sendfile(2) isn't always able to send to all file types, so current code only detected the ability to send by having sent something. This commit tries a little harder to detect permanent data-send problems on the first try. I'm unable to find a case where this fails on Linux, in particular because we don't enter this block if the source file isn't S_IFREG. Now, on Linux, there's no file-copy pump: openat(AT_FDCWD, "dummy.o", O_RDONLY|O_CLOEXEC) = 4 openat(AT_FDCWD, "/tmp/tmp", O_RDWR|O_CLOEXEC|O_TMPFILE, 0600) = 5 ioctl(5, BTRFS_IOC_CLONE or FICLONE, 4) = -1 EXDEV sendfile(5, 4, NULL, 2147479552) = -1 ENOSPC lseek(5, 0, SEEK_SET) = 0 close(5) = 0 write(2, "\"Could not copy to /tmp/tmp/f: No space left on device\"") = 56 close(4) = 0 Change-Id: I88ce23447b0d7341848cfffd2c469e15cb69f5a6 Reviewed-by: Ahmad Samir <a.samirh78@gmail.com>