*/
/*
* Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved.
+ * Copyright (c) 2011 by Delphix. All rights reserved.
*/
#include <sys/dsl_pool.h>
int zfs_no_write_throttle = 0;
int zfs_write_limit_shift = 3; /* 1/8th of physical memory */
int zfs_txg_synctime_ms = 1000; /* target millisecs to sync a txg */
+int zfs_txg_history = 60; /* statistics for the last N txgs */
-uint64_t zfs_write_limit_min = 32 << 20; /* min write limit is 32MB */
-uint64_t zfs_write_limit_max = 0; /* max data payload per txg */
-uint64_t zfs_write_limit_inflated = 0;
-uint64_t zfs_write_limit_override = 0;
+unsigned long zfs_write_limit_min = 32 << 20; /* min write limit is 32MB */
+unsigned long zfs_write_limit_max = 0; /* max data payload per txg */
+unsigned long zfs_write_limit_inflated = 0;
+unsigned long zfs_write_limit_override = 0;
kmutex_t zfs_write_limit_lock;
static pgcnt_t old_physmem = 0;
+static int
+dsl_pool_txg_history_update(kstat_t *ksp, int rw)
+{
+ dsl_pool_t *dp = ksp->ks_private;
+ txg_history_t *th;
+ int i = 0;
+
+ if (rw == KSTAT_WRITE)
+ return (EACCES);
+
+ if (ksp->ks_data)
+ kmem_free(ksp->ks_data, ksp->ks_data_size);
+
+ mutex_enter(&dp->dp_lock);
+
+ ksp->ks_ndata = dp->dp_txg_history_size;
+ ksp->ks_data_size = dp->dp_txg_history_size * sizeof(kstat_txg_t);
+ if (ksp->ks_data_size > 0)
+ ksp->ks_data = kmem_alloc(ksp->ks_data_size, KM_PUSHPAGE);
+
+ /* Traversed oldest to youngest for the most readable kstat output */
+ for (th = list_tail(&dp->dp_txg_history); th != NULL;
+ th = list_prev(&dp->dp_txg_history, th)) {
+ mutex_enter(&th->th_lock);
+ ASSERT3S(i + sizeof(kstat_txg_t), <=, ksp->ks_data_size);
+ memcpy(ksp->ks_data + i, &th->th_kstat, sizeof(kstat_txg_t));
+ i += sizeof(kstat_txg_t);
+ mutex_exit(&th->th_lock);
+ }
+
+ mutex_exit(&dp->dp_lock);
+
+ return (0);
+}
+
+static void
+dsl_pool_txg_history_init(dsl_pool_t *dp, uint64_t txg)
+{
+ char name[KSTAT_STRLEN];
+
+ list_create(&dp->dp_txg_history, sizeof (txg_history_t),
+ offsetof(txg_history_t, th_link));
+ dsl_pool_txg_history_add(dp, txg);
+
+ (void) snprintf(name, KSTAT_STRLEN, "txgs-%s", spa_name(dp->dp_spa));
+ dp->dp_txg_kstat = kstat_create("zfs", 0, name, "misc",
+ KSTAT_TYPE_TXG, 0, KSTAT_FLAG_VIRTUAL);
+ if (dp->dp_txg_kstat) {
+ dp->dp_txg_kstat->ks_data = NULL;
+ dp->dp_txg_kstat->ks_private = dp;
+ dp->dp_txg_kstat->ks_update = dsl_pool_txg_history_update;
+ kstat_install(dp->dp_txg_kstat);
+ }
+}
+
+static void
+dsl_pool_txg_history_destroy(dsl_pool_t *dp)
+{
+ txg_history_t *th;
+
+ if (dp->dp_txg_kstat) {
+ if (dp->dp_txg_kstat->ks_data)
+ kmem_free(dp->dp_txg_kstat->ks_data,
+ dp->dp_txg_kstat->ks_data_size);
+
+ kstat_delete(dp->dp_txg_kstat);
+ }
+
+ mutex_enter(&dp->dp_lock);
+ while ((th = list_remove_head(&dp->dp_txg_history))) {
+ dp->dp_txg_history_size--;
+ mutex_destroy(&th->th_lock);
+ kmem_free(th, sizeof(txg_history_t));
+ }
+
+ ASSERT3U(dp->dp_txg_history_size, ==, 0);
+ list_destroy(&dp->dp_txg_history);
+ mutex_exit(&dp->dp_lock);
+}
+
+txg_history_t *
+dsl_pool_txg_history_add(dsl_pool_t *dp, uint64_t txg)
+{
+ txg_history_t *th, *rm;
+
+ th = kmem_zalloc(sizeof(txg_history_t), KM_SLEEP);
+ mutex_init(&th->th_lock, NULL, MUTEX_DEFAULT, NULL);
+ th->th_kstat.txg = txg;
+ th->th_kstat.state = TXG_STATE_OPEN;
+ th->th_kstat.birth = gethrtime();
+
+ mutex_enter(&dp->dp_lock);
+
+ list_insert_head(&dp->dp_txg_history, th);
+ dp->dp_txg_history_size++;
+
+ while (dp->dp_txg_history_size > zfs_txg_history) {
+ dp->dp_txg_history_size--;
+ rm = list_remove_tail(&dp->dp_txg_history);
+ mutex_destroy(&rm->th_lock);
+ kmem_free(rm, sizeof(txg_history_t));
+ }
+
+ mutex_exit(&dp->dp_lock);
+
+ return (th);
+}
+
+/*
+ * Traversed youngest to oldest because lookups are only done for open
+ * or syncing txgs which are guaranteed to be at the head of the list.
+ * The txg_history_t structure will be returned locked.
+ */
+txg_history_t *
+dsl_pool_txg_history_get(dsl_pool_t *dp, uint64_t txg)
+{
+ txg_history_t *th;
+
+ mutex_enter(&dp->dp_lock);
+ for (th = list_head(&dp->dp_txg_history); th != NULL;
+ th = list_next(&dp->dp_txg_history, th)) {
+ if (th->th_kstat.txg == txg) {
+ mutex_enter(&th->th_lock);
+ break;
+ }
+ }
+ mutex_exit(&dp->dp_lock);
+
+ return (th);
+}
+
+void
+dsl_pool_txg_history_put(txg_history_t *th)
+{
+ mutex_exit(&th->th_lock);
+}
+
int
dsl_pool_open_special_dir(dsl_pool_t *dp, const char *name, dsl_dir_t **ddp)
{
mutex_init(&dp->dp_lock, NULL, MUTEX_DEFAULT, NULL);
- dp->dp_vnrele_taskq = taskq_create("zfs_vn_rele_taskq", 1, minclsyspri,
+ dp->dp_iput_taskq = taskq_create("zfs_iput_taskq", 1, minclsyspri,
1, 4, 0);
+ dsl_pool_txg_history_init(dp, txg);
+
return (dp);
}
arc_flush(dp->dp_spa);
txg_fini(dp);
dsl_scan_fini(dp);
+ dsl_pool_txg_history_destroy(dp);
rw_destroy(&dp->dp_config_rwlock);
mutex_destroy(&dp->dp_lock);
- taskq_destroy(dp->dp_vnrele_taskq);
+ taskq_destroy(dp->dp_iput_taskq);
if (dp->dp_blkstats)
kmem_free(dp->dp_blkstats, sizeof (zfs_all_blkstats_t));
kmem_free(dp, sizeof (dsl_pool_t));
/* create the root objset */
VERIFY(0 == dsl_dataset_hold_obj(dp, obj, FTAG, &ds));
- os = dmu_objset_create_impl(dp->dp_spa, ds,
- dsl_dataset_get_blkptr(ds), DMU_OST_ZFS, tx);
+ VERIFY(NULL != (os = dmu_objset_create_impl(dp->dp_spa, ds,
+ dsl_dataset_get_blkptr(ds), DMU_OST_ZFS, tx)));
#ifdef _KERNEL
zfs_create_fs(os, kcred, zplprops, tx);
#endif
deadlist_enqueue_cb(void *arg, const blkptr_t *bp, dmu_tx_t *tx)
{
dsl_deadlist_t *dl = arg;
+ dsl_pool_t *dp = dmu_objset_pool(dl->dl_os);
+ rw_enter(&dp->dp_config_rwlock, RW_READER);
dsl_deadlist_insert(dl, bp, tx);
+ rw_exit(&dp->dp_config_rwlock);
return (0);
}
start = gethrtime();
zio = zio_root(dp->dp_spa, NULL, NULL, ZIO_FLAG_MUSTSUCCEED);
- while (ds = txg_list_remove(&dp->dp_dirty_datasets, txg)) {
+ while ((ds = txg_list_remove(&dp->dp_dirty_datasets, txg))) {
/*
* We must not sync any non-MOS datasets twice, because
* we may have taken a snapshot of them. However, we
* whose ds_bp will be rewritten when we do this 2nd sync.
*/
zio = zio_root(dp->dp_spa, NULL, NULL, ZIO_FLAG_MUSTSUCCEED);
- while (ds = txg_list_remove(&dp->dp_dirty_datasets, txg)) {
+ while ((ds = txg_list_remove(&dp->dp_dirty_datasets, txg))) {
ASSERT(list_link_active(&ds->ds_synced_link));
dmu_buf_rele(ds->ds_dbuf, ds);
dsl_dataset_sync(ds, zio, tx);
deadlist_enqueue_cb, &ds->ds_deadlist, tx);
}
- while (dstg = txg_list_remove(&dp->dp_sync_tasks, txg)) {
+ while ((dstg = txg_list_remove(&dp->dp_sync_tasks, txg))) {
/*
* No more sync tasks should have been added while we
* were syncing.
DTRACE_PROBE(pool_sync__3task);
start = gethrtime();
- while (dd = txg_list_remove(&dp->dp_dirty_dirs, txg))
+ while ((dd = txg_list_remove(&dp->dp_dirty_dirs, txg)))
dsl_dir_sync(dd, tx);
write_time += gethrtime() - start;
dsl_dataset_t *ds;
objset_t *os;
- while (ds = list_head(&dp->dp_synced_datasets)) {
+ while ((ds = list_head(&dp->dp_synced_datasets))) {
list_remove(&dp->dp_synced_datasets, ds);
os = ds->ds_objset;
zil_clean(os->os_zil, txg);
reserved = dp->dp_space_towrite[tx->tx_txg & TXG_MASK]
+ dp->dp_tempreserved[tx->tx_txg & TXG_MASK] / 2;
- if (reserved && reserved > write_limit)
+ if (reserved && reserved > write_limit) {
+ DMU_TX_STAT_BUMP(dmu_tx_write_limit);
return (ERESTART);
+ }
}
atomic_add_64(&dp->dp_tempreserved[tx->tx_txg & TXG_MASK], space);
void
dsl_pool_upgrade_dir_clones(dsl_pool_t *dp, dmu_tx_t *tx)
{
- ASSERT(dmu_tx_is_syncing(tx));
uint64_t obj;
+ ASSERT(dmu_tx_is_syncing(tx));
+
(void) dsl_dir_create_sync(dp, dp->dp_root_dir, FREE_DIR_NAME, tx);
VERIFY(0 == dsl_pool_open_special_dir(dp,
FREE_DIR_NAME, &dp->dp_free_dir));
}
taskq_t *
-dsl_pool_vnrele_taskq(dsl_pool_t *dp)
+dsl_pool_iput_taskq(dsl_pool_t *dp)
{
- return (dp->dp_vnrele_taskq);
+ return (dp->dp_iput_taskq);
}
/*
return (dsl_pool_user_hold_rele_impl(dp, dsobj, tag, NULL,
tx, B_FALSE));
}
+
+#if defined(_KERNEL) && defined(HAVE_SPL)
+module_param(zfs_no_write_throttle, int, 0644);
+MODULE_PARM_DESC(zfs_no_write_throttle, "Disable write throttling");
+
+module_param(zfs_write_limit_shift, int, 0444);
+MODULE_PARM_DESC(zfs_write_limit_shift, "log2(fraction of memory) per txg");
+
+module_param(zfs_txg_synctime_ms, int, 0644);
+MODULE_PARM_DESC(zfs_txg_synctime_ms, "Target milliseconds between txg sync");
+
+module_param(zfs_txg_history, int, 0644);
+MODULE_PARM_DESC(zfs_txg_history, "Historic statistics for the last N txgs");
+
+module_param(zfs_write_limit_min, ulong, 0444);
+MODULE_PARM_DESC(zfs_write_limit_min, "Min txg write limit");
+
+module_param(zfs_write_limit_max, ulong, 0444);
+MODULE_PARM_DESC(zfs_write_limit_max, "Max txg write limit");
+
+module_param(zfs_write_limit_inflated, ulong, 0444);
+MODULE_PARM_DESC(zfs_write_limit_inflated, "Inflated txg write limit");
+
+module_param(zfs_write_limit_override, ulong, 0444);
+MODULE_PARM_DESC(zfs_write_limit_override, "Override txg write limit");
+#endif