I've been missing this feature from Azureus, so I thought that I'll implement it as is a fairly simple thing : patch.

Now 5G file is allocated in less than 1s instead of 20s, and 1024 * 3M are allocated in 10s vs. 1m05s.

It depends on xfsprogs(ktorrent will compile without it too, but won't use XFS methods; didn't test that ).

It should be possible to implement similar feature for Reiser4 and Ext4, but since both of those FS-es are experimental and undocumented it has to wait.


I hope that I didn't break many things and someone else will find this patch useful too.

Athantor wrote:I've been missing this feature from Azureus, so I thought that I'll implement it as is a fairly simple thing : patch.

Now 5G file is allocated in less than 1s instead of 20s, and 1024 * 3M are allocated in 10s vs. 1m05s.

It depends on xfsprogs(ktorrent will compile without it too, but won't use XFS methods; didn't test that ).

It should be possible to implement similar feature for Reiser4 and Ext4, but since both of those FS-es are experimental and undocumented it has to wait.


I hope that I didn't break many things and someone else will find this patch useful too.

Nice, I need to make sure it doesn't conflict with our full disk preallocate we added last month.

Does this xfs preallocate avoids fragmentation ?

George wrote:Nice,

Thank you.

George wrote:I need to make sure it doesn't conflict with our full disk preallocate we added last month.

I've based my 'XfsPreallocate' on your 'FatPreallocate', so I hope it won't.

George wrote:Does this xfs preallocate avoids fragmentation ?

Yes, it's more modern way of preallocation, but filesystem has to support it, and every FS has its own way of using it. [1], [2], [3] (allocate-on-flash and extents), [4].

Raiser4 has its own undocumented, strange calls in libreiser4 and ext4 has 'EXT4_IOC_PREALLOCATE' ioctl, but I wasn't able to use it. I've also found on LKML patches adding this functionality to ext3, and messages suggesting using delalloc as posix_fallocate();

BTW:
While I was 'playing' with this, I've found that the labels showing free space in 'Open' file selector won't show correct values if 'Download to'/m_downloadLocation points to the location on other drive. Maybe this is intended, but personally I think that this is bit illogical, so I came with this.

OK, XFS patch is committed

Wow, thanks!

The other patch to.

A big thanks for providing these patches.

Great!

As the XFS includes are b0rked, the '-ansi' flag breaks the compilation (at least for me ): fix.

Adding this fix breaks my compilation :

cd '/home/joris/ktorrent/newstuff/debug' && WANT_AUTOCONF_2_5="1" WANT_AUTOMAKE_1_6="1" make
compiling /home/joris/ktorrent/newstuff/libktorrent/util/fileops.cpp (/home/joris/ktorrent/newstuff/libktorrent/util/fileops.cpp)
compiling fileops.TUlo (g++)
/usr/include/xfs/platform_defs.h:41: error: conflicting declaration ‘typedef long long unsigned int __u64’
/home/joris/ktorrent/newstuff/libktorrent/util/fileops.cpp:46: error: ‘__u64’ has a previous declaration as ‘typedef uint64_t __u64’
/usr/include/xfs/platform_defs.h:42: error: conflicting declaration ‘typedef long long int __s64’
/home/joris/ktorrent/newstuff/libktorrent/util/fileops.cpp:47: error: ‘__s64’ has a previous declaration as ‘typedef int64_t __s64’
*** Exited with status: 2 ***

I've found something like this:

/usr/include/asm/types.h wrote:

Code: Select all

#if (defined(__GNUC__) && !defined(__STRICT_ANSI__)) || (defined(__STDC_VERSION__) && __STDC_VERSION__ >= 199901L)

guarding missing types; and this file is from:

equery wrote:

Code: Select all

athantor@athantor ~ % zsh belongs /usr/include/asm/types.h
[ Searching for file(s) /usr/include/asm/types.h in *... ]
sys-kernel/linux-headers-2.6.21 (/usr/include/asm/types.h)

This #if has changed between .20 and .21 (or guys from gentoo broke something )

How about this: patch v2?

It compiles so we will commit it.