Revert "drm: don't associate _DRM_DRIVER maps with a master"
[safe/jmp/linux-2.6] / drivers / firewire / fw-cdev.c
index ed03234..7eb6594 100644 (file)
  * Inc., 59 Temple Place - Suite 330, Boston, MA 02111-1307, USA.
  */
 
-#include <linux/module.h>
-#include <linux/kernel.h>
-#include <linux/wait.h>
-#include <linux/errno.h>
+#include <linux/compat.h>
+#include <linux/delay.h>
 #include <linux/device.h>
-#include <linux/vmalloc.h>
+#include <linux/errno.h>
+#include <linux/firewire-cdev.h>
+#include <linux/idr.h>
+#include <linux/jiffies.h>
+#include <linux/kernel.h>
+#include <linux/kref.h>
+#include <linux/mm.h>
+#include <linux/module.h>
+#include <linux/mutex.h>
 #include <linux/poll.h>
 #include <linux/preempt.h>
+#include <linux/spinlock.h>
 #include <linux/time.h>
-#include <linux/delay.h>
-#include <linux/mm.h>
-#include <linux/idr.h>
-#include <linux/compat.h>
-#include <linux/firewire-cdev.h>
+#include <linux/vmalloc.h>
+#include <linux/wait.h>
+#include <linux/workqueue.h>
+
 #include <asm/system.h>
 #include <asm/uaccess.h>
-#include "fw-transaction.h"
-#include "fw-topology.h"
+
 #include "fw-device.h"
+#include "fw-topology.h"
+#include "fw-transaction.h"
+
+struct client {
+       u32 version;
+       struct fw_device *device;
+
+       spinlock_t lock;
+       bool in_shutdown;
+       struct idr resource_idr;
+       struct list_head event_list;
+       wait_queue_head_t wait;
+       u64 bus_reset_closure;
+
+       struct fw_iso_context *iso_context;
+       u64 iso_closure;
+       struct fw_iso_buffer buffer;
+       unsigned long vm_start;
 
-struct client;
-struct client_resource {
        struct list_head link;
-       void (*release)(struct client *client, struct client_resource *r);
-       u32 handle;
+       struct kref kref;
 };
 
+static inline void client_get(struct client *client)
+{
+       kref_get(&client->kref);
+}
+
+static void client_release(struct kref *kref)
+{
+       struct client *client = container_of(kref, struct client, kref);
+
+       fw_device_put(client->device);
+       kfree(client);
+}
+
+static void client_put(struct client *client)
+{
+       kref_put(&client->kref, client_release);
+}
+
+struct client_resource;
+typedef void (*client_resource_release_fn_t)(struct client *,
+                                            struct client_resource *);
+struct client_resource {
+       client_resource_release_fn_t release;
+       int handle;
+};
+
+struct address_handler_resource {
+       struct client_resource resource;
+       struct fw_address_handler handler;
+       __u64 closure;
+       struct client *client;
+};
+
+struct outbound_transaction_resource {
+       struct client_resource resource;
+       struct fw_transaction transaction;
+};
+
+struct inbound_transaction_resource {
+       struct client_resource resource;
+       struct fw_request *request;
+       void *data;
+       size_t length;
+};
+
+struct descriptor_resource {
+       struct client_resource resource;
+       struct fw_descriptor descriptor;
+       u32 data[0];
+};
+
+struct iso_resource {
+       struct client_resource resource;
+       struct client *client;
+       /* Schedule work and access todo only with client->lock held. */
+       struct delayed_work work;
+       enum {ISO_RES_ALLOC, ISO_RES_REALLOC, ISO_RES_DEALLOC,
+             ISO_RES_ALLOC_ONCE, ISO_RES_DEALLOC_ONCE,} todo;
+       int generation;
+       u64 channels;
+       s32 bandwidth;
+       struct iso_resource_event *e_alloc, *e_dealloc;
+};
+
+static void schedule_iso_resource(struct iso_resource *);
+static void release_iso_resource(struct client *, struct client_resource *);
+
 /*
  * dequeue_event() just kfree()'s the event, so the event has to be
- * the first field in the struct.
+ * the first field in a struct XYZ_event.
  */
-
 struct event {
        struct { void *data; size_t size; } v[2];
        struct list_head link;
 };
 
-struct bus_reset {
+struct bus_reset_event {
        struct event event;
        struct fw_cdev_event_bus_reset reset;
 };
 
-struct response {
+struct outbound_transaction_event {
        struct event event;
-       struct fw_transaction transaction;
        struct client *client;
-       struct client_resource resource;
+       struct outbound_transaction_resource r;
        struct fw_cdev_event_response response;
 };
 
-struct iso_interrupt {
+struct inbound_transaction_event {
        struct event event;
-       struct fw_cdev_event_iso_interrupt interrupt;
+       struct fw_cdev_event_request request;
 };
 
-struct client {
-       u32 version;
-       struct fw_device *device;
-       spinlock_t lock;
-       u32 resource_handle;
-       struct list_head resource_list;
-       struct list_head event_list;
-       wait_queue_head_t wait;
-       u64 bus_reset_closure;
-
-       struct fw_iso_context *iso_context;
-       u64 iso_closure;
-       struct fw_iso_buffer buffer;
-       unsigned long vm_start;
+struct iso_interrupt_event {
+       struct event event;
+       struct fw_cdev_event_iso_interrupt interrupt;
+};
 
-       struct list_head link;
+struct iso_resource_event {
+       struct event event;
+       struct fw_cdev_event_iso_resource resource;
 };
 
-static inline void __user *
-u64_to_uptr(__u64 value)
+static inline void __user *u64_to_uptr(__u64 value)
 {
        return (void __user *)(unsigned long)value;
 }
 
-static inline __u64
-uptr_to_u64(void __user *ptr)
+static inline __u64 uptr_to_u64(void __user *ptr)
 {
        return (__u64)(unsigned long)ptr;
 }
@@ -107,7 +182,6 @@ static int fw_device_op_open(struct inode *inode, struct file *file)
 {
        struct fw_device *device;
        struct client *client;
-       unsigned long flags;
 
        device = fw_device_get_by_devt(inode->i_rdev);
        if (device == NULL)
@@ -125,16 +199,17 @@ static int fw_device_op_open(struct inode *inode, struct file *file)
        }
 
        client->device = device;
-       INIT_LIST_HEAD(&client->event_list);
-       INIT_LIST_HEAD(&client->resource_list);
        spin_lock_init(&client->lock);
+       idr_init(&client->resource_idr);
+       INIT_LIST_HEAD(&client->event_list);
        init_waitqueue_head(&client->wait);
+       kref_init(&client->kref);
 
        file->private_data = client;
 
-       spin_lock_irqsave(&device->card->lock, flags);
+       mutex_lock(&device->client_list_mutex);
        list_add_tail(&client->link, &device->client_list);
-       spin_unlock_irqrestore(&device->card->lock, flags);
+       mutex_unlock(&device->client_list_mutex);
 
        return 0;
 }
@@ -150,68 +225,69 @@ static void queue_event(struct client *client, struct event *event,
        event->v[1].size = size1;
 
        spin_lock_irqsave(&client->lock, flags);
-       list_add_tail(&event->link, &client->event_list);
+       if (client->in_shutdown)
+               kfree(event);
+       else
+               list_add_tail(&event->link, &client->event_list);
        spin_unlock_irqrestore(&client->lock, flags);
 
        wake_up_interruptible(&client->wait);
 }
 
-static int
-dequeue_event(struct client *client, char __user *buffer, size_t count)
+static int dequeue_event(struct client *client,
+                        char __user *buffer, size_t count)
 {
-       unsigned long flags;
        struct event *event;
        size_t size, total;
-       int i, retval;
+       int i, ret;
 
-       retval = wait_event_interruptible(client->wait,
-                                         !list_empty(&client->event_list) ||
-                                         fw_device_is_shutdown(client->device));
-       if (retval < 0)
-               return retval;
+       ret = wait_event_interruptible(client->wait,
+                       !list_empty(&client->event_list) ||
+                       fw_device_is_shutdown(client->device));
+       if (ret < 0)
+               return ret;
 
        if (list_empty(&client->event_list) &&
                       fw_device_is_shutdown(client->device))
                return -ENODEV;
 
-       spin_lock_irqsave(&client->lock, flags);
-       event = container_of(client->event_list.next, struct event, link);
+       spin_lock_irq(&client->lock);
+       event = list_first_entry(&client->event_list, struct event, link);
        list_del(&event->link);
-       spin_unlock_irqrestore(&client->lock, flags);
+       spin_unlock_irq(&client->lock);
 
        total = 0;
        for (i = 0; i < ARRAY_SIZE(event->v) && total < count; i++) {
                size = min(event->v[i].size, count - total);
                if (copy_to_user(buffer + total, event->v[i].data, size)) {
-                       retval = -EFAULT;
+                       ret = -EFAULT;
                        goto out;
                }
                total += size;
        }
-       retval = total;
+       ret = total;
 
  out:
        kfree(event);
 
-       return retval;
+       return ret;
 }
 
-static ssize_t
-fw_device_op_read(struct file *file,
-                 char __user *buffer, size_t count, loff_t *offset)
+static ssize_t fw_device_op_read(struct file *file, char __user *buffer,
+                                size_t count, loff_t *offset)
 {
        struct client *client = file->private_data;
 
        return dequeue_event(client, buffer, count);
 }
 
-/* caller must hold card->lock so that node pointers can be dereferenced here */
-static void
-fill_bus_reset_event(struct fw_cdev_event_bus_reset *event,
-                    struct client *client)
+static void fill_bus_reset_event(struct fw_cdev_event_bus_reset *event,
+                                struct client *client)
 {
        struct fw_card *card = client->device->card;
 
+       spin_lock_irq(&card->lock);
+
        event->closure       = client->bus_reset_closure;
        event->type          = FW_CDEV_EVENT_BUS_RESET;
        event->generation    = client->device->generation;
@@ -220,39 +296,49 @@ fill_bus_reset_event(struct fw_cdev_event_bus_reset *event,
        event->bm_node_id    = 0; /* FIXME: We don't track the BM. */
        event->irm_node_id   = card->irm_node->node_id;
        event->root_node_id  = card->root_node->node_id;
+
+       spin_unlock_irq(&card->lock);
 }
 
-static void
-for_each_client(struct fw_device *device,
-               void (*callback)(struct client *client))
+static void for_each_client(struct fw_device *device,
+                           void (*callback)(struct client *client))
 {
-       struct fw_card *card = device->card;
        struct client *c;
-       unsigned long flags;
-
-       spin_lock_irqsave(&card->lock, flags);
 
+       mutex_lock(&device->client_list_mutex);
        list_for_each_entry(c, &device->client_list, link)
                callback(c);
+       mutex_unlock(&device->client_list_mutex);
+}
+
+static int schedule_reallocations(int id, void *p, void *data)
+{
+       struct client_resource *r = p;
 
-       spin_unlock_irqrestore(&card->lock, flags);
+       if (r->release == release_iso_resource)
+               schedule_iso_resource(container_of(r,
+                                       struct iso_resource, resource));
+       return 0;
 }
 
-static void
-queue_bus_reset_event(struct client *client)
+static void queue_bus_reset_event(struct client *client)
 {
-       struct bus_reset *bus_reset;
+       struct bus_reset_event *e;
 
-       bus_reset = kzalloc(sizeof(*bus_reset), GFP_ATOMIC);
-       if (bus_reset == NULL) {
+       e = kzalloc(sizeof(*e), GFP_KERNEL);
+       if (e == NULL) {
                fw_notify("Out of memory when allocating bus reset event\n");
                return;
        }
 
-       fill_bus_reset_event(&bus_reset->reset, client);
+       fill_bus_reset_event(&e->reset, client);
+
+       queue_event(client, &e->event,
+                   &e->reset, sizeof(e->reset), NULL, 0);
 
-       queue_event(client, &bus_reset->event,
-                   &bus_reset->reset, sizeof(bus_reset->reset), NULL, 0);
+       spin_lock_irq(&client->lock);
+       idr_for_each(&client->resource_idr, schedule_reallocations, client);
+       spin_unlock_irq(&client->lock);
 }
 
 void fw_device_cdev_update(struct fw_device *device)
@@ -274,11 +360,11 @@ static int ioctl_get_info(struct client *client, void *buffer)
 {
        struct fw_cdev_get_info *get_info = buffer;
        struct fw_cdev_event_bus_reset bus_reset;
-       struct fw_card *card = client->device->card;
        unsigned long ret = 0;
 
        client->version = get_info->version;
        get_info->version = FW_CDEV_VERSION;
+       get_info->card = client->device->card->index;
 
        down_read(&fw_device_rwsem);
 
@@ -300,49 +386,61 @@ static int ioctl_get_info(struct client *client, void *buffer)
        client->bus_reset_closure = get_info->bus_reset_closure;
        if (get_info->bus_reset != 0) {
                void __user *uptr = u64_to_uptr(get_info->bus_reset);
-               unsigned long flags;
 
-               spin_lock_irqsave(&card->lock, flags);
                fill_bus_reset_event(&bus_reset, client);
-               spin_unlock_irqrestore(&card->lock, flags);
-
                if (copy_to_user(uptr, &bus_reset, sizeof(bus_reset)))
                        return -EFAULT;
        }
 
-       get_info->card = card->index;
-
        return 0;
 }
 
-static void
-add_client_resource(struct client *client, struct client_resource *resource)
+static int add_client_resource(struct client *client,
+                              struct client_resource *resource, gfp_t gfp_mask)
 {
        unsigned long flags;
+       int ret;
+
+ retry:
+       if (idr_pre_get(&client->resource_idr, gfp_mask) == 0)
+               return -ENOMEM;
 
        spin_lock_irqsave(&client->lock, flags);
-       list_add_tail(&resource->link, &client->resource_list);
-       resource->handle = client->resource_handle++;
+       if (client->in_shutdown)
+               ret = -ECANCELED;
+       else
+               ret = idr_get_new(&client->resource_idr, resource,
+                                 &resource->handle);
+       if (ret >= 0) {
+               client_get(client);
+               if (resource->release == release_iso_resource)
+                       schedule_iso_resource(container_of(resource,
+                                               struct iso_resource, resource));
+       }
        spin_unlock_irqrestore(&client->lock, flags);
+
+       if (ret == -EAGAIN)
+               goto retry;
+
+       return ret < 0 ? ret : 0;
 }
 
-static int
-release_client_resource(struct client *client, u32 handle,
-                       struct client_resource **resource)
+static int release_client_resource(struct client *client, u32 handle,
+                                  client_resource_release_fn_t release,
+                                  struct client_resource **resource)
 {
        struct client_resource *r;
-       unsigned long flags;
 
-       spin_lock_irqsave(&client->lock, flags);
-       list_for_each_entry(r, &client->resource_list, link) {
-               if (r->handle == handle) {
-                       list_del(&r->link);
-                       break;
-               }
-       }
-       spin_unlock_irqrestore(&client->lock, flags);
+       spin_lock_irq(&client->lock);
+       if (client->in_shutdown)
+               r = NULL;
+       else
+               r = idr_find(&client->resource_idr, handle);
+       if (r && r->release == release)
+               idr_remove(&client->resource_idr, handle);
+       spin_unlock_irq(&client->lock);
 
-       if (&r->link == &client->resource_list)
+       if (!(r && r->release == release))
                return -EINVAL;
 
        if (resource)
@@ -350,203 +448,239 @@ release_client_resource(struct client *client, u32 handle,
        else
                r->release(client, r);
 
+       client_put(client);
+
        return 0;
 }
 
-static void
-release_transaction(struct client *client, struct client_resource *resource)
+static void release_transaction(struct client *client,
+                               struct client_resource *resource)
 {
-       struct response *response =
-               container_of(resource, struct response, resource);
+       struct outbound_transaction_resource *r = container_of(resource,
+                       struct outbound_transaction_resource, resource);
 
-       fw_cancel_transaction(client->device->card, &response->transaction);
+       fw_cancel_transaction(client->device->card, &r->transaction);
 }
 
-static void
-complete_transaction(struct fw_card *card, int rcode,
-                    void *payload, size_t length, void *data)
+static void complete_transaction(struct fw_card *card, int rcode,
+                                void *payload, size_t length, void *data)
 {
-       struct response *response = data;
-       struct client *client = response->client;
+       struct outbound_transaction_event *e = data;
+       struct fw_cdev_event_response *rsp = &e->response;
+       struct client *client = e->client;
        unsigned long flags;
-       struct fw_cdev_event_response *r = &response->response;
 
-       if (length < r->length)
-               r->length = length;
+       if (length < rsp->length)
+               rsp->length = length;
        if (rcode == RCODE_COMPLETE)
-               memcpy(r->data, payload, r->length);
+               memcpy(rsp->data, payload, rsp->length);
 
        spin_lock_irqsave(&client->lock, flags);
-       list_del(&response->resource.link);
+       /*
+        * 1. If called while in shutdown, the idr tree must be left untouched.
+        *    The idr handle will be removed and the client reference will be
+        *    dropped later.
+        * 2. If the call chain was release_client_resource ->
+        *    release_transaction -> complete_transaction (instead of a normal
+        *    conclusion of the transaction), i.e. if this resource was already
+        *    unregistered from the idr, the client reference will be dropped
+        *    by release_client_resource and we must not drop it here.
+        */
+       if (!client->in_shutdown &&
+           idr_find(&client->resource_idr, e->r.resource.handle)) {
+               idr_remove(&client->resource_idr, e->r.resource.handle);
+               /* Drop the idr's reference */
+               client_put(client);
+       }
        spin_unlock_irqrestore(&client->lock, flags);
 
-       r->type   = FW_CDEV_EVENT_RESPONSE;
-       r->rcode  = rcode;
+       rsp->type = FW_CDEV_EVENT_RESPONSE;
+       rsp->rcode = rcode;
 
        /*
-        * In the case that sizeof(*r) doesn't align with the position of the
+        * In the case that sizeof(*rsp) doesn't align with the position of the
         * data, and the read is short, preserve an extra copy of the data
         * to stay compatible with a pre-2.6.27 bug.  Since the bug is harmless
         * for short reads and some apps depended on it, this is both safe
         * and prudent for compatibility.
         */
-       if (r->length <= sizeof(*r) - offsetof(typeof(*r), data))
-               queue_event(client, &response->event, r, sizeof(*r),
-                           r->data, r->length);
+       if (rsp->length <= sizeof(*rsp) - offsetof(typeof(*rsp), data))
+               queue_event(client, &e->event, rsp, sizeof(*rsp),
+                           rsp->data, rsp->length);
        else
-               queue_event(client, &response->event, r, sizeof(*r) + r->length,
+               queue_event(client, &e->event, rsp, sizeof(*rsp) + rsp->length,
                            NULL, 0);
+
+       /* Drop the transaction callback's reference */
+       client_put(client);
 }
 
-static int ioctl_send_request(struct client *client, void *buffer)
+static int init_request(struct client *client,
+                       struct fw_cdev_send_request *request,
+                       int destination_id, int speed)
 {
-       struct fw_device *device = client->device;
-       struct fw_cdev_send_request *request = buffer;
-       struct response *response;
+       struct outbound_transaction_event *e;
+       int ret;
 
-       /* What is the biggest size we'll accept, really? */
-       if (request->length > 4096)
-               return -EINVAL;
+       if (request->tcode != TCODE_STREAM_DATA &&
+           (request->length > 4096 || request->length > 512 << speed))
+               return -EIO;
 
-       response = kmalloc(sizeof(*response) + request->length, GFP_KERNEL);
-       if (response == NULL)
+       e = kmalloc(sizeof(*e) + request->length, GFP_KERNEL);
+       if (e == NULL)
                return -ENOMEM;
 
-       response->client = client;
-       response->response.length = request->length;
-       response->response.closure = request->closure;
+       e->client = client;
+       e->response.length = request->length;
+       e->response.closure = request->closure;
 
        if (request->data &&
-           copy_from_user(response->response.data,
+           copy_from_user(e->response.data,
                           u64_to_uptr(request->data), request->length)) {
-               kfree(response);
-               return -EFAULT;
+               ret = -EFAULT;
+               goto failed;
        }
 
-       response->resource.release = release_transaction;
-       add_client_resource(client, &response->resource);
+       e->r.resource.release = release_transaction;
+       ret = add_client_resource(client, &e->r.resource, GFP_KERNEL);
+       if (ret < 0)
+               goto failed;
 
-       fw_send_request(device->card, &response->transaction,
-                       request->tcode & 0x1f,
-                       device->node->node_id,
-                       request->generation,
-                       device->max_speed,
-                       request->offset,
-                       response->response.data, request->length,
-                       complete_transaction, response);
+       /* Get a reference for the transaction callback */
+       client_get(client);
 
-       if (request->data)
-               return sizeof(request) + request->length;
-       else
-               return sizeof(request);
+       fw_send_request(client->device->card, &e->r.transaction,
+                       request->tcode, destination_id, request->generation,
+                       speed, request->offset, e->response.data,
+                       request->length, complete_transaction, e);
+       return 0;
+
+ failed:
+       kfree(e);
+
+       return ret;
 }
 
-struct address_handler {
-       struct fw_address_handler handler;
-       __u64 closure;
-       struct client *client;
-       struct client_resource resource;
-};
+static int ioctl_send_request(struct client *client, void *buffer)
+{
+       struct fw_cdev_send_request *request = buffer;
 
-struct request {
-       struct fw_request *request;
-       void *data;
-       size_t length;
-       struct client_resource resource;
-};
+       switch (request->tcode) {
+       case TCODE_WRITE_QUADLET_REQUEST:
+       case TCODE_WRITE_BLOCK_REQUEST:
+       case TCODE_READ_QUADLET_REQUEST:
+       case TCODE_READ_BLOCK_REQUEST:
+       case TCODE_LOCK_MASK_SWAP:
+       case TCODE_LOCK_COMPARE_SWAP:
+       case TCODE_LOCK_FETCH_ADD:
+       case TCODE_LOCK_LITTLE_ADD:
+       case TCODE_LOCK_BOUNDED_ADD:
+       case TCODE_LOCK_WRAP_ADD:
+       case TCODE_LOCK_VENDOR_DEPENDENT:
+               break;
+       default:
+               return -EINVAL;
+       }
 
-struct request_event {
-       struct event event;
-       struct fw_cdev_event_request request;
-};
+       return init_request(client, request, client->device->node_id,
+                           client->device->max_speed);
+}
 
-static void
-release_request(struct client *client, struct client_resource *resource)
+static void release_request(struct client *client,
+                           struct client_resource *resource)
 {
-       struct request *request =
-               container_of(resource, struct request, resource);
+       struct inbound_transaction_resource *r = container_of(resource,
+                       struct inbound_transaction_resource, resource);
 
-       fw_send_response(client->device->card, request->request,
+       fw_send_response(client->device->card, r->request,
                         RCODE_CONFLICT_ERROR);
-       kfree(request);
+       kfree(r);
 }
 
-static void
-handle_request(struct fw_card *card, struct fw_request *r,
-              int tcode, int destination, int source,
-              int generation, int speed,
-              unsigned long long offset,
-              void *payload, size_t length, void *callback_data)
+static void handle_request(struct fw_card *card, struct fw_request *request,
+                          int tcode, int destination, int source,
+                          int generation, int speed,
+                          unsigned long long offset,
+                          void *payload, size_t length, void *callback_data)
 {
-       struct address_handler *handler = callback_data;
-       struct request *request;
-       struct request_event *e;
-       struct client *client = handler->client;
+       struct address_handler_resource *handler = callback_data;
+       struct inbound_transaction_resource *r;
+       struct inbound_transaction_event *e;
+       int ret;
 
-       request = kmalloc(sizeof(*request), GFP_ATOMIC);
+       r = kmalloc(sizeof(*r), GFP_ATOMIC);
        e = kmalloc(sizeof(*e), GFP_ATOMIC);
-       if (request == NULL || e == NULL) {
-               kfree(request);
-               kfree(e);
-               fw_send_response(card, r, RCODE_CONFLICT_ERROR);
-               return;
-       }
+       if (r == NULL || e == NULL)
+               goto failed;
 
-       request->request = r;
-       request->data    = payload;
-       request->length  = length;
+       r->request = request;
+       r->data    = payload;
+       r->length  = length;
 
-       request->resource.release = release_request;
-       add_client_resource(client, &request->resource);
+       r->resource.release = release_request;
+       ret = add_client_resource(handler->client, &r->resource, GFP_ATOMIC);
+       if (ret < 0)
+               goto failed;
 
        e->request.type    = FW_CDEV_EVENT_REQUEST;
        e->request.tcode   = tcode;
        e->request.offset  = offset;
        e->request.length  = length;
-       e->request.handle  = request->resource.handle;
+       e->request.handle  = r->resource.handle;
        e->request.closure = handler->closure;
 
-       queue_event(client, &e->event,
+       queue_event(handler->client, &e->event,
                    &e->request, sizeof(e->request), payload, length);
+       return;
+
+ failed:
+       kfree(r);
+       kfree(e);
+       fw_send_response(card, request, RCODE_CONFLICT_ERROR);
 }
 
-static void
-release_address_handler(struct client *client,
-                       struct client_resource *resource)
+static void release_address_handler(struct client *client,
+                                   struct client_resource *resource)
 {
-       struct address_handler *handler =
-               container_of(resource, struct address_handler, resource);
+       struct address_handler_resource *r =
+           container_of(resource, struct address_handler_resource, resource);
 
-       fw_core_remove_address_handler(&handler->handler);
-       kfree(handler);
+       fw_core_remove_address_handler(&r->handler);
+       kfree(r);
 }
 
 static int ioctl_allocate(struct client *client, void *buffer)
 {
        struct fw_cdev_allocate *request = buffer;
-       struct address_handler *handler;
+       struct address_handler_resource *r;
        struct fw_address_region region;
+       int ret;
 
-       handler = kmalloc(sizeof(*handler), GFP_KERNEL);
-       if (handler == NULL)
+       r = kmalloc(sizeof(*r), GFP_KERNEL);
+       if (r == NULL)
                return -ENOMEM;
 
        region.start = request->offset;
        region.end = request->offset + request->length;
-       handler->handler.length = request->length;
-       handler->handler.address_callback = handle_request;
-       handler->handler.callback_data = handler;
-       handler->closure = request->closure;
-       handler->client = client;
-
-       if (fw_core_add_address_handler(&handler->handler, &region) < 0) {
-               kfree(handler);
-               return -EBUSY;
+       r->handler.length = request->length;
+       r->handler.address_callback = handle_request;
+       r->handler.callback_data = r;
+       r->closure = request->closure;
+       r->client = client;
+
+       ret = fw_core_add_address_handler(&r->handler, &region);
+       if (ret < 0) {
+               kfree(r);
+               return ret;
        }
 
-       handler->resource.release = release_address_handler;
-       add_client_resource(client, &handler->resource);
-       request->handle = handler->resource.handle;
+       r->resource.release = release_address_handler;
+       ret = add_client_resource(client, &r->resource, GFP_KERNEL);
+       if (ret < 0) {
+               release_address_handler(client, &r->resource);
+               return ret;
+       }
+       request->handle = r->resource.handle;
 
        return 0;
 }
@@ -555,18 +689,22 @@ static int ioctl_deallocate(struct client *client, void *buffer)
 {
        struct fw_cdev_deallocate *request = buffer;
 
-       return release_client_resource(client, request->handle, NULL);
+       return release_client_resource(client, request->handle,
+                                      release_address_handler, NULL);
 }
 
 static int ioctl_send_response(struct client *client, void *buffer)
 {
        struct fw_cdev_send_response *request = buffer;
        struct client_resource *resource;
-       struct request *r;
+       struct inbound_transaction_resource *r;
 
-       if (release_client_resource(client, request->handle, &resource) < 0)
+       if (release_client_resource(client, request->handle,
+                                   release_request, &resource) < 0)
                return -EINVAL;
-       r = container_of(resource, struct request, resource);
+
+       r = container_of(resource, struct inbound_transaction_resource,
+                        resource);
        if (request->length < r->length)
                r->length = request->length;
        if (copy_from_user(r->data, u64_to_uptr(request->data), r->length))
@@ -588,85 +726,92 @@ static int ioctl_initiate_bus_reset(struct client *client, void *buffer)
        return fw_core_initiate_bus_reset(client->device->card, short_reset);
 }
 
-struct descriptor {
-       struct fw_descriptor d;
-       struct client_resource resource;
-       u32 data[0];
-};
-
 static void release_descriptor(struct client *client,
                               struct client_resource *resource)
 {
-       struct descriptor *descriptor =
-               container_of(resource, struct descriptor, resource);
+       struct descriptor_resource *r =
+               container_of(resource, struct descriptor_resource, resource);
 
-       fw_core_remove_descriptor(&descriptor->d);
-       kfree(descriptor);
+       fw_core_remove_descriptor(&r->descriptor);
+       kfree(r);
 }
 
 static int ioctl_add_descriptor(struct client *client, void *buffer)
 {
        struct fw_cdev_add_descriptor *request = buffer;
-       struct descriptor *descriptor;
-       int retval;
+       struct fw_card *card = client->device->card;
+       struct descriptor_resource *r;
+       int ret;
+
+       /* Access policy: Allow this ioctl only on local nodes' device files. */
+       spin_lock_irq(&card->lock);
+       ret = client->device->node_id != card->local_node->node_id;
+       spin_unlock_irq(&card->lock);
+       if (ret)
+               return -ENOSYS;
 
        if (request->length > 256)
                return -EINVAL;
 
-       descriptor =
-               kmalloc(sizeof(*descriptor) + request->length * 4, GFP_KERNEL);
-       if (descriptor == NULL)
+       r = kmalloc(sizeof(*r) + request->length * 4, GFP_KERNEL);
+       if (r == NULL)
                return -ENOMEM;
 
-       if (copy_from_user(descriptor->data,
+       if (copy_from_user(r->data,
                           u64_to_uptr(request->data), request->length * 4)) {
-               kfree(descriptor);
-               return -EFAULT;
+               ret = -EFAULT;
+               goto failed;
        }
 
-       descriptor->d.length = request->length;
-       descriptor->d.immediate = request->immediate;
-       descriptor->d.key = request->key;
-       descriptor->d.data = descriptor->data;
+       r->descriptor.length    = request->length;
+       r->descriptor.immediate = request->immediate;
+       r->descriptor.key       = request->key;
+       r->descriptor.data      = r->data;
 
-       retval = fw_core_add_descriptor(&descriptor->d);
-       if (retval < 0) {
-               kfree(descriptor);
-               return retval;
-       }
+       ret = fw_core_add_descriptor(&r->descriptor);
+       if (ret < 0)
+               goto failed;
 
-       descriptor->resource.release = release_descriptor;
-       add_client_resource(client, &descriptor->resource);
-       request->handle = descriptor->resource.handle;
+       r->resource.release = release_descriptor;
+       ret = add_client_resource(client, &r->resource, GFP_KERNEL);
+       if (ret < 0) {
+               fw_core_remove_descriptor(&r->descriptor);
+               goto failed;
+       }
+       request->handle = r->resource.handle;
 
        return 0;
+ failed:
+       kfree(r);
+
+       return ret;
 }
 
 static int ioctl_remove_descriptor(struct client *client, void *buffer)
 {
        struct fw_cdev_remove_descriptor *request = buffer;
 
-       return release_client_resource(client, request->handle, NULL);
+       return release_client_resource(client, request->handle,
+                                      release_descriptor, NULL);
 }
 
-static void
-iso_callback(struct fw_iso_context *context, u32 cycle,
-            size_t header_length, void *header, void *data)
+static void iso_callback(struct fw_iso_context *context, u32 cycle,
+                        size_t header_length, void *header, void *data)
 {
        struct client *client = data;
-       struct iso_interrupt *irq;
+       struct iso_interrupt_event *e;
 
-       irq = kzalloc(sizeof(*irq) + header_length, GFP_ATOMIC);
-       if (irq == NULL)
+       e = kzalloc(sizeof(*e) + header_length, GFP_ATOMIC);
+       if (e == NULL)
                return;
 
-       irq->interrupt.type      = FW_CDEV_EVENT_ISO_INTERRUPT;
-       irq->interrupt.closure   = client->iso_closure;
-       irq->interrupt.cycle     = cycle;
-       irq->interrupt.header_length = header_length;
-       memcpy(irq->interrupt.header, header, header_length);
-       queue_event(client, &irq->event, &irq->interrupt,
-                   sizeof(irq->interrupt) + header_length, NULL, 0);
+       e->interrupt.type      = FW_CDEV_EVENT_ISO_INTERRUPT;
+       e->interrupt.closure   = client->iso_closure;
+       e->interrupt.cycle     = cycle;
+       e->interrupt.header_length = header_length;
+       memcpy(e->interrupt.header, header, header_length);
+       queue_event(client, &e->event, &e->interrupt,
+                   sizeof(e->interrupt) + header_length, NULL, 0);
 }
 
 static int ioctl_create_iso_context(struct client *client, void *buffer)
@@ -871,6 +1016,261 @@ static int ioctl_get_cycle_timer(struct client *client, void *buffer)
        return 0;
 }
 
+static void iso_resource_work(struct work_struct *work)
+{
+       struct iso_resource_event *e;
+       struct iso_resource *r =
+                       container_of(work, struct iso_resource, work.work);
+       struct client *client = r->client;
+       int generation, channel, bandwidth, todo;
+       bool skip, free, success;
+
+       spin_lock_irq(&client->lock);
+       generation = client->device->generation;
+       todo = r->todo;
+       /* Allow 1000ms grace period for other reallocations. */
+       if (todo == ISO_RES_ALLOC &&
+           time_is_after_jiffies(client->device->card->reset_jiffies + HZ)) {
+               if (schedule_delayed_work(&r->work, DIV_ROUND_UP(HZ, 3)))
+                       client_get(client);
+               skip = true;
+       } else {
+               /* We could be called twice within the same generation. */
+               skip = todo == ISO_RES_REALLOC &&
+                      r->generation == generation;
+       }
+       free = todo == ISO_RES_DEALLOC ||
+              todo == ISO_RES_ALLOC_ONCE ||
+              todo == ISO_RES_DEALLOC_ONCE;
+       r->generation = generation;
+       spin_unlock_irq(&client->lock);
+
+       if (skip)
+               goto out;
+
+       bandwidth = r->bandwidth;
+
+       fw_iso_resource_manage(client->device->card, generation,
+                       r->channels, &channel, &bandwidth,
+                       todo == ISO_RES_ALLOC ||
+                       todo == ISO_RES_REALLOC ||
+                       todo == ISO_RES_ALLOC_ONCE);
+       /*
+        * Is this generation outdated already?  As long as this resource sticks
+        * in the idr, it will be scheduled again for a newer generation or at
+        * shutdown.
+        */
+       if (channel == -EAGAIN &&
+           (todo == ISO_RES_ALLOC || todo == ISO_RES_REALLOC))
+               goto out;
+
+       success = channel >= 0 || bandwidth > 0;
+
+       spin_lock_irq(&client->lock);
+       /*
+        * Transit from allocation to reallocation, except if the client
+        * requested deallocation in the meantime.
+        */
+       if (r->todo == ISO_RES_ALLOC)
+               r->todo = ISO_RES_REALLOC;
+       /*
+        * Allocation or reallocation failure?  Pull this resource out of the
+        * idr and prepare for deletion, unless the client is shutting down.
+        */
+       if (r->todo == ISO_RES_REALLOC && !success &&
+           !client->in_shutdown &&
+           idr_find(&client->resource_idr, r->resource.handle)) {
+               idr_remove(&client->resource_idr, r->resource.handle);
+               client_put(client);
+               free = true;
+       }
+       spin_unlock_irq(&client->lock);
+
+       if (todo == ISO_RES_ALLOC && channel >= 0)
+               r->channels = 1ULL << channel;
+
+       if (todo == ISO_RES_REALLOC && success)
+               goto out;
+
+       if (todo == ISO_RES_ALLOC || todo == ISO_RES_ALLOC_ONCE) {
+               e = r->e_alloc;
+               r->e_alloc = NULL;
+       } else {
+               e = r->e_dealloc;
+               r->e_dealloc = NULL;
+       }
+       e->resource.handle      = r->resource.handle;
+       e->resource.channel     = channel;
+       e->resource.bandwidth   = bandwidth;
+
+       queue_event(client, &e->event,
+                   &e->resource, sizeof(e->resource), NULL, 0);
+
+       if (free) {
+               cancel_delayed_work(&r->work);
+               kfree(r->e_alloc);
+               kfree(r->e_dealloc);
+               kfree(r);
+       }
+ out:
+       client_put(client);
+}
+
+static void schedule_iso_resource(struct iso_resource *r)
+{
+       client_get(r->client);
+       if (!schedule_delayed_work(&r->work, 0))
+               client_put(r->client);
+}
+
+static void release_iso_resource(struct client *client,
+                                struct client_resource *resource)
+{
+       struct iso_resource *r =
+               container_of(resource, struct iso_resource, resource);
+
+       spin_lock_irq(&client->lock);
+       r->todo = ISO_RES_DEALLOC;
+       schedule_iso_resource(r);
+       spin_unlock_irq(&client->lock);
+}
+
+static int init_iso_resource(struct client *client,
+               struct fw_cdev_allocate_iso_resource *request, int todo)
+{
+       struct iso_resource_event *e1, *e2;
+       struct iso_resource *r;
+       int ret;
+
+       if ((request->channels == 0 && request->bandwidth == 0) ||
+           request->bandwidth > BANDWIDTH_AVAILABLE_INITIAL ||
+           request->bandwidth < 0)
+               return -EINVAL;
+
+       r  = kmalloc(sizeof(*r), GFP_KERNEL);
+       e1 = kmalloc(sizeof(*e1), GFP_KERNEL);
+       e2 = kmalloc(sizeof(*e2), GFP_KERNEL);
+       if (r == NULL || e1 == NULL || e2 == NULL) {
+               ret = -ENOMEM;
+               goto fail;
+       }
+
+       INIT_DELAYED_WORK(&r->work, iso_resource_work);
+       r->client       = client;
+       r->todo         = todo;
+       r->generation   = -1;
+       r->channels     = request->channels;
+       r->bandwidth    = request->bandwidth;
+       r->e_alloc      = e1;
+       r->e_dealloc    = e2;
+
+       e1->resource.closure    = request->closure;
+       e1->resource.type       = FW_CDEV_EVENT_ISO_RESOURCE_ALLOCATED;
+       e2->resource.closure    = request->closure;
+       e2->resource.type       = FW_CDEV_EVENT_ISO_RESOURCE_DEALLOCATED;
+
+       if (todo == ISO_RES_ALLOC) {
+               r->resource.release = release_iso_resource;
+               ret = add_client_resource(client, &r->resource, GFP_KERNEL);
+               if (ret < 0)
+                       goto fail;
+       } else {
+               r->resource.release = NULL;
+               r->resource.handle = -1;
+               schedule_iso_resource(r);
+       }
+       request->handle = r->resource.handle;
+
+       return 0;
+ fail:
+       kfree(r);
+       kfree(e1);
+       kfree(e2);
+
+       return ret;
+}
+
+static int ioctl_allocate_iso_resource(struct client *client, void *buffer)
+{
+       struct fw_cdev_allocate_iso_resource *request = buffer;
+
+       return init_iso_resource(client, request, ISO_RES_ALLOC);
+}
+
+static int ioctl_deallocate_iso_resource(struct client *client, void *buffer)
+{
+       struct fw_cdev_deallocate *request = buffer;
+
+       return release_client_resource(client, request->handle,
+                                      release_iso_resource, NULL);
+}
+
+static int ioctl_allocate_iso_resource_once(struct client *client, void *buffer)
+{
+       struct fw_cdev_allocate_iso_resource *request = buffer;
+
+       return init_iso_resource(client, request, ISO_RES_ALLOC_ONCE);
+}
+
+static int ioctl_deallocate_iso_resource_once(struct client *client, void *buffer)
+{
+       struct fw_cdev_allocate_iso_resource *request = buffer;
+
+       return init_iso_resource(client, request, ISO_RES_DEALLOC_ONCE);
+}
+
+/*
+ * Returns a speed code:  Maximum speed to or from this device,
+ * limited by the device's link speed, the local node's link speed,
+ * and all PHY port speeds between the two links.
+ */
+static int ioctl_get_speed(struct client *client, void *buffer)
+{
+       return client->device->max_speed;
+}
+
+static int ioctl_send_broadcast_request(struct client *client, void *buffer)
+{
+       struct fw_cdev_send_request *request = buffer;
+
+       switch (request->tcode) {
+       case TCODE_WRITE_QUADLET_REQUEST:
+       case TCODE_WRITE_BLOCK_REQUEST:
+               break;
+       default:
+               return -EINVAL;
+       }
+
+       /* Security policy: Only allow accesses to Units Space. */
+       if (request->offset < CSR_REGISTER_BASE + CSR_CONFIG_ROM_END)
+               return -EACCES;
+
+       return init_request(client, request, LOCAL_BUS | 0x3f, SCODE_100);
+}
+
+static int ioctl_send_stream_packet(struct client *client, void *buffer)
+{
+       struct fw_cdev_send_stream_packet *p = buffer;
+       struct fw_cdev_send_request request;
+       int dest;
+
+       if (p->speed > client->device->card->link_speed ||
+           p->length > 1024 << p->speed)
+               return -EIO;
+
+       if (p->tag > 3 || p->channel > 63 || p->sy > 15)
+               return -EINVAL;
+
+       dest = fw_stream_packet_destination_id(p->tag, p->channel, p->sy);
+       request.tcode           = TCODE_STREAM_DATA;
+       request.length          = p->length;
+       request.closure         = p->closure;
+       request.data            = p->data;
+       request.generation      = p->generation;
+
+       return init_request(client, &request, dest, p->speed);
+}
+
 static int (* const ioctl_handlers[])(struct client *client, void *buffer) = {
        ioctl_get_info,
        ioctl_send_request,
@@ -885,13 +1285,20 @@ static int (* const ioctl_handlers[])(struct client *client, void *buffer) = {
        ioctl_start_iso,
        ioctl_stop_iso,
        ioctl_get_cycle_timer,
+       ioctl_allocate_iso_resource,
+       ioctl_deallocate_iso_resource,
+       ioctl_allocate_iso_resource_once,
+       ioctl_deallocate_iso_resource_once,
+       ioctl_get_speed,
+       ioctl_send_broadcast_request,
+       ioctl_send_stream_packet,
 };
 
-static int
-dispatch_ioctl(struct client *client, unsigned int cmd, void __user *arg)
+static int dispatch_ioctl(struct client *client,
+                         unsigned int cmd, void __user *arg)
 {
        char buffer[256];
-       int retval;
+       int ret;
 
        if (_IOC_TYPE(cmd) != '#' ||
            _IOC_NR(cmd) >= ARRAY_SIZE(ioctl_handlers))
@@ -903,9 +1310,9 @@ dispatch_ioctl(struct client *client, unsigned int cmd, void __user *arg)
                        return -EFAULT;
        }
 
-       retval = ioctl_handlers[_IOC_NR(cmd)](client, buffer);
-       if (retval < 0)
-               return retval;
+       ret = ioctl_handlers[_IOC_NR(cmd)](client, buffer);
+       if (ret < 0)
+               return ret;
 
        if (_IOC_DIR(cmd) & _IOC_READ) {
                if (_IOC_SIZE(cmd) > sizeof(buffer) ||
@@ -913,12 +1320,11 @@ dispatch_ioctl(struct client *client, unsigned int cmd, void __user *arg)
                        return -EFAULT;
        }
 
-       return retval;
+       return ret;
 }
 
-static long
-fw_device_op_ioctl(struct file *file,
-                  unsigned int cmd, unsigned long arg)
+static long fw_device_op_ioctl(struct file *file,
+                              unsigned int cmd, unsigned long arg)
 {
        struct client *client = file->private_data;
 
@@ -929,9 +1335,8 @@ fw_device_op_ioctl(struct file *file,
 }
 
 #ifdef CONFIG_COMPAT
-static long
-fw_device_op_compat_ioctl(struct file *file,
-                         unsigned int cmd, unsigned long arg)
+static long fw_device_op_compat_ioctl(struct file *file,
+                                     unsigned int cmd, unsigned long arg)
 {
        struct client *client = file->private_data;
 
@@ -947,7 +1352,7 @@ static int fw_device_op_mmap(struct file *file, struct vm_area_struct *vma)
        struct client *client = file->private_data;
        enum dma_data_direction direction;
        unsigned long size;
-       int page_count, retval;
+       int page_count, ret;
 
        if (fw_device_is_shutdown(client->device))
                return -ENODEV;
@@ -973,48 +1378,57 @@ static int fw_device_op_mmap(struct file *file, struct vm_area_struct *vma)
        else
                direction = DMA_FROM_DEVICE;
 
-       retval = fw_iso_buffer_init(&client->buffer, client->device->card,
-                                   page_count, direction);
-       if (retval < 0)
-               return retval;
+       ret = fw_iso_buffer_init(&client->buffer, client->device->card,
+                                page_count, direction);
+       if (ret < 0)
+               return ret;
 
-       retval = fw_iso_buffer_map(&client->buffer, vma);
-       if (retval < 0)
+       ret = fw_iso_buffer_map(&client->buffer, vma);
+       if (ret < 0)
                fw_iso_buffer_destroy(&client->buffer, client->device->card);
 
-       return retval;
+       return ret;
+}
+
+static int shutdown_resource(int id, void *p, void *data)
+{
+       struct client_resource *r = p;
+       struct client *client = data;
+
+       r->release(client, r);
+       client_put(client);
+
+       return 0;
 }
 
 static int fw_device_op_release(struct inode *inode, struct file *file)
 {
        struct client *client = file->private_data;
        struct event *e, *next_e;
-       struct client_resource *r, *next_r;
-       unsigned long flags;
 
-       if (client->buffer.pages)
-               fw_iso_buffer_destroy(&client->buffer, client->device->card);
+       mutex_lock(&client->device->client_list_mutex);
+       list_del(&client->link);
+       mutex_unlock(&client->device->client_list_mutex);
 
        if (client->iso_context)
                fw_iso_context_destroy(client->iso_context);
 
-       list_for_each_entry_safe(r, next_r, &client->resource_list, link)
-               r->release(client, r);
+       if (client->buffer.pages)
+               fw_iso_buffer_destroy(&client->buffer, client->device->card);
 
-       /*
-        * FIXME: We should wait for the async tasklets to stop
-        * running before freeing the memory.
-        */
+       /* Freeze client->resource_idr and client->event_list */
+       spin_lock_irq(&client->lock);
+       client->in_shutdown = true;
+       spin_unlock_irq(&client->lock);
+
+       idr_for_each(&client->resource_idr, shutdown_resource, client);
+       idr_remove_all(&client->resource_idr);
+       idr_destroy(&client->resource_idr);
 
        list_for_each_entry_safe(e, next_e, &client->event_list, link)
                kfree(e);
 
-       spin_lock_irqsave(&client->device->card->lock, flags);
-       list_del(&client->link);
-       spin_unlock_irqrestore(&client->device->card->lock, flags);
-
-       fw_device_put(client->device);
-       kfree(client);
+       client_put(client);
 
        return 0;
 }