Fix ENOSPC in "Handle zap_add() failures in ..."

Commit cc63068 caused ENOSPC error when copy a large amount of files
between two directories. The reason is that the patch limits zap leaf
expansion to 2 retries, and return ENOSPC when failed.

The intent for limiting retries is to prevent pointlessly growing table
to max size when adding a block full of entries with same name in
different case in mixed mode. However, it turns out we cannot use any
limit on the retry. When we copy files from one directory in readdir
order, we are copying in hash order, one leaf block at a time. Which
means that if the leaf block in source directory has expanded 6 times,
and you copy those entries in that block, by the time you need to expand
the leaf in destination directory, you need to expand it 6 times in one
go. So any limit on the retry will result in error where it shouldn't.

Note that while we do use different salt for different directories, it
seems that the salt/hash function doesn't provide enough randomization
to the hash distance to prevent this from happening.

Since cc63068 has already been reverted. This patch adds it back and
removes the retry limit.

Also, as it turn out, failing on zap_add() has a serious side effect for
mzap_upgrade(). When upgrading from micro zap to fat zap, it will
call zap_add() to transfer entries one at a time. If it hit any error
halfway through, the remaining entries will be lost, causing those files
to become orphan. This patch add a VERIFY to catch it.

Reviewed-by: Sanjeev Bagewadi <[email protected]>
Reviewed-by: Richard Yao <[email protected]>
Reviewed-by: Tony Hutter <[email protected]>
Reviewed-by: Albert Lee <[email protected]>
Reviewed-by: Brian Behlendorf <[email protected]>
Reviewed by: Matthew Ahrens <[email protected]>
Signed-off-by: Chunwei Chen <[email protected]>
Closes #7401 
Closes #7421 

5 files changed, 130 insertions, 32 deletions

diff --git a/module/zfs/zap.c b/module/zfs/zap.c
index 2f6aed667..d899ccdb7 100644
--- a/module/zfs/zap.c
+++ b/module/zfs/zap.c
@@ -852,8 +852,16 @@ retry:
 	} else if (err == EAGAIN) {
 		err = zap_expand_leaf(zn, l, tag, tx, &l);
 		zap = zn->zn_zap;	/* zap_expand_leaf() may change zap */
-		if (err == 0)
+		if (err == 0) {
 			goto retry;
+		} else if (err == ENOSPC) {
+			/*
+			 * If we failed to expand the leaf, then bailout
+			 * as there is no point trying
+			 * zap_put_leaf_maybe_grow_ptrtbl().
+			 */
+			return (err);
+		}
 	}
 
 out:
diff --git a/module/zfs/zap_leaf.c b/module/zfs/zap_leaf.c
index 5341fc098..661fd747b 100644
--- a/module/zfs/zap_leaf.c
+++ b/module/zfs/zap_leaf.c
@@ -53,7 +53,7 @@ static uint16_t *zap_leaf_rehash_entry(zap_leaf_t *l, uint16_t entry);
 	((h) >> \
 	(64 - ZAP_LEAF_HASH_SHIFT(l) - zap_leaf_phys(l)->l_hdr.lh_prefix_len)))
 
-#define	LEAF_HASH_ENTPTR(l, h) (&zap_leaf_phys(l)->l_hash[LEAF_HASH(l, h)])
+#define	LEAF_HASH_ENTPTR(l, h)	(&zap_leaf_phys(l)->l_hash[LEAF_HASH(l, h)])
 
 extern inline zap_leaf_phys_t *zap_leaf_phys(zap_leaf_t *l);
 
diff --git a/module/zfs/zap_micro.c b/module/zfs/zap_micro.c
index b81a48a0f..791cbee88 100644
--- a/module/zfs/zap_micro.c
+++ b/module/zfs/zap_micro.c
@@ -363,6 +363,41 @@ mze_find_unused_cd(zap_t *zap, uint64_t hash)
 	return (cd);
 }
 
+/*
+ * Each mzap entry requires at max : 4 chunks
+ * 3 chunks for names + 1 chunk for value.
+ */
+#define	MZAP_ENT_CHUNKS	(1 + ZAP_LEAF_ARRAY_NCHUNKS(MZAP_NAME_LEN) + \
+	ZAP_LEAF_ARRAY_NCHUNKS(sizeof (uint64_t)))
+
+/*
+ * Check if the current entry keeps the colliding entries under the fatzap leaf
+ * size.
+ */
+static boolean_t
+mze_canfit_fzap_leaf(zap_name_t *zn, uint64_t hash)
+{
+	zap_t *zap = zn->zn_zap;
+	mzap_ent_t mze_tofind;
+	mzap_ent_t *mze;
+	avl_index_t idx;
+	avl_tree_t *avl = &zap->zap_m.zap_avl;
+	uint32_t mzap_ents = 0;
+
+	mze_tofind.mze_hash = hash;
+	mze_tofind.mze_cd = 0;
+
+	for (mze = avl_find(avl, &mze_tofind, &idx);
+	    mze && mze->mze_hash == hash; mze = AVL_NEXT(avl, mze)) {
+		mzap_ents++;
+	}
+
+	/* Include the new entry being added */
+	mzap_ents++;
+
+	return (ZAP_LEAF_NUMCHUNKS_DEF > (mzap_ents * MZAP_ENT_CHUNKS));
+}
+
 static void
 mze_remove(zap_t *zap, mzap_ent_t *mze)
 {
@@ -638,16 +673,15 @@ mzap_upgrade(zap_t **zapp, void *tag, dmu_tx_t *tx, zap_flags_t flags)
 		dprintf("adding %s=%llu\n",
 		    mze->mze_name, mze->mze_value);
 		zn = zap_name_alloc(zap, mze->mze_name, 0);
-		err = fzap_add_cd(zn, 8, 1, &mze->mze_value, mze->mze_cd,
-		    tag, tx);
+		/* If we fail here, we would end up losing entries */
+		VERIFY0(fzap_add_cd(zn, 8, 1, &mze->mze_value, mze->mze_cd,
+		    tag, tx));
 		zap = zn->zn_zap;	/* fzap_add_cd() may change zap */
 		zap_name_free(zn);
-		if (err)
-			break;
 	}
 	vmem_free(mzp, sz);
 	*zapp = zap;
-	return (err);
+	return (0);
 }
 
 /*
@@ -1190,7 +1224,8 @@ zap_add_impl(zap_t *zap, const char *key,
 		err = fzap_add(zn, integer_size, num_integers, val, tag, tx);
 		zap = zn->zn_zap;	/* fzap_add() may change zap */
 	} else if (integer_size != 8 || num_integers != 1 ||
-	    strlen(key) >= MZAP_NAME_LEN) {
+	    strlen(key) >= MZAP_NAME_LEN ||
+	    !mze_canfit_fzap_leaf(zn, zn->zn_hash)) {
 		err = mzap_upgrade(&zn->zn_zap, tag, tx, 0);
 		if (err == 0) {
 			err = fzap_add(zn, integer_size, num_integers, val,
diff --git a/module/zfs/zfs_dir.c b/module/zfs/zfs_dir.c
index 76f79cc11..7eb426b78 100644
--- a/module/zfs/zfs_dir.c
+++ b/module/zfs/zfs_dir.c
@@ -742,7 +742,11 @@ zfs_dirent(znode_t *zp, uint64_t mode)
 }
 
 /*
- * Link zp into dl.  Can only fail if zp has been unlinked.
+ * Link zp into dl.  Can fail in the following cases :
+ * - if zp has been unlinked.
+ * - if the number of entries with the same hash (aka. colliding entries)
+ *    exceed the capacity of a leaf-block of fatzap and splitting of the
+ *    leaf-block does not help.
  */
 int
 zfs_link_create(zfs_dirlock_t *dl, znode_t *zp, dmu_tx_t *tx, int flag)
@@ -776,6 +780,24 @@ zfs_link_create(zfs_dirlock_t *dl, znode_t *zp, dmu_tx_t *tx, int flag)
 			    NULL, &links, sizeof (links));
 		}
 	}
+
+	value = zfs_dirent(zp, zp->z_mode);
+	error = zap_add(ZTOZSB(zp)->z_os, dzp->z_id, dl->dl_name, 8, 1,
+	    &value, tx);
+
+	/*
+	 * zap_add could fail to add the entry if it exceeds the capacity of the
+	 * leaf-block and zap_leaf_split() failed to help.
+	 * The caller of this routine is responsible for failing the transaction
+	 * which will rollback the SA updates done above.
+	 */
+	if (error != 0) {
+		if (!(flag & ZRENAMING) && !(flag & ZNEW))
+			drop_nlink(ZTOI(zp));
+		mutex_exit(&zp->z_lock);
+		return (error);
+	}
+
 	SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_PARENT(zfsvfs), NULL,
 	    &dzp->z_id, sizeof (dzp->z_id));
 	SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_FLAGS(zfsvfs), NULL,
@@ -813,11 +835,6 @@ zfs_link_create(zfs_dirlock_t *dl, znode_t *zp, dmu_tx_t *tx, int flag)
 	ASSERT(error == 0);
 	mutex_exit(&dzp->z_lock);
 
-	value = zfs_dirent(zp, zp->z_mode);
-	error = zap_add(ZTOZSB(zp)->z_os, dzp->z_id, dl->dl_name,
-	    8, 1, &value, tx);
-	ASSERT(error == 0);
-
 	return (0);
 }
 
diff --git a/module/zfs/zfs_vnops.c b/module/zfs/zfs_vnops.c
index f7f2f6167..0eeb2bb42 100644
--- a/module/zfs/zfs_vnops.c
+++ b/module/zfs/zfs_vnops.c
@@ -1450,6 +1450,7 @@ top:
 			dmu_tx_hold_write(tx, DMU_NEW_OBJECT,
 			    0, acl_ids.z_aclp->z_acl_bytes);
 		}
+
 		error = dmu_tx_assign(tx,
 		    (waited ? TXG_NOTHROTTLE : 0) | TXG_NOWAIT);
 		if (error) {
@@ -1467,10 +1468,22 @@ top:
 		}
 		zfs_mknode(dzp, vap, tx, cr, 0, &zp, &acl_ids);
 
+		error = zfs_link_create(dl, zp, tx, ZNEW);
+		if (error != 0) {
+			/*
+			 * Since, we failed to add the directory entry for it,
+			 * delete the newly created dnode.
+			 */
+			zfs_znode_delete(zp, tx);
+			remove_inode_hash(ZTOI(zp));
+			zfs_acl_ids_free(&acl_ids);
+			dmu_tx_commit(tx);
+			goto out;
+		}
+
 		if (fuid_dirtied)
 			zfs_fuid_sync(zfsvfs, tx);
 
-		(void) zfs_link_create(dl, zp, tx, ZNEW);
 		txtype = zfs_log_create_txtype(Z_FILE, vsecp, vap);
 		if (flag & FIGNORECASE)
 			txtype |= TX_CI;
@@ -2064,13 +2077,18 @@ top:
 	 */
 	zfs_mknode(dzp, vap, tx, cr, 0, &zp, &acl_ids);
 
-	if (fuid_dirtied)
-		zfs_fuid_sync(zfsvfs, tx);
-
 	/*
 	 * Now put new name in parent dir.
 	 */
-	(void) zfs_link_create(dl, zp, tx, ZNEW);
+	error = zfs_link_create(dl, zp, tx, ZNEW);
+	if (error != 0) {
+		zfs_znode_delete(zp, tx);
+		remove_inode_hash(ZTOI(zp));
+		goto out;
+	}
+
+	if (fuid_dirtied)
+		zfs_fuid_sync(zfsvfs, tx);
 
 	*ipp = ZTOI(zp);
 
@@ -2080,6 +2098,7 @@ top:
 	zfs_log_create(zilog, tx, txtype, dzp, zp, dirname, vsecp,
 	    acl_ids.z_fuidp, vap);
 
+out:
 	zfs_acl_ids_free(&acl_ids);
 
 	dmu_tx_commit(tx);
@@ -2089,10 +2108,14 @@ top:
 	if (zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS)
 		zil_commit(zilog, 0);
 
-	zfs_inode_update(dzp);
-	zfs_inode_update(zp);
+	if (error != 0) {
+		iput(ZTOI(zp));
+	} else {
+		zfs_inode_update(dzp);
+		zfs_inode_update(zp);
+	}
 	ZFS_EXIT(zfsvfs);
-	return (0);
+	return (error);
 }
 
 /*
@@ -3950,6 +3973,13 @@ top:
 				VERIFY3U(zfs_link_destroy(tdl, szp, tx,
 				    ZRENAMING, NULL), ==, 0);
 			}
+		} else {
+			/*
+			 * If we had removed the existing target, subsequent
+			 * call to zfs_link_create() to add back the same entry
+			 * but, the new dnode (szp) should not fail.
+			 */
+			ASSERT(tzp == NULL);
 		}
 	}
 
@@ -4120,14 +4150,18 @@ top:
 	/*
 	 * Insert the new object into the directory.
 	 */
-	(void) zfs_link_create(dl, zp, tx, ZNEW);
-
-	if (flags & FIGNORECASE)
-		txtype |= TX_CI;
-	zfs_log_symlink(zilog, tx, txtype, dzp, zp, name, link);
+	error = zfs_link_create(dl, zp, tx, ZNEW);
+	if (error != 0) {
+		zfs_znode_delete(zp, tx);
+		remove_inode_hash(ZTOI(zp));
+	} else {
+		if (flags & FIGNORECASE)
+			txtype |= TX_CI;
+		zfs_log_symlink(zilog, tx, txtype, dzp, zp, name, link);
 
-	zfs_inode_update(dzp);
-	zfs_inode_update(zp);
+		zfs_inode_update(dzp);
+		zfs_inode_update(zp);
+	}
 
 	zfs_acl_ids_free(&acl_ids);
 
@@ -4135,10 +4169,14 @@ top:
 
 	zfs_dirent_unlock(dl);
 
-	*ipp = ZTOI(zp);
+	if (error == 0) {
+		*ipp = ZTOI(zp);
 
-	if (zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS)
-		zil_commit(zilog, 0);
+		if (zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS)
+			zil_commit(zilog, 0);
+	} else {
+		iput(ZTOI(zp));
+	}
 
 	ZFS_EXIT(zfsvfs);
 	return (error);