Skip to content

Commit

Permalink
Browse files Browse the repository at this point in the history
…fs#4106

4101 metaslab_debug should allow for fine-grained control
4102 space_maps should store more information about themselves
4103 space map object blocksize should be increased
4105 removing a mirrored log device results in a leaked object
4106 asynchronously load metaslab
Reviewed by: Matthew Ahrens <[email protected]>
Reviewed by: Adam Leventhal <[email protected]>
Reviewed by: Sebastien Roy <[email protected]>
Approved by: Garrett D'Amore <[email protected]>

Prior to this patch, space_maps were preferred solely based on the
amount of free space left in each. Unfortunately, this heuristic didn't
contain any information about the make-up of that free space, which
meant we could keep preferring and loading a highly fragmented space map
that wouldn't actually have enough contiguous space to satisfy the
allocation; then unloading that space_map and repeating the process.

This change modifies the space_map's to store additional information
about the contiguous space in the space_map, so that we can use this
information to make a better decision about which space_map to load.
This requires reallocating all space_map objects to increase their
bonus buffer size sizes enough to fit the new metadata.

The above feature can be enabled via a new feature flag introduced by
this change: com.delphix:spacemap_histogram

In addition to the above, this patch allows the space_map block size to
be increase. Currently the block size is set to be 4K in size, which has
certain implications including the following:

    * 4K sector devices will not see any compression benefit
    * large space_maps require more metadata on-disk
    * large space_maps require more time to load (typically random reads)

Now the space_map block size can adjust as needed up to the maximum size
set via the space_map_max_blksz variable.

A bug was fixed which resulted in potentially leaking an object when
removing a mirrored log device. The previous logic for vdev_remove() did
not deal with removing top-level vdevs that are interior vdevs (i.e.
mirror) correctly. The problem would occur when removing a mirrored log
device, and result in the DTL space map object being leaked; because
top-level vdevs don't have DTL space map objects associated with them.

References:
  https://www.illumos.org/issues/4101
  https://www.illumos.org/issues/4102
  https://www.illumos.org/issues/4103
  https://www.illumos.org/issues/4105
  https://www.illumos.org/issues/4106
  illumos/illumos-gate@0713e23

Porting notes:

A handful of kmem_alloc() calls were converted to kmem_zalloc(). Also,
the KM_PUSHPAGE and TQ_PUSHPAGE flags were used as necessary.

Ported-by: Tim Chase <[email protected]>
Signed-off-by: Prakash Surya <[email protected]>
Signed-off-by: Brian Behlendorf <[email protected]>
Closes openzfs#2488
  • Loading branch information
grwilson authored and behlendorf committed Jul 22, 2014
1 parent 1be627f commit 93cf207
Show file tree
Hide file tree
Showing 24 changed files with 2,466 additions and 1,399 deletions.
233 changes: 156 additions & 77 deletions cmd/zdb/zdb.c
Original file line number Diff line number Diff line change
Expand Up @@ -246,7 +246,7 @@ const char histo_stars[] = "****************************************";
const int histo_width = sizeof (histo_stars) - 1;

static void
dump_histogram(const uint64_t *histo, int size)
dump_histogram(const uint64_t *histo, int size, int offset)
{
int i;
int minidx = size - 1;
Expand All @@ -267,7 +267,7 @@ dump_histogram(const uint64_t *histo, int size)

for (i = minidx; i <= maxidx; i++) {
(void) printf("\t\t\t%3u: %6llu %s\n",
i, (u_longlong_t)histo[i],
i + offset, (u_longlong_t)histo[i],
&histo_stars[(max - histo[i]) * histo_width / max]);
}
}
Expand Down Expand Up @@ -320,19 +320,19 @@ dump_zap_stats(objset_t *os, uint64_t object)
(u_longlong_t)zs.zs_salt);

(void) printf("\t\tLeafs with 2^n pointers:\n");
dump_histogram(zs.zs_leafs_with_2n_pointers, ZAP_HISTOGRAM_SIZE);
dump_histogram(zs.zs_leafs_with_2n_pointers, ZAP_HISTOGRAM_SIZE, 0);

(void) printf("\t\tBlocks with n*5 entries:\n");
dump_histogram(zs.zs_blocks_with_n5_entries, ZAP_HISTOGRAM_SIZE);
dump_histogram(zs.zs_blocks_with_n5_entries, ZAP_HISTOGRAM_SIZE, 0);

(void) printf("\t\tBlocks n/10 full:\n");
dump_histogram(zs.zs_blocks_n_tenths_full, ZAP_HISTOGRAM_SIZE);
dump_histogram(zs.zs_blocks_n_tenths_full, ZAP_HISTOGRAM_SIZE, 0);

(void) printf("\t\tEntries with n chunks:\n");
dump_histogram(zs.zs_entries_using_n_chunks, ZAP_HISTOGRAM_SIZE);
dump_histogram(zs.zs_entries_using_n_chunks, ZAP_HISTOGRAM_SIZE, 0);

(void) printf("\t\tBuckets with n entries:\n");
dump_histogram(zs.zs_buckets_with_n_entries, ZAP_HISTOGRAM_SIZE);
dump_histogram(zs.zs_buckets_with_n_entries, ZAP_HISTOGRAM_SIZE, 0);
}

/*ARGSUSED*/
Expand Down Expand Up @@ -521,26 +521,87 @@ dump_zpldir(objset_t *os, uint64_t object, void *data, size_t size)
zap_cursor_fini(&zc);
}

int
get_dtl_refcount(vdev_t *vd)
{
int refcount = 0;
int c;

if (vd->vdev_ops->vdev_op_leaf) {
space_map_t *sm = vd->vdev_dtl_sm;

if (sm != NULL &&
sm->sm_dbuf->db_size == sizeof (space_map_phys_t))
return (1);
return (0);
}

for (c = 0; c < vd->vdev_children; c++)
refcount += get_dtl_refcount(vd->vdev_child[c]);
return (refcount);
}

int
get_metaslab_refcount(vdev_t *vd)
{
int refcount = 0;
int c, m;

if (vd->vdev_top == vd) {
for (m = 0; m < vd->vdev_ms_count; m++) {
space_map_t *sm = vd->vdev_ms[m]->ms_sm;

if (sm != NULL &&
sm->sm_dbuf->db_size == sizeof (space_map_phys_t))
refcount++;
}
}
for (c = 0; c < vd->vdev_children; c++)
refcount += get_metaslab_refcount(vd->vdev_child[c]);

return (refcount);
}

static int
verify_spacemap_refcounts(spa_t *spa)
{
int expected_refcount, actual_refcount;

expected_refcount = spa_feature_get_refcount(spa,
&spa_feature_table[SPA_FEATURE_SPACEMAP_HISTOGRAM]);
actual_refcount = get_dtl_refcount(spa->spa_root_vdev);
actual_refcount += get_metaslab_refcount(spa->spa_root_vdev);

if (expected_refcount != actual_refcount) {
(void) printf("space map refcount mismatch: expected %d != "
"actual %d\n", expected_refcount, actual_refcount);
return (2);
}
return (0);
}

static void
dump_spacemap(objset_t *os, space_map_obj_t *smo, space_map_t *sm)
dump_spacemap(objset_t *os, space_map_t *sm)
{
uint64_t alloc, offset, entry;
uint8_t mapshift = sm->sm_shift;
uint64_t mapstart = sm->sm_start;
char *ddata[] = { "ALLOC", "FREE", "CONDENSE", "INVALID",
"INVALID", "INVALID", "INVALID", "INVALID" };

if (smo->smo_object == 0)
if (sm == NULL)
return;

/*
* Print out the freelist entries in both encoded and decoded form.
*/
alloc = 0;
for (offset = 0; offset < smo->smo_objsize; offset += sizeof (entry)) {
VERIFY3U(0, ==, dmu_read(os, smo->smo_object, offset,
for (offset = 0; offset < space_map_length(sm);
offset += sizeof (entry)) {
uint8_t mapshift = sm->sm_shift;

VERIFY0(dmu_read(os, space_map_object(sm), offset,
sizeof (entry), &entry, DMU_READ_PREFETCH));
if (SM_DEBUG_DECODE(entry)) {

(void) printf("\t [%6llu] %s: txg %llu, pass %llu\n",
(u_longlong_t)(offset / sizeof (entry)),
ddata[SM_DEBUG_ACTION_DECODE(entry)],
Expand All @@ -552,71 +613,85 @@ dump_spacemap(objset_t *os, space_map_obj_t *smo, space_map_t *sm)
(u_longlong_t)(offset / sizeof (entry)),
SM_TYPE_DECODE(entry) == SM_ALLOC ? 'A' : 'F',
(u_longlong_t)((SM_OFFSET_DECODE(entry) <<
mapshift) + mapstart),
mapshift) + sm->sm_start),
(u_longlong_t)((SM_OFFSET_DECODE(entry) <<
mapshift) + mapstart + (SM_RUN_DECODE(entry) <<
mapshift)),
mapshift) + sm->sm_start +
(SM_RUN_DECODE(entry) << mapshift)),
(u_longlong_t)(SM_RUN_DECODE(entry) << mapshift));
if (SM_TYPE_DECODE(entry) == SM_ALLOC)
alloc += SM_RUN_DECODE(entry) << mapshift;
else
alloc -= SM_RUN_DECODE(entry) << mapshift;
}
}
if (alloc != smo->smo_alloc) {
if (alloc != space_map_allocated(sm)) {
(void) printf("space_map_object alloc (%llu) INCONSISTENT "
"with space map summary (%llu)\n",
(u_longlong_t)smo->smo_alloc, (u_longlong_t)alloc);
(u_longlong_t)space_map_allocated(sm), (u_longlong_t)alloc);
}
}

static void
dump_metaslab_stats(metaslab_t *msp)
{
char maxbuf[32];
space_map_t *sm = msp->ms_map;
avl_tree_t *t = sm->sm_pp_root;
int free_pct = sm->sm_space * 100 / sm->sm_size;
range_tree_t *rt = msp->ms_tree;
avl_tree_t *t = &msp->ms_size_tree;
int free_pct = range_tree_space(rt) * 100 / msp->ms_size;

zdb_nicenum(space_map_maxsize(sm), maxbuf);
zdb_nicenum(metaslab_block_maxsize(msp), maxbuf);

(void) printf("\t %25s %10lu %7s %6s %4s %4d%%\n",
"segments", avl_numnodes(t), "maxsize", maxbuf,
"freepct", free_pct);
(void) printf("\tIn-memory histogram:\n");
dump_histogram(rt->rt_histogram, RANGE_TREE_HISTOGRAM_SIZE, 0);
}

static void
dump_metaslab(metaslab_t *msp)
{
vdev_t *vd = msp->ms_group->mg_vd;
spa_t *spa = vd->vdev_spa;
space_map_t *sm = msp->ms_map;
space_map_obj_t *smo = &msp->ms_smo;
space_map_t *sm = msp->ms_sm;
char freebuf[32];

zdb_nicenum(sm->sm_size - smo->smo_alloc, freebuf);
zdb_nicenum(msp->ms_size - space_map_allocated(sm), freebuf);

(void) printf(
"\tmetaslab %6llu offset %12llx spacemap %6llu free %5s\n",
(u_longlong_t)(sm->sm_start / sm->sm_size),
(u_longlong_t)sm->sm_start, (u_longlong_t)smo->smo_object, freebuf);
(u_longlong_t)msp->ms_id, (u_longlong_t)msp->ms_start,
(u_longlong_t)space_map_object(sm), freebuf);

if (dump_opt['m'] > 1 && !dump_opt['L']) {
if (dump_opt['m'] > 2 && !dump_opt['L']) {
mutex_enter(&msp->ms_lock);
space_map_load_wait(sm);
if (!sm->sm_loaded)
VERIFY(space_map_load(sm, zfs_metaslab_ops,
SM_FREE, smo, spa->spa_meta_objset) == 0);
metaslab_load_wait(msp);
if (!msp->ms_loaded) {
VERIFY0(metaslab_load(msp));
range_tree_stat_verify(msp->ms_tree);
}
dump_metaslab_stats(msp);
space_map_unload(sm);
metaslab_unload(msp);
mutex_exit(&msp->ms_lock);
}

if (dump_opt['d'] > 5 || dump_opt['m'] > 2) {
ASSERT(sm->sm_size == (1ULL << vd->vdev_ms_shift));
if (dump_opt['m'] > 1 && sm != NULL &&
spa_feature_is_active(spa,
&spa_feature_table[SPA_FEATURE_SPACEMAP_HISTOGRAM])) {
/*
* The space map histogram represents free space in chunks
* of sm_shift (i.e. bucket 0 refers to 2^sm_shift).
*/
(void) printf("\tOn-disk histogram:\n");
dump_histogram(sm->sm_phys->smp_histogram,
SPACE_MAP_HISTOGRAM_SIZE(sm), sm->sm_shift);
}

if (dump_opt['d'] > 5 || dump_opt['m'] > 3) {
ASSERT(msp->ms_size == (1ULL << vd->vdev_ms_shift));

mutex_enter(&msp->ms_lock);
dump_spacemap(spa->spa_meta_objset, smo, sm);
dump_spacemap(spa->spa_meta_objset, msp->ms_sm);
mutex_exit(&msp->ms_lock);
}
}
Expand Down Expand Up @@ -812,9 +887,9 @@ dump_all_ddts(spa_t *spa)
}

static void
dump_dtl_seg(space_map_t *sm, uint64_t start, uint64_t size)
dump_dtl_seg(void *arg, uint64_t start, uint64_t size)
{
char *prefix = (void *)sm;
char *prefix = arg;

(void) printf("%s [%llu,%llu) length %llu\n",
prefix,
Expand Down Expand Up @@ -845,17 +920,17 @@ dump_dtl(vdev_t *vd, int indent)
required ? "DTL-required" : "DTL-expendable");

for (t = 0; t < DTL_TYPES; t++) {
space_map_t *sm = &vd->vdev_dtl[t];
if (sm->sm_space == 0)
range_tree_t *rt = vd->vdev_dtl[t];
if (range_tree_space(rt) == 0)
continue;
(void) snprintf(prefix, sizeof (prefix), "\t%*s%s",
indent + 2, "", name[t]);
mutex_enter(sm->sm_lock);
space_map_walk(sm, dump_dtl_seg, (void *)prefix);
mutex_exit(sm->sm_lock);
mutex_enter(rt->rt_lock);
range_tree_walk(rt, dump_dtl_seg, prefix);
mutex_exit(rt->rt_lock);
if (dump_opt['d'] > 5 && vd->vdev_children == 0)
dump_spacemap(spa->spa_meta_objset,
&vd->vdev_dtl_smo, sm);
vd->vdev_dtl_sm);
}

for (c = 0; c < vd->vdev_children; c++)
Expand Down Expand Up @@ -2261,39 +2336,17 @@ zdb_blkptr_cb(spa_t *spa, zilog_t *zilog, const blkptr_t *bp,
}

static void
zdb_leak(space_map_t *sm, uint64_t start, uint64_t size)
zdb_leak(void *arg, uint64_t start, uint64_t size)
{
vdev_t *vd = sm->sm_ppd;
vdev_t *vd = arg;

(void) printf("leaked space: vdev %llu, offset 0x%llx, size %llu\n",
(u_longlong_t)vd->vdev_id, (u_longlong_t)start, (u_longlong_t)size);
}

/* ARGSUSED */
static void
zdb_space_map_load(space_map_t *sm)
{
}

static void
zdb_space_map_unload(space_map_t *sm)
{
space_map_vacate(sm, zdb_leak, sm);
}

/* ARGSUSED */
static void
zdb_space_map_claim(space_map_t *sm, uint64_t start, uint64_t size)
{
}

static space_map_ops_t zdb_space_map_ops = {
zdb_space_map_load,
zdb_space_map_unload,
static metaslab_ops_t zdb_metaslab_ops = {
NULL, /* alloc */
zdb_space_map_claim,
NULL, /* free */
NULL /* maxsize */
NULL /* fragmented */
};

static void
Expand Down Expand Up @@ -2350,11 +2403,21 @@ zdb_leak_init(spa_t *spa, zdb_cb_t *zcb)
for (m = 0; m < vd->vdev_ms_count; m++) {
metaslab_t *msp = vd->vdev_ms[m];
mutex_enter(&msp->ms_lock);
space_map_unload(msp->ms_map);
VERIFY(space_map_load(msp->ms_map,
&zdb_space_map_ops, SM_ALLOC, &msp->ms_smo,
spa->spa_meta_objset) == 0);
msp->ms_map->sm_ppd = vd;
metaslab_unload(msp);

/*
* For leak detection, we overload the metaslab
* ms_tree to contain allocated segments
* instead of free segments. As a result,
* we can't use the normal metaslab_load/unload
* interfaces.
*/
if (msp->ms_sm != NULL) {
msp->ms_ops = &zdb_metaslab_ops;
VERIFY0(space_map_load(msp->ms_sm,
msp->ms_tree, SM_ALLOC));
msp->ms_loaded = B_TRUE;
}
mutex_exit(&msp->ms_lock);
}
}
Expand All @@ -2379,7 +2442,20 @@ zdb_leak_fini(spa_t *spa)
for (m = 0; m < vd->vdev_ms_count; m++) {
metaslab_t *msp = vd->vdev_ms[m];
mutex_enter(&msp->ms_lock);
space_map_unload(msp->ms_map);

/*
* The ms_tree has been overloaded to
* contain allocated segments. Now that we
* finished traversing all blocks, any
* block that remains in the ms_tree
* represents an allocated block that we
* did not claim during the traversal.
* Claimed blocks would have been removed
* from the ms_tree.
*/
range_tree_vacate(msp->ms_tree, zdb_leak, vd);
msp->ms_loaded = B_FALSE;

mutex_exit(&msp->ms_lock);
}
}
Expand Down Expand Up @@ -2596,7 +2672,7 @@ dump_block_stats(spa_t *spa)
"(in 512-byte sectors): "
"number of blocks\n");
dump_histogram(zb->zb_psize_histogram,
PSIZE_HISTO_SIZE);
PSIZE_HISTO_SIZE, 0);
}
}
}
Expand Down Expand Up @@ -2769,6 +2845,9 @@ dump_zpool(spa_t *spa)
if (dump_opt['b'] || dump_opt['c'])
rc = dump_block_stats(spa);

if (rc == 0)
rc = verify_spacemap_refcounts(spa);

if (dump_opt['s'])
show_pool_stats(spa);

Expand Down
Loading

0 comments on commit 93cf207

Please sign in to comment.