Android NDK R8e does not provide ftruncate64, but bionic has the
assembler code to access kernel function. We borrowed those
ftruncate64.S files from android source code repository. It turns out
that x86 asm.h in NDK R8e is also broken, so latest asm.h was also
borrowed.
I tried CreateFile but the subsequent ReadFile fails with Access
Denied if sparse file is read on NTFS. I mostly reverted previous
changes and use _wsopen with read/write share enabled instead of
CreateFile.
This change also includes --enable-mmap support for MinGW32
build. Memory mapped file may be useful for 64-bits OS and lots of
RAM. Currently, FlushViewOfFile is not called during the download, so
it is slightly vulnerable against sudden power loss. I found lots of
read when resuming download due to page fault. So for now it is useful
for the initial download. I recommend not to use
--file-allocation=prealloc with --enable-mmap for MinGW32, because it
triggers page faults even in the initial download. Anyway, the option
is experimental.
Using off_t, at least, in DiskAdaptor layer is problematic because
torrent can contain under 2GiB files but total sum of those files may
exceed 2GiB limit, which makes off_t overflow in 32 bit system without
large file support. So we use int64_t in API. We'll check the file
length before download so that it does not exceed max off_t.
~AbstractDiskWriter calles closeFile(), but suppresses exception.
MultiDiskAdaptor::closeFile() logs error if child
DiskWriter::closeFile() throws exception. This exception is not
rethrown. If at least one exception is caught,
MultiDiskAdaptor::closeFile() throws new DlAbortEx.
RequestGroupMan::closeFile() just logs exception and suppress each
exception. Generally, don't call closeFile() in destructor. If you
need to call it, it must suppress the exception.
We replaced most of std::cerr with global::cerr. windows.h is now
included from common.h. Before including it, we define WINVER. We
renamed some variable name because some macros in windows.h collide
with them.
We use SetFilePointerEx and SetEndOfFile to allocate extents. This
only works with NTFS. To enable this feature, --file-allocation=falloc
must be given.
When allocating disk space, for Linux system with fallocate()
system call, first check file system supports fallocate. This
just run fallocate with small chunk and see it succeeds or fails.
If it succeeds, use fallocate() to allocate entire file otherwise
fall back to traditional slower method: writing zeros. This
behavior is enabled in --file-allocation=prealloc, so this is
enabled by default for most modern Linux.
* configure.ac
* src/AbstractDiskWriter.cc
* src/AbstractDiskWriter.h
* src/AbstractSingleDiskAdaptor.cc
* src/AdaptiveFileAllocationIterator.cc
* src/AdaptiveFileAllocationIterator.h
* src/DefaultPieceStorage.cc
* src/DiskAdaptor.cc
* src/DiskAdaptor.h
* src/FallocFileAllocationIterator.cc
* src/Makefile.am
* src/MultiFileAllocationIterator.cc
* src/OptionHandlerFactory.cc
* test/FallocFileAllocationIteratorTest.cc
* test/Makefile.am
Tatsuhiro Tsujikawa