static void
mmp_thread(void *arg)
{
spa_t *spa = (spa_t *)arg;
mmp_thread_t *mmp = &spa->spa_mmp;
boolean_t last_spa_suspended = spa_suspended(spa);
boolean_t last_spa_multihost = spa_multihost(spa);
callb_cpr_t cpr;
hrtime_t max_fail_ns = zfs_multihost_fail_intervals *
MSEC2NSEC(MAX(zfs_multihost_interval, MMP_MIN_INTERVAL));
mmp_thread_enter(mmp, &cpr);
/*
* The mmp_write_done() function calculates mmp_delay based on the
* prior value of mmp_delay and the elapsed time since the last write.
* For the first mmp write, there is no "last write", so we start
* with fake, but reasonable, default non-zero values.
*/
mmp->mmp_delay = MSEC2NSEC(MAX(zfs_multihost_interval,
MMP_MIN_INTERVAL)) / MAX(vdev_count_leaves(spa), 1);
mmp->mmp_last_write = gethrtime() - mmp->mmp_delay;
while (!mmp->mmp_thread_exiting) {
uint64_t mmp_fail_intervals = zfs_multihost_fail_intervals;
uint64_t mmp_interval = MSEC2NSEC(
MAX(zfs_multihost_interval, MMP_MIN_INTERVAL));
boolean_t suspended = spa_suspended(spa);
boolean_t multihost = spa_multihost(spa);
hrtime_t start, next_time;
start = gethrtime();
if (multihost) {
next_time = start + mmp_interval /
MAX(vdev_count_leaves(spa), 1);
} else {
next_time = start + MSEC2NSEC(MMP_DEFAULT_INTERVAL);
}
/*
* When MMP goes off => on, or spa goes suspended =>
* !suspended, we know no writes occurred recently. We
* update mmp_last_write to give us some time to try.
*/
if ((!last_spa_multihost && multihost) ||
(last_spa_suspended && !suspended)) {
mutex_enter(&mmp->mmp_io_lock);
mmp->mmp_last_write = gethrtime();
mutex_exit(&mmp->mmp_io_lock);
} else if (last_spa_multihost && !multihost) {
mutex_enter(&mmp->mmp_io_lock);
mmp->mmp_delay = 0;
mutex_exit(&mmp->mmp_io_lock);
}
last_spa_multihost = multihost;
last_spa_suspended = suspended;
/*
* Smooth max_fail_ns when its factors are decreased, because
* making (max_fail_ns < mmp_interval) results in the pool being
* immediately suspended before writes can occur at the new
* higher frequency.
*/
if ((mmp_interval * mmp_fail_intervals) < max_fail_ns) {
max_fail_ns = ((31 * max_fail_ns) + (mmp_interval *
mmp_fail_intervals)) / 32;
} else {
max_fail_ns = mmp_interval * mmp_fail_intervals;
}
/*
* Suspend the pool if no MMP write has succeeded in over
* mmp_interval * mmp_fail_intervals nanoseconds.
*/
if (!suspended && mmp_fail_intervals && multihost &&
(start - mmp->mmp_last_write) > max_fail_ns) {
cmn_err(CE_WARN, "MMP writes to pool '%s' have not "
"succeeded in over %llus; suspending pool",
spa_name(spa),
NSEC2SEC(start - mmp->mmp_last_write));
zio_suspend(spa, NULL);
}
if (multihost)
mmp_write_uberblock(spa);
CALLB_CPR_SAFE_BEGIN(&cpr);
(void) cv_timedwait_sig(&mmp->mmp_thread_cv,
&mmp->mmp_thread_lock, ddi_get_lbolt() +
((next_time - gethrtime()) / (NANOSEC / hz)));
CALLB_CPR_SAFE_END(&cpr, &mmp->mmp_thread_lock);
}
/* Outstanding writes are allowed to complete. */
if (mmp->mmp_zio_root)
zio_wait(mmp->mmp_zio_root);
mmp->mmp_zio_root = NULL;
mmp_thread_exit(mmp, &mmp->mmp_thread, &cpr);
}
/* ARGSUSED */
static int
spa_condense_indirect_thread(void *arg, zthr_t *zthr)
{
spa_t *spa = arg;
vdev_t *vd;
ASSERT3P(spa->spa_condensing_indirect, !=, NULL);
spa_config_enter(spa, SCL_VDEV, FTAG, RW_READER);
vd = vdev_lookup_top(spa, spa->spa_condensing_indirect_phys.scip_vdev);
ASSERT3P(vd, !=, NULL);
spa_config_exit(spa, SCL_VDEV, FTAG);
spa_condensing_indirect_t *sci = spa->spa_condensing_indirect;
spa_condensing_indirect_phys_t *scip =
&spa->spa_condensing_indirect_phys;
uint32_t *counts;
uint64_t start_index;
vdev_indirect_mapping_t *old_mapping = vd->vdev_indirect_mapping;
space_map_t *prev_obsolete_sm = NULL;
ASSERT3U(vd->vdev_id, ==, scip->scip_vdev);
ASSERT(scip->scip_next_mapping_object != 0);
ASSERT(scip->scip_prev_obsolete_sm_object != 0);
ASSERT3P(vd->vdev_ops, ==, &vdev_indirect_ops);
for (int i = 0; i < TXG_SIZE; i++) {
/*
* The list must start out empty in order for the
* _commit_sync() sync task to be properly registered
* on the first call to _commit_entry(); so it's wise
* to double check and ensure we actually are starting
* with empty lists.
*/
ASSERT(list_is_empty(&sci->sci_new_mapping_entries[i]));
}
VERIFY0(space_map_open(&prev_obsolete_sm, spa->spa_meta_objset,
scip->scip_prev_obsolete_sm_object, 0, vd->vdev_asize, 0));
space_map_update(prev_obsolete_sm);
counts = vdev_indirect_mapping_load_obsolete_counts(old_mapping);
if (prev_obsolete_sm != NULL) {
vdev_indirect_mapping_load_obsolete_spacemap(old_mapping,
counts, prev_obsolete_sm);
}
space_map_close(prev_obsolete_sm);
/*
* Generate new mapping. Determine what index to continue from
* based on the max offset that we've already written in the
* new mapping.
*/
uint64_t max_offset =
vdev_indirect_mapping_max_offset(sci->sci_new_mapping);
if (max_offset == 0) {
/* We haven't written anything to the new mapping yet. */
start_index = 0;
} else {
/*
* Pick up from where we left off. _entry_for_offset()
* returns a pointer into the vim_entries array. If
* max_offset is greater than any of the mappings
* contained in the table NULL will be returned and
* that indicates we've exhausted our iteration of the
* old_mapping.
*/
vdev_indirect_mapping_entry_phys_t *entry =
vdev_indirect_mapping_entry_for_offset_or_next(old_mapping,
max_offset);
if (entry == NULL) {
/*
* We've already written the whole new mapping.
* This special value will cause us to skip the
* generate_new_mapping step and just do the sync
* task to complete the condense.
*/
start_index = UINT64_MAX;
} else {
start_index = entry - old_mapping->vim_entries;
ASSERT3U(start_index, <,
vdev_indirect_mapping_num_entries(old_mapping));
}
}
spa_condense_indirect_generate_new_mapping(vd, counts,
start_index, zthr);
vdev_indirect_mapping_free_obsolete_counts(old_mapping, counts);
/*
* If the zthr has received a cancellation signal while running
* in generate_new_mapping() or at any point after that, then bail
* early. We don't want to complete the condense if the spa is
* shutting down.
*/
if (zthr_iscancelled(zthr))
return (0);
VERIFY0(dsl_sync_task(spa_name(spa), NULL,
spa_condense_indirect_complete_sync, sci, 0,
ZFS_SPACE_CHECK_EXTRA_RESERVED));
return (0);
thread_exit();
}
int
dsl_scan_cancel(dsl_pool_t *dp)
{
return (dsl_sync_task(spa_name(dp->dp_spa), dsl_scan_cancel_check,
dsl_scan_cancel_sync, NULL, 3));
}
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);
}
void
dataset_kstats_create(dataset_kstats_t *dk, objset_t *objset)
{
/*
* There should not be anything wrong with having kstats for
* snapshots. Since we are not sure how useful they would be
* though nor how much their memory overhead would matter in
* a filesystem with many snapshots, we skip them for now.
*/
if (dmu_objset_is_snapshot(objset))
return;
/*
* At the time of this writing, KSTAT_STRLEN is 255 in Linux,
* and the spa_name can theoretically be up to 256 characters.
* In reality though the spa_name can be 240 characters max
* [see origin directory name check in pool_namecheck()]. Thus,
* the naming scheme for the module name below should not cause
* any truncations. In the event that a truncation does happen
* though, due to some future change, we silently skip creating
* the kstat and log the event.
*/
char kstat_module_name[KSTAT_STRLEN];
int n = snprintf(kstat_module_name, sizeof (kstat_module_name),
"zfs/%s", spa_name(dmu_objset_spa(objset)));
if (n < 0) {
zfs_dbgmsg("failed to create dataset kstat for objset %lld: "
" snprintf() for kstat module name returned %d",
(unsigned long long)dmu_objset_id(objset), n);
return;
} else if (n >= KSTAT_STRLEN) {
zfs_dbgmsg("failed to create dataset kstat for objset %lld: "
"kstat module name length (%d) exceeds limit (%d)",
(unsigned long long)dmu_objset_id(objset),
n, KSTAT_STRLEN);
return;
}
char kstat_name[KSTAT_STRLEN];
n = snprintf(kstat_name, sizeof (kstat_name), "objset-0x%llx",
(unsigned long long)dmu_objset_id(objset));
if (n < 0) {
zfs_dbgmsg("failed to create dataset kstat for objset %lld: "
" snprintf() for kstat name returned %d",
(unsigned long long)dmu_objset_id(objset), n);
return;
}
ASSERT3U(n, <, KSTAT_STRLEN);
kstat_t *kstat = kstat_create(kstat_module_name, 0, kstat_name,
"dataset", KSTAT_TYPE_NAMED,
sizeof (empty_dataset_kstats) / sizeof (kstat_named_t),
KSTAT_FLAG_VIRTUAL);
if (kstat == NULL)
return;
dataset_kstat_values_t *dk_kstats =
kmem_alloc(sizeof (empty_dataset_kstats), KM_SLEEP);
bcopy(&empty_dataset_kstats, dk_kstats,
sizeof (empty_dataset_kstats));
char *ds_name = kmem_zalloc(ZFS_MAX_DATASET_NAME_LEN, KM_SLEEP);
dsl_dataset_name(objset->os_dsl_dataset, ds_name);
KSTAT_NAMED_STR_PTR(&dk_kstats->dkv_ds_name) = ds_name;
KSTAT_NAMED_STR_BUFLEN(&dk_kstats->dkv_ds_name) =
ZFS_MAX_DATASET_NAME_LEN;
kstat->ks_data = dk_kstats;
kstat->ks_update = dataset_kstats_update;
kstat->ks_private = dk;
kstat_install(kstat);
dk->dk_kstats = kstat;
aggsum_init(&dk->dk_aggsums.das_writes, 0);
aggsum_init(&dk->dk_aggsums.das_nwritten, 0);
aggsum_init(&dk->dk_aggsums.das_reads, 0);
aggsum_init(&dk->dk_aggsums.das_nread, 0);
aggsum_init(&dk->dk_aggsums.das_nunlinks, 0);
aggsum_init(&dk->dk_aggsums.das_nunlinked, 0);
}
/*
* Generate the pool's configuration based on the current in-core state.
* We infer whether to generate a complete config or just one top-level config
* based on whether vd is the root vdev.
*/
nvlist_t *
spa_config_generate(spa_t *spa, vdev_t *vd, uint64_t txg, int getstats)
{
nvlist_t *config, *nvroot;
vdev_t *rvd = spa->spa_root_vdev;
unsigned long hostid = 0;
ASSERT(spa_config_held(spa, RW_READER) ||
spa_config_held(spa, RW_WRITER));
if (vd == NULL)
vd = rvd;
/*
* If txg is -1, report the current value of spa->spa_config_txg.
*/
if (txg == -1ULL)
txg = spa->spa_config_txg;
VERIFY(nvlist_alloc(&config, NV_UNIQUE_NAME, KM_SLEEP) == 0);
VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_VERSION,
spa_version(spa)) == 0);
VERIFY(nvlist_add_string(config, ZPOOL_CONFIG_POOL_NAME,
spa_name(spa)) == 0);
VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_POOL_STATE,
spa_state(spa)) == 0);
VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_POOL_TXG,
txg) == 0);
VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_POOL_GUID,
spa_guid(spa)) == 0);
(void) ddi_strtoul(hw_serial, NULL, 10, &hostid);
if (hostid != 0) {
VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_HOSTID,
hostid) == 0);
}
VERIFY(nvlist_add_string(config, ZPOOL_CONFIG_HOSTNAME,
utsname.nodename) == 0);
if (vd != rvd) {
VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_TOP_GUID,
vd->vdev_top->vdev_guid) == 0);
VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_GUID,
vd->vdev_guid) == 0);
if (vd->vdev_isspare)
VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_IS_SPARE,
1ULL) == 0);
if (vd->vdev_islog)
VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_IS_LOG,
1ULL) == 0);
vd = vd->vdev_top; /* label contains top config */
}
nvroot = vdev_config_generate(spa, vd, getstats, B_FALSE, B_FALSE);
VERIFY(nvlist_add_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, nvroot) == 0);
nvlist_free(nvroot);
#ifdef DEBUG
if ( config ) {
printf("spa_config_generate::config=0x%x\n", config);
}
#endif
return (config);
}
int
__dbuf_stats_hash_table_data(char *buf, size_t size, dmu_buf_impl_t *db)
{
arc_buf_info_t abi = { 0 };
dmu_object_info_t doi = { 0 };
dnode_t *dn = DB_DNODE(db);
size_t nwritten;
if (db->db_buf)
arc_buf_info(db->db_buf, &abi, zfs_dbuf_state_index);
if (dn)
__dmu_object_info_from_dnode(dn, &doi);
nwritten = snprintf(buf, size,
"%-16s %-8llu %-8lld %-8lld %-8lld %-8llu %-8llu %-5d %-5d %-5lu | "
"%-5d %-5d 0x%-6x %-6lu %-8llu %-12llu "
"%-6lu %-6lu %-6lu %-6lu %-6lu %-8llu %-8llu %-8d %-5lu | "
"%-6d %-6d %-8lu %-8lu %-6llu %-6lu %-5lu %-8llu %-8llu\n",
/* dmu_buf_impl_t */
spa_name(dn->dn_objset->os_spa),
(u_longlong_t)dmu_objset_id(db->db_objset),
(longlong_t)db->db.db_object,
(longlong_t)db->db_level,
(longlong_t)db->db_blkid,
(u_longlong_t)db->db.db_offset,
(u_longlong_t)db->db.db_size,
!!dbuf_is_metadata(db),
db->db_state,
(ulong_t)refcount_count(&db->db_holds),
/* arc_buf_info_t */
abi.abi_state_type,
abi.abi_state_contents,
abi.abi_flags,
(ulong_t)abi.abi_bufcnt,
(u_longlong_t)abi.abi_size,
(u_longlong_t)abi.abi_access,
(ulong_t)abi.abi_mru_hits,
(ulong_t)abi.abi_mru_ghost_hits,
(ulong_t)abi.abi_mfu_hits,
(ulong_t)abi.abi_mfu_ghost_hits,
(ulong_t)abi.abi_l2arc_hits,
(u_longlong_t)abi.abi_l2arc_dattr,
(u_longlong_t)abi.abi_l2arc_asize,
abi.abi_l2arc_compress,
(ulong_t)abi.abi_holds,
/* dmu_object_info_t */
doi.doi_type,
doi.doi_bonus_type,
(ulong_t)doi.doi_data_block_size,
(ulong_t)doi.doi_metadata_block_size,
(u_longlong_t)doi.doi_bonus_size,
(ulong_t)doi.doi_indirection,
(ulong_t)refcount_count(&dn->dn_holds),
(u_longlong_t)doi.doi_fill_count,
(u_longlong_t)doi.doi_max_offset);
if (nwritten >= size)
return (size);
return (nwritten + 1);
}
static void
zhack_do_feature_ref(int argc, char **argv)
{
char c;
char *target;
boolean_t decr = B_FALSE;
spa_t *spa;
objset_t *mos;
zfeature_info_t feature;
zfeature_info_t *nodeps[] = { NULL };
/*
* fi_desc does not matter here because it was written to disk
* when the feature was enabled, but we need to properly set the
* feature for read or write based on the information we read off
* disk later.
*/
feature.fi_uname = "zhack";
feature.fi_mos = B_FALSE;
feature.fi_desc = NULL;
feature.fi_depends = nodeps;
optind = 1;
while ((c = getopt(argc, argv, "md")) != -1) {
switch (c) {
case 'm':
feature.fi_mos = B_TRUE;
break;
case 'd':
decr = B_TRUE;
break;
default:
usage();
break;
}
}
argc -= optind;
argv += optind;
if (argc < 2) {
(void) fprintf(stderr, "error: missing feature or pool name\n");
usage();
}
target = argv[0];
feature.fi_guid = argv[1];
if (!zfeature_is_valid_guid(feature.fi_guid))
fatal("invalid feature guid: %s", feature.fi_guid);
zhack_spa_open(target, B_FALSE, FTAG, &spa);
mos = spa->spa_meta_objset;
if (0 == zfeature_lookup_guid(feature.fi_guid, NULL))
fatal("'%s' is a real feature, will not change refcount");
if (0 == zap_contains(mos, spa->spa_feat_for_read_obj,
feature.fi_guid)) {
feature.fi_can_readonly = B_FALSE;
} else if (0 == zap_contains(mos, spa->spa_feat_for_write_obj,
feature.fi_guid)) {
feature.fi_can_readonly = B_TRUE;
} else {
fatal("feature is not enabled: %s", feature.fi_guid);
}
if (decr && !spa_feature_is_active(spa, &feature))
fatal("feature refcount already 0: %s", feature.fi_guid);
VERIFY0(dsl_sync_task(spa_name(spa), NULL,
decr ? feature_decr_sync : feature_incr_sync, &feature, 5));
spa_close(spa, FTAG);
}
/*
* Generate the pool's configuration based on the current in-core state.
*
* We infer whether to generate a complete config or just one top-level config
* based on whether vd is the root vdev.
*/
nvlist_t *
spa_config_generate(spa_t *spa, vdev_t *vd, uint64_t txg, int getstats)
{
nvlist_t *config, *nvroot;
vdev_t *rvd = spa->spa_root_vdev;
unsigned long hostid = 0;
boolean_t locked = B_FALSE;
uint64_t split_guid;
char *pool_name;
if (vd == NULL) {
vd = rvd;
locked = B_TRUE;
spa_config_enter(spa, SCL_CONFIG | SCL_STATE, FTAG, RW_READER);
}
ASSERT(spa_config_held(spa, SCL_CONFIG | SCL_STATE, RW_READER) ==
(SCL_CONFIG | SCL_STATE));
/*
* If txg is -1, report the current value of spa->spa_config_txg.
*/
if (txg == -1ULL)
txg = spa->spa_config_txg;
/*
* Originally, users had to handle spa namespace collisions by either
* exporting the already imported pool or by specifying a new name for
* the pool with a conflicting name. In the case of root pools from
* virtual guests, neither approach to collision resolution is
* reasonable. This is addressed by extending the new name syntax with
* an option to specify that the new name is temporary. When specified,
* ZFS_IMPORT_TEMP_NAME will be set in spa->spa_import_flags to tell us
* to use the previous name, which we do below.
*/
if (spa->spa_import_flags & ZFS_IMPORT_TEMP_NAME) {
VERIFY0(nvlist_lookup_string(spa->spa_config,
ZPOOL_CONFIG_POOL_NAME, &pool_name));
} else
pool_name = spa_name(spa);
VERIFY(nvlist_alloc(&config, NV_UNIQUE_NAME, KM_SLEEP) == 0);
VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_VERSION,
spa_version(spa)) == 0);
VERIFY(nvlist_add_string(config, ZPOOL_CONFIG_POOL_NAME,
pool_name) == 0);
VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_POOL_STATE,
spa_state(spa)) == 0);
VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_POOL_TXG,
txg) == 0);
VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_POOL_GUID,
spa_guid(spa)) == 0);
VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_ERRATA,
spa->spa_errata) == 0);
VERIFY(spa->spa_comment == NULL || nvlist_add_string(config,
ZPOOL_CONFIG_COMMENT, spa->spa_comment) == 0);
#ifdef _KERNEL
hostid = zone_get_hostid(NULL);
#else /* _KERNEL */
/*
* We're emulating the system's hostid in userland, so we can't use
* zone_get_hostid().
*/
(void) ddi_strtoul(hw_serial, NULL, 10, &hostid);
#endif /* _KERNEL */
if (hostid != 0) {
VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_HOSTID,
hostid) == 0);
}
VERIFY0(nvlist_add_string(config, ZPOOL_CONFIG_HOSTNAME,
utsname()->nodename));
if (vd != rvd) {
VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_TOP_GUID,
vd->vdev_top->vdev_guid) == 0);
VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_GUID,
vd->vdev_guid) == 0);
if (vd->vdev_isspare)
VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_IS_SPARE,
1ULL) == 0);
if (vd->vdev_islog)
VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_IS_LOG,
1ULL) == 0);
vd = vd->vdev_top; /* label contains top config */
} else {
/*
* Only add the (potentially large) split information
* in the mos config, and not in the vdev labels
*/
if (spa->spa_config_splitting != NULL)
VERIFY(nvlist_add_nvlist(config, ZPOOL_CONFIG_SPLIT,
spa->spa_config_splitting) == 0);
}
/*
* Add the top-level config. We even add this on pools which
* don't support holes in the namespace.
*/
vdev_top_config_generate(spa, config);
/*
* If we're splitting, record the original pool's guid.
*/
if (spa->spa_config_splitting != NULL &&
nvlist_lookup_uint64(spa->spa_config_splitting,
ZPOOL_CONFIG_SPLIT_GUID, &split_guid) == 0) {
VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_SPLIT_GUID,
split_guid) == 0);
}
nvroot = vdev_config_generate(spa, vd, getstats, 0);
VERIFY(nvlist_add_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, nvroot) == 0);
nvlist_free(nvroot);
/*
* Store what's necessary for reading the MOS in the label.
*/
VERIFY(nvlist_add_nvlist(config, ZPOOL_CONFIG_FEATURES_FOR_READ,
spa->spa_label_features) == 0);
if (getstats && spa_load_state(spa) == SPA_LOAD_NONE) {
ddt_histogram_t *ddh;
ddt_stat_t *dds;
ddt_object_t *ddo;
ddh = kmem_zalloc(sizeof (ddt_histogram_t), KM_SLEEP);
ddt_get_dedup_histogram(spa, ddh);
VERIFY(nvlist_add_uint64_array(config,
ZPOOL_CONFIG_DDT_HISTOGRAM,
(uint64_t *)ddh, sizeof (*ddh) / sizeof (uint64_t)) == 0);
kmem_free(ddh, sizeof (ddt_histogram_t));
ddo = kmem_zalloc(sizeof (ddt_object_t), KM_SLEEP);
ddt_get_dedup_object_stats(spa, ddo);
VERIFY(nvlist_add_uint64_array(config,
ZPOOL_CONFIG_DDT_OBJ_STATS,
(uint64_t *)ddo, sizeof (*ddo) / sizeof (uint64_t)) == 0);
kmem_free(ddo, sizeof (ddt_object_t));
dds = kmem_zalloc(sizeof (ddt_stat_t), KM_SLEEP);
ddt_get_dedup_stats(spa, dds);
VERIFY(nvlist_add_uint64_array(config,
ZPOOL_CONFIG_DDT_STATS,
(uint64_t *)dds, sizeof (*dds) / sizeof (uint64_t)) == 0);
kmem_free(dds, sizeof (ddt_stat_t));
}
if (locked)
spa_config_exit(spa, SCL_CONFIG | SCL_STATE, FTAG);
return (config);
}
/*
* The full semantics of this function are described in the comment above
* lzc_release().
*
* To summarize:
* Releases holds specified in the nvl holds.
*
* holds is nvl of snapname -> { holdname, ... }
* errlist will be filled in with snapname -> error
*
* If tmpdp is not NULL the names for holds should be the dsobj's of snapshots,
* otherwise they should be the names of shapshots.
*
* As a release may cause snapshots to be destroyed this trys to ensure they
* aren't mounted.
*
* The release of non-existent holds are skipped.
*
* At least one hold must have been released for the this function to succeed
* and return 0.
*/
static int
dsl_dataset_user_release_impl(nvlist_t *holds, nvlist_t *errlist,
dsl_pool_t *tmpdp)
{
dsl_dataset_user_release_arg_t ddura;
nvpair_t *pair;
char *pool;
int error;
pair = nvlist_next_nvpair(holds, NULL);
if (pair == NULL)
return (0);
/*
* The release may cause snapshots to be destroyed; make sure they
* are not mounted.
*/
if (tmpdp != NULL) {
/* Temporary holds are specified by dsobj string. */
ddura.ddura_holdfunc = dsl_dataset_hold_obj_string;
pool = spa_name(tmpdp->dp_spa);
#ifdef _KERNEL
for (pair = nvlist_next_nvpair(holds, NULL); pair != NULL;
pair = nvlist_next_nvpair(holds, pair)) {
dsl_dataset_t *ds;
dsl_pool_config_enter(tmpdp, FTAG);
error = dsl_dataset_hold_obj_string(tmpdp,
nvpair_name(pair), FTAG, &ds);
if (error == 0) {
char name[MAXNAMELEN];
dsl_dataset_name(ds, name);
dsl_pool_config_exit(tmpdp, FTAG);
dsl_dataset_rele(ds, FTAG);
(void) zfs_unmount_snap(name);
} else {
dsl_pool_config_exit(tmpdp, FTAG);
}
}
#endif
} else {
/* Non-temporary holds are specified by name. */
ddura.ddura_holdfunc = dsl_dataset_hold;
pool = nvpair_name(pair);
#ifdef _KERNEL
for (pair = nvlist_next_nvpair(holds, NULL); pair != NULL;
pair = nvlist_next_nvpair(holds, pair)) {
(void) zfs_unmount_snap(nvpair_name(pair));
}
#endif
}
ddura.ddura_holds = holds;
ddura.ddura_errlist = errlist;
ddura.ddura_todelete = fnvlist_alloc();
ddura.ddura_chkholds = fnvlist_alloc();
error = dsl_sync_task(pool, dsl_dataset_user_release_check,
dsl_dataset_user_release_sync, &ddura, 0);
fnvlist_free(ddura.ddura_todelete);
fnvlist_free(ddura.ddura_chkholds);
return (error);
}
/*
* Generate the pool's configuration based on the current in-core state.
*
* We infer whether to generate a complete config or just one top-level config
* based on whether vd is the root vdev.
*/
nvlist_t *
spa_config_generate(spa_t *spa, vdev_t *vd, uint64_t txg, int getstats)
{
nvlist_t *config, *nvroot;
vdev_t *rvd = spa->spa_root_vdev;
unsigned long hostid = 0;
boolean_t locked = B_FALSE;
uint64_t split_guid;
if (vd == NULL) {
vd = rvd;
locked = B_TRUE;
spa_config_enter(spa, SCL_CONFIG | SCL_STATE, FTAG, RW_READER);
}
ASSERT(spa_config_held(spa, SCL_CONFIG | SCL_STATE, RW_READER) ==
(SCL_CONFIG | SCL_STATE));
/*
* If txg is -1, report the current value of spa->spa_config_txg.
*/
if (txg == -1ULL)
txg = spa->spa_config_txg;
VERIFY(nvlist_alloc(&config, NV_UNIQUE_NAME, KM_SLEEP) == 0);
VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_VERSION,
spa_version(spa)) == 0);
VERIFY(nvlist_add_string(config, ZPOOL_CONFIG_POOL_NAME,
spa_name(spa)) == 0);
VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_POOL_STATE,
spa_state(spa)) == 0);
VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_POOL_TXG,
txg) == 0);
VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_POOL_GUID,
spa_guid(spa)) == 0);
VERIFY(spa->spa_comment == NULL || nvlist_add_string(config,
ZPOOL_CONFIG_COMMENT, spa->spa_comment) == 0);
#ifdef _KERNEL
hostid = zone_get_hostid(NULL);
#else /* _KERNEL */
/*
* We're emulating the system's hostid in userland, so we can't use
* zone_get_hostid().
*/
(void) ddi_strtoul(hw_serial, NULL, 10, &hostid);
#endif /* _KERNEL */
if (hostid != 0) {
VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_HOSTID,
hostid) == 0);
}
VERIFY(nvlist_add_string(config, ZPOOL_CONFIG_HOSTNAME,
utsname.nodename) == 0);
if (vd != rvd) {
VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_TOP_GUID,
vd->vdev_top->vdev_guid) == 0);
VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_GUID,
vd->vdev_guid) == 0);
if (vd->vdev_isspare)
VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_IS_SPARE,
1ULL) == 0);
if (vd->vdev_islog)
VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_IS_LOG,
1ULL) == 0);
vd = vd->vdev_top; /* label contains top config */
} else {
/*
* Only add the (potentially large) split information
* in the mos config, and not in the vdev labels
*/
if (spa->spa_config_splitting != NULL)
VERIFY(nvlist_add_nvlist(config, ZPOOL_CONFIG_SPLIT,
spa->spa_config_splitting) == 0);
}
/*
* Add the top-level config. We even add this on pools which
* don't support holes in the namespace.
*/
vdev_top_config_generate(spa, config);
/*
* If we're splitting, record the original pool's guid.
*/
if (spa->spa_config_splitting != NULL &&
nvlist_lookup_uint64(spa->spa_config_splitting,
ZPOOL_CONFIG_SPLIT_GUID, &split_guid) == 0) {
VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_SPLIT_GUID,
split_guid) == 0);
}
nvroot = vdev_config_generate(spa, vd, getstats, 0);
VERIFY(nvlist_add_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, nvroot) == 0);
nvlist_free(nvroot);
/*
* Store what's necessary for reading the MOS in the label.
*/
VERIFY(nvlist_add_nvlist(config, ZPOOL_CONFIG_FEATURES_FOR_READ,
spa->spa_label_features) == 0);
if (getstats && spa_load_state(spa) == SPA_LOAD_NONE) {
ddt_histogram_t *ddh;
ddt_stat_t *dds;
ddt_object_t *ddo;
ddh = kmem_zalloc(sizeof (ddt_histogram_t), KM_SLEEP);
ddt_get_dedup_histogram(spa, ddh);
VERIFY(nvlist_add_uint64_array(config,
ZPOOL_CONFIG_DDT_HISTOGRAM,
(uint64_t *)ddh, sizeof (*ddh) / sizeof (uint64_t)) == 0);
kmem_free(ddh, sizeof (ddt_histogram_t));
ddo = kmem_zalloc(sizeof (ddt_object_t), KM_SLEEP);
ddt_get_dedup_object_stats(spa, ddo);
VERIFY(nvlist_add_uint64_array(config,
ZPOOL_CONFIG_DDT_OBJ_STATS,
(uint64_t *)ddo, sizeof (*ddo) / sizeof (uint64_t)) == 0);
kmem_free(ddo, sizeof (ddt_object_t));
dds = kmem_zalloc(sizeof (ddt_stat_t), KM_SLEEP);
ddt_get_dedup_stats(spa, dds);
VERIFY(nvlist_add_uint64_array(config,
ZPOOL_CONFIG_DDT_STATS,
(uint64_t *)dds, sizeof (*dds) / sizeof (uint64_t)) == 0);
kmem_free(dds, sizeof (ddt_stat_t));
}
if (locked)
spa_config_exit(spa, SCL_CONFIG | SCL_STATE, FTAG);
return (config);
}
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);
}
/*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));
}
/*
* Generate the pool's configuration based on the current in-core state.
* We infer whether to generate a complete config or just one top-level config
* based on whether vd is the root vdev.
*/
nvlist_t *
spa_config_generate(spa_t *spa, vdev_t *vd, uint64_t txg, int getstats)
{
nvlist_t *config, *nvroot;
vdev_t *rvd = spa->spa_root_vdev;
unsigned long hostid = 0;
boolean_t locked = B_FALSE;
if (vd == NULL) {
vd = rvd;
locked = B_TRUE;
spa_config_enter(spa, SCL_CONFIG | SCL_STATE, FTAG, RW_READER);
}
ASSERT(spa_config_held(spa, SCL_CONFIG | SCL_STATE, RW_READER) ==
(SCL_CONFIG | SCL_STATE));
/*
* If txg is -1, report the current value of spa->spa_config_txg.
*/
if (txg == -1ULL)
txg = spa->spa_config_txg;
VERIFY(nvlist_alloc(&config, NV_UNIQUE_NAME, KM_SLEEP) == 0);
VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_VERSION,
spa_version(spa)) == 0);
VERIFY(nvlist_add_string(config, ZPOOL_CONFIG_POOL_NAME,
spa_name(spa)) == 0);
VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_POOL_STATE,
spa_state(spa)) == 0);
VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_POOL_TXG,
txg) == 0);
VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_POOL_GUID,
spa_guid(spa)) == 0);
#ifdef _KERNEL
hostid = zone_get_hostid(NULL);
#else /* _KERNEL */
/*
* We're emulating the system's hostid in userland, so we can't use
* zone_get_hostid().
*/
(void) ddi_strtoul(hw_serial, NULL, 10, &hostid);
#endif /* _KERNEL */
if (hostid != 0) {
VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_HOSTID,
hostid) == 0);
}
VERIFY(nvlist_add_string(config, ZPOOL_CONFIG_HOSTNAME,
utsname.nodename) == 0);
if (vd != rvd) {
VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_TOP_GUID,
vd->vdev_top->vdev_guid) == 0);
VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_GUID,
vd->vdev_guid) == 0);
if (vd->vdev_isspare)
VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_IS_SPARE,
1ULL) == 0);
if (vd->vdev_islog)
VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_IS_LOG,
1ULL) == 0);
vd = vd->vdev_top; /* label contains top config */
}
nvroot = vdev_config_generate(spa, vd, getstats, B_FALSE, B_FALSE);
VERIFY(nvlist_add_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, nvroot) == 0);
nvlist_free(nvroot);
if (locked)
spa_config_exit(spa, SCL_CONFIG | SCL_STATE, FTAG);
return (config);
}
/*
* Return the dsl_dir_t, and possibly the last component which couldn't
* be found in *tail. The name must be in the specified dsl_pool_t. This
* thread must hold the dp_config_rwlock for the pool. Returns NULL if the
* path is bogus, or if tail==NULL and we couldn't parse the whole name.
* (*tail)[0] == '@' means that the last component is a snapshot.
*/
int
dsl_dir_hold(dsl_pool_t *dp, const char *name, void *tag,
dsl_dir_t **ddp, const char **tailp)
{
char buf[MAXNAMELEN];
const char *spaname, *next, *nextnext = NULL;
int err;
dsl_dir_t *dd;
uint64_t ddobj;
err = getcomponent(name, buf, &next);
if (err != 0)
return (err);
/* Make sure the name is in the specified pool. */
spaname = spa_name(dp->dp_spa);
if (strcmp(buf, spaname) != 0)
return (SET_ERROR(EINVAL));
ASSERT(dsl_pool_config_held(dp));
err = dsl_dir_hold_obj(dp, dp->dp_root_dir_obj, NULL, tag, &dd);
if (err != 0) {
return (err);
}
while (next != NULL) {
dsl_dir_t *child_ds;
err = getcomponent(next, buf, &nextnext);
if (err != 0)
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 != 0) {
if (err == ENOENT)
err = 0;
break;
}
err = dsl_dir_hold_obj(dp, ddobj, buf, tag, &child_ds);
if (err != 0)
break;
dsl_dir_rele(dd, tag);
dd = child_ds;
next = nextnext;
}
if (err != 0) {
dsl_dir_rele(dd, tag);
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_rele(dd, tag);
dprintf("next=%p (%s) tail=%p\n", next, next?next:"", tailp);
err = SET_ERROR(ENOENT);
}
if (tailp != NULL)
*tailp = next;
*ddp = dd;
return (err);
}
static void
zhack_do_feature_enable(int argc, char **argv)
{
char c;
char *desc, *target;
spa_t *spa;
objset_t *mos;
zfeature_info_t feature;
spa_feature_t nodeps[] = { SPA_FEATURE_NONE };
/*
* Features are not added to the pool's label until their refcounts
* are incremented, so fi_mos can just be left as false for now.
*/
desc = NULL;
feature.fi_uname = "zhack";
feature.fi_flags = 0;
feature.fi_depends = nodeps;
feature.fi_feature = SPA_FEATURE_NONE;
optind = 1;
while ((c = getopt(argc, argv, "rmd:")) != -1) {
switch (c) {
case 'r':
feature.fi_flags |= ZFEATURE_FLAG_READONLY_COMPAT;
break;
case 'd':
desc = strdup(optarg);
break;
default:
usage();
break;
}
}
if (desc == NULL)
desc = strdup("zhack injected");
feature.fi_desc = desc;
argc -= optind;
argv += optind;
if (argc < 2) {
(void) fprintf(stderr, "error: missing feature or pool name\n");
usage();
}
target = argv[0];
feature.fi_guid = argv[1];
if (!zfeature_is_valid_guid(feature.fi_guid))
fatal(NULL, FTAG, "invalid feature guid: %s", feature.fi_guid);
zhack_spa_open(target, B_FALSE, FTAG, &spa);
mos = spa->spa_meta_objset;
if (zfeature_is_supported(feature.fi_guid))
fatal(spa, FTAG, "'%s' is a real feature, will not enable");
if (0 == zap_contains(mos, spa->spa_feat_desc_obj, feature.fi_guid))
fatal(spa, FTAG, "feature already enabled: %s",
feature.fi_guid);
VERIFY0(dsl_sync_task(spa_name(spa), NULL,
zhack_feature_enable_sync, &feature, 5, ZFS_SPACE_CHECK_NORMAL));
spa_close(spa, FTAG);
free(desc);
}
static int
vdev_file_io_start(zio_t *zio)
{
vdev_t *vd = zio->io_vd;
vdev_file_t *vf = vd->vdev_tsd;
#ifdef LINUX_AIO
struct iocb *iocbp = &zio->io_aio;
#endif
ssize_t resid;
int error;
if (zio->io_type == ZIO_TYPE_IOCTL) {
zio_vdev_io_bypass(zio);
/* XXPOLICY */
if (!vdev_readable(vd)) {
zio->io_error = ENXIO;
return (ZIO_PIPELINE_CONTINUE);
}
switch (zio->io_cmd) {
case DKIOCFLUSHWRITECACHE:
if (zfs_nocacheflush)
break;
/* This doesn't actually do much with O_DIRECT... */
zio->io_error = VOP_FSYNC(vf->vf_vnode, FSYNC | FDSYNC,
kcred, NULL);
dprintf("fsync(%s) = %d\n", vdev_description(vd),
zio->io_error);
if (vd->vdev_nowritecache) {
zio->io_error = ENOTSUP;
break;
}
/* Flush the write cache */
error = flushwc(vf->vf_vnode);
dprintf("flushwc(%s) = %d\n", vdev_description(vd),
error);
if (error) {
#ifdef _KERNEL
cmn_err(CE_WARN, "Failed to flush write cache "
"on device '%s'. Data on pool '%s' may be lost "
"if power fails. No further warnings will "
"be given.", vdev_description(vd),
spa_name(vd->vdev_spa));
#endif
vd->vdev_nowritecache = B_TRUE;
zio->io_error = error;
}
break;
default:
zio->io_error = ENOTSUP;
}
return (ZIO_PIPELINE_CONTINUE);
}
/*
* In the kernel, don't bother double-caching, but in userland,
* we want to test the vdev_cache code.
*/
if (zio->io_type == ZIO_TYPE_READ && vdev_cache_read(zio) == 0)
return (ZIO_PIPELINE_STOP);
if ((zio = vdev_queue_io(zio)) == NULL)
return (ZIO_PIPELINE_STOP);
/* XXPOLICY */
if (zio->io_type == ZIO_TYPE_WRITE)
error = vdev_writeable(vd) ? vdev_error_inject(vd, zio) : ENXIO;
else
error = vdev_readable(vd) ? vdev_error_inject(vd, zio) : ENXIO;
error = (vd->vdev_remove_wanted || vd->vdev_is_failing) ? ENXIO : error;
if (error) {
zio->io_error = error;
zio_interrupt(zio);
return (ZIO_PIPELINE_STOP);
}
#ifdef LINUX_AIO
if (zio->io_aio_ctx && zio->io_aio_ctx->zac_enabled) {
if (zio->io_type == ZIO_TYPE_READ)
io_prep_pread(&zio->io_aio, vf->vf_vnode->v_fd,
zio->io_data, zio->io_size, zio->io_offset);
else
io_prep_pwrite(&zio->io_aio, vf->vf_vnode->v_fd,
zio->io_data, zio->io_size, zio->io_offset);
zio->io_aio.data = zio;
do {
error = io_submit(zio->io_aio_ctx->zac_ctx, 1, &iocbp);
} while (error == -EINTR);
if (error < 0) {
zio->io_error = -error;
zio_interrupt(zio);
} else
VERIFY(error == 1);
return (ZIO_PIPELINE_STOP);
}
#endif
zio->io_error = vn_rdwr(zio->io_type == ZIO_TYPE_READ ?
UIO_READ : UIO_WRITE, vf->vf_vnode, zio->io_data,
zio->io_size, zio->io_offset, UIO_SYSSPACE,
0, RLIM64_INFINITY, kcred, &resid);
if (resid != 0 && zio->io_error == 0)
zio->io_error = ENOSPC;
zio_interrupt(zio);
return (ZIO_PIPELINE_STOP);
}
static void
zhack_do_feature_ref(int argc, char **argv)
{
char c;
char *target;
boolean_t decr = B_FALSE;
spa_t *spa;
objset_t *mos;
zfeature_info_t feature;
spa_feature_t nodeps[] = { SPA_FEATURE_NONE };
/*
* fi_desc does not matter here because it was written to disk
* when the feature was enabled, but we need to properly set the
* feature for read or write based on the information we read off
* disk later.
*/
feature.fi_uname = "zhack";
feature.fi_flags = 0;
feature.fi_desc = NULL;
feature.fi_depends = nodeps;
feature.fi_feature = SPA_FEATURE_NONE;
optind = 1;
while ((c = getopt(argc, argv, "md")) != -1) {
switch (c) {
case 'm':
feature.fi_flags |= ZFEATURE_FLAG_MOS;
break;
case 'd':
decr = B_TRUE;
break;
default:
usage();
break;
}
}
argc -= optind;
argv += optind;
if (argc < 2) {
(void) fprintf(stderr, "error: missing feature or pool name\n");
usage();
}
target = argv[0];
feature.fi_guid = argv[1];
if (!zfeature_is_valid_guid(feature.fi_guid))
fatal(NULL, FTAG, "invalid feature guid: %s", feature.fi_guid);
zhack_spa_open(target, B_FALSE, FTAG, &spa);
mos = spa->spa_meta_objset;
if (zfeature_is_supported(feature.fi_guid)) {
fatal(spa, FTAG,
"'%s' is a real feature, will not change refcount");
}
if (0 == zap_contains(mos, spa->spa_feat_for_read_obj,
feature.fi_guid)) {
feature.fi_flags &= ~ZFEATURE_FLAG_READONLY_COMPAT;
} else if (0 == zap_contains(mos, spa->spa_feat_for_write_obj,
feature.fi_guid)) {
feature.fi_flags |= ZFEATURE_FLAG_READONLY_COMPAT;
} else {
fatal(spa, FTAG, "feature is not enabled: %s", feature.fi_guid);
}
if (decr) {
uint64_t count;
if (feature_get_refcount_from_disk(spa, &feature,
&count) == 0 && count == 0) {
fatal(spa, FTAG, "feature refcount already 0: %s",
feature.fi_guid);
}
}
VERIFY0(dsl_sync_task(spa_name(spa), NULL,
decr ? feature_decr_sync : feature_incr_sync, &feature,
5, ZFS_SPACE_CHECK_NORMAL));
spa_close(spa, FTAG);
}
