diff options
Diffstat (limited to 'net/irda/irlmp.c')
-rw-r--r-- | net/irda/irlmp.c | 1996 |
1 files changed, 0 insertions, 1996 deletions
diff --git a/net/irda/irlmp.c b/net/irda/irlmp.c deleted file mode 100644 index 43964594aa12..000000000000 --- a/net/irda/irlmp.c +++ /dev/null @@ -1,1996 +0,0 @@ -/********************************************************************* - * - * Filename: irlmp.c - * Version: 1.0 - * Description: IrDA Link Management Protocol (LMP) layer - * Status: Stable. - * Author: Dag Brattli <dagb@cs.uit.no> - * Created at: Sun Aug 17 20:54:32 1997 - * Modified at: Wed Jan 5 11:26:03 2000 - * Modified by: Dag Brattli <dagb@cs.uit.no> - * - * Copyright (c) 1998-2000 Dag Brattli <dagb@cs.uit.no>, - * All Rights Reserved. - * Copyright (c) 2000-2003 Jean Tourrilhes <jt@hpl.hp.com> - * - * This program is free software; you can redistribute it and/or - * modify it under the terms of the GNU General Public License as - * published by the Free Software Foundation; either version 2 of - * the License, or (at your option) any later version. - * - * Neither Dag Brattli nor University of Tromsø admit liability nor - * provide warranty for any of this software. This material is - * provided "AS-IS" and at no charge. - * - ********************************************************************/ - -#include <linux/module.h> -#include <linux/slab.h> -#include <linux/string.h> -#include <linux/skbuff.h> -#include <linux/types.h> -#include <linux/proc_fs.h> -#include <linux/init.h> -#include <linux/kmod.h> -#include <linux/random.h> -#include <linux/seq_file.h> - -#include <net/irda/irda.h> -#include <net/irda/timer.h> -#include <net/irda/qos.h> -#include <net/irda/irlap.h> -#include <net/irda/iriap.h> -#include <net/irda/irlmp.h> -#include <net/irda/irlmp_frame.h> - -#include <asm/unaligned.h> - -static __u8 irlmp_find_free_slsap(void); -static int irlmp_slsap_inuse(__u8 slsap_sel); - -/* Master structure */ -struct irlmp_cb *irlmp = NULL; - -/* These can be altered by the sysctl interface */ -int sysctl_discovery = 0; -int sysctl_discovery_timeout = 3; /* 3 seconds by default */ -int sysctl_discovery_slots = 6; /* 6 slots by default */ -int sysctl_lap_keepalive_time = LM_IDLE_TIMEOUT * 1000 / HZ; -char sysctl_devname[65]; - -static const char *irlmp_reasons[] = { - "ERROR, NOT USED", - "LM_USER_REQUEST", - "LM_LAP_DISCONNECT", - "LM_CONNECT_FAILURE", - "LM_LAP_RESET", - "LM_INIT_DISCONNECT", - "ERROR, NOT USED", - "UNKNOWN", -}; - -const char *irlmp_reason_str(LM_REASON reason) -{ - reason = min_t(size_t, reason, ARRAY_SIZE(irlmp_reasons) - 1); - return irlmp_reasons[reason]; -} - -/* - * Function irlmp_init (void) - * - * Create (allocate) the main IrLMP structure - * - */ -int __init irlmp_init(void) -{ - /* Initialize the irlmp structure. */ - irlmp = kzalloc( sizeof(struct irlmp_cb), GFP_KERNEL); - if (irlmp == NULL) - return -ENOMEM; - - irlmp->magic = LMP_MAGIC; - - irlmp->clients = hashbin_new(HB_LOCK); - irlmp->services = hashbin_new(HB_LOCK); - irlmp->links = hashbin_new(HB_LOCK); - irlmp->unconnected_lsaps = hashbin_new(HB_LOCK); - irlmp->cachelog = hashbin_new(HB_NOLOCK); - - if ((irlmp->clients == NULL) || - (irlmp->services == NULL) || - (irlmp->links == NULL) || - (irlmp->unconnected_lsaps == NULL) || - (irlmp->cachelog == NULL)) { - return -ENOMEM; - } - - spin_lock_init(&irlmp->cachelog->hb_spinlock); - - irlmp->last_lsap_sel = 0x0f; /* Reserved 0x00-0x0f */ - strcpy(sysctl_devname, "Linux"); - - init_timer(&irlmp->discovery_timer); - - /* Do discovery every 3 seconds, conditionally */ - if (sysctl_discovery) - irlmp_start_discovery_timer(irlmp, - sysctl_discovery_timeout*HZ); - - return 0; -} - -/* - * Function irlmp_cleanup (void) - * - * Remove IrLMP layer - * - */ -void irlmp_cleanup(void) -{ - /* Check for main structure */ - IRDA_ASSERT(irlmp != NULL, return;); - IRDA_ASSERT(irlmp->magic == LMP_MAGIC, return;); - - del_timer(&irlmp->discovery_timer); - - hashbin_delete(irlmp->links, (FREE_FUNC) kfree); - hashbin_delete(irlmp->unconnected_lsaps, (FREE_FUNC) kfree); - hashbin_delete(irlmp->clients, (FREE_FUNC) kfree); - hashbin_delete(irlmp->services, (FREE_FUNC) kfree); - hashbin_delete(irlmp->cachelog, (FREE_FUNC) kfree); - - /* De-allocate main structure */ - kfree(irlmp); - irlmp = NULL; -} - -/* - * Function irlmp_open_lsap (slsap, notify) - * - * Register with IrLMP and create a local LSAP, - * returns handle to LSAP. - */ -struct lsap_cb *irlmp_open_lsap(__u8 slsap_sel, notify_t *notify, __u8 pid) -{ - struct lsap_cb *self; - - IRDA_ASSERT(notify != NULL, return NULL;); - IRDA_ASSERT(irlmp != NULL, return NULL;); - IRDA_ASSERT(irlmp->magic == LMP_MAGIC, return NULL;); - IRDA_ASSERT(notify->instance != NULL, return NULL;); - - /* Does the client care which Source LSAP selector it gets? */ - if (slsap_sel == LSAP_ANY) { - slsap_sel = irlmp_find_free_slsap(); - if (!slsap_sel) - return NULL; - } else if (irlmp_slsap_inuse(slsap_sel)) - return NULL; - - /* Allocate new instance of a LSAP connection */ - self = kzalloc(sizeof(struct lsap_cb), GFP_ATOMIC); - if (self == NULL) - return NULL; - - self->magic = LMP_LSAP_MAGIC; - self->slsap_sel = slsap_sel; - - /* Fix connectionless LSAP's */ - if (slsap_sel == LSAP_CONNLESS) { -#ifdef CONFIG_IRDA_ULTRA - self->dlsap_sel = LSAP_CONNLESS; - self->pid = pid; -#endif /* CONFIG_IRDA_ULTRA */ - } else - self->dlsap_sel = LSAP_ANY; - /* self->connected = FALSE; -> already NULL via memset() */ - - init_timer(&self->watchdog_timer); - - self->notify = *notify; - - self->lsap_state = LSAP_DISCONNECTED; - - /* Insert into queue of unconnected LSAPs */ - hashbin_insert(irlmp->unconnected_lsaps, (irda_queue_t *) self, - (long) self, NULL); - - return self; -} -EXPORT_SYMBOL(irlmp_open_lsap); - -/* - * Function __irlmp_close_lsap (self) - * - * Remove an instance of LSAP - */ -static void __irlmp_close_lsap(struct lsap_cb *self) -{ - IRDA_ASSERT(self != NULL, return;); - IRDA_ASSERT(self->magic == LMP_LSAP_MAGIC, return;); - - /* - * Set some of the variables to preset values - */ - self->magic = 0; - del_timer(&self->watchdog_timer); /* Important! */ - - if (self->conn_skb) - dev_kfree_skb(self->conn_skb); - - kfree(self); -} - -/* - * Function irlmp_close_lsap (self) - * - * Close and remove LSAP - * - */ -void irlmp_close_lsap(struct lsap_cb *self) -{ - struct lap_cb *lap; - struct lsap_cb *lsap = NULL; - - IRDA_ASSERT(self != NULL, return;); - IRDA_ASSERT(self->magic == LMP_LSAP_MAGIC, return;); - - /* - * Find out if we should remove this LSAP from a link or from the - * list of unconnected lsaps (not associated with a link) - */ - lap = self->lap; - if (lap) { - IRDA_ASSERT(lap->magic == LMP_LAP_MAGIC, return;); - /* We might close a LSAP before it has completed the - * connection setup. In those case, higher layers won't - * send a proper disconnect request. Harmless, except - * that we will forget to close LAP... - Jean II */ - if(self->lsap_state != LSAP_DISCONNECTED) { - self->lsap_state = LSAP_DISCONNECTED; - irlmp_do_lap_event(self->lap, - LM_LAP_DISCONNECT_REQUEST, NULL); - } - /* Now, remove from the link */ - lsap = hashbin_remove(lap->lsaps, (long) self, NULL); -#ifdef CONFIG_IRDA_CACHE_LAST_LSAP - lap->cache.valid = FALSE; -#endif - } - self->lap = NULL; - /* Check if we found the LSAP! If not then try the unconnected lsaps */ - if (!lsap) { - lsap = hashbin_remove(irlmp->unconnected_lsaps, (long) self, - NULL); - } - if (!lsap) { - pr_debug("%s(), Looks like somebody has removed me already!\n", - __func__); - return; - } - __irlmp_close_lsap(self); -} -EXPORT_SYMBOL(irlmp_close_lsap); - -/* - * Function irlmp_register_irlap (saddr, notify) - * - * Register IrLAP layer with IrLMP. There is possible to have multiple - * instances of the IrLAP layer, each connected to different IrDA ports - * - */ -void irlmp_register_link(struct irlap_cb *irlap, __u32 saddr, notify_t *notify) -{ - struct lap_cb *lap; - - IRDA_ASSERT(irlmp != NULL, return;); - IRDA_ASSERT(irlmp->magic == LMP_MAGIC, return;); - IRDA_ASSERT(notify != NULL, return;); - - /* - * Allocate new instance of a LSAP connection - */ - lap = kzalloc(sizeof(struct lap_cb), GFP_KERNEL); - if (lap == NULL) - return; - - lap->irlap = irlap; - lap->magic = LMP_LAP_MAGIC; - lap->saddr = saddr; - lap->daddr = DEV_ADDR_ANY; -#ifdef CONFIG_IRDA_CACHE_LAST_LSAP - lap->cache.valid = FALSE; -#endif - lap->lsaps = hashbin_new(HB_LOCK); - if (lap->lsaps == NULL) { - net_warn_ratelimited("%s(), unable to kmalloc lsaps\n", - __func__); - kfree(lap); - return; - } - - lap->lap_state = LAP_STANDBY; - - init_timer(&lap->idle_timer); - - /* - * Insert into queue of LMP links - */ - hashbin_insert(irlmp->links, (irda_queue_t *) lap, lap->saddr, NULL); - - /* - * We set only this variable so IrLAP can tell us on which link the - * different events happened on - */ - irda_notify_init(notify); - notify->instance = lap; -} - -/* - * Function irlmp_unregister_irlap (saddr) - * - * IrLAP layer has been removed! - * - */ -void irlmp_unregister_link(__u32 saddr) -{ - struct lap_cb *link; - - /* We must remove ourselves from the hashbin *first*. This ensure - * that no more LSAPs will be open on this link and no discovery - * will be triggered anymore. Jean II */ - link = hashbin_remove(irlmp->links, saddr, NULL); - if (link) { - IRDA_ASSERT(link->magic == LMP_LAP_MAGIC, return;); - - /* Kill all the LSAPs on this link. Jean II */ - link->reason = LAP_DISC_INDICATION; - link->daddr = DEV_ADDR_ANY; - irlmp_do_lap_event(link, LM_LAP_DISCONNECT_INDICATION, NULL); - - /* Remove all discoveries discovered at this link */ - irlmp_expire_discoveries(irlmp->cachelog, link->saddr, TRUE); - - /* Final cleanup */ - del_timer(&link->idle_timer); - link->magic = 0; - hashbin_delete(link->lsaps, (FREE_FUNC) __irlmp_close_lsap); - kfree(link); - } -} - -/* - * Function irlmp_connect_request (handle, dlsap, userdata) - * - * Connect with a peer LSAP - * - */ -int irlmp_connect_request(struct lsap_cb *self, __u8 dlsap_sel, - __u32 saddr, __u32 daddr, - struct qos_info *qos, struct sk_buff *userdata) -{ - struct sk_buff *tx_skb = userdata; - struct lap_cb *lap; - struct lsap_cb *lsap; - int ret; - - IRDA_ASSERT(self != NULL, return -EBADR;); - IRDA_ASSERT(self->magic == LMP_LSAP_MAGIC, return -EBADR;); - - pr_debug("%s(), slsap_sel=%02x, dlsap_sel=%02x, saddr=%08x, daddr=%08x\n", - __func__, self->slsap_sel, dlsap_sel, saddr, daddr); - - if (test_bit(0, &self->connected)) { - ret = -EISCONN; - goto err; - } - - /* Client must supply destination device address */ - if (!daddr) { - ret = -EINVAL; - goto err; - } - - /* Any userdata? */ - if (tx_skb == NULL) { - tx_skb = alloc_skb(LMP_MAX_HEADER, GFP_ATOMIC); - if (!tx_skb) - return -ENOMEM; - - skb_reserve(tx_skb, LMP_MAX_HEADER); - } - - /* Make room for MUX control header (3 bytes) */ - IRDA_ASSERT(skb_headroom(tx_skb) >= LMP_CONTROL_HEADER, return -1;); - skb_push(tx_skb, LMP_CONTROL_HEADER); - - self->dlsap_sel = dlsap_sel; - - /* - * Find the link to where we should try to connect since there may - * be more than one IrDA port on this machine. If the client has - * passed us the saddr (and already knows which link to use), then - * we use that to find the link, if not then we have to look in the - * discovery log and check if any of the links has discovered a - * device with the given daddr - */ - if ((!saddr) || (saddr == DEV_ADDR_ANY)) { - discovery_t *discovery; - unsigned long flags; - - spin_lock_irqsave(&irlmp->cachelog->hb_spinlock, flags); - if (daddr != DEV_ADDR_ANY) - discovery = hashbin_find(irlmp->cachelog, daddr, NULL); - else { - pr_debug("%s(), no daddr\n", __func__); - discovery = (discovery_t *) - hashbin_get_first(irlmp->cachelog); - } - - if (discovery) { - saddr = discovery->data.saddr; - daddr = discovery->data.daddr; - } - spin_unlock_irqrestore(&irlmp->cachelog->hb_spinlock, flags); - } - lap = hashbin_lock_find(irlmp->links, saddr, NULL); - if (lap == NULL) { - pr_debug("%s(), Unable to find a usable link!\n", __func__); - ret = -EHOSTUNREACH; - goto err; - } - - /* Check if LAP is disconnected or already connected */ - if (lap->daddr == DEV_ADDR_ANY) - lap->daddr = daddr; - else if (lap->daddr != daddr) { - /* Check if some LSAPs are active on this LAP */ - if (HASHBIN_GET_SIZE(lap->lsaps) == 0) { - /* No active connection, but LAP hasn't been - * disconnected yet (waiting for timeout in LAP). - * Maybe we could give LAP a bit of help in this case. - */ - pr_debug("%s(), sorry, but I'm waiting for LAP to timeout!\n", - __func__); - ret = -EAGAIN; - goto err; - } - - /* LAP is already connected to a different node, and LAP - * can only talk to one node at a time */ - pr_debug("%s(), sorry, but link is busy!\n", __func__); - ret = -EBUSY; - goto err; - } - - self->lap = lap; - - /* - * Remove LSAP from list of unconnected LSAPs and insert it into the - * list of connected LSAPs for the particular link - */ - lsap = hashbin_remove(irlmp->unconnected_lsaps, (long) self, NULL); - - IRDA_ASSERT(lsap != NULL, return -1;); - IRDA_ASSERT(lsap->magic == LMP_LSAP_MAGIC, return -1;); - IRDA_ASSERT(lsap->lap != NULL, return -1;); - IRDA_ASSERT(lsap->lap->magic == LMP_LAP_MAGIC, return -1;); - - hashbin_insert(self->lap->lsaps, (irda_queue_t *) self, (long) self, - NULL); - - set_bit(0, &self->connected); /* TRUE */ - - /* - * User supplied qos specifications? - */ - if (qos) - self->qos = *qos; - - irlmp_do_lsap_event(self, LM_CONNECT_REQUEST, tx_skb); - - /* Drop reference count - see irlap_data_request(). */ - dev_kfree_skb(tx_skb); - - return 0; - -err: - /* Cleanup */ - if(tx_skb) - dev_kfree_skb(tx_skb); - return ret; -} -EXPORT_SYMBOL(irlmp_connect_request); - -/* - * Function irlmp_connect_indication (self) - * - * Incoming connection - * - */ -void irlmp_connect_indication(struct lsap_cb *self, struct sk_buff *skb) -{ - int max_seg_size; - int lap_header_size; - int max_header_size; - - IRDA_ASSERT(self != NULL, return;); - IRDA_ASSERT(self->magic == LMP_LSAP_MAGIC, return;); - IRDA_ASSERT(skb != NULL, return;); - IRDA_ASSERT(self->lap != NULL, return;); - - pr_debug("%s(), slsap_sel=%02x, dlsap_sel=%02x\n", - __func__, self->slsap_sel, self->dlsap_sel); - - /* Note : self->lap is set in irlmp_link_data_indication(), - * (case CONNECT_CMD:) because we have no way to set it here. - * Similarly, self->dlsap_sel is usually set in irlmp_find_lsap(). - * Jean II */ - - self->qos = *self->lap->qos; - - max_seg_size = self->lap->qos->data_size.value-LMP_HEADER; - lap_header_size = IRLAP_GET_HEADER_SIZE(self->lap->irlap); - max_header_size = LMP_HEADER + lap_header_size; - - /* Hide LMP_CONTROL_HEADER header from layer above */ - skb_pull(skb, LMP_CONTROL_HEADER); - - if (self->notify.connect_indication) { - /* Don't forget to refcount it - see irlap_driver_rcv(). */ - skb_get(skb); - self->notify.connect_indication(self->notify.instance, self, - &self->qos, max_seg_size, - max_header_size, skb); - } -} - -/* - * Function irlmp_connect_response (handle, userdata) - * - * Service user is accepting connection - * - */ -int irlmp_connect_response(struct lsap_cb *self, struct sk_buff *userdata) -{ - IRDA_ASSERT(self != NULL, return -1;); - IRDA_ASSERT(self->magic == LMP_LSAP_MAGIC, return -1;); - IRDA_ASSERT(userdata != NULL, return -1;); - - /* We set the connected bit and move the lsap to the connected list - * in the state machine itself. Jean II */ - - pr_debug("%s(), slsap_sel=%02x, dlsap_sel=%02x\n", - __func__, self->slsap_sel, self->dlsap_sel); - - /* Make room for MUX control header (3 bytes) */ - IRDA_ASSERT(skb_headroom(userdata) >= LMP_CONTROL_HEADER, return -1;); - skb_push(userdata, LMP_CONTROL_HEADER); - - irlmp_do_lsap_event(self, LM_CONNECT_RESPONSE, userdata); - - /* Drop reference count - see irlap_data_request(). */ - dev_kfree_skb(userdata); - - return 0; -} -EXPORT_SYMBOL(irlmp_connect_response); - -/* - * Function irlmp_connect_confirm (handle, skb) - * - * LSAP connection confirmed peer device! - */ -void irlmp_connect_confirm(struct lsap_cb *self, struct sk_buff *skb) -{ - int max_header_size; - int lap_header_size; - int max_seg_size; - - IRDA_ASSERT(skb != NULL, return;); - IRDA_ASSERT(self != NULL, return;); - IRDA_ASSERT(self->magic == LMP_LSAP_MAGIC, return;); - IRDA_ASSERT(self->lap != NULL, return;); - - self->qos = *self->lap->qos; - - max_seg_size = self->lap->qos->data_size.value-LMP_HEADER; - lap_header_size = IRLAP_GET_HEADER_SIZE(self->lap->irlap); - max_header_size = LMP_HEADER + lap_header_size; - - pr_debug("%s(), max_header_size=%d\n", - __func__, max_header_size); - - /* Hide LMP_CONTROL_HEADER header from layer above */ - skb_pull(skb, LMP_CONTROL_HEADER); - - if (self->notify.connect_confirm) { - /* Don't forget to refcount it - see irlap_driver_rcv() */ - skb_get(skb); - self->notify.connect_confirm(self->notify.instance, self, - &self->qos, max_seg_size, - max_header_size, skb); - } -} - -/* - * Function irlmp_dup (orig, instance) - * - * Duplicate LSAP, can be used by servers to confirm a connection on a - * new LSAP so it can keep listening on the old one. - * - */ -struct lsap_cb *irlmp_dup(struct lsap_cb *orig, void *instance) -{ - struct lsap_cb *new; - unsigned long flags; - - spin_lock_irqsave(&irlmp->unconnected_lsaps->hb_spinlock, flags); - - /* Only allowed to duplicate unconnected LSAP's, and only LSAPs - * that have received a connect indication. Jean II */ - if ((!hashbin_find(irlmp->unconnected_lsaps, (long) orig, NULL)) || - (orig->lap == NULL)) { - pr_debug("%s(), invalid LSAP (wrong state)\n", - __func__); - spin_unlock_irqrestore(&irlmp->unconnected_lsaps->hb_spinlock, - flags); - return NULL; - } - - /* Allocate a new instance */ - new = kmemdup(orig, sizeof(*new), GFP_ATOMIC); - if (!new) { - pr_debug("%s(), unable to kmalloc\n", __func__); - spin_unlock_irqrestore(&irlmp->unconnected_lsaps->hb_spinlock, - flags); - return NULL; - } - /* new->lap = orig->lap; => done in the memcpy() */ - /* new->slsap_sel = orig->slsap_sel; => done in the memcpy() */ - new->conn_skb = NULL; - - spin_unlock_irqrestore(&irlmp->unconnected_lsaps->hb_spinlock, flags); - - /* Not everything is the same */ - new->notify.instance = instance; - - init_timer(&new->watchdog_timer); - - hashbin_insert(irlmp->unconnected_lsaps, (irda_queue_t *) new, - (long) new, NULL); - -#ifdef CONFIG_IRDA_CACHE_LAST_LSAP - /* Make sure that we invalidate the LSAP cache */ - new->lap->cache.valid = FALSE; -#endif /* CONFIG_IRDA_CACHE_LAST_LSAP */ - - return new; -} - -/* - * Function irlmp_disconnect_request (handle, userdata) - * - * The service user is requesting disconnection, this will not remove the - * LSAP, but only mark it as disconnected - */ -int irlmp_disconnect_request(struct lsap_cb *self, struct sk_buff *userdata) -{ - struct lsap_cb *lsap; - - IRDA_ASSERT(self != NULL, return -1;); - IRDA_ASSERT(self->magic == LMP_LSAP_MAGIC, return -1;); - IRDA_ASSERT(userdata != NULL, return -1;); - - /* Already disconnected ? - * There is a race condition between irlmp_disconnect_indication() - * and us that might mess up the hashbins below. This fixes it. - * Jean II */ - if (! test_and_clear_bit(0, &self->connected)) { - pr_debug("%s(), already disconnected!\n", __func__); - dev_kfree_skb(userdata); - return -1; - } - - skb_push(userdata, LMP_CONTROL_HEADER); - - /* - * Do the event before the other stuff since we must know - * which lap layer that the frame should be transmitted on - */ - irlmp_do_lsap_event(self, LM_DISCONNECT_REQUEST, userdata); - - /* Drop reference count - see irlap_data_request(). */ - dev_kfree_skb(userdata); - - /* - * Remove LSAP from list of connected LSAPs for the particular link - * and insert it into the list of unconnected LSAPs - */ - IRDA_ASSERT(self->lap != NULL, return -1;); - IRDA_ASSERT(self->lap->magic == LMP_LAP_MAGIC, return -1;); - IRDA_ASSERT(self->lap->lsaps != NULL, return -1;); - - lsap = hashbin_remove(self->lap->lsaps, (long) self, NULL); -#ifdef CONFIG_IRDA_CACHE_LAST_LSAP - self->lap->cache.valid = FALSE; -#endif - - IRDA_ASSERT(lsap != NULL, return -1;); - IRDA_ASSERT(lsap->magic == LMP_LSAP_MAGIC, return -1;); - IRDA_ASSERT(lsap == self, return -1;); - - hashbin_insert(irlmp->unconnected_lsaps, (irda_queue_t *) self, - (long) self, NULL); - - /* Reset some values */ - self->dlsap_sel = LSAP_ANY; - self->lap = NULL; - - return 0; -} -EXPORT_SYMBOL(irlmp_disconnect_request); - -/* - * Function irlmp_disconnect_indication (reason, userdata) - * - * LSAP is being closed! - */ -void irlmp_disconnect_indication(struct lsap_cb *self, LM_REASON reason, - struct sk_buff *skb) -{ - struct lsap_cb *lsap; - - pr_debug("%s(), reason=%s [%d]\n", __func__, - irlmp_reason_str(reason), reason); - IRDA_ASSERT(self != NULL, return;); - IRDA_ASSERT(self->magic == LMP_LSAP_MAGIC, return;); - - pr_debug("%s(), slsap_sel=%02x, dlsap_sel=%02x\n", - __func__, self->slsap_sel, self->dlsap_sel); - - /* Already disconnected ? - * There is a race condition between irlmp_disconnect_request() - * and us that might mess up the hashbins below. This fixes it. - * Jean II */ - if (! test_and_clear_bit(0, &self->connected)) { - pr_debug("%s(), already disconnected!\n", __func__); - return; - } - - /* - * Remove association between this LSAP and the link it used - */ - IRDA_ASSERT(self->lap != NULL, return;); - IRDA_ASSERT(self->lap->lsaps != NULL, return;); - - lsap = hashbin_remove(self->lap->lsaps, (long) self, NULL); -#ifdef CONFIG_IRDA_CACHE_LAST_LSAP - self->lap->cache.valid = FALSE; -#endif - - IRDA_ASSERT(lsap != NULL, return;); - IRDA_ASSERT(lsap == self, return;); - hashbin_insert(irlmp->unconnected_lsaps, (irda_queue_t *) lsap, - (long) lsap, NULL); - - self->dlsap_sel = LSAP_ANY; - self->lap = NULL; - - /* - * Inform service user - */ - if (self->notify.disconnect_indication) { - /* Don't forget to refcount it - see irlap_driver_rcv(). */ - if(skb) - skb_get(skb); - self->notify.disconnect_indication(self->notify.instance, - self, reason, skb); - } else { - pr_debug("%s(), no handler\n", __func__); - } -} - -/* - * Function irlmp_do_expiry (void) - * - * Do a cleanup of the discovery log (remove old entries) - * - * Note : separate from irlmp_do_discovery() so that we can handle - * passive discovery properly. - */ -void irlmp_do_expiry(void) -{ - struct lap_cb *lap; - - /* - * Expire discovery on all links which are *not* connected. - * On links which are connected, we can't do discovery - * anymore and can't refresh the log, so we freeze the - * discovery log to keep info about the device we are - * connected to. - * This info is mandatory if we want irlmp_connect_request() - * to work properly. - Jean II - */ - lap = (struct lap_cb *) hashbin_get_first(irlmp->links); - while (lap != NULL) { - IRDA_ASSERT(lap->magic == LMP_LAP_MAGIC, return;); - - if (lap->lap_state == LAP_STANDBY) { - /* Expire discoveries discovered on this link */ - irlmp_expire_discoveries(irlmp->cachelog, lap->saddr, - FALSE); - } - lap = (struct lap_cb *) hashbin_get_next(irlmp->links); - } -} - -/* - * Function irlmp_do_discovery (nslots) - * - * Do some discovery on all links - * - * Note : log expiry is done above. - */ -void irlmp_do_discovery(int nslots) -{ - struct lap_cb *lap; - __u16 *data_hintsp; - - /* Make sure the value is sane */ - if ((nslots != 1) && (nslots != 6) && (nslots != 8) && (nslots != 16)){ - net_warn_ratelimited("%s: invalid value for number of slots!\n", - __func__); - nslots = sysctl_discovery_slots = 8; - } - - /* Construct new discovery info to be used by IrLAP, */ - data_hintsp = (__u16 *) irlmp->discovery_cmd.data.hints; - put_unaligned(irlmp->hints.word, data_hintsp); - - /* - * Set character set for device name (we use ASCII), and - * copy device name. Remember to make room for a \0 at the - * end - */ - irlmp->discovery_cmd.data.charset = CS_ASCII; - strncpy(irlmp->discovery_cmd.data.info, sysctl_devname, - NICKNAME_MAX_LEN); - irlmp->discovery_cmd.name_len = strlen(irlmp->discovery_cmd.data.info); - irlmp->discovery_cmd.nslots = nslots; - - /* - * Try to send discovery packets on all links - */ - lap = (struct lap_cb *) hashbin_get_first(irlmp->links); - while (lap != NULL) { - IRDA_ASSERT(lap->magic == LMP_LAP_MAGIC, return;); - - if (lap->lap_state == LAP_STANDBY) { - /* Try to discover */ - irlmp_do_lap_event(lap, LM_LAP_DISCOVERY_REQUEST, - NULL); - } - lap = (struct lap_cb *) hashbin_get_next(irlmp->links); - } -} - -/* - * Function irlmp_discovery_request (nslots) - * - * Do a discovery of devices in front of the computer - * - * If the caller has registered a client discovery callback, this - * allow him to receive the full content of the discovery log through - * this callback (as normally he will receive only new discoveries). - */ -void irlmp_discovery_request(int nslots) -{ - /* Return current cached discovery log (in full) */ - irlmp_discovery_confirm(irlmp->cachelog, DISCOVERY_LOG); - - /* - * Start a single discovery operation if discovery is not already - * running - */ - if (!sysctl_discovery) { - /* Check if user wants to override the default */ - if (nslots == DISCOVERY_DEFAULT_SLOTS) - nslots = sysctl_discovery_slots; - - irlmp_do_discovery(nslots); - /* Note : we never do expiry here. Expiry will run on the - * discovery timer regardless of the state of sysctl_discovery - * Jean II */ - } -} -EXPORT_SYMBOL(irlmp_discovery_request); - -/* - * Function irlmp_get_discoveries (pn, mask, slots) - * - * Return the current discovery log - * - * If discovery is not enabled, you should call this function again - * after 1 or 2 seconds (i.e. after discovery has been done). - */ -struct irda_device_info *irlmp_get_discoveries(int *pn, __u16 mask, int nslots) -{ - /* If discovery is not enabled, it's likely that the discovery log - * will be empty. So, we trigger a single discovery, so that next - * time the user call us there might be some results in the log. - * Jean II - */ - if (!sysctl_discovery) { - /* Check if user wants to override the default */ - if (nslots == DISCOVERY_DEFAULT_SLOTS) - nslots = sysctl_discovery_slots; - - /* Start discovery - will complete sometime later */ - irlmp_do_discovery(nslots); - /* Note : we never do expiry here. Expiry will run on the - * discovery timer regardless of the state of sysctl_discovery - * Jean II */ - } - - /* Return current cached discovery log */ - return irlmp_copy_discoveries(irlmp->cachelog, pn, mask, TRUE); -} -EXPORT_SYMBOL(irlmp_get_discoveries); - -/* - * Function irlmp_notify_client (log) - * - * Notify all about discovered devices - * - * Clients registered with IrLMP are : - * o IrComm - * o IrLAN - * o Any socket (in any state - ouch, that may be a lot !) - * The client may have defined a callback to be notified in case of - * partial/selective discovery based on the hints that it passed to IrLMP. - */ -static inline void -irlmp_notify_client(irlmp_client_t *client, - hashbin_t *log, DISCOVERY_MODE mode) -{ - discinfo_t *discoveries; /* Copy of the discovery log */ - int number; /* Number of nodes in the log */ - int i; - - /* Check if client wants or not partial/selective log (optimisation) */ - if (!client->disco_callback) - return; - - /* - * Locking notes : - * the old code was manipulating the log directly, which was - * very racy. Now, we use copy_discoveries, that protects - * itself while dumping the log for us. - * The overhead of the copy is compensated by the fact that - * we only pass new discoveries in normal mode and don't - * pass the same old entry every 3s to the caller as we used - * to do (virtual function calling is expensive). - * Jean II - */ - - /* - * Now, check all discovered devices (if any), and notify client - * only about the services that the client is interested in - * We also notify only about the new devices unless the caller - * explicitly request a dump of the log. Jean II - */ - discoveries = irlmp_copy_discoveries(log, &number, - client->hint_mask.word, - (mode == DISCOVERY_LOG)); - /* Check if the we got some results */ - if (discoveries == NULL) - return; /* No nodes discovered */ - - /* Pass all entries to the listener */ - for(i = 0; i < number; i++) - client->disco_callback(&(discoveries[i]), mode, client->priv); - - /* Free up our buffer */ - kfree(discoveries); -} - -/* - * Function irlmp_discovery_confirm ( self, log) - * - * Some device(s) answered to our discovery request! Check to see which - * device it is, and give indication to the client(s) - * - */ -void irlmp_discovery_confirm(hashbin_t *log, DISCOVERY_MODE mode) -{ - irlmp_client_t *client; - irlmp_client_t *client_next; - - IRDA_ASSERT(log != NULL, return;); - - if (!(HASHBIN_GET_SIZE(log))) - return; - - /* For each client - notify callback may touch client list */ - client = (irlmp_client_t *) hashbin_get_first(irlmp->clients); - while (NULL != hashbin_find_next(irlmp->clients, (long) client, NULL, - (void *) &client_next) ) { - /* Check if we should notify client */ - irlmp_notify_client(client, log, mode); - - client = client_next; - } -} - -/* - * Function irlmp_discovery_expiry (expiry) - * - * This device is no longer been discovered, and therefore it is being - * purged from the discovery log. Inform all clients who have - * registered for this event... - * - * Note : called exclusively from discovery.c - * Note : this is no longer called under discovery spinlock, so the - * client can do whatever he wants in the callback. - */ -void irlmp_discovery_expiry(discinfo_t *expiries, int number) -{ - irlmp_client_t *client; - irlmp_client_t *client_next; - int i; - - IRDA_ASSERT(expiries != NULL, return;); - - /* For each client - notify callback may touch client list */ - client = (irlmp_client_t *) hashbin_get_first(irlmp->clients); - while (NULL != hashbin_find_next(irlmp->clients, (long) client, NULL, - (void *) &client_next) ) { - - /* Pass all entries to the listener */ - for(i = 0; i < number; i++) { - /* Check if we should notify client */ - if ((client->expir_callback) && - (client->hint_mask.word & - get_unaligned((__u16 *)expiries[i].hints) - & 0x7f7f) ) - client->expir_callback(&(expiries[i]), - EXPIRY_TIMEOUT, - client->priv); - } - - /* Next client */ - client = client_next; - } -} - -/* - * Function irlmp_get_discovery_response () - * - * Used by IrLAP to get the discovery info it needs when answering - * discovery requests by other devices. - */ -discovery_t *irlmp_get_discovery_response(void) -{ - IRDA_ASSERT(irlmp != NULL, return NULL;); - - put_unaligned(irlmp->hints.word, (__u16 *)irlmp->discovery_rsp.data.hints); - - /* - * Set character set for device name (we use ASCII), and - * copy device name. Remember to make room for a \0 at the - * end - */ - irlmp->discovery_rsp.data.charset = CS_ASCII; - - strncpy(irlmp->discovery_rsp.data.info, sysctl_devname, - NICKNAME_MAX_LEN); - irlmp->discovery_rsp.name_len = strlen(irlmp->discovery_rsp.data.info); - - return &irlmp->discovery_rsp; -} - -/* - * Function irlmp_data_request (self, skb) - * - * Send some data to peer device - * - * Note on skb management : - * After calling the lower layers of the IrDA stack, we always - * kfree() the skb, which drop the reference count (and potentially - * destroy it). - * IrLMP and IrLAP may queue the packet, and in those cases will need - * to use skb_get() to keep it around. - * Jean II - */ -int irlmp_data_request(struct lsap_cb *self, struct sk_buff *userdata) -{ - int ret; - - IRDA_ASSERT(self != NULL, return -1;); - IRDA_ASSERT(self->magic == LMP_LSAP_MAGIC, return -1;); - - /* Make room for MUX header */ - IRDA_ASSERT(skb_headroom(userdata) >= LMP_HEADER, return -1;); - skb_push(userdata, LMP_HEADER); - - ret = irlmp_do_lsap_event(self, LM_DATA_REQUEST, userdata); - - /* Drop reference count - see irlap_data_request(). */ - dev_kfree_skb(userdata); - - return ret; -} -EXPORT_SYMBOL(irlmp_data_request); - -/* - * Function irlmp_data_indication (handle, skb) - * - * Got data from LAP layer so pass it up to upper layer - * - */ -void irlmp_data_indication(struct lsap_cb *self, struct sk_buff *skb) -{ - /* Hide LMP header from layer above */ - skb_pull(skb, LMP_HEADER); - - if (self->notify.data_indication) { - /* Don't forget to refcount it - see irlap_driver_rcv(). */ - skb_get(skb); - self->notify.data_indication(self->notify.instance, self, skb); - } -} - -/* - * Function irlmp_udata_request (self, skb) - */ -int irlmp_udata_request(struct lsap_cb *self, struct sk_buff *userdata) -{ - int ret; - - IRDA_ASSERT(userdata != NULL, return -1;); - - /* Make room for MUX header */ - IRDA_ASSERT(skb_headroom(userdata) >= LMP_HEADER, return -1;); - skb_push(userdata, LMP_HEADER); - - ret = irlmp_do_lsap_event(self, LM_UDATA_REQUEST, userdata); - - /* Drop reference count - see irlap_data_request(). */ - dev_kfree_skb(userdata); - - return ret; -} - -/* - * Function irlmp_udata_indication (self, skb) - * - * Send unreliable data (but still within the connection) - * - */ -void irlmp_udata_indication(struct lsap_cb *self, struct sk_buff *skb) -{ - IRDA_ASSERT(self != NULL, return;); - IRDA_ASSERT(self->magic == LMP_LSAP_MAGIC, return;); - IRDA_ASSERT(skb != NULL, return;); - - /* Hide LMP header from layer above */ - skb_pull(skb, LMP_HEADER); - - if (self->notify.udata_indication) { - /* Don't forget to refcount it - see irlap_driver_rcv(). */ - skb_get(skb); - self->notify.udata_indication(self->notify.instance, self, - skb); - } -} - -/* - * Function irlmp_connless_data_request (self, skb) - */ -#ifdef CONFIG_IRDA_ULTRA -int irlmp_connless_data_request(struct lsap_cb *self, struct sk_buff *userdata, - __u8 pid) -{ - struct sk_buff *clone_skb; - struct lap_cb *lap; - - IRDA_ASSERT(userdata != NULL, return -1;); - - /* Make room for MUX and PID header */ - IRDA_ASSERT(skb_headroom(userdata) >= LMP_HEADER+LMP_PID_HEADER, - return -1;); - - /* Insert protocol identifier */ - skb_push(userdata, LMP_PID_HEADER); - if(self != NULL) - userdata->data[0] = self->pid; - else - userdata->data[0] = pid; - - /* Connectionless sockets must use 0x70 */ - skb_push(userdata, LMP_HEADER); - userdata->data[0] = userdata->data[1] = LSAP_CONNLESS; - - /* Try to send Connectionless packets out on all links */ - lap = (struct lap_cb *) hashbin_get_first(irlmp->links); - while (lap != NULL) { - IRDA_ASSERT(lap->magic == LMP_LAP_MAGIC, return -1;); - - clone_skb = skb_clone(userdata, GFP_ATOMIC); - if (!clone_skb) { - dev_kfree_skb(userdata); - return -ENOMEM; - } - - irlap_unitdata_request(lap->irlap, clone_skb); - /* irlap_unitdata_request() don't increase refcount, - * so no dev_kfree_skb() - Jean II */ - - lap = (struct lap_cb *) hashbin_get_next(irlmp->links); - } - dev_kfree_skb(userdata); - - return 0; -} -#endif /* CONFIG_IRDA_ULTRA */ - -/* - * Function irlmp_connless_data_indication (self, skb) - * - * Receive unreliable data outside any connection. Mostly used by Ultra - * - */ -#ifdef CONFIG_IRDA_ULTRA -void irlmp_connless_data_indication(struct lsap_cb *self, struct sk_buff *skb) -{ - IRDA_ASSERT(self != NULL, return;); - IRDA_ASSERT(self->magic == LMP_LSAP_MAGIC, return;); - IRDA_ASSERT(skb != NULL, return;); - - /* Hide LMP and PID header from layer above */ - skb_pull(skb, LMP_HEADER+LMP_PID_HEADER); - - if (self->notify.udata_indication) { - /* Don't forget to refcount it - see irlap_driver_rcv(). */ - skb_get(skb); - self->notify.udata_indication(self->notify.instance, self, - skb); - } -} -#endif /* CONFIG_IRDA_ULTRA */ - -/* - * Propagate status indication from LAP to LSAPs (via LMP) - * This don't trigger any change of state in lap_cb, lmp_cb or lsap_cb, - * and the event is stateless, therefore we can bypass both state machines - * and send the event direct to the LSAP user. - * Jean II - */ -void irlmp_status_indication(struct lap_cb *self, - LINK_STATUS link, LOCK_STATUS lock) -{ - struct lsap_cb *next; - struct lsap_cb *curr; - - /* Send status_indication to all LSAPs using this link */ - curr = (struct lsap_cb *) hashbin_get_first( self->lsaps); - while (NULL != hashbin_find_next(self->lsaps, (long) curr, NULL, - (void *) &next) ) { - IRDA_ASSERT(curr->magic == LMP_LSAP_MAGIC, return;); - /* - * Inform service user if he has requested it - */ - if (curr->notify.status_indication != NULL) - curr->notify.status_indication(curr->notify.instance, - link, lock); - else - pr_debug("%s(), no handler\n", __func__); - - curr = next; - } -} - -/* - * Receive flow control indication from LAP. - * LAP want us to send it one more frame. We implement a simple round - * robin scheduler between the active sockets so that we get a bit of - * fairness. Note that the round robin is far from perfect, but it's - * better than nothing. - * We then poll the selected socket so that we can do synchronous - * refilling of IrLAP (which allow to minimise the number of buffers). - * Jean II - */ -void irlmp_flow_indication(struct lap_cb *self, LOCAL_FLOW flow) -{ - struct lsap_cb *next; - struct lsap_cb *curr; - int lsap_todo; - - IRDA_ASSERT(self->magic == LMP_LAP_MAGIC, return;); - IRDA_ASSERT(flow == FLOW_START, return;); - - /* Get the number of lsap. That's the only safe way to know - * that we have looped around... - Jean II */ - lsap_todo = HASHBIN_GET_SIZE(self->lsaps); - pr_debug("%s() : %d lsaps to scan\n", __func__, lsap_todo); - - /* Poll lsap in order until the queue is full or until we - * tried them all. - * Most often, the current LSAP will have something to send, - * so we will go through this loop only once. - Jean II */ - while((lsap_todo--) && - (IRLAP_GET_TX_QUEUE_LEN(self->irlap) < LAP_HIGH_THRESHOLD)) { - /* Try to find the next lsap we should poll. */ - next = self->flow_next; - /* If we have no lsap, restart from first one */ - if(next == NULL) - next = (struct lsap_cb *) hashbin_get_first(self->lsaps); - /* Verify current one and find the next one */ - curr = hashbin_find_next(self->lsaps, (long) next, NULL, - (void *) &self->flow_next); - /* Uh-oh... Paranoia */ - if(curr == NULL) - break; - pr_debug("%s() : curr is %p, next was %p and is now %p, still %d to go - queue len = %d\n", - __func__, curr, next, self->flow_next, lsap_todo, - IRLAP_GET_TX_QUEUE_LEN(self->irlap)); - - /* Inform lsap user that it can send one more packet. */ - if (curr->notify.flow_indication != NULL) - curr->notify.flow_indication(curr->notify.instance, - curr, flow); - else - pr_debug("%s(), no handler\n", __func__); - } -} - -#if 0 -/* - * Function irlmp_hint_to_service (hint) - * - * Returns a list of all servics contained in the given hint bits. This - * function assumes that the hint bits have the size of two bytes only - */ -__u8 *irlmp_hint_to_service(__u8 *hint) -{ - __u8 *service; - int i = 0; - - /* - * Allocate array to store services in. 16 entries should be safe - * since we currently only support 2 hint bytes - */ - service = kmalloc(16, GFP_ATOMIC); - if (!service) - return NULL; - - if (!hint[0]) { - pr_debug("<None>\n"); - kfree(service); - return NULL; - } - if (hint[0] & HINT_PNP) - pr_debug("PnP Compatible "); - if (hint[0] & HINT_PDA) - pr_debug("PDA/Palmtop "); - if (hint[0] & HINT_COMPUTER) - pr_debug("Computer "); - if (hint[0] & HINT_PRINTER) { - pr_debug("Printer "); - service[i++] = S_PRINTER; - } - if (hint[0] & HINT_MODEM) - pr_debug("Modem "); - if (hint[0] & HINT_FAX) - pr_debug("Fax "); - if (hint[0] & HINT_LAN) { - pr_debug("LAN Access "); - service[i++] = S_LAN; - } - /* - * Test if extension byte exists. This byte will usually be - * there, but this is not really required by the standard. - * (IrLMP p. 29) - */ - if (hint[0] & HINT_EXTENSION) { - if (hint[1] & HINT_TELEPHONY) { - pr_debug("Telephony "); - service[i++] = S_TELEPHONY; - } - if (hint[1] & HINT_FILE_SERVER) - pr_debug("File Server "); - - if (hint[1] & HINT_COMM) { - pr_debug("IrCOMM "); - service[i++] = S_COMM; - } - if (hint[1] & HINT_OBEX) { - pr_debug("IrOBEX "); - service[i++] = S_OBEX; - } - } - pr_debug("\n"); - - /* So that client can be notified about any discovery */ - service[i++] = S_ANY; - - service[i] = S_END; - - return service; -} -#endif - -static const __u16 service_hint_mapping[S_END][2] = { - { HINT_PNP, 0 }, /* S_PNP */ - { HINT_PDA, 0 }, /* S_PDA */ - { HINT_COMPUTER, 0 }, /* S_COMPUTER */ - { HINT_PRINTER, 0 }, /* S_PRINTER */ - { HINT_MODEM, 0 }, /* S_MODEM */ - { HINT_FAX, 0 }, /* S_FAX */ - { HINT_LAN, 0 }, /* S_LAN */ - { HINT_EXTENSION, HINT_TELEPHONY }, /* S_TELEPHONY */ - { HINT_EXTENSION, HINT_COMM }, /* S_COMM */ - { HINT_EXTENSION, HINT_OBEX }, /* S_OBEX */ - { 0xFF, 0xFF }, /* S_ANY */ -}; - -/* - * Function irlmp_service_to_hint (service) - * - * Converts a service type, to a hint bit - * - * Returns: a 16 bit hint value, with the service bit set - */ -__u16 irlmp_service_to_hint(int service) -{ - __u16_host_order hint; - - hint.byte[0] = service_hint_mapping[service][0]; - hint.byte[1] = service_hint_mapping[service][1]; - - return hint.word; -} -EXPORT_SYMBOL(irlmp_service_to_hint); - -/* - * Function irlmp_register_service (service) - * - * Register local service with IrLMP - * - */ -void *irlmp_register_service(__u16 hints) -{ - irlmp_service_t *service; - - pr_debug("%s(), hints = %04x\n", __func__, hints); - - /* Make a new registration */ - service = kmalloc(sizeof(irlmp_service_t), GFP_ATOMIC); - if (!service) - return NULL; - - service->hints.word = hints; - hashbin_insert(irlmp->services, (irda_queue_t *) service, - (long) service, NULL); - - irlmp->hints.word |= hints; - - return (void *)service; -} -EXPORT_SYMBOL(irlmp_register_service); - -/* - * Function irlmp_unregister_service (handle) - * - * Unregister service with IrLMP. - * - * Returns: 0 on success, -1 on error - */ -int irlmp_unregister_service(void *handle) -{ - irlmp_service_t *service; - unsigned long flags; - - if (!handle) - return -1; - - /* Caller may call with invalid handle (it's legal) - Jean II */ - service = hashbin_lock_find(irlmp->services, (long) handle, NULL); - if (!service) { - pr_debug("%s(), Unknown service!\n", __func__); - return -1; - } - - hashbin_remove_this(irlmp->services, (irda_queue_t *) service); - kfree(service); - - /* Remove old hint bits */ - irlmp->hints.word = 0; - - /* Refresh current hint bits */ - spin_lock_irqsave(&irlmp->services->hb_spinlock, flags); - service = (irlmp_service_t *) hashbin_get_first(irlmp->services); - while (service) { - irlmp->hints.word |= service->hints.word; - - service = (irlmp_service_t *)hashbin_get_next(irlmp->services); - } - spin_unlock_irqrestore(&irlmp->services->hb_spinlock, flags); - return 0; -} -EXPORT_SYMBOL(irlmp_unregister_service); - -/* - * Function irlmp_register_client (hint_mask, callback1, callback2) - * - * Register a local client with IrLMP - * First callback is selective discovery (based on hints) - * Second callback is for selective discovery expiries - * - * Returns: handle > 0 on success, 0 on error - */ -void *irlmp_register_client(__u16 hint_mask, DISCOVERY_CALLBACK1 disco_clb, - DISCOVERY_CALLBACK2 expir_clb, void *priv) -{ - irlmp_client_t *client; - - IRDA_ASSERT(irlmp != NULL, return NULL;); - - /* Make a new registration */ - client = kmalloc(sizeof(irlmp_client_t), GFP_ATOMIC); - if (!client) - return NULL; - - /* Register the details */ - client->hint_mask.word = hint_mask; - client->disco_callback = disco_clb; - client->expir_callback = expir_clb; - client->priv = priv; - - hashbin_insert(irlmp->clients, (irda_queue_t *) client, - (long) client, NULL); - - return (void *) client; -} -EXPORT_SYMBOL(irlmp_register_client); - -/* - * Function irlmp_update_client (handle, hint_mask, callback1, callback2) - * - * Updates specified client (handle) with possibly new hint_mask and - * callback - * - * Returns: 0 on success, -1 on error - */ -int irlmp_update_client(void *handle, __u16 hint_mask, - DISCOVERY_CALLBACK1 disco_clb, - DISCOVERY_CALLBACK2 expir_clb, void *priv) -{ - irlmp_client_t *client; - - if (!handle) - return -1; - - client = hashbin_lock_find(irlmp->clients, (long) handle, NULL); - if (!client) { - pr_debug("%s(), Unknown client!\n", __func__); - return -1; - } - - client->hint_mask.word = hint_mask; - client->disco_callback = disco_clb; - client->expir_callback = expir_clb; - client->priv = priv; - - return 0; -} -EXPORT_SYMBOL(irlmp_update_client); - -/* - * Function irlmp_unregister_client (handle) - * - * Returns: 0 on success, -1 on error - * - */ -int irlmp_unregister_client(void *handle) -{ - struct irlmp_client *client; - - if (!handle) - return -1; - - /* Caller may call with invalid handle (it's legal) - Jean II */ - client = hashbin_lock_find(irlmp->clients, (long) handle, NULL); - if (!client) { - pr_debug("%s(), Unknown client!\n", __func__); - return -1; - } - - pr_debug("%s(), removing client!\n", __func__); - hashbin_remove_this(irlmp->clients, (irda_queue_t *) client); - kfree(client); - - return 0; -} -EXPORT_SYMBOL(irlmp_unregister_client); - -/* - * Function irlmp_slsap_inuse (slsap) - * - * Check if the given source LSAP selector is in use - * - * This function is clearly not very efficient. On the mitigating side, the - * stack make sure that in 99% of the cases, we are called only once - * for each socket allocation. We could probably keep a bitmap - * of the allocated LSAP, but I'm not sure the complexity is worth it. - * Jean II - */ -static int irlmp_slsap_inuse(__u8 slsap_sel) -{ - struct lsap_cb *self; - struct lap_cb *lap; - unsigned long flags; - - IRDA_ASSERT(irlmp != NULL, return TRUE;); - IRDA_ASSERT(irlmp->magic == LMP_MAGIC, return TRUE;); - IRDA_ASSERT(slsap_sel != LSAP_ANY, return TRUE;); - -#ifdef CONFIG_IRDA_ULTRA - /* Accept all bindings to the connectionless LSAP */ - if (slsap_sel == LSAP_CONNLESS) - return FALSE; -#endif /* CONFIG_IRDA_ULTRA */ - - /* Valid values are between 0 and 127 (0x0-0x6F) */ - if (slsap_sel > LSAP_MAX) - return TRUE; - - /* - * Check if slsap is already in use. To do this we have to loop over - * every IrLAP connection and check every LSAP associated with each - * the connection. - */ - spin_lock_irqsave_nested(&irlmp->links->hb_spinlock, flags, - SINGLE_DEPTH_NESTING); - lap = (struct lap_cb *) hashbin_get_first(irlmp->links); - while (lap != NULL) { - IRDA_ASSERT(lap->magic == LMP_LAP_MAGIC, goto errlap;); - - /* Careful for priority inversions here ! - * irlmp->links is never taken while another IrDA - * spinlock is held, so we are safe. Jean II */ - spin_lock(&lap->lsaps->hb_spinlock); - - /* For this IrLAP, check all the LSAPs */ - self = (struct lsap_cb *) hashbin_get_first(lap->lsaps); - while (self != NULL) { - IRDA_ASSERT(self->magic == LMP_LSAP_MAGIC, - goto errlsap;); - - if ((self->slsap_sel == slsap_sel)) { - pr_debug("Source LSAP selector=%02x in use\n", - self->slsap_sel); - goto errlsap; - } - self = (struct lsap_cb*) hashbin_get_next(lap->lsaps); - } - spin_unlock(&lap->lsaps->hb_spinlock); - - /* Next LAP */ - lap = (struct lap_cb *) hashbin_get_next(irlmp->links); - } - spin_unlock_irqrestore(&irlmp->links->hb_spinlock, flags); - - /* - * Server sockets are typically waiting for connections and - * therefore reside in the unconnected list. We don't want - * to give out their LSAPs for obvious reasons... - * Jean II - */ - spin_lock_irqsave(&irlmp->unconnected_lsaps->hb_spinlock, flags); - - self = (struct lsap_cb *) hashbin_get_first(irlmp->unconnected_lsaps); - while (self != NULL) { - IRDA_ASSERT(self->magic == LMP_LSAP_MAGIC, goto erruncon;); - if ((self->slsap_sel == slsap_sel)) { - pr_debug("Source LSAP selector=%02x in use (unconnected)\n", - self->slsap_sel); - goto erruncon; - } - self = (struct lsap_cb*) hashbin_get_next(irlmp->unconnected_lsaps); - } - spin_unlock_irqrestore(&irlmp->unconnected_lsaps->hb_spinlock, flags); - - return FALSE; - - /* Error exit from within one of the two nested loops. - * Make sure we release the right spinlock in the righ order. - * Jean II */ -errlsap: - spin_unlock(&lap->lsaps->hb_spinlock); -IRDA_ASSERT_LABEL(errlap:) - spin_unlock_irqrestore(&irlmp->links->hb_spinlock, flags); - return TRUE; - - /* Error exit from within the unconnected loop. - * Just one spinlock to release... Jean II */ -erruncon: - spin_unlock_irqrestore(&irlmp->unconnected_lsaps->hb_spinlock, flags); - return TRUE; -} - -/* - * Function irlmp_find_free_slsap () - * - * Find a free source LSAP to use. This function is called if the service - * user has requested a source LSAP equal to LM_ANY - */ -static __u8 irlmp_find_free_slsap(void) -{ - __u8 lsap_sel; - int wrapped = 0; - - IRDA_ASSERT(irlmp != NULL, return -1;); - IRDA_ASSERT(irlmp->magic == LMP_MAGIC, return -1;); - - /* Most users don't really care which LSAPs they are given, - * and therefore we automatically give them a free LSAP. - * This function try to find a suitable LSAP, i.e. which is - * not in use and is within the acceptable range. Jean II */ - - do { - /* Always increment to LSAP number before using it. - * In theory, we could reuse the last LSAP number, as long - * as it is no longer in use. Some IrDA stack do that. - * However, the previous socket may be half closed, i.e. - * we closed it, we think it's no longer in use, but the - * other side did not receive our close and think it's - * active and still send data on it. - * This is similar to what is done with PIDs and TCP ports. - * Also, this reduce the number of calls to irlmp_slsap_inuse() - * which is an expensive function to call. - * Jean II */ - irlmp->last_lsap_sel++; - - /* Check if we need to wraparound (0x70-0x7f are reserved) */ - if (irlmp->last_lsap_sel > LSAP_MAX) { - /* 0x00-0x10 are also reserved for well know ports */ - irlmp->last_lsap_sel = 0x10; - - /* Make sure we terminate the loop */ - if (wrapped++) { - net_err_ratelimited("%s: no more free LSAPs !\n", - __func__); - return 0; - } - } - - /* If the LSAP is in use, try the next one. - * Despite the autoincrement, we need to check if the lsap - * is really in use or not, first because LSAP may be - * directly allocated in irlmp_open_lsap(), and also because - * we may wraparound on old sockets. Jean II */ - } while (irlmp_slsap_inuse(irlmp->last_lsap_sel)); - - /* Got it ! */ - lsap_sel = irlmp->last_lsap_sel; - pr_debug("%s(), found free lsap_sel=%02x\n", - __func__, lsap_sel); - - return lsap_sel; -} - -/* - * Function irlmp_convert_lap_reason (lap_reason) - * - * Converts IrLAP disconnect reason codes to IrLMP disconnect reason - * codes - * - */ -LM_REASON irlmp_convert_lap_reason( LAP_REASON lap_reason) -{ - int reason = LM_LAP_DISCONNECT; - - switch (lap_reason) { - case LAP_DISC_INDICATION: /* Received a disconnect request from peer */ - pr_debug("%s(), LAP_DISC_INDICATION\n", __func__); - reason = LM_USER_REQUEST; - break; - case LAP_NO_RESPONSE: /* To many retransmits without response */ - pr_debug("%s(), LAP_NO_RESPONSE\n", __func__); - reason = LM_LAP_DISCONNECT; - break; - case LAP_RESET_INDICATION: - pr_debug("%s(), LAP_RESET_INDICATION\n", __func__); - reason = LM_LAP_RESET; - break; - case LAP_FOUND_NONE: - case LAP_MEDIA_BUSY: - case LAP_PRIMARY_CONFLICT: - pr_debug("%s(), LAP_FOUND_NONE, LAP_MEDIA_BUSY or LAP_PRIMARY_CONFLICT\n", - __func__); - reason = LM_CONNECT_FAILURE; - break; - default: - pr_debug("%s(), Unknown IrLAP disconnect reason %d!\n", - __func__, lap_reason); - reason = LM_LAP_DISCONNECT; - break; - } - - return reason; -} - -#ifdef CONFIG_PROC_FS - -struct irlmp_iter_state { - hashbin_t *hashbin; -}; - -#define LSAP_START_TOKEN ((void *)1) -#define LINK_START_TOKEN ((void *)2) - -static void *irlmp_seq_hb_idx(struct irlmp_iter_state *iter, loff_t *off) -{ - void *element; - - spin_lock_irq(&iter->hashbin->hb_spinlock); - for (element = hashbin_get_first(iter->hashbin); - element != NULL; - element = hashbin_get_next(iter->hashbin)) { - if (!off || (*off)-- == 0) { - /* NB: hashbin left locked */ - return element; - } - } - spin_unlock_irq(&iter->hashbin->hb_spinlock); - iter->hashbin = NULL; - return NULL; -} - - -static void *irlmp_seq_start(struct seq_file *seq, loff_t *pos) -{ - struct irlmp_iter_state *iter = seq->private; - void *v; - loff_t off = *pos; - - iter->hashbin = NULL; - if (off-- == 0) - return LSAP_START_TOKEN; - - iter->hashbin = irlmp->unconnected_lsaps; - v = irlmp_seq_hb_idx(iter, &off); - if (v) - return v; - - if (off-- == 0) - return LINK_START_TOKEN; - - iter->hashbin = irlmp->links; - return irlmp_seq_hb_idx(iter, &off); -} - -static void *irlmp_seq_next(struct seq_file *seq, void *v, loff_t *pos) -{ - struct irlmp_iter_state *iter = seq->private; - - ++*pos; - - if (v == LSAP_START_TOKEN) { /* start of list of lsaps */ - iter->hashbin = irlmp->unconnected_lsaps; - v = irlmp_seq_hb_idx(iter, NULL); - return v ? v : LINK_START_TOKEN; - } - - if (v == LINK_START_TOKEN) { /* start of list of links */ - iter->hashbin = irlmp->links; - return irlmp_seq_hb_idx(iter, NULL); - } - - v = hashbin_get_next(iter->hashbin); - - if (v == NULL) { /* no more in this hash bin */ - spin_unlock_irq(&iter->hashbin->hb_spinlock); - - if (iter->hashbin == irlmp->unconnected_lsaps) - v = LINK_START_TOKEN; - - iter->hashbin = NULL; - } - return v; -} - -static void irlmp_seq_stop(struct seq_file *seq, void *v) -{ - struct irlmp_iter_state *iter = seq->private; - - if (iter->hashbin) - spin_unlock_irq(&iter->hashbin->hb_spinlock); -} - -static int irlmp_seq_show(struct seq_file *seq, void *v) -{ - const struct irlmp_iter_state *iter = seq->private; - struct lsap_cb *self = v; - - if (v == LSAP_START_TOKEN) - seq_puts(seq, "Unconnected LSAPs:\n"); - else if (v == LINK_START_TOKEN) - seq_puts(seq, "\nRegistered Link Layers:\n"); - else if (iter->hashbin == irlmp->unconnected_lsaps) { - self = v; - IRDA_ASSERT(self->magic == LMP_LSAP_MAGIC, return -EINVAL; ); - seq_printf(seq, "lsap state: %s, ", - irlsap_state[ self->lsap_state]); - seq_printf(seq, - "slsap_sel: %#02x, dlsap_sel: %#02x, ", - self->slsap_sel, self->dlsap_sel); - seq_printf(seq, "(%s)", self->notify.name); - seq_printf(seq, "\n"); - } else if (iter->hashbin == irlmp->links) { - struct lap_cb *lap = v; - - seq_printf(seq, "lap state: %s, ", - irlmp_state[lap->lap_state]); - - seq_printf(seq, "saddr: %#08x, daddr: %#08x, ", - lap->saddr, lap->daddr); - seq_printf(seq, "num lsaps: %d", - HASHBIN_GET_SIZE(lap->lsaps)); - seq_printf(seq, "\n"); - - /* Careful for priority inversions here ! - * All other uses of attrib spinlock are independent of - * the object spinlock, so we are safe. Jean II */ - spin_lock(&lap->lsaps->hb_spinlock); - - seq_printf(seq, "\n Connected LSAPs:\n"); - for (self = (struct lsap_cb *) hashbin_get_first(lap->lsaps); - self != NULL; - self = (struct lsap_cb *)hashbin_get_next(lap->lsaps)) { - IRDA_ASSERT(self->magic == LMP_LSAP_MAGIC, - goto outloop;); - seq_printf(seq, " lsap state: %s, ", - irlsap_state[ self->lsap_state]); - seq_printf(seq, - "slsap_sel: %#02x, dlsap_sel: %#02x, ", - self->slsap_sel, self->dlsap_sel); - seq_printf(seq, "(%s)", self->notify.name); - seq_putc(seq, '\n'); - - } - IRDA_ASSERT_LABEL(outloop:) - spin_unlock(&lap->lsaps->hb_spinlock); - seq_putc(seq, '\n'); - } else - return -EINVAL; - - return 0; -} - -static const struct seq_operations irlmp_seq_ops = { - .start = irlmp_seq_start, - .next = irlmp_seq_next, - .stop = irlmp_seq_stop, - .show = irlmp_seq_show, -}; - -static int irlmp_seq_open(struct inode *inode, struct file *file) -{ - IRDA_ASSERT(irlmp != NULL, return -EINVAL;); - - return seq_open_private(file, &irlmp_seq_ops, - sizeof(struct irlmp_iter_state)); -} - -const struct file_operations irlmp_seq_fops = { - .owner = THIS_MODULE, - .open = irlmp_seq_open, - .read = seq_read, - .llseek = seq_lseek, - .release = seq_release_private, -}; - -#endif /* PROC_FS */ |