]> git.proxmox.com Git - mirror_ubuntu-bionic-kernel.git/blobdiff - drivers/net/bonding/bond_main.c
[NETNS]: Add namespace parameter to ip_route_output_key.
[mirror_ubuntu-bionic-kernel.git] / drivers / net / bonding / bond_main.c
index 6937ef0e7275f40a14b227055f0f65886b24a052..2039f7838f2df713a0afe9a4bbe1aecc8955e617 100644 (file)
@@ -74,6 +74,7 @@
 #include <linux/ethtool.h>
 #include <linux/if_vlan.h>
 #include <linux/if_bonding.h>
+#include <linux/jiffies.h>
 #include <net/route.h>
 #include <net/net_namespace.h>
 #include "bonding.h"
@@ -174,6 +175,7 @@ struct bond_parm_tbl bond_mode_tbl[] = {
 struct bond_parm_tbl xmit_hashtype_tbl[] = {
 {      "layer2",               BOND_XMIT_POLICY_LAYER2},
 {      "layer3+4",             BOND_XMIT_POLICY_LAYER34},
+{      "layer2+3",             BOND_XMIT_POLICY_LAYER23},
 {      NULL,                   -1},
 };
 
@@ -1744,7 +1746,9 @@ int bond_release(struct net_device *bond_dev, struct net_device *slave_dev)
                 * has been cleared (if our_slave == old_current),
                 * but before a new active slave is selected.
                 */
+               write_unlock_bh(&bond->lock);
                bond_alb_deinit_slave(bond, slave);
+               write_lock_bh(&bond->lock);
        }
 
        if (oldcurrent == slave) {
@@ -1847,9 +1851,9 @@ int bond_release(struct net_device *bond_dev, struct net_device *slave_dev)
 */
 void bond_destroy(struct bonding *bond)
 {
-       unregister_netdevice(bond->dev);
        bond_deinit(bond->dev);
        bond_destroy_sysfs_entry(bond);
+       unregister_netdevice(bond->dev);
 }
 
 /*
@@ -1903,6 +1907,12 @@ static int bond_release_all(struct net_device *bond_dev)
                slave_dev = slave->dev;
                bond_detach_slave(bond, slave);
 
+               /* now that the slave is detached, unlock and perform
+                * all the undo steps that should not be called from
+                * within a lock.
+                */
+               write_unlock_bh(&bond->lock);
+
                if ((bond->params.mode == BOND_MODE_TLB) ||
                    (bond->params.mode == BOND_MODE_ALB)) {
                        /* must be called only after the slave
@@ -1913,12 +1923,6 @@ static int bond_release_all(struct net_device *bond_dev)
 
                bond_compute_features(bond);
 
-               /* now that the slave is detached, unlock and perform
-                * all the undo steps that should not be called from
-                * within a lock.
-                */
-               write_unlock_bh(&bond->lock);
-
                bond_destroy_slave_symlinks(bond_dev, slave_dev);
                bond_del_vlans_from_slave(bond, slave_dev);
 
@@ -2382,7 +2386,9 @@ void bond_mii_monitor(struct work_struct *work)
                rtnl_lock();
                read_lock(&bond->lock);
                __bond_mii_monitor(bond, 1);
-               rtnl_unlock();
+               read_unlock(&bond->lock);
+               rtnl_unlock();  /* might sleep, hold no other locks */
+               read_lock(&bond->lock);
        }
 
        delay = ((bond->params.miimon * HZ) / 1000) ? : 1;
@@ -2511,7 +2517,7 @@ static void bond_arp_send_all(struct bonding *bond, struct slave *slave)
                fl.fl4_dst = targets[i];
                fl.fl4_tos = RTO_ONLINK;
 
-               rv = ip_route_output_key(&rt, &fl);
+               rv = ip_route_output_key(&init_net, &rt, &fl);
                if (rv) {
                        if (net_ratelimit()) {
                                printk(KERN_WARNING DRV_NAME
@@ -2722,8 +2728,8 @@ void bond_loadbalance_arp_mon(struct work_struct *work)
         */
        bond_for_each_slave(bond, slave, i) {
                if (slave->link != BOND_LINK_UP) {
-                       if (((jiffies - slave->dev->trans_start) <= delta_in_ticks) &&
-                           ((jiffies - slave->dev->last_rx) <= delta_in_ticks)) {
+                       if (time_before_eq(jiffies, slave->dev->trans_start + delta_in_ticks) &&
+                           time_before_eq(jiffies, slave->dev->last_rx + delta_in_ticks)) {
 
                                slave->link  = BOND_LINK_UP;
                                slave->state = BOND_STATE_ACTIVE;
@@ -2754,8 +2760,8 @@ void bond_loadbalance_arp_mon(struct work_struct *work)
                         * when the source ip is 0, so don't take the link down
                         * if we don't know our ip yet
                         */
-                       if (((jiffies - slave->dev->trans_start) >= (2*delta_in_ticks)) ||
-                           (((jiffies - slave->dev->last_rx) >= (2*delta_in_ticks)) &&
+                       if (time_after_eq(jiffies, slave->dev->trans_start + 2*delta_in_ticks) ||
+                           (time_after_eq(jiffies, slave->dev->last_rx + 2*delta_in_ticks) &&
                             bond_has_ip(bond))) {
 
                                slave->link  = BOND_LINK_DOWN;
@@ -2848,8 +2854,8 @@ void bond_activebackup_arp_mon(struct work_struct *work)
         */
        bond_for_each_slave(bond, slave, i) {
                if (slave->link != BOND_LINK_UP) {
-                       if ((jiffies - slave_last_rx(bond, slave)) <=
-                            delta_in_ticks) {
+                       if (time_before_eq(jiffies,
+                           slave_last_rx(bond, slave) + delta_in_ticks)) {
 
                                slave->link = BOND_LINK_UP;
 
@@ -2858,7 +2864,7 @@ void bond_activebackup_arp_mon(struct work_struct *work)
                                write_lock_bh(&bond->curr_slave_lock);
 
                                if ((!bond->curr_active_slave) &&
-                                   ((jiffies - slave->dev->trans_start) <= delta_in_ticks)) {
+                                   time_before_eq(jiffies, slave->dev->trans_start + delta_in_ticks)) {
                                        bond_change_active_slave(bond, slave);
                                        bond->current_arp_slave = NULL;
                                } else if (bond->curr_active_slave != slave) {
@@ -2897,7 +2903,7 @@ void bond_activebackup_arp_mon(struct work_struct *work)
 
                        if ((slave != bond->curr_active_slave) &&
                            (!bond->current_arp_slave) &&
-                           (((jiffies - slave_last_rx(bond, slave)) >= 3*delta_in_ticks) &&
+                           (time_after_eq(jiffies, slave_last_rx(bond, slave) + 3*delta_in_ticks) &&
                             bond_has_ip(bond))) {
                                /* a backup slave has gone down; three times
                                 * the delta allows the current slave to be
@@ -2943,10 +2949,10 @@ void bond_activebackup_arp_mon(struct work_struct *work)
                 * before being taken out. if a primary is being used, check
                 * if it is up and needs to take over as the curr_active_slave
                 */
-               if ((((jiffies - slave->dev->trans_start) >= (2*delta_in_ticks)) ||
-           (((jiffies - slave_last_rx(bond, slave)) >= (2*delta_in_ticks)) &&
-            bond_has_ip(bond))) &&
-                   ((jiffies - slave->jiffies) >= 2*delta_in_ticks)) {
+               if ((time_after_eq(jiffies, slave->dev->trans_start + 2*delta_in_ticks) ||
+                       (time_after_eq(jiffies, slave_last_rx(bond, slave) + 2*delta_in_ticks) &&
+                        bond_has_ip(bond))) &&
+                       time_after_eq(jiffies, slave->jiffies + 2*delta_in_ticks)) {
 
                        slave->link  = BOND_LINK_DOWN;
 
@@ -3397,9 +3403,7 @@ static int bond_master_netdev_event(unsigned long event, struct net_device *bond
        case NETDEV_CHANGENAME:
                return bond_event_changename(event_bond);
        case NETDEV_UNREGISTER:
-               /*
-                * TODO: remove a bond from the list?
-                */
+               bond_release_all(event_bond->dev);
                break;
        default:
                break;
@@ -3603,6 +3607,24 @@ void bond_unregister_arp(struct bonding *bond)
 
 /*---------------------------- Hashing Policies -----------------------------*/
 
+/*
+ * Hash for the output device based upon layer 2 and layer 3 data. If
+ * the packet is not IP mimic bond_xmit_hash_policy_l2()
+ */
+static int bond_xmit_hash_policy_l23(struct sk_buff *skb,
+                                    struct net_device *bond_dev, int count)
+{
+       struct ethhdr *data = (struct ethhdr *)skb->data;
+       struct iphdr *iph = ip_hdr(skb);
+
+       if (skb->protocol == __constant_htons(ETH_P_IP)) {
+               return ((ntohl(iph->saddr ^ iph->daddr) & 0xffff) ^
+                       (data->h_dest[5] ^ bond_dev->dev_addr[5])) % count;
+       }
+
+       return (data->h_dest[5] ^ bond_dev->dev_addr[5]) % count;
+}
+
 /*
  * Hash for the output device based upon layer 3 and layer 4 data. If
  * the packet is a frag or not TCP or UDP, just use layer 3 data.  If it is
@@ -4305,6 +4327,22 @@ out:
 
 /*------------------------- Device initialization ---------------------------*/
 
+static void bond_set_xmit_hash_policy(struct bonding *bond)
+{
+       switch (bond->params.xmit_policy) {
+       case BOND_XMIT_POLICY_LAYER23:
+               bond->xmit_hash_policy = bond_xmit_hash_policy_l23;
+               break;
+       case BOND_XMIT_POLICY_LAYER34:
+               bond->xmit_hash_policy = bond_xmit_hash_policy_l34;
+               break;
+       case BOND_XMIT_POLICY_LAYER2:
+       default:
+               bond->xmit_hash_policy = bond_xmit_hash_policy_l2;
+               break;
+       }
+}
+
 /*
  * set bond mode specific net device operations
  */
@@ -4321,10 +4359,7 @@ void bond_set_mode_ops(struct bonding *bond, int mode)
                break;
        case BOND_MODE_XOR:
                bond_dev->hard_start_xmit = bond_xmit_xor;
-               if (bond->params.xmit_policy == BOND_XMIT_POLICY_LAYER34)
-                       bond->xmit_hash_policy = bond_xmit_hash_policy_l34;
-               else
-                       bond->xmit_hash_policy = bond_xmit_hash_policy_l2;
+               bond_set_xmit_hash_policy(bond);
                break;
        case BOND_MODE_BROADCAST:
                bond_dev->hard_start_xmit = bond_xmit_broadcast;
@@ -4332,10 +4367,7 @@ void bond_set_mode_ops(struct bonding *bond, int mode)
        case BOND_MODE_8023AD:
                bond_set_master_3ad_flags(bond);
                bond_dev->hard_start_xmit = bond_3ad_xmit_xor;
-               if (bond->params.xmit_policy == BOND_XMIT_POLICY_LAYER34)
-                       bond->xmit_hash_policy = bond_xmit_hash_policy_l34;
-               else
-                       bond->xmit_hash_policy = bond_xmit_hash_policy_l2;
+               bond_set_xmit_hash_policy(bond);
                break;
        case BOND_MODE_ALB:
                bond_set_master_alb_flags(bond);
@@ -4405,6 +4437,7 @@ static int bond_init(struct net_device *bond_dev, struct bond_params *params)
        bond_dev->set_multicast_list = bond_set_multicast_list;
        bond_dev->change_mtu = bond_change_mtu;
        bond_dev->set_mac_address = bond_set_mac_address;
+       bond_dev->validate_addr = NULL;
 
        bond_set_mode_ops(bond, bond->params.mode);
 
@@ -4461,6 +4494,27 @@ static void bond_deinit(struct net_device *bond_dev)
 #endif
 }
 
+static void bond_work_cancel_all(struct bonding *bond)
+{
+       write_lock_bh(&bond->lock);
+       bond->kill_timers = 1;
+       write_unlock_bh(&bond->lock);
+
+       if (bond->params.miimon && delayed_work_pending(&bond->mii_work))
+               cancel_delayed_work(&bond->mii_work);
+
+       if (bond->params.arp_interval && delayed_work_pending(&bond->arp_work))
+               cancel_delayed_work(&bond->arp_work);
+
+       if (bond->params.mode == BOND_MODE_ALB &&
+           delayed_work_pending(&bond->alb_work))
+               cancel_delayed_work(&bond->alb_work);
+
+       if (bond->params.mode == BOND_MODE_8023AD &&
+           delayed_work_pending(&bond->ad_work))
+               cancel_delayed_work(&bond->ad_work);
+}
+
 /* Unregister and free all bond devices.
  * Caller must hold rtnl_lock.
  */
@@ -4471,11 +4525,12 @@ static void bond_free_all(void)
        list_for_each_entry_safe(bond, nxt, &bond_dev_list, bond_list) {
                struct net_device *bond_dev = bond->dev;
 
+               bond_work_cancel_all(bond);
                bond_mc_list_destroy(bond);
                /* Release the bonded slaves */
                bond_release_all(bond_dev);
-               unregister_netdevice(bond_dev);
                bond_deinit(bond_dev);
+               unregister_netdevice(bond_dev);
        }
 
 #ifdef CONFIG_PROC_FS
@@ -4487,19 +4542,27 @@ static void bond_free_all(void)
 
 /*
  * Convert string input module parms.  Accept either the
- * number of the mode or its string name.
+ * number of the mode or its string name.  A bit complicated because
+ * some mode names are substrings of other names, and calls from sysfs
+ * may have whitespace in the name (trailing newlines, for example).
  */
-int bond_parse_parm(char *mode_arg, struct bond_parm_tbl *tbl)
+int bond_parse_parm(const char *buf, struct bond_parm_tbl *tbl)
 {
-       int i;
+       int mode = -1, i, rv;
+       char modestr[BOND_MAX_MODENAME_LEN + 1] = { 0, };
+
+       rv = sscanf(buf, "%d", &mode);
+       if (!rv) {
+               rv = sscanf(buf, "%20s", modestr);
+               if (!rv)
+                       return -1;
+       }
 
        for (i = 0; tbl[i].modename; i++) {
-               if ((isdigit(*mode_arg) &&
-                    tbl[i].mode == simple_strtol(mode_arg, NULL, 0)) ||
-                   (strncmp(mode_arg, tbl[i].modename,
-                            strlen(tbl[i].modename)) == 0)) {
+               if (mode == tbl[i].mode)
+                       return tbl[i].mode;
+               if (strcmp(modestr, tbl[i].modename) == 0)
                        return tbl[i].mode;
-               }
        }
 
        return -1;
@@ -4813,9 +4876,22 @@ static struct lock_class_key bonding_netdev_xmit_lock_key;
 int bond_create(char *name, struct bond_params *params, struct bonding **newbond)
 {
        struct net_device *bond_dev;
+       struct bonding *bond, *nxt;
        int res;
 
        rtnl_lock();
+       down_write(&bonding_rwsem);
+
+       /* Check to see if the bond already exists. */
+       list_for_each_entry_safe(bond, nxt, &bond_dev_list, bond_list)
+               if (strnicmp(bond->dev->name, name, IFNAMSIZ) == 0) {
+                       printk(KERN_ERR DRV_NAME
+                              ": cannot add bond %s; it already exists\n",
+                              name);
+                       res = -EPERM;
+                       goto out_rtnl;
+               }
+
        bond_dev = alloc_netdev(sizeof(struct bonding), name ? name : "",
                                ether_setup);
        if (!bond_dev) {
@@ -4854,10 +4930,12 @@ int bond_create(char *name, struct bond_params *params, struct bonding **newbond
 
        netif_carrier_off(bond_dev);
 
+       up_write(&bonding_rwsem);
        rtnl_unlock(); /* allows sysfs registration of net device */
        res = bond_create_sysfs_entry(bond_dev->priv);
        if (res < 0) {
                rtnl_lock();
+               down_write(&bonding_rwsem);
                goto out_bond;
        }
 
@@ -4868,31 +4946,11 @@ out_bond:
 out_netdev:
        free_netdev(bond_dev);
 out_rtnl:
+       up_write(&bonding_rwsem);
        rtnl_unlock();
        return res;
 }
 
-static void bond_work_cancel_all(struct bonding *bond)
-{
-       write_lock_bh(&bond->lock);
-       bond->kill_timers = 1;
-       write_unlock_bh(&bond->lock);
-
-       if (bond->params.miimon && delayed_work_pending(&bond->mii_work))
-               cancel_delayed_work(&bond->mii_work);
-
-       if (bond->params.arp_interval && delayed_work_pending(&bond->arp_work))
-               cancel_delayed_work(&bond->arp_work);
-
-       if (bond->params.mode == BOND_MODE_ALB &&
-           delayed_work_pending(&bond->alb_work))
-               cancel_delayed_work(&bond->alb_work);
-
-       if (bond->params.mode == BOND_MODE_8023AD &&
-           delayed_work_pending(&bond->ad_work))
-               cancel_delayed_work(&bond->ad_work);
-}
-
 static int __init bonding_init(void)
 {
        int i;
@@ -4909,6 +4967,9 @@ static int __init bonding_init(void)
 #ifdef CONFIG_PROC_FS
        bond_create_proc_dir();
 #endif
+
+       init_rwsem(&bonding_rwsem);
+
        for (i = 0; i < max_bonds; i++) {
                res = bond_create(NULL, &bonding_defaults, NULL);
                if (res)