]> git.saurik.com Git - apple/xnu.git/blobdiff - bsd/net/kpi_protocol.c
xnu-6153.61.1.tar.gz
[apple/xnu.git] / bsd / net / kpi_protocol.c
index 5dda02574d2e405c299143661a4e11445e4e8deb..3b3c50f5f2947465c9a893c415506ef07addbe06 100644 (file)
@@ -1,8 +1,8 @@
 /*
- * Copyright (c) 2004 Apple Computer, Inc. All rights reserved.
+ * Copyright (c) 2004-2013 Apple Inc. All rights reserved.
  *
  * @APPLE_OSREFERENCE_LICENSE_HEADER_START@
- * 
+ *
  * This file contains Original Code and/or Modifications of Original Code
  * as defined in and that are subject to the Apple Public Source License
  * Version 2.0 (the 'License'). You may not use this file except in
  * unlawful or unlicensed copies of an Apple operating system, or to
  * circumvent, violate, or enable the circumvention or violation of, any
  * terms of an Apple operating system software license agreement.
- * 
+ *
  * Please obtain a copy of the License at
  * http://www.opensource.apple.com/apsl/ and read it before using this file.
- * 
+ *
  * The Original Code and all software distributed under the License are
  * distributed on an 'AS IS' basis, WITHOUT WARRANTY OF ANY KIND, EITHER
  * EXPRESS OR IMPLIED, AND APPLE HEREBY DISCLAIMS ALL SUCH WARRANTIES,
@@ -22,7 +22,7 @@
  * FITNESS FOR A PARTICULAR PURPOSE, QUIET ENJOYMENT OR NON-INFRINGEMENT.
  * Please see the License for the specific language governing rights and
  * limitations under the License.
- * 
+ *
  * @APPLE_OSREFERENCE_LICENSE_HEADER_END@
  */
 
 #include <net/dlil.h>
 #include <libkern/OSAtomic.h>
 
-void proto_kpi_init(void);
 void proto_input_run(void);
 
-typedef int (*attach_t)(struct ifnet *ifp, u_long protocol_family);
-typedef int (*detach_t)(struct ifnet *ifp, u_long protocol_family);
+typedef int (*attach_t)(struct ifnet *ifp, uint32_t protocol_family);
+typedef int (*detach_t)(struct ifnet *ifp, uint32_t protocol_family);
 
-/****************************************************************************/
-/* WARNING: Big assumption made here - there can be only one input thread      */
 struct proto_input_entry {
-       struct proto_input_entry                *next;
-       int                                                             detach;
-       struct domain                                   *domain;
-       
-       protocol_family_t                               protocol;
-       proto_input_handler                             input;
-       proto_input_detached_handler    detached;
-       
-       mbuf_t                                                  first_packet;
-       mbuf_t                                                  last_packet;
-};
+       struct proto_input_entry        *next;
+       int                             detach;
+       struct domain                   *domain;
+       int                             hash;
+       int                             chain;
 
-#define PROTO_HASH_SLOTS       5
+       protocol_family_t               protocol;
+       proto_input_handler             input;
+       proto_input_detached_handler    detached;
 
-static struct proto_input_entry        *proto_hash[PROTO_HASH_SLOTS];
-static struct proto_input_entry *proto_input_add_list;
-static lck_mtx_t                               *proto_input_lock = 0;
-__private_extern__ u_int32_t   inject_buckets = 0;
+       mbuf_t                          inject_first;
+       mbuf_t                          inject_last;
 
-extern thread_t        dlil_input_thread_ptr;
-extern int             dlil_input_thread_wakeup;
+       struct proto_input_entry        *input_next;
+       mbuf_t                          input_first;
+       mbuf_t                          input_last;
+};
 
-static int
-proto_hash_value(
-       protocol_family_t protocol)
-{
-       switch(protocol) {
-               case PF_INET:
-                       return 0;
-               case PF_INET6:
-                       return 1;
-               case PF_APPLETALK:
-                       return 2;
-               case PF_VLAN:
-                       return 3;
-       }
-       return 4;
-}
+
+struct proto_family_str {
+       TAILQ_ENTRY(proto_family_str)   proto_fam_next;
+       protocol_family_t               proto_family;
+       ifnet_family_t                  if_family;
+       proto_plumb_handler             attach_proto;
+       proto_unplumb_handler           detach_proto;
+};
+
+static struct proto_input_entry *proto_hash[PROTO_HASH_SLOTS];
+static int proto_total_waiting = 0;
+static struct proto_input_entry *proto_input_add_list = NULL;
+decl_lck_mtx_data(static, proto_family_mutex_data);
+static lck_mtx_t *proto_family_mutex = &proto_family_mutex_data;
+static TAILQ_HEAD(, proto_family_str) proto_family_head =
+    TAILQ_HEAD_INITIALIZER(proto_family_head);
 
 __private_extern__ void
 proto_kpi_init(void)
 {
-       lck_grp_attr_t  *grp_attrib = 0;
-       lck_attr_t              *lck_attrib = 0;
-       lck_grp_t               *lck_group = 0;
-       
+       lck_grp_attr_t  *grp_attrib = NULL;
+       lck_attr_t      *lck_attrib = NULL;
+       lck_grp_t       *lck_group = NULL;
+
        /* Allocate a mtx lock */
        grp_attrib = lck_grp_attr_alloc_init();
-       lck_grp_attr_setdefault(grp_attrib);
        lck_group = lck_grp_alloc_init("protocol kpi", grp_attrib);
        lck_grp_attr_free(grp_attrib);
        lck_attrib = lck_attr_alloc_init();
-       lck_attr_setdefault(lck_attrib);
-       proto_input_lock = lck_mtx_alloc_init(lck_group, lck_attrib);
+       lck_mtx_init(proto_family_mutex, lck_group, lck_attrib);
        lck_grp_free(lck_group);
        lck_attr_free(lck_attrib);
+
+       bzero(proto_hash, sizeof(proto_hash));
 }
 
 __private_extern__ errno_t
-proto_register_input(
-       protocol_family_t protocol,
-       proto_input_handler input,
-       proto_input_detached_handler detached)
+proto_register_input(protocol_family_t protocol, proto_input_handler input,
+    proto_input_detached_handler detached, int  chains)
 {
-       
        struct proto_input_entry *entry;
-       
-       entry = _MALLOC(sizeof(*entry), M_IFADDR, M_WAITOK);
-       
-       if (entry == NULL)
+       struct dlil_threading_info *inp = dlil_main_input_thread;
+       struct domain *dp;
+       domain_guard_t guard;
+
+       entry = _MALLOC(sizeof(*entry), M_IFADDR, M_WAITOK | M_ZERO);
+       if (entry == NULL) {
                return ENOMEM;
-       
-       bzero(entry, sizeof(*entry));
+       }
+
        entry->protocol = protocol;
        entry->input = input;
        entry->detached = detached;
-       
-       {
-               struct domain *dp = domains;
-               extern lck_mtx_t *domain_proto_mtx;
-
-               lck_mtx_assert(domain_proto_mtx, LCK_MTX_ASSERT_NOTOWNED);
-               lck_mtx_lock(domain_proto_mtx);
-               while (dp && dp->dom_family != protocol)
-                       dp = dp->dom_next;
-               entry->domain = dp;
-               lck_mtx_unlock(domain_proto_mtx);       
+       entry->hash = proto_hash_value(protocol);
+       entry->chain = chains;
+
+       guard = domain_guard_deploy();
+       TAILQ_FOREACH(dp, &domains, dom_entry) {
+               if (dp->dom_family == (int)protocol) {
+                       break;
+               }
+       }
+       domain_guard_release(guard);
+       if (dp == NULL) {
+               return EINVAL;
+       }
+
+       entry->domain = dp;
+
+       lck_mtx_lock(&inp->input_lck);
+       entry->next = proto_input_add_list;
+       proto_input_add_list = entry;
+
+       inp->input_waiting |= DLIL_PROTO_REGISTER;
+       if ((inp->input_waiting & DLIL_INPUT_RUNNING) == 0) {
+               wakeup((caddr_t)&inp->input_waiting);
        }
+       lck_mtx_unlock(&inp->input_lck);
 
-       
-       do {
-               entry->next = proto_input_add_list;
-       } while(!OSCompareAndSwap((UInt32)entry->next, (UInt32)entry, (UInt32*)&proto_input_add_list));
-       
-       wakeup((caddr_t)&dlil_input_thread_wakeup);
-       
        return 0;
 }
 
-
 __private_extern__ void
-proto_unregister_input(
-       protocol_family_t       protocol)
+proto_unregister_input(protocol_family_t protocol)
 {
        struct proto_input_entry *entry = NULL;
-       
-       for (entry = proto_hash[proto_hash_value(protocol)]; entry; entry = entry->next)
-               if (entry->protocol == protocol)
+
+       for (entry = proto_hash[proto_hash_value(protocol)]; entry != NULL;
+           entry = entry->next) {
+               if (entry->protocol == protocol) {
                        break;
-       
-       if (entry)
+               }
+       }
+
+       if (entry != NULL) {
                entry->detach = 1;
+       }
 }
 
-
 static void
-proto_delayed_attach(
-       struct proto_input_entry *entry)
+proto_delayed_attach(struct proto_input_entry *entry)
 {
        struct proto_input_entry *next_entry;
-       for (next_entry = entry->next; entry; entry = next_entry) {
+
+       for (next_entry = entry->next; entry != NULL; entry = next_entry) {
                struct proto_input_entry *exist;
                int hash_slot;
-               
+
                hash_slot = proto_hash_value(entry->protocol);
                next_entry = entry->next;
-               
-               for (exist = proto_hash[hash_slot]; exist; exist = exist->next)
-                       if (exist->protocol == entry->protocol)
+
+               for (exist = proto_hash[hash_slot]; exist != NULL;
+                   exist = exist->next) {
+                       if (exist->protocol == entry->protocol) {
                                break;
-               
+                       }
+               }
+
                /* If the entry already exists, call detached and dispose */
-               if (exist) {
-                       if (entry->detached)
+               if (exist != NULL) {
+                       if (entry->detached) {
                                entry->detached(entry->protocol);
+                       }
                        FREE(entry, M_IFADDR);
-               }
-               else {
+               } else {
                        entry->next = proto_hash[hash_slot];
                        proto_hash[hash_slot] = entry;
                }
        }
 }
 
-static void
-proto_delayed_inject(
-       struct proto_input_entry *entry)
-{
-       mbuf_t  packet_list;
-       mbuf_t  packet;
-       int             locked = 0;
-       
-       lck_mtx_lock(proto_input_lock);
-       packet_list = entry->first_packet;
-       entry->first_packet = entry->last_packet = 0;
-       lck_mtx_unlock(proto_input_lock);
-       
-       if (packet_list == NULL)
-               return;
-       
-       if (entry->domain && (entry->domain->dom_flags & DOM_REENTRANT) == 0) {
-               lck_mtx_lock(entry->domain->dom_mtx);
-               locked = 1;
-       }
-       
-       for (packet = packet_list; packet; packet = packet_list) {
-               packet_list = mbuf_nextpkt(packet);
-               mbuf_setnextpkt(packet, NULL);
-               entry->input(entry->protocol, packet);
-       }
-       
-       if (locked) {
-               lck_mtx_unlock(entry->domain->dom_mtx);
-       }
-}
-
-/* This function must be called from a single dlil input thread */
 __private_extern__ void
 proto_input_run(void)
 {
-       struct proto_input_entry        *entry;
-       u_int32_t                                       inject;
-       int                                                     i;
-       
-       if (current_thread() != dlil_input_thread_ptr)
-               panic("proto_input_run called from a thread other than dlil_input_thread!\n");
-
-       do {
+       struct proto_input_entry *entry;
+       struct dlil_threading_info *inp = dlil_main_input_thread;
+       mbuf_t packet_list;
+       int i, locked = 0;
+
+       LCK_MTX_ASSERT(&inp->input_lck, LCK_MTX_ASSERT_NOTOWNED);
+
+       if (inp->input_waiting & DLIL_PROTO_REGISTER) {
+               lck_mtx_lock_spin(&inp->input_lck);
                entry = proto_input_add_list;
-       } while (entry && !OSCompareAndSwap((UInt32)entry, 0, (UInt32*)&proto_input_add_list));
-       
-       if (entry)
+               proto_input_add_list = NULL;
+               inp->input_waiting &= ~DLIL_PROTO_REGISTER;
+               lck_mtx_unlock(&inp->input_lck);
                proto_delayed_attach(entry);
-       
-       do {
-               inject = inject_buckets;
-       } while (inject && !OSCompareAndSwap(inject, 0, (UInt32*)&inject_buckets));
-       
-       if (inject) {
-               for (i = 0; i < PROTO_HASH_SLOTS; i++) {
-                       if ((inject & (1L << i)) != 0) {
-                               for (entry = proto_hash[i]; entry; entry = entry->next) {
-                                       if (entry->first_packet) {
-                                               proto_delayed_inject(entry);
+       }
+
+       /*
+        * Move everything from the lock protected list to the thread
+        * specific list.
+        */
+       for (i = 0; proto_total_waiting != 0 && i < PROTO_HASH_SLOTS; i++) {
+               for (entry = proto_hash[i];
+                   entry != NULL && proto_total_waiting; entry = entry->next) {
+                       if (entry->inject_first != NULL) {
+                               lck_mtx_lock_spin(&inp->input_lck);
+                               inp->input_waiting &= ~DLIL_PROTO_WAITING;
+
+                               packet_list = entry->inject_first;
+
+                               entry->inject_first = NULL;
+                               entry->inject_last = NULL;
+                               proto_total_waiting--;
+
+                               lck_mtx_unlock(&inp->input_lck);
+
+                               if (entry->domain != NULL && !(entry->domain->
+                                   dom_flags & DOM_REENTRANT)) {
+                                       lck_mtx_lock(entry->domain->dom_mtx);
+                                       locked = 1;
+                               }
+
+                               if (entry->chain) {
+                                       entry->input(entry->protocol,
+                                           packet_list);
+                               } else {
+                                       mbuf_t  packet;
+
+                                       for (packet = packet_list;
+                                           packet != NULL;
+                                           packet = packet_list) {
+                                               packet_list =
+                                                   mbuf_nextpkt(packet);
+                                               mbuf_setnextpkt(packet, NULL);
+                                               entry->input(entry->protocol,
+                                                   packet);
                                        }
                                }
+                               if (locked) {
+                                       locked = 0;
+                                       lck_mtx_unlock(entry->domain->dom_mtx);
+                               }
                        }
                }
        }
 }
 
 errno_t
-proto_input(
-       protocol_family_t       protocol,
-       mbuf_t                          packet_list)
+proto_input(protocol_family_t protocol, mbuf_t packet_list)
 {
        struct proto_input_entry *entry;
-       
-       if (current_thread() != dlil_input_thread_ptr)
-               panic("proto_input called from a thread other than dlil_input_thread!\n");
-       
-       for (entry = proto_hash[proto_hash_value(protocol)]; entry; entry = entry->next) {
-               if (entry->protocol == protocol)
+       errno_t locked = 0, result = 0;
+
+       for (entry = proto_hash[proto_hash_value(protocol)]; entry != NULL;
+           entry = entry->next) {
+               if (entry->protocol == protocol) {
                        break;
+               }
+       }
+
+       if (entry == NULL) {
+               return -1;
        }
 
-       if (entry) {
-               mbuf_t  packet;
-#if DIRECT_PROTO_INPUT
-               // See <rdar://problem/3687868> for why this is disabled
-               // We need to release the dlil lock before taking the protocol lock
-               for (packet = packet_list; packet; packet = packet_list) {
+       if (entry->domain && !(entry->domain->dom_flags & DOM_REENTRANT)) {
+               lck_mtx_lock(entry->domain->dom_mtx);
+               locked = 1;
+       }
+
+       if (entry->chain) {
+               entry->input(entry->protocol, packet_list);
+       } else {
+               mbuf_t  packet;
+
+               for (packet = packet_list; packet != NULL;
+                   packet = packet_list) {
                        packet_list = mbuf_nextpkt(packet);
                        mbuf_setnextpkt(packet, NULL);
                        entry->input(entry->protocol, packet);
                }
-#else
-               mbuf_t  last_packet;
-               int             hash_slot = proto_hash_value(protocol);
-               
-               for (last_packet = packet_list; mbuf_nextpkt(last_packet);
-                        last_packet = mbuf_nextpkt(last_packet))
-                       /* find the last packet */;
-               
-               lck_mtx_lock(proto_input_lock);
-               if (entry->first_packet == NULL) {
-                       entry->first_packet = packet_list;
-               }
-               else {
-                       mbuf_setnextpkt(entry->last_packet, packet_list);
-               }
-               entry->last_packet = last_packet;
-               lck_mtx_unlock(proto_input_lock);
-       
-               OSBitOrAtomic((1L << hash_slot), (UInt32*)&inject_buckets);
-#endif
        }
-       else
-       {
-               return ENOENT;
+
+       if (locked) {
+               lck_mtx_unlock(entry->domain->dom_mtx);
        }
-       
-       return 0;
+       return result;
 }
 
 errno_t
-proto_inject(
-       protocol_family_t       protocol,
-       mbuf_t                          packet_list)
+proto_inject(protocol_family_t protocol, mbuf_t packet_list)
 {
-       struct proto_input_entry        *entry;
-       mbuf_t                                          last_packet;
-       int                                                     hash_slot = proto_hash_value(protocol);
-       
-       for (last_packet = packet_list; mbuf_nextpkt(last_packet);
-                last_packet = mbuf_nextpkt(last_packet))
+       struct proto_input_entry *entry;
+       mbuf_t last_packet;
+       int hash_slot = proto_hash_value(protocol);
+       struct dlil_threading_info *inp = dlil_main_input_thread;
+
+       for (last_packet = packet_list; mbuf_nextpkt(last_packet) != NULL;
+           last_packet = mbuf_nextpkt(last_packet)) {
                /* find the last packet */;
-       
-       for (entry = proto_hash[hash_slot]; entry; entry = entry->next) {
-               if (entry->protocol == protocol)
+       }
+
+       for (entry = proto_hash[hash_slot]; entry != NULL;
+           entry = entry->next) {
+               if (entry->protocol == protocol) {
                        break;
+               }
        }
-       
-       if (entry) {
-               lck_mtx_lock(proto_input_lock);
-               if (entry->first_packet == NULL) {
-                       entry->first_packet = packet_list;
+
+       if (entry != NULL) {
+               lck_mtx_lock(&inp->input_lck);
+               if (entry->inject_first == NULL) {
+                       proto_total_waiting++;
+                       inp->input_waiting |= DLIL_PROTO_WAITING;
+                       entry->inject_first = packet_list;
+               } else {
+                       mbuf_setnextpkt(entry->inject_last, packet_list);
                }
-               else {
-                       mbuf_setnextpkt(entry->last_packet, packet_list);
+               entry->inject_last = last_packet;
+               if ((inp->input_waiting & DLIL_INPUT_RUNNING) == 0) {
+                       wakeup((caddr_t)&inp->input_waiting);
                }
-               entry->last_packet = last_packet;
-               lck_mtx_unlock(proto_input_lock);
-       
-               OSBitOrAtomic((1L << hash_slot), (UInt32*)&inject_buckets);
-               
-               wakeup((caddr_t)&dlil_input_thread_wakeup);
-       }
-       else
-       {
+               lck_mtx_unlock(&inp->input_lck);
+       } else {
                return ENOENT;
        }
 
        return 0;
 }
 
+static struct proto_family_str *
+proto_plumber_find(protocol_family_t proto_family, ifnet_family_t if_family)
+{
+       struct proto_family_str  *mod = NULL;
+
+       TAILQ_FOREACH(mod, &proto_family_head, proto_fam_next) {
+               if ((mod->proto_family == (proto_family & 0xffff)) &&
+                   (mod->if_family == (if_family & 0xffff))) {
+                       break;
+               }
+       }
+
+       return mod;
+}
+
 errno_t
-proto_register_plumber(
-       protocol_family_t proto_fam,
-       ifnet_family_t if_fam,
-       proto_plumb_handler plumb,
-       proto_unplumb_handler unplumb)
+proto_register_plumber(protocol_family_t protocol_family,
+    ifnet_family_t interface_family, proto_plumb_handler attach,
+    proto_unplumb_handler detach)
 {
-       return dlil_reg_proto_module(proto_fam, if_fam, (attach_t)plumb, (detach_t)unplumb);
+       struct proto_family_str *proto_family;
+
+       if (attach == NULL) {
+               return EINVAL;
+       }
+
+       lck_mtx_lock(proto_family_mutex);
+
+       TAILQ_FOREACH(proto_family, &proto_family_head, proto_fam_next) {
+               if (proto_family->proto_family == protocol_family &&
+                   proto_family->if_family == interface_family) {
+                       lck_mtx_unlock(proto_family_mutex);
+                       return EEXIST;
+               }
+       }
+
+       proto_family = (struct proto_family_str *)
+           _MALLOC(sizeof(struct proto_family_str), M_IFADDR,
+           M_WAITOK | M_ZERO);
+       if (!proto_family) {
+               lck_mtx_unlock(proto_family_mutex);
+               return ENOMEM;
+       }
+
+       proto_family->proto_family      = protocol_family;
+       proto_family->if_family         = interface_family & 0xffff;
+       proto_family->attach_proto      = attach;
+       proto_family->detach_proto      = detach;
+
+       TAILQ_INSERT_TAIL(&proto_family_head, proto_family, proto_fam_next);
+       lck_mtx_unlock(proto_family_mutex);
+       return 0;
 }
 
 void
-proto_unregister_plumber(
-       protocol_family_t proto_fam,
-       ifnet_family_t if_fam)
+proto_unregister_plumber(protocol_family_t protocol_family,
+    ifnet_family_t interface_family)
+{
+       struct proto_family_str  *proto_family;
+
+       lck_mtx_lock(proto_family_mutex);
+
+       proto_family = proto_plumber_find(protocol_family, interface_family);
+       if (proto_family == NULL) {
+               lck_mtx_unlock(proto_family_mutex);
+               return;
+       }
+
+       TAILQ_REMOVE(&proto_family_head, proto_family, proto_fam_next);
+       FREE(proto_family, M_IFADDR);
+
+       lck_mtx_unlock(proto_family_mutex);
+}
+
+__private_extern__ errno_t
+proto_plumb(protocol_family_t protocol_family, ifnet_t ifp)
+{
+       struct proto_family_str  *proto_family;
+       int ret = 0;
+
+       lck_mtx_lock(proto_family_mutex);
+       proto_family = proto_plumber_find(protocol_family, ifp->if_family);
+       if (proto_family == NULL) {
+               lck_mtx_unlock(proto_family_mutex);
+               return ENXIO;
+       }
+
+       ret = proto_family->attach_proto(ifp, protocol_family);
+
+       lck_mtx_unlock(proto_family_mutex);
+       return ret;
+}
+
+
+__private_extern__ errno_t
+proto_unplumb(protocol_family_t protocol_family, ifnet_t ifp)
 {
-       (void)dlil_dereg_proto_module(proto_fam, if_fam);
+       struct proto_family_str  *proto_family;
+       int ret = 0;
+
+       lck_mtx_lock(proto_family_mutex);
+
+       proto_family = proto_plumber_find(protocol_family, ifp->if_family);
+       if (proto_family != NULL && proto_family->detach_proto) {
+               proto_family->detach_proto(ifp, protocol_family);
+       } else {
+               ret = ifnet_detach_protocol(ifp, protocol_family);
+       }
+
+       lck_mtx_unlock(proto_family_mutex);
+       return ret;
 }