[patch 43/56] pkt_*read fix

Subject: [patch @num@/@total@] pkt_*read fix

fix error recognition in pkt*read calls (some caused leaks, some others caused crashes ...)


Index: linux/hpc/kcomd.c
===================================================================
--- linux.orig/hpc/kcomd.c	2006-11-02 22:52:39.000000000 +0100
+++ linux/hpc/kcomd.c	2006-11-02 22:52:42.000000000 +0100
@@ -329,24 +329,24 @@
 {
 	struct kcom_pkt *recv_kcom_pkt;
 	int i=0;
-	task_t *sltsk;
+
 
 	// read in hdr
-	recv_kcom_pkt=pkt_hdr_read(node);
-	if (recv_kcom_pkt==NULL) {
-		printk(KERN_ERR "ERROR: incomplete header pkt\n");
-		goto error_recv;
-	}
+	recv_kcom_pkt = pkt_hdr_read(node);
+
+	if (recv_kcom_pkt==NULL) goto error_recv;
 
 	// read in any data
 	if (recv_kcom_pkt->len > 0) {
 		if ((recv_kcom_pkt->type & MSG_MASK) == PKT_NEW_MSG) {
-			recv_kcom_pkt->data=kzalloc(recv_kcom_pkt->len, GFP_KERNEL);
-			i=pkt_data_read(node, recv_kcom_pkt, recv_kcom_pkt->len, recv_kcom_pkt->data);
+
+			recv_kcom_pkt->data = kzalloc(recv_kcom_pkt->len, GFP_KERNEL);
+			i = pkt_data_read(node, recv_kcom_pkt, recv_kcom_pkt->len, recv_kcom_pkt->data);
 		} else {
-			i=pkt_data_read(node, recv_kcom_pkt, recv_kcom_pkt->len, recv_kcom_pkt->resp);
+			i = pkt_data_read(node, recv_kcom_pkt, recv_kcom_pkt->len, recv_kcom_pkt->resp);
 		}
-		if (i<recv_kcom_pkt->len) {
+
+		if (i < recv_kcom_pkt->len) {
 			printk(KERN_ERR "ERROR: incomplete data pkt\n");
 			goto error_recv;
 		}
@@ -362,12 +362,18 @@
 				mig_do_receive_home(node, recv_kcom_pkt);
 				break;
 			case MIG_COME_HOME:
-				sltsk=find_task_by_pid(recv_kcom_pkt->rpid);
+			{
+				task_t *sltsk;
+				sltsk = find_task_by_pid(recv_kcom_pkt->rpid);
+				if (!sltsk) {
+				    printk(KERN_ERR "openMosix: %s Unable to find the task %d ", __FUNCTION__, recv_kcom_pkt->rpid);
+				    goto error_recv;
+				}
 				task_register_migration(sltsk);
 				break;
+			}
 			default:
 				append_in_packs(recv_kcom_pkt);
-
 				break;
 		}
 	} else { // PKT_ACK and PKT_RESP go straight to in_packs
@@ -380,21 +386,10 @@
 	sys_close(node->fd);
 	sock_release(node->sock);
 
-	#if 0
-	node->fd=-1;
-	printk("kfree(data)\n");
-	kfree(pkt);
-	printk("kfree(node->sock)\n");
-	kfree(node->sock);
-	printk("list_del\n");
-	#endif
-
 	list_del(&node->list);
 	kmem_cache_free(kcom_node_cachep, node);
 
-	#if 0
-	kfree(node);
-	#endif
+
 	return -1;
 
 }
Index: linux/hpc/kcom.c
===================================================================
--- linux.orig/hpc/kcom.c	2006-11-02 22:52:33.000000000 +0100
+++ linux/hpc/kcom.c	2006-11-02 22:52:42.000000000 +0100
@@ -1,5 +1,6 @@
 /*
  *	Copyright (C) 2006 Matt Dew <[email protected]>
+ *	Copyright (C) 2006 Florian Delizy <[email protected]>
  *
  * This program is free software; you can redistribute it and/or modify
  * it under the terms of the GNU General Public License as published
@@ -40,6 +41,7 @@
 	struct msghdr msg = { NULL, 0, &iov, 1, NULL, 0, MSG_WAITALL | MSG_NOSIGNAL };
 	mm_segment_t oldfs;
 	int i;
+	int nb_retries = 0;
 	char buf[32];
 
 	iov.iov_base = data;
@@ -55,26 +57,39 @@
 	set_fs(KERNEL_DS);
 
 	while (iov.iov_len > 0) {
+
 		i = sock_recvmsg(sock, &msg, iov.iov_len, msg.msg_flags);
 		if ((i == -ENOSPC) || (i == -EAGAIN)) {
+
+			/* Prevent infinite loop 60s */
+			if (60000 > nb_retries) {
+				printk(KERN_ERR "openMosix %s too many retries\n", __FUNCTION__);
+				len = -1;
+				goto read_exit;
+			}
+
 			schedule_timeout(HZ/1000);
 			continue;
 		}
+
 		if (i < 0) {
-			printk(KERN_ERR "ERROR %d receiving data.\n", i);
-			set_fs(oldfs);
-			return -1;
+			printk(KERN_ERR "openMosix %s: %d receiving data.\n", __FUNCTION__, i);
+			len = -1;
+			goto read_exit;
 		}
 		iov.iov_base += i;
  	}
 
-	set_fs(oldfs);
 
 	if (len < 32) {
 		memcpy(data, buf, len);
 	}
 
+read_exit:
+	set_fs(oldfs);
+
 	return len;
+
 }
 EXPORT_SYMBOL_GPL(pkt_data_read);
 
@@ -94,32 +109,62 @@
 	mm_segment_t oldfs;
 	struct socket *sock=node->sock;
 	int i;
+ 	int retry = 0;
 
-	recv_kcom_pkt=kmem_cache_alloc(kcom_pkt_cachep, SLAB_KERNEL);
+ 	recv_kcom_pkt = kmem_cache_alloc(kcom_pkt_cachep, SLAB_KERNEL);
 	iov.iov_base = recv_kcom_pkt;
-	iov.iov_len = sizeof(*recv_kcom_pkt);
+ 	iov.iov_len = sizeof(struct kcom_pkt);
 
 	oldfs = get_fs();
 	set_fs(KERNEL_DS);
-	while (iov.iov_len > 0) {
-		i = sock_recvmsg(sock, &msg, iov.iov_len, msg.msg_flags);
-		if ((i == -ENOSPC) || (i == -EAGAIN)) {
-			schedule_timeout(HZ/1000);
-			continue;
-		}
-		if (i < 0) {
-			printk(KERN_ERR "ERROR %d receiving header.\n", i);
-			set_fs(oldfs);
-			return NULL;
-		}
-		iov.iov_base += i;
+
+
+ receive_fragment:
+
+ 	i = sock_recvmsg(sock, &msg, iov.iov_len, msg.msg_flags);
+ 	if ((i == -ENOSPC) || (i == -EAGAIN)) {
+ 		/* prevent infinite loop */
+ 		if (60000 > retry) goto receive_timeout;
+
+ 		schedule_timeout(HZ/1000);
+ 		goto receive_fragment;
+ 	}
+
+ 	if (i < 0) {
+ 		printk(KERN_ERR "openMosix: %s %d receiving header.\n", __FUNCTION__, i);
+ 		goto receive_error;
  	}
+ 	iov.iov_base += i;
+
+ 	/* sock_revmsg update the iov struct, and len */
+ 	if (iov.iov_len > 0) goto receive_fragment;
+
+
+ 	if (iov.iov_len != 0) goto receive_incomplete;
+
+
 	set_fs(oldfs);
 
-	if (iov.iov_len==0) // all expected data received.
-		return recv_kcom_pkt;
-	else
-		return NULL;
+ 	return recv_kcom_pkt;
+
+
+receive_timeout:
+ 	printk(KERN_ERR "openMosix %s Can't receive header %d fragment, too many (%d)\n"
+ 			 " retries)", __FUNCTION__, i, retry);
+ 	goto exit_error;
+
+receive_error:
+ 	printk(KERN_ERR "openMosix %s, %d receiving header.\n", __FUNCTION__,  i);
+ 	goto exit_error;
+
+receive_incomplete:
+ 	printk(KERN_ERR "openMosix %s incomplete packet received\n", __FUNCTION__);
+
+exit_error:
+
+ 	kmem_cache_free(kcom_pkt_cachep, recv_kcom_pkt);
+ 	set_fs(oldfs);
+ 	return NULL;
 }
 EXPORT_SYMBOL_GPL(pkt_hdr_read);