]> git.proxmox.com Git - mirror_zfs.git/blobdiff - module/zfs/zap_leaf.c
ZAP: Some cleanups/micro-optimizations
[mirror_zfs.git] / module / zfs / zap_leaf.c
index da498b6bc9e3f0ded069f243d4d39ba739d089e0..032aca92695e506d666d80040e624016ca91b8fe 100644 (file)
@@ -6,7 +6,7 @@
  * 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
- * or http://www.opensolaris.org/os/licensing.
+ * or https://opensource.org/licenses/CDDL-1.0.
  * See the License for the specific language governing permissions
  * and limitations under the License.
  *
  *
  * CDDL HEADER END
  */
+
 /*
- * Copyright 2007 Sun Microsystems, Inc.  All rights reserved.
- * Use is subject to license terms.
+ * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved.
+ * Copyright (c) 2013, 2016 by Delphix. All rights reserved.
+ * Copyright 2017 Nexenta Systems, Inc.
  */
 
-#pragma ident  "%Z%%M% %I%     %E% SMI"
-
 /*
  * The 512-byte leaf is broken into 32 16-byte chunks.
  * chunk number n means l_chunk[n], even though the header precedes it.
  * the names are stored null-terminated.
  */
 
+#include <sys/zio.h>
+#include <sys/spa.h>
+#include <sys/dmu.h>
 #include <sys/zfs_context.h>
+#include <sys/fs/zfs.h>
 #include <sys/zap.h>
 #include <sys/zap_impl.h>
 #include <sys/zap_leaf.h>
-#include <sys/spa.h>
-#include <sys/dmu.h>
+#include <sys/arc.h>
 
-static uint16_t *zap_leaf_rehash_entry(zap_leaf_t *l, uint16_t entry);
+static uint16_t *zap_leaf_rehash_entry(zap_leaf_t *l, struct zap_leaf_entry *le,
+    uint16_t entry);
 
 #define        CHAIN_END 0xffff /* end of the chunk chain */
 
-/* half the (current) minimum block size */
-#define        MAX_ARRAY_BYTES (8<<10)
-
 #define        LEAF_HASH(l, h) \
        ((ZAP_LEAF_HASH_NUMENTRIES(l)-1) & \
-       ((h) >> (64 - ZAP_LEAF_HASH_SHIFT(l)-(l)->l_phys->l_hdr.lh_prefix_len)))
-
-#define        LEAF_HASH_ENTPTR(l, h) (&(l)->l_phys->l_hash[LEAF_HASH(l, h)])
+       ((h) >> \
+       (64 - ZAP_LEAF_HASH_SHIFT(l) - zap_leaf_phys(l)->l_hdr.lh_prefix_len)))
 
-
-static void
-zap_memset(void *a, int c, size_t n)
-{
-       char *cp = a;
-       char *cpend = cp + n;
-
-       while (cp < cpend)
-               *cp++ = c;
-}
+#define        LEAF_HASH_ENTPTR(l, h)  (&zap_leaf_phys(l)->l_hash[LEAF_HASH(l, h)])
 
 static void
 stv(int len, void *addr, uint64_t value)
@@ -78,8 +69,9 @@ stv(int len, void *addr, uint64_t value)
        case 8:
                *(uint64_t *)addr = value;
                return;
+       default:
+               PANIC("bad int len %d", len);
        }
-       ASSERT(!"bad int len");
 }
 
 static uint64_t
@@ -94,31 +86,34 @@ ldv(int len, const void *addr)
                return (*(uint32_t *)addr);
        case 8:
                return (*(uint64_t *)addr);
+       default:
+               PANIC("bad int len %d", len);
        }
-       ASSERT(!"bad int len");
        return (0xFEEDFACEDEADBEEFULL);
 }
 
 void
-zap_leaf_byteswap(zap_leaf_phys_t *buf, int size)
+zap_leaf_byteswap(zap_leaf_phys_t *buf, size_t size)
 {
-       int i;
        zap_leaf_t l;
-       l.l_bs = highbit(size)-1;
-       l.l_phys = buf;
-
-       buf->l_hdr.lh_block_type =      BSWAP_64(buf->l_hdr.lh_block_type);
-       buf->l_hdr.lh_prefix =          BSWAP_64(buf->l_hdr.lh_prefix);
-       buf->l_hdr.lh_magic =           BSWAP_32(buf->l_hdr.lh_magic);
-       buf->l_hdr.lh_nfree =           BSWAP_16(buf->l_hdr.lh_nfree);
-       buf->l_hdr.lh_nentries =        BSWAP_16(buf->l_hdr.lh_nentries);
-       buf->l_hdr.lh_prefix_len =      BSWAP_16(buf->l_hdr.lh_prefix_len);
-       buf->l_hdr.lh_freelist =        BSWAP_16(buf->l_hdr.lh_freelist);
-
-       for (i = 0; i < ZAP_LEAF_HASH_NUMENTRIES(&l); i++)
+       dmu_buf_t l_dbuf;
+
+       l_dbuf.db_data = buf;
+       l.l_bs = highbit64(size) - 1;
+       l.l_dbuf = &l_dbuf;
+
+       buf->l_hdr.lh_block_type =      BSWAP_64(buf->l_hdr.lh_block_type);
+       buf->l_hdr.lh_prefix =          BSWAP_64(buf->l_hdr.lh_prefix);
+       buf->l_hdr.lh_magic =           BSWAP_32(buf->l_hdr.lh_magic);
+       buf->l_hdr.lh_nfree =           BSWAP_16(buf->l_hdr.lh_nfree);
+       buf->l_hdr.lh_nentries =        BSWAP_16(buf->l_hdr.lh_nentries);
+       buf->l_hdr.lh_prefix_len =      BSWAP_16(buf->l_hdr.lh_prefix_len);
+       buf->l_hdr.lh_freelist =        BSWAP_16(buf->l_hdr.lh_freelist);
+
+       for (uint_t i = 0; i < ZAP_LEAF_HASH_NUMENTRIES(&l); i++)
                buf->l_hash[i] = BSWAP_16(buf->l_hash[i]);
 
-       for (i = 0; i < ZAP_LEAF_NUMCHUNKS(&l); i++) {
+       for (uint_t i = 0; i < ZAP_LEAF_NUMCHUNKS(&l); i++) {
                zap_leaf_chunk_t *lc = &ZAP_LEAF_CHUNK(&l, i);
                struct zap_leaf_entry *le;
 
@@ -127,12 +122,12 @@ zap_leaf_byteswap(zap_leaf_phys_t *buf, int size)
                        le = &lc->l_entry;
 
                        le->le_type =           BSWAP_8(le->le_type);
-                       le->le_int_size =       BSWAP_8(le->le_int_size);
+                       le->le_value_intlen =   BSWAP_8(le->le_value_intlen);
                        le->le_next =           BSWAP_16(le->le_next);
                        le->le_name_chunk =     BSWAP_16(le->le_name_chunk);
-                       le->le_name_length =    BSWAP_16(le->le_name_length);
+                       le->le_name_numints =   BSWAP_16(le->le_name_numints);
                        le->le_value_chunk =    BSWAP_16(le->le_value_chunk);
-                       le->le_value_length =   BSWAP_16(le->le_value_length);
+                       le->le_value_numints =  BSWAP_16(le->le_value_numints);
                        le->le_cd =             BSWAP_32(le->le_cd);
                        le->le_hash =           BSWAP_64(le->le_hash);
                        break;
@@ -146,7 +141,8 @@ zap_leaf_byteswap(zap_leaf_phys_t *buf, int size)
                        /* la_array doesn't need swapping */
                        break;
                default:
-                       ASSERT(!"bad leaf type");
+                       cmn_err(CE_PANIC, "bad leaf type %d",
+                           lc->l_free.lf_type);
                }
        }
 }
@@ -154,21 +150,21 @@ zap_leaf_byteswap(zap_leaf_phys_t *buf, int size)
 void
 zap_leaf_init(zap_leaf_t *l, boolean_t sort)
 {
-       int i;
-
-       l->l_bs = highbit(l->l_dbuf->db_size)-1;
-       zap_memset(&l->l_phys->l_hdr, 0, sizeof (struct zap_leaf_header));
-       zap_memset(l->l_phys->l_hash, CHAIN_END, 2*ZAP_LEAF_HASH_NUMENTRIES(l));
-       for (i = 0; i < ZAP_LEAF_NUMCHUNKS(l); i++) {
+       l->l_bs = highbit64(l->l_dbuf->db_size) - 1;
+       memset(&zap_leaf_phys(l)->l_hdr, 0,
+           sizeof (struct zap_leaf_header));
+       memset(zap_leaf_phys(l)->l_hash, CHAIN_END,
+           2*ZAP_LEAF_HASH_NUMENTRIES(l));
+       for (uint_t i = 0; i < ZAP_LEAF_NUMCHUNKS(l); i++) {
                ZAP_LEAF_CHUNK(l, i).l_free.lf_type = ZAP_CHUNK_FREE;
                ZAP_LEAF_CHUNK(l, i).l_free.lf_next = i+1;
        }
        ZAP_LEAF_CHUNK(l, ZAP_LEAF_NUMCHUNKS(l)-1).l_free.lf_next = CHAIN_END;
-       l->l_phys->l_hdr.lh_block_type = ZBT_LEAF;
-       l->l_phys->l_hdr.lh_magic = ZAP_LEAF_MAGIC;
-       l->l_phys->l_hdr.lh_nfree = ZAP_LEAF_NUMCHUNKS(l);
+       zap_leaf_phys(l)->l_hdr.lh_block_type = ZBT_LEAF;
+       zap_leaf_phys(l)->l_hdr.lh_magic = ZAP_LEAF_MAGIC;
+       zap_leaf_phys(l)->l_hdr.lh_nfree = ZAP_LEAF_NUMCHUNKS(l);
        if (sort)
-               l->l_phys->l_hdr.lh_flags |= ZLF_ENTRIES_CDSORTED;
+               zap_leaf_phys(l)->l_hdr.lh_flags |= ZLF_ENTRIES_CDSORTED;
 }
 
 /*
@@ -178,17 +174,16 @@ zap_leaf_init(zap_leaf_t *l, boolean_t sort)
 static uint16_t
 zap_leaf_chunk_alloc(zap_leaf_t *l)
 {
-       int chunk;
-
-       ASSERT(l->l_phys->l_hdr.lh_nfree > 0);
+       ASSERT(zap_leaf_phys(l)->l_hdr.lh_nfree > 0);
 
-       chunk = l->l_phys->l_hdr.lh_freelist;
+       uint_t chunk = zap_leaf_phys(l)->l_hdr.lh_freelist;
        ASSERT3U(chunk, <, ZAP_LEAF_NUMCHUNKS(l));
        ASSERT3U(ZAP_LEAF_CHUNK(l, chunk).l_free.lf_type, ==, ZAP_CHUNK_FREE);
 
-       l->l_phys->l_hdr.lh_freelist = ZAP_LEAF_CHUNK(l, chunk).l_free.lf_next;
+       zap_leaf_phys(l)->l_hdr.lh_freelist =
+           ZAP_LEAF_CHUNK(l, chunk).l_free.lf_next;
 
-       l->l_phys->l_hdr.lh_nfree--;
+       zap_leaf_phys(l)->l_hdr.lh_nfree--;
 
        return (chunk);
 }
@@ -197,16 +192,16 @@ static void
 zap_leaf_chunk_free(zap_leaf_t *l, uint16_t chunk)
 {
        struct zap_leaf_free *zlf = &ZAP_LEAF_CHUNK(l, chunk).l_free;
-       ASSERT3U(l->l_phys->l_hdr.lh_nfree, <, ZAP_LEAF_NUMCHUNKS(l));
+       ASSERT3U(zap_leaf_phys(l)->l_hdr.lh_nfree, <, ZAP_LEAF_NUMCHUNKS(l));
        ASSERT3U(chunk, <, ZAP_LEAF_NUMCHUNKS(l));
        ASSERT(zlf->lf_type != ZAP_CHUNK_FREE);
 
        zlf->lf_type = ZAP_CHUNK_FREE;
-       zlf->lf_next = l->l_phys->l_hdr.lh_freelist;
-       bzero(zlf->lf_pad, sizeof (zlf->lf_pad)); /* help it to compress */
-       l->l_phys->l_hdr.lh_freelist = chunk;
+       zlf->lf_next = zap_leaf_phys(l)->l_hdr.lh_freelist;
+       memset(zlf->lf_pad, 0, sizeof (zlf->lf_pad)); /* help it to compress */
+       zap_leaf_phys(l)->l_hdr.lh_freelist = chunk;
 
-       l->l_phys->l_hdr.lh_nfree++;
+       zap_leaf_phys(l)->l_hdr.lh_nfree++;
 }
 
 /*
@@ -215,33 +210,33 @@ zap_leaf_chunk_free(zap_leaf_t *l, uint16_t chunk)
 
 static uint16_t
 zap_leaf_array_create(zap_leaf_t *l, const char *buf,
-       int integer_size, int num_integers)
+    int integer_size, int num_integers)
 {
        uint16_t chunk_head;
        uint16_t *chunkp = &chunk_head;
-       int byten = 0;
-       uint64_t value;
-       int shift = (integer_size-1)*8;
+       int byten = integer_size;
+       uint64_t value = 0;
+       int shift = (integer_size - 1) * 8;
        int len = num_integers;
 
-       ASSERT3U(num_integers * integer_size, <, MAX_ARRAY_BYTES);
+       ASSERT3U(num_integers * integer_size, <=, ZAP_MAXVALUELEN);
 
+       if (len > 0)
+               value = ldv(integer_size, buf);
        while (len > 0) {
                uint16_t chunk = zap_leaf_chunk_alloc(l);
                struct zap_leaf_array *la = &ZAP_LEAF_CHUNK(l, chunk).l_array;
-               int i;
 
                la->la_type = ZAP_CHUNK_ARRAY;
-               for (i = 0; i < ZAP_LEAF_ARRAY_BYTES; i++) {
-                       if (byten == 0)
-                               value = ldv(integer_size, buf);
+               for (int i = 0; i < ZAP_LEAF_ARRAY_BYTES; i++) {
                        la->la_array[i] = value >> shift;
                        value <<= 8;
-                       if (++byten == integer_size) {
-                               byten = 0;
-                               buf += integer_size;
+                       if (--byten == 0) {
                                if (--len == 0)
                                        break;
+                               byten = integer_size;
+                               buf += integer_size;
+                               value = ldv(integer_size, buf);
                        }
                }
 
@@ -261,7 +256,7 @@ zap_leaf_array_free(zap_leaf_t *l, uint16_t *chunkp)
        *chunkp = CHAIN_END;
 
        while (chunk != CHAIN_END) {
-               int nextchunk = ZAP_LEAF_CHUNK(l, chunk).l_array.la_next;
+               uint_t nextchunk = ZAP_LEAF_CHUNK(l, chunk).l_array.la_next;
                ASSERT3U(ZAP_LEAF_CHUNK(l, chunk).l_array.la_type, ==,
                    ZAP_CHUNK_ARRAY);
                zap_leaf_chunk_free(l, chunk);
@@ -273,11 +268,12 @@ zap_leaf_array_free(zap_leaf_t *l, uint16_t *chunkp)
 static void
 zap_leaf_array_read(zap_leaf_t *l, uint16_t chunk,
     int array_int_len, int array_len, int buf_int_len, uint64_t buf_len,
-    char *buf)
+    void *buf)
 {
        int len = MIN(array_len, buf_len);
        int byten = 0;
        uint64_t value = 0;
+       char *p = buf;
 
        ASSERT3U(array_int_len, <=, buf_int_len);
 
@@ -285,7 +281,7 @@ zap_leaf_array_read(zap_leaf_t *l, uint16_t chunk,
        if (array_int_len == 8 && buf_int_len == 8 && len == 1) {
                struct zap_leaf_array *la = &ZAP_LEAF_CHUNK(l, chunk).l_array;
                uint8_t *ip = la->la_array;
-               uint64_t *buf64 = (uint64_t *)buf;
+               uint64_t *buf64 = buf;
 
                *buf64 = (uint64_t)ip[0] << 56 | (uint64_t)ip[1] << 48 |
                    (uint64_t)ip[2] << 40 | (uint64_t)ip[3] << 32 |
@@ -300,8 +296,8 @@ zap_leaf_array_read(zap_leaf_t *l, uint16_t chunk,
                while (chunk != CHAIN_END) {
                        struct zap_leaf_array *la =
                            &ZAP_LEAF_CHUNK(l, chunk).l_array;
-                       bcopy(la->la_array, buf, ZAP_LEAF_ARRAY_BYTES);
-                       buf += ZAP_LEAF_ARRAY_BYTES;
+                       memcpy(p, la->la_array, ZAP_LEAF_ARRAY_BYTES);
+                       p += ZAP_LEAF_ARRAY_BYTES;
                        chunk = la->la_next;
                }
                return;
@@ -309,57 +305,72 @@ zap_leaf_array_read(zap_leaf_t *l, uint16_t chunk,
 
        while (len > 0) {
                struct zap_leaf_array *la = &ZAP_LEAF_CHUNK(l, chunk).l_array;
-               int i;
 
                ASSERT3U(chunk, <, ZAP_LEAF_NUMCHUNKS(l));
-               for (i = 0; i < ZAP_LEAF_ARRAY_BYTES && len > 0; i++) {
+               for (int i = 0; i < ZAP_LEAF_ARRAY_BYTES; i++) {
                        value = (value << 8) | la->la_array[i];
                        byten++;
                        if (byten == array_int_len) {
-                               stv(buf_int_len, buf, value);
+                               stv(buf_int_len, p, value);
                                byten = 0;
                                len--;
                                if (len == 0)
                                        return;
-                               buf += buf_int_len;
+                               p += buf_int_len;
                        }
                }
                chunk = la->la_next;
        }
 }
 
-/*
- * Only to be used on 8-bit arrays.
- * array_len is actual len in bytes (not encoded le_value_length).
- * namenorm is null-terminated.
- */
 static boolean_t
-zap_leaf_array_match(zap_leaf_t *l, zap_name_t *zn, int chunk, int array_len)
+zap_leaf_array_match(zap_leaf_t *l, zap_name_t *zn,
+    uint_t chunk, int array_numints)
 {
        int bseen = 0;
 
-       if (zn->zn_matchtype == MT_FIRST) {
-               char *thisname = kmem_alloc(array_len, KM_SLEEP);
-               boolean_t match;
+       if (zap_getflags(zn->zn_zap) & ZAP_FLAG_UINT64_KEY) {
+               uint64_t *thiskey =
+                   kmem_alloc(array_numints * sizeof (*thiskey), KM_SLEEP);
+               ASSERT(zn->zn_key_intlen == sizeof (*thiskey));
 
-               zap_leaf_array_read(l, chunk, 1, array_len, 1,
-                   array_len, thisname);
-               match = zap_match(zn, thisname);
-               kmem_free(thisname, array_len);
+               zap_leaf_array_read(l, chunk, sizeof (*thiskey), array_numints,
+                   sizeof (*thiskey), array_numints, thiskey);
+               boolean_t match = memcmp(thiskey, zn->zn_key_orig,
+                   array_numints * sizeof (*thiskey)) == 0;
+               kmem_free(thiskey, array_numints * sizeof (*thiskey));
                return (match);
        }
 
-       /* Fast path for exact matching */
-       while (bseen < array_len) {
+       ASSERT(zn->zn_key_intlen == 1);
+       if (zn->zn_matchtype & MT_NORMALIZE) {
+               char *thisname = kmem_alloc(array_numints, KM_SLEEP);
+
+               zap_leaf_array_read(l, chunk, sizeof (char), array_numints,
+                   sizeof (char), array_numints, thisname);
+               boolean_t match = zap_match(zn, thisname);
+               kmem_free(thisname, array_numints);
+               return (match);
+       }
+
+       /*
+        * Fast path for exact matching.
+        * First check that the lengths match, so that we don't read
+        * past the end of the zn_key_orig array.
+        */
+       if (array_numints != zn->zn_key_orig_numints)
+               return (B_FALSE);
+       while (bseen < array_numints) {
                struct zap_leaf_array *la = &ZAP_LEAF_CHUNK(l, chunk).l_array;
-               int toread = MIN(array_len - bseen, ZAP_LEAF_ARRAY_BYTES);
+               int toread = MIN(array_numints - bseen, ZAP_LEAF_ARRAY_BYTES);
                ASSERT3U(chunk, <, ZAP_LEAF_NUMCHUNKS(l));
-               if (bcmp(la->la_array, zn->zn_name_orij + bseen, toread))
+               if (memcmp(la->la_array, (char *)zn->zn_key_orig + bseen,
+                   toread))
                        break;
                chunk = la->la_next;
                bseen += toread;
        }
-       return (bseen == array_len);
+       return (bseen == array_numints);
 }
 
 /*
@@ -369,13 +380,11 @@ zap_leaf_array_match(zap_leaf_t *l, zap_name_t *zn, int chunk, int array_len)
 int
 zap_leaf_lookup(zap_leaf_t *l, zap_name_t *zn, zap_entry_handle_t *zeh)
 {
-       uint16_t *chunkp;
        struct zap_leaf_entry *le;
 
-       ASSERT3U(l->l_phys->l_hdr.lh_magic, ==, ZAP_LEAF_MAGIC);
+       ASSERT3U(zap_leaf_phys(l)->l_hdr.lh_magic, ==, ZAP_LEAF_MAGIC);
 
-again:
-       for (chunkp = LEAF_HASH_ENTPTR(l, zn->zn_hash);
+       for (uint16_t *chunkp = LEAF_HASH_ENTPTR(l, zn->zn_hash);
            *chunkp != CHAIN_END; chunkp = &le->le_next) {
                uint16_t chunk = *chunkp;
                le = ZAP_LEAF_ENTRY(l, chunk);
@@ -389,14 +398,14 @@ again:
                /*
                 * NB: the entry chain is always sorted by cd on
                 * normalized zap objects, so this will find the
-                * lowest-cd match for MT_FIRST.
+                * lowest-cd match for MT_NORMALIZE.
                 */
-               ASSERT(zn->zn_matchtype == MT_EXACT ||
-                   (l->l_phys->l_hdr.lh_flags & ZLF_ENTRIES_CDSORTED));
+               ASSERT((zn->zn_matchtype == 0) ||
+                   (zap_leaf_phys(l)->l_hdr.lh_flags & ZLF_ENTRIES_CDSORTED));
                if (zap_leaf_array_match(l, zn, le->le_name_chunk,
-                   le->le_name_length)) {
-                       zeh->zeh_num_integers = le->le_value_length;
-                       zeh->zeh_integer_size = le->le_int_size;
+                   le->le_name_numints)) {
+                       zeh->zeh_num_integers = le->le_value_numints;
+                       zeh->zeh_integer_size = le->le_value_intlen;
                        zeh->zeh_cd = le->le_cd;
                        zeh->zeh_hash = le->le_hash;
                        zeh->zeh_chunkp = chunkp;
@@ -405,16 +414,7 @@ again:
                }
        }
 
-       /*
-        * NB: we could of course do this in one pass, but that would be
-        * a pain.  We'll see if MT_BEST is even used much.
-        */
-       if (zn->zn_matchtype == MT_BEST) {
-               zn->zn_matchtype = MT_FIRST;
-               goto again;
-       }
-
-       return (ENOENT);
+       return (SET_ERROR(ENOENT));
 }
 
 /* Return (h1,cd1 >= h2,cd2) */
@@ -425,17 +425,15 @@ int
 zap_leaf_lookup_closest(zap_leaf_t *l,
     uint64_t h, uint32_t cd, zap_entry_handle_t *zeh)
 {
-       uint16_t chunk;
        uint64_t besth = -1ULL;
-       uint32_t bestcd = ZAP_MAXCD;
+       uint32_t bestcd = -1U;
        uint16_t bestlh = ZAP_LEAF_HASH_NUMENTRIES(l)-1;
-       uint16_t lh;
        struct zap_leaf_entry *le;
 
-       ASSERT3U(l->l_phys->l_hdr.lh_magic, ==, ZAP_LEAF_MAGIC);
+       ASSERT3U(zap_leaf_phys(l)->l_hdr.lh_magic, ==, ZAP_LEAF_MAGIC);
 
-       for (lh = LEAF_HASH(l, h); lh <= bestlh; lh++) {
-               for (chunk = l->l_phys->l_hash[lh];
+       for (uint16_t lh = LEAF_HASH(l, h); lh <= bestlh; lh++) {
+               for (uint16_t chunk = zap_leaf_phys(l)->l_hash[lh];
                    chunk != CHAIN_END; chunk = le->le_next) {
                        le = ZAP_LEAF_ENTRY(l, chunk);
 
@@ -449,8 +447,8 @@ zap_leaf_lookup_closest(zap_leaf_t *l,
                                besth = le->le_hash;
                                bestcd = le->le_cd;
 
-                               zeh->zeh_num_integers = le->le_value_length;
-                               zeh->zeh_integer_size = le->le_int_size;
+                               zeh->zeh_num_integers = le->le_value_numints;
+                               zeh->zeh_integer_size = le->le_value_intlen;
                                zeh->zeh_cd = le->le_cd;
                                zeh->zeh_hash = le->le_hash;
                                zeh->zeh_fakechunk = chunk;
@@ -460,7 +458,7 @@ zap_leaf_lookup_closest(zap_leaf_t *l,
                }
        }
 
-       return (bestcd == ZAP_MAXCD ? ENOENT : 0);
+       return (bestcd == -1U ? SET_ERROR(ENOENT) : 0);
 }
 
 int
@@ -471,73 +469,69 @@ zap_entry_read(const zap_entry_handle_t *zeh,
            ZAP_LEAF_ENTRY(zeh->zeh_leaf, *zeh->zeh_chunkp);
        ASSERT3U(le->le_type, ==, ZAP_CHUNK_ENTRY);
 
-       if (le->le_int_size > integer_size)
-               return (EINVAL);
+       if (le->le_value_intlen > integer_size)
+               return (SET_ERROR(EINVAL));
 
-       zap_leaf_array_read(zeh->zeh_leaf, le->le_value_chunk, le->le_int_size,
-           le->le_value_length, integer_size, num_integers, buf);
+       zap_leaf_array_read(zeh->zeh_leaf, le->le_value_chunk,
+           le->le_value_intlen, le->le_value_numints,
+           integer_size, num_integers, buf);
 
        if (zeh->zeh_num_integers > num_integers)
-               return (EOVERFLOW);
+               return (SET_ERROR(EOVERFLOW));
        return (0);
 
 }
 
 int
-zap_entry_read_name(const zap_entry_handle_t *zeh, uint16_t buflen, char *buf)
+zap_entry_read_name(zap_t *zap, const zap_entry_handle_t *zeh, uint16_t buflen,
+    char *buf)
 {
        struct zap_leaf_entry *le =
            ZAP_LEAF_ENTRY(zeh->zeh_leaf, *zeh->zeh_chunkp);
        ASSERT3U(le->le_type, ==, ZAP_CHUNK_ENTRY);
 
-       zap_leaf_array_read(zeh->zeh_leaf, le->le_name_chunk, 1,
-           le->le_name_length, 1, buflen, buf);
-       if (le->le_name_length > buflen)
-               return (EOVERFLOW);
+       if (zap_getflags(zap) & ZAP_FLAG_UINT64_KEY) {
+               zap_leaf_array_read(zeh->zeh_leaf, le->le_name_chunk, 8,
+                   le->le_name_numints, 8, buflen / 8, buf);
+       } else {
+               zap_leaf_array_read(zeh->zeh_leaf, le->le_name_chunk, 1,
+                   le->le_name_numints, 1, buflen, buf);
+       }
+       if (le->le_name_numints > buflen)
+               return (SET_ERROR(EOVERFLOW));
        return (0);
 }
 
 int
 zap_entry_update(zap_entry_handle_t *zeh,
-       uint8_t integer_size, uint64_t num_integers, const void *buf)
+    uint8_t integer_size, uint64_t num_integers, const void *buf)
 {
-       int delta_chunks;
        zap_leaf_t *l = zeh->zeh_leaf;
        struct zap_leaf_entry *le = ZAP_LEAF_ENTRY(l, *zeh->zeh_chunkp);
 
-       delta_chunks = ZAP_LEAF_ARRAY_NCHUNKS(num_integers * integer_size) -
-           ZAP_LEAF_ARRAY_NCHUNKS(le->le_value_length * le->le_int_size);
+       int delta_chunks = ZAP_LEAF_ARRAY_NCHUNKS(num_integers * integer_size) -
+           ZAP_LEAF_ARRAY_NCHUNKS(le->le_value_numints * le->le_value_intlen);
 
-       if ((int)l->l_phys->l_hdr.lh_nfree < delta_chunks)
-               return (EAGAIN);
-
-       /*
-        * We should search other chained leaves (via
-        * zap_entry_remove,create?) otherwise returning EAGAIN will
-        * just send us into an infinite loop if we have to chain
-        * another leaf block, rather than being able to split this
-        * block.
-        */
+       if ((int)zap_leaf_phys(l)->l_hdr.lh_nfree < delta_chunks)
+               return (SET_ERROR(EAGAIN));
 
        zap_leaf_array_free(l, &le->le_value_chunk);
        le->le_value_chunk =
            zap_leaf_array_create(l, buf, integer_size, num_integers);
-       le->le_value_length = num_integers;
-       le->le_int_size = integer_size;
+       le->le_value_numints = num_integers;
+       le->le_value_intlen = integer_size;
        return (0);
 }
 
 void
 zap_entry_remove(zap_entry_handle_t *zeh)
 {
-       uint16_t entry_chunk;
-       struct zap_leaf_entry *le;
        zap_leaf_t *l = zeh->zeh_leaf;
 
        ASSERT3P(zeh->zeh_chunkp, !=, &zeh->zeh_fakechunk);
 
-       entry_chunk = *zeh->zeh_chunkp;
-       le = ZAP_LEAF_ENTRY(l, entry_chunk);
+       uint16_t entry_chunk = *zeh->zeh_chunkp;
+       struct zap_leaf_entry *le = ZAP_LEAF_ENTRY(l, entry_chunk);
        ASSERT3U(le->le_type, ==, ZAP_CHUNK_ENTRY);
 
        zap_leaf_array_free(l, &le->le_name_chunk);
@@ -546,32 +540,28 @@ zap_entry_remove(zap_entry_handle_t *zeh)
        *zeh->zeh_chunkp = le->le_next;
        zap_leaf_chunk_free(l, entry_chunk);
 
-       l->l_phys->l_hdr.lh_nentries--;
+       zap_leaf_phys(l)->l_hdr.lh_nentries--;
 }
 
 int
-zap_entry_create(zap_leaf_t *l, const char *name, uint64_t h, uint32_t cd,
+zap_entry_create(zap_leaf_t *l, zap_name_t *zn, uint32_t cd,
     uint8_t integer_size, uint64_t num_integers, const void *buf,
     zap_entry_handle_t *zeh)
 {
        uint16_t chunk;
-       uint16_t *chunkp;
        struct zap_leaf_entry *le;
-       uint64_t namelen, valuelen;
-       int numchunks;
+       uint64_t h = zn->zn_hash;
 
-       valuelen = integer_size * num_integers;
-       namelen = strlen(name) + 1;
-       ASSERT(namelen >= 2);
+       uint64_t valuelen = integer_size * num_integers;
 
-       numchunks = 1 + ZAP_LEAF_ARRAY_NCHUNKS(namelen) +
-           ZAP_LEAF_ARRAY_NCHUNKS(valuelen);
+       uint_t numchunks = 1 + ZAP_LEAF_ARRAY_NCHUNKS(zn->zn_key_orig_numints *
+           zn->zn_key_intlen) + ZAP_LEAF_ARRAY_NCHUNKS(valuelen);
        if (numchunks > ZAP_LEAF_NUMCHUNKS(l))
-               return (E2BIG);
+               return (SET_ERROR(E2BIG));
 
-       if (cd == ZAP_MAXCD) {
+       if (cd == ZAP_NEED_CD) {
                /* find the lowest unused cd */
-               if (l->l_phys->l_hdr.lh_flags & ZLF_ENTRIES_CDSORTED) {
+               if (zap_leaf_phys(l)->l_hdr.lh_flags & ZLF_ENTRIES_CDSORTED) {
                        cd = 0;
 
                        for (chunk = *LEAF_HASH_ENTPTR(l, h);
@@ -586,7 +576,7 @@ zap_entry_create(zap_leaf_t *l, const char *name, uint64_t h, uint32_t cd,
                        }
                } else {
                        /* old unsorted format; do it the O(n^2) way */
-                       for (cd = 0; cd < ZAP_MAXCD; cd++) {
+                       for (cd = 0; ; cd++) {
                                for (chunk = *LEAF_HASH_ENTPTR(l, h);
                                    chunk != CHAIN_END; chunk = le->le_next) {
                                        le = ZAP_LEAF_ENTRY(l, chunk);
@@ -601,37 +591,38 @@ zap_entry_create(zap_leaf_t *l, const char *name, uint64_t h, uint32_t cd,
                        }
                }
                /*
-                * we would run out of space in a block before we could
-                * have ZAP_MAXCD entries
+                * We would run out of space in a block before we could
+                * store enough entries to run out of CD values.
                 */
-               ASSERT3U(cd, <, ZAP_MAXCD);
+               ASSERT3U(cd, <, zap_maxcd(zn->zn_zap));
        }
 
-       if (l->l_phys->l_hdr.lh_nfree < numchunks)
-               return (EAGAIN);
+       if (zap_leaf_phys(l)->l_hdr.lh_nfree < numchunks)
+               return (SET_ERROR(EAGAIN));
 
        /* make the entry */
        chunk = zap_leaf_chunk_alloc(l);
        le = ZAP_LEAF_ENTRY(l, chunk);
        le->le_type = ZAP_CHUNK_ENTRY;
-       le->le_name_chunk = zap_leaf_array_create(l, name, 1, namelen);
-       le->le_name_length = namelen;
+       le->le_name_chunk = zap_leaf_array_create(l, zn->zn_key_orig,
+           zn->zn_key_intlen, zn->zn_key_orig_numints);
+       le->le_name_numints = zn->zn_key_orig_numints;
        le->le_value_chunk =
            zap_leaf_array_create(l, buf, integer_size, num_integers);
-       le->le_value_length = num_integers;
-       le->le_int_size = integer_size;
+       le->le_value_numints = num_integers;
+       le->le_value_intlen = integer_size;
        le->le_hash = h;
        le->le_cd = cd;
 
        /* link it into the hash chain */
        /* XXX if we did the search above, we could just use that */
-       chunkp = zap_leaf_rehash_entry(l, chunk);
+       uint16_t *chunkp = zap_leaf_rehash_entry(l, le, chunk);
 
-       l->l_phys->l_hdr.lh_nentries++;
+       zap_leaf_phys(l)->l_hdr.lh_nentries++;
 
        zeh->zeh_leaf = l;
        zeh->zeh_num_integers = num_integers;
-       zeh->zeh_integer_size = le->le_int_size;
+       zeh->zeh_integer_size = le->le_value_intlen;
        zeh->zeh_cd = le->le_cd;
        zeh->zeh_hash = le->le_hash;
        zeh->zeh_chunkp = chunkp;
@@ -647,20 +638,19 @@ zap_entry_create(zap_leaf_t *l, const char *name, uint64_t h, uint32_t cd,
  * form of the name.  But all callers have one of these on hand anyway,
  * so might as well take advantage.  A cleaner but slower interface
  * would accept neither argument, and compute the normalized name as
- * needed (using zap_name_alloc(zap_entry_read_name(zeh))).
+ * needed (using zap_name_alloc_str(zap_entry_read_name(zeh))).
  */
 boolean_t
 zap_entry_normalization_conflict(zap_entry_handle_t *zeh, zap_name_t *zn,
     const char *name, zap_t *zap)
 {
-       uint64_t chunk;
        struct zap_leaf_entry *le;
        boolean_t allocdzn = B_FALSE;
 
        if (zap->zap_normflags == 0)
                return (B_FALSE);
 
-       for (chunk = *LEAF_HASH_ENTPTR(zeh->zeh_leaf, zeh->zeh_hash);
+       for (uint16_t chunk = *LEAF_HASH_ENTPTR(zeh->zeh_leaf, zeh->zeh_hash);
            chunk != CHAIN_END; chunk = le->le_next) {
                le = ZAP_LEAF_ENTRY(zeh->zeh_leaf, chunk);
                if (le->le_hash != zeh->zeh_hash)
@@ -669,11 +659,11 @@ zap_entry_normalization_conflict(zap_entry_handle_t *zeh, zap_name_t *zn,
                        continue;
 
                if (zn == NULL) {
-                       zn = zap_name_alloc(zap, name, MT_FIRST);
+                       zn = zap_name_alloc_str(zap, name, MT_NORMALIZE);
                        allocdzn = B_TRUE;
                }
                if (zap_leaf_array_match(zeh->zeh_leaf, zn,
-                   le->le_name_chunk, le->le_name_length)) {
+                   le->le_name_chunk, le->le_name_numints)) {
                        if (allocdzn)
                                zap_name_free(zn);
                        return (B_TRUE);
@@ -689,9 +679,8 @@ zap_entry_normalization_conflict(zap_entry_handle_t *zeh, zap_name_t *zn,
  */
 
 static uint16_t *
-zap_leaf_rehash_entry(zap_leaf_t *l, uint16_t entry)
+zap_leaf_rehash_entry(zap_leaf_t *l, struct zap_leaf_entry *le, uint16_t entry)
 {
-       struct zap_leaf_entry *le = ZAP_LEAF_ENTRY(l, entry);
        struct zap_leaf_entry *le2;
        uint16_t *chunkp;
 
@@ -724,7 +713,7 @@ zap_leaf_transfer_array(zap_leaf_t *l, uint16_t chunk, zap_leaf_t *nl)
                    &ZAP_LEAF_CHUNK(nl, nchunk).l_array;
                struct zap_leaf_array *la =
                    &ZAP_LEAF_CHUNK(l, chunk).l_array;
-               int nextchunk = la->la_next;
+               uint_t nextchunk = la->la_next;
 
                ASSERT3U(chunk, <, ZAP_LEAF_NUMCHUNKS(l));
                ASSERT3U(nchunk, <, ZAP_LEAF_NUMCHUNKS(l));
@@ -741,19 +730,16 @@ zap_leaf_transfer_array(zap_leaf_t *l, uint16_t chunk, zap_leaf_t *nl)
 }
 
 static void
-zap_leaf_transfer_entry(zap_leaf_t *l, int entry, zap_leaf_t *nl)
+zap_leaf_transfer_entry(zap_leaf_t *l, uint_t entry, zap_leaf_t *nl)
 {
-       struct zap_leaf_entry *le, *nle;
-       uint16_t chunk;
-
-       le = ZAP_LEAF_ENTRY(l, entry);
+       struct zap_leaf_entry *le = ZAP_LEAF_ENTRY(l, entry);
        ASSERT3U(le->le_type, ==, ZAP_CHUNK_ENTRY);
 
-       chunk = zap_leaf_chunk_alloc(nl);
-       nle = ZAP_LEAF_ENTRY(nl, chunk);
+       uint16_t chunk = zap_leaf_chunk_alloc(nl);
+       struct zap_leaf_entry *nle = ZAP_LEAF_ENTRY(nl, chunk);
        *nle = *le; /* structure assignment */
 
-       (void) zap_leaf_rehash_entry(nl, chunk);
+       (void) zap_leaf_rehash_entry(nl, nle, chunk);
 
        nle->le_name_chunk = zap_leaf_transfer_array(l, le->le_name_chunk, nl);
        nle->le_value_chunk =
@@ -761,8 +747,8 @@ zap_leaf_transfer_entry(zap_leaf_t *l, int entry, zap_leaf_t *nl)
 
        zap_leaf_chunk_free(l, entry);
 
-       l->l_phys->l_hdr.lh_nentries--;
-       nl->l_phys->l_hdr.lh_nentries++;
+       zap_leaf_phys(l)->l_hdr.lh_nentries--;
+       zap_leaf_phys(nl)->l_hdr.lh_nentries++;
 }
 
 /*
@@ -771,20 +757,22 @@ zap_leaf_transfer_entry(zap_leaf_t *l, int entry, zap_leaf_t *nl)
 void
 zap_leaf_split(zap_leaf_t *l, zap_leaf_t *nl, boolean_t sort)
 {
-       int i;
-       int bit = 64 - 1 - l->l_phys->l_hdr.lh_prefix_len;
+       uint_t bit = 64 - 1 - zap_leaf_phys(l)->l_hdr.lh_prefix_len;
 
        /* set new prefix and prefix_len */
-       l->l_phys->l_hdr.lh_prefix <<= 1;
-       l->l_phys->l_hdr.lh_prefix_len++;
-       nl->l_phys->l_hdr.lh_prefix = l->l_phys->l_hdr.lh_prefix | 1;
-       nl->l_phys->l_hdr.lh_prefix_len = l->l_phys->l_hdr.lh_prefix_len;
+       zap_leaf_phys(l)->l_hdr.lh_prefix <<= 1;
+       zap_leaf_phys(l)->l_hdr.lh_prefix_len++;
+       zap_leaf_phys(nl)->l_hdr.lh_prefix =
+           zap_leaf_phys(l)->l_hdr.lh_prefix | 1;
+       zap_leaf_phys(nl)->l_hdr.lh_prefix_len =
+           zap_leaf_phys(l)->l_hdr.lh_prefix_len;
 
        /* break existing hash chains */
-       zap_memset(l->l_phys->l_hash, CHAIN_END, 2*ZAP_LEAF_HASH_NUMENTRIES(l));
+       memset(zap_leaf_phys(l)->l_hash, CHAIN_END,
+           2*ZAP_LEAF_HASH_NUMENTRIES(l));
 
        if (sort)
-               l->l_phys->l_hdr.lh_flags |= ZLF_ENTRIES_CDSORTED;
+               zap_leaf_phys(l)->l_hdr.lh_flags |= ZLF_ENTRIES_CDSORTED;
 
        /*
         * Transfer entries whose hash bit 'bit' is set to nl; rehash
@@ -795,7 +783,7 @@ zap_leaf_split(zap_leaf_t *l, zap_leaf_t *nl, boolean_t sort)
         * but this accesses memory more sequentially, and when we're
         * called, the block is usually pretty full.
         */
-       for (i = 0; i < ZAP_LEAF_NUMCHUNKS(l); i++) {
+       for (uint_t i = 0; i < ZAP_LEAF_NUMCHUNKS(l); i++) {
                struct zap_leaf_entry *le = ZAP_LEAF_ENTRY(l, i);
                if (le->le_type != ZAP_CHUNK_ENTRY)
                        continue;
@@ -803,42 +791,40 @@ zap_leaf_split(zap_leaf_t *l, zap_leaf_t *nl, boolean_t sort)
                if (le->le_hash & (1ULL << bit))
                        zap_leaf_transfer_entry(l, i, nl);
                else
-                       (void) zap_leaf_rehash_entry(l, i);
+                       (void) zap_leaf_rehash_entry(l, le, i);
        }
 }
 
 void
 zap_leaf_stats(zap_t *zap, zap_leaf_t *l, zap_stats_t *zs)
 {
-       int i, n;
-
-       n = zap->zap_f.zap_phys->zap_ptrtbl.zt_shift -
-           l->l_phys->l_hdr.lh_prefix_len;
+       uint_t n = zap_f_phys(zap)->zap_ptrtbl.zt_shift -
+           zap_leaf_phys(l)->l_hdr.lh_prefix_len;
        n = MIN(n, ZAP_HISTOGRAM_SIZE-1);
        zs->zs_leafs_with_2n_pointers[n]++;
 
 
-       n = l->l_phys->l_hdr.lh_nentries/5;
+       n = zap_leaf_phys(l)->l_hdr.lh_nentries/5;
        n = MIN(n, ZAP_HISTOGRAM_SIZE-1);
        zs->zs_blocks_with_n5_entries[n]++;
 
        n = ((1<<FZAP_BLOCK_SHIFT(zap)) -
-           l->l_phys->l_hdr.lh_nfree * (ZAP_LEAF_ARRAY_BYTES+1))*10 /
+           zap_leaf_phys(l)->l_hdr.lh_nfree * (ZAP_LEAF_ARRAY_BYTES+1))*10 /
            (1<<FZAP_BLOCK_SHIFT(zap));
        n = MIN(n, ZAP_HISTOGRAM_SIZE-1);
        zs->zs_blocks_n_tenths_full[n]++;
 
-       for (i = 0; i < ZAP_LEAF_HASH_NUMENTRIES(l); i++) {
-               int nentries = 0;
-               int chunk = l->l_phys->l_hash[i];
+       for (uint_t i = 0; i < ZAP_LEAF_HASH_NUMENTRIES(l); i++) {
+               uint_t nentries = 0;
+               uint_t chunk = zap_leaf_phys(l)->l_hash[i];
 
                while (chunk != CHAIN_END) {
                        struct zap_leaf_entry *le =
                            ZAP_LEAF_ENTRY(l, chunk);
 
-                       n = 1 + ZAP_LEAF_ARRAY_NCHUNKS(le->le_name_length) +
-                           ZAP_LEAF_ARRAY_NCHUNKS(le->le_value_length *
-                           le->le_int_size);
+                       n = 1 + ZAP_LEAF_ARRAY_NCHUNKS(le->le_name_numints) +
+                           ZAP_LEAF_ARRAY_NCHUNKS(le->le_value_numints *
+                           le->le_value_intlen);
                        n = MIN(n, ZAP_HISTOGRAM_SIZE-1);
                        zs->zs_entries_using_n_chunks[n]++;