d/freebsd-dev
1
0
Fork 0
Code Issues Pull Requests Packages Projects Releases Wiki Activity

OpenZFS 7613 - ms_freetree[4] is only used in syncing context

Browse Source 
metaslab_t:ms_freetree[TXG_SIZE] is only used in syncing context. We
should replace it with two trees: the freeing tree (ranges that we are
freeing this syncing txg) and the freed tree (ranges which have been
freed this txg).

Authored by: Matthew Ahrens <mahrens@delphix.com>
Reviewed by: George Wilson <george.wilson@delphix.com>
Reviewed by: Alex Reece <alex@delphix.com>
Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov>
Ported-by: Tim Chase <tim@chase2k.com>

OpenZFS-issue: https://www.illumos.org/issues/7613
OpenZFS-commit: https://github.com/openzfs/openzfs/commit/a8698da2
Closes #5598
This commit is contained in:
Tim Chase 2017-01-26 17:27:19 -06:00 committed by Brian Behlendorf
parent 9c9531cb6f
commit 258553d3d7
2 changed files with 82 additions and 76 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

77
include/sys/metaslab_impl.h
View File

@ -254,21 +254,24 @@ struct metaslab_group {
#define MAX_LBAS 64
/*
* Each metaslab maintains a set of in-core trees to track metaslab operations.
* The in-core free tree (ms_tree) contains the current list of free segments.
* As blocks are allocated, the allocated segment are removed from the ms_tree
* and added to a per txg allocation tree (ms_alloctree). As blocks are freed,
* they are added to the per txg free tree (ms_freetree). These per txg
* trees allow us to process all allocations and frees in syncing context
* where it is safe to update the on-disk space maps. One additional in-core
* tree is maintained to track deferred frees (ms_defertree). Once a block
* is freed it will move from the ms_freetree to the ms_defertree. A deferred
* free means that a block has been freed but cannot be used by the pool
* until TXG_DEFER_SIZE transactions groups later. For example, a block
* that is freed in txg 50 will not be available for reallocation until
* txg 52 (50 + TXG_DEFER_SIZE). This provides a safety net for uberblock
* rollback. A pool could be safely rolled back TXG_DEFERS_SIZE
* transactions groups and ensure that no block has been reallocated.
* Each metaslab maintains a set of in-core trees to track metaslab
* operations. The in-core free tree (ms_tree) contains the list of
* free segments which are eligible for allocation. As blocks are
* allocated, the allocated segments are removed from the ms_tree and
* added to a per txg allocation tree (ms_alloctree). This allows us to
* process all allocations in syncing context where it is safe to update
* the on-disk space maps. Frees are also processed in syncing context.
* Most frees are generated from syncing context, and those that are not
* are held in the spa_free_bplist for processing in syncing context.
* An additional set of in-core trees is maintained to track deferred
* frees (ms_defertree). Once a block is freed it will move from the
* ms_freedtree to the ms_defertree. A deferred free means that a block
* has been freed but cannot be used by the pool until TXG_DEFER_SIZE
* transactions groups later. For example, a block that is freed in txg
* 50 will not be available for reallocation until txg 52 (50 +
* TXG_DEFER_SIZE). This provides a safety net for uberblock rollback.
* A pool could be safely rolled back TXG_DEFERS_SIZE transactions
* groups and ensure that no block has been reallocated.
*
* The simplified transition diagram looks like this:
*
@ -276,33 +279,34 @@ struct metaslab_group {
* ALLOCATE
* |
* V
* free segment (ms_tree) --------> ms_alloctree ----> (write to space map)
* free segment (ms_tree) -----> ms_alloctree[4] ----> (write to space map)
* ^
* |
* | ms_freetree <--- FREE
* | ms_freeingtree <--- FREE
* | |
* | v
* | ms_freedtree
* | |
* | |
* +----------- ms_defertree <-------+---------> (write to space map)
* +-------- ms_defertree[2] <-------+---------> (write to space map)
*
*
* Each metaslab's space is tracked in a single space map in the MOS,
* which is only updated in syncing context. Each time we sync a txg,
* we append the allocs and frees from that txg to the space map.
* The pool space is only updated once all metaslabs have finished syncing.
* which is only updated in syncing context. Each time we sync a txg,
* we append the allocs and frees from that txg to the space map. The
* pool space is only updated once all metaslabs have finished syncing.
*
* To load the in-core free tree we read the space map from disk.
* This object contains a series of alloc and free records that are
* combined to make up the list of all free segments in this metaslab. These
* To load the in-core free tree we read the space map from disk. This
* object contains a series of alloc and free records that are combined
* to make up the list of all free segments in this metaslab. These
* segments are represented in-core by the ms_tree and are stored in an
* AVL tree.
*
* As the space map grows (as a result of the appends) it will
* eventually become space-inefficient. When the metaslab's in-core free tree
* is zfs_condense_pct/100 times the size of the minimal on-disk
* representation, we rewrite it in its minimized form. If a metaslab
* needs to condense then we must set the ms_condensing flag to ensure
* that allocations are not performed on the metaslab that is being written.
* eventually become space-inefficient. When the metaslab's in-core
* free tree is zfs_condense_pct/100 times the size of the minimal
* on-disk representation, we rewrite it in its minimized form. If a
* metaslab needs to condense then we must set the ms_condensing flag to
* ensure that allocations are not performed on the metaslab that is
* being written.
*/
struct metaslab {
kmutex_t ms_lock;
@ -314,10 +318,17 @@ struct metaslab {
uint64_t ms_fragmentation;
range_tree_t *ms_alloctree[TXG_SIZE];
range_tree_t *ms_freetree[TXG_SIZE];
range_tree_t *ms_defertree[TXG_DEFER_SIZE];
range_tree_t *ms_tree;
/*
* The following range trees are accessed only from syncing context.
* ms_free*tree only have entries while syncing, and are empty
* between syncs.
*/
range_tree_t *ms_freeingtree; /* to free this syncing txg */
range_tree_t *ms_freedtree; /* already freed this syncing txg */
range_tree_t *ms_defertree[TXG_DEFER_SIZE];
boolean_t ms_condensing; /* condensing? */
boolean_t ms_condense_wanted;

81
module/zfs/metaslab.c
View File

@ -442,7 +442,6 @@ metaslab_verify_space(metaslab_t *msp, uint64_t txg)
{
spa_t *spa = msp->ms_group->mg_vd->vdev_spa;
uint64_t allocated = 0;
uint64_t freed = 0;
uint64_t sm_free_space, msp_free_space;
int t;
@ -473,10 +472,9 @@ metaslab_verify_space(metaslab_t *msp, uint64_t txg)
allocated +=
range_tree_space(msp->ms_alloctree[(txg + t) & TXG_MASK]);
}
freed = range_tree_space(msp->ms_freetree[TXG_CLEAN(txg) & TXG_MASK]);
msp_free_space = range_tree_space(msp->ms_tree) + allocated +
msp->ms_deferspace + freed;
msp->ms_deferspace + range_tree_space(msp->ms_freedtree);
VERIFY3U(sm_free_space, ==, msp_free_space);
}
@ -1423,7 +1421,7 @@ metaslab_init(metaslab_group_t *mg, uint64_t id, uint64_t object, uint64_t txg,
/*
* We create the main range tree here, but we don't create the
* alloctree and freetree until metaslab_sync_done(). This serves
* other range trees until metaslab_sync_done(). This serves
* two purposes: it allows metaslab_sync_done() to detect the
* addition of new space; and for debugging, it ensures that we'd
* data fault on any attempt to use this metaslab before it's ready.
@ -1483,10 +1481,11 @@ metaslab_fini(metaslab_t *msp)
metaslab_unload(msp);
range_tree_destroy(msp->ms_tree);
range_tree_destroy(msp->ms_freeingtree);
range_tree_destroy(msp->ms_freedtree);
for (t = 0; t < TXG_SIZE; t++) {
range_tree_destroy(msp->ms_alloctree[t]);
range_tree_destroy(msp->ms_freetree[t]);
}
for (t = 0; t < TXG_DEFER_SIZE; t++) {
@ -2103,7 +2102,6 @@ static void
metaslab_condense(metaslab_t *msp, uint64_t txg, dmu_tx_t *tx)
{
spa_t *spa = msp->ms_group->mg_vd->vdev_spa;
range_tree_t *freetree = msp->ms_freetree[txg & TXG_MASK];
range_tree_t *condense_tree;
space_map_t *sm = msp->ms_sm;
int t;
@ -2135,9 +2133,9 @@ metaslab_condense(metaslab_t *msp, uint64_t txg, dmu_tx_t *tx)
/*
* Remove what's been freed in this txg from the condense_tree.
* Since we're in sync_pass 1, we know that all the frees from
* this txg are in the freetree.
* this txg are in the freeingtree.
*/
range_tree_walk(freetree, range_tree_remove, condense_tree);
range_tree_walk(msp->ms_freeingtree, range_tree_remove, condense_tree);
for (t = 0; t < TXG_DEFER_SIZE; t++) {
range_tree_walk(msp->ms_defertree[t],
@ -2193,9 +2191,6 @@ metaslab_sync(metaslab_t *msp, uint64_t txg)
spa_t *spa = vd->vdev_spa;
objset_t *mos = spa_meta_objset(spa);
range_tree_t *alloctree = msp->ms_alloctree[txg & TXG_MASK];
range_tree_t **freetree = &msp->ms_freetree[txg & TXG_MASK];
range_tree_t **freed_tree =
&msp->ms_freetree[TXG_CLEAN(txg) & TXG_MASK];
dmu_tx_t *tx;
uint64_t object = space_map_object(msp->ms_sm);
@ -2204,14 +2199,14 @@ metaslab_sync(metaslab_t *msp, uint64_t txg)
/*
* This metaslab has just been added so there's no work to do now.
*/
if (*freetree == NULL) {
if (msp->ms_freeingtree == NULL) {
ASSERT3P(alloctree, ==, NULL);
return;
}
ASSERT3P(alloctree, !=, NULL);
ASSERT3P(*freetree, !=, NULL);
ASSERT3P(*freed_tree, !=, NULL);
ASSERT3P(msp->ms_freeingtree, !=, NULL);
ASSERT3P(msp->ms_freedtree, !=, NULL);
/*
* Normally, we don't want to process a metaslab if there
@ -2219,14 +2214,14 @@ metaslab_sync(metaslab_t *msp, uint64_t txg)
* is being forced to condense we need to let it through.
*/
if (range_tree_space(alloctree) == 0 &&
range_tree_space(*freetree) == 0 &&
range_tree_space(msp->ms_freeingtree) == 0 &&
!msp->ms_condense_wanted)
return;
/*
* The only state that can actually be changing concurrently with
* metaslab_sync() is the metaslab's ms_tree. No other thread can
* be modifying this txg's alloctree, freetree, freed_tree, or
* be modifying this txg's alloctree, freeingtree, freedtree, or
* space_map_phys_t. Therefore, we only hold ms_lock to satify
* space map ASSERTs. We drop it whenever we call into the DMU,
* because the DMU can call down to us (e.g. via zio_free()) at
@ -2263,7 +2258,7 @@ metaslab_sync(metaslab_t *msp, uint64_t txg)
metaslab_condense(msp, txg, tx);
} else {
space_map_write(msp->ms_sm, alloctree, SM_ALLOC, tx);
space_map_write(msp->ms_sm, *freetree, SM_FREE, tx);
space_map_write(msp->ms_sm, msp->ms_freeingtree, SM_FREE, tx);
}
if (msp->ms_loaded) {
@ -2285,7 +2280,7 @@ metaslab_sync(metaslab_t *msp, uint64_t txg)
* to accurately reflect all free space even if some space
* is not yet available for allocation (i.e. deferred).
*/
space_map_histogram_add(msp->ms_sm, *freed_tree, tx);
space_map_histogram_add(msp->ms_sm, msp->ms_freedtree, tx);
/*
* Add back any deferred free space that has not been
@ -2307,7 +2302,7 @@ metaslab_sync(metaslab_t *msp, uint64_t txg)
* then we will lose some accuracy but will correct it the next
* time we load the space map.
*/
space_map_histogram_add(msp->ms_sm, *freetree, tx);
space_map_histogram_add(msp->ms_sm, msp->ms_freeingtree, tx);
metaslab_group_histogram_add(mg, msp);
metaslab_group_histogram_verify(mg);
@ -2315,20 +2310,21 @@ metaslab_sync(metaslab_t *msp, uint64_t txg)
/*
* For sync pass 1, we avoid traversing this txg's free range tree
* and instead will just swap the pointers for freetree and
* freed_tree. We can safely do this since the freed_tree is
* and instead will just swap the pointers for freeingtree and
* freedtree. We can safely do this since the freed_tree is
* guaranteed to be empty on the initial pass.
*/
if (spa_sync_pass(spa) == 1) {
range_tree_swap(freetree, freed_tree);
range_tree_swap(&msp->ms_freeingtree, &msp->ms_freedtree);
} else {
range_tree_vacate(*freetree, range_tree_add, *freed_tree);
range_tree_vacate(msp->ms_freeingtree,
range_tree_add, msp->ms_freedtree);
}
range_tree_vacate(alloctree, NULL, NULL);
ASSERT0(range_tree_space(msp->ms_alloctree[txg & TXG_MASK]));
ASSERT0(range_tree_space(msp->ms_alloctree[TXG_CLEAN(txg) & TXG_MASK]));
ASSERT0(range_tree_space(msp->ms_freetree[txg & TXG_MASK]));
ASSERT0(range_tree_space(msp->ms_freeingtree));
mutex_exit(&msp->ms_lock);
@ -2350,7 +2346,6 @@ metaslab_sync_done(metaslab_t *msp, uint64_t txg)
metaslab_group_t *mg = msp->ms_group;
vdev_t *vd = mg->mg_vd;
spa_t *spa = vd->vdev_spa;
range_tree_t **freed_tree;
range_tree_t **defer_tree;
int64_t alloc_delta, defer_delta;
uint64_t free_space;
@ -2363,20 +2358,24 @@ metaslab_sync_done(metaslab_t *msp, uint64_t txg)
/*
* If this metaslab is just becoming available, initialize its
* alloctrees, freetrees, and defertree and add its capacity to
* the vdev.
* range trees and add its capacity to the vdev.
*/
if (msp->ms_freetree[TXG_CLEAN(txg) & TXG_MASK] == NULL) {
if (msp->ms_freedtree == NULL) {
for (t = 0; t < TXG_SIZE; t++) {
ASSERT(msp->ms_alloctree[t] == NULL);
ASSERT(msp->ms_freetree[t] == NULL);
msp->ms_alloctree[t] = range_tree_create(NULL, msp,
&msp->ms_lock);
msp->ms_freetree[t] = range_tree_create(NULL, msp,
&msp->ms_lock);
}
ASSERT3P(msp->ms_freeingtree, ==, NULL);
msp->ms_freeingtree = range_tree_create(NULL, msp,
&msp->ms_lock);
ASSERT3P(msp->ms_freedtree, ==, NULL);
msp->ms_freedtree = range_tree_create(NULL, msp,
&msp->ms_lock);
for (t = 0; t < TXG_DEFER_SIZE; t++) {
ASSERT(msp->ms_defertree[t] == NULL);
@ -2387,7 +2386,6 @@ metaslab_sync_done(metaslab_t *msp, uint64_t txg)
vdev_space_update(vd, 0, 0, msp->ms_size);
}
freed_tree = &msp->ms_freetree[TXG_CLEAN(txg) & TXG_MASK];
defer_tree = &msp->ms_defertree[txg % TXG_DEFER_SIZE];
free_space = metaslab_class_get_space(spa_normal_class(spa)) -
@ -2399,7 +2397,7 @@ metaslab_sync_done(metaslab_t *msp, uint64_t txg)
defer_delta = 0;
alloc_delta = space_map_alloc_delta(msp->ms_sm);
if (defer_allowed) {
defer_delta = range_tree_space(*freed_tree) -
defer_delta = range_tree_space(msp->ms_freedtree) -
range_tree_space(*defer_tree);
} else {
defer_delta -= range_tree_space(*defer_tree);
@ -2407,9 +2405,6 @@ metaslab_sync_done(metaslab_t *msp, uint64_t txg)
vdev_space_update(vd, alloc_delta + defer_delta, defer_delta, 0);
ASSERT0(range_tree_space(msp->ms_alloctree[txg & TXG_MASK]));
ASSERT0(range_tree_space(msp->ms_freetree[txg & TXG_MASK]));
/*
* If there's a metaslab_load() in progress, wait for it to complete
* so that we have a consistent view of the in-core space map.
@ -2425,9 +2420,9 @@ metaslab_sync_done(metaslab_t *msp, uint64_t txg)
range_tree_vacate(*defer_tree,
msp->ms_loaded ? range_tree_add : NULL, msp->ms_tree);
if (defer_allowed) {
range_tree_swap(freed_tree, defer_tree);
range_tree_swap(&msp->ms_freedtree, defer_tree);
} else {
range_tree_vacate(*freed_tree,
range_tree_vacate(msp->ms_freedtree,
msp->ms_loaded ? range_tree_add : NULL, msp->ms_tree);
}
@ -3252,10 +3247,10 @@ metaslab_free_dva(spa_t *spa, const dva_t *dva, uint64_t txg, boolean_t now)
range_tree_add(msp->ms_tree, offset, size);
msp->ms_max_size = metaslab_block_maxsize(msp);
} else {
if (range_tree_space(msp->ms_freetree[txg & TXG_MASK]) == 0)
VERIFY3U(txg, ==, spa->spa_syncing_txg);
if (range_tree_space(msp->ms_freeingtree) == 0)
vdev_dirty(vd, VDD_METASLAB, msp, txg);
range_tree_add(msp->ms_freetree[txg & TXG_MASK],
offset, size);
range_tree_add(msp->ms_freeingtree, offset, size);
}
mutex_exit(&msp->ms_lock);
@ -3543,8 +3538,8 @@ metaslab_check_free(spa_t *spa, const blkptr_t *bp)
if (msp->ms_loaded)
range_tree_verify(msp->ms_tree, offset, size);
for (j = 0; j < TXG_SIZE; j++)
range_tree_verify(msp->ms_freetree[j], offset, size);
range_tree_verify(msp->ms_freeingtree, offset, size);
range_tree_verify(msp->ms_freedtree, offset, size);
for (j = 0; j < TXG_DEFER_SIZE; j++)
range_tree_verify(msp->ms_defertree[j], offset, size);
}