On Thu, Nov 08, 2001 at 08:10:07AM +0200, Ville Herva wrote:
> On Wed, Nov 07, 2001 at 11:40:25PM +0100, you [Peter Seiderer] claimed:
> > Mhhh,
> > the strace output from the 'login: root' one (the one which was good)
> > looks the same till the EFBIG place:
> > 
> > 	write(1, "\10\10\10\10\10", 5)          = 5
> > 	write(1, "16/44", 5)                    = 5
> > 	_llseek(4, 18446744071562084352, [2147500032], SEEK_SET) = 0
> > 	write(4, "\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0"..., 32768) = 32768
> > 	_llseek(4, 18446744071562117120, [2147532800], SEEK_SET) = 0
> > 	write(4, "\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0"..., 32768) = 32768
> > 	_llseek(4, 18446744071562149888, [2147565568], SEEK_SET) = 0
> > 	write(4, "\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0"..., 32768) = 32768
> > 	_llseek(4, 18446744071562182656, [2147598336], SEEK_SET) = 0
> > 	write(4, "\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0"..., 32768) = 32768
> 
> Weird. Perhaps strace gets that wrong and the problem is elsewhere.
> 
> Did you make sure that fd 4 is the same _partition_ in both cases (using
> strace)? The only thing I could imagine exposing 2GB limit is writing to a
> file.
> 
> > > > 	zodiak login: seiderer
> > > > 	Password:
> > > > 	seiderer@zodiak:~ > su -
> > > > 	Password:
> > > > 	zodiak:~ #
> > > > 	zodiak:~ # mkfs.ext2 /dev/hdc4
> > > > 	mke2fs 1.18, 11-Nov-1999 for EXT2 FS 0.5b, 95/08/09
> > > > 	Filesystem label=
> > > > 	OS type: Linux
> > > > 	Block size=4096 (log=2)
> > > > 	Fragment size=4096 (log=2)
> > > > 	716672 inodes, 1432116 blocks
> > > > 	71605 blocks (5.00%) reserved for the super user
> > > > 	First data block=0
> > > > 	44 block groups
> > > > 	32768 blocks per group, 32768 fragments per group
> > > > 	16288 inodes per group
> > > > 	Superblock backups stored on blocks:
> > > > 	        32768, 98304, 163840, 229376, 294912, 819200, 884736
> > > > 
> > > > 	Writing inode tables: 16/44File size limit exceeded
> > > > 
> > > > strace showed that write returned wit EFBIG and the process ended with SIGXFSZ:
> > > > 
> > > > 	write(1, "\10\10\10\10\10", 5)          = 5
> > > > 	write(1, "16/44", 5)                    = 5
> > > > 	_llseek(4, 18446744071562084352, [2147500032], SEEK_SET) = 0
> > > > 	write(4, "\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0"..., 32768) = -1 EFBIG (File too large)
> > > > 	--- SIGXFSZ (File size limit exceeded) ---
> > > > 	+++ killed by SIGXFSZ +++
> > > 
> > > Hmm, 18446744071562084352 = 0xffffffff80004000, 2147500032 = 0x80004000...
> > > It looks a tad like llseek's offset_high would have been corrupted...
> > > Strange.
> > > 
> > > 1432116 blocks * 4096 bytes/block * 16/44 written = 2133071685.81818 so
> > > 2147500032 looks sane(ish).
> 
> -- v --
> 
> v@iki.fi
-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/