initial qnx vlan support #1931

MarcelJordense · 2024-02-08T08:38:04Z

Include the option to use raw ethernet transport and on QNX

src/core/ddsi/src/ddsi_raweth.c

Signed-off-by: Marcel Jordense <marcel@zettascale.tech>

eboasson

Thanks @MarcelJordense, I have only done a quick review, but it generally looks fine. I have a few minor comments, perhaps you could take a look?

I am curious to try it on macOS, but I haven't had a chance yet.

eboasson · 2024-02-22T16:11:23Z

src/core/ddsi/src/ddsi_raweth.c

+#if defined(__FreeBSD__) || defined(__APPLE__)
+#include <net/ethernet.h>
+#elif defined(__QNXNTO__)
+#define DDSI_BPF_IS_CONING_DEV (1)


I suspect a typo here, that is meant to be DDSI_BPF_IS_CLONING_DEV

eboasson · 2024-02-22T16:11:48Z

src/core/ddsi/src/ddsi_raweth.c

+    return DDS_RETCODE_ERROR;
+  }
+
+#if defined(DDSI_BPF_IS_CONING_DEV)


If my suspicion on this above is correct, then this should also be DDSI_BPF_IS_CLONING_DEV.

eboasson · 2024-02-22T16:13:59Z

src/core/ddsi/src/ddsi_raweth.c

+      ddsrt_close (sock);
+      DDS_CWARNING (&fact->gv->logconfig, "ddsi_raweth_create_conn %s port %u could not set directiion ... retcode = %d\n", mcast ? "multicast" : "unicast", port, r);
+  }
+#elif defined(__QNXNTO__) || defined(__APPLLE__)


Suggested change

#elif defined(__QNXNTO__) || defined(__APPLLE__)

#elif defined(__QNXNTO__) || defined(__APPLE__)

eboasson · 2024-02-22T16:18:53Z

src/core/ddsi/src/ddsi_raweth.c

+        port = (uint32_t)(ntohs (eth_hdr->proto));
+        if (vtag)
+        {
+          port += ((uint32_t)(vtag->tag & 0xfff) << 20) + ((uint32_t)(vtag->tag & 0xf000) << 16);


I think it makes sense not to duplicate this. I have no problem with the magic numbers and the bit fiddling, but it is a fiddly bit of code and it is nice to have only one copy to check.

Setting the source locator is also the same, I'd also de-dup that.

eboasson · 2024-02-22T16:20:39Z

src/core/ddsi/src/ddsi_raweth.c

+  ushort etype = (ushort)(port & 0xFFFF);
+  struct bpf_insn insns[] = {
+    BPF_STMT(BPF_LD+BPF_H+BPF_ABS, 12),
+	  BPF_JUMP(BPF_JMP+BPF_JEQ+BPF_K, etype, 3, 0),


Suggested change

BPF_JUMP(BPF_JMP+BPF_JEQ+BPF_K, etype, 3, 0),

BPF_JUMP(BPF_JMP+BPF_JEQ+BPF_K, etype, 3, 0),

eboasson · 2024-02-22T16:29:45Z

src/core/ddsi/src/ddsi_raweth.c

+    uc->bptr = uc->buffer;
+  }
+
+  if (uc->bptr < uc->buffer + uc->avail)


For all the bptr manipulations/checks: these structures come straight from the OS kernel, so I suppose we can trust them. I do think it would be good to point out in a comment that this is the reason there is so little checking, to help any future reader.

eboasson · 2024-02-22T16:31:10Z

src/core/ddsi/src/ddsi_raweth.c

+  memcpy(vhdr.e.dmac, dst->address + 10, 6);
+  memcpy(vhdr.e.smac, uc->m_base.m_base.gv->interfaces[0].loc.address + 10, 6);
+
+  if (vtag) {


Constructing the ethernet header also looks like a good candidate for de-duping.

eboasson · 2024-02-22T16:31:37Z

src/core/ddsi/src/ddsi_raweth.c

+    return DDS_RETCODE_ERROR;
+  }
+
+   buflen = gv->config.socket_rcvbuf_size.max.value;


Suggested change

buflen = gv->config.socket_rcvbuf_size.max.value;

buflen = gv->config.socket_rcvbuf_size.max.value;

eboasson · 2024-02-22T16:32:18Z

src/core/ddsi/src/ddsi_raweth.c

+      ddsrt_close (sock);
+      DDS_CWARNING (&fact->gv->logconfig, "ddsi_raweth_create_conn %s port %u could not set directiion ... retcode = %d\n", mcast ? "multicast" : "unicast", port, r);


Suggested change

ddsrt_close (sock);

DDS_CWARNING (&fact->gv->logconfig, "ddsi_raweth_create_conn %s port %u could not set directiion ... retcode = %d\n", mcast ? "multicast" : "unicast", port, r);

ddsrt_close (sock);

DDS_CWARNING (&fact->gv->logconfig, "ddsi_raweth_create_conn %s port %u could not set direction ... retcode = %d\n", mcast ? "multicast" : "unicast", port, r);

eboasson · 2024-02-22T16:32:33Z

src/core/ddsi/src/ddsi_raweth.c

+      ddsrt_close (sock);
+      DDS_CWARNING (&fact->gv->logconfig, "ddsi_raweth_create_conn %s port %u could not set directiion ... retcode = %d\n", mcast ? "multicast" : "unicast", port, r);


Suggested change

ddsrt_close (sock);

DDS_CWARNING (&fact->gv->logconfig, "ddsi_raweth_create_conn %s port %u could not set directiion ... retcode = %d\n", mcast ? "multicast" : "unicast", port, r);

ddsrt_close (sock);

DDS_CWARNING (&fact->gv->logconfig, "ddsi_raweth_create_conn %s port %u could not set direction ... retcode = %d\n", mcast ? "multicast" : "unicast", port, r);

Signed-off-by: Marcel Jordense <marcel@zettascale.tech>

eboasson

Thanks @MarcelJordense ☺️ Just a few tiny remarks left

eboasson · 2024-03-04T10:20:33Z

src/core/ddsi/src/ddsi_raweth.c

@@ -121,7 +151,7 @@ static ssize_t ddsi_raweth_conn_read (struct ddsi_tran_conn * conn, unsigned cha
  struct tpacket_auxdata *pauxd;
  struct cmsghdr *cptr;
  uint16_t vtag = 0;
-  uint32_t port;
+//  uint32_t port;


Suggested change

// uint32_t port;

eboasson · 2024-03-04T10:20:43Z

src/core/ddsi/src/ddsi_raweth.c

@@ -200,7 +223,7 @@ static ssize_t ddsi_raweth_conn_write (struct ddsi_tran_conn * conn, const ddsi_
  struct msghdr msg;
  struct sockaddr_ll dstaddr;
  struct ddsi_vlan_header vhdr;
-  uint16_t vtag;
+//  uint16_t vtag;


Suggested change

// uint16_t vtag;

eboasson · 2024-03-04T10:44:16Z

src/core/ddsi/src/ddsi_raweth.c

+static void set_locator(ddsi_locator_t *srcloc, const uint8_t * addr, uint16_t port, uint16_t vtag)
+{
+  srcloc->kind = DDSI_LOCATOR_KIND_RAWETH;
+  srcloc->port = (uint32_t)(port +  ((vtag & 0xfff) << 20) + ((vtag & 0xf000) << 4));


I'd prefer it if port and vtag were cast to a uint32_t before doing the arithmetic, so that all of it only involves unsigned 32-bit integers. I think the C99 section 6.5.7 "Bitwise shift operators" even says that if we are not careful we will get ourselves into trouble:

The integer promotions are performed on each of the operands. The type of the result is that of the promoted left operand.
The result of E1 << E2 is E1 left-shifted E2 bit positions; vacated bits are filled with zeros.
If E1 has a signed type and nonnegative value, and E1 × 2^E2 is representable in the result type, then that is the resulting value; otherwise, the behavior is undefined.

So if vtag & 0xfff ≥ 0x800, so say 0x800 = 2^11, then (vtag & 0xfff) << 20 = 2^11 × 2^20 = 2^31, and 2^31 isn't representable in a signed 32 bit int. That would make it undefined behaviour.

It is quite possible that this hasn't been introduced by the refactoring and that I just happened to spot it now. Even if that's the case, I think it makes sense to fix it now.

Suggested change

srcloc->port = (uint32_t)(port + ((vtag & 0xfff) << 20) + ((vtag & 0xf000) << 4));

srcloc->port = (uint32_t)port + (((uint32_t)vtag & 0xfff) << 20) + (((uint32_t)vtag & 0xf000) << 4));

eboasson · 2024-03-04T10:55:22Z

src/core/ddsi/src/ddsi_raweth.c

@@ -307,6 +331,7 @@ static dds_return_t ddsi_raweth_create_conn (struct ddsi_tran_conn **conn_out, s
  memset(&addr, 0, sizeof(addr));
  addr.sll_family = AF_PACKET;
  addr.sll_protocol = htons(ETH_P_ALL);
+


Suggested change

eboasson · 2024-03-04T10:57:55Z

src/core/ddsi/src/ddsi_raweth.c

-      memset(srcloc->address, 0, 10);
-      memcpy(srcloc->address + 10, src.sll_addr, 6);
-    }
+      set_locator(srcloc, src.sll_addr, ntohs (src.sll_protocol), vtag);


Is it really the case that vtag a.k.a. pauxd->tp_vlan_tci is already in native byte order? I'm asking only because it seems somewhat against the pattern, but it is not impossible.

I this case the vlan tag is already in host order

eboasson · 2024-03-04T10:58:44Z

src/core/ddsi/src/ddsi_raweth.c

+      {
+        memcpy(buf, ptr, (size_t)ret);
+        if (srcloc)
+          set_locator(srcloc, eth_hdr->smac, ntohs (eth_hdr->proto), (vtag ? vtag->tag : 0));


Like in the Linux version, on consideration I am somewhat surprised that the tag would be in native byte order already. Perhaps you can double-check?

In this case the vtag is of course in network order. I will correct this.

Signed-off-by: Marcel Jordense <marcel@zettascale.tech>

eboasson

Thanks @MarcelJordense !

While I don't like merging when there are CI failures, these are definitely unrelated and so it makes no sense to not merge this PR now.

github-advanced-security bot found potential problems Feb 8, 2024

View reviewed changes

src/core/ddsi/src/ddsi_raweth.c Fixed Show fixed Hide fixed

MarcelJordense force-pushed the qnx_vlan_support branch from 5db253b to b67ab2d Compare February 8, 2024 08:46

MarcelJordense added 3 commits February 22, 2024 13:42

initial qnx vlan support

f3589c3

Signed-off-by: Marcel Jordense <marcel@zettascale.tech>

resolve incorrect declared global variable

461bf6b

remove some compiler warnings

b9a6397

MarcelJordense force-pushed the qnx_vlan_support branch from 31c0530 to b9a6397 Compare February 22, 2024 12:44

eboasson reviewed Feb 22, 2024

View reviewed changes

process review comments, remove duplicate code

b67a2e1

Signed-off-by: Marcel Jordense <marcel@zettascale.tech>

eboasson reviewed Mar 4, 2024

View reviewed changes

convert vlan tag from network to host order

6cee3dc

Signed-off-by: Marcel Jordense <marcel@zettascale.tech>

eboasson approved these changes Mar 5, 2024

View reviewed changes

eboasson merged commit 8acd62b into eclipse-cyclonedds:master Mar 5, 2024
17 of 23 checks passed

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

initial qnx vlan support #1931

initial qnx vlan support #1931

MarcelJordense commented Feb 8, 2024

eboasson left a comment

eboasson Feb 22, 2024

eboasson Feb 22, 2024

eboasson Feb 22, 2024

eboasson Feb 22, 2024

eboasson Feb 22, 2024

eboasson Feb 22, 2024

eboasson Feb 22, 2024

eboasson Feb 22, 2024

eboasson Feb 22, 2024

eboasson Feb 22, 2024

eboasson left a comment

eboasson Mar 4, 2024

eboasson Mar 4, 2024

eboasson Mar 4, 2024

eboasson Mar 4, 2024

eboasson Mar 4, 2024

MarcelJordense Mar 4, 2024

eboasson Mar 4, 2024

MarcelJordense Mar 4, 2024

eboasson left a comment

	#elif defined(__QNXNTO__) \|\| defined(__APPLLE__)
	#elif defined(__QNXNTO__) \|\| defined(__APPLE__)

	BPF_JUMP(BPF_JMP+BPF_JEQ+BPF_K, etype, 3, 0),
	BPF_JUMP(BPF_JMP+BPF_JEQ+BPF_K, etype, 3, 0),

	buflen = gv->config.socket_rcvbuf_size.max.value;
	buflen = gv->config.socket_rcvbuf_size.max.value;

		ddsrt_close (sock);
		DDS_CWARNING (&fact->gv->logconfig, "ddsi_raweth_create_conn %s port %u could not set directiion ... retcode = %d\n", mcast ? "multicast" : "unicast", port, r);

	srcloc->port = (uint32_t)(port + ((vtag & 0xfff) << 20) + ((vtag & 0xf000) << 4));
	srcloc->port = (uint32_t)port + (((uint32_t)vtag & 0xfff) << 20) + (((uint32_t)vtag & 0xf000) << 4));

initial qnx vlan support #1931

initial qnx vlan support #1931

Conversation

MarcelJordense commented Feb 8, 2024

eboasson left a comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

eboasson left a comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

eboasson left a comment

Choose a reason for hiding this comment