#include <sys/zfs_znode.h>
#include <sys/spa_impl.h>
#include <sys/dsl_deadlist.h>
+#include <sys/vdev_impl.h>
+#include <sys/metaslab_impl.h>
#include <sys/bptree.h>
#include <sys/zfeature.h>
#include <sys/zil_impl.h>
#include <sys/dsl_userhold.h>
#include <sys/trace_txg.h>
+#include <sys/mmp.h>
/*
* ZFS Write Throttle
* zfs_dirty_data_max determines the dirty space limit. Once that value is
* exceeded, new writes are halted until space frees up.
*
- * The zfs_dirty_data_sync tunable dictates the threshold at which we
+ * The zfs_dirty_data_sync_percent tunable dictates the threshold at which we
* ensure that there is a txg syncing (see the comment in txg.c for a full
* description of transaction group stages).
*
int zfs_dirty_data_max_max_percent = 25;
/*
- * If there is at least this much dirty data, push out a txg.
+ * If there's at least this much dirty data (as a percentage of
+ * zfs_dirty_data_max), push out a txg. This should be less than
+ * zfs_vdev_async_write_active_min_dirty_percent.
*/
-unsigned long zfs_dirty_data_sync = 64 * 1024 * 1024;
+int zfs_dirty_data_sync_percent = 20;
/*
* Once there is this amount of dirty data, the dmu_tx_delay() will kick in
*/
unsigned long zfs_delay_scale = 1000 * 1000 * 1000 / 2000;
-hrtime_t zfs_throttle_delay = MSEC2NSEC(10);
-hrtime_t zfs_throttle_resolution = MSEC2NSEC(10);
-
/*
* This determines the number of threads used by the dp_sync_taskq.
*/
int zfs_sync_taskq_batch_pct = 75;
+/*
+ * These tunables determine the behavior of how zil_itxg_clean() is
+ * called via zil_clean() in the context of spa_sync(). When an itxg
+ * list needs to be cleaned, TQ_NOSLEEP will be used when dispatching.
+ * If the dispatch fails, the call to zil_itxg_clean() will occur
+ * synchronously in the context of spa_sync(), which can negatively
+ * impact the performance of spa_sync() (e.g. in the case of the itxg
+ * list having a large number of itxs that needs to be cleaned).
+ *
+ * Thus, these tunables can be used to manipulate the behavior of the
+ * taskq used by zil_clean(); they determine the number of taskq entries
+ * that are pre-populated when the taskq is first created (via the
+ * "zfs_zil_clean_taskq_minalloc" tunable) and the maximum number of
+ * taskq entries that are cached after an on-demand allocation (via the
+ * "zfs_zil_clean_taskq_maxalloc").
+ *
+ * The idea being, we want to try reasonably hard to ensure there will
+ * already be a taskq entry pre-allocated by the time that it is needed
+ * by zil_clean(). This way, we can avoid the possibility of an
+ * on-demand allocation of a new taskq entry from failing, which would
+ * result in zil_itxg_clean() being called synchronously from zil_clean()
+ * (which can adversely affect performance of spa_sync()).
+ *
+ * Additionally, the number of threads used by the taskq can be
+ * configured via the "zfs_zil_clean_taskq_nthr_pct" tunable.
+ */
+int zfs_zil_clean_taskq_nthr_pct = 100;
+int zfs_zil_clean_taskq_minalloc = 1024;
+int zfs_zil_clean_taskq_maxalloc = 1024 * 1024;
+
int
dsl_pool_open_special_dir(dsl_pool_t *dp, const char *name, dsl_dir_t **ddp)
{
dp->dp_meta_rootbp = *bp;
rrw_init(&dp->dp_config_rwlock, B_TRUE);
txg_init(dp, txg);
+ mmp_init(spa);
- txg_list_create(&dp->dp_dirty_datasets,
+ txg_list_create(&dp->dp_dirty_datasets, spa,
offsetof(dsl_dataset_t, ds_dirty_link));
- txg_list_create(&dp->dp_dirty_zilogs,
+ txg_list_create(&dp->dp_dirty_zilogs, spa,
offsetof(zilog_t, zl_dirty_link));
- txg_list_create(&dp->dp_dirty_dirs,
+ txg_list_create(&dp->dp_dirty_dirs, spa,
offsetof(dsl_dir_t, dd_dirty_link));
- txg_list_create(&dp->dp_sync_tasks,
+ txg_list_create(&dp->dp_sync_tasks, spa,
+ offsetof(dsl_sync_task_t, dst_node));
+ txg_list_create(&dp->dp_early_sync_tasks, spa,
offsetof(dsl_sync_task_t, dst_node));
dp->dp_sync_taskq = taskq_create("dp_sync_taskq",
zfs_sync_taskq_batch_pct, minclsyspri, 1, INT_MAX,
TASKQ_THREADS_CPU_PCT);
+ dp->dp_zil_clean_taskq = taskq_create("dp_zil_clean_taskq",
+ zfs_zil_clean_taskq_nthr_pct, minclsyspri,
+ zfs_zil_clean_taskq_minalloc,
+ zfs_zil_clean_taskq_maxalloc,
+ TASKQ_PREPOPULATE | TASKQ_THREADS_CPU_PCT);
+
mutex_init(&dp->dp_lock, NULL, MUTEX_DEFAULT, NULL);
cv_init(&dp->dp_spaceavail_cv, NULL, CV_DEFAULT, NULL);
dp->dp_iput_taskq = taskq_create("z_iput", max_ncpus, defclsyspri,
max_ncpus * 8, INT_MAX, TASKQ_PREPOPULATE | TASKQ_DYNAMIC);
+ dp->dp_unlinked_drain_taskq = taskq_create("z_unlinked_drain",
+ max_ncpus, defclsyspri, max_ncpus, INT_MAX,
+ TASKQ_PREPOPULATE | TASKQ_DYNAMIC);
return (dp);
}
dp->dp_meta_objset, obj));
}
+ if (spa_feature_is_active(dp->dp_spa, SPA_FEATURE_OBSOLETE_COUNTS)) {
+ err = zap_lookup(dp->dp_meta_objset, DMU_POOL_DIRECTORY_OBJECT,
+ DMU_POOL_OBSOLETE_BPOBJ, sizeof (uint64_t), 1, &obj);
+ if (err == 0) {
+ VERIFY0(bpobj_open(&dp->dp_obsolete_bpobj,
+ dp->dp_meta_objset, obj));
+ } else if (err == ENOENT) {
+ /*
+ * We might not have created the remap bpobj yet.
+ */
+ err = 0;
+ } else {
+ goto out;
+ }
+ }
+
/*
- * Note: errors ignored, because the leak dir will not exist if we
- * have not encountered a leak yet.
+ * Note: errors ignored, because the these special dirs, used for
+ * space accounting, are only created on demand.
*/
(void) dsl_pool_open_special_dir(dp, LEAK_DIR_NAME,
&dp->dp_leak_dir);
* includes pool-opening context), it actually only got a "ref"
* and not a hold, so just drop that here.
*/
- if (dp->dp_origin_snap)
+ if (dp->dp_origin_snap != NULL)
dsl_dataset_rele(dp->dp_origin_snap, dp);
- if (dp->dp_mos_dir)
+ if (dp->dp_mos_dir != NULL)
dsl_dir_rele(dp->dp_mos_dir, dp);
- if (dp->dp_free_dir)
+ if (dp->dp_free_dir != NULL)
dsl_dir_rele(dp->dp_free_dir, dp);
- if (dp->dp_leak_dir)
+ if (dp->dp_leak_dir != NULL)
dsl_dir_rele(dp->dp_leak_dir, dp);
- if (dp->dp_root_dir)
+ if (dp->dp_root_dir != NULL)
dsl_dir_rele(dp->dp_root_dir, dp);
bpobj_close(&dp->dp_free_bpobj);
+ bpobj_close(&dp->dp_obsolete_bpobj);
/* undo the dmu_objset_open_impl(mos) from dsl_pool_open() */
- if (dp->dp_meta_objset)
+ if (dp->dp_meta_objset != NULL)
dmu_objset_evict(dp->dp_meta_objset);
txg_list_destroy(&dp->dp_dirty_datasets);
txg_list_destroy(&dp->dp_dirty_zilogs);
txg_list_destroy(&dp->dp_sync_tasks);
+ txg_list_destroy(&dp->dp_early_sync_tasks);
txg_list_destroy(&dp->dp_dirty_dirs);
+ taskq_destroy(dp->dp_zil_clean_taskq);
taskq_destroy(dp->dp_sync_taskq);
/*
*/
arc_flush(dp->dp_spa, FALSE);
+ mmp_fini(dp->dp_spa);
txg_fini(dp);
dsl_scan_fini(dp);
dmu_buf_user_evict_wait();
rrw_destroy(&dp->dp_config_rwlock);
mutex_destroy(&dp->dp_lock);
+ cv_destroy(&dp->dp_spaceavail_cv);
+ taskq_destroy(dp->dp_unlinked_drain_taskq);
taskq_destroy(dp->dp_iput_taskq);
- if (dp->dp_blkstats)
+ if (dp->dp_blkstats != NULL) {
+ mutex_destroy(&dp->dp_blkstats->zab_lock);
vmem_free(dp->dp_blkstats, sizeof (zfs_all_blkstats_t));
+ }
kmem_free(dp, sizeof (dsl_pool_t));
}
+void
+dsl_pool_create_obsolete_bpobj(dsl_pool_t *dp, dmu_tx_t *tx)
+{
+ uint64_t obj;
+ /*
+ * Currently, we only create the obsolete_bpobj where there are
+ * indirect vdevs with referenced mappings.
+ */
+ ASSERT(spa_feature_is_active(dp->dp_spa, SPA_FEATURE_DEVICE_REMOVAL));
+ /* create and open the obsolete_bpobj */
+ obj = bpobj_alloc(dp->dp_meta_objset, SPA_OLD_MAXBLOCKSIZE, tx);
+ VERIFY0(bpobj_open(&dp->dp_obsolete_bpobj, dp->dp_meta_objset, obj));
+ VERIFY0(zap_add(dp->dp_meta_objset, DMU_POOL_DIRECTORY_OBJECT,
+ DMU_POOL_OBSOLETE_BPOBJ, sizeof (uint64_t), 1, &obj, tx));
+ spa_feature_incr(dp->dp_spa, SPA_FEATURE_OBSOLETE_COUNTS, tx);
+}
+
+void
+dsl_pool_destroy_obsolete_bpobj(dsl_pool_t *dp, dmu_tx_t *tx)
+{
+ spa_feature_decr(dp->dp_spa, SPA_FEATURE_OBSOLETE_COUNTS, tx);
+ VERIFY0(zap_remove(dp->dp_meta_objset,
+ DMU_POOL_DIRECTORY_OBJECT,
+ DMU_POOL_OBSOLETE_BPOBJ, tx));
+ bpobj_free(dp->dp_meta_objset,
+ dp->dp_obsolete_bpobj.bpo_object, tx);
+ bpobj_close(&dp->dp_obsolete_bpobj);
+}
+
dsl_pool_t *
-dsl_pool_create(spa_t *spa, nvlist_t *zplprops, uint64_t txg)
+dsl_pool_create(spa_t *spa, nvlist_t *zplprops, dsl_crypto_params_t *dcp,
+ uint64_t txg)
{
int err;
dsl_pool_t *dp = dsl_pool_open_impl(spa, txg);
dmu_tx_t *tx = dmu_tx_create_assigned(dp, txg);
+#ifdef _KERNEL
objset_t *os;
+#else
+ objset_t *os __attribute__((unused));
+#endif
dsl_dataset_t *ds;
uint64_t obj;
/* create and open the MOS (meta-objset) */
dp->dp_meta_objset = dmu_objset_create_impl(spa,
NULL, &dp->dp_meta_rootbp, DMU_OST_META, tx);
+ spa->spa_meta_objset = dp->dp_meta_objset;
/* create the pool directory */
err = zap_create_claim(dp->dp_meta_objset, DMU_POOL_DIRECTORY_OBJECT,
if (spa_version(spa) >= SPA_VERSION_DSL_SCRUB)
dsl_pool_create_origin(dp, tx);
+ /*
+ * Some features may be needed when creating the root dataset, so we
+ * create the feature objects here.
+ */
+ if (spa_version(spa) >= SPA_VERSION_FEATURES)
+ spa_feature_create_zap_objects(spa, tx);
+
+ if (dcp != NULL && dcp->cp_crypt != ZIO_CRYPT_OFF &&
+ dcp->cp_crypt != ZIO_CRYPT_INHERIT)
+ spa_feature_enable(spa, SPA_FEATURE_ENCRYPTION, tx);
+
/* create the root dataset */
- obj = dsl_dataset_create_sync_dd(dp->dp_root_dir, NULL, 0, tx);
+ obj = dsl_dataset_create_sync_dd(dp->dp_root_dir, NULL, dcp, 0, tx);
/* create the root objset */
- VERIFY0(dsl_dataset_hold_obj(dp, obj, FTAG, &ds));
+ VERIFY0(dsl_dataset_hold_obj_flags(dp, obj,
+ DS_HOLD_FLAG_DECRYPT, FTAG, &ds));
rrw_enter(&ds->ds_bp_rwlock, RW_READER, FTAG);
- VERIFY(NULL != (os = dmu_objset_create_impl(dp->dp_spa, ds,
- dsl_dataset_get_blkptr(ds), DMU_OST_ZFS, tx)));
+ os = dmu_objset_create_impl(dp->dp_spa, ds,
+ dsl_dataset_get_blkptr(ds), DMU_OST_ZFS, tx);
rrw_exit(&ds->ds_bp_rwlock, FTAG);
#ifdef _KERNEL
zfs_create_fs(os, kcred, zplprops, tx);
#endif
- dsl_dataset_rele(ds, FTAG);
+ dsl_dataset_rele_flags(ds, DS_HOLD_FLAG_DECRYPT, FTAG);
dmu_tx_commit(tx);
* Note: we signal even when increasing dp_dirty_total.
* This ensures forward progress -- each thread wakes the next waiter.
*/
- if (dp->dp_dirty_total <= zfs_dirty_data_max)
+ if (dp->dp_dirty_total < zfs_dirty_data_max)
cv_signal(&dp->dp_spaceavail_cv);
}
+#ifdef ZFS_DEBUG
+static boolean_t
+dsl_early_sync_task_verify(dsl_pool_t *dp, uint64_t txg)
+{
+ spa_t *spa = dp->dp_spa;
+ vdev_t *rvd = spa->spa_root_vdev;
+
+ for (uint64_t c = 0; c < rvd->vdev_children; c++) {
+ vdev_t *vd = rvd->vdev_child[c];
+ txg_list_t *tl = &vd->vdev_ms_list;
+ metaslab_t *ms;
+
+ for (ms = txg_list_head(tl, TXG_CLEAN(txg)); ms;
+ ms = txg_list_next(tl, ms, TXG_CLEAN(txg))) {
+ VERIFY(range_tree_is_empty(ms->ms_freeing));
+ VERIFY(range_tree_is_empty(ms->ms_checkpointing));
+ }
+ }
+
+ return (B_TRUE);
+}
+#endif
+
void
dsl_pool_sync(dsl_pool_t *dp, uint64_t txg)
{
tx = dmu_tx_create_assigned(dp, txg);
+ /*
+ * Run all early sync tasks before writing out any dirty blocks.
+ * For more info on early sync tasks see block comment in
+ * dsl_early_sync_task().
+ */
+ if (!txg_list_empty(&dp->dp_early_sync_tasks, txg)) {
+ dsl_sync_task_t *dst;
+
+ ASSERT3U(spa_sync_pass(dp->dp_spa), ==, 1);
+ while ((dst =
+ txg_list_remove(&dp->dp_early_sync_tasks, txg)) != NULL) {
+ ASSERT(dsl_early_sync_task_verify(dp, txg));
+ dsl_sync_task_sync(dst, tx);
+ }
+ ASSERT(dsl_early_sync_task_verify(dp, txg));
+ }
+
/*
* Write out all dirty blocks of dirty datasets.
*/
/*
* After the data blocks have been written (ensured by the zio_wait()
- * above), update the user/group space accounting. This happens
+ * above), update the user/group/project space accounting. This happens
* in tasks dispatched to dp_sync_taskq, so wait for them before
* continuing.
*/
*/
zio = zio_root(dp->dp_spa, NULL, NULL, ZIO_FLAG_MUSTSUCCEED);
while ((ds = txg_list_remove(&dp->dp_dirty_datasets, txg)) != NULL) {
+ objset_t *os = ds->ds_objset;
+
ASSERT(list_link_active(&ds->ds_synced_link));
dmu_buf_rele(ds->ds_dbuf, ds);
dsl_dataset_sync(ds, zio, tx);
+
+ /*
+ * Release any key mappings created by calls to
+ * dsl_dataset_dirty() from the userquota accounting
+ * code paths.
+ */
+ if (os->os_encrypted && !os->os_raw_receive &&
+ !os->os_next_write_raw[txg & TXG_MASK]) {
+ ASSERT3P(ds->ds_key_mapping, !=, NULL);
+ key_mapping_rele(dp->dp_spa, ds->ds_key_mapping, ds);
+ }
}
VERIFY0(zio_wait(zio));
*
* - move dead blocks from the pending deadlist to the on-disk deadlist
* - release hold from dsl_dataset_dirty()
+ * - release key mapping hold from dsl_dataset_dirty()
*/
while ((ds = list_remove_head(&synced_datasets)) != NULL) {
+ objset_t *os = ds->ds_objset;
+
+ if (os->os_encrypted && !os->os_raw_receive &&
+ !os->os_next_write_raw[txg & TXG_MASK]) {
+ ASSERT3P(ds->ds_key_mapping, !=, NULL);
+ key_mapping_rele(dp->dp_spa, ds->ds_key_mapping, ds);
+ }
+
dsl_dataset_sync_done(ds, tx);
}
taskq_member(dp->dp_sync_taskq, curthread));
}
+/*
+ * This function returns the amount of allocatable space in the pool
+ * minus whatever space is currently reserved by ZFS for specific
+ * purposes. Specifically:
+ *
+ * 1] Any reserved SLOP space
+ * 2] Any space used by the checkpoint
+ * 3] Any space used for deferred frees
+ *
+ * The latter 2 are especially important because they are needed to
+ * rectify the SPA's and DMU's different understanding of how much space
+ * is used. Now the DMU is aware of that extra space tracked by the SPA
+ * without having to maintain a separate special dir (e.g similar to
+ * $MOS, $FREEING, and $LEAKED).
+ *
+ * Note: By deferred frees here, we mean the frees that were deferred
+ * in spa_sync() after sync pass 1 (spa_deferred_bpobj), and not the
+ * segments placed in ms_defer trees during metaslab_sync_done().
+ */
uint64_t
-dsl_pool_adjustedsize(dsl_pool_t *dp, boolean_t netfree)
+dsl_pool_adjustedsize(dsl_pool_t *dp, zfs_space_check_t slop_policy)
{
- uint64_t space, resv;
-
- /*
- * If we're trying to assess whether it's OK to do a free,
- * cut the reservation in half to allow forward progress
- * (e.g. make it possible to rm(1) files from a full pool).
- */
- space = spa_get_dspace(dp->dp_spa);
- resv = spa_get_slop_space(dp->dp_spa);
- if (netfree)
+ spa_t *spa = dp->dp_spa;
+ uint64_t space, resv, adjustedsize;
+ uint64_t spa_deferred_frees =
+ spa->spa_deferred_bpobj.bpo_phys->bpo_bytes;
+
+ space = spa_get_dspace(spa)
+ - spa_get_checkpoint_space(spa) - spa_deferred_frees;
+ resv = spa_get_slop_space(spa);
+
+ switch (slop_policy) {
+ case ZFS_SPACE_CHECK_NORMAL:
+ break;
+ case ZFS_SPACE_CHECK_RESERVED:
resv >>= 1;
+ break;
+ case ZFS_SPACE_CHECK_EXTRA_RESERVED:
+ resv >>= 2;
+ break;
+ case ZFS_SPACE_CHECK_NONE:
+ resv = 0;
+ break;
+ default:
+ panic("invalid slop policy value: %d", slop_policy);
+ break;
+ }
+ adjustedsize = (space >= resv) ? (space - resv) : 0;
+
+ return (adjustedsize);
+}
- return (space - resv);
+uint64_t
+dsl_pool_unreserved_space(dsl_pool_t *dp, zfs_space_check_t slop_policy)
+{
+ uint64_t poolsize = dsl_pool_adjustedsize(dp, slop_policy);
+ uint64_t deferred =
+ metaslab_class_get_deferred(spa_normal_class(dp->dp_spa));
+ uint64_t quota = (poolsize >= deferred) ? (poolsize - deferred) : 0;
+ return (quota);
}
boolean_t
{
uint64_t delay_min_bytes =
zfs_dirty_data_max * zfs_delay_min_dirty_percent / 100;
+ uint64_t dirty_min_bytes =
+ zfs_dirty_data_max * zfs_dirty_data_sync_percent / 100;
boolean_t rv;
mutex_enter(&dp->dp_lock);
- if (dp->dp_dirty_total > zfs_dirty_data_sync)
+ if (dp->dp_dirty_total > dirty_min_bytes)
txg_kick(dp);
rv = (dp->dp_dirty_total > delay_min_bytes);
mutex_exit(&dp->dp_lock);
/* create the origin dir, ds, & snap-ds */
dsobj = dsl_dataset_create_sync(dp->dp_root_dir, ORIGIN_DIR_NAME,
- NULL, 0, kcred, tx);
+ NULL, 0, kcred, NULL, tx);
VERIFY0(dsl_dataset_hold_obj(dp, dsobj, FTAG, &ds));
dsl_dataset_snapshot_sync_impl(ds, ORIGIN_DIR_NAME, tx);
VERIFY0(dsl_dataset_hold_obj(dp, dsl_dataset_phys(ds)->ds_prev_snap_obj,
return (dp->dp_iput_taskq);
}
+taskq_t *
+dsl_pool_unlinked_drain_taskq(dsl_pool_t *dp)
+{
+ return (dp->dp_unlinked_drain_taskq);
+}
+
/*
* Walk through the pool-wide zap object of temporary snapshot user holds
* and release them.
return (RRW_WRITE_HELD(&dp->dp_config_rwlock));
}
-#if defined(_KERNEL) && defined(HAVE_SPL)
+#if defined(_KERNEL)
EXPORT_SYMBOL(dsl_pool_config_enter);
EXPORT_SYMBOL(dsl_pool_config_exit);
MODULE_PARM_DESC(zfs_dirty_data_max_max,
"zfs_dirty_data_max upper bound in bytes");
-module_param(zfs_dirty_data_sync, ulong, 0644);
-MODULE_PARM_DESC(zfs_dirty_data_sync, "sync txg when this much dirty data");
+module_param(zfs_dirty_data_sync_percent, int, 0644);
+MODULE_PARM_DESC(zfs_dirty_data_sync_percent,
+ "dirty data txg sync threshold as a percentage of zfs_dirty_data_max");
module_param(zfs_delay_scale, ulong, 0644);
MODULE_PARM_DESC(zfs_delay_scale, "how quickly delay approaches infinity");
module_param(zfs_sync_taskq_batch_pct, int, 0644);
MODULE_PARM_DESC(zfs_sync_taskq_batch_pct,
"max percent of CPUs that are used to sync dirty data");
+
+module_param(zfs_zil_clean_taskq_nthr_pct, int, 0644);
+MODULE_PARM_DESC(zfs_zil_clean_taskq_nthr_pct,
+ "max percent of CPUs that are used per dp_sync_taskq");
+
+module_param(zfs_zil_clean_taskq_minalloc, int, 0644);
+MODULE_PARM_DESC(zfs_zil_clean_taskq_minalloc,
+ "number of taskq entries that are pre-populated");
+
+module_param(zfs_zil_clean_taskq_maxalloc, int, 0644);
+MODULE_PARM_DESC(zfs_zil_clean_taskq_maxalloc,
+ "max number of taskq entries that are cached");
+
/* END CSTYLED */
#endif
projects / mirror_zfs.git / blobdiff