kdbus: add connection, queue handling and message validation code
authorDaniel Mack <daniel@zonque.org>
Thu, 11 Sep 2014 16:57:24 +0000 (18:57 +0200)
committerMaciej Wereski <m.wereski@partner.samsung.com>
Thu, 2 Apr 2015 15:24:17 +0000 (17:24 +0200)
This patch adds code to create and destroy connections, to validate
incoming messages and to maintain the queue of messages that are
associated with a connection.

Note that connection and queue have a 1:1 relation, the code is only
split in two parts for cleaner separation and better readability.

Signed-off-by: Daniel Mack <daniel@zonque.org>
Signed-off-by: David Herrmann <dh.herrmann@gmail.com>
Signed-off-by: Djalal Harouni <tixxdz@opendz.org>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
ipc/kdbus/connection.c [new file with mode: 0644]
ipc/kdbus/connection.h [new file with mode: 0644]
ipc/kdbus/item.c [new file with mode: 0644]
ipc/kdbus/item.h [new file with mode: 0644]
ipc/kdbus/message.c [new file with mode: 0644]
ipc/kdbus/message.h [new file with mode: 0644]
ipc/kdbus/queue.c [new file with mode: 0644]
ipc/kdbus/queue.h [new file with mode: 0644]
ipc/kdbus/reply.c [new file with mode: 0644]
ipc/kdbus/reply.h [new file with mode: 0644]
ipc/kdbus/util.h

diff --git a/ipc/kdbus/connection.c b/ipc/kdbus/connection.c
new file mode 100644 (file)
index 0000000..e554f1a
--- /dev/null
@@ -0,0 +1,2215 @@
+/*
+ * Copyright (C) 2013-2015 Kay Sievers
+ * Copyright (C) 2013-2015 Greg Kroah-Hartman <gregkh@linuxfoundation.org>
+ * Copyright (C) 2013-2015 Daniel Mack <daniel@zonque.org>
+ * Copyright (C) 2013-2015 David Herrmann <dh.herrmann@gmail.com>
+ * Copyright (C) 2013-2015 Linux Foundation
+ * Copyright (C) 2014-2015 Djalal Harouni <tixxdz@opendz.org>
+ *
+ * kdbus is free software; you can redistribute it and/or modify it under
+ * the terms of the GNU Lesser General Public License as published by the
+ * Free Software Foundation; either version 2.1 of the License, or (at
+ * your option) any later version.
+ */
+
+#include <linux/audit.h>
+#include <linux/file.h>
+#include <linux/fs.h>
+#include <linux/fs_struct.h>
+#include <linux/hashtable.h>
+#include <linux/idr.h>
+#include <linux/init.h>
+#include <linux/math64.h>
+#include <linux/mm.h>
+#include <linux/module.h>
+#include <linux/mutex.h>
+#include <linux/path.h>
+#include <linux/poll.h>
+#include <linux/sched.h>
+#include <linux/shmem_fs.h>
+#include <linux/sizes.h>
+#include <linux/slab.h>
+#include <linux/syscalls.h>
+#include <linux/uio.h>
+
+#include "bus.h"
+#include "connection.h"
+#include "endpoint.h"
+#include "handle.h"
+#include "match.h"
+#include "message.h"
+#include "metadata.h"
+#include "names.h"
+#include "domain.h"
+#include "item.h"
+#include "notify.h"
+#include "policy.h"
+#include "pool.h"
+#include "reply.h"
+#include "util.h"
+#include "queue.h"
+
+#define KDBUS_CONN_ACTIVE_BIAS (INT_MIN + 2)
+#define KDBUS_CONN_ACTIVE_NEW  (INT_MIN + 1)
+
+static struct kdbus_conn *kdbus_conn_new(struct kdbus_ep *ep, bool privileged,
+                                        struct kdbus_cmd_hello *hello,
+                                        const char *name,
+                                        const struct kdbus_creds *creds,
+                                        const struct kdbus_pids *pids,
+                                        const char *seclabel,
+                                        const char *conn_description)
+{
+#ifdef CONFIG_DEBUG_LOCK_ALLOC
+       static struct lock_class_key __key;
+#endif
+       struct kdbus_pool_slice *slice = NULL;
+       struct kdbus_bus *bus = ep->bus;
+       struct kdbus_conn *conn;
+       u64 attach_flags_send;
+       u64 attach_flags_recv;
+       u64 items_size = 0;
+       bool is_policy_holder;
+       bool is_activator;
+       bool is_monitor;
+       struct kvec kvec;
+       int ret;
+
+       struct {
+               u64 size;
+               u64 type;
+               struct kdbus_bloom_parameter bloom;
+       } bloom_item;
+
+       is_monitor = hello->flags & KDBUS_HELLO_MONITOR;
+       is_activator = hello->flags & KDBUS_HELLO_ACTIVATOR;
+       is_policy_holder = hello->flags & KDBUS_HELLO_POLICY_HOLDER;
+
+       if (!hello->pool_size || !IS_ALIGNED(hello->pool_size, PAGE_SIZE))
+               return ERR_PTR(-EINVAL);
+       if (is_monitor + is_activator + is_policy_holder > 1)
+               return ERR_PTR(-EINVAL);
+       if (name && !is_activator && !is_policy_holder)
+               return ERR_PTR(-EINVAL);
+       if (!name && (is_activator || is_policy_holder))
+               return ERR_PTR(-EINVAL);
+       if (name && !kdbus_name_is_valid(name, true))
+               return ERR_PTR(-EINVAL);
+       if (is_monitor && ep->user)
+               return ERR_PTR(-EOPNOTSUPP);
+       if (!privileged && (is_activator || is_policy_holder || is_monitor))
+               return ERR_PTR(-EPERM);
+       if ((creds || pids || seclabel) && !privileged)
+               return ERR_PTR(-EPERM);
+
+       ret = kdbus_sanitize_attach_flags(hello->attach_flags_send,
+                                         &attach_flags_send);
+       if (ret < 0)
+               return ERR_PTR(ret);
+
+       ret = kdbus_sanitize_attach_flags(hello->attach_flags_recv,
+                                         &attach_flags_recv);
+       if (ret < 0)
+               return ERR_PTR(ret);
+
+       /* The attach flags must always satisfy the bus requirements. */
+       if (bus->attach_flags_req & ~attach_flags_send)
+               return ERR_PTR(-ECONNREFUSED);
+
+       conn = kzalloc(sizeof(*conn), GFP_KERNEL);
+       if (!conn)
+               return ERR_PTR(-ENOMEM);
+
+       kref_init(&conn->kref);
+       atomic_set(&conn->active, KDBUS_CONN_ACTIVE_NEW);
+#ifdef CONFIG_DEBUG_LOCK_ALLOC
+       lockdep_init_map(&conn->dep_map, "s_active", &__key, 0);
+#endif
+       mutex_init(&conn->lock);
+       INIT_LIST_HEAD(&conn->names_list);
+       INIT_LIST_HEAD(&conn->names_queue_list);
+       INIT_LIST_HEAD(&conn->reply_list);
+       atomic_set(&conn->name_count, 0);
+       atomic_set(&conn->request_count, 0);
+       atomic_set(&conn->lost_count, 0);
+       INIT_DELAYED_WORK(&conn->work, kdbus_reply_list_scan_work);
+       conn->cred = get_current_cred();
+       init_waitqueue_head(&conn->wait);
+       kdbus_queue_init(&conn->queue);
+       conn->privileged = privileged;
+       conn->ep = kdbus_ep_ref(ep);
+       conn->id = atomic64_inc_return(&bus->domain->last_id);
+       conn->flags = hello->flags;
+       atomic64_set(&conn->attach_flags_send, attach_flags_send);
+       atomic64_set(&conn->attach_flags_recv, attach_flags_recv);
+       INIT_LIST_HEAD(&conn->monitor_entry);
+
+       if (conn_description) {
+               conn->description = kstrdup(conn_description, GFP_KERNEL);
+               if (!conn->description) {
+                       ret = -ENOMEM;
+                       goto exit_unref;
+               }
+       }
+
+       conn->pool = kdbus_pool_new(conn->description, hello->pool_size);
+       if (IS_ERR(conn->pool)) {
+               ret = PTR_ERR(conn->pool);
+               conn->pool = NULL;
+               goto exit_unref;
+       }
+
+       conn->match_db = kdbus_match_db_new();
+       if (IS_ERR(conn->match_db)) {
+               ret = PTR_ERR(conn->match_db);
+               conn->match_db = NULL;
+               goto exit_unref;
+       }
+
+       /* return properties of this connection to the caller */
+       hello->bus_flags = bus->bus_flags;
+       hello->id = conn->id;
+
+       BUILD_BUG_ON(sizeof(bus->id128) != sizeof(hello->id128));
+       memcpy(hello->id128, bus->id128, sizeof(hello->id128));
+
+       conn->meta = kdbus_meta_proc_new();
+       if (IS_ERR(conn->meta)) {
+               ret = PTR_ERR(conn->meta);
+               conn->meta = NULL;
+               goto exit_unref;
+       }
+
+       /* privileged processes can impersonate somebody else */
+       if (creds || pids || seclabel) {
+               ret = kdbus_meta_proc_fake(conn->meta, creds, pids, seclabel);
+               if (ret < 0)
+                       goto exit_unref;
+
+               conn->faked_meta = true;
+       } else {
+               ret = kdbus_meta_proc_collect(conn->meta,
+                                             KDBUS_ATTACH_CREDS |
+                                             KDBUS_ATTACH_PIDS |
+                                             KDBUS_ATTACH_AUXGROUPS |
+                                             KDBUS_ATTACH_TID_COMM |
+                                             KDBUS_ATTACH_PID_COMM |
+                                             KDBUS_ATTACH_EXE |
+                                             KDBUS_ATTACH_CMDLINE |
+                                             KDBUS_ATTACH_CGROUP |
+                                             KDBUS_ATTACH_CAPS |
+                                             KDBUS_ATTACH_SECLABEL |
+                                             KDBUS_ATTACH_AUDIT);
+               if (ret < 0)
+                       goto exit_unref;
+       }
+
+       /*
+        * Account the connection against the current user (UID), or for
+        * custom endpoints use the anonymous user assigned to the endpoint.
+        * Note that limits are always accounted against the real UID, not
+        * the effective UID (cred->user always points to the accounting of
+        * cred->uid, not cred->euid).
+        */
+       if (ep->user) {
+               conn->user = kdbus_user_ref(ep->user);
+       } else {
+               conn->user = kdbus_user_lookup(ep->bus->domain, current_uid());
+               if (IS_ERR(conn->user)) {
+                       ret = PTR_ERR(conn->user);
+                       conn->user = NULL;
+                       goto exit_unref;
+               }
+       }
+
+       if (atomic_inc_return(&conn->user->connections) > KDBUS_USER_MAX_CONN) {
+               /* decremented by destructor as conn->user is valid */
+               ret = -EMFILE;
+               goto exit_unref;
+       }
+
+       bloom_item.size = sizeof(bloom_item);
+       bloom_item.type = KDBUS_ITEM_BLOOM_PARAMETER;
+       bloom_item.bloom = bus->bloom;
+       kdbus_kvec_set(&kvec, &bloom_item, bloom_item.size, &items_size);
+
+       slice = kdbus_pool_slice_alloc(conn->pool, items_size, false);
+       if (IS_ERR(slice)) {
+               ret = PTR_ERR(slice);
+               slice = NULL;
+               goto exit_unref;
+       }
+
+       ret = kdbus_pool_slice_copy_kvec(slice, 0, &kvec, 1, items_size);
+       if (ret < 0)
+               goto exit_unref;
+
+       kdbus_pool_slice_publish(slice, &hello->offset, &hello->items_size);
+       kdbus_pool_slice_release(slice);
+
+       return conn;
+
+exit_unref:
+       kdbus_pool_slice_release(slice);
+       kdbus_conn_unref(conn);
+       return ERR_PTR(ret);
+}
+
+static void __kdbus_conn_free(struct kref *kref)
+{
+       struct kdbus_conn *conn = container_of(kref, struct kdbus_conn, kref);
+
+       WARN_ON(kdbus_conn_active(conn));
+       WARN_ON(delayed_work_pending(&conn->work));
+       WARN_ON(!list_empty(&conn->queue.msg_list));
+       WARN_ON(!list_empty(&conn->names_list));
+       WARN_ON(!list_empty(&conn->names_queue_list));
+       WARN_ON(!list_empty(&conn->reply_list));
+
+       if (conn->user) {
+               atomic_dec(&conn->user->connections);
+               kdbus_user_unref(conn->user);
+       }
+
+       kdbus_meta_proc_unref(conn->meta);
+       kdbus_match_db_free(conn->match_db);
+       kdbus_pool_free(conn->pool);
+       kdbus_ep_unref(conn->ep);
+       put_cred(conn->cred);
+       kfree(conn->description);
+       kfree(conn->quota);
+       kfree(conn);
+}
+
+/**
+ * kdbus_conn_ref() - take a connection reference
+ * @conn:              Connection, may be %NULL
+ *
+ * Return: the connection itself
+ */
+struct kdbus_conn *kdbus_conn_ref(struct kdbus_conn *conn)
+{
+       if (conn)
+               kref_get(&conn->kref);
+       return conn;
+}
+
+/**
+ * kdbus_conn_unref() - drop a connection reference
+ * @conn:              Connection (may be NULL)
+ *
+ * When the last reference is dropped, the connection's internal structure
+ * is freed.
+ *
+ * Return: NULL
+ */
+struct kdbus_conn *kdbus_conn_unref(struct kdbus_conn *conn)
+{
+       if (conn)
+               kref_put(&conn->kref, __kdbus_conn_free);
+       return NULL;
+}
+
+/**
+ * kdbus_conn_active() - connection is not disconnected
+ * @conn:              Connection to check
+ *
+ * Return true if the connection was not disconnected, yet. Note that a
+ * connection might be disconnected asynchronously, unless you hold the
+ * connection lock. If that's not suitable for you, see kdbus_conn_acquire() to
+ * suppress connection shutdown for a short period.
+ *
+ * Return: true if the connection is still active
+ */
+bool kdbus_conn_active(const struct kdbus_conn *conn)
+{
+       return atomic_read(&conn->active) >= 0;
+}
+
+/**
+ * kdbus_conn_acquire() - acquire an active connection reference
+ * @conn:              Connection
+ *
+ * Users can close a connection via KDBUS_BYEBYE (or by destroying the
+ * endpoint/bus/...) at any time. Whenever this happens, we should deny any
+ * user-visible action on this connection and signal ECONNRESET instead.
+ * To avoid testing for connection availability everytime you take the
+ * connection-lock, you can acquire a connection for short periods.
+ *
+ * By calling kdbus_conn_acquire(), you gain an "active reference" to the
+ * connection. You must also hold a regular reference at any time! As long as
+ * you hold the active-ref, the connection will not be shut down. However, if
+ * the connection was shut down, you can never acquire an active-ref again.
+ *
+ * kdbus_conn_disconnect() disables the connection and then waits for all active
+ * references to be dropped. It will also wake up any pending operation.
+ * However, you must not sleep for an indefinite period while holding an
+ * active-reference. Otherwise, kdbus_conn_disconnect() might stall. If you need
+ * to sleep for an indefinite period, either release the reference and try to
+ * acquire it again after waking up, or make kdbus_conn_disconnect() wake up
+ * your wait-queue.
+ *
+ * Return: 0 on success, negative error code on failure.
+ */
+int kdbus_conn_acquire(struct kdbus_conn *conn)
+{
+       if (!atomic_inc_unless_negative(&conn->active))
+               return -ECONNRESET;
+
+#ifdef CONFIG_DEBUG_LOCK_ALLOC
+       rwsem_acquire_read(&conn->dep_map, 0, 1, _RET_IP_);
+#endif
+
+       return 0;
+}
+
+/**
+ * kdbus_conn_release() - release an active connection reference
+ * @conn:              Connection
+ *
+ * This releases an active reference that has been acquired via
+ * kdbus_conn_acquire(). If the connection was already disabled and this is the
+ * last active-ref that is dropped, the disconnect-waiter will be woken up and
+ * properly close the connection.
+ */
+void kdbus_conn_release(struct kdbus_conn *conn)
+{
+       int v;
+
+       if (!conn)
+               return;
+
+#ifdef CONFIG_DEBUG_LOCK_ALLOC
+       rwsem_release(&conn->dep_map, 1, _RET_IP_);
+#endif
+
+       v = atomic_dec_return(&conn->active);
+       if (v != KDBUS_CONN_ACTIVE_BIAS)
+               return;
+
+       wake_up_all(&conn->wait);
+}
+
+static int kdbus_conn_connect(struct kdbus_conn *conn, const char *name)
+{
+       struct kdbus_ep *ep = conn->ep;
+       struct kdbus_bus *bus = ep->bus;
+       int ret;
+
+       if (WARN_ON(atomic_read(&conn->active) != KDBUS_CONN_ACTIVE_NEW))
+               return -EALREADY;
+
+       /* make sure the ep-node is active while we add our connection */
+       if (!kdbus_node_acquire(&ep->node))
+               return -ESHUTDOWN;
+
+       /* lock order: domain -> bus -> ep -> names -> conn */
+       mutex_lock(&ep->lock);
+       down_write(&bus->conn_rwlock);
+
+       /* link into monitor list */
+       if (kdbus_conn_is_monitor(conn))
+               list_add_tail(&conn->monitor_entry, &bus->monitors_list);
+
+       /* link into bus and endpoint */
+       list_add_tail(&conn->ep_entry, &ep->conn_list);
+       hash_add(bus->conn_hash, &conn->hentry, conn->id);
+
+       /* enable lookups and acquire active ref */
+       atomic_set(&conn->active, 1);
+#ifdef CONFIG_DEBUG_LOCK_ALLOC
+       rwsem_acquire_read(&conn->dep_map, 0, 1, _RET_IP_);
+#endif
+
+       up_write(&bus->conn_rwlock);
+       mutex_unlock(&ep->lock);
+
+       kdbus_node_release(&ep->node);
+
+       /*
+        * Notify subscribers about the new active connection, unless it is
+        * a monitor. Monitors are invisible on the bus, can't be addressed
+        * directly, and won't cause any notifications.
+        */
+       if (!kdbus_conn_is_monitor(conn)) {
+               ret = kdbus_notify_id_change(conn->ep->bus, KDBUS_ITEM_ID_ADD,
+                                            conn->id, conn->flags);
+               if (ret < 0)
+                       goto exit_disconnect;
+       }
+
+       if (kdbus_conn_is_activator(conn)) {
+               u64 flags = KDBUS_NAME_ACTIVATOR;
+
+               if (WARN_ON(!name)) {
+                       ret = -EINVAL;
+                       goto exit_disconnect;
+               }
+
+               ret = kdbus_name_acquire(bus->name_registry, conn, name,
+                                        flags, NULL);
+               if (ret < 0)
+                       goto exit_disconnect;
+       }
+
+       kdbus_conn_release(conn);
+       kdbus_notify_flush(bus);
+       return 0;
+
+exit_disconnect:
+       kdbus_conn_release(conn);
+       kdbus_conn_disconnect(conn, false);
+       return ret;
+}
+
+/**
+ * kdbus_conn_disconnect() - disconnect a connection
+ * @conn:              The connection to disconnect
+ * @ensure_queue_empty:        Flag to indicate if the call should fail in
+ *                     case the connection's message list is not
+ *                     empty
+ *
+ * If @ensure_msg_list_empty is true, and the connection has pending messages,
+ * -EBUSY is returned.
+ *
+ * Return: 0 on success, negative errno on failure
+ */
+int kdbus_conn_disconnect(struct kdbus_conn *conn, bool ensure_queue_empty)
+{
+       struct kdbus_queue_entry *entry, *tmp;
+       struct kdbus_bus *bus = conn->ep->bus;
+       struct kdbus_reply *r, *r_tmp;
+       struct kdbus_conn *c;
+       int i, v;
+
+       mutex_lock(&conn->lock);
+       v = atomic_read(&conn->active);
+       if (v == KDBUS_CONN_ACTIVE_NEW) {
+               /* was never connected */
+               mutex_unlock(&conn->lock);
+               return 0;
+       }
+       if (v < 0) {
+               /* already dead */
+               mutex_unlock(&conn->lock);
+               return -ECONNRESET;
+       }
+       if (ensure_queue_empty && !list_empty(&conn->queue.msg_list)) {
+               /* still busy */
+               mutex_unlock(&conn->lock);
+               return -EBUSY;
+       }
+
+       atomic_add(KDBUS_CONN_ACTIVE_BIAS, &conn->active);
+       mutex_unlock(&conn->lock);
+
+       wake_up_interruptible(&conn->wait);
+
+#ifdef CONFIG_DEBUG_LOCK_ALLOC
+       rwsem_acquire(&conn->dep_map, 0, 0, _RET_IP_);
+       if (atomic_read(&conn->active) != KDBUS_CONN_ACTIVE_BIAS)
+               lock_contended(&conn->dep_map, _RET_IP_);
+#endif
+
+       wait_event(conn->wait,
+                  atomic_read(&conn->active) == KDBUS_CONN_ACTIVE_BIAS);
+
+#ifdef CONFIG_DEBUG_LOCK_ALLOC
+       lock_acquired(&conn->dep_map, _RET_IP_);
+       rwsem_release(&conn->dep_map, 1, _RET_IP_);
+#endif
+
+       cancel_delayed_work_sync(&conn->work);
+       kdbus_policy_remove_owner(&conn->ep->bus->policy_db, conn);
+
+       /* lock order: domain -> bus -> ep -> names -> conn */
+       mutex_lock(&conn->ep->lock);
+       down_write(&bus->conn_rwlock);
+
+       /* remove from bus and endpoint */
+       hash_del(&conn->hentry);
+       list_del(&conn->monitor_entry);
+       list_del(&conn->ep_entry);
+
+       up_write(&bus->conn_rwlock);
+       mutex_unlock(&conn->ep->lock);
+
+       /*
+        * Remove all names associated with this connection; this possibly
+        * moves queued messages back to the activator connection.
+        */
+       kdbus_name_release_all(bus->name_registry, conn);
+
+       /* if we die while other connections wait for our reply, notify them */
+       mutex_lock(&conn->lock);
+       list_for_each_entry_safe(entry, tmp, &conn->queue.msg_list, entry) {
+               if (entry->reply)
+                       kdbus_notify_reply_dead(bus,
+                                               entry->reply->reply_dst->id,
+                                               entry->reply->cookie);
+               kdbus_queue_entry_free(entry);
+       }
+
+       list_for_each_entry_safe(r, r_tmp, &conn->reply_list, entry)
+               kdbus_reply_unlink(r);
+       mutex_unlock(&conn->lock);
+
+       /* lock order: domain -> bus -> ep -> names -> conn */
+       down_read(&bus->conn_rwlock);
+       hash_for_each(bus->conn_hash, i, c, hentry) {
+               mutex_lock(&c->lock);
+               list_for_each_entry_safe(r, r_tmp, &c->reply_list, entry) {
+                       if (r->reply_src == conn) {
+                               if (r->sync) {
+                                       kdbus_sync_reply_wakeup(r, -EPIPE);
+                                       kdbus_reply_unlink(r);
+                                       continue;
+                               }
+
+                               /* send a 'connection dead' notification */
+                               kdbus_notify_reply_dead(bus, c->id, r->cookie);
+                               kdbus_reply_unlink(r);
+                       }
+               }
+               mutex_unlock(&c->lock);
+       }
+       up_read(&bus->conn_rwlock);
+
+       if (!kdbus_conn_is_monitor(conn))
+               kdbus_notify_id_change(bus, KDBUS_ITEM_ID_REMOVE,
+                                      conn->id, conn->flags);
+
+       kdbus_notify_flush(bus);
+
+       return 0;
+}
+
+/**
+ * kdbus_conn_has_name() - check if a connection owns a name
+ * @conn:              Connection
+ * @name:              Well-know name to check for
+ *
+ * The caller must hold the registry lock of conn->ep->bus.
+ *
+ * Return: true if the name is currently owned by the connection
+ */
+bool kdbus_conn_has_name(struct kdbus_conn *conn, const char *name)
+{
+       struct kdbus_name_entry *e;
+
+       lockdep_assert_held(&conn->ep->bus->name_registry->rwlock);
+
+       list_for_each_entry(e, &conn->names_list, conn_entry)
+               if (strcmp(e->name, name) == 0)
+                       return true;
+
+       return false;
+}
+
+struct kdbus_quota {
+       uint32_t memory;
+       uint16_t msgs;
+       uint8_t fds;
+};
+
+/**
+ * kdbus_conn_quota_inc() - increase quota accounting
+ * @c:         connection owning the quota tracking
+ * @u:         user to account for (or NULL for kernel accounting)
+ * @memory:    size of memory to account for
+ * @fds:       number of FDs to account for
+ *
+ * This call manages the quotas on resource @c. That is, it's used if other
+ * users want to use the resources of connection @c, which so far only concerns
+ * the receive queue of the destination.
+ *
+ * This increases the quota-accounting for user @u by @memory bytes and @fds
+ * file descriptors. If the user has already reached the quota limits, this call
+ * will not do any accounting but return a negative error code indicating the
+ * failure.
+ *
+ * Return: 0 on success, negative error code on failure.
+ */
+int kdbus_conn_quota_inc(struct kdbus_conn *c, struct kdbus_user *u,
+                        size_t memory, size_t fds)
+{
+       struct kdbus_quota *quota;
+       size_t available, accounted;
+       unsigned int id;
+
+       /*
+        * Pool Layout:
+        * 50% of a pool is always owned by the connection. It is reserved for
+        * kernel queries, handling received messages and other tasks that are
+        * under control of the pool owner. The other 50% of the pool are used
+        * as incoming queue.
+        * As we optionally support user-space based policies, we need fair
+        * allocation schemes. Furthermore, resource utilization should be
+        * maximized, so only minimal resources stay reserved. However, we need
+        * to adapt to a dynamic number of users, as we cannot know how many
+        * users will talk to a connection. Therefore, the current allocations
+        * works like this:
+        * We limit the number of bytes in a destination's pool per sending
+        * user. The space available for a user is 33% of the unused pool space
+        * (whereas the space used by the user itself is also treated as
+        * 'unused'). This way, we favor users coming first, but keep enough
+        * pool space available for any following users. Given that messages are
+        * dequeued in FIFO order, this should balance nicely if the number of
+        * users grows. At the same time, this algorithm guarantees that the
+        * space available to a connection is reduced dynamically, the more
+        * concurrent users talk to a connection.
+        */
+
+       /* per user-accounting is expensive, so we keep state small */
+       BUILD_BUG_ON(sizeof(quota->memory) != 4);
+       BUILD_BUG_ON(sizeof(quota->msgs) != 2);
+       BUILD_BUG_ON(sizeof(quota->fds) != 1);
+       BUILD_BUG_ON(KDBUS_CONN_MAX_MSGS > U16_MAX);
+       BUILD_BUG_ON(KDBUS_CONN_MAX_FDS_PER_USER > U8_MAX);
+
+       id = u ? u->id : KDBUS_USER_KERNEL_ID;
+       if (id >= c->n_quota) {
+               unsigned int users;
+
+               users = max(KDBUS_ALIGN8(id) + 8, id);
+               quota = krealloc(c->quota, users * sizeof(*quota),
+                                GFP_KERNEL | __GFP_ZERO);
+               if (!quota)
+                       return -ENOMEM;
+
+               c->n_quota = users;
+               c->quota = quota;
+       }
+
+       quota = &c->quota[id];
+       kdbus_pool_accounted(c->pool, &available, &accounted);
+
+       /* half the pool is _always_ reserved for the pool owner */
+       available /= 2;
+
+       /*
+        * Pool owner slices are un-accounted slices; they can claim more
+        * than 50% of the queue. However, the slice we're dealing with here
+        * belong to the incoming queue, hence they are 'accounted' slices
+        * to which the 50%-limit applies.
+        */
+       if (available < accounted)
+               return -ENOBUFS;
+
+       /* 1/3 of the remaining space (including your own memory) */
+       available = (available - accounted + quota->memory) / 3;
+
+       if (available < quota->memory ||
+           available - quota->memory < memory ||
+           quota->memory + memory > U32_MAX)
+               return -ENOBUFS;
+       if (quota->msgs >= KDBUS_CONN_MAX_MSGS)
+               return -ENOBUFS;
+       if (quota->fds + fds < quota->fds ||
+           quota->fds + fds > KDBUS_CONN_MAX_FDS_PER_USER)
+               return -EMFILE;
+
+       quota->memory += memory;
+       quota->fds += fds;
+       ++quota->msgs;
+       return 0;
+}
+
+/**
+ * kdbus_conn_quota_dec() - decrease quota accounting
+ * @c:         connection owning the quota tracking
+ * @u:         user which was accounted for (or NULL for kernel accounting)
+ * @memory:    size of memory which was accounted for
+ * @fds:       number of FDs which were accounted for
+ *
+ * This does the reverse of kdbus_conn_quota_inc(). You have to release any
+ * accounted resources that you called kdbus_conn_quota_inc() for. However, you
+ * must not call kdbus_conn_quota_dec() if the accounting failed (that is,
+ * kdbus_conn_quota_inc() failed).
+ */
+void kdbus_conn_quota_dec(struct kdbus_conn *c, struct kdbus_user *u,
+                         size_t memory, size_t fds)
+{
+       struct kdbus_quota *quota;
+       unsigned int id;
+
+       id = u ? u->id : KDBUS_USER_KERNEL_ID;
+       if (WARN_ON(id >= c->n_quota))
+               return;
+
+       quota = &c->quota[id];
+
+       if (!WARN_ON(quota->msgs == 0))
+               --quota->msgs;
+       if (!WARN_ON(quota->memory < memory))
+               quota->memory -= memory;
+       if (!WARN_ON(quota->fds < fds))
+               quota->fds -= fds;
+}
+
+/**
+ * kdbus_conn_lost_message() - handle lost messages
+ * @c:         connection that lost a message
+ *
+ * kdbus is reliable. That means, we try hard to never lose messages. However,
+ * memory is limited, so we cannot rely on transmissions to never fail.
+ * Therefore, we use quota-limits to let callers know if there unicast message
+ * cannot be transmitted to a peer. This works fine for unicasts, but for
+ * broadcasts we cannot make the caller handle the transmission failure.
+ * Instead, we must let the destination know that it couldn't receive a
+ * broadcast.
+ * As this is an unlikely scenario, we keep it simple. A single lost-counter
+ * remembers the number of lost messages since the last call to RECV. The next
+ * message retrieval will notify the connection that it lost messages since the
+ * last message retrieval and thus should resync its state.
+ */
+void kdbus_conn_lost_message(struct kdbus_conn *c)
+{
+       if (atomic_inc_return(&c->lost_count) == 1)
+               wake_up_interruptible(&c->wait);
+}
+
+/* Callers should take the conn_dst lock */
+static struct kdbus_queue_entry *
+kdbus_conn_entry_make(struct kdbus_conn *conn_dst,
+                     const struct kdbus_kmsg *kmsg,
+                     struct kdbus_user *user)
+{
+       struct kdbus_queue_entry *entry;
+
+       /* The remote connection was disconnected */
+       if (!kdbus_conn_active(conn_dst))
+               return ERR_PTR(-ECONNRESET);
+
+       /*
+        * If the connection does not accept file descriptors but the message
+        * has some attached, refuse it.
+        *
+        * If this is a monitor connection, accept the message. In that
+        * case, all file descriptors will be set to -1 at receive time.
+        */
+       if (!kdbus_conn_is_monitor(conn_dst) &&
+           !(conn_dst->flags & KDBUS_HELLO_ACCEPT_FD) &&
+           kmsg->res && kmsg->res->fds_count > 0)
+               return ERR_PTR(-ECOMM);
+
+       entry = kdbus_queue_entry_new(conn_dst, kmsg, user);
+       if (IS_ERR(entry))
+               return entry;
+
+       return entry;
+}
+
+/*
+ * Synchronously responding to a message, allocate a queue entry
+ * and attach it to the reply tracking object.
+ * The connection's queue will never get to see it.
+ */
+static int kdbus_conn_entry_sync_attach(struct kdbus_conn *conn_dst,
+                                       const struct kdbus_kmsg *kmsg,
+                                       struct kdbus_reply *reply_wake)
+{
+       struct kdbus_queue_entry *entry;
+       int remote_ret;
+       int ret = 0;
+
+       mutex_lock(&reply_wake->reply_dst->lock);
+
+       /*
+        * If we are still waiting then proceed, allocate a queue
+        * entry and attach it to the reply object
+        */
+       if (reply_wake->waiting) {
+               entry = kdbus_conn_entry_make(conn_dst, kmsg,
+                                             reply_wake->reply_src->user);
+               if (IS_ERR(entry))
+                       ret = PTR_ERR(entry);
+               else
+                       /* Attach the entry to the reply object */
+                       reply_wake->queue_entry = entry;
+       } else {
+               ret = -ECONNRESET;
+       }
+
+       /*
+        * Update the reply object and wake up remote peer only
+        * on appropriate return codes
+        *
+        * * -ECOMM: if the replying connection failed with -ECOMM
+        *           then wakeup remote peer with -EREMOTEIO
+        *
+        *           We do this to differenciate between -ECOMM errors
+        *           from the original sender perspective:
+        *           -ECOMM error during the sync send and
+        *           -ECOMM error during the sync reply, this last
+        *           one is rewritten to -EREMOTEIO
+        *
+        * * Wake up on all other return codes.
+        */
+       remote_ret = ret;
+
+       if (ret == -ECOMM)
+               remote_ret = -EREMOTEIO;
+
+       kdbus_sync_reply_wakeup(reply_wake, remote_ret);
+       kdbus_reply_unlink(reply_wake);
+       mutex_unlock(&reply_wake->reply_dst->lock);
+
+       return ret;
+}
+
+/**
+ * kdbus_conn_entry_insert() - enqueue a message into the receiver's pool
+ * @conn_src:          The sending connection
+ * @conn_dst:          The connection to queue into
+ * @kmsg:              The kmsg to queue
+ * @reply:             The reply tracker to attach to the queue entry
+ *
+ * Return: 0 on success. negative error otherwise.
+ */
+int kdbus_conn_entry_insert(struct kdbus_conn *conn_src,
+                           struct kdbus_conn *conn_dst,
+                           const struct kdbus_kmsg *kmsg,
+                           struct kdbus_reply *reply)
+{
+       struct kdbus_queue_entry *entry;
+       int ret;
+
+       kdbus_conn_lock2(conn_src, conn_dst);
+
+       entry = kdbus_conn_entry_make(conn_dst, kmsg,
+                                     conn_src ? conn_src->user : NULL);
+       if (IS_ERR(entry)) {
+               ret = PTR_ERR(entry);
+               goto exit_unlock;
+       }
+
+       if (reply) {
+               kdbus_reply_link(reply);
+               if (!reply->sync)
+                       schedule_delayed_work(&conn_src->work, 0);
+       }
+
+       kdbus_queue_entry_enqueue(entry, reply);
+       wake_up_interruptible(&conn_dst->wait);
+
+       ret = 0;
+
+exit_unlock:
+       kdbus_conn_unlock2(conn_src, conn_dst);
+       return ret;
+}
+
+static int kdbus_conn_wait_reply(struct kdbus_conn *conn_src,
+                                struct kdbus_cmd_send *cmd_send,
+                                struct file *ioctl_file,
+                                struct file *cancel_fd,
+                                struct kdbus_reply *reply_wait,
+                                ktime_t expire)
+{
+       struct kdbus_queue_entry *entry;
+       struct poll_wqueues pwq = {};
+       int ret;
+
+       if (WARN_ON(!reply_wait))
+               return -EIO;
+
+       /*
+        * Block until the reply arrives. reply_wait is left untouched
+        * by the timeout scans that might be conducted for other,
+        * asynchronous replies of conn_src.
+        */
+
+       poll_initwait(&pwq);
+       poll_wait(ioctl_file, &conn_src->wait, &pwq.pt);
+
+       for (;;) {
+               /*
+                * Any of the following conditions will stop our synchronously
+                * blocking SEND command:
+                *
+                * a) The origin sender closed its connection
+                * b) The remote peer answered, setting reply_wait->waiting = 0
+                * c) The cancel FD was written to
+                * d) A signal was received
+                * e) The specified timeout was reached, and none of the above
+                *    conditions kicked in.
+                */
+
+               /*
+                * We have already acquired an active reference when
+                * entering here, but another thread may call
+                * KDBUS_CMD_BYEBYE which does not acquire an active
+                * reference, therefore kdbus_conn_disconnect() will
+                * not wait for us.
+                */
+               if (!kdbus_conn_active(conn_src)) {
+                       ret = -ECONNRESET;
+                       break;
+               }
+
+               /*
+                * After the replying peer unset the waiting variable
+                * it will wake up us.
+                */
+               if (!reply_wait->waiting) {
+                       ret = reply_wait->err;
+                       break;
+               }
+
+               if (cancel_fd) {
+                       unsigned int r;
+
+                       r = cancel_fd->f_op->poll(cancel_fd, &pwq.pt);
+                       if (r & POLLIN) {
+                               ret = -ECANCELED;
+                               break;
+                       }
+               }
+
+               if (signal_pending(current)) {
+                       ret = -EINTR;
+                       break;
+               }
+
+               if (!poll_schedule_timeout(&pwq, TASK_INTERRUPTIBLE,
+                                          &expire, 0)) {
+                       ret = -ETIMEDOUT;
+                       break;
+               }
+
+               /*
+                * Reset the poll worker func, so the waitqueues are not
+                * added to the poll table again. We just reuse what we've
+                * collected earlier for further iterations.
+                */
+               init_poll_funcptr(&pwq.pt, NULL);
+       }
+
+       poll_freewait(&pwq);
+
+       if (ret == -EINTR) {
+               /*
+                * Interrupted system call. Unref the reply object, and pass
+                * the return value down the chain. Mark the reply as
+                * interrupted, so the cleanup work can remove it, but do not
+                * unlink it from the list. Once the syscall restarts, we'll
+                * pick it up and wait on it again.
+                */
+               mutex_lock(&conn_src->lock);
+               reply_wait->interrupted = true;
+               schedule_delayed_work(&conn_src->work, 0);
+               mutex_unlock(&conn_src->lock);
+
+               return -ERESTARTSYS;
+       }
+
+       mutex_lock(&conn_src->lock);
+       reply_wait->waiting = false;
+       entry = reply_wait->queue_entry;
+       if (entry) {
+               ret = kdbus_queue_entry_install(entry,
+                                               &cmd_send->reply.return_flags,
+                                               true);
+               kdbus_pool_slice_publish(entry->slice, &cmd_send->reply.offset,
+                                        &cmd_send->reply.msg_size);
+               kdbus_queue_entry_free(entry);
+       }
+       kdbus_reply_unlink(reply_wait);
+       mutex_unlock(&conn_src->lock);
+
+       return ret;
+}
+
+static int kdbus_pin_dst(struct kdbus_bus *bus,
+                        struct kdbus_kmsg *kmsg,
+                        struct kdbus_name_entry **out_name,
+                        struct kdbus_conn **out_dst)
+{
+       struct kdbus_msg_resources *res = kmsg->res;
+       struct kdbus_name_entry *name = NULL;
+       struct kdbus_conn *dst = NULL;
+       struct kdbus_msg *msg = &kmsg->msg;
+       int ret;
+
+       if (WARN_ON(!res))
+               return -EINVAL;
+
+       lockdep_assert_held(&bus->name_registry->rwlock);
+
+       if (!res->dst_name) {
+               dst = kdbus_bus_find_conn_by_id(bus, msg->dst_id);
+               if (!dst)
+                       return -ENXIO;
+
+               if (!kdbus_conn_is_ordinary(dst)) {
+                       ret = -ENXIO;
+                       goto error;
+               }
+       } else {
+               name = kdbus_name_lookup_unlocked(bus->name_registry,
+                                                 res->dst_name);
+               if (!name)
+                       return -ESRCH;
+
+               /*
+                * If both a name and a connection ID are given as destination
+                * of a message, check that the currently owning connection of
+                * the name matches the specified ID.
+                * This way, we allow userspace to send the message to a
+                * specific connection by ID only if the connection currently
+                * owns the given name.
+                */
+               if (msg->dst_id != KDBUS_DST_ID_NAME &&
+                   msg->dst_id != name->conn->id)
+                       return -EREMCHG;
+
+               if (!name->conn && name->activator)
+                       dst = kdbus_conn_ref(name->activator);
+               else
+                       dst = kdbus_conn_ref(name->conn);
+
+               if ((msg->flags & KDBUS_MSG_NO_AUTO_START) &&
+                   kdbus_conn_is_activator(dst)) {
+                       ret = -EADDRNOTAVAIL;
+                       goto error;
+               }
+
+               /*
+                * Record the sequence number of the registered name; it will
+                * be passed on to the queue, in case messages addressed to a
+                * name need to be moved from or to an activator.
+                */
+               kmsg->dst_name_id = name->name_id;
+       }
+
+       *out_name = name;
+       *out_dst = dst;
+       return 0;
+
+error:
+       kdbus_conn_unref(dst);
+       return ret;
+}
+
+static int kdbus_conn_reply(struct kdbus_conn *src, struct kdbus_kmsg *kmsg)
+{
+       struct kdbus_name_entry *name = NULL;
+       struct kdbus_reply *reply, *wake = NULL;
+       struct kdbus_conn *dst = NULL;
+       struct kdbus_bus *bus = src->ep->bus;
+       u64 attach;
+       int ret;
+
+       if (WARN_ON(kmsg->msg.dst_id == KDBUS_DST_ID_BROADCAST) ||
+           WARN_ON(kmsg->msg.flags & KDBUS_MSG_EXPECT_REPLY) ||
+           WARN_ON(kmsg->msg.flags & KDBUS_MSG_SIGNAL))
+               return -EINVAL;
+
+       /* name-registry must be locked for lookup *and* collecting data */
+       down_read(&bus->name_registry->rwlock);
+
+       /* find and pin destination */
+
+       ret = kdbus_pin_dst(bus, kmsg, &name, &dst);
+       if (ret < 0)
+               goto exit;
+
+       mutex_lock(&dst->lock);
+       reply = kdbus_reply_find(src, dst, kmsg->msg.cookie_reply);
+       if (reply) {
+               if (reply->sync)
+                       wake = kdbus_reply_ref(reply);
+               kdbus_reply_unlink(reply);
+       }
+       mutex_unlock(&dst->lock);
+
+       if (!reply) {
+               ret = -EPERM;
+               goto exit;
+       }
+
+       /* attach metadata */
+
+       attach = kdbus_meta_calc_attach_flags(src, dst);
+
+       if (!src->faked_meta) {
+               ret = kdbus_meta_proc_collect(kmsg->proc_meta, attach);
+               if (ret < 0)
+                       goto exit;
+       }
+
+       ret = kdbus_meta_conn_collect(kmsg->conn_meta, kmsg, src, attach);
+       if (ret < 0)
+               goto exit;
+
+       /* send message */
+
+       kdbus_bus_eavesdrop(bus, src, kmsg);
+
+       if (wake)
+               ret = kdbus_conn_entry_sync_attach(dst, kmsg, wake);
+       else
+               ret = kdbus_conn_entry_insert(src, dst, kmsg, NULL);
+
+exit:
+       up_read(&bus->name_registry->rwlock);
+       kdbus_reply_unref(wake);
+       kdbus_conn_unref(dst);
+       return ret;
+}
+
+static struct kdbus_reply *kdbus_conn_call(struct kdbus_conn *src,
+                                          struct kdbus_kmsg *kmsg,
+                                          ktime_t exp)
+{
+       struct kdbus_name_entry *name = NULL;
+       struct kdbus_reply *wait = NULL;
+       struct kdbus_conn *dst = NULL;
+       struct kdbus_bus *bus = src->ep->bus;
+       u64 attach;
+       int ret;
+
+       if (WARN_ON(kmsg->msg.dst_id == KDBUS_DST_ID_BROADCAST) ||
+           WARN_ON(kmsg->msg.flags & KDBUS_MSG_SIGNAL) ||
+           WARN_ON(!(kmsg->msg.flags & KDBUS_MSG_EXPECT_REPLY)))
+               return ERR_PTR(-EINVAL);
+
+       /* resume previous wait-context, if available */
+
+       mutex_lock(&src->lock);
+       wait = kdbus_reply_find(NULL, src, kmsg->msg.cookie);
+       if (wait) {
+               if (wait->interrupted) {
+                       kdbus_reply_ref(wait);
+                       wait->interrupted = false;
+               } else {
+                       wait = NULL;
+               }
+       }
+       mutex_unlock(&src->lock);
+
+       if (wait)
+               return wait;
+
+       if (ktime_compare(ktime_get(), exp) >= 0)
+               return ERR_PTR(-ETIMEDOUT);
+
+       /* name-registry must be locked for lookup *and* collecting data */
+       down_read(&bus->name_registry->rwlock);
+
+       /* find and pin destination */
+
+       ret = kdbus_pin_dst(bus, kmsg, &name, &dst);
+       if (ret < 0)
+               goto exit;
+
+       if (!kdbus_conn_policy_talk(src, current_cred(), dst)) {
+               ret = -EPERM;
+               goto exit;
+       }
+
+       wait = kdbus_reply_new(dst, src, &kmsg->msg, name, true);
+       if (IS_ERR(wait)) {
+               ret = PTR_ERR(wait);
+               wait = NULL;
+               goto exit;
+       }
+
+       /* attach metadata */
+
+       attach = kdbus_meta_calc_attach_flags(src, dst);
+
+       if (!src->faked_meta) {
+               ret = kdbus_meta_proc_collect(kmsg->proc_meta, attach);
+               if (ret < 0)
+                       goto exit;
+       }
+
+       ret = kdbus_meta_conn_collect(kmsg->conn_meta, kmsg, src, attach);
+       if (ret < 0)
+               goto exit;
+
+       /* send message */
+
+       kdbus_bus_eavesdrop(bus, src, kmsg);
+
+       ret = kdbus_conn_entry_insert(src, dst, kmsg, wait);
+       if (ret < 0)
+               goto exit;
+
+       ret = 0;
+
+exit:
+       up_read(&bus->name_registry->rwlock);
+       if (ret < 0) {
+               kdbus_reply_unref(wait);
+               wait = ERR_PTR(ret);
+       }
+       kdbus_conn_unref(dst);
+       return wait;
+}
+
+static int kdbus_conn_unicast(struct kdbus_conn *src, struct kdbus_kmsg *kmsg)
+{
+       struct kdbus_name_entry *name = NULL;
+       struct kdbus_reply *wait = NULL;
+       struct kdbus_conn *dst = NULL;
+       struct kdbus_bus *bus = src->ep->bus;
+       bool is_signal = (kmsg->msg.flags & KDBUS_MSG_SIGNAL);
+       u64 attach;
+       int ret = 0;
+
+       if (WARN_ON(kmsg->msg.dst_id == KDBUS_DST_ID_BROADCAST) ||
+           WARN_ON(!(kmsg->msg.flags & KDBUS_MSG_EXPECT_REPLY) &&
+                   kmsg->msg.cookie_reply != 0))
+               return -EINVAL;
+
+       /* name-registry must be locked for lookup *and* collecting data */
+       down_read(&bus->name_registry->rwlock);
+
+       /* find and pin destination */
+
+       ret = kdbus_pin_dst(bus, kmsg, &name, &dst);
+       if (ret < 0)
+               goto exit;
+
+       if (is_signal) {
+               /* like broadcasts we eavesdrop even if the msg is dropped */
+               kdbus_bus_eavesdrop(bus, src, kmsg);
+
+               /* drop silently if peer is not interested or not privileged */
+               if (!kdbus_match_db_match_kmsg(dst->match_db, src, kmsg) ||
+                   !kdbus_conn_policy_talk(dst, NULL, src))
+                       goto exit;
+       } else if (!kdbus_conn_policy_talk(src, current_cred(), dst)) {
+               ret = -EPERM;
+               goto exit;
+       } else if (kmsg->msg.flags & KDBUS_MSG_EXPECT_REPLY) {
+               wait = kdbus_reply_new(dst, src, &kmsg->msg, name, false);
+               if (IS_ERR(wait)) {
+                       ret = PTR_ERR(wait);
+                       wait = NULL;
+                       goto exit;
+               }
+       }
+
+       /* attach metadata */
+
+       attach = kdbus_meta_calc_attach_flags(src, dst);
+
+       if (!src->faked_meta) {
+               ret = kdbus_meta_proc_collect(kmsg->proc_meta, attach);
+               if (ret < 0 && !is_signal)
+                       goto exit;
+       }
+
+       ret = kdbus_meta_conn_collect(kmsg->conn_meta, kmsg, src, attach);
+       if (ret < 0 && !is_signal)
+               goto exit;
+
+       /* send message */
+
+       if (!is_signal)
+               kdbus_bus_eavesdrop(bus, src, kmsg);
+
+       ret = kdbus_conn_entry_insert(src, dst, kmsg, wait);
+       if (ret < 0 && !is_signal)
+               goto exit;
+
+       /* signals are treated like broadcasts, recv-errors are ignored */
+       ret = 0;
+
+exit:
+       up_read(&bus->name_registry->rwlock);
+       kdbus_reply_unref(wait);
+       kdbus_conn_unref(dst);
+       return ret;
+}
+
+/**
+ * kdbus_conn_move_messages() - move messages from one connection to another
+ * @conn_dst:          Connection to copy to
+ * @conn_src:          Connection to copy from
+ * @name_id:           Filter for the sequence number of the registered
+ *                     name, 0 means no filtering.
+ *
+ * Move all messages from one connection to another. This is used when
+ * an implementer connection is taking over/giving back a well-known name
+ * from/to an activator connection.
+ */
+void kdbus_conn_move_messages(struct kdbus_conn *conn_dst,
+                             struct kdbus_conn *conn_src,
+                             u64 name_id)
+{
+       struct kdbus_queue_entry *e, *e_tmp;
+       struct kdbus_reply *r, *r_tmp;
+       struct kdbus_bus *bus;
+       struct kdbus_conn *c;
+       LIST_HEAD(msg_list);
+       int i, ret = 0;
+
+       if (WARN_ON(conn_src == conn_dst))
+               return;
+
+       bus = conn_src->ep->bus;
+
+       /* lock order: domain -> bus -> ep -> names -> conn */
+       down_read(&bus->conn_rwlock);
+       hash_for_each(bus->conn_hash, i, c, hentry) {
+               if (c == conn_src || c == conn_dst)
+                       continue;
+
+               mutex_lock(&c->lock);
+               list_for_each_entry_safe(r, r_tmp, &c->reply_list, entry) {
+                       if (r->reply_src != conn_src)
+                               continue;
+
+                       /* filter messages for a specific name */
+                       if (name_id > 0 && r->name_id != name_id)
+                               continue;
+
+                       kdbus_conn_unref(r->reply_src);
+                       r->reply_src = kdbus_conn_ref(conn_dst);
+               }
+               mutex_unlock(&c->lock);
+       }
+       up_read(&bus->conn_rwlock);
+
+       kdbus_conn_lock2(conn_src, conn_dst);
+       list_for_each_entry_safe(e, e_tmp, &conn_src->queue.msg_list, entry) {
+               /* filter messages for a specific name */
+               if (name_id > 0 && e->dst_name_id != name_id)
+                       continue;
+
+               if (!(conn_dst->flags & KDBUS_HELLO_ACCEPT_FD) &&
+                   e->msg_res && e->msg_res->fds_count > 0) {
+                       kdbus_conn_lost_message(conn_dst);
+                       kdbus_queue_entry_free(e);
+                       continue;
+               }
+
+               ret = kdbus_queue_entry_move(e, conn_dst);
+               if (ret < 0) {
+                       kdbus_conn_lost_message(conn_dst);
+                       kdbus_queue_entry_free(e);
+                       continue;
+               }
+       }
+       kdbus_conn_unlock2(conn_src, conn_dst);
+
+       /* wake up poll() */
+       wake_up_interruptible(&conn_dst->wait);
+}
+
+/* query the policy-database for all names of @whom */
+static bool kdbus_conn_policy_query_all(struct kdbus_conn *conn,
+                                       const struct cred *conn_creds,
+                                       struct kdbus_policy_db *db,
+                                       struct kdbus_conn *whom,
+                                       unsigned int access)
+{
+       struct kdbus_name_entry *ne;
+       bool pass = false;
+       int res;
+
+       lockdep_assert_held(&conn->ep->bus->name_registry->rwlock);
+
+       down_read(&db->entries_rwlock);
+       mutex_lock(&whom->lock);
+
+       list_for_each_entry(ne, &whom->names_list, conn_entry) {
+               res = kdbus_policy_query_unlocked(db, conn_creds ? : conn->cred,
+                                                 ne->name,
+                                                 kdbus_strhash(ne->name));
+               if (res >= (int)access) {
+                       pass = true;
+                       break;
+               }
+       }
+
+       mutex_unlock(&whom->lock);
+       up_read(&db->entries_rwlock);
+
+       return pass;
+}
+
+/**
+ * kdbus_conn_policy_own_name() - verify a connection can own the given name
+ * @conn:              Connection
+ * @conn_creds:                Credentials of @conn to use for policy check
+ * @name:              Name
+ *
+ * This verifies that @conn is allowed to acquire the well-known name @name.
+ *
+ * Return: true if allowed, false if not.
+ */
+bool kdbus_conn_policy_own_name(struct kdbus_conn *conn,
+                               const struct cred *conn_creds,
+                               const char *name)
+{
+       unsigned int hash = kdbus_strhash(name);
+       int res;
+
+       if (!conn_creds)
+               conn_creds = conn->cred;
+
+       if (conn->ep->user) {
+               res = kdbus_policy_query(&conn->ep->policy_db, conn_creds,
+                                        name, hash);
+               if (res < KDBUS_POLICY_OWN)
+                       return false;
+       }
+
+       if (conn->privileged)
+               return true;
+
+       res = kdbus_policy_query(&conn->ep->bus->policy_db, conn_creds,
+                                name, hash);
+       return res >= KDBUS_POLICY_OWN;
+}
+
+/**
+ * kdbus_conn_policy_talk() - verify a connection can talk to a given peer
+ * @conn:              Connection that tries to talk
+ * @conn_creds:                Credentials of @conn to use for policy check
+ * @to:                        Connection that is talked to
+ *
+ * This verifies that @conn is allowed to talk to @to.
+ *
+ * Return: true if allowed, false if not.
+ */
+bool kdbus_conn_policy_talk(struct kdbus_conn *conn,
+                           const struct cred *conn_creds,
+                           struct kdbus_conn *to)
+{
+       if (!conn_creds)
+               conn_creds = conn->cred;
+
+       if (conn->ep->user &&
+           !kdbus_conn_policy_query_all(conn, conn_creds, &conn->ep->policy_db,
+                                        to, KDBUS_POLICY_TALK))
+               return false;
+
+       if (conn->privileged)
+               return true;
+       if (uid_eq(conn_creds->euid, to->cred->uid))
+               return true;
+
+       return kdbus_conn_policy_query_all(conn, conn_creds,
+                                          &conn->ep->bus->policy_db, to,
+                                          KDBUS_POLICY_TALK);
+}
+
+/**
+ * kdbus_conn_policy_see_name_unlocked() - verify a connection can see a given
+ *                                        name
+ * @conn:              Connection
+ * @conn_creds:                Credentials of @conn to use for policy check
+ * @name:              Name
+ *
+ * This verifies that @conn is allowed to see the well-known name @name. Caller
+ * must hold policy-lock.
+ *
+ * Return: true if allowed, false if not.
+ */
+bool kdbus_conn_policy_see_name_unlocked(struct kdbus_conn *conn,
+                                        const struct cred *conn_creds,
+                                        const char *name)
+{
+       int res;
+
+       /*
+        * By default, all names are visible on a bus. SEE policies can only be
+        * installed on custom endpoints, where by default no name is visible.
+        */
+       if (!conn->ep->user)
+               return true;
+
+       res = kdbus_policy_query_unlocked(&conn->ep->policy_db,
+                                         conn_creds ? : conn->cred,
+                                         name, kdbus_strhash(name));
+       return res >= KDBUS_POLICY_SEE;
+}
+
+static bool kdbus_conn_policy_see_name(struct kdbus_conn *conn,
+                                      const struct cred *conn_creds,
+                                      const char *name)
+{
+       bool res;
+
+       down_read(&conn->ep->policy_db.entries_rwlock);
+       res = kdbus_conn_policy_see_name_unlocked(conn, conn_creds, name);
+       up_read(&conn->ep->policy_db.entries_rwlock);
+
+       return res;
+}
+
+static bool kdbus_conn_policy_see(struct kdbus_conn *conn,
+                                 const struct cred *conn_creds,
+                                 struct kdbus_conn *whom)
+{
+       /*
+        * By default, all names are visible on a bus, so a connection can
+        * always see other connections. SEE policies can only be installed on
+        * custom endpoints, where by default no name is visible and we hide
+        * peers from each other, unless you see at least _one_ name of the
+        * peer.
+        */
+       return !conn->ep->user ||
+              kdbus_conn_policy_query_all(conn, conn_creds,
+                                          &conn->ep->policy_db, whom,
+                                          KDBUS_POLICY_SEE);
+}
+
+/**
+ * kdbus_conn_policy_see_notification() - verify a connection is allowed to
+ *                                       receive a given kernel notification
+ * @conn:              Connection
+ * @conn_creds:                Credentials of @conn to use for policy check
+ * @kmsg:              The message carrying the notification
+ *
+ * This checks whether @conn is allowed to see the kernel notification @kmsg.
+ *
+ * Return: true if allowed, false if not.
+ */
+bool kdbus_conn_policy_see_notification(struct kdbus_conn *conn,
+                                       const struct cred *conn_creds,
+                                       const struct kdbus_kmsg *kmsg)
+{
+       if (WARN_ON(kmsg->msg.src_id != KDBUS_SRC_ID_KERNEL))
+               return false;
+
+       /*
+        * Depending on the notification type, broadcasted kernel notifications
+        * have to be filtered:
+        *
+        * KDBUS_ITEM_NAME_{ADD,REMOVE,CHANGE}: This notification is forwarded
+        *     to a peer if, and only if, that peer can see the name this
+        *     notification is for.
+        *
+        * KDBUS_ITEM_ID_{ADD,REMOVE}: As new peers cannot have names, and all
+        *     names are dropped before a peer is removed, those notifications
+        *     cannot be seen on custom endpoints. Thus, we only pass them
+        *     through on default endpoints.
+        */
+
+       switch (kmsg->notify_type) {
+       case KDBUS_ITEM_NAME_ADD:
+       case KDBUS_ITEM_NAME_REMOVE:
+       case KDBUS_ITEM_NAME_CHANGE:
+               return kdbus_conn_policy_see_name(conn, conn_creds,
+                                                 kmsg->notify_name);
+
+       case KDBUS_ITEM_ID_ADD:
+       case KDBUS_ITEM_ID_REMOVE:
+               return !conn->ep->user;
+
+       default:
+               WARN(1, "Invalid type for notification broadcast: %llu\n",
+                    (unsigned long long)kmsg->notify_type);
+               return false;
+       }
+}
+
+/**
+ * kdbus_cmd_hello() - handle KDBUS_CMD_HELLO
+ * @ep:                        Endpoint to operate on
+ * @privileged:                Whether the caller is privileged
+ * @argp:              Command payload
+ *
+ * Return: Newly created connection on success, ERR_PTR on failure.
+ */
+struct kdbus_conn *kdbus_cmd_hello(struct kdbus_ep *ep, bool privileged,
+                                  void __user *argp)
+{
+       struct kdbus_cmd_hello *cmd;
+       struct kdbus_conn *c = NULL;
+       const char *item_name;
+       int ret;
+
+       struct kdbus_arg argv[] = {
+               { .type = KDBUS_ITEM_NEGOTIATE },
+               { .type = KDBUS_ITEM_NAME },
+               { .type = KDBUS_ITEM_CREDS },
+               { .type = KDBUS_ITEM_PIDS },
+               { .type = KDBUS_ITEM_SECLABEL },
+               { .type = KDBUS_ITEM_CONN_DESCRIPTION },
+               { .type = KDBUS_ITEM_POLICY_ACCESS, .multiple = true },
+       };
+       struct kdbus_args args = {
+               .allowed_flags = KDBUS_FLAG_NEGOTIATE |
+                                KDBUS_HELLO_ACCEPT_FD |
+                                KDBUS_HELLO_ACTIVATOR |
+                                KDBUS_HELLO_POLICY_HOLDER |
+                                KDBUS_HELLO_MONITOR,
+               .argv = argv,
+               .argc = ARRAY_SIZE(argv),
+       };
+
+       ret = kdbus_args_parse(&args, argp, &cmd);
+       if (ret < 0)
+               return ERR_PTR(ret);
+       if (ret > 0)
+               return NULL;
+
+       item_name = argv[1].item ? argv[1].item->str : NULL;
+
+       c = kdbus_conn_new(ep, privileged, cmd, item_name,
+                          argv[2].item ? &argv[2].item->creds : NULL,
+                          argv[3].item ? &argv[3].item->pids : NULL,
+                          argv[4].item ? argv[4].item->str : NULL,
+                          argv[5].item ? argv[5].item->str : NULL);
+       if (IS_ERR(c)) {
+               ret = PTR_ERR(c);
+               c = NULL;
+               goto exit;
+       }
+
+       ret = kdbus_conn_connect(c, item_name);
+       if (ret < 0)
+               goto exit;
+
+       if (kdbus_conn_is_activator(c) || kdbus_conn_is_policy_holder(c)) {
+               ret = kdbus_conn_acquire(c);
+               if (ret < 0)
+                       goto exit;
+
+               ret = kdbus_policy_set(&c->ep->bus->policy_db, args.items,
+                                      args.items_size, 1,
+                                      kdbus_conn_is_policy_holder(c), c);
+               kdbus_conn_release(c);
+               if (ret < 0)
+                       goto exit;
+       }
+
+       if (copy_to_user(argp, cmd, sizeof(*cmd)))
+               ret = -EFAULT;
+
+exit:
+       ret = kdbus_args_clear(&args, ret);
+       if (ret < 0) {
+               if (c) {
+                       kdbus_conn_disconnect(c, false);
+                       kdbus_conn_unref(c);
+               }
+               return ERR_PTR(ret);
+       }
+       return c;
+}
+
+/**
+ * kdbus_cmd_byebye_unlocked() - handle KDBUS_CMD_BYEBYE
+ * @conn:              connection to operate on
+ * @argp:              command payload
+ *
+ * The caller must not hold any active reference to @conn or this will deadlock.
+ *
+ * Return: 0 on success, negative error code on failure.
+ */
+int kdbus_cmd_byebye_unlocked(struct kdbus_conn *conn, void __user *argp)
+{
+       struct kdbus_cmd *cmd;
+       int ret;
+
+       struct kdbus_arg argv[] = {
+               { .type = KDBUS_ITEM_NEGOTIATE },
+       };
+       struct kdbus_args args = {
+               .allowed_flags = KDBUS_FLAG_NEGOTIATE,
+               .argv = argv,
+               .argc = ARRAY_SIZE(argv),
+       };
+
+       if (!kdbus_conn_is_ordinary(conn))
+               return -EOPNOTSUPP;
+
+       ret = kdbus_args_parse(&args, argp, &cmd);
+       if (ret != 0)
+               return ret;
+
+       ret = kdbus_conn_disconnect(conn, true);
+       return kdbus_args_clear(&args, ret);
+}
+
+/**
+ * kdbus_cmd_conn_info() - handle KDBUS_CMD_CONN_INFO
+ * @conn:              connection to operate on
+ * @argp:              command payload
+ *
+ * Return: 0 on success, negative error code on failure.
+ */
+int kdbus_cmd_conn_info(struct kdbus_conn *conn, void __user *argp)
+{
+       struct kdbus_meta_conn *conn_meta = NULL;
+       struct kdbus_pool_slice *slice = NULL;
+       struct kdbus_name_entry *entry = NULL;
+       struct kdbus_conn *owner_conn = NULL;
+       struct kdbus_info info = {};
+       struct kdbus_cmd_info *cmd;
+       struct kdbus_bus *bus = conn->ep->bus;
+       struct kvec kvec;
+       size_t meta_size;
+       const char *name;
+       u64 attach_flags;
+       int ret;
+
+       struct kdbus_arg argv[] = {
+               { .type = KDBUS_ITEM_NEGOTIATE },
+               { .type = KDBUS_ITEM_NAME },
+       };
+       struct kdbus_args args = {
+               .allowed_flags = KDBUS_FLAG_NEGOTIATE,
+               .argv = argv,
+               .argc = ARRAY_SIZE(argv),
+       };
+
+       ret = kdbus_args_parse(&args, argp, &cmd);
+       if (ret != 0)
+               return ret;
+
+       /* registry must be held throughout lookup *and* collecting data */
+       down_read(&bus->name_registry->rwlock);
+
+       ret = kdbus_sanitize_attach_flags(cmd->attach_flags, &attach_flags);
+       if (ret < 0)
+               goto exit;
+
+       name = argv[1].item ? argv[1].item->str : NULL;
+
+       if (name) {
+               entry = kdbus_name_lookup_unlocked(bus->name_registry, name);
+               if (!entry || !entry->conn ||
+                   !kdbus_conn_policy_see_name(conn, current_cred(), name) ||
+                   (cmd->id != 0 && entry->conn->id != cmd->id)) {
+                       /* pretend a name doesn't exist if you cannot see it */
+                       ret = -ESRCH;
+                       goto exit;
+               }
+
+               owner_conn = kdbus_conn_ref(entry->conn);
+       } else if (cmd->id > 0) {
+               owner_conn = kdbus_bus_find_conn_by_id(bus, cmd->id);
+               if (!owner_conn || !kdbus_conn_policy_see(conn, current_cred(),
+                                                         owner_conn)) {
+                       /* pretend an id doesn't exist if you cannot see it */
+                       ret = -ENXIO;
+                       goto exit;
+               }
+       } else {
+               ret = -EINVAL;
+               goto exit;
+       }
+
+       info.id = owner_conn->id;
+       info.flags = owner_conn->flags;
+       kdbus_kvec_set(&kvec, &info, sizeof(info), &info.size);
+
+       attach_flags &= atomic64_read(&owner_conn->attach_flags_send);
+
+       conn_meta = kdbus_meta_conn_new();
+       if (IS_ERR(conn_meta)) {
+               ret = PTR_ERR(conn_meta);
+               conn_meta = NULL;
+               goto exit;
+       }
+
+       ret = kdbus_meta_conn_collect(conn_meta, NULL, owner_conn,
+                                     attach_flags);
+       if (ret < 0)
+               goto exit;
+
+       ret = kdbus_meta_export_prepare(owner_conn->meta, conn_meta,
+                                       &attach_flags, &meta_size);
+       if (ret < 0)
+               goto exit;
+
+       slice = kdbus_pool_slice_alloc(conn->pool,
+                                      info.size + meta_size, false);
+       if (IS_ERR(slice)) {
+               ret = PTR_ERR(slice);
+               slice = NULL;
+               goto exit;
+       }
+
+       ret = kdbus_meta_export(owner_conn->meta, conn_meta, attach_flags,
+                               slice, sizeof(info), &meta_size);
+       if (ret < 0)
+               goto exit;
+
+       info.size += meta_size;
+
+       ret = kdbus_pool_slice_copy_kvec(slice, 0, &kvec, 1, sizeof(info));
+       if (ret < 0)
+               goto exit;
+
+       kdbus_pool_slice_publish(slice, &cmd->offset, &cmd->info_size);
+
+       if (kdbus_member_set_user(&cmd->offset, argp, typeof(*cmd), offset) ||
+           kdbus_member_set_user(&cmd->info_size, argp,
+                                 typeof(*cmd), info_size)) {
+               ret = -EFAULT;
+               goto exit;
+       }
+
+       ret = 0;
+
+exit:
+       up_read(&bus->name_registry->rwlock);
+       kdbus_pool_slice_release(slice);
+       kdbus_meta_conn_unref(conn_meta);
+       kdbus_conn_unref(owner_conn);
+       return kdbus_args_clear(&args, ret);
+}
+
+/**
+ * kdbus_cmd_update() - handle KDBUS_CMD_UPDATE
+ * @conn:              connection to operate on
+ * @argp:              command payload
+ *
+ * Return: 0 on success, negative error code on failure.
+ */
+int kdbus_cmd_update(struct kdbus_conn *conn, void __user *argp)
+{
+       struct kdbus_bus *bus = conn->ep->bus;
+       struct kdbus_item *item_policy;
+       u64 *item_attach_send = NULL;
+       u64 *item_attach_recv = NULL;
+       struct kdbus_cmd *cmd;
+       u64 attach_send;
+       u64 attach_recv;
+       int ret;
+
+       struct kdbus_arg argv[] = {
+               { .type = KDBUS_ITEM_NEGOTIATE },
+               { .type = KDBUS_ITEM_ATTACH_FLAGS_SEND },
+               { .type = KDBUS_ITEM_ATTACH_FLAGS_RECV },
+               { .type = KDBUS_ITEM_NAME, .multiple = true },
+               { .type = KDBUS_ITEM_POLICY_ACCESS, .multiple = true },
+       };
+       struct kdbus_args args = {
+               .allowed_flags = KDBUS_FLAG_NEGOTIATE,
+               .argv = argv,
+               .argc = ARRAY_SIZE(argv),
+       };
+
+       ret = kdbus_args_parse(&args, argp, &cmd);
+       if (ret != 0)
+               return ret;
+
+       item_attach_send = argv[1].item ? &argv[1].item->data64[0] : NULL;
+       item_attach_recv = argv[2].item ? &argv[2].item->data64[0] : NULL;
+       item_policy = argv[3].item ? : argv[4].item;
+
+       if (item_attach_send) {
+               if (!kdbus_conn_is_ordinary(conn) &&
+                   !kdbus_conn_is_monitor(conn)) {
+                       ret = -EOPNOTSUPP;
+                       goto exit;
+               }
+
+               ret = kdbus_sanitize_attach_flags(*item_attach_send,
+                                                 &attach_send);
+               if (ret < 0)
+                       goto exit;
+
+               if (bus->attach_flags_req & ~attach_send) {
+                       ret = -EINVAL;
+                       goto exit;
+               }
+       }
+
+       if (item_attach_recv) {
+               if (!kdbus_conn_is_ordinary(conn) &&
+                   !kdbus_conn_is_monitor(conn) &&
+                   !kdbus_conn_is_activator(conn)) {
+                       ret = -EOPNOTSUPP;
+                       goto exit;
+               }
+
+               ret = kdbus_sanitize_attach_flags(*item_attach_recv,
+                                                 &attach_recv);
+               if (ret < 0)
+                       goto exit;
+       }
+
+       if (item_policy && !kdbus_conn_is_policy_holder(conn)) {
+               ret = -EOPNOTSUPP;
+               goto exit;
+       }
+
+       /* now that we verified the input, update the connection */
+
+       if (item_policy) {
+               ret = kdbus_policy_set(&conn->ep->bus->policy_db, cmd->items,
+                                      KDBUS_ITEMS_SIZE(cmd, items),
+                                      1, true, conn);
+               if (ret < 0)
+                       goto exit;
+       }
+
+       if (item_attach_send)
+               atomic64_set(&conn->attach_flags_send, attach_send);
+
+       if (item_attach_recv)
+               atomic64_set(&conn->attach_flags_recv, attach_recv);
+
+exit:
+       return kdbus_args_clear(&args, ret);
+}
+
+/**
+ * kdbus_cmd_send() - handle KDBUS_CMD_SEND
+ * @conn:              connection to operate on
+ * @f:                 file this command was called on
+ * @argp:              command payload
+ *
+ * Return: 0 on success, negative error code on failure.
+ */
+int kdbus_cmd_send(struct kdbus_conn *conn, struct file *f, void __user *argp)
+{
+       struct kdbus_cmd_send *cmd;
+       struct kdbus_kmsg *kmsg = NULL;
+       struct file *cancel_fd = NULL;
+       int ret;
+
+       struct kdbus_arg argv[] = {
+               { .type = KDBUS_ITEM_NEGOTIATE },
+               { .type = KDBUS_ITEM_CANCEL_FD },
+       };
+       struct kdbus_args args = {
+               .allowed_flags = KDBUS_FLAG_NEGOTIATE |
+                                KDBUS_SEND_SYNC_REPLY,
+               .argv = argv,
+               .argc = ARRAY_SIZE(argv),
+       };
+
+       if (!kdbus_conn_is_ordinary(conn))
+               return -EOPNOTSUPP;
+
+       ret = kdbus_args_parse(&args, argp, &cmd);
+       if (ret != 0)
+               return ret;
+
+       cmd->reply.return_flags = 0;
+       kdbus_pool_publish_empty(conn->pool, &cmd->reply.offset,
+                                &cmd->reply.msg_size);
+
+       if (argv[1].item) {
+               cancel_fd = fget(argv[1].item->fds[0]);
+               if (IS_ERR(cancel_fd)) {
+                       ret = PTR_ERR(cancel_fd);
+                       cancel_fd = NULL;
+                       goto exit;
+               }
+
+               if (!cancel_fd->f_op->poll) {
+                       ret = -EINVAL;
+                       goto exit;
+               }
+       }
+
+       kmsg = kdbus_kmsg_new_from_cmd(conn, cmd);
+       if (IS_ERR(kmsg)) {
+               ret = PTR_ERR(kmsg);
+               kmsg = NULL;
+               goto exit;
+       }
+
+       if (kmsg->msg.dst_id == KDBUS_DST_ID_BROADCAST) {
+               down_read(&conn->ep->bus->name_registry->rwlock);
+               kdbus_bus_broadcast(conn->ep->bus, conn, kmsg);
+               up_read(&conn->ep->bus->name_registry->rwlock);
+       } else if (cmd->flags & KDBUS_SEND_SYNC_REPLY) {
+               struct kdbus_reply *r;
+               ktime_t exp;
+
+               exp = ns_to_ktime(kmsg->msg.timeout_ns);
+               r = kdbus_conn_call(conn, kmsg, exp);
+               if (IS_ERR(r)) {
+                       ret = PTR_ERR(r);
+                       goto exit;
+               }
+
+               ret = kdbus_conn_wait_reply(conn, cmd, f, cancel_fd, r, exp);
+               kdbus_reply_unref(r);
+               if (ret < 0)
+                       goto exit;
+       } else if ((kmsg->msg.flags & KDBUS_MSG_EXPECT_REPLY) ||
+                  kmsg->msg.cookie_reply == 0) {
+               ret = kdbus_conn_unicast(conn, kmsg);
+               if (ret < 0)
+                       goto exit;
+       } else {
+               ret = kdbus_conn_reply(conn, kmsg);
+               if (ret < 0)
+                       goto exit;
+       }
+
+       if (kdbus_member_set_user(&cmd->reply, argp, typeof(*cmd), reply))
+               ret = -EFAULT;
+
+exit:
+       if (cancel_fd)
+               fput(cancel_fd);
+       kdbus_kmsg_free(kmsg);
+       return kdbus_args_clear(&args, ret);
+}
+
+/**
+ * kdbus_cmd_recv() - handle KDBUS_CMD_RECV
+ * @conn:              connection to operate on
+ * @argp:              command payload
+ *
+ * Return: 0 on success, negative error code on failure.
+ */
+int kdbus_cmd_recv(struct kdbus_conn *conn, void __user *argp)
+{
+       struct kdbus_queue_entry *entry;
+       struct kdbus_cmd_recv *cmd;
+       int ret;
+
+       struct kdbus_arg argv[] = {
+               { .type = KDBUS_ITEM_NEGOTIATE },
+       };
+       struct kdbus_args args = {
+               .allowed_flags = KDBUS_FLAG_NEGOTIATE |
+                                KDBUS_RECV_PEEK |
+                                KDBUS_RECV_DROP |
+                                KDBUS_RECV_USE_PRIORITY,
+               .argv = argv,
+               .argc = ARRAY_SIZE(argv),
+       };
+
+       if (!kdbus_conn_is_ordinary(conn) &&
+           !kdbus_conn_is_monitor(conn) &&
+           !kdbus_conn_is_activator(conn))
+               return -EOPNOTSUPP;
+
+       ret = kdbus_args_parse(&args, argp, &cmd);
+       if (ret != 0)
+               return ret;
+
+       cmd->dropped_msgs = 0;
+       cmd->msg.return_flags = 0;
+       kdbus_pool_publish_empty(conn->pool, &cmd->msg.offset,
+                                &cmd->msg.msg_size);
+
+       /* DROP+priority is not realiably, so prevent it */
+       if ((cmd->flags & KDBUS_RECV_DROP) &&
+           (cmd->flags & KDBUS_RECV_USE_PRIORITY)) {
+               ret = -EINVAL;
+               goto exit;
+       }
+
+       mutex_lock(&conn->lock);
+
+       entry = kdbus_queue_peek(&conn->queue, cmd->priority,
+                                cmd->flags & KDBUS_RECV_USE_PRIORITY);
+       if (!entry) {
+               mutex_unlock(&conn->lock);
+               ret = -EAGAIN;
+       } else if (cmd->flags & KDBUS_RECV_DROP) {
+               struct kdbus_reply *reply = kdbus_reply_ref(entry->reply);
+
+               kdbus_queue_entry_free(entry);
+
+               mutex_unlock(&conn->lock);
+
+               if (reply) {
+                       mutex_lock(&reply->reply_dst->lock);
+                       if (!list_empty(&reply->entry)) {
+                               kdbus_reply_unlink(reply);
+                               if (reply->sync)
+                                       kdbus_sync_reply_wakeup(reply, -EPIPE);
+                               else
+                                       kdbus_notify_reply_dead(conn->ep->bus,
+                                                       reply->reply_dst->id,
+                                                       reply->cookie);
+                       }
+                       mutex_unlock(&reply->reply_dst->lock);
+                       kdbus_notify_flush(conn->ep->bus);
+               }
+
+               kdbus_reply_unref(reply);
+       } else {
+               bool install_fds;
+
+               /*
+                * PEEK just returns the location of the next message. Do not
+                * install FDs nor memfds nor anything else. The only
+                * information of interest should be the message header and
+                * metadata. Any FD numbers in the payload is undefined for
+                * PEEK'ed messages.
+                * Also make sure to never install fds into a connection that
+                * has refused to receive any. Ordinary connections will not get
+                * messages with FDs queued (the receiver will get -ECOMM), but
+                * eavesdroppers might.
+                */
+               install_fds = (conn->flags & KDBUS_HELLO_ACCEPT_FD) &&
+                             !(cmd->flags & KDBUS_RECV_PEEK);
+
+               ret = kdbus_queue_entry_install(entry,
+                                               &cmd->msg.return_flags,
+                                               install_fds);
+               if (ret < 0) {
+                       mutex_unlock(&conn->lock);
+                       goto exit;
+               }
+
+               kdbus_pool_slice_publish(entry->slice, &cmd->msg.offset,
+                                        &cmd->msg.msg_size);
+
+               if (!(cmd->flags & KDBUS_RECV_PEEK))
+                       kdbus_queue_entry_free(entry);
+
+               mutex_unlock(&conn->lock);
+       }
+
+       cmd->dropped_msgs = atomic_xchg(&conn->lost_count, 0);
+       if (cmd->dropped_msgs > 0)
+               cmd->return_flags |= KDBUS_RECV_RETURN_DROPPED_MSGS;
+
+       if (kdbus_member_set_user(&cmd->msg, argp, typeof(*cmd), msg) ||
+           kdbus_member_set_user(&cmd->dropped_msgs, argp, typeof(*cmd),
+                                 dropped_msgs))
+               ret = -EFAULT;
+
+exit:
+       return kdbus_args_clear(&args, ret);
+}
+
+/**
+ * kdbus_cmd_free() - handle KDBUS_CMD_FREE
+ * @conn:              connection to operate on
+ * @argp:              command payload
+ *
+ * Return: 0 on success, negative error code on failure.
+ */
+int kdbus_cmd_free(struct kdbus_conn *conn, void __user *argp)
+{
+       struct kdbus_cmd_free *cmd;
+       int ret;
+
+       struct kdbus_arg argv[] = {
+               { .type = KDBUS_ITEM_NEGOTIATE },
+       };
+       struct kdbus_args args = {
+               .allowed_flags = KDBUS_FLAG_NEGOTIATE,
+               .argv = argv,
+               .argc = ARRAY_SIZE(argv),
+       };
+
+       if (!kdbus_conn_is_ordinary(conn) &&
+           !kdbus_conn_is_monitor(conn) &&
+           !kdbus_conn_is_activator(conn))
+               return -EOPNOTSUPP;
+
+       ret = kdbus_args_parse(&args, argp, &cmd);
+       if (ret != 0)
+               return ret;
+
+       ret = kdbus_pool_release_offset(conn->pool, cmd->offset);
+
+       return kdbus_args_clear(&args, ret);
+}
diff --git a/ipc/kdbus/connection.h b/ipc/kdbus/connection.h
new file mode 100644 (file)
index 0000000..d1ffe90
--- /dev/null
@@ -0,0 +1,257 @@
+/*
+ * Copyright (C) 2013-2015 Kay Sievers
+ * Copyright (C) 2013-2015 Greg Kroah-Hartman <gregkh@linuxfoundation.org>
+ * Copyright (C) 2013-2015 Daniel Mack <daniel@zonque.org>
+ * Copyright (C) 2013-2015 David Herrmann <dh.herrmann@gmail.com>
+ * Copyright (C) 2013-2015 Linux Foundation
+ * Copyright (C) 2014-2015 Djalal Harouni
+ *
+ * kdbus is free software; you can redistribute it and/or modify it under
+ * the terms of the GNU Lesser General Public License as published by the
+ * Free Software Foundation; either version 2.1 of the License, or (at
+ * your option) any later version.
+ */
+
+#ifndef __KDBUS_CONNECTION_H
+#define __KDBUS_CONNECTION_H
+
+#include <linux/atomic.h>
+#include <linux/kref.h>
+#include <linux/lockdep.h>
+#include <linux/path.h>
+
+#include "limits.h"
+#include "metadata.h"
+#include "pool.h"
+#include "queue.h"
+#include "util.h"
+
+#define KDBUS_HELLO_SPECIAL_CONN       (KDBUS_HELLO_ACTIVATOR | \
+                                        KDBUS_HELLO_POLICY_HOLDER | \
+                                        KDBUS_HELLO_MONITOR)
+
+struct kdbus_quota;
+struct kdbus_kmsg;
+
+/**
+ * struct kdbus_conn - connection to a bus
+ * @kref:              Reference count
+ * @active:            Active references to the connection
+ * @id:                        Connection ID
+ * @flags:             KDBUS_HELLO_* flags
+ * @attach_flags_send: KDBUS_ATTACH_* flags for sending
+ * @attach_flags_recv: KDBUS_ATTACH_* flags for receiving
+ * @description:       Human-readable connection description, used for
+ *                     debugging. This field is only set when the
+ *                     connection is created.
+ * @ep:                        The endpoint this connection belongs to
+ * @lock:              Connection data lock
+ * @hentry:            Entry in ID <-> connection map
+ * @ep_entry:          Entry in endpoint
+ * @monitor_entry:     Entry in monitor, if the connection is a monitor
+ * @reply_list:                List of connections this connection should
+ *                     reply to
+ * @work:              Delayed work to handle timeouts
+ *                     activator for
+ * @match_db:          Subscription filter to broadcast messages
+ * @meta:              Active connection creator's metadata/credentials,
+ *                     either from the handle or from HELLO
+ * @pool:              The user's buffer to receive messages
+ * @user:              Owner of the connection
+ * @cred:              The credentials of the connection at creation time
+ * @name_count:                Number of owned well-known names
+ * @request_count:     Number of pending requests issued by this
+ *                     connection that are waiting for replies from
+ *                     other peers
+ * @lost_count:                Number of lost broadcast messages
+ * @wait:              Wake up this endpoint
+ * @queue:             The message queue associated with this connection
+ * @quota:             Array of per-user quota indexed by user->id
+ * @n_quota:           Number of elements in quota array
+ * @activator_of:      Well-known name entry this connection acts as an
+ * @names_list:                List of well-known names
+ * @names_queue_list:  Well-known names this connection waits for
+ * @privileged:                Whether this connection is privileged on the bus
+ * @faked_meta:                Whether the metadata was faked on HELLO
+ */
+struct kdbus_conn {
+       struct kref kref;
+       atomic_t active;
+#ifdef CONFIG_DEBUG_LOCK_ALLOC
+       struct lockdep_map dep_map;
+#endif
+       u64 id;
+       u64 flags;
+       atomic64_t attach_flags_send;
+       atomic64_t attach_flags_recv;
+       const char *description;
+       struct kdbus_ep *ep;
+       struct mutex lock;
+       struct hlist_node hentry;
+       struct list_head ep_entry;
+       struct list_head monitor_entry;
+       struct list_head reply_list;
+       struct delayed_work work;
+       struct kdbus_match_db *match_db;
+       struct kdbus_meta_proc *meta;
+       struct kdbus_pool *pool;
+       struct kdbus_user *user;
+       const struct cred *cred;
+       atomic_t name_count;
+       atomic_t request_count;
+       atomic_t lost_count;
+       wait_queue_head_t wait;
+       struct kdbus_queue queue;
+
+       struct kdbus_quota *quota;
+       unsigned int n_quota;
+
+       /* protected by registry->rwlock */
+       struct kdbus_name_entry *activator_of;
+       struct list_head names_list;
+       struct list_head names_queue_list;
+
+       bool privileged:1;
+       bool faked_meta:1;
+};
+
+struct kdbus_conn *kdbus_conn_ref(struct kdbus_conn *conn);
+struct kdbus_conn *kdbus_conn_unref(struct kdbus_conn *conn);
+bool kdbus_conn_active(const struct kdbus_conn *conn);
+int kdbus_conn_acquire(struct kdbus_conn *conn);
+void kdbus_conn_release(struct kdbus_conn *conn);
+int kdbus_conn_disconnect(struct kdbus_conn *conn, bool ensure_queue_empty);
+bool kdbus_conn_has_name(struct kdbus_conn *conn, const char *name);
+int kdbus_conn_quota_inc(struct kdbus_conn *c, struct kdbus_user *u,
+                        size_t memory, size_t fds);
+void kdbus_conn_quota_dec(struct kdbus_conn *c, struct kdbus_user *u,
+                         size_t memory, size_t fds);
+void kdbus_conn_lost_message(struct kdbus_conn *c);
+int kdbus_conn_entry_insert(struct kdbus_conn *conn_src,
+                           struct kdbus_conn *conn_dst,
+                           const struct kdbus_kmsg *kmsg,
+                           struct kdbus_reply *reply);
+void kdbus_conn_move_messages(struct kdbus_conn *conn_dst,
+                             struct kdbus_conn *conn_src,
+                             u64 name_id);
+
+/* policy */
+bool kdbus_conn_policy_own_name(struct kdbus_conn *conn,
+                               const struct cred *conn_creds,
+                               const char *name);
+bool kdbus_conn_policy_talk(struct kdbus_conn *conn,
+                           const struct cred *conn_creds,
+                           struct kdbus_conn *to);
+bool kdbus_conn_policy_see_name_unlocked(struct kdbus_conn *conn,
+                                        const struct cred *curr_creds,
+                                        const char *name);
+bool kdbus_conn_policy_see_notification(struct kdbus_conn *conn,
+                                       const struct cred *curr_creds,
+                                       const struct kdbus_kmsg *kmsg);
+
+/* command dispatcher */
+struct kdbus_conn *kdbus_cmd_hello(struct kdbus_ep *ep, bool privileged,
+                                  void __user *argp);
+int kdbus_cmd_byebye_unlocked(struct kdbus_conn *conn, void __user *argp);
+int kdbus_cmd_conn_info(struct kdbus_conn *conn, void __user *argp);
+int kdbus_cmd_update(struct kdbus_conn *conn, void __user *argp);
+int kdbus_cmd_send(struct kdbus_conn *conn, struct file *f, void __user *argp);
+int kdbus_cmd_recv(struct kdbus_conn *conn, void __user *argp);
+int kdbus_cmd_free(struct kdbus_conn *conn, void __user *argp);
+
+/**
+ * kdbus_conn_is_ordinary() - Check if connection is ordinary
+ * @conn:              The connection to check
+ *
+ * Return: Non-zero if the connection is an ordinary connection
+ */
+static inline int kdbus_conn_is_ordinary(const struct kdbus_conn *conn)
+{
+       return !(conn->flags & KDBUS_HELLO_SPECIAL_CONN);
+}
+
+/**
+ * kdbus_conn_is_activator() - Check if connection is an activator
+ * @conn:              The connection to check
+ *
+ * Return: Non-zero if the connection is an activator
+ */
+static inline int kdbus_conn_is_activator(const struct kdbus_conn *conn)
+{
+       return conn->flags & KDBUS_HELLO_ACTIVATOR;
+}
+
+/**
+ * kdbus_conn_is_policy_holder() - Check if connection is a policy holder
+ * @conn:              The connection to check
+ *
+ * Return: Non-zero if the connection is a policy holder
+ */
+static inline int kdbus_conn_is_policy_holder(const struct kdbus_conn *conn)
+{
+       return conn->flags & KDBUS_HELLO_POLICY_HOLDER;
+}
+
+/**
+ * kdbus_conn_is_monitor() - Check if connection is a monitor
+ * @conn:              The connection to check
+ *
+ * Return: Non-zero if the connection is a monitor
+ */
+static inline int kdbus_conn_is_monitor(const struct kdbus_conn *conn)
+{
+       return conn->flags & KDBUS_HELLO_MONITOR;
+}
+
+/**
+ * kdbus_conn_lock2() - Lock two connections
+ * @a:         connection A to lock or NULL
+ * @b:         connection B to lock or NULL
+ *
+ * Lock two connections at once. As we need to have a stable locking order, we
+ * always lock the connection with lower memory address first.
+ */
+static inline void kdbus_conn_lock2(struct kdbus_conn *a, struct kdbus_conn *b)
+{
+       if (a < b) {
+               if (a)
+                       mutex_lock(&a->lock);
+               if (b && b != a)
+                       mutex_lock_nested(&b->lock, !!a);
+       } else {
+               if (b)
+                       mutex_lock(&b->lock);
+               if (a && a != b)
+                       mutex_lock_nested(&a->lock, !!b);
+       }
+}
+
+/**
+ * kdbus_conn_unlock2() - Unlock two connections
+ * @a:         connection A to unlock or NULL
+ * @b:         connection B to unlock or NULL
+ *
+ * Unlock two connections at once. See kdbus_conn_lock2().
+ */
+static inline void kdbus_conn_unlock2(struct kdbus_conn *a,
+                                     struct kdbus_conn *b)
+{
+       if (a)
+               mutex_unlock(&a->lock);
+       if (b && b != a)
+               mutex_unlock(&b->lock);
+}
+
+/**
+ * kdbus_conn_assert_active() - lockdep assert on active lock
+ * @conn:      connection that shall be active
+ *
+ * This verifies via lockdep that the caller holds an active reference to the
+ * given connection.
+ */
+static inline void kdbus_conn_assert_active(struct kdbus_conn *conn)
+{
+       lockdep_assert_held(conn);
+}
+
+#endif
diff --git a/ipc/kdbus/item.c b/ipc/kdbus/item.c
new file mode 100644 (file)
index 0000000..745ad54
--- /dev/null
@@ -0,0 +1,339 @@
+/*
+ * Copyright (C) 2013-2015 Kay Sievers
+ * Copyright (C) 2013-2015 Greg Kroah-Hartman <gregkh@linuxfoundation.org>
+ * Copyright (C) 2013-2015 Daniel Mack <daniel@zonque.org>
+ * Copyright (C) 2013-2015 David Herrmann <dh.herrmann@gmail.com>
+ * Copyright (C) 2013-2015 Linux Foundation
+ * Copyright (C) 2014-2015 Djalal Harouni <tixxdz@opendz.org>
+ *
+ * kdbus is free software; you can redistribute it and/or modify it under
+ * the terms of the GNU Lesser General Public License as published by the
+ * Free Software Foundation; either version 2.1 of the License, or (at
+ * your option) any later version.
+ */
+
+#include <linux/ctype.h>
+#include <linux/fs.h>
+#include <linux/string.h>
+
+#include "item.h"
+#include "limits.h"
+#include "util.h"
+
+/*
+ * This verifies the string at position @str with size @size is properly
+ * zero-terminated and does not contain a 0-byte but at the end.
+ */
+static bool kdbus_str_valid(const char *str, size_t size)
+{
+       return size > 0 && memchr(str, '\0', size) == str + size - 1;
+}
+
+/**
+ * kdbus_item_validate_name() - validate an item containing a name
+ * @item:              Item to validate
+ *
+ * Return: zero on success or an negative error code on failure
+ */
+int kdbus_item_validate_name(const struct kdbus_item *item)
+{
+       const char *name = item->str;
+       unsigned int i;
+       size_t len;
+
+       if (item->size < KDBUS_ITEM_HEADER_SIZE + 2)
+               return -EINVAL;
+
+       if (item->size > KDBUS_ITEM_HEADER_SIZE +
+                        KDBUS_SYSNAME_MAX_LEN + 1)
+               return -ENAMETOOLONG;
+
+       if (!kdbus_str_valid(name, KDBUS_ITEM_PAYLOAD_SIZE(item)))
+               return -EINVAL;
+
+       len = strlen(name);
+       if (len == 0)
+               return -EINVAL;
+
+       for (i = 0; i < len; i++) {
+               if (isalpha(name[i]))
+                       continue;
+               if (isdigit(name[i]))
+                       continue;
+               if (name[i] == '_')
+                       continue;
+               if (i > 0 && i + 1 < len && (name[i] == '-' || name[i] == '.'))
+                       continue;
+
+               return -EINVAL;
+       }
+
+       return 0;
+}
+
+/**
+ * kdbus_item_validate() - validate a single item
+ * @item:      item to validate
+ *
+ * Return: 0 if item is valid, negative error code if not.
+ */
+int kdbus_item_validate(const struct kdbus_item *item)
+{
+       size_t payload_size = KDBUS_ITEM_PAYLOAD_SIZE(item);
+       size_t l;
+       int ret;
+
+       BUILD_BUG_ON(KDBUS_ITEM_HEADER_SIZE !=
+                    sizeof(struct kdbus_item_header));
+
+       if (item->size < KDBUS_ITEM_HEADER_SIZE)
+               return -EINVAL;
+
+       switch (item->type) {
+       case KDBUS_ITEM_NEGOTIATE:
+               if (payload_size % sizeof(u64) != 0)
+                       return -EINVAL;
+               break;
+
+       case KDBUS_ITEM_PAYLOAD_VEC:
+               if (payload_size != sizeof(struct kdbus_vec))
+                       return -EINVAL;
+               if (item->vec.size == 0 || item->vec.size > SIZE_MAX)
+                       return -EINVAL;
+               break;
+
+       case KDBUS_ITEM_PAYLOAD_OFF:
+               if (payload_size != sizeof(struct kdbus_vec))
+                       return -EINVAL;
+               if (item->vec.size == 0 || item->vec.size > SIZE_MAX)
+                       return -EINVAL;
+               break;
+
+       case KDBUS_ITEM_PAYLOAD_MEMFD:
+               if (payload_size != sizeof(struct kdbus_memfd))
+                       return -EINVAL;
+               if (item->memfd.size == 0 || item->memfd.size > SIZE_MAX)
+                       return -EINVAL;
+               if (item->memfd.fd < 0)
+                       return -EBADF;
+               break;
+
+       case KDBUS_ITEM_FDS:
+               if (payload_size % sizeof(int) != 0)
+                       return -EINVAL;
+               break;
+
+       case KDBUS_ITEM_CANCEL_FD:
+               if (payload_size != sizeof(int))
+                       return -EINVAL;
+               break;
+
+       case KDBUS_ITEM_BLOOM_PARAMETER:
+               if (payload_size != sizeof(struct kdbus_bloom_parameter))
+                       return -EINVAL;
+               break;
+
+       case KDBUS_ITEM_BLOOM_FILTER:
+               /* followed by the bloom-mask, depends on the bloom-size */
+               if (payload_size < sizeof(struct kdbus_bloom_filter))
+                       return -EINVAL;
+               break;
+
+       case KDBUS_ITEM_BLOOM_MASK:
+               /* size depends on bloom-size of bus */
+               break;
+
+       case KDBUS_ITEM_CONN_DESCRIPTION:
+       case KDBUS_ITEM_MAKE_NAME:
+               ret = kdbus_item_validate_name(item);
+               if (ret < 0)
+                       return ret;
+               break;
+
+       case KDBUS_ITEM_ATTACH_FLAGS_SEND:
+       case KDBUS_ITEM_ATTACH_FLAGS_RECV:
+       case KDBUS_ITEM_ID:
+               if (payload_size != sizeof(u64))
+                       return -EINVAL;
+               break;
+
+       case KDBUS_ITEM_TIMESTAMP:
+               if (payload_size != sizeof(struct kdbus_timestamp))
+                       return -EINVAL;
+               break;
+
+       case KDBUS_ITEM_CREDS:
+               if (payload_size != sizeof(struct kdbus_creds))
+                       return -EINVAL;
+               break;
+
+       case KDBUS_ITEM_AUXGROUPS:
+               if (payload_size % sizeof(u32) != 0)
+                       return -EINVAL;
+               break;
+
+       case KDBUS_ITEM_NAME:
+       case KDBUS_ITEM_DST_NAME:
+       case KDBUS_ITEM_PID_COMM:
+       case KDBUS_ITEM_TID_COMM:
+       case KDBUS_ITEM_EXE:
+       case KDBUS_ITEM_CMDLINE:
+       case KDBUS_ITEM_CGROUP:
+       case KDBUS_ITEM_SECLABEL:
+               if (!kdbus_str_valid(item->str, payload_size))
+                       return -EINVAL;
+               break;
+
+       case KDBUS_ITEM_CAPS:
+               if (payload_size < sizeof(u32))
+                       return -EINVAL;
+               if (payload_size < sizeof(u32) +
+                   4 * CAP_TO_INDEX(item->caps.last_cap) * sizeof(u32))
+                       return -EINVAL;
+               break;
+
+       case KDBUS_ITEM_AUDIT:
+               if (payload_size != sizeof(struct kdbus_audit))
+                       return -EINVAL;
+               break;
+
+       case KDBUS_ITEM_POLICY_ACCESS:
+               if (payload_size != sizeof(struct kdbus_policy_access))
+                       return -EINVAL;
+               break;
+
+       case KDBUS_ITEM_NAME_ADD:
+       case KDBUS_ITEM_NAME_REMOVE:
+       case KDBUS_ITEM_NAME_CHANGE:
+               if (payload_size < sizeof(struct kdbus_notify_name_change))
+                       return -EINVAL;
+               l = payload_size - offsetof(struct kdbus_notify_name_change,
+                                           name);
+               if (l > 0 && !kdbus_str_valid(item->name_change.name, l))
+                       return -EINVAL;
+               break;
+
+       case KDBUS_ITEM_ID_ADD:
+       case KDBUS_ITEM_ID_REMOVE:
+               if (payload_size != sizeof(struct kdbus_notify_id_change))
+                       return -EINVAL;
+               break;
+
+       case KDBUS_ITEM_REPLY_TIMEOUT:
+       case KDBUS_ITEM_REPLY_DEAD:
+               if (payload_size != 0)
+                       return -EINVAL;
+               break;
+
+       default:
+               break;
+       }
+
+       return 0;
+}
+
+/**
+ * kdbus_items_validate() - validate items passed by user-space
+ * @items:             items to validate
+ * @items_size:                number of items
+ *
+ * This verifies that the passed items pointer is consistent and valid.
+ * Furthermore, each item is checked for:
+ *  - valid "size" value
+ *  - payload is of expected type
+ *  - payload is fully included in the item
+ *  - string payloads are zero-terminated
+ *
+ * Return: 0 on success, negative error code on failure.
+ */
+int kdbus_items_validate(const struct kdbus_item *items, size_t items_size)
+{
+       const struct kdbus_item *item;
+       int ret;
+
+       KDBUS_ITEMS_FOREACH(item, items, items_size) {
+               if (!KDBUS_ITEM_VALID(item, items, items_size))
+                       return -EINVAL;
+
+               ret = kdbus_item_validate(item);
+               if (ret < 0)
+                       return ret;
+       }
+
+       if (!KDBUS_ITEMS_END(item, items, items_size))
+               return -EINVAL;
+
+       return 0;
+}
+
+static struct kdbus_item *kdbus_items_get(const struct kdbus_item *items,
+                                         size_t items_size,
+                                         unsigned int item_type)
+{
+       const struct kdbus_item *iter, *found = NULL;
+
+       KDBUS_ITEMS_FOREACH(iter, items, items_size) {
+               if (iter->type == item_type) {
+                       if (found)
+                               return ERR_PTR(-EEXIST);
+                       found = iter;
+               }
+       }
+
+       return (struct kdbus_item *)found ? : ERR_PTR(-EBADMSG);
+}
+
+/**
+ * kdbus_items_get_str() - get string from a list of items
+ * @items:             The items to walk
+ * @items_size:                The size of all items
+ * @item_type:         The item type to look for
+ *
+ * This function walks a list of items and searches for items of type
+ * @item_type. If it finds exactly one such item, @str_ret will be set to
+ * the .str member of the item.
+ *
+ * Return: the string, if the item was found exactly once, ERR_PTR(-EEXIST)
+ * if the item was found more than once, and ERR_PTR(-EBADMSG) if there was
+ * no item of the given type.
+ */
+const char *kdbus_items_get_str(const struct kdbus_item *items,
+                               size_t items_size,
+                               unsigned int item_type)
+{
+       const struct kdbus_item *item;
+
+       item = kdbus_items_get(items, items_size, item_type);
+       return IS_ERR(item) ? ERR_CAST(item) : item->str;
+}
+
+/**
+ * kdbus_item_set() - Set item content
+ * @item:      The item to modify
+ * @type:      The item type to set (KDBUS_ITEM_*)
+ * @data:      Data to copy to item->data, may be %NULL
+ * @len:       Number of bytes in @data
+ *
+ * This sets type, size and data fields of an item. If @data is NULL, the data
+ * memory is cleared.
+ *
+ * Note that you must align your @data memory to 8 bytes. Trailing padding (in
+ * case @len is not 8byte aligned) is cleared by this call.
+ *
+ * Returns: Pointer to the following item.
+ */
+struct kdbus_item *kdbus_item_set(struct kdbus_item *item, u64 type,
+                                 const void *data, size_t len)
+{
+       item->type = type;
+       item->size = KDBUS_ITEM_HEADER_SIZE + len;
+
+       if (data) {
+               memcpy(item->data, data, len);
+               memset(item->data + len, 0, KDBUS_ALIGN8(len) - len);
+       } else {
+               memset(item->data, 0, KDBUS_ALIGN8(len));
+       }
+
+       return KDBUS_ITEM_NEXT(item);
+}
diff --git a/ipc/kdbus/item.h b/ipc/kdbus/item.h
new file mode 100644 (file)
index 0000000..eeefd8b
--- /dev/null
@@ -0,0 +1,64 @@
+/*
+ * Copyright (C) 2013-2015 Kay Sievers
+ * Copyright (C) 2013-2015 Greg Kroah-Hartman <gregkh@linuxfoundation.org>
+ * Copyright (C) 2013-2015 Daniel Mack <daniel@zonque.org>
+ * Copyright (C) 2013-2015 David Herrmann <dh.herrmann@gmail.com>
+ * Copyright (C) 2013-2015 Linux Foundation
+ * Copyright (C) 2014-2015 Djalal Harouni <tixxdz@opendz.org>
+ *
+ * kdbus is free software; you can redistribute it and/or modify it under
+ * the terms of the GNU Lesser General Public License as published by the
+ * Free Software Foundation; either version 2.1 of the License, or (at
+ * your option) any later version.
+ */
+
+#ifndef __KDBUS_ITEM_H
+#define __KDBUS_ITEM_H
+
+#include <linux/kernel.h>
+#include <uapi/linux/kdbus.h>
+
+#include "util.h"
+
+/* generic access and iterators over a stream of items */
+#define KDBUS_ITEM_NEXT(_i) (typeof(_i))(((u8 *)_i) + KDBUS_ALIGN8((_i)->size))
+#define KDBUS_ITEMS_SIZE(_h, _is) ((_h)->size - offsetof(typeof(*_h), _is))
+#define KDBUS_ITEM_HEADER_SIZE offsetof(struct kdbus_item, data)
+#define KDBUS_ITEM_SIZE(_s) KDBUS_ALIGN8(KDBUS_ITEM_HEADER_SIZE + (_s))
+#define KDBUS_ITEM_PAYLOAD_SIZE(_i) ((_i)->size - KDBUS_ITEM_HEADER_SIZE)
+
+#define KDBUS_ITEMS_FOREACH(_i, _is, _s)                               \
+       for (_i = _is;                                                  \
+            ((u8 *)(_i) < (u8 *)(_is) + (_s)) &&                       \
+              ((u8 *)(_i) >= (u8 *)(_is));                             \
+            _i = KDBUS_ITEM_NEXT(_i))
+
+#define KDBUS_ITEM_VALID(_i, _is, _s)                                  \
+       ((_i)->size >= KDBUS_ITEM_HEADER_SIZE &&                        \
+        (u8 *)(_i) + (_i)->size > (u8 *)(_i) &&                        \
+        (u8 *)(_i) + (_i)->size <= (u8 *)(_is) + (_s) &&               \
+        (u8 *)(_i) >= (u8 *)(_is))
+
+#define KDBUS_ITEMS_END(_i, _is, _s)                                   \
+       ((u8 *)_i == ((u8 *)(_is) + KDBUS_ALIGN8(_s)))
+
+/**
+ * struct kdbus_item_header - Describes the fix part of an item
+ * @size:      The total size of the item
+ * @type:      The item type, one of KDBUS_ITEM_*
+ */
+struct kdbus_item_header {
+       u64 size;
+       u64 type;
+};
+
+int kdbus_item_validate_name(const struct kdbus_item *item);
+int kdbus_item_validate(const struct kdbus_item *item);
+int kdbus_items_validate(const struct kdbus_item *items, size_t items_size);
+const char *kdbus_items_get_str(const struct kdbus_item *items,
+                               size_t items_size,
+                               unsigned int item_type);
+struct kdbus_item *kdbus_item_set(struct kdbus_item *item, u64 type,
+                                 const void *data, size_t len);
+
+#endif
diff --git a/ipc/kdbus/message.c b/ipc/kdbus/message.c
new file mode 100644 (file)
index 0000000..8096075
--- /dev/null
@@ -0,0 +1,616 @@
+/*
+ * Copyright (C) 2013-2015 Kay Sievers
+ * Copyright (C) 2013-2015 Greg Kroah-Hartman <gregkh@linuxfoundation.org>
+ * Copyright (C) 2013-2015 Daniel Mack <daniel@zonque.org>
+ * Copyright (C) 2013-2015 David Herrmann <dh.herrmann@gmail.com>
+ * Copyright (C) 2013-2015 Linux Foundation
+ * Copyright (C) 2014-2015 Djalal Harouni <tixxdz@opendz.org>
+ *
+ * kdbus is free software; you can redistribute it and/or modify it under
+ * the terms of the GNU Lesser General Public License as published by the
+ * Free Software Foundation; either version 2.1 of the License, or (at
+ * your option) any later version.
+ */
+
+#include <linux/capability.h>
+#include <linux/cgroup.h>
+#include <linux/cred.h>
+#include <linux/file.h>
+#include <linux/init.h>
+#include <linux/module.h>
+#include <linux/mutex.h>
+#include <linux/sched.h>
+#include <linux/shmem_fs.h>
+#include <linux/sizes.h>
+#include <linux/slab.h>
+#include <linux/uaccess.h>
+#include <net/sock.h>
+
+#include "bus.h"
+#include "connection.h"
+#include "domain.h"
+#include "endpoint.h"
+#include "handle.h"
+#include "item.h"
+#include "match.h"
+#include "message.h"
+#include "names.h"
+#include "policy.h"
+
+#define KDBUS_KMSG_HEADER_SIZE offsetof(struct kdbus_kmsg, msg)
+
+static struct kdbus_msg_resources *kdbus_msg_resources_new(void)
+{
+       struct kdbus_msg_resources *r;
+
+       r = kzalloc(sizeof(*r), GFP_KERNEL);
+       if (!r)
+               return ERR_PTR(-ENOMEM);
+
+       kref_init(&r->kref);
+
+       return r;
+}
+
+static void __kdbus_msg_resources_free(struct kref *kref)
+{
+       struct kdbus_msg_resources *r =
+               container_of(kref, struct kdbus_msg_resources, kref);
+       size_t i;
+
+       for (i = 0; i < r->data_count; ++i) {
+               switch (r->data[i].type) {
+               case KDBUS_MSG_DATA_VEC:
+                       /* nothing to do */
+                       break;
+               case KDBUS_MSG_DATA_MEMFD:
+                       if (r->data[i].memfd.file)
+                               fput(r->data[i].memfd.file);
+                       break;
+               }
+       }
+
+       for (i = 0; i < r->fds_count; i++)
+               if (r->fds[i])
+                       fput(r->fds[i]);
+
+       kfree(r->dst_name);
+       kfree(r->data);
+       kfree(r->fds);
+       kfree(r);
+}
+
+/**
+ * kdbus_msg_resources_ref() - Acquire reference to msg resources
+ * @r:         resources to acquire ref to
+ *
+ * Return: The acquired resource
+ */
+struct kdbus_msg_resources *
+kdbus_msg_resources_ref(struct kdbus_msg_resources *r)
+{
+       if (r)
+               kref_get(&r->kref);
+       return r;
+}
+
+/**
+ * kdbus_msg_resources_unref() - Drop reference to msg resources
+ * @r:         resources to drop reference of
+ *
+ * Return: NULL
+ */
+struct kdbus_msg_resources *
+kdbus_msg_resources_unref(struct kdbus_msg_resources *r)
+{
+       if (r)
+               kref_put(&r->kref, __kdbus_msg_resources_free);
+       return NULL;
+}
+
+/**
+ * kdbus_kmsg_free() - free allocated message
+ * @kmsg:              Message
+ */
+void kdbus_kmsg_free(struct kdbus_kmsg *kmsg)
+{
+       if (!kmsg)
+               return;
+
+       kdbus_msg_resources_unref(kmsg->res);
+       kdbus_meta_conn_unref(kmsg->conn_meta);
+       kdbus_meta_proc_unref(kmsg->proc_meta);
+       kfree(kmsg->iov);
+       kfree(kmsg);
+}
+
+/**
+ * kdbus_kmsg_new() - allocate message
+ * @bus:               Bus this message is allocated on
+ * @extra_size:                Additional size to reserve for data
+ *
+ * Return: new kdbus_kmsg on success, ERR_PTR on failure.
+ */
+struct kdbus_kmsg *kdbus_kmsg_new(struct kdbus_bus *bus, size_t extra_size)
+{
+       struct kdbus_kmsg *m;
+       size_t size;
+       int ret;
+
+       size = sizeof(struct kdbus_kmsg) + KDBUS_ITEM_SIZE(extra_size);
+       m = kzalloc(size, GFP_KERNEL);
+       if (!m)
+               return ERR_PTR(-ENOMEM);
+
+       m->seq = atomic64_inc_return(&bus->domain->last_id);
+       m->msg.size = size - KDBUS_KMSG_HEADER_SIZE;
+       m->msg.items[0].size = KDBUS_ITEM_SIZE(extra_size);
+
+       m->proc_meta = kdbus_meta_proc_new();
+       if (IS_ERR(m->proc_meta)) {
+               ret = PTR_ERR(m->proc_meta);
+               m->proc_meta = NULL;
+               goto exit;
+       }
+
+       m->conn_meta = kdbus_meta_conn_new();
+       if (IS_ERR(m->conn_meta)) {
+               ret = PTR_ERR(m->conn_meta);
+               m->conn_meta = NULL;
+               goto exit;
+       }
+
+       return m;
+
+exit:
+       kdbus_kmsg_free(m);
+       return ERR_PTR(ret);
+}
+
+static int kdbus_handle_check_file(struct file *file)
+{
+       struct inode *inode = file_inode(file);
+       struct socket *sock;
+
+       /*
+        * Don't allow file descriptors in the transport that themselves allow
+        * file descriptor queueing. This will eventually be allowed once both
+        * unix domain sockets and kdbus share a generic garbage collector.
+        */
+
+       if (file->f_op == &kdbus_handle_ops)
+               return -EOPNOTSUPP;
+
+       if (!S_ISSOCK(inode->i_mode))
+               return 0;
+
+       if (file->f_mode & FMODE_PATH)
+               return 0;
+
+       sock = SOCKET_I(inode);
+       if (sock->sk && sock->ops && sock->ops->family == PF_UNIX)
+               return -EOPNOTSUPP;
+
+       return 0;
+}
+
+static const char * const zeros = "\0\0\0\0\0\0\0";
+
+/*
+ * kdbus_msg_scan_items() - validate incoming data and prepare parsing
+ * @kmsg:              Message
+ * @bus:               Bus the message is sent over
+ *
+ * Return: 0 on success, negative errno on failure.
+ *
+ * Files references in MEMFD or FDS items are pinned.
+ *
+ * On errors, the caller should drop any taken reference with
+ * kdbus_kmsg_free()
+ */
+static int kdbus_msg_scan_items(struct kdbus_kmsg *kmsg,
+                               struct kdbus_bus *bus)
+{
+       struct kdbus_msg_resources *res = kmsg->res;
+       const struct kdbus_msg *msg = &kmsg->msg;
+       const struct kdbus_item *item;
+       size_t n, n_vecs, n_memfds;
+       bool has_bloom = false;
+       bool has_name = false;
+       bool has_fds = false;
+       bool is_broadcast;
+       bool is_signal;
+       u64 vec_size;
+
+       is_broadcast = (msg->dst_id == KDBUS_DST_ID_BROADCAST);
+       is_signal = !!(msg->flags & KDBUS_MSG_SIGNAL);
+
+       /* count data payloads */
+       n_vecs = 0;
+       n_memfds = 0;
+       KDBUS_ITEMS_FOREACH(item, msg->items, KDBUS_ITEMS_SIZE(msg, items)) {
+               switch (item->type) {
+               case KDBUS_ITEM_PAYLOAD_VEC:
+                       ++n_vecs;
+                       break;
+               case KDBUS_ITEM_PAYLOAD_MEMFD:
+                       ++n_memfds;
+                       if (item->memfd.size % 8)
+                               ++n_vecs;
+                       break;
+               default:
+                       break;
+               }
+       }
+
+       n = n_vecs + n_memfds;
+       if (n > 0) {
+               res->data = kcalloc(n, sizeof(*res->data), GFP_KERNEL);
+               if (!res->data)
+                       return -ENOMEM;
+       }
+
+       if (n_vecs > 0) {
+               kmsg->iov = kcalloc(n_vecs, sizeof(*kmsg->iov), GFP_KERNEL);
+               if (!kmsg->iov)
+                       return -ENOMEM;
+       }
+
+       /* import data payloads */
+       n = 0;
+       vec_size = 0;
+       KDBUS_ITEMS_FOREACH(item, msg->items, KDBUS_ITEMS_SIZE(msg, items)) {
+               size_t payload_size = KDBUS_ITEM_PAYLOAD_SIZE(item);
+               struct iovec *iov = kmsg->iov + kmsg->iov_count;
+
+               if (++n > KDBUS_MSG_MAX_ITEMS)
+                       return -E2BIG;
+
+               switch (item->type) {
+               case KDBUS_ITEM_PAYLOAD_VEC: {
+                       struct kdbus_msg_data *d = res->data + res->data_count;
+                       void __force __user *ptr = KDBUS_PTR(item->vec.address);
+                       size_t size = item->vec.size;
+
+                       if (vec_size + size < vec_size)
+                               return -EMSGSIZE;
+                       if (vec_size + size > KDBUS_MSG_MAX_PAYLOAD_VEC_SIZE)
+                               return -EMSGSIZE;
+
+                       d->type = KDBUS_MSG_DATA_VEC;
+                       d->size = size;
+
+                       if (ptr) {
+                               if (unlikely(!access_ok(VERIFY_READ, ptr,
+                                                       size)))
+                                       return -EFAULT;
+
+                               d->vec.off = kmsg->pool_size;
+                               iov->iov_base = ptr;
+                               iov->iov_len = size;
+                       } else {
+                               d->vec.off = ~0ULL;
+                               iov->iov_base = (char __user *)zeros;
+                               iov->iov_len = size % 8;
+                       }
+
+                       if (kmsg->pool_size + iov->iov_len < kmsg->pool_size)
+                               return -EMSGSIZE;
+
+                       kmsg->pool_size += iov->iov_len;
+                       ++kmsg->iov_count;
+                       ++res->vec_count;
+                       ++res->data_count;
+                       vec_size += size;
+
+                       break;
+               }
+
+               case KDBUS_ITEM_PAYLOAD_MEMFD: {
+                       struct kdbus_msg_data *d = res->data + res->data_count;
+                       u64 start = item->memfd.start;
+                       u64 size = item->memfd.size;
+                       size_t pad = size % 8;
+                       int seals, mask;
+                       struct file *f;
+
+                       if (kmsg->pool_size + size % 8 < kmsg->pool_size)
+                               return -EMSGSIZE;
+                       if (start + size < start)
+                               return -EMSGSIZE;
+
+                       if (item->memfd.fd < 0)
+                               return -EBADF;
+
+                       if (res->memfd_count >= KDBUS_MSG_MAX_MEMFD_ITEMS)
+                               return -E2BIG;
+
+                       f = fget(item->memfd.fd);
+                       if (!f)
+                               return -EBADF;
+
+                       if (pad) {
+                               iov->iov_base = (char __user *)zeros;
+                               iov->iov_len = pad;
+
+                               kmsg->pool_size += pad;
+                               ++kmsg->iov_count;
+                       }
+
+                       ++res->data_count;
+                       ++res->memfd_count;
+
+                       d->type = KDBUS_MSG_DATA_MEMFD;
+                       d->size = size;
+                       d->memfd.start = start;
+                       d->memfd.file = f;
+
+                       /*
+                        * We only accept a sealed memfd file whose content
+                        * cannot be altered by the sender or anybody else
+                        * while it is shared or in-flight. Other files need
+                        * to be passed with KDBUS_MSG_FDS.
+                        */
+                       seals = shmem_get_seals(f);
+                       if (seals < 0)
+                               return -EMEDIUMTYPE;
+
+                       mask = F_SEAL_SHRINK | F_SEAL_GROW |
+                               F_SEAL_WRITE | F_SEAL_SEAL;
+                       if ((seals & mask) != mask)
+                               return -ETXTBSY;
+
+                       if (start + size > (u64)i_size_read(file_inode(f)))
+                               return -EBADF;
+
+                       break;
+               }
+
+               case KDBUS_ITEM_FDS: {
+                       unsigned int i;
+                       unsigned int fds_count = payload_size / sizeof(int);
+
+                       /* do not allow multiple fd arrays */
+                       if (has_fds)
+                               return -EEXIST;
+                       has_fds = true;
+
+                       /* Do not allow to broadcast file descriptors */
+                       if (is_broadcast)
+                               return -ENOTUNIQ;
+
+                       if (fds_count > KDBUS_CONN_MAX_FDS_PER_USER)
+                               return -EMFILE;
+
+                       res->fds = kcalloc(fds_count, sizeof(struct file *),
+                                          GFP_KERNEL);
+                       if (!res->fds)
+                               return -ENOMEM;
+
+                       for (i = 0; i < fds_count; i++) {
+                               int fd = item->fds[i];
+                               int ret;
+
+                               /*
+                                * Verify the fd and increment the usage count.
+                                * Use fget_raw() to allow passing O_PATH fds.
+                                */
+                               if (fd < 0)
+                                       return -EBADF;
+
+                               res->fds[i] = fget_raw(fd);
+                               if (!res->fds[i])
+                                       return -EBADF;
+
+                               res->fds_count++;
+
+                               ret = kdbus_handle_check_file(res->fds[i]);
+                               if (ret < 0)
+                                       return ret;
+                       }
+
+                       break;
+               }
+
+               case KDBUS_ITEM_BLOOM_FILTER: {
+                       u64 bloom_size;
+
+                       /* do not allow multiple bloom filters */
+                       if (has_bloom)
+                               return -EEXIST;
+                       has_bloom = true;
+
+                       bloom_size = payload_size -
+                                    offsetof(struct kdbus_bloom_filter, data);
+
+                       /*
+                       * Allow only bloom filter sizes of a multiple of 64bit.
+                       */
+                       if (!KDBUS_IS_ALIGNED8(bloom_size))
+                               return -EFAULT;
+
+                       /* do not allow mismatching bloom filter sizes */
+                       if (bloom_size != bus->bloom.size)
+                               return -EDOM;
+
+                       kmsg->bloom_filter = &item->bloom_filter;
+                       break;
+               }
+
+               case KDBUS_ITEM_DST_NAME:
+                       /* do not allow multiple names */
+                       if (has_name)
+                               return -EEXIST;
+                       has_name = true;
+
+                       if (!kdbus_name_is_valid(item->str, false))
+                               return -EINVAL;
+
+                       res->dst_name = kstrdup(item->str, GFP_KERNEL);
+                       if (!res->dst_name)
+                               return -ENOMEM;
+                       break;
+
+               default:
+                       return -EINVAL;
+               }
+       }
+
+       /* name is needed if no ID is given */
+       if (msg->dst_id == KDBUS_DST_ID_NAME && !has_name)
+               return -EDESTADDRREQ;
+
+       if (is_broadcast) {
+               /* Broadcasts can't take names */
+               if (has_name)
+                       return -EBADMSG;
+
+               /* All broadcasts have to be signals */
+               if (!is_signal)
+                       return -EBADMSG;
+
+               /* Timeouts are not allowed for broadcasts */
+               if (msg->timeout_ns > 0)
+                       return -ENOTUNIQ;
+       }
+
+       /*
+        * Signal messages require a bloom filter, and bloom filters are
+        * only valid with signals.
+        */
+       if (is_signal ^ has_bloom)
+               return -EBADMSG;
+
+       return 0;
+}
+
+/**
+ * kdbus_kmsg_new_from_cmd() - create kernel message from send payload
+ * @conn:              Connection
+ * @cmd_send:          Payload of KDBUS_CMD_SEND
+ *
+ * Return: a new kdbus_kmsg on success, ERR_PTR on failure.
+ */
+struct kdbus_kmsg *kdbus_kmsg_new_from_cmd(struct kdbus_conn *conn,
+                                          struct kdbus_cmd_send *cmd_send)
+{
+       struct kdbus_kmsg *m;
+       u64 size;
+       int ret;
+
+       ret = kdbus_copy_from_user(&size, KDBUS_PTR(cmd_send->msg_address),
+                                  sizeof(size));
+       if (ret < 0)
+               return ERR_PTR(ret);
+
+       if (size < sizeof(struct kdbus_msg) || size > KDBUS_MSG_MAX_SIZE)
+               return ERR_PTR(-EINVAL);
+
+       m = kmalloc(size + KDBUS_KMSG_HEADER_SIZE, GFP_KERNEL);
+       if (!m)
+               return ERR_PTR(-ENOMEM);
+
+       memset(m, 0, KDBUS_KMSG_HEADER_SIZE);
+       m->seq = atomic64_inc_return(&conn->ep->bus->domain->last_id);
+
+       m->proc_meta = kdbus_meta_proc_new();
+       if (IS_ERR(m->proc_meta)) {
+               ret = PTR_ERR(m->proc_meta);
+               m->proc_meta = NULL;
+               goto exit_free;
+       }
+
+       m->conn_meta = kdbus_meta_conn_new();
+       if (IS_ERR(m->conn_meta)) {
+               ret = PTR_ERR(m->conn_meta);
+               m->conn_meta = NULL;
+               goto exit_free;
+       }
+
+       if (copy_from_user(&m->msg, KDBUS_PTR(cmd_send->msg_address), size)) {
+               ret = -EFAULT;
+               goto exit_free;
+       }
+
+       if (m->msg.size != size) {
+               ret = -EINVAL;
+               goto exit_free;
+       }
+
+       if (m->msg.flags & ~(KDBUS_MSG_EXPECT_REPLY |
+                            KDBUS_MSG_NO_AUTO_START |
+                            KDBUS_MSG_SIGNAL)) {
+               ret = -EINVAL;
+               goto exit_free;
+       }
+
+       ret = kdbus_items_validate(m->msg.items,
+                                  KDBUS_ITEMS_SIZE(&m->msg, items));
+       if (ret < 0)
+               goto exit_free;
+
+       m->res = kdbus_msg_resources_new();
+       if (IS_ERR(m->res)) {
+               ret = PTR_ERR(m->res);
+               m->res = NULL;
+               goto exit_free;
+       }
+
+       /* do not accept kernel-generated messages */
+       if (m->msg.payload_type == KDBUS_PAYLOAD_KERNEL) {
+               ret = -EINVAL;
+               goto exit_free;
+       }
+
+       if (m->msg.flags & KDBUS_MSG_EXPECT_REPLY) {
+               /* requests for replies need timeout and cookie */
+               if (m->msg.timeout_ns == 0 || m->msg.cookie == 0) {
+                       ret = -EINVAL;
+                       goto exit_free;
+               }
+
+               /* replies may not be expected for broadcasts */
+               if (m->msg.dst_id == KDBUS_DST_ID_BROADCAST) {
+                       ret = -ENOTUNIQ;
+                       goto exit_free;
+               }
+
+               /* replies may not be expected for signals */
+               if (m->msg.flags & KDBUS_MSG_SIGNAL) {
+                       ret = -EINVAL;
+                       goto exit_free;
+               }
+       } else {
+               /*
+                * KDBUS_SEND_SYNC_REPLY is only valid together with
+                * KDBUS_MSG_EXPECT_REPLY
+                */
+               if (cmd_send->flags & KDBUS_SEND_SYNC_REPLY) {
+                       ret = -EINVAL;
+                       goto exit_free;
+               }
+
+               /* replies cannot be signals */
+               if (m->msg.cookie_reply && (m->msg.flags & KDBUS_MSG_SIGNAL)) {
+                       ret = -EINVAL;
+                       goto exit_free;
+               }
+       }
+
+       ret = kdbus_msg_scan_items(m, conn->ep->bus);
+       if (ret < 0)
+               goto exit_free;
+
+       /* patch-in the source of this message */
+       if (m->msg.src_id > 0 && m->msg.src_id != conn->id) {
+               ret = -EINVAL;
+               goto exit_free;
+       }
+       m->msg.src_id = conn->id;
+
+       return m;
+
+exit_free:
+       kdbus_kmsg_free(m);
+       return ERR_PTR(ret);
+}
diff --git a/ipc/kdbus/message.h b/ipc/kdbus/message.h
new file mode 100644 (file)
index 0000000..af47758
--- /dev/null
@@ -0,0 +1,133 @@
+/*
+ * Copyright (C) 2013-2015 Kay Sievers
+ * Copyright (C) 2013-2015 Greg Kroah-Hartman <gregkh@linuxfoundation.org>
+ * Copyright (C) 2013-2015 Daniel Mack <daniel@zonque.org>
+ * Copyright (C) 2013-2015 David Herrmann <dh.herrmann@gmail.com>
+ * Copyright (C) 2013-2015 Linux Foundation
+ *
+ * kdbus is free software; you can redistribute it and/or modify it under
+ * the terms of the GNU Lesser General Public License as published by the
+ * Free Software Foundation; either version 2.1 of the License, or (at
+ * your option) any later version.
+ */
+
+#ifndef __KDBUS_MESSAGE_H
+#define __KDBUS_MESSAGE_H
+
+#include "util.h"
+#include "metadata.h"
+
+/**
+ * enum kdbus_msg_data_type - Type of kdbus_msg_data payloads
+ * @KDBUS_MSG_DATA_VEC:                Data vector provided by user-space
+ * @KDBUS_MSG_DATA_MEMFD:      Memfd payload
+ */
+enum kdbus_msg_data_type {
+       KDBUS_MSG_DATA_VEC,
+       KDBUS_MSG_DATA_MEMFD,
+};
+
+/**
+ * struct kdbus_msg_data - Data payload as stored by messages
+ * @type:      Type of payload (KDBUS_MSG_DATA_*)
+ * @size:      Size of the described payload
+ * @off:       The offset, relative to the vec slice
+ * @start:     Offset inside the memfd
+ * @file:      Backing file referenced by the memfd
+ */
+struct kdbus_msg_data {
+       unsigned int type;
+       u64 size;
+
+       union {
+               struct {
+                       u64 off;
+               } vec;
+               struct {
+                       u64 start;
+                       struct file *file;
+               } memfd;
+       };
+};
+
+/**
+ * struct kdbus_kmsg_resources - resources of a message
+ * @kref:              Reference counter
+ * @dst_name:          Short-cut to msg for faster lookup
+ * @fds:               Array of file descriptors to pass
+ * @fds_count:         Number of file descriptors to pass
+ * @data:              Array of data payloads
+ * @vec_count:         Number of VEC entries
+ * @memfd_count:       Number of MEMFD entries in @data
+ * @data_count:                Sum of @vec_count + @memfd_count
+ */
+struct kdbus_msg_resources {
+       struct kref kref;
+       const char *dst_name;
+
+       struct file **fds;
+       unsigned int fds_count;
+
+       struct kdbus_msg_data *data;
+       size_t vec_count;
+       size_t memfd_count;
+       size_t data_count;
+};
+
+struct kdbus_msg_resources *
+kdbus_msg_resources_ref(struct kdbus_msg_resources *r);
+struct kdbus_msg_resources *
+kdbus_msg_resources_unref(struct kdbus_msg_resources *r);
+
+/**
+ * struct kdbus_kmsg - internal message handling data
+ * @seq:               Domain-global message sequence number
+ * @notify_type:       Short-cut for faster lookup
+ * @notify_old_id:     Short-cut for faster lookup
+ * @notify_new_id:     Short-cut for faster lookup
+ * @notify_name:       Short-cut for faster lookup
+ * @dst_name_id:       Short-cut to msg for faster lookup
+ * @bloom_filter:      Bloom filter to match message properties
+ * @bloom_generation:  Generation of bloom element set
+ * @notify_entry:      List of kernel-generated notifications
+ * @iov:               Array of iovec, describing the payload to copy
+ * @iov_count:         Number of array members in @iov
+ * @pool_size:         Overall size of inlined data referenced by @iov
+ * @proc_meta:         Appended SCM-like metadata of the sending process
+ * @conn_meta:         Appended SCM-like metadata of the sending connection
+ * @res:               Message resources
+ * @msg:               Message from or to userspace
+ */
+struct kdbus_kmsg {
+       u64 seq;
+       u64 notify_type;
+       u64 notify_old_id;
+       u64 notify_new_id;
+       const char *notify_name;
+
+       u64 dst_name_id;
+       const struct kdbus_bloom_filter *bloom_filter;
+       u64 bloom_generation;
+       struct list_head notify_entry;
+
+       struct iovec *iov;
+       size_t iov_count;
+       u64 pool_size;
+
+       struct kdbus_meta_proc *proc_meta;
+       struct kdbus_meta_conn *conn_meta;
+       struct kdbus_msg_resources *res;
+
+       /* variable size, must be the last member */
+       struct kdbus_msg msg;
+};
+
+struct kdbus_bus;
+struct kdbus_conn;
+
+struct kdbus_kmsg *kdbus_kmsg_new(struct kdbus_bus *bus, size_t extra_size);
+struct kdbus_kmsg *kdbus_kmsg_new_from_cmd(struct kdbus_conn *conn,
+                                          struct kdbus_cmd_send *cmd_send);
+void kdbus_kmsg_free(struct kdbus_kmsg *kmsg);
+
+#endif
diff --git a/ipc/kdbus/queue.c b/ipc/kdbus/queue.c
new file mode 100644 (file)
index 0000000..a449464
--- /dev/null
@@ -0,0 +1,678 @@
+/*
+ * Copyright (C) 2013-2015 Kay Sievers
+ * Copyright (C) 2013-2015 Greg Kroah-Hartman <gregkh@linuxfoundation.org>
+ * Copyright (C) 2013-2015 Daniel Mack <daniel@zonque.org>
+ * Copyright (C) 2013-2015 David Herrmann <dh.herrmann@gmail.com>
+ * Copyright (C) 2013-2015 Linux Foundation
+ * Copyright (C) 2014-2015 Djalal Harouni <tixxdz@opendz.org>
+ *
+ * kdbus is free software; you can redistribute it and/or modify it under
+ * the terms of the GNU Lesser General Public License as published by the
+ * Free Software Foundation; either version 2.1 of the License, or (at
+ * your option) any later version.
+ */
+
+#include <linux/audit.h>
+#include <linux/file.h>
+#include <linux/fs.h>
+#include <linux/hashtable.h>
+#include <linux/idr.h>
+#include <linux/init.h>
+#include <linux/math64.h>
+#include <linux/mm.h>
+#include <linux/module.h>
+#include <linux/mutex.h>
+#include <linux/poll.h>
+#include <linux/sched.h>
+#include <linux/sizes.h>
+#include <linux/slab.h>
+#include <linux/syscalls.h>
+#include <linux/uio.h>
+
+#include "util.h"
+#include "domain.h"
+#include "connection.h"
+#include "item.h"
+#include "message.h"
+#include "metadata.h"
+#include "queue.h"
+#include "reply.h"
+
+/**
+ * kdbus_queue_init() - initialize data structure related to a queue
+ * @queue:     The queue to initialize
+ */
+void kdbus_queue_init(struct kdbus_queue *queue)
+{
+       INIT_LIST_HEAD(&queue->msg_list);
+       queue->msg_prio_queue = RB_ROOT;
+}
+
+/**
+ * kdbus_queue_peek() - Retrieves an entry from a queue
+ * @queue:             The queue
+ * @priority:          The minimum priority of the entry to peek
+ * @use_priority:      Boolean flag whether or not to peek by priority
+ *
+ * Look for a entry in a queue, either by priority, or the oldest one (FIFO).
+ * The entry is not freed, put off the queue's lists or anything else.
+ *
+ * Return: the peeked queue entry on success, NULL if no suitable msg is found
+ */
+struct kdbus_queue_entry *kdbus_queue_peek(struct kdbus_queue *queue,
+                                          s64 priority, bool use_priority)
+{
+       struct kdbus_queue_entry *e;
+
+       if (list_empty(&queue->msg_list))
+               return NULL;
+
+       if (use_priority) {
+               /* get next entry with highest priority */
+               e = rb_entry(queue->msg_prio_highest,
+                            struct kdbus_queue_entry, prio_node);
+
+               /* no entry with the requested priority */
+               if (e->priority > priority)
+                       return NULL;
+       } else {
+               /* ignore the priority, return the next entry in the entry */
+               e = list_first_entry(&queue->msg_list,
+                                    struct kdbus_queue_entry, entry);
+       }
+
+       return e;
+}
+
+static void kdbus_queue_entry_link(struct kdbus_queue_entry *entry)
+{
+       struct kdbus_queue *queue = &entry->conn->queue;
+       struct rb_node **n, *pn = NULL;
+       bool highest = true;
+
+       lockdep_assert_held(&entry->conn->lock);
+       if (WARN_ON(!list_empty(&entry->entry)))
+               return;
+
+       /* sort into priority entry tree */
+       n = &queue->msg_prio_queue.rb_node;
+       while (*n) {
+               struct kdbus_queue_entry *e;
+
+               pn = *n;
+               e = rb_entry(pn, struct kdbus_queue_entry, prio_node);
+
+               /* existing node for this priority, add to its list */
+               if (likely(entry->priority == e->priority)) {
+                       list_add_tail(&entry->prio_entry, &e->prio_entry);
+                       goto prio_done;
+               }
+
+               if (entry->priority < e->priority) {
+                       n = &pn->rb_left;
+               } else {
+                       n = &pn->rb_right;
+                       highest = false;
+               }
+       }
+
+       /* cache highest-priority entry */
+       if (highest)
+               queue->msg_prio_highest = &entry->prio_node;
+
+       /* new node for this priority */
+       rb_link_node(&entry->prio_node, pn, n);
+       rb_insert_color(&entry->prio_node, &queue->msg_prio_queue);
+       INIT_LIST_HEAD(&entry->prio_entry);
+
+prio_done:
+       /* add to unsorted fifo list */
+       list_add_tail(&entry->entry, &queue->msg_list);
+}
+
+static void kdbus_queue_entry_unlink(struct kdbus_queue_entry *entry)
+{
+       struct kdbus_queue *queue = &entry->conn->queue;
+
+       lockdep_assert_held(&entry->conn->lock);
+       if (list_empty(&entry->entry))
+               return;
+
+       list_del_init(&entry->entry);
+
+       if (list_empty(&entry->prio_entry)) {
+               /*
+                * Single entry for this priority, update cached
+                * highest-priority entry, remove the tree node.
+                */
+               if (queue->msg_prio_highest == &entry->prio_node)
+                       queue->msg_prio_highest = rb_next(&entry->prio_node);
+
+               rb_erase(&entry->prio_node, &queue->msg_prio_queue);
+       } else {
+               struct kdbus_queue_entry *q;
+
+               /*
+                * Multiple entries for this priority entry, get next one in
+                * the list. Update cached highest-priority entry, store the
+                * new one as the tree node.
+                */
+               q = list_first_entry(&entry->prio_entry,
+                                    struct kdbus_queue_entry, prio_entry);
+               list_del(&entry->prio_entry);
+
+               if (queue->msg_prio_highest == &entry->prio_node)
+                       queue->msg_prio_highest = &q->prio_node;
+
+               rb_replace_node(&entry->prio_node, &q->prio_node,
+                               &queue->msg_prio_queue);
+       }
+}
+
+/**
+ * kdbus_queue_entry_new() - allocate a queue entry
+ * @conn_dst:  destination connection
+ * @kmsg:      kmsg object the queue entry should track
+ * @user:      user to account message on (or NULL for kernel messages)
+ *
+ * Allocates a queue entry based on a given kmsg and allocate space for
+ * the message payload and the requested metadata in the connection's pool.
+ * The entry is not actually added to the queue's lists at this point.
+ *
+ * Return: the allocated entry on success, or an ERR_PTR on failures.
+ */
+struct kdbus_queue_entry *kdbus_queue_entry_new(struct kdbus_conn *conn_dst,
+                                               const struct kdbus_kmsg *kmsg,
+                                               struct kdbus_user *user)
+{
+       struct kdbus_msg_resources *res = kmsg->res;
+       const struct kdbus_msg *msg = &kmsg->msg;
+       struct kdbus_queue_entry *entry;
+       size_t memfd_cnt = 0;
+       struct kvec kvec[2];
+       size_t meta_size;
+       size_t msg_size;
+       u64 payload_off;
+       u64 size = 0;
+       int ret = 0;
+
+       entry = kzalloc(sizeof(*entry), GFP_KERNEL);
+       if (!entry)
+               return ERR_PTR(-ENOMEM);
+
+       INIT_LIST_HEAD(&entry->entry);
+       entry->priority = msg->priority;
+       entry->dst_name_id = kmsg->dst_name_id;
+       entry->msg_res = kdbus_msg_resources_ref(res);
+       entry->proc_meta = kdbus_meta_proc_ref(kmsg->proc_meta);
+       entry->conn_meta = kdbus_meta_conn_ref(kmsg->conn_meta);
+       entry->conn = kdbus_conn_ref(conn_dst);
+
+       if (kmsg->msg.src_id == KDBUS_SRC_ID_KERNEL)
+               msg_size = msg->size;
+       else
+               msg_size = offsetof(struct kdbus_msg, items);
+
+       /* sum up the size of the needed slice */
+       size = msg_size;
+
+       if (res) {
+               size += res->vec_count *
+                       KDBUS_ITEM_SIZE(sizeof(struct kdbus_vec));
+
+               if (res->memfd_count) {
+                       entry->memfd_offset =
+                               kcalloc(res->memfd_count, sizeof(size_t),
+                                       GFP_KERNEL);
+                       if (!entry->memfd_offset) {
+                               ret = -ENOMEM;
+                               goto exit_free_entry;
+                       }
+
+                       size += res->memfd_count *
+                               KDBUS_ITEM_SIZE(sizeof(struct kdbus_memfd));
+               }
+
+               if (res->fds_count)
+                       size += KDBUS_ITEM_SIZE(sizeof(int) * res->fds_count);
+
+               if (res->dst_name)
+                       size += KDBUS_ITEM_SIZE(strlen(res->dst_name) + 1);
+       }
+
+       /*
+        * Remember the offset of the metadata part, so we can override
+        * this part later during kdbus_queue_entry_install().
+        */
+       entry->meta_offset = size;
+
+       if (entry->proc_meta || entry->conn_meta) {
+               entry->attach_flags =
+                       atomic64_read(&conn_dst->attach_flags_recv);
+
+               ret = kdbus_meta_export_prepare(entry->proc_meta,
+                                               entry->conn_meta,
+                                               &entry->attach_flags,
+                                               &meta_size);
+               if (ret < 0)
+                       goto exit_free_entry;
+
+               size += meta_size;
+       }
+
+       payload_off = size;
+       size += kmsg->pool_size;
+       size = KDBUS_ALIGN8(size);
+
+       ret = kdbus_conn_quota_inc(conn_dst, user, size,
+                                  res ? res->fds_count : 0);
+       if (ret < 0)
+               goto exit_free_entry;
+
+       entry->slice = kdbus_pool_slice_alloc(conn_dst->pool, size, true);
+       if (IS_ERR(entry->slice)) {
+               ret = PTR_ERR(entry->slice);
+               entry->slice = NULL;
+               kdbus_conn_quota_dec(conn_dst, user, size,
+                                    res ? res->fds_count : 0);
+               goto exit_free_entry;
+       }
+
+       /* we accounted for exactly 'size' bytes, make sure it didn't grow */
+       WARN_ON(kdbus_pool_slice_size(entry->slice) != size);
+       entry->user = kdbus_user_ref(user);
+
+       /* copy message header */
+       kvec[0].iov_base = (char *)msg;
+       kvec[0].iov_len = msg_size;
+
+       ret = kdbus_pool_slice_copy_kvec(entry->slice, 0, kvec, 1, msg_size);
+       if (ret < 0)
+               goto exit_free_entry;
+
+       /* 'size' will now track the write position */
+       size = msg_size;
+
+       /* create message payload items */
+       if (res) {
+               size_t dst_name_len = 0;
+               unsigned int i;
+               size_t sz = 0;
+
+               if (res->dst_name) {
+                       dst_name_len = strlen(res->dst_name) + 1;
+                       sz += KDBUS_ITEM_SIZE(dst_name_len);
+               }
+
+               for (i = 0; i < res->data_count; ++i) {
+                       struct kdbus_vec v;
+                       struct kdbus_memfd m;
+
+                       switch (res->data[i].type) {
+                       case KDBUS_MSG_DATA_VEC:
+                               sz += KDBUS_ITEM_SIZE(sizeof(v));
+                               break;
+
+                       case KDBUS_MSG_DATA_MEMFD:
+                               sz += KDBUS_ITEM_SIZE(sizeof(m));
+                               break;
+                       }
+               }
+
+               if (sz) {
+                       struct kdbus_item *items, *item;
+
+                       items = kmalloc(sz, GFP_KERNEL);
+                       if (!items) {
+                               ret = -ENOMEM;
+                               goto exit_free_entry;
+                       }
+
+                       item = items;
+
+                       if (res->dst_name)
+                               item = kdbus_item_set(item, KDBUS_ITEM_DST_NAME,
+                                                     res->dst_name,
+                                                     dst_name_len);
+
+                       for (i = 0; i < res->data_count; ++i) {
+                               struct kdbus_msg_data *d = res->data + i;
+                               struct kdbus_memfd m = {};
+                               struct kdbus_vec v = {};
+
+                               switch (d->type) {
+                               case KDBUS_MSG_DATA_VEC:
+                                       v.size = d->size;
+                                       v.offset = d->vec.off;
+                                       if (v.offset != ~0ULL)
+                                               v.offset += payload_off;
+
+                                       item = kdbus_item_set(item,
+                                                       KDBUS_ITEM_PAYLOAD_OFF,
+                                                       &v, sizeof(v));
+                                       break;
+
+                               case KDBUS_MSG_DATA_MEMFD:
+                                       /*
+                                        * Remember the location of memfds, so
+                                        * we can override the content from
+                                        * kdbus_queue_entry_install().
+                                        */
+                                       entry->memfd_offset[memfd_cnt++] =
+                                               msg_size +
+                                               (char *)item - (char *)items +
+                                               offsetof(struct kdbus_item,
+                                                        memfd);
+
+                                       item = kdbus_item_set(item,
+                                                      KDBUS_ITEM_PAYLOAD_MEMFD,
+                                                      &m, sizeof(m));
+                                       break;
+                               }
+                       }
+
+                       kvec[0].iov_base = items;
+                       kvec[0].iov_len = sz;
+
+                       ret = kdbus_pool_slice_copy_kvec(entry->slice, size,
+                                                        kvec, 1, sz);
+                       kfree(items);
+
+                       if (ret < 0)
+                               goto exit_free_entry;
+
+                       size += sz;
+               }
+
+               /*
+                * Remember the location of the FD part, so we can override the
+                * content in kdbus_queue_entry_install().
+                */
+               if (res->fds_count) {
+                       entry->fds_offset = size;
+                       size += KDBUS_ITEM_SIZE(sizeof(int) * res->fds_count);
+               }
+       }
+
+       /* finally, copy over the actual message payload */
+       if (kmsg->iov_count) {
+               ret = kdbus_pool_slice_copy_iovec(entry->slice, payload_off,
+                                                 kmsg->iov,
+                                                 kmsg->iov_count,
+                                                 kmsg->pool_size);
+               if (ret < 0)
+                       goto exit_free_entry;
+       }
+
+       return entry;
+
+exit_free_entry:
+       kdbus_queue_entry_free(entry);
+       return ERR_PTR(ret);
+}
+
+/**
+ * kdbus_queue_entry_free() - free resources of an entry
+ * @entry:     The entry to free
+ *
+ * Removes resources allocated by a queue entry, along with the entry itself.
+ * Note that the entry's slice is not freed at this point.
+ */
+void kdbus_queue_entry_free(struct kdbus_queue_entry *entry)
+{
+       if (!entry)
+               return;
+
+       lockdep_assert_held(&entry->conn->lock);
+
+       kdbus_queue_entry_unlink(entry);
+       kdbus_reply_unref(entry->reply);
+
+       if (entry->slice) {
+               kdbus_conn_quota_dec(entry->conn, entry->user,
+                                    kdbus_pool_slice_size(entry->slice),
+                                    entry->msg_res ?
+                                               entry->msg_res->fds_count : 0);
+               kdbus_pool_slice_release(entry->slice);
+               kdbus_user_unref(entry->user);
+       }
+
+       kdbus_msg_resources_unref(entry->msg_res);
+       kdbus_meta_conn_unref(entry->conn_meta);
+       kdbus_meta_proc_unref(entry->proc_meta);
+       kdbus_conn_unref(entry->conn);
+       kfree(entry->memfd_offset);
+       kfree(entry);
+}
+
+/**
+ * kdbus_queue_entry_install() - install message components into the
+ *                              receiver's process
+ * @entry:             The queue entry to install
+ * @return_flags:      Pointer to store the return flags for userspace
+ * @install_fds:       Whether or not to install associated file descriptors
+ *
+ * This function will create a slice to transport the message header, the
+ * metadata items and other items for information stored in @entry, and
+ * store it as entry->slice.
+ *
+ * If @install_fds is %true, file descriptors will as well be installed.
+ * This function must always be called from the task context of the receiver.
+ *
+ * Return: 0 on success.
+ */
+int kdbus_queue_entry_install(struct kdbus_queue_entry *entry,
+                             u64 *return_flags, bool install_fds)
+{
+       u64 msg_size = entry->meta_offset;
+       struct kdbus_conn *conn_dst = entry->conn;
+       struct kdbus_msg_resources *res;
+       bool incomplete_fds = false;
+       struct kvec kvec[2];
+       size_t memfds = 0;
+       int i, ret;
+
+       lockdep_assert_held(&conn_dst->lock);
+
+       if (entry->proc_meta || entry->conn_meta) {
+               size_t meta_size;
+
+               ret = kdbus_meta_export(entry->proc_meta,
+                                       entry->conn_meta,
+                                       entry->attach_flags,
+                                       entry->slice,
+                                       entry->meta_offset,
+                                       &meta_size);
+               if (ret < 0)
+                       return ret;
+
+               msg_size += meta_size;
+       }
+
+       /* Update message size at offset 0 */
+       kvec[0].iov_base = &msg_size;
+       kvec[0].iov_len = sizeof(msg_size);
+
+       ret = kdbus_pool_slice_copy_kvec(entry->slice, 0, kvec, 1,
+                                        sizeof(msg_size));
+       if (ret < 0)
+               return ret;
+
+       res = entry->msg_res;
+
+       if (!res)
+               return 0;
+
+       if (res->fds_count) {
+               struct kdbus_item_header hdr;
+               size_t off;
+               int *fds;
+
+               fds = kmalloc_array(res->fds_count, sizeof(int), GFP_KERNEL);
+               if (!fds)
+                       return -ENOMEM;
+
+               for (i = 0; i < res->fds_count; i++) {
+                       if (install_fds) {
+                               fds[i] = get_unused_fd_flags(O_CLOEXEC);
+                               if (fds[i] >= 0)
+                                       fd_install(fds[i],
+                                                  get_file(res->fds[i]));
+                               else
+                                       incomplete_fds = true;
+                       } else {
+                               fds[i] = -1;
+                       }
+               }
+
+               off = entry->fds_offset;
+
+               hdr.type = KDBUS_ITEM_FDS;
+               hdr.size = KDBUS_ITEM_HEADER_SIZE +
+                          sizeof(int) * res->fds_count;
+
+               kvec[0].iov_base = &hdr;
+               kvec[0].iov_len = sizeof(hdr);
+
+               kvec[1].iov_base = fds;
+               kvec[1].iov_len = sizeof(int) * res->fds_count;
+
+               ret = kdbus_pool_slice_copy_kvec(entry->slice, off,
+                                                kvec, 2, hdr.size);
+               kfree(fds);
+
+               if (ret < 0)
+                       return ret;
+       }
+
+       for (i = 0; i < res->data_count; ++i) {
+               struct kdbus_msg_data *d = res->data + i;
+               struct kdbus_memfd m;
+
+               if (d->type != KDBUS_MSG_DATA_MEMFD)
+                       continue;
+
+               m.start = d->memfd.start;
+               m.size = d->size;
+               m.fd = -1;
+
+               if (install_fds) {
+                       m.fd = get_unused_fd_flags(O_CLOEXEC);
+                       if (m.fd < 0) {
+                               m.fd = -1;
+                               incomplete_fds = true;
+                       } else {
+                               fd_install(m.fd,
+                                          get_file(d->memfd.file));
+                       }
+               }
+
+               kvec[0].iov_base = &m;
+               kvec[0].iov_len = sizeof(m);
+
+               ret = kdbus_pool_slice_copy_kvec(entry->slice,
+                                                entry->memfd_offset[memfds++],
+                                                kvec, 1, sizeof(m));
+               if (ret < 0)
+                       return ret;
+       }
+
+       if (incomplete_fds)
+               *return_flags |= KDBUS_RECV_RETURN_INCOMPLETE_FDS;
+
+       return 0;
+}
+
+/**
+ * kdbus_queue_entry_enqueue() - enqueue an entry
+ * @entry:             entry to enqueue
+ * @reply:             reply to link to this entry (or NULL if none)
+ *
+ * This enqueues an unqueued entry into the message queue of the linked
+ * connection. It also binds a reply object to the entry so we can remember it
+ * when the message is moved.
+ *
+ * Once this call returns (and the connection lock is released), this entry can
+ * be dequeued by the target connection. Note that the entry will not be removed
+ * from the queue until it is destroyed.
+ */
+void kdbus_queue_entry_enqueue(struct kdbus_queue_entry *entry,
+                              struct kdbus_reply *reply)
+{
+       lockdep_assert_held(&entry->conn->lock);
+
+       if (WARN_ON(entry->reply) || WARN_ON(!list_empty(&entry->entry)))
+               return;
+
+       entry->reply = kdbus_reply_ref(reply);
+       kdbus_queue_entry_link(entry);
+}
+
+/**
+ * kdbus_queue_entry_move() - move queue entry
+ * @e:         queue entry to move
+ * @dst:       destination connection to queue the entry on
+ *
+ * This moves a queue entry onto a different connection. It allocates a new
+ * slice on the target connection and copies the message over. If the copy
+ * succeeded, we move the entry from @src to @dst.
+ *
+ * On failure, the entry is left untouched.
+ *
+ * The queue entry must be queued right now, and after the call succeeds it will
+ * be queued on the destination, but no longer on the source.
+ *
+ * The caller must hold the connection lock of the source *and* destination.
+ *
+ * Return: 0 on success, negative error code on failure.
+ */
+int kdbus_queue_entry_move(struct kdbus_queue_entry *e,
+                          struct kdbus_conn *dst)
+{
+       struct kdbus_pool_slice *slice = NULL;
+       struct kdbus_conn *src = e->conn;
+       size_t size, fds;
+       int ret;
+
+       lockdep_assert_held(&src->lock);
+       lockdep_assert_held(&dst->lock);
+
+       if (WARN_ON(IS_ERR(e->user)) || WARN_ON(list_empty(&e->entry)))
+               return -EINVAL;
+       if (src == dst)
+               return 0;
+
+       size = kdbus_pool_slice_size(e->slice);
+       fds = e->msg_res ? e->msg_res->fds_count : 0;
+
+       ret = kdbus_conn_quota_inc(dst, e->user, size, fds);
+       if (ret < 0)
+               return ret;
+
+       slice = kdbus_pool_slice_alloc(dst->pool, size, true);
+       if (IS_ERR(slice)) {
+               ret = PTR_ERR(slice);
+               slice = NULL;
+               goto error;
+       }
+
+       ret = kdbus_pool_slice_copy(slice, e->slice);
+       if (ret < 0)
+               goto error;
+
+       kdbus_queue_entry_unlink(e);
+       kdbus_conn_quota_dec(src, e->user, size, fds);
+       kdbus_pool_slice_release(e->slice);
+       kdbus_conn_unref(e->conn);
+
+       e->slice = slice;
+       e->conn = kdbus_conn_ref(dst);
+       kdbus_queue_entry_link(e);
+
+       return 0;
+
+error:
+       kdbus_pool_slice_release(slice);
+       kdbus_conn_quota_dec(dst, e->user, size, fds);
+       return ret;
+}
diff --git a/ipc/kdbus/queue.h b/ipc/kdbus/queue.h
new file mode 100644 (file)
index 0000000..7f2db96
--- /dev/null
@@ -0,0 +1,92 @@
+/*
+ * Copyright (C) 2013-2015 Kay Sievers
+ * Copyright (C) 2013-2015 Greg Kroah-Hartman <gregkh@linuxfoundation.org>
+ * Copyright (C) 2013-2015 Daniel Mack <daniel@zonque.org>
+ * Copyright (C) 2013-2015 David Herrmann <dh.herrmann@gmail.com>
+ * Copyright (C) 2013-2015 Linux Foundation
+ * Copyright (C) 2014-2015 Djalal Harouni <tixxdz@opendz.org>
+ *
+ * kdbus is free software; you can redistribute it and/or modify it under
+ * the terms of the GNU Lesser General Public License as published by the
+ * Free Software Foundation; either version 2.1 of the License, or (at
+ * your option) any later version.
+ */
+
+#ifndef __KDBUS_QUEUE_H
+#define __KDBUS_QUEUE_H
+
+struct kdbus_user;
+
+/**
+ * struct kdbus_queue - a connection's message queue
+ * @msg_list:          List head for kdbus_queue_entry objects
+ * @msg_prio_queue:    RB tree root for messages, sorted by priority
+ * @msg_prio_highest:  Link to the RB node referencing the message with the
+ *                     highest priority in the tree.
+ */
+struct kdbus_queue {
+       struct list_head msg_list;
+       struct rb_root msg_prio_queue;
+       struct rb_node *msg_prio_highest;
+};
+
+/**
+ * struct kdbus_queue_entry - messages waiting to be read
+ * @entry:             Entry in the connection's list
+ * @prio_node:         Entry in the priority queue tree
+ * @prio_entry:                Queue tree node entry in the list of one priority
+ * @slice:             Slice in the receiver's pool for the message
+ * @attach_flags:      Attach flags used during slice allocation
+ * @meta_offset:       Offset of first metadata item in slice
+ * @fds_offset:                Offset of FD item in slice
+ * @memfd_offset:      Array of slice-offsets for all memfd items
+ * @priority:          Message priority
+ * @dst_name_id:       The sequence number of the name this message is
+ *                     addressed to, 0 for messages sent to an ID
+ * @msg_res:           Message resources
+ * @proc_meta:         Process metadata, captured at message arrival
+ * @conn_meta:         Connection metadata, captured at message arrival
+ * @reply:             The reply block if a reply to this message is expected
+ * @user:              User used for accounting
+ */
+struct kdbus_queue_entry {
+       struct list_head entry;
+       struct rb_node prio_node;
+       struct list_head prio_entry;
+
+       struct kdbus_pool_slice *slice;
+
+       u64 attach_flags;
+       size_t meta_offset;
+       size_t fds_offset;
+       size_t *memfd_offset;
+
+       s64 priority;
+       u64 dst_name_id;
+
+       struct kdbus_msg_resources *msg_res;
+       struct kdbus_meta_proc *proc_meta;
+       struct kdbus_meta_conn *conn_meta;
+       struct kdbus_reply *reply;
+       struct kdbus_conn *conn;
+       struct kdbus_user *user;
+};
+
+struct kdbus_kmsg;
+
+void kdbus_queue_init(struct kdbus_queue *queue);
+struct kdbus_queue_entry *kdbus_queue_peek(struct kdbus_queue *queue,
+                                          s64 priority, bool use_priority);
+
+struct kdbus_queue_entry *kdbus_queue_entry_new(struct kdbus_conn *conn_dst,
+                                               const struct kdbus_kmsg *kmsg,
+                                               struct kdbus_user *user);
+void kdbus_queue_entry_free(struct kdbus_queue_entry *entry);
+int kdbus_queue_entry_install(struct kdbus_queue_entry *entry,
+                             u64 *return_flags, bool install_fds);
+void kdbus_queue_entry_enqueue(struct kdbus_queue_entry *entry,
+                              struct kdbus_reply *reply);
+int kdbus_queue_entry_move(struct kdbus_queue_entry *entry,
+                          struct kdbus_conn *dst);
+
+#endif /* __KDBUS_QUEUE_H */
diff --git a/ipc/kdbus/reply.c b/ipc/kdbus/reply.c
new file mode 100644 (file)
index 0000000..6b3bd81
--- /dev/null
@@ -0,0 +1,259 @@
+#include <linux/init.h>
+#include <linux/mm.h>
+#include <linux/module.h>
+#include <linux/mutex.h>
+#include <linux/slab.h>
+#include <linux/uio.h>
+
+#include "bus.h"
+#include "connection.h"
+#include "endpoint.h"
+#include "message.h"
+#include "metadata.h"
+#include "names.h"
+#include "domain.h"
+#include "item.h"
+#include "notify.h"
+#include "policy.h"
+#include "reply.h"
+#include "util.h"
+
+/**
+ * kdbus_reply_new() - Allocate and set up a new kdbus_reply object
+ * @reply_src:         The connection a reply is expected from
+ * @reply_dst:         The connection this reply object belongs to
+ * @msg:               Message associated with the reply
+ * @name_entry:                Name entry used to send the message
+ * @sync:              Whether or not to make this reply synchronous
+ *
+ * Allocate and fill a new kdbus_reply object.
+ *
+ * Return: New kdbus_conn object on success, ERR_PTR on error.
+ */
+struct kdbus_reply *kdbus_reply_new(struct kdbus_conn *reply_src,
+                                   struct kdbus_conn *reply_dst,
+                                   const struct kdbus_msg *msg,
+                                   struct kdbus_name_entry *name_entry,
+                                   bool sync)
+{
+       struct kdbus_reply *r;
+       int ret = 0;
+
+       if (atomic_inc_return(&reply_dst->request_count) >
+           KDBUS_CONN_MAX_REQUESTS_PENDING) {
+               ret = -EMLINK;
+               goto exit_dec_request_count;
+       }
+
+       r = kzalloc(sizeof(*r), GFP_KERNEL);
+       if (!r) {
+               ret = -ENOMEM;
+               goto exit_dec_request_count;
+       }
+
+       kref_init(&r->kref);
+       INIT_LIST_HEAD(&r->entry);
+       r->reply_src = kdbus_conn_ref(reply_src);
+       r->reply_dst = kdbus_conn_ref(reply_dst);
+       r->cookie = msg->cookie;
+       r->name_id = name_entry ? name_entry->name_id : 0;
+       r->deadline_ns = msg->timeout_ns;
+
+       if (sync) {
+               r->sync = true;
+               r->waiting = true;
+       }
+
+exit_dec_request_count:
+       if (ret < 0) {
+               atomic_dec(&reply_dst->request_count);
+               return ERR_PTR(ret);
+       }
+
+       return r;
+}
+
+static void __kdbus_reply_free(struct kref *kref)
+{
+       struct kdbus_reply *reply =
+               container_of(kref, struct kdbus_reply, kref);
+
+       atomic_dec(&reply->reply_dst->request_count);
+       kdbus_conn_unref(reply->reply_src);
+       kdbus_conn_unref(reply->reply_dst);
+       kfree(reply);
+}
+
+/**
+ * kdbus_reply_ref() - Increase reference on kdbus_reply
+ * @r:         The reply, may be %NULL
+ *
+ * Return: The reply object with an extra reference
+ */
+struct kdbus_reply *kdbus_reply_ref(struct kdbus_reply *r)
+{
+       if (r)
+               kref_get(&r->kref);
+       return r;
+}
+
+/**
+ * kdbus_reply_unref() - Decrease reference on kdbus_reply
+ * @r:         The reply, may be %NULL
+ *
+ * Return: NULL
+ */
+struct kdbus_reply *kdbus_reply_unref(struct kdbus_reply *r)
+{
+       if (r)
+               kref_put(&r->kref, __kdbus_reply_free);
+       return NULL;
+}
+
+/**
+ * kdbus_reply_link() - Link reply object into target connection
+ * @r:         Reply to link
+ */
+void kdbus_reply_link(struct kdbus_reply *r)
+{
+       if (WARN_ON(!list_empty(&r->entry)))
+               return;
+
+       list_add(&r->entry, &r->reply_dst->reply_list);
+       kdbus_reply_ref(r);
+}
+
+/**
+ * kdbus_reply_unlink() - Unlink reply object from target connection
+ * @r:         Reply to unlink
+ */
+void kdbus_reply_unlink(struct kdbus_reply *r)
+{
+       if (!list_empty(&r->entry)) {
+               list_del_init(&r->entry);
+               kdbus_reply_unref(r);
+       }
+}
+
+/**
+ * kdbus_sync_reply_wakeup() - Wake a synchronously blocking reply
+ * @reply:     The reply object
+ * @err:       Error code to set on the remote side
+ *
+ * Remove the synchronous reply object from its connection reply_list, and
+ * wake up remote peer (method origin) with the appropriate synchronous reply
+ * code.
+ */
+void kdbus_sync_reply_wakeup(struct kdbus_reply *reply, int err)
+{
+       if (WARN_ON(!reply->sync))
+               return;
+
+       reply->waiting = false;
+       reply->err = err;
+       wake_up_interruptible(&reply->reply_dst->wait);
+}
+
+/**
+ * kdbus_reply_find() - Find the corresponding reply object
+ * @replying:  The replying connection or NULL
+ * @reply_dst: The connection the reply will be sent to
+ *             (method origin)
+ * @cookie:    The cookie of the requesting message
+ *
+ * Lookup a reply object that should be sent as a reply by
+ * @replying to @reply_dst with the given cookie.
+ *
+ * Callers must take the @reply_dst lock.
+ *
+ * Return: the corresponding reply object or NULL if not found
+ */
+struct kdbus_reply *kdbus_reply_find(struct kdbus_conn *replying,
+                                    struct kdbus_conn *reply_dst,
+                                    u64 cookie)
+{
+       struct kdbus_reply *r, *reply = NULL;
+
+       list_for_each_entry(r, &reply_dst->reply_list, entry) {
+               if (r->cookie == cookie &&
+                   (!replying || r->reply_src == replying)) {
+                       reply = r;
+                       break;
+               }
+       }
+
+       return reply;
+}
+
+/**
+ * kdbus_reply_list_scan_work() - Worker callback to scan the replies of a
+ *                               connection for exceeded timeouts
+ * @work:              Work struct of the connection to scan
+ *
+ * Walk the list of replies stored with a connection and look for entries
+ * that have exceeded their timeout. If such an entry is found, a timeout
+ * notification is sent to the waiting peer, and the reply is removed from
+ * the list.
+ *
+ * The work is rescheduled to the nearest timeout found during the list
+ * iteration.
+ */
+void kdbus_reply_list_scan_work(struct work_struct *work)
+{
+       struct kdbus_conn *conn =
+               container_of(work, struct kdbus_conn, work.work);
+       struct kdbus_reply *reply, *reply_tmp;
+       u64 deadline = ~0ULL;
+       struct timespec64 ts;
+       u64 now;
+
+       ktime_get_ts64(&ts);
+       now = timespec64_to_ns(&ts);
+
+       mutex_lock(&conn->lock);
+       if (!kdbus_conn_active(conn)) {
+               mutex_unlock(&conn->lock);
+               return;
+       }
+
+       list_for_each_entry_safe(reply, reply_tmp, &conn->reply_list, entry) {
+               /*
+                * If the reply block is waiting for synchronous I/O,
+                * the timeout is handled by wait_event_*_timeout(),
+                * so we don't have to care for it here.
+                */
+               if (reply->sync && !reply->interrupted)
+                       continue;
+
+               WARN_ON(reply->reply_dst != conn);
+
+               if (reply->deadline_ns > now) {
+                       /* remember next timeout */
+                       if (deadline > reply->deadline_ns)
+                               deadline = reply->deadline_ns;
+
+                       continue;
+               }
+
+               /*
+                * A zero deadline means the connection died, was
+                * cleaned up already and the notification was sent.
+                * Don't send notifications for reply trackers that were
+                * left in an interrupted syscall state.
+                */
+               if (reply->deadline_ns != 0 && !reply->interrupted)
+                       kdbus_notify_reply_timeout(conn->ep->bus, conn->id,
+                                                  reply->cookie);
+
+               kdbus_reply_unlink(reply);
+       }
+
+       /* rearm delayed work with next timeout */
+       if (deadline != ~0ULL)
+               schedule_delayed_work(&conn->work,
+                                     nsecs_to_jiffies(deadline - now));
+
+       mutex_unlock(&conn->lock);
+
+       kdbus_notify_flush(conn->ep->bus);
+}
diff --git a/ipc/kdbus/reply.h b/ipc/kdbus/reply.h
new file mode 100644 (file)
index 0000000..68d5232
--- /dev/null
@@ -0,0 +1,68 @@
+/*
+ * Copyright (C) 2013-2015 Kay Sievers
+ * Copyright (C) 2013-2015 Greg Kroah-Hartman <gregkh@linuxfoundation.org>
+ * Copyright (C) 2013-2015 Daniel Mack <daniel@zonque.org>
+ * Copyright (C) 2013-2015 David Herrmann <dh.herrmann@gmail.com>
+ * Copyright (C) 2013-2015 Linux Foundation
+ * Copyright (C) 2014-2015 Djalal Harouni <tixxdz@opendz.org>
+ *
+ * kdbus is free software; you can redistribute it and/or modify it under
+ * the terms of the GNU Lesser General Public License as published by the
+ * Free Software Foundation; either version 2.1 of the License, or (at
+ * your option) any later version.
+ */
+
+#ifndef __KDBUS_REPLY_H
+#define __KDBUS_REPLY_H
+
+/**
+ * struct kdbus_reply - an entry of kdbus_conn's list of replies
+ * @kref:              Ref-count of this object
+ * @entry:             The entry of the connection's reply_list
+ * @reply_src:         The connection the reply will be sent from
+ * @reply_dst:         The connection the reply will be sent to
+ * @queue_entry:       The queue entry item that is prepared by the replying
+ *                     connection
+ * @deadline_ns:       The deadline of the reply, in nanoseconds
+ * @cookie:            The cookie of the requesting message
+ * @name_id:           ID of the well-known name the original msg was sent to
+ * @sync:              The reply block is waiting for synchronous I/O
+ * @waiting:           The condition to synchronously wait for
+ * @interrupted:       The sync reply was left in an interrupted state
+ * @err:               The error code for the synchronous reply
+ */
+struct kdbus_reply {
+       struct kref kref;
+       struct list_head entry;
+       struct kdbus_conn *reply_src;
+       struct kdbus_conn *reply_dst;
+       struct kdbus_queue_entry *queue_entry;
+       u64 deadline_ns;
+       u64 cookie;
+       u64 name_id;
+       bool sync:1;
+       bool waiting:1;
+       bool interrupted:1;
+       int err;
+};
+
+struct kdbus_reply *kdbus_reply_new(struct kdbus_conn *reply_src,
+                                   struct kdbus_conn *reply_dst,
+                                   const struct kdbus_msg *msg,
+                                   struct kdbus_name_entry *name_entry,
+                                   bool sync);
+
+struct kdbus_reply *kdbus_reply_ref(struct kdbus_reply *r);
+struct kdbus_reply *kdbus_reply_unref(struct kdbus_reply *r);
+
+void kdbus_reply_link(struct kdbus_reply *r);
+void kdbus_reply_unlink(struct kdbus_reply *r);
+
+struct kdbus_reply *kdbus_reply_find(struct kdbus_conn *replying,
+                                    struct kdbus_conn *reply_dst,
+                                    u64 cookie);
+
+void kdbus_sync_reply_wakeup(struct kdbus_reply *reply, int err);
+void kdbus_reply_list_scan_work(struct work_struct *work);
+
+#endif /* __KDBUS_REPLY_H */
index 9caadb3..740b198 100644 (file)
@@ -18,7 +18,7 @@
 #include <linux/dcache.h>
 #include <linux/ioctl.h>
 
-#include "kdbus.h"
+#include <uapi/linux/kdbus.h>
 
 /* all exported addresses are 64 bit */
 #define KDBUS_PTR(addr) ((void __user *)(uintptr_t)(addr))