Make opening a non-existent interface on Linux return 'no such interface'

[afipay]

• Attempts to open a non-existent interface should always return "no such interface" even if you don't have capture permissions #1538
• Different behavior with zero-padded interface numbers tcpdump#1334

Fix interface existence check on Linux before privileged operations

When attempting to open a non-existent network interface on Linux,
libpcap would return PCAP_ERROR_PERM_DENIED instead of the correct
PCAP_ERROR_NO_SUCH_DEVICE. This occurred because the privileged
socket(PF_PACKET, SOCK_RAW, 0) call in setup_socket() would fail
with permission errors before interface validation could occur.

This fix adds an interface existence check using an unprivileged
AF_INET socket and SIOCGIFINDEX ioctl before attempting to create
the privileged packet socket. Non-existent interfaces now correctly
return PCAP_ERROR_NO_SUCH_DEVICE.

This resolves the issue where applications like tcpdump would report
"Permission denied" instead of "No such device exists" for non-existent
interfaces, breaking their fallback logic from interface names to indices.

Note: BSD/macOS platforms have the same underlying issue where BPF device
access fails with permission errors before interface validation. This
should be addressed in a separate commit.

Fixes: libpcap issue #1538
Related: tcpdump issue #1334

./tcpdump -i nonexistent
----------------------------------------
tcpdump: nonexistent: No such device exists
(No such device exists)

./tcpdump -i eth999
----------------------------------------
tcpdump: eth999: No such device exists
(No such device exists)

./tcpdump -i 000000000000008
----------------------------------------
tcpdump: Invalid adapter index 8: only 7 interfaces found

./tcpdump -i 00000000000000073
----------------------------------------
tcpdump: Invalid adapter index 73: only 7 interfaces found

[infrastation]

Thank you for preparing this change. The comments that clarify the problem and the solution should be in the commit message.

[infrastation]

On Linux the only proposed change is to iface_get_ts_types(), thus the "check for existence before trying a privileged operation" logic does not apply to Linux:

$ ./testprogs/activatetest 
Trying to use capture device "nosuchdevice"...
FAIL: Permission denied from pcap_activate(), retry with higher privileges.

It looks like the two proposed changes are not related and should be two different commits, each stating and solving a separate problem. Also, since pcap-linux.c already implements helper functions such as iface_get_id(), it would be easy to implement the same logic for Linux. That said, the logic increases the race condition window, which would require some attention.

This requires a bit more work before it is ready.

[infrastation]

As noted.

[fxlb]

Rebased.

[guyharris]

Should the check whether a device exists be done before trying to bind it, or after the attempt to bind the device fails with a permission error?

There's a time-of-check/time-of-use issue for both of those, but, if interfaces are appearing or disappearing, there's a risk of an error being reported for a condition that's not true by the time the user sees the error no matter what the code does, so I'm not bothered by that.

The one advantage of doing it afterwards is that fewer system calls are made in the case where everything succeeds, i.e. if the program does have permission to capture. I'm not sure that's a big enough difference to worry about.

[infrastation]

Make it work, make it right, make it fast?