Bug 912790

If strace attempts to stat a file which returns an inode number above 32 bits, it will fail.  Although ext4 doesn't have 64-bit inode nrs today, btrfs, xfs, and others will.

This can be tested on ext4 by using this systemtap script which bumps up any ext4 inode number above 2^32; the line numbers work for F18 or so:

probe kernel.statement("ext4_getattr@fs/ext4/inode.c+21")
{
	$stat->ino = $stat->ino + 4294967295;
}

probe begin { log("starting probe") }

We can see how it behaves when running:

# stat testfile | grep Inode
Device: fd01h/64769d	Inode: 281242      Links: 1

# stap -g change-ino.stp  &
starting probe

# stat testfile | grep Inode
Device: fd01h/64769d	Inode: 4295248537  Links: 1

If we attempt to use strace while ext4 is returning large inode numbers, it fails:

# strace /bin/ls
strace: Can't stat '/bin/ls': Value too large for defined data type

because the 32-bit stat call returns EOVERFLOW due to the large inode nr.

Unfortunately, strace would be useful for evaluating other binaries which fail in this way, so it'd be nice to fix.

Using -D_FILE_OFFSET_BITS=64 at compile time (or putting it in a header) will invoke the 64-bit variants of stat etc; then just be sure local code handles the larger values as needed.