dsl_pool.c revision 2acef22db7808606888f8f92715629ff3ba555b9
/*
* CDDL HEADER START
*
* The contents of this file are subject to the terms of the
* Common Development and Distribution License (the "License").
* You may not use this file except in compliance with the License.
*
* You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
* See the License for the specific language governing permissions
* and limitations under the License.
*
* When distributing Covered Code, include this CDDL HEADER in each
* file and include the License file at usr/src/OPENSOLARIS.LICENSE.
* If applicable, add the following below this CDDL HEADER, with the
* fields enclosed by brackets "[]" replaced with your own identifying
* information: Portions Copyright [yyyy] [name of copyright owner]
*
* CDDL HEADER END
*/
/*
* Copyright (c) 2013 by Delphix. All rights reserved.
* Copyright (c) 2013 Steven Hartland. All rights reserved.
*/
#include <sys/dsl_pool.h>
#include <sys/dsl_dataset.h>
#include <sys/dsl_prop.h>
#include <sys/dsl_synctask.h>
#include <sys/dsl_scan.h>
#include <sys/dmu_objset.h>
#include <sys/zfs_context.h>
#include <sys/zfs_znode.h>
#include <sys/spa_impl.h>
#include <sys/dsl_deadlist.h>
#include <sys/zfeature.h>
#include <sys/zil_impl.h>
#include <sys/dsl_userhold.h>
/*
* ZFS Write Throttle
* ------------------
*
* ZFS must limit the rate of incoming writes to the rate at which it is able
* to sync data modifications to the backend storage. Throttling by too much
* creates an artificial limit; throttling by too little can only be sustained
* for short periods and would lead to highly lumpy performance. On a per-pool
* basis, ZFS tracks the amount of modified (dirty) data. As operations change
* data, the amount of dirty data increases; as ZFS syncs out data, the amount
* of dirty data decreases. When the amount of dirty data exceeds a
* predetermined threshold further modifications are blocked until the amount
* of dirty data decreases (as data is synced out).
*
* The limit on dirty data is tunable, and should be adjusted according to
* both the IO capacity and available memory of the system. The larger the
* window, the more ZFS is able to aggregate and amortize metadata (and data)
* changes. However, memory is a limited resource, and allowing for more dirty
* data comes at the cost of keeping other useful data in memory (for example
* ZFS data cached by the ARC).
*
* Implementation
*
* As buffers are modified dsl_pool_willuse_space() increments both the per-
* txg (dp_dirty_pertxg[]) and poolwide (dp_dirty_total) accounting of
* dirty space used; dsl_pool_dirty_space() decrements those values as data
* is synced out from dsl_pool_sync(). While only the poolwide value is
* relevant, the per-txg value is useful for debugging. The tunable
* 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
* ensure that there is a txg syncing (see the comment in txg.c for a full
* description of transaction group stages).
*
* The IO scheduler uses both the dirty space limit and current amount of
* dirty data as inputs. Those values affect the number of concurrent IOs ZFS
* issues. See the comment in vdev_queue.c for details of the IO scheduler.
*
* The delay is also calculated based on the amount of dirty data. See the
* comment above dmu_tx_delay() for details.
*/
/*
* zfs_dirty_data_max will be set to zfs_dirty_data_max_percent% of all memory,
*/
int zfs_dirty_data_max_percent = 10;
/*
* If there is at least this much dirty data, push out a txg.
*/
/*
* Once there is this amount of dirty data, the dmu_tx_delay() will kick in
* and delay each transaction.
* This value should be >= zfs_vdev_async_write_active_max_dirty_percent.
*/
int zfs_delay_min_dirty_percent = 60;
/*
* This controls how quickly the delay approaches infinity.
* Larger values cause it to delay less for a given amount of dirty data.
* Therefore larger values will cause there to be more dirty data for a
* given throughput.
*
* For the smoothest delay, this value should be about 1 billion divided
* by the maximum number of operations per second. This will smoothly
* handle between 10x and 1/10th this number.
*
* Note: zfs_delay_scale * zfs_dirty_data_max must be < 2^64, due to the
* multiply in dmu_tx_delay().
*/
/*
* XXX someday maybe turn these into #defines, and you have to tune it on a
* per-pool basis using zfs.conf.
*/
int
{
int err;
if (err)
return (err);
}
static dsl_pool_t *
{
dsl_pool_t *dp;
1, 4, 0);
return (dp);
}
int
{
int err;
&dp->dp_meta_objset);
if (err != 0)
else
return (err);
}
int
{
int err;
&dp->dp_root_dir_obj);
if (err)
goto out;
if (err)
goto out;
if (err)
goto out;
if (err)
goto out;
if (err == 0) {
&dp->dp_origin_snap);
}
if (err)
goto out;
}
&dp->dp_free_dir);
if (err)
goto out;
if (err)
goto out;
}
&dp->dp_bptree_obj);
if (err != 0)
goto out;
}
&dp->dp_empty_bpobj);
if (err != 0)
goto out;
}
err = 0;
if (err)
goto out;
out:
return (err);
}
void
{
/*
* Drop our references from dsl_pool_open().
*
* Since we held the origin_snap from "syncing" context (which
* 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_mos_dir)
if (dp->dp_free_dir)
if (dp->dp_root_dir)
/* undo the dmu_objset_open_impl(mos) from dsl_pool_open() */
if (dp->dp_meta_objset)
if (dp->dp_blkstats)
}
{
int err;
/* create and open the MOS (meta-objset) */
/* create the pool directory */
/* Initialize scan structures */
/* create and open the root dir */
/* create and open the meta-objset dir */
/* create and open the free dir */
FREE_DIR_NAME, tx);
/* create and open the free_bplist */
}
/* create the root dataset */
/* create the root objset */
#ifdef _KERNEL
#endif
return (dp);
}
/*
* Account for the meta-objset space in its placeholder dsl_dir.
*/
void
{
}
static int
{
return (0);
}
static void
{
}
static void
{
if (delta < 0)
/*
* Note: we signal even when increasing dp_dirty_total.
* This ensures forward progress -- each thread wakes the next waiter.
*/
}
void
{
/*
* Write out all dirty blocks of dirty datasets.
*/
/*
* We must not sync any non-MOS datasets twice, because
* we may have taken a snapshot of them. However, we
* may sync newly-created datasets on pass 2.
*/
}
/*
* We have written all of the accounted dirty data, so our
* dp_space_towrite should now be zero. However, some seldom-used
* code paths do not adhere to this (e.g. dbuf_undirty(), also
* rounding error in dbuf_write_physdone).
* Shore up the accounting of any dirtied space now.
*/
/*
* After the data blocks have been written (ensured by the zio_wait()
*/
}
/*
* Sync the datasets again to push out the changes due to
* userspace updates. This must be done before we process the
* sync tasks, so that any snapshots will have the correct
* user accounting information (and we won't get confused
* about which blocks are part of the snapshot).
*/
}
/*
* Now that the datasets have been completely synced, we can
* clean up our in-memory structures accumulated while syncing:
*
* - move dead blocks from the pending deadlist to the on-disk deadlist
* - release hold from dsl_dataset_dirty()
*/
}
}
/*
* The MOS's space is accounted for in the pool/$MOS
* (dp_mos_dir). We can't modify the mos while we're syncing
* it, so we remember the deltas and apply them here.
*/
dp->dp_mos_uncompressed_delta != 0) {
dp->dp_mos_used_delta = 0;
dp->dp_mos_compressed_delta = 0;
dp->dp_mos_uncompressed_delta = 0;
}
}
/*
* If we modify a dataset in the same txg that we want to destroy it,
* its dsl_dir's dd_dbuf will be dirty, and thus have a hold on it.
* dsl_dir_destroy_check() will fail if there are unexpected holds.
* Therefore, we want to sync the MOS (thus syncing the dd_dbuf
* and clearing the hold on it) before we process the sync_tasks.
* The MOS data dirtied by the sync_tasks will be synced on the next
* pass.
*/
/*
* No more sync tasks should have been added while we
* were syncing.
*/
}
}
void
{
}
}
/*
* TRUE if the current thread is the tx_sync_thread or if we
* are being called from SPA context during pool initialization.
*/
int
{
}
{
/*
* Reserve about 1.6% (1/64), or at least 32MB, for allocation
* efficiency.
* XXX The intent log is not accounted for, so it must fit
* within this slop.
*
* 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).
*/
if (netfree)
resv >>= 1;
}
{
return (rv);
}
void
{
if (space > 0) {
}
}
void
{
if (space == 0)
return;
/* XXX writing something we didn't dirty? */
}
}
/* ARGSUSED */
static int
{
int err;
if (err)
return (err);
if (err) {
return (err);
}
break;
}
/*
* The $ORIGIN can't have any data, or the accounting
* will be wrong.
*/
/* The origin doesn't get attached to itself */
return (0);
}
}
}
}
return (0);
}
void
{
tx, DS_FIND_CHILDREN));
}
/* ARGSUSED */
static int
{
}
}
return (0);
}
void
{
/*
* We can't use bpobj_alloc(), because spa_version() still
* returns the old version, and we need a new-version bpobj with
* subobj support. So call dmu_object_alloc() directly.
*/
}
void
{
/* create the origin dir, ds, & snap-ds */
}
taskq_t *
{
return (dp->dp_vnrele_taskq);
}
/*
* Walk through the pool-wide zap object of temporary snapshot user holds
* and release them.
*/
void
{
if (zapobj == 0)
return;
holds = fnvlist_alloc();
zap_cursor_advance(&zc)) {
char *htag;
*htag = '\0';
++htag;
tags = fnvlist_alloc();
} else {
}
}
}
/*
* Create the pool-wide zap object for storing temporary snapshot holds.
*/
void
{
}
static int
{
char *name;
int error;
/*
* If the pool was created prior to SPA_VERSION_USERREFS, the
* zap object for temporary holds might not exist yet.
*/
if (zapobj == 0) {
if (holding) {
} else {
}
}
if (holding)
else
return (error);
}
/*
* Add a temporary hold for the given dataset object and tag.
*/
int
{
}
/*
* Release a temporary hold for the given dataset object and tag.
*/
int
{
}
/*
* DSL Pool Configuration Lock
*
* The dp_config_rwlock protects against changes to DSL state (e.g. dataset
* creation / destruction / rename / property setting). It must be held for
* read to hold a dataset or dsl_dir. I.e. you must call
* dsl_pool_config_enter() or dsl_pool_hold() before calling
* dsl_{dataset,dir}_hold{_obj}. In most circumstances, the dp_config_rwlock
* must be held continuously until all datasets and dsl_dirs are released.
*
* The only exception to this rule is that if a "long hold" is placed on
* a dataset, then the dp_config_rwlock may be dropped while the dataset
* is still held. The long hold will prevent the dataset from being
* destroyed -- the destroy will fail with EBUSY. A long hold can be
* obtained by calling dsl_dataset_long_hold(), or by "owning" a dataset
* (by calling dsl_{dataset,objset}_{try}own{_obj}).
*
* Legitimate long-holders (including owners) should be long-running, cancelable
* tasks that should cause "zfs destroy" to fail. This includes DMU
* consumers (i.e. a ZPL filesystem being mounted or ZVOL being open),
* "zfs send", and "zfs diff". There are several other long-holders whose
* uses are suboptimal (e.g. "zfs promote", and zil_suspend()).
*
* The usual formula for long-holding would be:
* dsl_pool_hold()
* dsl_dataset_hold()
* ... perform checks ...
* dsl_dataset_long_hold()
* dsl_pool_rele()
* ... perform long-running task ...
* dsl_dataset_long_rele()
* dsl_dataset_rele()
*
* Note that when the long hold is released, the dataset is still held but
* the pool is not held. The dataset may change arbitrarily during this time
* (e.g. it could be destroyed). Therefore you shouldn't do anything to the
* dataset except release it.
*
* User-initiated operations (e.g. ioctls, zfs_ioc_*()) are either read-only
* or modifying operations.
*
* Modifying operations should generally use dsl_sync_task(). The synctask
* infrastructure enforces proper locking strategy with respect to the
* dp_config_rwlock. See the comment above dsl_sync_task() for details.
*
* Read-only operations will manually hold the pool, then the dataset, obtain
* information from the dataset, then release the pool and dataset.
* dmu_objset_{hold,rele}() are convenience routines that also do the pool
*/
int
{
int error;
if (error == 0) {
}
return (error);
}
void
{
}
void
{
/*
* We use a "reentrant" reader-writer lock, but not reentrantly.
*
* The rrwlock can (with the track_all flag) track all reading threads,
* which is very useful for debugging which code path failed to release
* the lock, and for verifying that the *current* thread does hold
* the lock.
*
* (Unlike a rwlock, which knows that N threads hold it for
* read, but not *which* threads, so rw_held(RW_READER) returns TRUE
* if any thread holds it for read, even if this thread doesn't).
*/
}
void
{
}
{
}