PosixMmapReadableFile fails on FreeBSD with large dataset

[GoogleCodeExporter]

What steps will reproduce the problem?
See https://github.com/bitcoin/bitcoin/issues/4362 
I think it happens when it reads very large dataset like blockchain on FreeBSD

What is the expected output? What do you see instead?
expected: Valid database is readed without exceptions.
instead: Invalid argument: not an sstable (bad magic number)

What version of the product are you using? On what operating system?
1.17, FreeBSD amd64

Please provide any additional information below.
I read about previous OSX unmap bug which was corrupting files and find out 
that if i comment out lines 322-337 in file util/env_posix.cc and force program 
to use PosixRandomAccessFile instead of PosixMmapReadableFile everything works 
as expected. There are reported similar issues on ARM. (I am not C++ programmer 
so i cant tell whats wrong with mmap on FreeBSD nor if it is connected to ARM 
problems)

Original issue reported on code.google.com by efe...@gmail.com on 19 Jun 2014 at 8:15

[GoogleCodeExporter]

How large are the datasets? I'm seeing a similar reports with that error 
message on Android, but haven't been able to repro it yet.

Original comment by r...@google.com on 27 Jun 2014 at 5:42

[GoogleCodeExporter]

Also, would you have links to the similar issues on ARM you referred to? Thanks.

Original comment by r...@google.com on 27 Jun 2014 at 6:01

[GoogleCodeExporter]

Sorry, blockchain is not in leveldb. Leveldb is used only for index of 
blockchain and for db of unspent transactions. Actualy all these database files 
are now about 43MB and 473MB. I am trying to separate behaviour which is 
causing error but with no luck yet (imho mostly because i dont understand C++)

Original comment by efe...@gmail.com on 1 Jul 2014 at 8:02