linuxdcpp-team team mailing list archive

[LadyStardust <930247@xxxxxxxxxxxxxxxxxx>]

Externad HDD when unplugged caused a crash while hashing
http://pastebin.com/RWrYpL73

-- 
You received this bug notification because you are a member of
Dcplusplus-team, which is subscribed to DC++.
https://bugs.launchpad.net/bugs/930247

Title:
  Crash when hashing a bad file with the new filereader (r2781 and
  newer)

Status in DC++:
  New

Bug description:
  [16:35:19] <Night> when hashing files it tries to use overlapped read by default and if it fails then it uses mapped read.
  [16:35:53] <Night> but mapped read doesnt seem to be able to handle errors like ERROR_CRC and such data invalid errors
  [16:36:06] <Night> so it will crash when hashing such a file.
  [16:37:08] <eMTee> well we've heard about crashes during hashing so as you may found out the cause, this could go to the bug tracker right away
  [16:37:42] <eMTee> any details on why would the mapped reader crash on this error?
  [16:37:51] <Night> it just reads and reads
  [16:38:03] <Night> until the pointer gets false
  [16:38:02] <eMTee> what error is this? low level error reading the disk or what
  [16:38:25] <Night> at my case it seemed to be a bad file
  [16:38:34] <Night> the file cannot be read
  [16:38:58] <eMTee> because of hardware error or logical error on the filesystem?
  [16:39:00] <Night> the user couldnt even upload the file
  [16:39:18] <Night> hardware is still under checks
  [16:39:34] <Night> but was only that one file
  [16:39:43] <Night> everytime
  [16:39:52] <eMTee> is it in your system?
  [16:39:56] <Night> no
  [16:40:14] <Night> the error the read gives is ERROR_CRC
  [16:41:07] <Night> if skip mapped read when hashing it will work since it will throw on the error and hash just fails
  [16:42:05] <Night> 
  	AirDC.exe!DCUnhandledExceptionFilter(_EXCEPTION_POINTERS * e)  Line 171 + 0x21 bytes	C++
   	kernel32.dll!0000000076f7cb7f() 	
   	ntdll.dll!0000000077063318() 	
   	ntdll.dll!0000000077029334() 	
   	ntdll.dll!00000000770258dd() 	
   	ntdll.dll!00000000770296d7() 	
   	ntdll.dll!0000000077036e08() 	
   	AirDC.exe!memcpy()  Line 119	Asm
   	AirDC.exe!dcpp::TigerHash::update(const void * data, unsigned __int64 length)  Line 174	C++
   	AirDC.exe!dcpp::MerkleTree<dcpp::TigerHash,1024>::update(const void * data, unsigned __int64 len)  Line 111	C++
  >	AirDC.exe!std::tr1::_Impl_no_alloc2<std::tr1::_Callable_obj<dcpp::`anonymous namespace'::<lambda1>,0>,bool,void const * __ptr64,unsigned __int64>::_Do_call(const void * _Ax0, unsigned __int64 _Ax1)  Line 65 + 0xae bytes	C++
   	AirDC.exe!dcpp::FileReader::readMapped(const std::basic_string<char,std::char_traits<char>,std::allocator<char> > & file, const std::tr1::function<bool __cdecl(void const *,unsigned __int64)> & callback)  Line 214 + 0x17 bytes	C++
   	AirDC.exe!dcpp::FileReader::read(const std::basic_string<char,std::char_traits<char>,std::allocator<char> > & file, const std::tr1::function<bool __cdecl(void const *,unsigned __int64)> & callback)  Line 28	C++
  [16:44:08] <Night> i fixed it by skipping mapped read with direct read, as i dont know if mapped read can do any good anyway if direct failed
  [16:45:39] <Night> i think arne should be able to understand the issue with the ERROR_CRC handling
  ...
  [17:19:17] <Night> he cant reproduce it with DC++ anymore
  [17:19:46] <Night> figured it might be if its not the first overlapped read that fails
  [17:21:47] <Night> if its a disk error or a filesystem error what so ever it might get thru to the second one and it just reports that it cant read the file
  [17:24:11] <Night> Datafeil (sirkulerende overflødighetskontroll, CRC)   if that makes any sense to you, its the ERROR_CRC that's what it should report
  [17:24:45] <Night> but if it fails on the first overlapped read it will try mapped read and that means crash

To manage notifications about this bug go to:
https://bugs.launchpad.net/dcplusplus/+bug/930247/+subscriptions