ZAP: Massively switch to _by_dnode() interfaces

Before this change ZAP called dnode_hold() for almost every block
access, that was clearly visible in profiler under heavy load, such
as BRT.  This patch makes it always hold the dnode reference between
zap_lockdir() and zap_unlockdir().  It allows to avoid most of dnode
operations between those.  It also adds several new _by_dnode() APIs
to ZAP and uses them in BRT code.  Also adds dmu_prefetch_by_dnode()
variant and uses it in the ZAP code.

After this there remains only one call to dmu_buf_dnode_enter(),
which seems to be unneeded.  So remove the call and the functions.

Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov>
Signed-off-by:	Alexander Motin <mav@FreeBSD.org>
Sponsored by:	iXsystems, Inc.
Closes #15951

include/sys/dmu.h

@@ -752,8 +752,6 @@ void dmu_buf_sub_user_size(dmu_buf_t *db, uint64_t nsub);
 void *dmu_buf_get_user(dmu_buf_t *db);
 
 objset_t *dmu_buf_get_objset(dmu_buf_t *db);
-dnode_t *dmu_buf_dnode_enter(dmu_buf_t *db);
-void dmu_buf_dnode_exit(dmu_buf_t *db);
 
 /* Block until any in-progress dmu buf user evictions complete. */
 void dmu_buf_user_evict_wait(void);
@@ -902,6 +900,8 @@ extern uint_t zfs_max_recordsize;
  */
 void dmu_prefetch(objset_t *os, uint64_t object, int64_t level, uint64_t offset,
 	uint64_t len, enum zio_priority pri);
+void dmu_prefetch_by_dnode(dnode_t *dn, int64_t level, uint64_t offset,
+	uint64_t len, enum zio_priority pri);
 void dmu_prefetch_dnode(objset_t *os, uint64_t object, enum zio_priority pri);
 
 typedef struct dmu_object_info {

include/sys/zap.h

@@ -253,6 +253,9 @@ int zap_add_by_dnode(dnode_t *dn, const char *key,
 int zap_add_uint64(objset_t *ds, uint64_t zapobj, const uint64_t *key,
     int key_numints, int integer_size, uint64_t num_integers,
     const void *val, dmu_tx_t *tx);
+int zap_add_uint64_by_dnode(dnode_t *dn, const uint64_t *key,
+    int key_numints, int integer_size, uint64_t num_integers,
+    const void *val, dmu_tx_t *tx);
 
 /*
  * Set the attribute with the given name to the given value.  If an
@@ -267,6 +270,9 @@ int zap_update(objset_t *ds, uint64_t zapobj, const char *name,
 int zap_update_uint64(objset_t *os, uint64_t zapobj, const uint64_t *key,
     int key_numints,
     int integer_size, uint64_t num_integers, const void *val, dmu_tx_t *tx);
+int zap_update_uint64_by_dnode(dnode_t *dn, const uint64_t *key,
+    int key_numints,
+    int integer_size, uint64_t num_integers, const void *val, dmu_tx_t *tx);
 
 /*
  * Get the length (in integers) and the integer size of the specified
@@ -292,6 +298,8 @@ int zap_remove_norm(objset_t *ds, uint64_t zapobj, const char *name,
 int zap_remove_by_dnode(dnode_t *dn, const char *name, dmu_tx_t *tx);
 int zap_remove_uint64(objset_t *os, uint64_t zapobj, const uint64_t *key,
     int key_numints, dmu_tx_t *tx);
+int zap_remove_uint64_by_dnode(dnode_t *dn, const uint64_t *key,
+    int key_numints, dmu_tx_t *tx);
 
 /*
  * Returns (in *count) the number of attributes in the specified zap

include/sys/zap_impl.h

@@ -145,6 +145,7 @@ typedef struct zap {
 	dmu_buf_user_t zap_dbu;
 	objset_t *zap_objset;
 	uint64_t zap_object;
+	dnode_t *zap_dnode;
 	struct dmu_buf *zap_dbuf;
 	krwlock_t zap_rwlock;
 	boolean_t zap_ismicro;