From: Alan Cox <alan@linux.intel.com>
Date: Fri, 25 Mar 2016 21:07:07 +0000 (+0000)
Subject: alloc_socket: interrupt lock
X-Git-Url: https://git.ndcode.org/public/gitweb.cgi?a=commitdiff_plain;h=a31db315645b286a5f1f0901b8c63668a7e57a15;p=FUZIX.git

alloc_socket: interrupt lock

In the normal case this is safe because we always allocate from in kernel.
Likewise it's safe for accept with the native TCP/IP. However it's not safe
if we have a hardware TCP/IP and it implements accept() and thus allocates
accepting sockets from an interrupt.

So block IRQs for the brief scan.

[noted by Brett]
---

diff --git a/Kernel/syscall_net.c b/Kernel/syscall_net.c
index d8d46853..c4a4a8fa 100644
--- a/Kernel/syscall_net.c
+++ b/Kernel/syscall_net.c
@@ -90,12 +90,17 @@ static int sock_wait_enter(struct socket *s, uint8_t flag, uint8_t state)
 
 static struct socket *alloc_socket(void)
 {
+	irqflags_t irq = di();
 	struct socket *s = sockets;
 	while (s < sockets + NSOCKET) {
-		if (s->s_state == SS_UNUSED)
+		if (s->s_state == SS_UNUSED) {
+			s->s_state = SS_INIT;
+			irqrestore(irq);
 			return s;
+		}
 		s++;
 	}
+	irqrestore(irq);
 	return NULL;
 }