Add the rt linux 4.1.3-rt3 as base
[kvmfornfv.git] / kernel / drivers / net / ppp / ppp_async.c
diff --git a/kernel/drivers/net/ppp/ppp_async.c b/kernel/drivers/net/ppp/ppp_async.c
new file mode 100644 (file)
index 0000000..9c889e0
--- /dev/null
@@ -0,0 +1,1028 @@
+/*
+ * PPP async serial channel driver for Linux.
+ *
+ * Copyright 1999 Paul Mackerras.
+ *
+ *  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.
+ *
+ * This driver provides the encapsulation and framing for sending
+ * and receiving PPP frames over async serial lines.  It relies on
+ * the generic PPP layer to give it frames to send and to process
+ * received frames.  It implements the PPP line discipline.
+ *
+ * Part of the code in this driver was inspired by the old async-only
+ * PPP driver, written by Michael Callahan and Al Longyear, and
+ * subsequently hacked by Paul Mackerras.
+ */
+
+#include <linux/module.h>
+#include <linux/kernel.h>
+#include <linux/skbuff.h>
+#include <linux/tty.h>
+#include <linux/netdevice.h>
+#include <linux/poll.h>
+#include <linux/crc-ccitt.h>
+#include <linux/ppp_defs.h>
+#include <linux/ppp-ioctl.h>
+#include <linux/ppp_channel.h>
+#include <linux/spinlock.h>
+#include <linux/init.h>
+#include <linux/interrupt.h>
+#include <linux/jiffies.h>
+#include <linux/slab.h>
+#include <asm/unaligned.h>
+#include <asm/uaccess.h>
+#include <asm/string.h>
+
+#define PPP_VERSION    "2.4.2"
+
+#define OBUFSIZE       4096
+
+/* Structure for storing local state. */
+struct asyncppp {
+       struct tty_struct *tty;
+       unsigned int    flags;
+       unsigned int    state;
+       unsigned int    rbits;
+       int             mru;
+       spinlock_t      xmit_lock;
+       spinlock_t      recv_lock;
+       unsigned long   xmit_flags;
+       u32             xaccm[8];
+       u32             raccm;
+       unsigned int    bytes_sent;
+       unsigned int    bytes_rcvd;
+
+       struct sk_buff  *tpkt;
+       int             tpkt_pos;
+       u16             tfcs;
+       unsigned char   *optr;
+       unsigned char   *olim;
+       unsigned long   last_xmit;
+
+       struct sk_buff  *rpkt;
+       int             lcp_fcs;
+       struct sk_buff_head rqueue;
+
+       struct tasklet_struct tsk;
+
+       atomic_t        refcnt;
+       struct semaphore dead_sem;
+       struct ppp_channel chan;        /* interface to generic ppp layer */
+       unsigned char   obuf[OBUFSIZE];
+};
+
+/* Bit numbers in xmit_flags */
+#define XMIT_WAKEUP    0
+#define XMIT_FULL      1
+#define XMIT_BUSY      2
+
+/* State bits */
+#define SC_TOSS                1
+#define SC_ESCAPE      2
+#define SC_PREV_ERROR  4
+
+/* Bits in rbits */
+#define SC_RCV_BITS    (SC_RCV_B7_1|SC_RCV_B7_0|SC_RCV_ODDP|SC_RCV_EVNP)
+
+static int flag_time = HZ;
+module_param(flag_time, int, 0);
+MODULE_PARM_DESC(flag_time, "ppp_async: interval between flagged packets (in clock ticks)");
+MODULE_LICENSE("GPL");
+MODULE_ALIAS_LDISC(N_PPP);
+
+/*
+ * Prototypes.
+ */
+static int ppp_async_encode(struct asyncppp *ap);
+static int ppp_async_send(struct ppp_channel *chan, struct sk_buff *skb);
+static int ppp_async_push(struct asyncppp *ap);
+static void ppp_async_flush_output(struct asyncppp *ap);
+static void ppp_async_input(struct asyncppp *ap, const unsigned char *buf,
+                           char *flags, int count);
+static int ppp_async_ioctl(struct ppp_channel *chan, unsigned int cmd,
+                          unsigned long arg);
+static void ppp_async_process(unsigned long arg);
+
+static void async_lcp_peek(struct asyncppp *ap, unsigned char *data,
+                          int len, int inbound);
+
+static const struct ppp_channel_ops async_ops = {
+       .start_xmit = ppp_async_send,
+       .ioctl      = ppp_async_ioctl,
+};
+
+/*
+ * Routines implementing the PPP line discipline.
+ */
+
+/*
+ * We have a potential race on dereferencing tty->disc_data,
+ * because the tty layer provides no locking at all - thus one
+ * cpu could be running ppp_asynctty_receive while another
+ * calls ppp_asynctty_close, which zeroes tty->disc_data and
+ * frees the memory that ppp_asynctty_receive is using.  The best
+ * way to fix this is to use a rwlock in the tty struct, but for now
+ * we use a single global rwlock for all ttys in ppp line discipline.
+ *
+ * FIXME: this is no longer true. The _close path for the ldisc is
+ * now guaranteed to be sane.
+ */
+static DEFINE_RWLOCK(disc_data_lock);
+
+static struct asyncppp *ap_get(struct tty_struct *tty)
+{
+       struct asyncppp *ap;
+
+       read_lock(&disc_data_lock);
+       ap = tty->disc_data;
+       if (ap != NULL)
+               atomic_inc(&ap->refcnt);
+       read_unlock(&disc_data_lock);
+       return ap;
+}
+
+static void ap_put(struct asyncppp *ap)
+{
+       if (atomic_dec_and_test(&ap->refcnt))
+               up(&ap->dead_sem);
+}
+
+/*
+ * Called when a tty is put into PPP line discipline. Called in process
+ * context.
+ */
+static int
+ppp_asynctty_open(struct tty_struct *tty)
+{
+       struct asyncppp *ap;
+       int err;
+       int speed;
+
+       if (tty->ops->write == NULL)
+               return -EOPNOTSUPP;
+
+       err = -ENOMEM;
+       ap = kzalloc(sizeof(*ap), GFP_KERNEL);
+       if (!ap)
+               goto out;
+
+       /* initialize the asyncppp structure */
+       ap->tty = tty;
+       ap->mru = PPP_MRU;
+       spin_lock_init(&ap->xmit_lock);
+       spin_lock_init(&ap->recv_lock);
+       ap->xaccm[0] = ~0U;
+       ap->xaccm[3] = 0x60000000U;
+       ap->raccm = ~0U;
+       ap->optr = ap->obuf;
+       ap->olim = ap->obuf;
+       ap->lcp_fcs = -1;
+
+       skb_queue_head_init(&ap->rqueue);
+       tasklet_init(&ap->tsk, ppp_async_process, (unsigned long) ap);
+
+       atomic_set(&ap->refcnt, 1);
+       sema_init(&ap->dead_sem, 0);
+
+       ap->chan.private = ap;
+       ap->chan.ops = &async_ops;
+       ap->chan.mtu = PPP_MRU;
+       speed = tty_get_baud_rate(tty);
+       ap->chan.speed = speed;
+       err = ppp_register_channel(&ap->chan);
+       if (err)
+               goto out_free;
+
+       tty->disc_data = ap;
+       tty->receive_room = 65536;
+       return 0;
+
+ out_free:
+       kfree(ap);
+ out:
+       return err;
+}
+
+/*
+ * Called when the tty is put into another line discipline
+ * or it hangs up.  We have to wait for any cpu currently
+ * executing in any of the other ppp_asynctty_* routines to
+ * finish before we can call ppp_unregister_channel and free
+ * the asyncppp struct.  This routine must be called from
+ * process context, not interrupt or softirq context.
+ */
+static void
+ppp_asynctty_close(struct tty_struct *tty)
+{
+       struct asyncppp *ap;
+
+       write_lock_irq(&disc_data_lock);
+       ap = tty->disc_data;
+       tty->disc_data = NULL;
+       write_unlock_irq(&disc_data_lock);
+       if (!ap)
+               return;
+
+       /*
+        * We have now ensured that nobody can start using ap from now
+        * on, but we have to wait for all existing users to finish.
+        * Note that ppp_unregister_channel ensures that no calls to
+        * our channel ops (i.e. ppp_async_send/ioctl) are in progress
+        * by the time it returns.
+        */
+       if (!atomic_dec_and_test(&ap->refcnt))
+               down(&ap->dead_sem);
+       tasklet_kill(&ap->tsk);
+
+       ppp_unregister_channel(&ap->chan);
+       kfree_skb(ap->rpkt);
+       skb_queue_purge(&ap->rqueue);
+       kfree_skb(ap->tpkt);
+       kfree(ap);
+}
+
+/*
+ * Called on tty hangup in process context.
+ *
+ * Wait for I/O to driver to complete and unregister PPP channel.
+ * This is already done by the close routine, so just call that.
+ */
+static int ppp_asynctty_hangup(struct tty_struct *tty)
+{
+       ppp_asynctty_close(tty);
+       return 0;
+}
+
+/*
+ * Read does nothing - no data is ever available this way.
+ * Pppd reads and writes packets via /dev/ppp instead.
+ */
+static ssize_t
+ppp_asynctty_read(struct tty_struct *tty, struct file *file,
+                 unsigned char __user *buf, size_t count)
+{
+       return -EAGAIN;
+}
+
+/*
+ * Write on the tty does nothing, the packets all come in
+ * from the ppp generic stuff.
+ */
+static ssize_t
+ppp_asynctty_write(struct tty_struct *tty, struct file *file,
+                  const unsigned char *buf, size_t count)
+{
+       return -EAGAIN;
+}
+
+/*
+ * Called in process context only. May be re-entered by multiple
+ * ioctl calling threads.
+ */
+
+static int
+ppp_asynctty_ioctl(struct tty_struct *tty, struct file *file,
+                  unsigned int cmd, unsigned long arg)
+{
+       struct asyncppp *ap = ap_get(tty);
+       int err, val;
+       int __user *p = (int __user *)arg;
+
+       if (!ap)
+               return -ENXIO;
+       err = -EFAULT;
+       switch (cmd) {
+       case PPPIOCGCHAN:
+               err = -EFAULT;
+               if (put_user(ppp_channel_index(&ap->chan), p))
+                       break;
+               err = 0;
+               break;
+
+       case PPPIOCGUNIT:
+               err = -EFAULT;
+               if (put_user(ppp_unit_number(&ap->chan), p))
+                       break;
+               err = 0;
+               break;
+
+       case TCFLSH:
+               /* flush our buffers and the serial port's buffer */
+               if (arg == TCIOFLUSH || arg == TCOFLUSH)
+                       ppp_async_flush_output(ap);
+               err = n_tty_ioctl_helper(tty, file, cmd, arg);
+               break;
+
+       case FIONREAD:
+               val = 0;
+               if (put_user(val, p))
+                       break;
+               err = 0;
+               break;
+
+       default:
+               /* Try the various mode ioctls */
+               err = tty_mode_ioctl(tty, file, cmd, arg);
+       }
+
+       ap_put(ap);
+       return err;
+}
+
+/* No kernel lock - fine */
+static unsigned int
+ppp_asynctty_poll(struct tty_struct *tty, struct file *file, poll_table *wait)
+{
+       return 0;
+}
+
+/* May sleep, don't call from interrupt level or with interrupts disabled */
+static void
+ppp_asynctty_receive(struct tty_struct *tty, const unsigned char *buf,
+                 char *cflags, int count)
+{
+       struct asyncppp *ap = ap_get(tty);
+       unsigned long flags;
+
+       if (!ap)
+               return;
+       spin_lock_irqsave(&ap->recv_lock, flags);
+       ppp_async_input(ap, buf, cflags, count);
+       spin_unlock_irqrestore(&ap->recv_lock, flags);
+       if (!skb_queue_empty(&ap->rqueue))
+               tasklet_schedule(&ap->tsk);
+       ap_put(ap);
+       tty_unthrottle(tty);
+}
+
+static void
+ppp_asynctty_wakeup(struct tty_struct *tty)
+{
+       struct asyncppp *ap = ap_get(tty);
+
+       clear_bit(TTY_DO_WRITE_WAKEUP, &tty->flags);
+       if (!ap)
+               return;
+       set_bit(XMIT_WAKEUP, &ap->xmit_flags);
+       tasklet_schedule(&ap->tsk);
+       ap_put(ap);
+}
+
+
+static struct tty_ldisc_ops ppp_ldisc = {
+       .owner  = THIS_MODULE,
+       .magic  = TTY_LDISC_MAGIC,
+       .name   = "ppp",
+       .open   = ppp_asynctty_open,
+       .close  = ppp_asynctty_close,
+       .hangup = ppp_asynctty_hangup,
+       .read   = ppp_asynctty_read,
+       .write  = ppp_asynctty_write,
+       .ioctl  = ppp_asynctty_ioctl,
+       .poll   = ppp_asynctty_poll,
+       .receive_buf = ppp_asynctty_receive,
+       .write_wakeup = ppp_asynctty_wakeup,
+};
+
+static int __init
+ppp_async_init(void)
+{
+       int err;
+
+       err = tty_register_ldisc(N_PPP, &ppp_ldisc);
+       if (err != 0)
+               printk(KERN_ERR "PPP_async: error %d registering line disc.\n",
+                      err);
+       return err;
+}
+
+/*
+ * The following routines provide the PPP channel interface.
+ */
+static int
+ppp_async_ioctl(struct ppp_channel *chan, unsigned int cmd, unsigned long arg)
+{
+       struct asyncppp *ap = chan->private;
+       void __user *argp = (void __user *)arg;
+       int __user *p = argp;
+       int err, val;
+       u32 accm[8];
+
+       err = -EFAULT;
+       switch (cmd) {
+       case PPPIOCGFLAGS:
+               val = ap->flags | ap->rbits;
+               if (put_user(val, p))
+                       break;
+               err = 0;
+               break;
+       case PPPIOCSFLAGS:
+               if (get_user(val, p))
+                       break;
+               ap->flags = val & ~SC_RCV_BITS;
+               spin_lock_irq(&ap->recv_lock);
+               ap->rbits = val & SC_RCV_BITS;
+               spin_unlock_irq(&ap->recv_lock);
+               err = 0;
+               break;
+
+       case PPPIOCGASYNCMAP:
+               if (put_user(ap->xaccm[0], (u32 __user *)argp))
+                       break;
+               err = 0;
+               break;
+       case PPPIOCSASYNCMAP:
+               if (get_user(ap->xaccm[0], (u32 __user *)argp))
+                       break;
+               err = 0;
+               break;
+
+       case PPPIOCGRASYNCMAP:
+               if (put_user(ap->raccm, (u32 __user *)argp))
+                       break;
+               err = 0;
+               break;
+       case PPPIOCSRASYNCMAP:
+               if (get_user(ap->raccm, (u32 __user *)argp))
+                       break;
+               err = 0;
+               break;
+
+       case PPPIOCGXASYNCMAP:
+               if (copy_to_user(argp, ap->xaccm, sizeof(ap->xaccm)))
+                       break;
+               err = 0;
+               break;
+       case PPPIOCSXASYNCMAP:
+               if (copy_from_user(accm, argp, sizeof(accm)))
+                       break;
+               accm[2] &= ~0x40000000U;        /* can't escape 0x5e */
+               accm[3] |= 0x60000000U;         /* must escape 0x7d, 0x7e */
+               memcpy(ap->xaccm, accm, sizeof(ap->xaccm));
+               err = 0;
+               break;
+
+       case PPPIOCGMRU:
+               if (put_user(ap->mru, p))
+                       break;
+               err = 0;
+               break;
+       case PPPIOCSMRU:
+               if (get_user(val, p))
+                       break;
+               if (val < PPP_MRU)
+                       val = PPP_MRU;
+               ap->mru = val;
+               err = 0;
+               break;
+
+       default:
+               err = -ENOTTY;
+       }
+
+       return err;
+}
+
+/*
+ * This is called at softirq level to deliver received packets
+ * to the ppp_generic code, and to tell the ppp_generic code
+ * if we can accept more output now.
+ */
+static void ppp_async_process(unsigned long arg)
+{
+       struct asyncppp *ap = (struct asyncppp *) arg;
+       struct sk_buff *skb;
+
+       /* process received packets */
+       while ((skb = skb_dequeue(&ap->rqueue)) != NULL) {
+               if (skb->cb[0])
+                       ppp_input_error(&ap->chan, 0);
+               ppp_input(&ap->chan, skb);
+       }
+
+       /* try to push more stuff out */
+       if (test_bit(XMIT_WAKEUP, &ap->xmit_flags) && ppp_async_push(ap))
+               ppp_output_wakeup(&ap->chan);
+}
+
+/*
+ * Procedures for encapsulation and framing.
+ */
+
+/*
+ * Procedure to encode the data for async serial transmission.
+ * Does octet stuffing (escaping), puts the address/control bytes
+ * on if A/C compression is disabled, and does protocol compression.
+ * Assumes ap->tpkt != 0 on entry.
+ * Returns 1 if we finished the current frame, 0 otherwise.
+ */
+
+#define PUT_BYTE(ap, buf, c, islcp)    do {            \
+       if ((islcp && c < 0x20) || (ap->xaccm[c >> 5] & (1 << (c & 0x1f)))) {\
+               *buf++ = PPP_ESCAPE;                    \
+               *buf++ = c ^ PPP_TRANS;                 \
+       } else                                          \
+               *buf++ = c;                             \
+} while (0)
+
+static int
+ppp_async_encode(struct asyncppp *ap)
+{
+       int fcs, i, count, c, proto;
+       unsigned char *buf, *buflim;
+       unsigned char *data;
+       int islcp;
+
+       buf = ap->obuf;
+       ap->olim = buf;
+       ap->optr = buf;
+       i = ap->tpkt_pos;
+       data = ap->tpkt->data;
+       count = ap->tpkt->len;
+       fcs = ap->tfcs;
+       proto = get_unaligned_be16(data);
+
+       /*
+        * LCP packets with code values between 1 (configure-reqest)
+        * and 7 (code-reject) must be sent as though no options
+        * had been negotiated.
+        */
+       islcp = proto == PPP_LCP && 1 <= data[2] && data[2] <= 7;
+
+       if (i == 0) {
+               if (islcp)
+                       async_lcp_peek(ap, data, count, 0);
+
+               /*
+                * Start of a new packet - insert the leading FLAG
+                * character if necessary.
+                */
+               if (islcp || flag_time == 0 ||
+                   time_after_eq(jiffies, ap->last_xmit + flag_time))
+                       *buf++ = PPP_FLAG;
+               ap->last_xmit = jiffies;
+               fcs = PPP_INITFCS;
+
+               /*
+                * Put in the address/control bytes if necessary
+                */
+               if ((ap->flags & SC_COMP_AC) == 0 || islcp) {
+                       PUT_BYTE(ap, buf, 0xff, islcp);
+                       fcs = PPP_FCS(fcs, 0xff);
+                       PUT_BYTE(ap, buf, 0x03, islcp);
+                       fcs = PPP_FCS(fcs, 0x03);
+               }
+       }
+
+       /*
+        * Once we put in the last byte, we need to put in the FCS
+        * and closing flag, so make sure there is at least 7 bytes
+        * of free space in the output buffer.
+        */
+       buflim = ap->obuf + OBUFSIZE - 6;
+       while (i < count && buf < buflim) {
+               c = data[i++];
+               if (i == 1 && c == 0 && (ap->flags & SC_COMP_PROT))
+                       continue;       /* compress protocol field */
+               fcs = PPP_FCS(fcs, c);
+               PUT_BYTE(ap, buf, c, islcp);
+       }
+
+       if (i < count) {
+               /*
+                * Remember where we are up to in this packet.
+                */
+               ap->olim = buf;
+               ap->tpkt_pos = i;
+               ap->tfcs = fcs;
+               return 0;
+       }
+
+       /*
+        * We have finished the packet.  Add the FCS and flag.
+        */
+       fcs = ~fcs;
+       c = fcs & 0xff;
+       PUT_BYTE(ap, buf, c, islcp);
+       c = (fcs >> 8) & 0xff;
+       PUT_BYTE(ap, buf, c, islcp);
+       *buf++ = PPP_FLAG;
+       ap->olim = buf;
+
+       consume_skb(ap->tpkt);
+       ap->tpkt = NULL;
+       return 1;
+}
+
+/*
+ * Transmit-side routines.
+ */
+
+/*
+ * Send a packet to the peer over an async tty line.
+ * Returns 1 iff the packet was accepted.
+ * If the packet was not accepted, we will call ppp_output_wakeup
+ * at some later time.
+ */
+static int
+ppp_async_send(struct ppp_channel *chan, struct sk_buff *skb)
+{
+       struct asyncppp *ap = chan->private;
+
+       ppp_async_push(ap);
+
+       if (test_and_set_bit(XMIT_FULL, &ap->xmit_flags))
+               return 0;       /* already full */
+       ap->tpkt = skb;
+       ap->tpkt_pos = 0;
+
+       ppp_async_push(ap);
+       return 1;
+}
+
+/*
+ * Push as much data as possible out to the tty.
+ */
+static int
+ppp_async_push(struct asyncppp *ap)
+{
+       int avail, sent, done = 0;
+       struct tty_struct *tty = ap->tty;
+       int tty_stuffed = 0;
+
+       /*
+        * We can get called recursively here if the tty write
+        * function calls our wakeup function.  This can happen
+        * for example on a pty with both the master and slave
+        * set to PPP line discipline.
+        * We use the XMIT_BUSY bit to detect this and get out,
+        * leaving the XMIT_WAKEUP bit set to tell the other
+        * instance that it may now be able to write more now.
+        */
+       if (test_and_set_bit(XMIT_BUSY, &ap->xmit_flags))
+               return 0;
+       spin_lock_bh(&ap->xmit_lock);
+       for (;;) {
+               if (test_and_clear_bit(XMIT_WAKEUP, &ap->xmit_flags))
+                       tty_stuffed = 0;
+               if (!tty_stuffed && ap->optr < ap->olim) {
+                       avail = ap->olim - ap->optr;
+                       set_bit(TTY_DO_WRITE_WAKEUP, &tty->flags);
+                       sent = tty->ops->write(tty, ap->optr, avail);
+                       if (sent < 0)
+                               goto flush;     /* error, e.g. loss of CD */
+                       ap->optr += sent;
+                       if (sent < avail)
+                               tty_stuffed = 1;
+                       continue;
+               }
+               if (ap->optr >= ap->olim && ap->tpkt) {
+                       if (ppp_async_encode(ap)) {
+                               /* finished processing ap->tpkt */
+                               clear_bit(XMIT_FULL, &ap->xmit_flags);
+                               done = 1;
+                       }
+                       continue;
+               }
+               /*
+                * We haven't made any progress this time around.
+                * Clear XMIT_BUSY to let other callers in, but
+                * after doing so we have to check if anyone set
+                * XMIT_WAKEUP since we last checked it.  If they
+                * did, we should try again to set XMIT_BUSY and go
+                * around again in case XMIT_BUSY was still set when
+                * the other caller tried.
+                */
+               clear_bit(XMIT_BUSY, &ap->xmit_flags);
+               /* any more work to do? if not, exit the loop */
+               if (!(test_bit(XMIT_WAKEUP, &ap->xmit_flags) ||
+                     (!tty_stuffed && ap->tpkt)))
+                       break;
+               /* more work to do, see if we can do it now */
+               if (test_and_set_bit(XMIT_BUSY, &ap->xmit_flags))
+                       break;
+       }
+       spin_unlock_bh(&ap->xmit_lock);
+       return done;
+
+flush:
+       clear_bit(XMIT_BUSY, &ap->xmit_flags);
+       if (ap->tpkt) {
+               kfree_skb(ap->tpkt);
+               ap->tpkt = NULL;
+               clear_bit(XMIT_FULL, &ap->xmit_flags);
+               done = 1;
+       }
+       ap->optr = ap->olim;
+       spin_unlock_bh(&ap->xmit_lock);
+       return done;
+}
+
+/*
+ * Flush output from our internal buffers.
+ * Called for the TCFLSH ioctl. Can be entered in parallel
+ * but this is covered by the xmit_lock.
+ */
+static void
+ppp_async_flush_output(struct asyncppp *ap)
+{
+       int done = 0;
+
+       spin_lock_bh(&ap->xmit_lock);
+       ap->optr = ap->olim;
+       if (ap->tpkt != NULL) {
+               kfree_skb(ap->tpkt);
+               ap->tpkt = NULL;
+               clear_bit(XMIT_FULL, &ap->xmit_flags);
+               done = 1;
+       }
+       spin_unlock_bh(&ap->xmit_lock);
+       if (done)
+               ppp_output_wakeup(&ap->chan);
+}
+
+/*
+ * Receive-side routines.
+ */
+
+/* see how many ordinary chars there are at the start of buf */
+static inline int
+scan_ordinary(struct asyncppp *ap, const unsigned char *buf, int count)
+{
+       int i, c;
+
+       for (i = 0; i < count; ++i) {
+               c = buf[i];
+               if (c == PPP_ESCAPE || c == PPP_FLAG ||
+                   (c < 0x20 && (ap->raccm & (1 << c)) != 0))
+                       break;
+       }
+       return i;
+}
+
+/* called when a flag is seen - do end-of-packet processing */
+static void
+process_input_packet(struct asyncppp *ap)
+{
+       struct sk_buff *skb;
+       unsigned char *p;
+       unsigned int len, fcs, proto;
+
+       skb = ap->rpkt;
+       if (ap->state & (SC_TOSS | SC_ESCAPE))
+               goto err;
+
+       if (skb == NULL)
+               return;         /* 0-length packet */
+
+       /* check the FCS */
+       p = skb->data;
+       len = skb->len;
+       if (len < 3)
+               goto err;       /* too short */
+       fcs = PPP_INITFCS;
+       for (; len > 0; --len)
+               fcs = PPP_FCS(fcs, *p++);
+       if (fcs != PPP_GOODFCS)
+               goto err;       /* bad FCS */
+       skb_trim(skb, skb->len - 2);
+
+       /* check for address/control and protocol compression */
+       p = skb->data;
+       if (p[0] == PPP_ALLSTATIONS) {
+               /* chop off address/control */
+               if (p[1] != PPP_UI || skb->len < 3)
+                       goto err;
+               p = skb_pull(skb, 2);
+       }
+       proto = p[0];
+       if (proto & 1) {
+               /* protocol is compressed */
+               skb_push(skb, 1)[0] = 0;
+       } else {
+               if (skb->len < 2)
+                       goto err;
+               proto = (proto << 8) + p[1];
+               if (proto == PPP_LCP)
+                       async_lcp_peek(ap, p, skb->len, 1);
+       }
+
+       /* queue the frame to be processed */
+       skb->cb[0] = ap->state;
+       skb_queue_tail(&ap->rqueue, skb);
+       ap->rpkt = NULL;
+       ap->state = 0;
+       return;
+
+ err:
+       /* frame had an error, remember that, reset SC_TOSS & SC_ESCAPE */
+       ap->state = SC_PREV_ERROR;
+       if (skb) {
+               /* make skb appear as freshly allocated */
+               skb_trim(skb, 0);
+               skb_reserve(skb, - skb_headroom(skb));
+       }
+}
+
+/* Called when the tty driver has data for us. Runs parallel with the
+   other ldisc functions but will not be re-entered */
+
+static void
+ppp_async_input(struct asyncppp *ap, const unsigned char *buf,
+               char *flags, int count)
+{
+       struct sk_buff *skb;
+       int c, i, j, n, s, f;
+       unsigned char *sp;
+
+       /* update bits used for 8-bit cleanness detection */
+       if (~ap->rbits & SC_RCV_BITS) {
+               s = 0;
+               for (i = 0; i < count; ++i) {
+                       c = buf[i];
+                       if (flags && flags[i] != 0)
+                               continue;
+                       s |= (c & 0x80)? SC_RCV_B7_1: SC_RCV_B7_0;
+                       c = ((c >> 4) ^ c) & 0xf;
+                       s |= (0x6996 & (1 << c))? SC_RCV_ODDP: SC_RCV_EVNP;
+               }
+               ap->rbits |= s;
+       }
+
+       while (count > 0) {
+               /* scan through and see how many chars we can do in bulk */
+               if ((ap->state & SC_ESCAPE) && buf[0] == PPP_ESCAPE)
+                       n = 1;
+               else
+                       n = scan_ordinary(ap, buf, count);
+
+               f = 0;
+               if (flags && (ap->state & SC_TOSS) == 0) {
+                       /* check the flags to see if any char had an error */
+                       for (j = 0; j < n; ++j)
+                               if ((f = flags[j]) != 0)
+                                       break;
+               }
+               if (f != 0) {
+                       /* start tossing */
+                       ap->state |= SC_TOSS;
+
+               } else if (n > 0 && (ap->state & SC_TOSS) == 0) {
+                       /* stuff the chars in the skb */
+                       skb = ap->rpkt;
+                       if (!skb) {
+                               skb = dev_alloc_skb(ap->mru + PPP_HDRLEN + 2);
+                               if (!skb)
+                                       goto nomem;
+                               ap->rpkt = skb;
+                       }
+                       if (skb->len == 0) {
+                               /* Try to get the payload 4-byte aligned.
+                                * This should match the
+                                * PPP_ALLSTATIONS/PPP_UI/compressed tests in
+                                * process_input_packet, but we do not have
+                                * enough chars here to test buf[1] and buf[2].
+                                */
+                               if (buf[0] != PPP_ALLSTATIONS)
+                                       skb_reserve(skb, 2 + (buf[0] & 1));
+                       }
+                       if (n > skb_tailroom(skb)) {
+                               /* packet overflowed MRU */
+                               ap->state |= SC_TOSS;
+                       } else {
+                               sp = skb_put(skb, n);
+                               memcpy(sp, buf, n);
+                               if (ap->state & SC_ESCAPE) {
+                                       sp[0] ^= PPP_TRANS;
+                                       ap->state &= ~SC_ESCAPE;
+                               }
+                       }
+               }
+
+               if (n >= count)
+                       break;
+
+               c = buf[n];
+               if (flags != NULL && flags[n] != 0) {
+                       ap->state |= SC_TOSS;
+               } else if (c == PPP_FLAG) {
+                       process_input_packet(ap);
+               } else if (c == PPP_ESCAPE) {
+                       ap->state |= SC_ESCAPE;
+               } else if (I_IXON(ap->tty)) {
+                       if (c == START_CHAR(ap->tty))
+                               start_tty(ap->tty);
+                       else if (c == STOP_CHAR(ap->tty))
+                               stop_tty(ap->tty);
+               }
+               /* otherwise it's a char in the recv ACCM */
+               ++n;
+
+               buf += n;
+               if (flags)
+                       flags += n;
+               count -= n;
+       }
+       return;
+
+ nomem:
+       printk(KERN_ERR "PPPasync: no memory (input pkt)\n");
+       ap->state |= SC_TOSS;
+}
+
+/*
+ * We look at LCP frames going past so that we can notice
+ * and react to the LCP configure-ack from the peer.
+ * In the situation where the peer has been sent a configure-ack
+ * already, LCP is up once it has sent its configure-ack
+ * so the immediately following packet can be sent with the
+ * configured LCP options.  This allows us to process the following
+ * packet correctly without pppd needing to respond quickly.
+ *
+ * We only respond to the received configure-ack if we have just
+ * sent a configure-request, and the configure-ack contains the
+ * same data (this is checked using a 16-bit crc of the data).
+ */
+#define CONFREQ                1       /* LCP code field values */
+#define CONFACK                2
+#define LCP_MRU                1       /* LCP option numbers */
+#define LCP_ASYNCMAP   2
+
+static void async_lcp_peek(struct asyncppp *ap, unsigned char *data,
+                          int len, int inbound)
+{
+       int dlen, fcs, i, code;
+       u32 val;
+
+       data += 2;              /* skip protocol bytes */
+       len -= 2;
+       if (len < 4)            /* 4 = code, ID, length */
+               return;
+       code = data[0];
+       if (code != CONFACK && code != CONFREQ)
+               return;
+       dlen = get_unaligned_be16(data + 2);
+       if (len < dlen)
+               return;         /* packet got truncated or length is bogus */
+
+       if (code == (inbound? CONFACK: CONFREQ)) {
+               /*
+                * sent confreq or received confack:
+                * calculate the crc of the data from the ID field on.
+                */
+               fcs = PPP_INITFCS;
+               for (i = 1; i < dlen; ++i)
+                       fcs = PPP_FCS(fcs, data[i]);
+
+               if (!inbound) {
+                       /* outbound confreq - remember the crc for later */
+                       ap->lcp_fcs = fcs;
+                       return;
+               }
+
+               /* received confack, check the crc */
+               fcs ^= ap->lcp_fcs;
+               ap->lcp_fcs = -1;
+               if (fcs != 0)
+                       return;
+       } else if (inbound)
+               return; /* not interested in received confreq */
+
+       /* process the options in the confack */
+       data += 4;
+       dlen -= 4;
+       /* data[0] is code, data[1] is length */
+       while (dlen >= 2 && dlen >= data[1] && data[1] >= 2) {
+               switch (data[0]) {
+               case LCP_MRU:
+                       val = get_unaligned_be16(data + 2);
+                       if (inbound)
+                               ap->mru = val;
+                       else
+                               ap->chan.mtu = val;
+                       break;
+               case LCP_ASYNCMAP:
+                       val = get_unaligned_be32(data + 2);
+                       if (inbound)
+                               ap->raccm = val;
+                       else
+                               ap->xaccm[0] = val;
+                       break;
+               }
+               dlen -= data[1];
+               data += data[1];
+       }
+}
+
+static void __exit ppp_async_cleanup(void)
+{
+       if (tty_unregister_ldisc(N_PPP) != 0)
+               printk(KERN_ERR "failed to unregister PPP line discipline\n");
+}
+
+module_init(ppp_async_init);
+module_exit(ppp_async_cleanup);