DDT: Add/use zap_lookup_length_uint64_by_dnode()

Unlike other ZAP consumers due to compression DDT does not know
how big entry it is reading from ZAP.  Due to this it called
zap_length_uint64_by_dnode() and zap_lookup_uint64_by_dnode(),
each of which does full ZAP entry lookup.

Introduction of the combined ZAP method dramatically reduces the
CPU overhead and locks contention at DBUF layer.

Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov>
Signed-off-by: Alexander Motin <alexander.motin@TrueNAS.com>
Closes #18048
This commit is contained in:
Alexander Motin 2025-12-15 17:38:34 -05:00 committed by GitHub
parent ff5414406f
commit 3b1ff816bd
No known key found for this signature in database
GPG key ID: B5690EEEBB952194
5 changed files with 46 additions and 26 deletions

View file

@ -226,6 +226,9 @@ int zap_lookup_uint64(objset_t *os, uint64_t zapobj, const uint64_t *key,
int key_numints, uint64_t integer_size, uint64_t num_integers, void *buf);
int zap_lookup_uint64_by_dnode(dnode_t *dn, const uint64_t *key,
int key_numints, uint64_t integer_size, uint64_t num_integers, void *buf);
int zap_lookup_length_uint64_by_dnode(dnode_t *dn, const uint64_t *key,
int key_numints, uint64_t integer_size, uint64_t num_integers, void *buf,
uint64_t *actual_num_integers);
int zap_contains(objset_t *ds, uint64_t zapobj, const char *name);
int zap_prefetch(objset_t *os, uint64_t zapobj, const char *name);
int zap_prefetch_object(objset_t *os, uint64_t zapobj);

View file

@ -219,7 +219,8 @@ void fzap_byteswap(void *buf, size_t size);
int fzap_count(zap_t *zap, uint64_t *count);
int fzap_lookup(zap_name_t *zn,
uint64_t integer_size, uint64_t num_integers, void *buf,
char *realname, int rn_len, boolean_t *normalization_conflictp);
char *realname, int rn_len, boolean_t *normalization_conflictp,
uint64_t *actual_num_integers);
void fzap_prefetch(zap_name_t *zn);
int fzap_add(zap_name_t *zn, uint64_t integer_size, uint64_t num_integers,
const void *val, const void *tag, dmu_tx_t *tx);

View file

@ -124,25 +124,19 @@ static int
ddt_zap_lookup(dnode_t *dn, const ddt_key_t *ddk, void *phys, size_t psize)
{
uchar_t *cbuf;
uint64_t one, csize;
uint64_t csize;
int error;
error = zap_length_uint64_by_dnode(dn, (uint64_t *)ddk,
DDT_KEY_WORDS, &one, &csize);
if (error)
return (error);
cbuf = kmem_alloc(psize + 1, KM_SLEEP);
ASSERT3U(one, ==, 1);
ASSERT3U(csize, <=, psize + 1);
cbuf = kmem_alloc(csize, KM_SLEEP);
error = zap_lookup_uint64_by_dnode(dn, (uint64_t *)ddk,
DDT_KEY_WORDS, 1, csize, cbuf);
if (error == 0)
error = zap_lookup_length_uint64_by_dnode(dn, (uint64_t *)ddk,
DDT_KEY_WORDS, 1, psize + 1, cbuf, &csize);
if (error == 0) {
ASSERT3U(csize, <=, psize + 1);
ddt_zap_decompress(cbuf, phys, csize, psize);
}
kmem_free(cbuf, csize);
kmem_free(cbuf, psize + 1);
return (error);
}

View file

@ -878,7 +878,8 @@ fzap_check(zap_name_t *zn, uint64_t integer_size, uint64_t num_integers)
int
fzap_lookup(zap_name_t *zn,
uint64_t integer_size, uint64_t num_integers, void *buf,
char *realname, int rn_len, boolean_t *ncp)
char *realname, int rn_len, boolean_t *ncp,
uint64_t *actual_num_integers)
{
zap_leaf_t *l;
zap_entry_handle_t zeh;
@ -898,6 +899,8 @@ fzap_lookup(zap_name_t *zn,
}
err = zap_entry_read(&zeh, integer_size, num_integers, buf);
if (err == 0 && actual_num_integers != NULL)
*actual_num_integers = zeh.zeh_num_integers;
(void) zap_entry_read_name(zn->zn_zap, &zeh, rn_len, realname);
if (ncp) {
*ncp = zap_entry_normalization_conflict(&zeh,

View file

@ -1145,7 +1145,7 @@ zap_lookup_impl(zap_t *zap, const char *name,
if (!zap->zap_ismicro) {
err = fzap_lookup(zn, integer_size, num_integers, buf,
realname, rn_len, ncp);
realname, rn_len, ncp, NULL);
} else {
zfs_btree_index_t idx;
mzap_ent_t *mze = mze_find(zn, &idx);
@ -1300,8 +1300,9 @@ zap_prefetch_uint64_by_dnode(dnode_t *dn, const uint64_t *key, int key_numints)
}
static int
zap_lookup_uint64_impl(zap_t *zap, const uint64_t *key,
int key_numints, uint64_t integer_size, uint64_t num_integers, void *buf)
zap_lookup_length_uint64_impl(zap_t *zap, const uint64_t *key,
int key_numints, uint64_t integer_size, uint64_t num_integers, void *buf,
uint64_t *actual_num_integers)
{
zap_name_t *zn = zap_name_alloc_uint64(zap, key, key_numints);
if (zn == NULL) {
@ -1310,7 +1311,7 @@ zap_lookup_uint64_impl(zap_t *zap, const uint64_t *key,
}
int err = fzap_lookup(zn, integer_size, num_integers, buf,
NULL, 0, NULL);
NULL, 0, NULL, actual_num_integers);
zap_name_free(zn);
zap_unlockdir(zap, FTAG);
return (err);
@ -1326,9 +1327,9 @@ zap_lookup_uint64(objset_t *os, uint64_t zapobj, const uint64_t *key,
zap_lockdir(os, zapobj, NULL, RW_READER, TRUE, FALSE, FTAG, &zap);
if (err != 0)
return (err);
err = zap_lookup_uint64_impl(zap, key, key_numints, integer_size,
num_integers, buf);
/* zap_lookup_uint64_impl() calls zap_unlockdir() */
err = zap_lookup_length_uint64_impl(zap, key, key_numints,
integer_size, num_integers, buf, NULL);
/* zap_lookup_length_uint64_impl() calls zap_unlockdir() */
return (err);
}
@ -1342,9 +1343,26 @@ zap_lookup_uint64_by_dnode(dnode_t *dn, const uint64_t *key,
zap_lockdir_by_dnode(dn, NULL, RW_READER, TRUE, FALSE, FTAG, &zap);
if (err != 0)
return (err);
err = zap_lookup_uint64_impl(zap, key, key_numints, integer_size,
num_integers, buf);
/* zap_lookup_uint64_impl() calls zap_unlockdir() */
err = zap_lookup_length_uint64_impl(zap, key, key_numints,
integer_size, num_integers, buf, NULL);
/* zap_lookup_length_uint64_impl() calls zap_unlockdir() */
return (err);
}
int
zap_lookup_length_uint64_by_dnode(dnode_t *dn, const uint64_t *key,
int key_numints, uint64_t integer_size, uint64_t num_integers, void *buf,
uint64_t *actual_num_integers)
{
zap_t *zap;
int err =
zap_lockdir_by_dnode(dn, NULL, RW_READER, TRUE, FALSE, FTAG, &zap);
if (err != 0)
return (err);
err = zap_lookup_length_uint64_impl(zap, key, key_numints,
integer_size, num_integers, buf, actual_num_integers);
/* zap_lookup_length_uint64_impl() calls zap_unlockdir() */
return (err);
}
@ -2042,6 +2060,7 @@ EXPORT_SYMBOL(zap_lookup);
EXPORT_SYMBOL(zap_lookup_by_dnode);
EXPORT_SYMBOL(zap_lookup_norm);
EXPORT_SYMBOL(zap_lookup_uint64);
EXPORT_SYMBOL(zap_lookup_length_uint64_by_dnode);
EXPORT_SYMBOL(zap_contains);
EXPORT_SYMBOL(zap_prefetch);
EXPORT_SYMBOL(zap_prefetch_uint64);