/*
* Copyright (c) 2009, 2010, Oracle and/or its affiliates. All rights reserved.
* Copyright (c) 2012, 2016 by Delphix. All rights reserved.
+ * Copyright (c) 2022 by Pawel Jakub Dawidek
*/
#include <sys/zfs_context.h>
#include <sys/spa_impl.h>
#include <sys/zio.h>
#include <sys/ddt.h>
+#include <sys/ddt_impl.h>
#include <sys/zap.h>
#include <sys/dmu_tx.h>
#include <sys/arc.h>
#include <sys/dsl_pool.h>
#include <sys/zio_checksum.h>
-#include <sys/zio_compress.h>
#include <sys/dsl_scan.h>
#include <sys/abd.h>
ddt_object_name(ddt, type, class, name);
- ASSERT(*objectp == 0);
- VERIFY(ddt_ops[type]->ddt_op_create(os, objectp, tx, prehash) == 0);
- ASSERT(*objectp != 0);
+ ASSERT3U(*objectp, ==, 0);
+ VERIFY0(ddt_ops[type]->ddt_op_create(os, objectp, tx, prehash));
+ ASSERT3U(*objectp, !=, 0);
- VERIFY(zap_add(os, DMU_POOL_DIRECTORY_OBJECT, name,
- sizeof (uint64_t), 1, objectp, tx) == 0);
+ VERIFY0(zap_add(os, DMU_POOL_DIRECTORY_OBJECT, name,
+ sizeof (uint64_t), 1, objectp, tx));
- VERIFY(zap_add(os, spa->spa_ddt_stat_object, name,
+ VERIFY0(zap_add(os, spa->spa_ddt_stat_object, name,
sizeof (uint64_t), sizeof (ddt_histogram_t) / sizeof (uint64_t),
- &ddt->ddt_histogram[type][class], tx) == 0);
+ &ddt->ddt_histogram[type][class], tx));
}
static void
ddt_object_name(ddt, type, class, name);
- ASSERT(*objectp != 0);
+ ASSERT3U(*objectp, !=, 0);
ASSERT(ddt_histogram_empty(&ddt->ddt_histogram[type][class]));
- VERIFY(ddt_object_count(ddt, type, class, &count) == 0 && count == 0);
- VERIFY(zap_remove(os, DMU_POOL_DIRECTORY_OBJECT, name, tx) == 0);
- VERIFY(zap_remove(os, spa->spa_ddt_stat_object, name, tx) == 0);
- VERIFY(ddt_ops[type]->ddt_op_destroy(os, *objectp, tx) == 0);
+ VERIFY0(ddt_object_count(ddt, type, class, &count));
+ VERIFY0(count);
+ VERIFY0(zap_remove(os, DMU_POOL_DIRECTORY_OBJECT, name, tx));
+ VERIFY0(zap_remove(os, spa->spa_ddt_stat_object, name, tx));
+ VERIFY0(ddt_ops[type]->ddt_op_destroy(os, *objectp, tx));
memset(&ddt->ddt_object_stats[type][class], 0, sizeof (ddt_object_t));
*objectp = 0;
ddt_object_name(ddt, type, class, name);
- VERIFY(zap_update(ddt->ddt_os, ddt->ddt_spa->spa_ddt_stat_object, name,
+ VERIFY0(zap_update(ddt->ddt_os, ddt->ddt_spa->spa_ddt_stat_object, name,
sizeof (uint64_t), sizeof (ddt_histogram_t) / sizeof (uint64_t),
- &ddt->ddt_histogram[type][class], tx) == 0);
+ &ddt->ddt_histogram[type][class], tx));
/*
* Cache DDT statistics; this is the only time they'll change.
*/
- VERIFY(ddt_object_info(ddt, type, class, &doi) == 0);
- VERIFY(ddt_object_count(ddt, type, class, &count) == 0);
+ VERIFY0(ddt_object_info(ddt, type, class, &doi));
+ VERIFY0(ddt_object_count(ddt, type, class, &count));
ddo->ddo_count = count;
ddo->ddo_dspace = doi.doi_physical_blocks_512 << 9;
ddo->ddo_mspace = doi.doi_fill_count * doi.doi_data_block_size;
}
+static boolean_t
+ddt_object_exists(ddt_t *ddt, enum ddt_type type, enum ddt_class class)
+{
+ return (!!ddt->ddt_object[type][class]);
+}
+
static int
ddt_object_lookup(ddt_t *ddt, enum ddt_type type, enum ddt_class class,
ddt_entry_t *dde)
ddt->ddt_object[type][class], dde);
}
-int
+static int
ddt_object_update(ddt_t *ddt, enum ddt_type type, enum ddt_class class,
ddt_entry_t *dde, dmu_tx_t *tx)
{
doi));
}
-boolean_t
-ddt_object_exists(ddt_t *ddt, enum ddt_type type, enum ddt_class class)
-{
- return (!!ddt->ddt_object[type][class]);
-}
-
void
ddt_object_name(ddt_t *ddt, enum ddt_type type, enum ddt_class class,
char *name)
void
ddt_bp_fill(const ddt_phys_t *ddp, blkptr_t *bp, uint64_t txg)
{
- ASSERT(txg != 0);
+ ASSERT3U(txg, !=, 0);
for (int d = 0; d < SPA_DVAS_PER_BP; d++)
bp->blk_dva[d] = ddp->ddp_dva[d];
void
ddt_phys_fill(ddt_phys_t *ddp, const blkptr_t *bp)
{
- ASSERT(ddp->ddp_phys_birth == 0);
+ ASSERT0(ddp->ddp_phys_birth);
for (int d = 0; d < SPA_DVAS_PER_BP; d++)
ddp->ddp_dva[d] = bp->blk_dva[d];
ddt_phys_decref(ddt_phys_t *ddp)
{
if (ddp) {
- ASSERT(ddp->ddp_refcnt > 0);
+ ASSERT3U(ddp->ddp_refcnt, >, 0);
ddp->ddp_refcnt--;
}
}
-void
+static void
ddt_phys_free(ddt_t *ddt, ddt_key_t *ddk, ddt_phys_t *ddp, uint64_t txg)
{
blkptr_t blk;
return (refcnt);
}
-static void
-ddt_stat_generate(ddt_t *ddt, ddt_entry_t *dde, ddt_stat_t *dds)
-{
- spa_t *spa = ddt->ddt_spa;
- ddt_phys_t *ddp = dde->dde_phys;
- ddt_key_t *ddk = &dde->dde_key;
- uint64_t lsize = DDK_GET_LSIZE(ddk);
- uint64_t psize = DDK_GET_PSIZE(ddk);
-
- memset(dds, 0, sizeof (*dds));
-
- for (int p = 0; p < DDT_PHYS_TYPES; p++, ddp++) {
- uint64_t dsize = 0;
- uint64_t refcnt = ddp->ddp_refcnt;
-
- if (ddp->ddp_phys_birth == 0)
- continue;
-
- for (int d = 0; d < DDE_GET_NDVAS(dde); d++)
- dsize += dva_get_dsize_sync(spa, &ddp->ddp_dva[d]);
-
- dds->dds_blocks += 1;
- dds->dds_lsize += lsize;
- dds->dds_psize += psize;
- dds->dds_dsize += dsize;
-
- dds->dds_ref_blocks += refcnt;
- dds->dds_ref_lsize += lsize * refcnt;
- dds->dds_ref_psize += psize * refcnt;
- dds->dds_ref_dsize += dsize * refcnt;
- }
-}
-
-void
-ddt_stat_add(ddt_stat_t *dst, const ddt_stat_t *src, uint64_t neg)
-{
- const uint64_t *s = (const uint64_t *)src;
- uint64_t *d = (uint64_t *)dst;
- uint64_t *d_end = (uint64_t *)(dst + 1);
-
- ASSERT(neg == 0 || neg == -1ULL); /* add or subtract */
-
- for (int i = 0; i < d_end - d; i++)
- d[i] += (s[i] ^ neg) - neg;
-}
-
-static void
-ddt_stat_update(ddt_t *ddt, ddt_entry_t *dde, uint64_t neg)
-{
- ddt_stat_t dds;
- ddt_histogram_t *ddh;
- int bucket;
-
- ddt_stat_generate(ddt, dde, &dds);
-
- bucket = highbit64(dds.dds_ref_blocks) - 1;
- ASSERT(bucket >= 0);
-
- ddh = &ddt->ddt_histogram[dde->dde_type][dde->dde_class];
-
- ddt_stat_add(&ddh->ddh_stat[bucket], &dds, neg);
-}
-
-void
-ddt_histogram_add(ddt_histogram_t *dst, const ddt_histogram_t *src)
-{
- for (int h = 0; h < 64; h++)
- ddt_stat_add(&dst->ddh_stat[h], &src->ddh_stat[h], 0);
-}
-
-void
-ddt_histogram_stat(ddt_stat_t *dds, const ddt_histogram_t *ddh)
-{
- memset(dds, 0, sizeof (*dds));
-
- for (int h = 0; h < 64; h++)
- ddt_stat_add(dds, &ddh->ddh_stat[h], 0);
-}
-
-boolean_t
-ddt_histogram_empty(const ddt_histogram_t *ddh)
-{
- const uint64_t *s = (const uint64_t *)ddh;
- const uint64_t *s_end = (const uint64_t *)(ddh + 1);
-
- while (s < s_end)
- if (*s++ != 0)
- return (B_FALSE);
-
- return (B_TRUE);
-}
-
-void
-ddt_get_dedup_object_stats(spa_t *spa, ddt_object_t *ddo_total)
-{
- /* Sum the statistics we cached in ddt_object_sync(). */
- for (enum zio_checksum c = 0; c < ZIO_CHECKSUM_FUNCTIONS; c++) {
- ddt_t *ddt = spa->spa_ddt[c];
- for (enum ddt_type type = 0; type < DDT_TYPES; type++) {
- for (enum ddt_class class = 0; class < DDT_CLASSES;
- class++) {
- ddt_object_t *ddo =
- &ddt->ddt_object_stats[type][class];
- ddo_total->ddo_count += ddo->ddo_count;
- ddo_total->ddo_dspace += ddo->ddo_dspace;
- ddo_total->ddo_mspace += ddo->ddo_mspace;
- }
- }
- }
-
- /* ... and compute the averages. */
- if (ddo_total->ddo_count != 0) {
- ddo_total->ddo_dspace /= ddo_total->ddo_count;
- ddo_total->ddo_mspace /= ddo_total->ddo_count;
- }
-}
-
-void
-ddt_get_dedup_histogram(spa_t *spa, ddt_histogram_t *ddh)
-{
- for (enum zio_checksum c = 0; c < ZIO_CHECKSUM_FUNCTIONS; c++) {
- ddt_t *ddt = spa->spa_ddt[c];
- for (enum ddt_type type = 0; type < DDT_TYPES && ddt; type++) {
- for (enum ddt_class class = 0; class < DDT_CLASSES;
- class++) {
- ddt_histogram_add(ddh,
- &ddt->ddt_histogram_cache[type][class]);
- }
- }
- }
-}
-
-void
-ddt_get_dedup_stats(spa_t *spa, ddt_stat_t *dds_total)
-{
- ddt_histogram_t *ddh_total;
-
- ddh_total = kmem_zalloc(sizeof (ddt_histogram_t), KM_SLEEP);
- ddt_get_dedup_histogram(spa, ddh_total);
- ddt_histogram_stat(dds_total, ddh_total);
- kmem_free(ddh_total, sizeof (ddt_histogram_t));
-}
-
-uint64_t
-ddt_get_dedup_dspace(spa_t *spa)
-{
- ddt_stat_t dds_total;
-
- if (spa->spa_dedup_dspace != ~0ULL)
- return (spa->spa_dedup_dspace);
-
- memset(&dds_total, 0, sizeof (ddt_stat_t));
-
- /* Calculate and cache the stats */
- ddt_get_dedup_stats(spa, &dds_total);
- spa->spa_dedup_dspace = dds_total.dds_ref_dsize - dds_total.dds_dsize;
- return (spa->spa_dedup_dspace);
-}
-
-uint64_t
-ddt_get_pool_dedup_ratio(spa_t *spa)
-{
- ddt_stat_t dds_total = { 0 };
-
- ddt_get_dedup_stats(spa, &dds_total);
- if (dds_total.dds_dsize == 0)
- return (100);
-
- return (dds_total.dds_ref_dsize * 100 / dds_total.dds_dsize);
-}
-
-size_t
-ddt_compress(void *src, uchar_t *dst, size_t s_len, size_t d_len)
-{
- uchar_t *version = dst++;
- int cpfunc = ZIO_COMPRESS_ZLE;
- zio_compress_info_t *ci = &zio_compress_table[cpfunc];
- size_t c_len;
-
- ASSERT(d_len >= s_len + 1); /* no compression plus version byte */
-
- c_len = ci->ci_compress(src, dst, s_len, d_len - 1, ci->ci_level);
-
- if (c_len == s_len) {
- cpfunc = ZIO_COMPRESS_OFF;
- memcpy(dst, src, s_len);
- }
-
- *version = cpfunc;
- if (ZFS_HOST_BYTEORDER)
- *version |= DDT_COMPRESS_BYTEORDER_MASK;
-
- return (c_len + 1);
-}
-
-void
-ddt_decompress(uchar_t *src, void *dst, size_t s_len, size_t d_len)
-{
- uchar_t version = *src++;
- int cpfunc = version & DDT_COMPRESS_FUNCTION_MASK;
- zio_compress_info_t *ci = &zio_compress_table[cpfunc];
-
- if (ci->ci_decompress != NULL)
- (void) ci->ci_decompress(src, dst, s_len, d_len, ci->ci_level);
- else
- memcpy(dst, src, d_len);
-
- if (((version & DDT_COMPRESS_BYTEORDER_MASK) != 0) !=
- (ZFS_HOST_BYTEORDER != 0))
- byteswap_uint64_array(dst, d_len);
-}
-
ddt_t *
ddt_select(spa_t *spa, const blkptr_t *bp)
{
ASSERT(!dde->dde_loading);
for (int p = 0; p < DDT_PHYS_TYPES; p++)
- ASSERT(dde->dde_lead_zio[p] == NULL);
+ ASSERT3P(dde->dde_lead_zio[p], ==, NULL);
if (dde->dde_repair_abd != NULL)
abd_free(dde->dde_repair_abd);
ddt_entry_t *
ddt_lookup(ddt_t *ddt, const blkptr_t *bp, boolean_t add)
{
- ddt_entry_t *dde, dde_search;
+ ddt_key_t search;
+ ddt_entry_t *dde;
enum ddt_type type;
enum ddt_class class;
avl_index_t where;
ASSERT(MUTEX_HELD(&ddt->ddt_lock));
- ddt_key_fill(&dde_search.dde_key, bp);
+ ddt_key_fill(&search, bp);
- dde = avl_find(&ddt->ddt_tree, &dde_search, &where);
+ dde = avl_find(&ddt->ddt_tree, &search, &where);
if (dde == NULL) {
if (!add)
return (NULL);
- dde = ddt_alloc(&dde_search.dde_key);
+ dde = ddt_alloc(&search);
avl_insert(&ddt->ddt_tree, dde, where);
}
ddt_enter(ddt);
- ASSERT(dde->dde_loaded == B_FALSE);
- ASSERT(dde->dde_loading == B_TRUE);
+ ASSERT(!dde->dde_loaded);
+ ASSERT(dde->dde_loading);
dde->dde_type = type; /* will be DDT_TYPES if no entry found */
dde->dde_class = class; /* will be DDT_CLASSES if no entry found */
}
/*
- * Opaque struct used for ddt_key comparison
+ * Key comparison. Any struct wanting to make use of this function must have
+ * the key as the first element.
*/
#define DDT_KEY_CMP_LEN (sizeof (ddt_key_t) / sizeof (uint16_t))
} ddt_key_cmp_t;
int
-ddt_entry_compare(const void *x1, const void *x2)
+ddt_key_compare(const void *x1, const void *x2)
{
- const ddt_entry_t *dde1 = x1;
- const ddt_entry_t *dde2 = x2;
- const ddt_key_cmp_t *k1 = (const ddt_key_cmp_t *)&dde1->dde_key;
- const ddt_key_cmp_t *k2 = (const ddt_key_cmp_t *)&dde2->dde_key;
+ const ddt_key_cmp_t *k1 = (const ddt_key_cmp_t *)x1;
+ const ddt_key_cmp_t *k2 = (const ddt_key_cmp_t *)x2;
int32_t cmp = 0;
for (int i = 0; i < DDT_KEY_CMP_LEN; i++) {
memset(ddt, 0, sizeof (ddt_t));
mutex_init(&ddt->ddt_lock, NULL, MUTEX_DEFAULT, NULL);
- avl_create(&ddt->ddt_tree, ddt_entry_compare,
+ avl_create(&ddt->ddt_tree, ddt_key_compare,
sizeof (ddt_entry_t), offsetof(ddt_entry_t, dde_node));
- avl_create(&ddt->ddt_repair_tree, ddt_entry_compare,
+ avl_create(&ddt->ddt_repair_tree, ddt_key_compare,
sizeof (ddt_entry_t), offsetof(ddt_entry_t, dde_node));
ddt->ddt_checksum = c;
ddt->ddt_spa = spa;
static void
ddt_table_free(ddt_t *ddt)
{
- ASSERT(avl_numnodes(&ddt->ddt_tree) == 0);
- ASSERT(avl_numnodes(&ddt->ddt_repair_tree) == 0);
+ ASSERT0(avl_numnodes(&ddt->ddt_tree));
+ ASSERT0(avl_numnodes(&ddt->ddt_repair_tree));
avl_destroy(&ddt->ddt_tree);
avl_destroy(&ddt->ddt_repair_tree);
mutex_destroy(&ddt->ddt_lock);
ASSERT(!dde->dde_loading);
for (int p = 0; p < DDT_PHYS_TYPES; p++, ddp++) {
- ASSERT(dde->dde_lead_zio[p] == NULL);
+ ASSERT3P(dde->dde_lead_zio[p], ==, NULL);
if (ddp->ddp_phys_birth == 0) {
- ASSERT(ddp->ddp_refcnt == 0);
+ ASSERT0(ddp->ddp_refcnt);
continue;
}
if (p == DDT_PHYS_DITTO) {
if (otype != DDT_TYPES &&
(otype != ntype || oclass != nclass || total_refcnt == 0)) {
- VERIFY(ddt_object_remove(ddt, otype, oclass, dde, tx) == 0);
- ASSERT(ddt_object_lookup(ddt, otype, oclass, dde) == ENOENT);
+ VERIFY0(ddt_object_remove(ddt, otype, oclass, dde, tx));
+ ASSERT3U(
+ ddt_object_lookup(ddt, otype, oclass, dde), ==, ENOENT);
}
if (total_refcnt != 0) {
ddt_stat_update(ddt, dde, 0);
if (!ddt_object_exists(ddt, ntype, nclass))
ddt_object_create(ddt, ntype, nclass, tx);
- VERIFY(ddt_object_update(ddt, ntype, nclass, dde, tx) == 0);
+ VERIFY0(ddt_object_update(ddt, ntype, nclass, dde, tx));
/*
* If the class changes, the order that we scan this bp
if (avl_numnodes(&ddt->ddt_tree) == 0)
return;
- ASSERT(spa->spa_uberblock.ub_version >= SPA_VERSION_DEDUP);
+ ASSERT3U(spa->spa_uberblock.ub_version, >=, SPA_VERSION_DEDUP);
if (spa->spa_ddt_stat_object == 0) {
spa->spa_ddt_stat_object = zap_create_link(ddt->ddt_os,
for (enum ddt_class class = 0; class < DDT_CLASSES; class++) {
if (ddt_object_exists(ddt, type, class)) {
ddt_object_sync(ddt, type, class, tx);
- VERIFY(ddt_object_count(ddt, type, class,
- &add) == 0);
+ VERIFY0(ddt_object_count(ddt, type, class,
+ &add));
count += add;
}
}
dmu_tx_t *tx;
zio_t *rio;
- ASSERT(spa_syncing_txg(spa) == txg);
+ ASSERT3U(spa_syncing_txg(spa), ==, txg);
tx = dmu_tx_create_assigned(spa->spa_dsl_pool, txg);
return (SET_ERROR(ENOENT));
}
+/*
+ * This function is used by Block Cloning (brt.c) to increase reference
+ * counter for the DDT entry if the block is already in DDT.
+ *
+ * Return false if the block, despite having the D bit set, is not present
+ * in the DDT. Currently this is not possible but might be in the future.
+ * See the comment below.
+ */
+boolean_t
+ddt_addref(spa_t *spa, const blkptr_t *bp)
+{
+ ddt_t *ddt;
+ ddt_entry_t *dde;
+ boolean_t result;
+
+ spa_config_enter(spa, SCL_ZIO, FTAG, RW_READER);
+ ddt = ddt_select(spa, bp);
+ ddt_enter(ddt);
+
+ dde = ddt_lookup(ddt, bp, B_TRUE);
+ ASSERT3P(dde, !=, NULL);
+
+ if (dde->dde_type < DDT_TYPES) {
+ ddt_phys_t *ddp;
+
+ ASSERT3S(dde->dde_class, <, DDT_CLASSES);
+
+ ddp = &dde->dde_phys[BP_GET_NDVAS(bp)];
+
+ /*
+ * This entry already existed (dde_type is real), so it must
+ * have refcnt >0 at the start of this txg. We are called from
+ * brt_pending_apply(), before frees are issued, so the refcnt
+ * can't be lowered yet. Therefore, it must be >0. We assert
+ * this because if the order of BRT and DDT interactions were
+ * ever to change and the refcnt was ever zero here, then
+ * likely further action is required to fill out the DDT entry,
+ * and this is a place that is likely to be missed in testing.
+ */
+ ASSERT3U(ddp->ddp_refcnt, >, 0);
+
+ ddt_phys_addref(ddp);
+ result = B_TRUE;
+ } else {
+ /*
+ * At the time of implementating this if the block has the
+ * DEDUP flag set it must exist in the DEDUP table, but
+ * there are many advocates that want ability to remove
+ * entries from DDT with refcnt=1. If this will happen,
+ * we may have a block with the DEDUP set, but which doesn't
+ * have a corresponding entry in the DDT. Be ready.
+ */
+ ASSERT3S(dde->dde_class, ==, DDT_CLASSES);
+ ddt_remove(ddt, dde);
+ result = B_FALSE;
+ }
+
+ ddt_exit(ddt);
+ spa_config_exit(spa, SCL_ZIO, FTAG);
+
+ return (result);
+}
+
ZFS_MODULE_PARAM(zfs_dedup, zfs_dedup_, prefetch, INT, ZMOD_RW,
"Enable prefetching dedup-ed blks");