[fuse-devel] Fstat64 succeeding, but libc returning EOVERFLOW

[Han-Wen N. <ha...@gm...>]

Hi there,

I'm debugging an odd problem with my FUSE library.

When I set timeout = 0.0 on lookup/getattr results, some programs
start failing with EOVERFLOW.

>From inspection of the source code of the userspace program, this
happens because it is using fstat() rather than fstat64().

However, strace  -v has the following to say

fstat64(3, {st_dev=makedev(0, 82), st_ino=178635819520,
st_mode=S_IFREG|0666, st_nlink=1, st_uid=500, st_gid=500,
st_blksize=4096, st_blocks=24, st_size=11133,
st_atime=2011/07/21-01:54:46, st_mtime=2010/12/13-02:23:53,
st_ctime=2011/07/13-10:35:51}) = 0
close(3)                                = 0
lseek(2, 0, SEEK_CUR)                   = -1 ESPIPE (Illegal seek)
write(2, "Could not open input file '", 27) = 27
write(2, "X86.td", 6)                   = 6
write(2, "': ", 3)                      = 3
write(2, "Value too large for defined data type", 37) = 37
write(2, "\n", 1)                       = 1
exit_group(1)                           = ?

It looks as if st_ino is overflowing.  According to the manpage, it
EOVERFLOW only happens for the size field, but the inode number is a
candidate too (its sizeof is 4 on 32-bit).

How do the daemon's inode numbers translate to the VFS' inodes anyway?
 AFAICS they are not passed on verbatim.

On a related note, what is the purpose of the generation field
-- 
Han-Wen Nienhuys - ha...@xs... - http://www.xs4all.nl/~hanwen