From c2c6c3c8c66855d9ccbe079015355ae8db8ff5e5 Mon Sep 17 00:00:00 2001
From: Kaz Kylheku <kaz@kylheku.com>
Date: Thu, 31 Mar 2016 21:33:32 -0700
Subject: Bugfix: support abstract UNIX socket addresses on Linux.

Making it work as already documented.

* socket.c (MIN): New macro.
(sockaddr_pack): Use utf8_dup_to_buf to convert
Unix socket path to a buffer of UTF-8 bytes, possibly with one
or more embedded null bytes. Copy as much of this as fits into
the sun_path member of struct sockaddr_un.

* txr.1: Improve documentation about the abstract names
on Linux.
---
 socket.c |  7 +++++--
 txr.1    | 38 ++++++++++++++++++++++++++++++++++----
 2 files changed, 39 insertions(+), 6 deletions(-)

diff --git a/socket.c b/socket.c
index 39c4d931..e90d374e 100644
--- a/socket.c
+++ b/socket.c
@@ -51,6 +51,8 @@
 #include "arith.h"
 #include "socket.h"
 
+#define MIN(A, B) ((A) < (B) ? (A) : (B))
+
 struct dgram_stream {
   struct strm_base a;
   val stream;
@@ -244,11 +246,12 @@ static void sockaddr_pack(val sockaddr, val family,
     *len = sizeof *sa;
   } else if (addr_type == sockaddr_un_s) {
     val path = slot(sockaddr, path_s);
-    char *path_u8 = utf8_dup_to(c_str(path));
     struct sockaddr_un *sa = coerce(struct sockaddr_un *, buf);
+    size_t size;
+    unsigned char *path_u8 = utf8_dup_to_buf(c_str(path), &size, 0);
     memset(sa, 0, sizeof *sa);
     sa->sun_family = AF_UNIX;
-    strncpy(sa->sun_path, path_u8, sizeof sa->sun_path - 1);
+    memcpy(sa->sun_path, path_u8, MIN(size, sizeof sa->sun_path));
     free(path_u8);
     *len = sizeof *sa;
   } else {
diff --git a/txr.1 b/txr.1
index c2f9e06d..01cd4ea6 100644
--- a/txr.1
+++ b/txr.1
@@ -37932,10 +37932,40 @@ address family.
 This structure has only one slot,
 .code path
 which holds the rendezvous name for connecting pairs of socket endpoints.
-This names appears in the filesystem. Linux systems have support for "abstract"
-names which do not appear in the filesystem. These abstract names are distinguished
-by starting with a null byte. In \*(TX, a null byte can be encoded in a string
-using the pseudo-null code point U+DC00.
+This name appears in the filesystem.
+
+When the
+.code sockaddr-un
+structure is converted to the C structure
+.code "struct sockaddr_un" ,
+the
+.code path
+slot undergoes conversion to UTF-8. The resulting bytes are stored in the
+.code sun_path
+member of the C structure. If the resulting UTF-8 byte string
+is larger than the
+.code sun_path
+array, it is silently truncated.
+
+Note: Linux systems have support for "abstract" names which do not appear in
+the filesystem. These abstract names are distinguished by starting with a null
+byte. For more information, consult Linux documentation.
+This convention is supported in the
+.code path
+slot of the
+.code sockaddr-un
+structure. If
+.code path
+contains occurrences of the pseudo-null character U+DC00, these translate
+to null bytes in the
+.code sun_path
+member of the corresponding C structure
+.codn "struct sockaddr_un" .
+For example, the path
+.str "\exDC00;foo"
+is valid and represents an abstract address consisting of the three bytes
+.str "foo"
+followed by null padding bytes.
 
 .coNP Structure @ addrinfo
 .synb
-- 
cgit v1.2.3