I'm using the nftw() system call in a program I am writing and it returns -1 (error) every time it comes across a directory that is part of an active torrent download. I was wondering if there is a conventional way to cause nftw() to skip the file that gives it trouble and continue walking the file tree.

Thanks,
Matt Bragano

If you are getting (-1) and EINVAL, EINVAL is not an error code set by either fstat or ftw(). It has to be coming from your callback function. Something in that function fails, it is not an nftw() problem.

This is how nftw works - if the nftw function itself fails it returns -1. If the (l/f)stat call it makes it returns (-1) nftw quits and returns (-1). You can control this behavior. If your fn() returns with anything but zero,
nftw bails out and returns that value. Since errno is EINVAL and stat and nftw do not set that error code, then your function has to be doing it.

Thanks for the quick response. I went ahead and implemented a quick test callback function that just prints the name of the files encountered by nftw() to cout and returns 0, and nftw() still returns -1 when it encounters the same problem file described earlier. Also, the program runs solid on other computers I've tried it on.

Any ideas on tracking down the source of the problem? Or even just finding a way to skip the problem file and let nftw() continue?

Thanks,
Matt Bragano

I just discovered the meaning behind the error using perror(). It is returning EOVERFLOW because "Value too large for defined data type." This occurs when:

A field in the stat structure cannot be represented correctly in the current programming environment for one or more files found in the file hierarchy.

I'll keep playing with it to see if I can work around this somehow. It is failing on a rather large file (a DVD iso image) so that is probably the source.

Matt

struct stat.st_size may be overflowing.