Commit 67058b9b70

Andrew Kelley <andrew@ziglang.org>
2019-10-29 07:19:22
basic DNS address resolution for linux without libc
1 parent d5865f5
lib/std/event/loop.zig
@@ -466,6 +466,10 @@ pub const Loop = struct {
         return self.linuxWaitFd(fd, os.EPOLLET | os.EPOLLIN);
     }
 
+    pub fn waitUntilFdWritable(self: *Loop, fd: os.fd_t) !void {
+        return self.linuxWaitFd(fd, os.EPOLLET | os.EPOLLOUT);
+    }
+
     pub async fn bsdWaitKev(self: *Loop, ident: usize, filter: i16, fflags: u32) !os.Kevent {
         var resume_node = ResumeNode.Basic{
             .base = ResumeNode{
lib/std/fs/file.zig
@@ -41,7 +41,7 @@ pub const File = struct {
             const path_w = try windows.cStrToPrefixedFileW(path);
             return openReadW(&path_w);
         }
-        const flags = os.O_LARGEFILE | os.O_RDONLY;
+        const flags = os.O_LARGEFILE | os.O_RDONLY | os.O_CLOEXEC;
         const fd = try os.openC(path, flags, 0);
         return openHandle(fd);
     }
lib/std/os/bits/linux.zig
@@ -1431,3 +1431,23 @@ pub const IPPROTO_UDPLITE = 136;
 pub const IPPROTO_MPLS = 137;
 pub const IPPROTO_RAW = 255;
 pub const IPPROTO_MAX = 256;
+
+pub const RR_A = 1;
+pub const RR_CNAME = 5;
+pub const RR_AAAA = 28;
+
+pub const nfds_t = usize;
+pub const pollfd = extern struct {
+    fd: fd_t,
+    events: i16,
+    revents: i16,
+};
+
+pub const POLLIN = 0x001;
+pub const POLLPRI = 0x002;
+pub const POLLOUT = 0x004;
+pub const POLLERR = 0x008;
+pub const POLLHUP = 0x010;
+pub const POLLNVAL = 0x020;
+pub const POLLRDNORM = 0x040;
+pub const POLLRDBAND = 0x080;
lib/std/os/linux.zig
@@ -226,6 +226,28 @@ pub fn munmap(address: [*]const u8, length: usize) usize {
     return syscall2(SYS_munmap, @ptrToInt(address), length);
 }
 
+pub fn poll(fds: [*]pollfd, n: nfds_t, timeout: i32) usize {
+    if (@hasDecl(@This(), "SYS_poll")) {
+        return syscall3(SYS_poll, @ptrToInt(fds), n, @bitCast(u32, timeout));
+    } else {
+        return syscall6(
+            SYS_ppoll,
+            @ptrToInt(fds),
+            n,
+            @ptrToInt(if (timeout >= 0)
+                &timespec{
+                    .tv_sec = timeout / 1000,
+                    .tv_nsec = (timeout % 1000) * 1000000,
+                }
+            else
+                null),
+            0,
+            0,
+            NSIG / 8,
+        );
+    }
+}
+
 pub fn read(fd: i32, buf: [*]u8, count: usize) usize {
     return syscall3(SYS_read, @bitCast(usize, isize(fd)), @ptrToInt(buf), count);
 }
lib/std/buffer.zig
@@ -72,11 +72,11 @@ pub const Buffer = struct {
         self.list.deinit();
     }
 
-    pub fn toSlice(self: *const Buffer) []u8 {
+    pub fn toSlice(self: Buffer) []u8 {
         return self.list.toSlice()[0..self.len()];
     }
 
-    pub fn toSliceConst(self: *const Buffer) []const u8 {
+    pub fn toSliceConst(self: Buffer) []const u8 {
         return self.list.toSliceConst()[0..self.len()];
     }
 
@@ -91,11 +91,11 @@ pub const Buffer = struct {
         self.list.items[self.len()] = 0;
     }
 
-    pub fn isNull(self: *const Buffer) bool {
+    pub fn isNull(self: Buffer) bool {
         return self.list.len == 0;
     }
 
-    pub fn len(self: *const Buffer) usize {
+    pub fn len(self: Buffer) usize {
         return self.list.len - 1;
     }
 
@@ -111,16 +111,16 @@ pub const Buffer = struct {
         self.list.toSlice()[old_len] = byte;
     }
 
-    pub fn eql(self: *const Buffer, m: []const u8) bool {
+    pub fn eql(self: Buffer, m: []const u8) bool {
         return mem.eql(u8, self.toSliceConst(), m);
     }
 
-    pub fn startsWith(self: *const Buffer, m: []const u8) bool {
+    pub fn startsWith(self: Buffer, m: []const u8) bool {
         if (self.len() < m.len) return false;
         return mem.eql(u8, self.list.items[0..m.len], m);
     }
 
-    pub fn endsWith(self: *const Buffer, m: []const u8) bool {
+    pub fn endsWith(self: Buffer, m: []const u8) bool {
         const l = self.len();
         if (l < m.len) return false;
         const start = l - m.len;
@@ -133,7 +133,7 @@ pub const Buffer = struct {
     }
 
     /// For passing to C functions.
-    pub fn ptr(self: *const Buffer) [*]u8 {
+    pub fn ptr(self: Buffer) [*]u8 {
         return self.list.items.ptr;
     }
 };
lib/std/c.zig
@@ -116,6 +116,26 @@ pub extern "c" fn getsockname(sockfd: fd_t, noalias addr: *sockaddr, noalias add
 pub extern "c" fn connect(sockfd: fd_t, sock_addr: *const sockaddr, addrlen: socklen_t) c_int;
 pub extern "c" fn accept4(sockfd: fd_t, addr: *sockaddr, addrlen: *socklen_t, flags: c_uint) c_int;
 pub extern "c" fn getsockopt(sockfd: fd_t, level: c_int, optname: c_int, optval: *c_void, optlen: *socklen_t) c_int;
+pub extern "c" fn send(sockfd: fd_t, buf: *const c_void, len: usize, flags: u32) isize;
+pub extern "c" fn sendto(
+    sockfd: fd_t,
+    buf: *const c_void,
+    len: usize,
+    flags: u32,
+    dest_addr: *const sockaddr,
+    addrlen: socklen_t,
+) isize;
+
+pub extern fn recv(sockfd: fd_t, arg1: ?*c_void, arg2: usize, arg3: c_int) isize;
+pub extern fn recvfrom(
+    sockfd: fd_t,
+    noalias buf: *c_void,
+    len: usize,
+    flags: u32,
+    noalias src_addr: ?*sockaddr,
+    noalias addrlen: ?*socklen_t,
+) isize;
+
 pub extern "c" fn kill(pid: pid_t, sig: c_int) c_int;
 pub extern "c" fn getdirentries(fd: fd_t, buf_ptr: [*]u8, nbytes: usize, basep: *i64) isize;
 pub extern "c" fn setgid(ruid: c_uint, euid: c_uint) c_int;
@@ -169,3 +189,5 @@ pub extern "c" fn getnameinfo(
 ) c_int;
 
 pub extern "c" fn gai_strerror(errcode: c_int) [*]const u8;
+
+pub extern "c" fn poll(fds: [*]pollfd, nfds: nfds_t, timeout: c_int) c_int;
lib/std/fs.zig
@@ -704,7 +704,7 @@ pub const Dir = struct {
 
     /// Call `File.close` on the result when done.
     pub fn openReadC(self: Dir, sub_path: [*]const u8) File.OpenError!File {
-        const flags = os.O_LARGEFILE | os.O_RDONLY;
+        const flags = os.O_LARGEFILE | os.O_RDONLY | os.O_CLOEXEC;
         const fd = try os.openatC(self.fd, sub_path, flags, 0);
         return File.openHandle(fd);
     }
lib/std/net.zig
@@ -4,6 +4,7 @@ const assert = std.debug.assert;
 const net = @This();
 const mem = std.mem;
 const os = std.os;
+const fs = std.fs;
 
 pub const TmpWinAddr = struct {
     family: u8,
@@ -285,7 +286,7 @@ test "std.net.parseIp6" {
     std.testing.expect(mem.eql(u8, "[ff01::fb]:80", printed));
 }
 
-pub fn connectUnixSocket(path: []const u8) !std.fs.File {
+pub fn connectUnixSocket(path: []const u8) !fs.File {
     const opt_non_block = if (std.event.Loop.instance != null) os.SOCK_NONBLOCK else 0;
     const sockfd = try os.socket(
         os.AF_UNIX,
@@ -312,7 +313,7 @@ pub fn connectUnixSocket(path: []const u8) !std.fs.File {
         try os.connect(sockfd, &sock_addr, size);
     }
 
-    return std.fs.File.openHandle(sockfd);
+    return fs.File.openHandle(sockfd);
 }
 
 pub const AddressList = struct {
@@ -356,7 +357,7 @@ pub fn getAddressList(allocator: *mem.Allocator, name: []const u8, port: u16) !*
 
         const hints = os.addrinfo{
             .flags = c.AI_NUMERICSERV,
-            .family = os.AF_UNSPEC,
+            .family = os.AF_INET, // TODO os.AF_UNSPEC,
             .socktype = os.SOCK_STREAM,
             .protocol = os.IPPROTO_TCP,
             .canonname = null,
@@ -413,40 +414,41 @@ pub fn getAddressList(allocator: *mem.Allocator, name: []const u8, port: u16) !*
         return result;
     }
     if (builtin.os == .linux) {
-        const flags = os.AI_NUMERICSERV;
+        const flags = std.c.AI_NUMERICSERV;
         const family = os.AF_INET; //TODO os.AF_UNSPEC;
         // The limit of 48 results is a non-sharp bound on the number of addresses
         // that can fit in one 512-byte DNS packet full of v4 results and a second
         // packet full of v6 results. Due to headers, the actual limit is lower.
-        var buf: [48]LookupAddr = undefined;
-        var canon_buf: [256]u8 = undefined;
-        var canon_len: usize = 0;
-        const cnt = try linuxLookupName(buf[0..], &canon_buf, &canon_len, name, family, flags);
+        var addrs = std.ArrayList(LookupAddr).init(allocator);
+        defer addrs.deinit();
 
-        result.addrs = try arena.alloc(Address, cnt);
+        var canon = std.Buffer.initNull(allocator);
+        defer canon.deinit();
 
-        if (canon_len != 0) {
-            result.canon_name = try mem.dupe(arena, u8, canon_buf[0..canon_len]);
+        try linuxLookupName(&addrs, &canon, name, family, flags);
+
+        result.addrs = try arena.alloc(Address, addrs.len);
+        if (!canon.isNull()) {
+            result.canon_name = canon.toOwnedSlice();
         }
 
-        var i: usize = 0;
-        while (i < cnt) : (i += 1) {
-            const os_addr = if (buf[i].family == os.AF_INET6)
+        for (addrs.toSliceConst()) |addr, i| {
+            const os_addr = if (addr.family == os.AF_INET6)
                 os.sockaddr{
                     .in6 = os.sockaddr_in6{
-                        .family = buf[i].family,
+                        .family = addr.family,
                         .port = mem.nativeToBig(u16, port),
                         .flowinfo = 0,
-                        .addr = buf[i].addr,
-                        .scope_id = buf[i].scope_id,
+                        .addr = addr.addr,
+                        .scope_id = addr.scope_id,
                     },
                 }
             else
                 os.sockaddr{
                     .in = os.sockaddr_in{
-                        .family = buf[i].family,
+                        .family = addr.family,
                         .port = mem.nativeToBig(u16, port),
-                        .addr = @ptrCast(*align(1) u32, &buf[i].addr).*,
+                        .addr = @ptrCast(*align(1) const u32, &addr.addr).*,
                         .zero = [8]u8{ 0, 0, 0, 0, 0, 0, 0, 0 },
                     },
                 };
@@ -466,51 +468,52 @@ const LookupAddr = struct {
 };
 
 fn linuxLookupName(
-    buf: []LookupAddr,
-    canon_buf: []u8,
-    canon_len: *usize,
+    addrs: *std.ArrayList(LookupAddr),
+    canon: *std.Buffer,
     opt_name: ?[]const u8,
     family: i32,
     flags: u32,
-) !usize {
-    var cnt: usize = 0;
+) !void {
     if (opt_name) |name| {
         // reject empty name and check len so it fits into temp bufs
-        if (name.len >= 254) return error.UnknownName;
-        mem.copy(u8, canon_buf, name);
-        canon_len.* = name.len;
-
-        cnt = (linuxLookupNameFromNumeric(buf, name, family) catch |err| switch (err) {
-            error.ExpectedIPv6ButFoundIPv4 => unreachable,
-            error.ExpectedIPv4ButFoundIPv6 => unreachable,
-        });
-        if (cnt == 0 and (flags & os.AI_NUMERICHOST) == 0) {
-            cnt = try linuxLookupNameFromHosts(buf, canon_buf, canon_len, name, family);
-            if (cnt == 0) {
-                cnt = try linuxLookupNameFromDnsSearch(buf, canon_buf, canon_len, name, family);
+        try canon.replaceContents(name);
+        try linuxLookupNameFromNumeric(addrs, name, family);
+        if (addrs.len == 0 and (flags & std.c.AI_NUMERICHOST) == 0) {
+            try linuxLookupNameFromHosts(addrs, canon, name, family);
+            if (addrs.len == 0) {
+                try linuxLookupNameFromDnsSearch(addrs, canon, name, family);
             }
         }
     } else {
-        canon_len.* = 0;
-        cnt = linuxLookupNameFromNull(buf, family, flags);
+        try canon.resize(0);
+        try linuxLookupNameFromNull(addrs, family, flags);
     }
-    if (cnt == 0) return error.UnknownName;
+    if (addrs.len == 0) return error.UnknownName;
 
     // No further processing is needed if there are fewer than 2
     // results or if there are only IPv4 results.
-    if (cnt == 1 or family == os.AF_INET) return cnt;
+    if (addrs.len == 1 or family == os.AF_INET) return;
 
     @panic("port the RFC 3484/6724 destination address selection from musl libc");
 }
 
-fn linuxLookupNameFromNumeric(buf: []LookupAddr, name: []const u8, family: i32) !usize {
+fn linuxLookupNameFromNumericUnspec(addrs: *std.ArrayList(LookupAddr), name: []const u8) !void {
+    return linuxLookupNameFromNumeric(addrs, name, os.AF_UNSPEC) catch |err| switch (err) {
+        error.ExpectedIPv6ButFoundIPv4 => unreachable,
+        error.ExpectedIPv4ButFoundIPv6 => unreachable,
+        else => |e| return e,
+    };
+}
+
+fn linuxLookupNameFromNumeric(addrs: *std.ArrayList(LookupAddr), name: []const u8, family: i32) !void {
     if (parseIp4(name)) |ip4| {
         if (family == os.AF_INET6) return error.ExpectedIPv6ButFoundIPv4;
+        const item = try addrs.addOne();
         // TODO [0..4] should return *[4]u8, making this pointer cast unnecessary
-        mem.writeIntNative(u32, @ptrCast(*[4]u8, &buf[0].addr), ip4);
-        buf[0].family = os.AF_INET;
-        buf[0].scope_id = 0;
-        return 1;
+        mem.writeIntNative(u32, @ptrCast(*[4]u8, &item.addr), ip4);
+        item.family = os.AF_INET;
+        item.scope_id = 0;
+        return;
     } else |err| switch (err) {
         error.Overflow,
         error.InvalidEnd,
@@ -521,10 +524,11 @@ fn linuxLookupNameFromNumeric(buf: []LookupAddr, name: []const u8, family: i32)
 
     if (parseIp6(name)) |ip6| {
         if (family == os.AF_INET) return error.ExpectedIPv4ButFoundIPv6;
-        @memcpy(&buf[0].addr, &ip6.addr, 16);
-        buf[0].family = os.AF_INET6;
-        buf[0].scope_id = ip6.scope_id;
-        return 1;
+        const item = try addrs.addOne();
+        @memcpy(&item.addr, &ip6.addr, 16);
+        item.family = os.AF_INET6;
+        item.scope_id = ip6.scope_id;
+        return;
     } else |err| switch (err) {
         error.Overflow,
         error.InvalidEnd,
@@ -532,64 +536,54 @@ fn linuxLookupNameFromNumeric(buf: []LookupAddr, name: []const u8, family: i32)
         error.Incomplete,
         => {},
     }
-
-    return 0;
 }
 
-fn linuxLookupNameFromNull(buf: []LookupAddr, family: i32, flags: u32) usize {
-    var cnt: usize = 0;
-    if ((flags & os.AI_PASSIVE) != 0) {
+fn linuxLookupNameFromNull(addrs: *std.ArrayList(LookupAddr), family: i32, flags: u32) !void {
+    if ((flags & std.c.AI_PASSIVE) != 0) {
         if (family != os.AF_INET6) {
-            buf[cnt] = LookupAddr{
+            (try addrs.addOne()).* = LookupAddr{
                 .family = os.AF_INET,
                 .addr = [1]u8{0} ** 16,
             };
-            cnt += 1;
         }
         if (family != os.AF_INET) {
-            buf[cnt] = LookupAddr{
+            (try addrs.addOne()).* = LookupAddr{
                 .family = os.AF_INET6,
                 .addr = [1]u8{0} ** 16,
             };
-            cnt += 1;
         }
     } else {
         if (family != os.AF_INET6) {
-            buf[cnt] = LookupAddr{
+            (try addrs.addOne()).* = LookupAddr{
                 .family = os.AF_INET,
                 .addr = [4]u8{ 127, 0, 0, 1 } ++ ([1]u8{0} ** 12),
             };
-            cnt += 1;
         }
         if (family != os.AF_INET) {
-            buf[cnt] = LookupAddr{
+            (try addrs.addOne()).* = LookupAddr{
                 .family = os.AF_INET6,
                 .addr = ([1]u8{0} ** 15) ++ [1]u8{1},
             };
-            cnt += 1;
         }
     }
-    return cnt;
 }
 
 fn linuxLookupNameFromHosts(
-    buf: []LookupAddr,
-    canon_buf: []u8,
-    canon_len: *usize,
+    addrs: *std.ArrayList(LookupAddr),
+    canon: *std.Buffer,
     name: []const u8,
     family: i32,
-) !usize {
-    const file = std.fs.File.openReadC(c"/etc/hosts") catch |err| switch (err) {
+) !void {
+    const file = fs.File.openReadC(c"/etc/hosts") catch |err| switch (err) {
         error.FileNotFound,
         error.NotDir,
         error.AccessDenied,
-        => return 0,
+        => return,
         else => |e| return e,
     };
     defer file.close();
 
-    var cnt: usize = 0;
-    const stream = &std.io.BufferedInStream(std.fs.File.ReadError).init(&file.inStream().stream).stream;
+    const stream = &std.io.BufferedInStream(fs.File.ReadError).init(&file.inStream().stream).stream;
     var line_buf: [512]u8 = undefined;
     while (stream.readUntilDelimiterOrEof(&line_buf, '\n') catch |err| switch (err) {
         error.StreamTooLong => blk: {
@@ -612,26 +606,20 @@ fn linuxLookupNameFromHosts(
             }
         } else continue;
 
-        switch (linuxLookupNameFromNumeric(buf[cnt..], ip_text, family) catch |err| switch (err) {
+        const prev_len = addrs.len;
+        linuxLookupNameFromNumeric(addrs, ip_text, family) catch |err| switch (err) {
             error.ExpectedIPv6ButFoundIPv4 => continue,
             error.ExpectedIPv4ButFoundIPv6 => continue,
-        }) {
-            0 => continue,
-            1 => {
-                // first name is canonical name
-                const name_text = first_name_text.?;
-                if (isValidHostName(name_text)) {
-                    mem.copy(u8, canon_buf, name_text);
-                    canon_len.* = name_text.len;
-                }
-
-                cnt += 1;
-                if (cnt == buf.len) break;
-            },
-            else => unreachable,
+            error.OutOfMemory => |e| return e,
+        };
+        if (addrs.len > prev_len) {
+            // first name is canonical name
+            const name_text = first_name_text.?;
+            if (isValidHostName(name_text)) {
+                try canon.replaceContents(name_text);
+            }
         }
     }
-    return cnt;
 }
 
 pub fn isValidHostName(hostname: []const u8) bool {
@@ -647,50 +635,414 @@ pub fn isValidHostName(hostname: []const u8) bool {
 }
 
 fn linuxLookupNameFromDnsSearch(
-    buf: []LookupAddr,
-    canon_buf: []u8,
-    canon_len: *usize,
+    addrs: *std.ArrayList(LookupAddr),
+    canon: *std.Buffer,
     name: []const u8,
     family: i32,
-) !usize {
-    var search: [256]u8 = undefined;
-    const resolv_conf = try getResolvConf(&search);
+) !void {
+    var rc: ResolvConf = undefined;
+    try getResolvConf(addrs.allocator, &rc);
+    defer rc.deinit();
 
     // Count dots, suppress search when >=ndots or name ends in
     // a dot, which is an explicit request for global scope.
-    //var dots: usize = 0;
-    //for (name) |byte| {
-    //    if (byte == '.') dots += 1;
-    //}
+    var dots: usize = 0;
+    for (name) |byte| {
+        if (byte == '.') dots += 1;
+    }
+
+    const search = if (rc.search.isNull() or dots >= rc.ndots or mem.endsWith(u8, name, "."))
+        [_]u8{}
+    else
+        rc.search.toSliceConst();
+
+    var canon_name = name;
+
+    // Strip final dot for canon, fail if multiple trailing dots.
+    if (mem.endsWith(u8, canon_name, ".")) canon_name.len -= 1;
+    if (mem.endsWith(u8, canon_name, ".")) return error.UnknownName;
+
+    // Name with search domain appended is setup in canon[]. This both
+    // provides the desired default canonical name (if the requested
+    // name is not a CNAME record) and serves as a buffer for passing
+    // the full requested name to name_from_dns.
+    try canon.resize(canon_name.len);
+    mem.copy(u8, canon.toSlice(), canon_name);
+    try canon.appendByte('.');
+
+    var tok_it = mem.tokenize(search, " \t");
+    while (tok_it.next()) |tok| {
+        canon.shrink(canon_name.len + 1);
+        try canon.append(tok);
+        try linuxLookupNameFromDns(addrs, canon, canon.toSliceConst(), family, rc);
+        if (addrs.len != 0) return;
+    }
+
+    canon.shrink(canon_name.len);
+    return linuxLookupNameFromDns(addrs, canon, name, family, rc);
+}
+
+const dpc_ctx = struct {
+    addrs: *std.ArrayList(LookupAddr),
+    canon: *std.Buffer,
+};
+
+fn linuxLookupNameFromDns(
+    addrs: *std.ArrayList(LookupAddr),
+    canon: *std.Buffer,
+    name: []const u8,
+    family: i32,
+    rc: ResolvConf,
+) !void {
+    var ctx = dpc_ctx{
+        .addrs = addrs,
+        .canon = canon,
+    };
+    const AfRr = struct {
+        af: i32,
+        rr: u8,
+    };
+    const afrrs = [_]AfRr{
+        AfRr{ .af = os.AF_INET6, .rr = os.RR_A },
+        AfRr{ .af = os.AF_INET, .rr = os.RR_AAAA },
+    };
+    var qbuf: [2][280]u8 = undefined;
+    var abuf: [2][512]u8 = undefined;
+    var qp: [2][]const u8 = undefined;
+    const apbuf = [2][]u8{ &abuf[0], &abuf[1] };
+    var nq: usize = 0;
+
+    for (afrrs) |afrr| {
+        if (family != afrr.af) {
+            const len = os.res_mkquery(0, name, 1, afrr.rr, [_]u8{}, null, &qbuf[nq]);
+            qp[nq] = qbuf[nq][0..len];
+            nq += 1;
+        }
+    }
+
+    var ap = [2][]u8{ apbuf[0][0..0], apbuf[1][0..0] };
+    try resMSendRc(qp[0..nq], ap[0..nq], apbuf[0..nq], rc);
+
+    var i: usize = 0;
+    while (i < nq) : (i += 1) {
+        dnsParse(ap[i], ctx, dnsParseCallback) catch {};
+    }
+
+    if (addrs.len != 0) return;
+    if (ap[0].len < 4 or (ap[0][3] & 15) == 2) return error.TemporaryNameServerFailure;
+    if ((ap[0][3] & 15) == 0) return error.UnknownName;
+    if ((ap[0][3] & 15) == 3) return;
+    return error.NameServerFailure;
+}
+
+const ResolvConf = struct {
+    attempts: u32,
+    ndots: u32,
+    timeout: u32,
+    search: std.Buffer,
+    ns: std.ArrayList(LookupAddr),
+
+    fn deinit(rc: *ResolvConf) void {
+        rc.ns.deinit();
+        rc.search.deinit();
+        rc.* = undefined;
+    }
+};
+
+/// Ignores lines longer than 512 bytes.
+/// TODO: https://github.com/ziglang/zig/issues/2765 and https://github.com/ziglang/zig/issues/2761
+fn getResolvConf(allocator: *mem.Allocator, rc: *ResolvConf) !void {
+    rc.* = ResolvConf{
+        .ns = std.ArrayList(LookupAddr).init(allocator),
+        .search = std.Buffer.initNull(allocator),
+        .ndots = 1,
+        .timeout = 5,
+        .attempts = 2,
+    };
+    errdefer rc.deinit();
+
+    const file = fs.File.openReadC(c"/etc/resolv.conf") catch |err| switch (err) {
+        error.FileNotFound,
+        error.NotDir,
+        error.AccessDenied,
+        => return linuxLookupNameFromNumericUnspec(&rc.ns, "127.0.0.1"),
+        else => |e| return e,
+    };
+    defer file.close();
+
+    var cnt: usize = 0;
+    const stream = &std.io.BufferedInStream(fs.File.ReadError).init(&file.inStream().stream).stream;
+    var line_buf: [512]u8 = undefined;
+    while (stream.readUntilDelimiterOrEof(&line_buf, '\n') catch |err| switch (err) {
+        error.StreamTooLong => blk: {
+            // Skip to the delimiter in the stream, to fix parsing
+            try stream.skipUntilDelimiterOrEof('\n');
+            // Give an empty line to the while loop, which will be skipped.
+            break :blk line_buf[0..0];
+        },
+        else => |e| return e,
+    }) |line| {
+        const no_comment_line = mem.separate(line, "#").next().?;
+        var line_it = mem.tokenize(no_comment_line, " \t");
+
+        const token = line_it.next() orelse continue;
+        if (mem.eql(u8, token, "options")) {
+            while (line_it.next()) |sub_tok| {
+                var colon_it = mem.separate(sub_tok, ":");
+                const name = colon_it.next().?;
+                const value_txt = colon_it.next() orelse continue;
+                const value = std.fmt.parseInt(u8, value_txt, 10) catch |err| switch (err) {
+                    error.Overflow => 255,
+                    error.InvalidCharacter => continue,
+                };
+                if (mem.eql(u8, name, "ndots")) {
+                    rc.ndots = std.math.min(value, 15);
+                } else if (mem.eql(u8, name, "attempts")) {
+                    rc.attempts = std.math.min(value, 10);
+                } else if (mem.eql(u8, name, "timeout")) {
+                    rc.timeout = std.math.min(value, 60);
+                }
+            }
+        } else if (mem.eql(u8, token, "nameserver")) {
+            const ip_txt = line_it.next() orelse continue;
+            try linuxLookupNameFromNumericUnspec(&rc.ns, ip_txt);
+        } else if (mem.eql(u8, token, "domain") or mem.eql(u8, token, "search")) {
+            try rc.search.replaceContents(line_it.rest());
+        }
+    }
+
+    if (rc.ns.len == 0) {
+        return linuxLookupNameFromNumericUnspec(&rc.ns, "127.0.0.1");
+    }
+}
+
+fn eqlSockAddr(a: *const os.sockaddr, b: *const os.sockaddr, len: usize) bool {
+    const a_bytes = @ptrCast([*]const u8, a)[0..len];
+    const b_bytes = @ptrCast([*]const u8, b)[0..len];
+    return mem.eql(u8, a_bytes, b_bytes);
+}
+
+fn resMSendRc(
+    queries: []const []const u8,
+    answers: [][]u8,
+    answer_bufs: []const []u8,
+    rc: ResolvConf,
+) !void {
+    const timeout = 1000 * rc.timeout;
+    const attempts = rc.attempts;
+
+    var sl: os.socklen_t = @sizeOf(os.sockaddr_in);
+    var family: os.sa_family_t = os.AF_INET;
+
+    var ns_list = std.ArrayList(os.sockaddr).init(rc.ns.allocator);
+    defer ns_list.deinit();
 
-    //if (dots >= conf.ndots || name[l-1]=='.') *search = 0;
-
-    //// Strip final dot for canon, fail if multiple trailing dots.
-    //if (name[l-1]=='.') l--;
-    //if (!l || name[l-1]=='.') return EAI_NONAME;
-
-    //// This can never happen; the caller already checked length.
-    //if (l >= 256) return EAI_NONAME;
-
-    //// Name with search domain appended is setup in canon[]. This both
-    //// provides the desired default canonical name (if the requested
-    //// name is not a CNAME record) and serves as a buffer for passing
-    //// the full requested name to name_from_dns.
-    //memcpy(canon, name, l);
-    //canon[l] = '.';
-
-    //for (p=search; *p; p=z) {
-    //    for (; isspace(*p); p++);
-    //    for (z=p; *z && !isspace(*z); z++);
-    //    if (z==p) break;
-    //    if (z-p < 256 - l - 1) {
-    //        memcpy(canon+l+1, p, z-p);
-    //        canon[z-p+1+l] = 0;
-    //        int cnt = name_from_dns(buf, canon, canon, family, &conf);
-    //        if (cnt) return cnt;
+    try ns_list.resize(rc.ns.len);
+    const ns = ns_list.toSlice();
+
+    for (rc.ns.toSliceConst()) |iplit, i| {
+        if (iplit.family == os.AF_INET) {
+            ns[i] = os.sockaddr{
+                .in = os.sockaddr_in{
+                    .family = os.AF_INET,
+                    .port = mem.nativeToBig(u16, 53),
+                    .addr = mem.readIntNative(u32, @ptrCast(*const [4]u8, &iplit.addr)),
+                    .zero = [8]u8{ 0, 0, 0, 0, 0, 0, 0, 0 },
+                },
+            };
+        } else {
+            ns[i] = os.sockaddr{
+                .in6 = os.sockaddr_in6{
+                    .family = os.AF_INET6,
+                    .port = mem.nativeToBig(u16, 53),
+                    .flowinfo = 0,
+                    .addr = iplit.addr,
+                    .scope_id = iplit.scope_id,
+                },
+            };
+            sl = @sizeOf(os.sockaddr_in6);
+            family = os.AF_INET6;
+        }
+    }
+
+    // Get local address and open/bind a socket
+    var sa: os.sockaddr = undefined;
+    @memset(@ptrCast([*]u8, &sa), 0, @sizeOf(os.sockaddr));
+    sa.in.family = family;
+    const flags = os.SOCK_DGRAM | os.SOCK_CLOEXEC | os.SOCK_NONBLOCK;
+    const fd = os.socket(family, flags, 0) catch |err| switch (err) {
+        error.AddressFamilyNotSupported => blk: {
+            // Handle case where system lacks IPv6 support
+            if (family == os.AF_INET6) {
+                family = os.AF_INET;
+                break :blk try os.socket(os.AF_INET, flags, 0);
+            }
+            return err;
+        },
+        else => |e| return e,
+    };
+    defer os.close(fd);
+    try os.bind(fd, &sa, sl);
+
+    // Past this point, there are no errors. Each individual query will
+    // yield either no reply (indicated by zero length) or an answer
+    // packet which is up to the caller to interpret.
+
+    // Convert any IPv4 addresses in a mixed environment to v4-mapped
+    // TODO
+    //if (family == AF_INET6) {
+    //    setsockopt(fd, IPPROTO_IPV6, IPV6_V6ONLY, &(int){0}, sizeof 0);
+    //    for (i=0; i<nns; i++) {
+    //        if (ns[i].sin.sin_family != AF_INET) continue;
+    //        memcpy(ns[i].sin6.sin6_addr.s6_addr+12,
+    //            &ns[i].sin.sin_addr, 4);
+    //        memcpy(ns[i].sin6.sin6_addr.s6_addr,
+    //            "\0\0\0\0\0\0\0\0\0\0\xff\xff", 12);
+    //        ns[i].sin6.sin6_family = AF_INET6;
+    //        ns[i].sin6.sin6_flowinfo = 0;
+    //        ns[i].sin6.sin6_scope_id = 0;
     //    }
     //}
 
-    //canon[l] = 0;
-    //return name_from_dns(buf, canon, name, family, &conf);
+    var pfd = [1]os.pollfd{os.pollfd{
+        .fd = fd,
+        .events = os.POLLIN,
+        .revents = undefined,
+    }};
+    const retry_interval = timeout / attempts;
+    var next: u32 = 0;
+    var t2: usize = std.time.milliTimestamp();
+    var t0 = t2;
+    var t1 = t2 - retry_interval;
+
+    var servfail_retry: usize = undefined;
+
+    outer: while (t2 - t0 < timeout) : (t2 = std.time.milliTimestamp()) {
+        if (t2 - t1 >= retry_interval) {
+            // Query all configured nameservers in parallel
+            var i: usize = 0;
+            while (i < queries.len) : (i += 1) {
+                if (answers[i].len == 0) {
+                    var j: usize = 0;
+                    while (j < ns.len) : (j += 1) {
+                        _ = os.sendto(fd, queries[i], os.MSG_NOSIGNAL, &ns[j], sl) catch undefined;
+                    }
+                }
+            }
+            t1 = t2;
+            servfail_retry = 2 * queries.len;
+        }
+
+        // Wait for a response, or until time to retry
+        const clamped_timeout = std.math.min(u31(std.math.maxInt(u31)), t1 + retry_interval - t2);
+        const nevents = os.poll(&pfd, clamped_timeout) catch 0;
+        if (nevents == 0) continue;
+
+        while (true) {
+            var sl_copy = sl;
+            const rlen = os.recvfrom(fd, answer_bufs[next], 0, &sa, &sl_copy) catch break;
+
+            // Ignore non-identifiable packets
+            if (rlen < 4) continue;
+
+            // Ignore replies from addresses we didn't send to
+            var j: usize = 0;
+            while (j < ns.len and !eqlSockAddr(&ns[j], &sa, sl)) : (j += 1) {}
+            if (j == ns.len) continue;
+
+            // Find which query this answer goes with, if any
+            var i: usize = next;
+            while (i < queries.len and (answer_bufs[next][0] != queries[i][0] or
+                answer_bufs[next][1] != queries[i][1])) : (i += 1)
+            {}
+
+            if (i == queries.len) continue;
+            if (answers[i].len != 0) continue;
+
+            // Only accept positive or negative responses;
+            // retry immediately on server failure, and ignore
+            // all other codes such as refusal.
+            switch (answer_bufs[next][3] & 15) {
+                0, 3 => {},
+                2 => if (servfail_retry != 0) {
+                    servfail_retry -= 1;
+                    _ = os.sendto(fd, queries[i], os.MSG_NOSIGNAL, &ns[j], sl) catch undefined;
+                },
+                else => continue,
+            }
+
+            // Store answer in the right slot, or update next
+            // available temp slot if it's already in place.
+            answers[i].len = rlen;
+            if (i == next) {
+                while (next < queries.len and answers[next].len != 0) : (next += 1) {}
+            } else {
+                mem.copy(u8, answer_bufs[i], answer_bufs[next][0..rlen]);
+            }
+
+            if (next == queries.len) break :outer;
+        }
+    }
+}
+
+fn dnsParse(
+    r: []const u8,
+    ctx: var,
+    comptime callback: var,
+) !void {
+    if (r.len < 12) return error.InvalidDnsPacket;
+    if ((r[3] & 15) != 0) return;
+    var p = r.ptr + 12;
+    var qdcount = r[4] * usize(256) + r[5];
+    var ancount = r[6] * usize(256) + r[7];
+    if (qdcount + ancount > 64) return error.InvalidDnsPacket;
+    while (qdcount != 0) {
+        qdcount -= 1;
+        while (@ptrToInt(p) - @ptrToInt(r.ptr) < r.len and p[0] -% 1 < 127) p += 1;
+        if (p[0] > 193 or (p[0] == 193 and p[1] > 254) or @ptrToInt(p) > @ptrToInt(r.ptr) + r.len - 6)
+            return error.InvalidDnsPacket;
+        p += usize(5) + @boolToInt(p[0] != 0);
+    }
+    while (ancount != 0) {
+        ancount -= 1;
+        while (@ptrToInt(p) - @ptrToInt(r.ptr) < r.len and p[0] -% 1 < 127) p += 1;
+        if (p[0] > 193 or (p[0] == 193 and p[1] > 254) or @ptrToInt(p) > @ptrToInt(r.ptr) + r.len - 6)
+            return error.InvalidDnsPacket;
+        p += usize(1) + @boolToInt(p[0] != 0);
+        const len = p[8] * usize(256) + p[9];
+        if (@ptrToInt(p) + len > @ptrToInt(r.ptr) + r.len) return error.InvalidDnsPacket;
+        try callback(ctx, p[1], p[10 .. 10 + len], r);
+        p += 10 + len;
+    }
+}
+
+fn dnsParseCallback(ctx: dpc_ctx, rr: u8, data: []const u8, packet: []const u8) !void {
+    var tmp: [256]u8 = undefined;
+    switch (rr) {
+        os.RR_A => {
+            if (data.len != 4) return error.InvalidDnsARecord;
+            const new_addr = try ctx.addrs.addOne();
+            new_addr.* = LookupAddr{
+                .family = os.AF_INET,
+                .addr = undefined,
+            };
+            mem.copy(u8, &new_addr.addr, data);
+        },
+        os.RR_AAAA => {
+            if (data.len != 16) return error.InvalidDnsAAAARecord;
+            const new_addr = try ctx.addrs.addOne();
+            new_addr.* = LookupAddr{
+                .family = os.AF_INET6,
+                .addr = undefined,
+            };
+            mem.copy(u8, &new_addr.addr, data);
+        },
+        os.RR_CNAME => {
+            @panic("TODO dn_expand");
+            //if (__dn_expand(packet, (const unsigned char *)packet + 512,
+            //    data, tmp, sizeof tmp) > 0 && is_valid_hostname(tmp))
+            //    strcpy(ctx->canon, tmp);
+        },
+        else => return,
+    }
 }
lib/std/os.zig
@@ -1508,16 +1508,17 @@ pub const SocketError = error{
     ProtocolNotSupported,
 } || UnexpectedError;
 
-pub fn socket(domain: u32, socket_type: u32, protocol: u32) SocketError!i32 {
+pub fn socket(domain: u32, socket_type: u32, protocol: u32) SocketError!fd_t {
     const rc = system.socket(domain, socket_type, protocol);
     switch (errno(rc)) {
-        0 => return @intCast(i32, rc),
+        0 => return @intCast(fd_t, rc),
         EACCES => return error.PermissionDenied,
         EAFNOSUPPORT => return error.AddressFamilyNotSupported,
         EINVAL => return error.ProtocolFamilyNotAvailable,
         EMFILE => return error.ProcessFdQuotaExceeded,
         ENFILE => return error.SystemFdQuotaExceeded,
-        ENOBUFS, ENOMEM => return error.SystemResources,
+        ENOBUFS => return error.SystemResources,
+        ENOMEM => return error.SystemResources,
         EPROTONOSUPPORT => return error.ProtocolNotSupported,
         else => |err| return unexpectedErrno(err),
     }
@@ -1559,17 +1560,17 @@ pub const BindError = error{
 } || UnexpectedError;
 
 /// addr is `*const T` where T is one of the sockaddr
-pub fn bind(fd: i32, addr: *const sockaddr) BindError!void {
-    const rc = system.bind(fd, addr, @sizeOf(sockaddr));
+pub fn bind(sockfd: fd_t, addr: *const sockaddr, len: socklen_t) BindError!void {
+    const rc = system.bind(sockfd, addr, len);
     switch (errno(rc)) {
         0 => return,
         EACCES => return error.AccessDenied,
         EADDRINUSE => return error.AddressInUse,
         EBADF => unreachable, // always a race condition if this error is returned
-        EINVAL => unreachable,
-        ENOTSOCK => unreachable,
+        EINVAL => unreachable, // invalid parameters
+        ENOTSOCK => unreachable, // invalid `sockfd`
         EADDRNOTAVAIL => return error.AddressNotAvailable,
-        EFAULT => unreachable,
+        EFAULT => unreachable, // invalid `addr` pointer
         ELOOP => return error.SymLinkLoop,
         ENAMETOOLONG => return error.NameTooLong,
         ENOENT => return error.FileNotFound,
@@ -2833,3 +2834,245 @@ pub fn gethostname(name_buffer: *[HOST_NAME_MAX]u8) GetHostNameError![]u8 {
 
     @compileError("TODO implement gethostname for this OS");
 }
+
+pub fn res_mkquery(
+    op: u4,
+    dname: []const u8,
+    class: u8,
+    ty: u8,
+    data: []const u8,
+    newrr: ?[*]const u8,
+    buf: []u8,
+) usize {
+    var name = dname;
+    if (mem.endsWith(u8, name, ".")) name.len -= 1;
+    assert(name.len <= 253);
+    const n = 17 + name.len + @boolToInt(name.len != 0);
+
+    // Construct query template - ID will be filled later
+    var q: [280]u8 = undefined;
+    @memset(&q, 0, n);
+    q[2] = u8(op) * 8 + 1;
+    q[5] = 1;
+    mem.copy(u8, q[13..], name);
+    var i: usize = 13;
+    var j: usize = undefined;
+    while (q[i] != 0) : (i = j + 1) {
+        j = i;
+        while (q[j] != 0 and q[j] != '.') : (j += 1) {}
+        // TODO determine the circumstances for this and whether or
+        // not this should be an error.
+        if (j - i - 1 > 62) unreachable;
+        q[i - 1] = @intCast(u8, j - i);
+    }
+    q[i + 1] = ty;
+    q[i + 3] = class;
+
+    // Make a reasonably unpredictable id
+    var ts: timespec = undefined;
+    clock_gettime(CLOCK_REALTIME, &ts) catch {};
+    const UInt = @IntType(false, @typeOf(ts.tv_nsec).bit_count);
+    const unsec = @bitCast(UInt, ts.tv_nsec);
+    const id = @truncate(u32, unsec + unsec / 65536);
+    q[0] = @truncate(u8, id / 256);
+    q[1] = @truncate(u8, id);
+
+    mem.copy(u8, buf, q[0..n]);
+    return n;
+}
+
+pub const SendError = error{
+    /// (For UNIX domain sockets, which are identified by pathname) Write permission is  denied
+    /// on  the destination socket file, or search permission is denied for one of the
+    /// directories the path prefix.  (See path_resolution(7).)
+    /// (For UDP sockets) An attempt was made to send to a network/broadcast address as  though
+    /// it was a unicast address.
+    AccessDenied,
+
+    /// The socket is marked nonblocking and the requested operation would block, and
+    /// there is no global event loop configured.
+    /// It's also possible to get this error under the following condition:
+    /// (Internet  domain datagram sockets) The socket referred to by sockfd had not previously
+    /// been bound to an address and, upon attempting to bind it to an ephemeral port,  it  was
+    /// determined that all port numbers in the ephemeral port range are currently in use.  See
+    /// the discussion of /proc/sys/net/ipv4/ip_local_port_range in ip(7).
+    WouldBlock,
+
+    /// Another Fast Open is already in progress.
+    FastOpenAlreadyInProgress,
+
+    /// Connection reset by peer.
+    ConnectionResetByPeer,
+
+    /// The  socket  type requires that message be sent atomically, and the size of the message
+    /// to be sent made this impossible. The message is not transmitted.
+    ///
+    MessageTooBig,
+
+    /// The output queue for a network interface was full.  This generally indicates  that  the
+    /// interface  has  stopped sending, but may be caused by transient congestion.  (Normally,
+    /// this does not occur in Linux.  Packets are just silently dropped when  a  device  queue
+    /// overflows.)
+    /// This is also caused when there is not enough kernel memory available.
+    SystemResources,
+
+    /// The  local  end  has been shut down on a connection oriented socket.  In this case, the
+    /// process will also receive a SIGPIPE unless MSG_NOSIGNAL is set.
+    BrokenPipe,
+} || UnexpectedError;
+
+/// Transmit a message to another socket.
+///
+/// The `sendto` call may be used only when the socket is in a connected state (so that the intended
+/// recipient  is  known). The  following call
+///
+///     send(sockfd, buf, len, flags);
+///
+/// is equivalent to
+///
+///     sendto(sockfd, buf, len, flags, NULL, 0);
+///
+/// If  sendto()  is used on a connection-mode (`SOCK_STREAM`, `SOCK_SEQPACKET`) socket, the arguments
+/// `dest_addr` and `addrlen` are asserted to be `null` and `0` respectively, and asserted
+/// that the socket was actually connected.
+/// Otherwise, the address of the target is given by `dest_addr` with `addrlen` specifying  its  size.
+///
+/// If the message is too long to pass atomically through the underlying protocol,
+/// `SendError.MessageTooBig` is returned, and the message is not transmitted.
+///
+/// There is no  indication  of  failure  to  deliver.
+///
+/// When the message does not fit into the send buffer of  the  socket,  `sendto`  normally  blocks,
+/// unless  the socket has been placed in nonblocking I/O mode.  In nonblocking mode it would fail
+/// with `SendError.WouldBlock`.  The `select` call may be used  to  determine when it is
+/// possible to send more data.
+pub fn sendto(
+    /// The file descriptor of the sending socket.
+    sockfd: fd_t,
+    /// Message to send.
+    buf: []const u8,
+    flags: u32,
+    dest_addr: ?*const sockaddr,
+    addrlen: socklen_t,
+) SendError!usize {
+    while (true) {
+        const rc = system.sendto(sockfd, buf.ptr, buf.len, flags, dest_addr, addrlen);
+        switch (errno(rc)) {
+            0 => return rc,
+
+            EACCES => return error.AccessDenied,
+            EAGAIN => if (std.event.Loop.instance) |loop| {
+                loop.waitUntilFdWritable(sockfd) catch return error.WouldBlock;
+                continue;
+            } else {
+                return error.WouldBlock;
+            },
+            EALREADY => return error.FastOpenAlreadyInProgress,
+            EBADF => unreachable, // always a race condition
+            ECONNRESET => return error.ConnectionResetByPeer,
+            EDESTADDRREQ => unreachable, // The socket is not connection-mode, and no peer address is set.
+            EFAULT => unreachable, // An invalid user space address was specified for an argument.
+            EINTR => continue,
+            EINVAL => unreachable, // Invalid argument passed.
+            EISCONN => unreachable, // connection-mode socket was connected already but a recipient was specified
+            EMSGSIZE => return error.MessageTooBig,
+            ENOBUFS => return error.SystemResources,
+            ENOMEM => return error.SystemResources,
+            ENOTCONN => unreachable, // The socket is not connected, and no target has been given.
+            ENOTSOCK => unreachable, // The file descriptor sockfd does not refer to a socket.
+            EOPNOTSUPP => unreachable, // Some bit in the flags argument is inappropriate for the socket type.
+            EPIPE => return error.BrokenPipe,
+            else => |err| return unexpectedErrno(err),
+        }
+    }
+}
+
+/// Transmit a message to another socket.
+///
+/// The `send` call may be used only when the socket is in a connected state (so that the intended
+/// recipient  is  known).   The  only  difference  between `send` and `write` is the presence of
+/// flags.  With a zero flags argument, `send` is equivalent to  `write`.   Also,  the  following
+/// call
+///
+///     send(sockfd, buf, len, flags);
+///
+/// is equivalent to
+///
+///     sendto(sockfd, buf, len, flags, NULL, 0);
+///
+/// There is no  indication  of  failure  to  deliver.
+///
+/// When the message does not fit into the send buffer of  the  socket,  `send`  normally  blocks,
+/// unless  the socket has been placed in nonblocking I/O mode.  In nonblocking mode it would fail
+/// with `SendError.WouldBlock`.  The `select` call may be used  to  determine when it is
+/// possible to send more data.
+pub fn send(
+    /// The file descriptor of the sending socket.
+    sockfd: fd_t,
+    buf: []const u8,
+    flags: u32,
+) SendError!usize {
+    return sendto(sockfd, buf, flags, null, 0);
+}
+
+pub const PollError = error{
+    /// The kernel had no space to allocate file descriptor tables.
+    SystemResources,
+} || UnexpectedError;
+
+pub fn poll(fds: []pollfd, timeout: i32) PollError!usize {
+    while (true) {
+        const rc = system.poll(fds.ptr, fds.len, timeout);
+        switch (errno(rc)) {
+            0 => return rc,
+            EFAULT => unreachable,
+            EINTR => continue,
+            EINVAL => unreachable,
+            ENOMEM => return error.SystemResources,
+            else => |err| return unexpectedErrno(err),
+        }
+    }
+}
+
+pub const RecvFromError = error{
+    /// The socket is marked nonblocking and the requested operation would block, and
+    /// there is no global event loop configured.
+    WouldBlock,
+
+    /// A remote host refused to allow the network connection, typically because it is not
+    /// running the requested service.
+    ConnectionRefused,
+
+    /// Could not allocate kernel memory.
+    SystemResources,
+} || UnexpectedError;
+
+pub fn recvfrom(
+    sockfd: fd_t,
+    buf: []u8,
+    flags: u32,
+    src_addr: ?*sockaddr,
+    addrlen: ?*socklen_t,
+) RecvFromError!usize {
+    while (true) {
+        const rc = system.recvfrom(sockfd, buf.ptr, buf.len, flags, src_addr, addrlen);
+        switch (errno(rc)) {
+            0 => return rc,
+            EBADF => unreachable, // always a race condition
+            EFAULT => unreachable,
+            EINVAL => unreachable,
+            ENOTCONN => unreachable,
+            ENOTSOCK => unreachable,
+            EINTR => continue,
+            EAGAIN => if (std.event.Loop.instance) |loop| {
+                loop.waitUntilFdReadable(sockfd) catch return error.WouldBlock;
+                continue;
+            } else {
+                return error.WouldBlock;
+            },
+            ENOMEM => return error.SystemResources,
+            ECONNREFUSED => return error.ConnectionRefused,
+            else => |err| return unexpectedErrno(err),
+        }
+    }
+}