Add the rt linux 4.1.3-rt3 as base
[kvmfornfv.git] / kernel / drivers / net / netconsole.c
diff --git a/kernel/drivers/net/netconsole.c b/kernel/drivers/net/netconsole.c
new file mode 100644 (file)
index 0000000..15731d1
--- /dev/null
@@ -0,0 +1,859 @@
+/*
+ *  linux/drivers/net/netconsole.c
+ *
+ *  Copyright (C) 2001  Ingo Molnar <mingo@redhat.com>
+ *
+ *  This file contains the implementation of an IRQ-safe, crash-safe
+ *  kernel console implementation that outputs kernel messages to the
+ *  network.
+ *
+ * Modification history:
+ *
+ * 2001-09-17    started by Ingo Molnar.
+ * 2003-08-11    2.6 port by Matt Mackall
+ *               simplified options
+ *               generic card hooks
+ *               works non-modular
+ * 2003-09-07    rewritten with netpoll api
+ */
+
+/****************************************************************
+ *      This program is free software; you can redistribute it and/or modify
+ *      it under the terms of the GNU General Public License as published by
+ *      the Free Software Foundation; either version 2, or (at your option)
+ *      any later version.
+ *
+ *      This program is distributed in the hope that it will be useful,
+ *      but WITHOUT ANY WARRANTY; without even the implied warranty of
+ *      MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
+ *      GNU General Public License for more details.
+ *
+ *      You should have received a copy of the GNU General Public License
+ *      along with this program; if not, write to the Free Software
+ *      Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA.
+ *
+ ****************************************************************/
+
+#define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
+
+#include <linux/mm.h>
+#include <linux/init.h>
+#include <linux/module.h>
+#include <linux/slab.h>
+#include <linux/console.h>
+#include <linux/moduleparam.h>
+#include <linux/kernel.h>
+#include <linux/string.h>
+#include <linux/netpoll.h>
+#include <linux/inet.h>
+#include <linux/configfs.h>
+#include <linux/etherdevice.h>
+
+MODULE_AUTHOR("Maintainer: Matt Mackall <mpm@selenic.com>");
+MODULE_DESCRIPTION("Console driver for network interfaces");
+MODULE_LICENSE("GPL");
+
+#define MAX_PARAM_LENGTH       256
+#define MAX_PRINT_CHUNK                1000
+
+static char config[MAX_PARAM_LENGTH];
+module_param_string(netconsole, config, MAX_PARAM_LENGTH, 0);
+MODULE_PARM_DESC(netconsole, " netconsole=[src-port]@[src-ip]/[dev],[tgt-port]@<tgt-ip>/[tgt-macaddr]");
+
+static bool oops_only = false;
+module_param(oops_only, bool, 0600);
+MODULE_PARM_DESC(oops_only, "Only log oops messages");
+
+#ifndef        MODULE
+static int __init option_setup(char *opt)
+{
+       strlcpy(config, opt, MAX_PARAM_LENGTH);
+       return 1;
+}
+__setup("netconsole=", option_setup);
+#endif /* MODULE */
+
+/* Linked list of all configured targets */
+static LIST_HEAD(target_list);
+
+/* This needs to be a spinlock because write_msg() cannot sleep */
+static DEFINE_SPINLOCK(target_list_lock);
+
+/**
+ * struct netconsole_target - Represents a configured netconsole target.
+ * @list:      Links this target into the target_list.
+ * @item:      Links us into the configfs subsystem hierarchy.
+ * @enabled:   On / off knob to enable / disable target.
+ *             Visible from userspace (read-write).
+ *             We maintain a strict 1:1 correspondence between this and
+ *             whether the corresponding netpoll is active or inactive.
+ *             Also, other parameters of a target may be modified at
+ *             runtime only when it is disabled (enabled == 0).
+ * @np:                The netpoll structure for this target.
+ *             Contains the other userspace visible parameters:
+ *             dev_name        (read-write)
+ *             local_port      (read-write)
+ *             remote_port     (read-write)
+ *             local_ip        (read-write)
+ *             remote_ip       (read-write)
+ *             local_mac       (read-only)
+ *             remote_mac      (read-write)
+ */
+struct netconsole_target {
+       struct list_head        list;
+#ifdef CONFIG_NETCONSOLE_DYNAMIC
+       struct config_item      item;
+#endif
+       int                     enabled;
+       struct mutex            mutex;
+       struct netpoll          np;
+};
+
+#ifdef CONFIG_NETCONSOLE_DYNAMIC
+
+static struct configfs_subsystem netconsole_subsys;
+
+static int __init dynamic_netconsole_init(void)
+{
+       config_group_init(&netconsole_subsys.su_group);
+       mutex_init(&netconsole_subsys.su_mutex);
+       return configfs_register_subsystem(&netconsole_subsys);
+}
+
+static void __exit dynamic_netconsole_exit(void)
+{
+       configfs_unregister_subsystem(&netconsole_subsys);
+}
+
+/*
+ * Targets that were created by parsing the boot/module option string
+ * do not exist in the configfs hierarchy (and have NULL names) and will
+ * never go away, so make these a no-op for them.
+ */
+static void netconsole_target_get(struct netconsole_target *nt)
+{
+       if (config_item_name(&nt->item))
+               config_item_get(&nt->item);
+}
+
+static void netconsole_target_put(struct netconsole_target *nt)
+{
+       if (config_item_name(&nt->item))
+               config_item_put(&nt->item);
+}
+
+#else  /* !CONFIG_NETCONSOLE_DYNAMIC */
+
+static int __init dynamic_netconsole_init(void)
+{
+       return 0;
+}
+
+static void __exit dynamic_netconsole_exit(void)
+{
+}
+
+/*
+ * No danger of targets going away from under us when dynamic
+ * reconfigurability is off.
+ */
+static void netconsole_target_get(struct netconsole_target *nt)
+{
+}
+
+static void netconsole_target_put(struct netconsole_target *nt)
+{
+}
+
+#endif /* CONFIG_NETCONSOLE_DYNAMIC */
+
+/* Allocate new target (from boot/module param) and setup netpoll for it */
+static struct netconsole_target *alloc_param_target(char *target_config)
+{
+       int err = -ENOMEM;
+       struct netconsole_target *nt;
+
+       /*
+        * Allocate and initialize with defaults.
+        * Note that these targets get their config_item fields zeroed-out.
+        */
+       nt = kzalloc(sizeof(*nt), GFP_KERNEL);
+       if (!nt)
+               goto fail;
+
+       nt->np.name = "netconsole";
+       strlcpy(nt->np.dev_name, "eth0", IFNAMSIZ);
+       nt->np.local_port = 6665;
+       nt->np.remote_port = 6666;
+       mutex_init(&nt->mutex);
+       eth_broadcast_addr(nt->np.remote_mac);
+
+       /* Parse parameters and setup netpoll */
+       err = netpoll_parse_options(&nt->np, target_config);
+       if (err)
+               goto fail;
+
+       err = netpoll_setup(&nt->np);
+       if (err)
+               goto fail;
+
+       nt->enabled = 1;
+
+       return nt;
+
+fail:
+       kfree(nt);
+       return ERR_PTR(err);
+}
+
+/* Cleanup netpoll for given target (from boot/module param) and free it */
+static void free_param_target(struct netconsole_target *nt)
+{
+       netpoll_cleanup(&nt->np);
+       kfree(nt);
+}
+
+#ifdef CONFIG_NETCONSOLE_DYNAMIC
+
+/*
+ * Our subsystem hierarchy is:
+ *
+ * /sys/kernel/config/netconsole/
+ *                             |
+ *                             <target>/
+ *                             |       enabled
+ *                             |       dev_name
+ *                             |       local_port
+ *                             |       remote_port
+ *                             |       local_ip
+ *                             |       remote_ip
+ *                             |       local_mac
+ *                             |       remote_mac
+ *                             |
+ *                             <target>/...
+ */
+
+struct netconsole_target_attr {
+       struct configfs_attribute       attr;
+       ssize_t                         (*show)(struct netconsole_target *nt,
+                                               char *buf);
+       ssize_t                         (*store)(struct netconsole_target *nt,
+                                                const char *buf,
+                                                size_t count);
+};
+
+static struct netconsole_target *to_target(struct config_item *item)
+{
+       return item ?
+               container_of(item, struct netconsole_target, item) :
+               NULL;
+}
+
+/*
+ * Attribute operations for netconsole_target.
+ */
+
+static ssize_t show_enabled(struct netconsole_target *nt, char *buf)
+{
+       return snprintf(buf, PAGE_SIZE, "%d\n", nt->enabled);
+}
+
+static ssize_t show_dev_name(struct netconsole_target *nt, char *buf)
+{
+       return snprintf(buf, PAGE_SIZE, "%s\n", nt->np.dev_name);
+}
+
+static ssize_t show_local_port(struct netconsole_target *nt, char *buf)
+{
+       return snprintf(buf, PAGE_SIZE, "%d\n", nt->np.local_port);
+}
+
+static ssize_t show_remote_port(struct netconsole_target *nt, char *buf)
+{
+       return snprintf(buf, PAGE_SIZE, "%d\n", nt->np.remote_port);
+}
+
+static ssize_t show_local_ip(struct netconsole_target *nt, char *buf)
+{
+       if (nt->np.ipv6)
+               return snprintf(buf, PAGE_SIZE, "%pI6c\n", &nt->np.local_ip.in6);
+       else
+               return snprintf(buf, PAGE_SIZE, "%pI4\n", &nt->np.local_ip);
+}
+
+static ssize_t show_remote_ip(struct netconsole_target *nt, char *buf)
+{
+       if (nt->np.ipv6)
+               return snprintf(buf, PAGE_SIZE, "%pI6c\n", &nt->np.remote_ip.in6);
+       else
+               return snprintf(buf, PAGE_SIZE, "%pI4\n", &nt->np.remote_ip);
+}
+
+static ssize_t show_local_mac(struct netconsole_target *nt, char *buf)
+{
+       struct net_device *dev = nt->np.dev;
+       static const u8 bcast[ETH_ALEN] = { 0xff, 0xff, 0xff, 0xff, 0xff, 0xff };
+
+       return snprintf(buf, PAGE_SIZE, "%pM\n", dev ? dev->dev_addr : bcast);
+}
+
+static ssize_t show_remote_mac(struct netconsole_target *nt, char *buf)
+{
+       return snprintf(buf, PAGE_SIZE, "%pM\n", nt->np.remote_mac);
+}
+
+/*
+ * This one is special -- targets created through the configfs interface
+ * are not enabled (and the corresponding netpoll activated) by default.
+ * The user is expected to set the desired parameters first (which
+ * would enable him to dynamically add new netpoll targets for new
+ * network interfaces as and when they come up).
+ */
+static ssize_t store_enabled(struct netconsole_target *nt,
+                            const char *buf,
+                            size_t count)
+{
+       unsigned long flags;
+       int enabled;
+       int err;
+
+       err = kstrtoint(buf, 10, &enabled);
+       if (err < 0)
+               return err;
+       if (enabled < 0 || enabled > 1)
+               return -EINVAL;
+       if (enabled == nt->enabled) {
+               pr_info("network logging has already %s\n",
+                       nt->enabled ? "started" : "stopped");
+               return -EINVAL;
+       }
+
+       if (enabled) {  /* 1 */
+               /*
+                * Skip netpoll_parse_options() -- all the attributes are
+                * already configured via configfs. Just print them out.
+                */
+               netpoll_print_options(&nt->np);
+
+               err = netpoll_setup(&nt->np);
+               if (err)
+                       return err;
+
+               pr_info("netconsole: network logging started\n");
+       } else {        /* 0 */
+               /* We need to disable the netconsole before cleaning it up
+                * otherwise we might end up in write_msg() with
+                * nt->np.dev == NULL and nt->enabled == 1
+                */
+               spin_lock_irqsave(&target_list_lock, flags);
+               nt->enabled = 0;
+               spin_unlock_irqrestore(&target_list_lock, flags);
+               netpoll_cleanup(&nt->np);
+       }
+
+       nt->enabled = enabled;
+
+       return strnlen(buf, count);
+}
+
+static ssize_t store_dev_name(struct netconsole_target *nt,
+                             const char *buf,
+                             size_t count)
+{
+       size_t len;
+
+       if (nt->enabled) {
+               pr_err("target (%s) is enabled, disable to update parameters\n",
+                      config_item_name(&nt->item));
+               return -EINVAL;
+       }
+
+       strlcpy(nt->np.dev_name, buf, IFNAMSIZ);
+
+       /* Get rid of possible trailing newline from echo(1) */
+       len = strnlen(nt->np.dev_name, IFNAMSIZ);
+       if (nt->np.dev_name[len - 1] == '\n')
+               nt->np.dev_name[len - 1] = '\0';
+
+       return strnlen(buf, count);
+}
+
+static ssize_t store_local_port(struct netconsole_target *nt,
+                               const char *buf,
+                               size_t count)
+{
+       int rv;
+
+       if (nt->enabled) {
+               pr_err("target (%s) is enabled, disable to update parameters\n",
+                      config_item_name(&nt->item));
+               return -EINVAL;
+       }
+
+       rv = kstrtou16(buf, 10, &nt->np.local_port);
+       if (rv < 0)
+               return rv;
+       return strnlen(buf, count);
+}
+
+static ssize_t store_remote_port(struct netconsole_target *nt,
+                                const char *buf,
+                                size_t count)
+{
+       int rv;
+
+       if (nt->enabled) {
+               pr_err("target (%s) is enabled, disable to update parameters\n",
+                      config_item_name(&nt->item));
+               return -EINVAL;
+       }
+
+       rv = kstrtou16(buf, 10, &nt->np.remote_port);
+       if (rv < 0)
+               return rv;
+       return strnlen(buf, count);
+}
+
+static ssize_t store_local_ip(struct netconsole_target *nt,
+                             const char *buf,
+                             size_t count)
+{
+       if (nt->enabled) {
+               pr_err("target (%s) is enabled, disable to update parameters\n",
+                      config_item_name(&nt->item));
+               return -EINVAL;
+       }
+
+       if (strnchr(buf, count, ':')) {
+               const char *end;
+               if (in6_pton(buf, count, nt->np.local_ip.in6.s6_addr, -1, &end) > 0) {
+                       if (*end && *end != '\n') {
+                               pr_err("invalid IPv6 address at: <%c>\n", *end);
+                               return -EINVAL;
+                       }
+                       nt->np.ipv6 = true;
+               } else
+                       return -EINVAL;
+       } else {
+               if (!nt->np.ipv6) {
+                       nt->np.local_ip.ip = in_aton(buf);
+               } else
+                       return -EINVAL;
+       }
+
+       return strnlen(buf, count);
+}
+
+static ssize_t store_remote_ip(struct netconsole_target *nt,
+                              const char *buf,
+                              size_t count)
+{
+       if (nt->enabled) {
+               pr_err("target (%s) is enabled, disable to update parameters\n",
+                      config_item_name(&nt->item));
+               return -EINVAL;
+       }
+
+       if (strnchr(buf, count, ':')) {
+               const char *end;
+               if (in6_pton(buf, count, nt->np.remote_ip.in6.s6_addr, -1, &end) > 0) {
+                       if (*end && *end != '\n') {
+                               pr_err("invalid IPv6 address at: <%c>\n", *end);
+                               return -EINVAL;
+                       }
+                       nt->np.ipv6 = true;
+               } else
+                       return -EINVAL;
+       } else {
+               if (!nt->np.ipv6) {
+                       nt->np.remote_ip.ip = in_aton(buf);
+               } else
+                       return -EINVAL;
+       }
+
+       return strnlen(buf, count);
+}
+
+static ssize_t store_remote_mac(struct netconsole_target *nt,
+                               const char *buf,
+                               size_t count)
+{
+       u8 remote_mac[ETH_ALEN];
+
+       if (nt->enabled) {
+               pr_err("target (%s) is enabled, disable to update parameters\n",
+                      config_item_name(&nt->item));
+               return -EINVAL;
+       }
+
+       if (!mac_pton(buf, remote_mac))
+               return -EINVAL;
+       if (buf[3 * ETH_ALEN - 1] && buf[3 * ETH_ALEN - 1] != '\n')
+               return -EINVAL;
+       memcpy(nt->np.remote_mac, remote_mac, ETH_ALEN);
+
+       return strnlen(buf, count);
+}
+
+/*
+ * Attribute definitions for netconsole_target.
+ */
+
+#define NETCONSOLE_TARGET_ATTR_RO(_name)                               \
+static struct netconsole_target_attr netconsole_target_##_name =       \
+       __CONFIGFS_ATTR(_name, S_IRUGO, show_##_name, NULL)
+
+#define NETCONSOLE_TARGET_ATTR_RW(_name)                               \
+static struct netconsole_target_attr netconsole_target_##_name =       \
+       __CONFIGFS_ATTR(_name, S_IRUGO | S_IWUSR, show_##_name, store_##_name)
+
+NETCONSOLE_TARGET_ATTR_RW(enabled);
+NETCONSOLE_TARGET_ATTR_RW(dev_name);
+NETCONSOLE_TARGET_ATTR_RW(local_port);
+NETCONSOLE_TARGET_ATTR_RW(remote_port);
+NETCONSOLE_TARGET_ATTR_RW(local_ip);
+NETCONSOLE_TARGET_ATTR_RW(remote_ip);
+NETCONSOLE_TARGET_ATTR_RO(local_mac);
+NETCONSOLE_TARGET_ATTR_RW(remote_mac);
+
+static struct configfs_attribute *netconsole_target_attrs[] = {
+       &netconsole_target_enabled.attr,
+       &netconsole_target_dev_name.attr,
+       &netconsole_target_local_port.attr,
+       &netconsole_target_remote_port.attr,
+       &netconsole_target_local_ip.attr,
+       &netconsole_target_remote_ip.attr,
+       &netconsole_target_local_mac.attr,
+       &netconsole_target_remote_mac.attr,
+       NULL,
+};
+
+/*
+ * Item operations and type for netconsole_target.
+ */
+
+static void netconsole_target_release(struct config_item *item)
+{
+       kfree(to_target(item));
+}
+
+static ssize_t netconsole_target_attr_show(struct config_item *item,
+                                          struct configfs_attribute *attr,
+                                          char *buf)
+{
+       ssize_t ret = -EINVAL;
+       struct netconsole_target *nt = to_target(item);
+       struct netconsole_target_attr *na =
+               container_of(attr, struct netconsole_target_attr, attr);
+
+       if (na->show)
+               ret = na->show(nt, buf);
+
+       return ret;
+}
+
+static ssize_t netconsole_target_attr_store(struct config_item *item,
+                                           struct configfs_attribute *attr,
+                                           const char *buf,
+                                           size_t count)
+{
+       ssize_t ret = -EINVAL;
+       struct netconsole_target *nt = to_target(item);
+       struct netconsole_target_attr *na =
+               container_of(attr, struct netconsole_target_attr, attr);
+
+       mutex_lock(&nt->mutex);
+       if (na->store)
+               ret = na->store(nt, buf, count);
+       mutex_unlock(&nt->mutex);
+
+       return ret;
+}
+
+static struct configfs_item_operations netconsole_target_item_ops = {
+       .release                = netconsole_target_release,
+       .show_attribute         = netconsole_target_attr_show,
+       .store_attribute        = netconsole_target_attr_store,
+};
+
+static struct config_item_type netconsole_target_type = {
+       .ct_attrs               = netconsole_target_attrs,
+       .ct_item_ops            = &netconsole_target_item_ops,
+       .ct_owner               = THIS_MODULE,
+};
+
+/*
+ * Group operations and type for netconsole_subsys.
+ */
+
+static struct config_item *make_netconsole_target(struct config_group *group,
+                                                 const char *name)
+{
+       unsigned long flags;
+       struct netconsole_target *nt;
+
+       /*
+        * Allocate and initialize with defaults.
+        * Target is disabled at creation (enabled == 0).
+        */
+       nt = kzalloc(sizeof(*nt), GFP_KERNEL);
+       if (!nt)
+               return ERR_PTR(-ENOMEM);
+
+       nt->np.name = "netconsole";
+       strlcpy(nt->np.dev_name, "eth0", IFNAMSIZ);
+       nt->np.local_port = 6665;
+       nt->np.remote_port = 6666;
+       mutex_init(&nt->mutex);
+       eth_broadcast_addr(nt->np.remote_mac);
+
+       /* Initialize the config_item member */
+       config_item_init_type_name(&nt->item, name, &netconsole_target_type);
+
+       /* Adding, but it is disabled */
+       spin_lock_irqsave(&target_list_lock, flags);
+       list_add(&nt->list, &target_list);
+       spin_unlock_irqrestore(&target_list_lock, flags);
+
+       return &nt->item;
+}
+
+static void drop_netconsole_target(struct config_group *group,
+                                  struct config_item *item)
+{
+       unsigned long flags;
+       struct netconsole_target *nt = to_target(item);
+
+       spin_lock_irqsave(&target_list_lock, flags);
+       list_del(&nt->list);
+       spin_unlock_irqrestore(&target_list_lock, flags);
+
+       /*
+        * The target may have never been enabled, or was manually disabled
+        * before being removed so netpoll may have already been cleaned up.
+        */
+       if (nt->enabled)
+               netpoll_cleanup(&nt->np);
+
+       config_item_put(&nt->item);
+}
+
+static struct configfs_group_operations netconsole_subsys_group_ops = {
+       .make_item      = make_netconsole_target,
+       .drop_item      = drop_netconsole_target,
+};
+
+static struct config_item_type netconsole_subsys_type = {
+       .ct_group_ops   = &netconsole_subsys_group_ops,
+       .ct_owner       = THIS_MODULE,
+};
+
+/* The netconsole configfs subsystem */
+static struct configfs_subsystem netconsole_subsys = {
+       .su_group       = {
+               .cg_item        = {
+                       .ci_namebuf     = "netconsole",
+                       .ci_type        = &netconsole_subsys_type,
+               },
+       },
+};
+
+#endif /* CONFIG_NETCONSOLE_DYNAMIC */
+
+/* Handle network interface device notifications */
+static int netconsole_netdev_event(struct notifier_block *this,
+                                  unsigned long event, void *ptr)
+{
+       unsigned long flags;
+       struct netconsole_target *nt;
+       struct net_device *dev = netdev_notifier_info_to_dev(ptr);
+       bool stopped = false;
+
+       if (!(event == NETDEV_CHANGENAME || event == NETDEV_UNREGISTER ||
+             event == NETDEV_RELEASE || event == NETDEV_JOIN))
+               goto done;
+
+       spin_lock_irqsave(&target_list_lock, flags);
+restart:
+       list_for_each_entry(nt, &target_list, list) {
+               netconsole_target_get(nt);
+               if (nt->np.dev == dev) {
+                       switch (event) {
+                       case NETDEV_CHANGENAME:
+                               strlcpy(nt->np.dev_name, dev->name, IFNAMSIZ);
+                               break;
+                       case NETDEV_RELEASE:
+                       case NETDEV_JOIN:
+                       case NETDEV_UNREGISTER:
+                               /* rtnl_lock already held
+                                * we might sleep in __netpoll_cleanup()
+                                */
+                               spin_unlock_irqrestore(&target_list_lock, flags);
+
+                               __netpoll_cleanup(&nt->np);
+
+                               spin_lock_irqsave(&target_list_lock, flags);
+                               dev_put(nt->np.dev);
+                               nt->np.dev = NULL;
+                               nt->enabled = 0;
+                               stopped = true;
+                               netconsole_target_put(nt);
+                               goto restart;
+                       }
+               }
+               netconsole_target_put(nt);
+       }
+       spin_unlock_irqrestore(&target_list_lock, flags);
+       if (stopped) {
+               const char *msg = "had an event";
+               switch (event) {
+               case NETDEV_UNREGISTER:
+                       msg = "unregistered";
+                       break;
+               case NETDEV_RELEASE:
+                       msg = "released slaves";
+                       break;
+               case NETDEV_JOIN:
+                       msg = "is joining a master device";
+                       break;
+               }
+               pr_info("network logging stopped on interface %s as it %s\n",
+                       dev->name, msg);
+       }
+
+done:
+       return NOTIFY_DONE;
+}
+
+static struct notifier_block netconsole_netdev_notifier = {
+       .notifier_call  = netconsole_netdev_event,
+};
+
+static void write_msg(struct console *con, const char *msg, unsigned int len)
+{
+       int frag, left;
+       unsigned long flags;
+       struct netconsole_target *nt;
+       const char *tmp;
+
+       if (oops_only && !oops_in_progress)
+               return;
+       /* Avoid taking lock and disabling interrupts unnecessarily */
+       if (list_empty(&target_list))
+               return;
+
+       spin_lock_irqsave(&target_list_lock, flags);
+       list_for_each_entry(nt, &target_list, list) {
+               netconsole_target_get(nt);
+               if (nt->enabled && netif_running(nt->np.dev)) {
+                       /*
+                        * We nest this inside the for-each-target loop above
+                        * so that we're able to get as much logging out to
+                        * at least one target if we die inside here, instead
+                        * of unnecessarily keeping all targets in lock-step.
+                        */
+                       tmp = msg;
+                       for (left = len; left;) {
+                               frag = min(left, MAX_PRINT_CHUNK);
+                               netpoll_send_udp(&nt->np, tmp, frag);
+                               tmp += frag;
+                               left -= frag;
+                       }
+               }
+               netconsole_target_put(nt);
+       }
+       spin_unlock_irqrestore(&target_list_lock, flags);
+}
+
+static struct console netconsole = {
+       .name   = "netcon",
+       .flags  = CON_ENABLED,
+       .write  = write_msg,
+};
+
+static int __init init_netconsole(void)
+{
+       int err;
+       struct netconsole_target *nt, *tmp;
+       unsigned long flags;
+       char *target_config;
+       char *input = config;
+
+       if (strnlen(input, MAX_PARAM_LENGTH)) {
+               while ((target_config = strsep(&input, ";"))) {
+                       nt = alloc_param_target(target_config);
+                       if (IS_ERR(nt)) {
+                               err = PTR_ERR(nt);
+                               goto fail;
+                       }
+                       /* Dump existing printks when we register */
+                       netconsole.flags |= CON_PRINTBUFFER;
+
+                       spin_lock_irqsave(&target_list_lock, flags);
+                       list_add(&nt->list, &target_list);
+                       spin_unlock_irqrestore(&target_list_lock, flags);
+               }
+       }
+
+       err = register_netdevice_notifier(&netconsole_netdev_notifier);
+       if (err)
+               goto fail;
+
+       err = dynamic_netconsole_init();
+       if (err)
+               goto undonotifier;
+
+       register_console(&netconsole);
+       pr_info("network logging started\n");
+
+       return err;
+
+undonotifier:
+       unregister_netdevice_notifier(&netconsole_netdev_notifier);
+
+fail:
+       pr_err("cleaning up\n");
+
+       /*
+        * Remove all targets and destroy them (only targets created
+        * from the boot/module option exist here). Skipping the list
+        * lock is safe here, and netpoll_cleanup() will sleep.
+        */
+       list_for_each_entry_safe(nt, tmp, &target_list, list) {
+               list_del(&nt->list);
+               free_param_target(nt);
+       }
+
+       return err;
+}
+
+static void __exit cleanup_netconsole(void)
+{
+       struct netconsole_target *nt, *tmp;
+
+       unregister_console(&netconsole);
+       dynamic_netconsole_exit();
+       unregister_netdevice_notifier(&netconsole_netdev_notifier);
+
+       /*
+        * Targets created via configfs pin references on our module
+        * and would first be rmdir(2)'ed from userspace. We reach
+        * here only when they are already destroyed, and only those
+        * created from the boot/module option are left, so remove and
+        * destroy them. Skipping the list lock is safe here, and
+        * netpoll_cleanup() will sleep.
+        */
+       list_for_each_entry_safe(nt, tmp, &target_list, list) {
+               list_del(&nt->list);
+               free_param_target(nt);
+       }
+}
+
+/*
+ * Use late_initcall to ensure netconsole is
+ * initialized after network device driver if built-in.
+ *
+ * late_initcall() and module_init() are identical if built as module.
+ */
+late_initcall(init_netconsole);
+module_exit(cleanup_netconsole);