Additional linux AF_PACKET functionality

[nathaniel-bennett]

Added socket options, structs and other miscellaneous constants that make memory-mapped packet sockets possible in Linux (PACKET_TX_RING, PACKET_RX_RING, etc.).

[rustbot]

Thanks for the pull request, and welcome! The Rust team is excited to review your changes, and you should hear from @JohnTitor (or someone else) soon.

Please see the contribution instructions for more information.

[nathaniel-bennett]

(Note that this change removes the netpacket/packet.h header file and replaces it with linux/if_packet.h. This is because if_packet.h is a proper superset of packet.h, and the inclusion of both would lead to build errors with structs/constants being defined twice)

[JohnTitor]

@bors r+

[bors]

📌 Commit 0ffc639 has been approved by JohnTitor

It is now in the queue for this repository.

[bors]

⌛ Testing commit 0ffc639 with merge feacf46...

[bors]

💔 Test failed - checks-actions

[JohnTitor]

The failure is valid, I guess the cause is our musl image on CI is old, we could skip tests for that arch.

[nathaniel-bennett]

It seems like fanout_args was the main culprit that was causing the musl build to fail, so I've added a check that skips that particular struct for musl tests.

[JohnTitor]

@bors r+

[bors]

📌 Commit c412376 has been approved by JohnTitor

It is now in the queue for this repository.

[bors]

⌛ Testing commit c412376 with merge eb58fcc...

[bors]

💔 Test failed - checks-actions

[nathaniel-bennett]

Alright, so this time the musl build failed due to some issue with sockaddr_pkt (likely something to do with endianness). I've opted to just remove sockaddr_pkt, since it's considered obsolete. It was part of the original SOCK_PACKET raw socket interface for Linux that was superceded by AF_PACKET and the accompanying sockaddr_ll socket type many years ago.

Fingers crossed that this time we actually pass the extended checks :)

[workingjubilee]

@bors r=JohnTitor

[JohnTitor]

Failed on mips: @rustbot author

[nathaniel-bennett]

Did a rebase for this--the issues with mips and BSD should (hopefully) all be fixed now

[JohnTitor]

@bors r+

[bors]

📌 Commit a3d0958 has been approved by JohnTitor

It is now in the queue for this repository.

[bors]

⌛ Testing commit a3d0958 with merge 9fbadf8...

[bors]

💔 Test failed - checks-actions

[nathaniel-bennett]

Most recent issue was with MIPS--it's in the same camp as sparc64 and musl with big endian alignment. I've added an exception for sockaddr_pkt for mips, similar to what I did for the others.

     Running test/main.rs (target/mips-unknown-linux-gnu/debug/deps/main-ae76ee3ebd17d880)
RUNNING ALL TESTS
rust[16] = 0 != 16 (C): Rust "sockaddr_pkt" -> C
rust[17] = 0 != 17 (C): Rust "sockaddr_pkt" -> C

[JohnTitor]

@bors r+

[bors]

📌 Commit 40c5858 has been approved by JohnTitor

It is now in the queue for this repository.

[bors]

⌛ Testing commit 40c5858 with merge 662cd46...

[bors]

💔 Test failed - checks-actions

[JohnTitor]

Seems something is wrong on powerpc64:

 bad TPACKET3_HDRLEN value at byte 7: rust: 84 (0x54) != c 68 (0x44)
bad tpacket_auxdata size: rust: 24 (0x18) != c 20 (0x14)
bad field offset tp_vlan_tci of tpacket_auxdata: rust: 20 (0x14) != c 16 (0x10)
bad field type tp_vlan_tci of tpacket_auxdata: rust: 0x40027409dc != c 0x40027409d8
bad field offset tp_vlan_tpid of tpacket_auxdata: rust: 22 (0x16) != c 18 (0x12)
bad field type tp_vlan_tpid of tpacket_auxdata: rust: 0x4002740a2e != c 0x4002740a2a
size of struct tpacket_auxdata is 20 in C and 24 in Rust
bad tpacket_hdr_variant1 size: rust: 16 (0x10) != c 12 (0xc)
bad field offset tp_vlan_tpid of tpacket_hdr_variant1: rust: 12 (0xc) != c 8 (0x8)
bad field type tp_vlan_tpid of tpacket_hdr_variant1: rust: 0x4002740ae4 != c 0x4002740ae0
bad field offset tp_padding of tpacket_hdr_variant1: rust: 14 (0xe) != c 10 (0xa)
bad field type tp_padding of tpacket_hdr_variant1: rust: 0x4002740b1e != c 0x4002740b1a
size of struct tpacket_hdr_variant1 is 12 in C and 16 in Rust
bad tpacket_req3 size: rust: 32 (0x20) != c 28 (0x1c)
bad field offset tp_feature_req_word of tpacket_req3: rust: 28 (0x1c) != c 24 (0x18)
bad field type tp_feature_req_word of tpacket_req3: rust: 0x4002740a2c != c 0x4002740a28
size of struct tpacket_req3 is 28 in C and 32 in Rust
bad tpacket_stats_v3 size: rust: 16 (0x10) != c 12 (0xc)
bad field offset tp_freeze_q_cnt of tpacket_stats_v3: rust: 12 (0xc) != c 8 (0x8)
bad field type tp_freeze_q_cnt of tpacket_stats_v3: rust: 0x4002740b6c != c 0x4002740b68
size of struct tpacket_stats_v3 is 12 in C and 16 in Rust
bad tpacket3_hdr size: rust: 52 (0x34) != c 48 (0x30)
bad field offset tp_padding of tpacket3_hdr: rust: 44 (0x2c) != c 40 (0x28)
bad field type tp_padding of tpacket3_hdr: rust: 0x400274097c != c 0x4002740978
size of struct tpacket3_hdr is 48 in C and 52 in Rust

@rustbot author

[JohnTitor]

Closing as inactive, feel free to re-submit a PR if you're still interested in this change. Thanks anyway!

[nathaniel-bennett]

I kept running into struct alignment issues on 64-bit big-endian architectures (currently being tracked at rust-lang/rust#43894); these would lead to bors failing. Once rust-lang resolves those, I'll resubmit a PR and get this functionality merged in.