Wed Jun  5 18:52:04 1996  Theodore Ts'o  <tytso@rsts-11.mit.edu>

	* tty_io.c (do_tty_hangup): 
	* pty.c (pty_close): When closing a pty, make sure packet mode is
	 	cleared.

Sun May 26 09:33:52 1996  Theodore Ts'o  <tytso@rsts-11.mit.edu>

	* vesa_blank.c (set_vesa_blanking): Add missing verify_area() call.

	* selection.c (set_selection): Add missing verify_area() call.

	* tty_io.c (tty_ioctl): Add missing verify_area() calls.

	* serial.c (rs_ioctl): Add missing verify_area() calls.
		(rs_init): Allow initialization of serial driver
		configuration from a module.

	* random.c (extract_entropy): Add missing verify_area call.
		Don't limit number of characters returned to
		32,768. Extract entropy is now no longer a inlined
		function.

		(random_read): Check return value in case extract_entropy
		returns an error.

		(secure_tcp_sequence_number): New function which returns a
		secure TCP sequence number.  This is needed to prevent some
		nasty TCP hijacking attacks.
	
		(init_std_data): Initialize using gettimeofday() instead of
		struct timeval xtime.

		(fast_add_entropy_word, add_entropy_word): Rename the
		inline function add_entropy_word() to
		fast_add_entropy_word().  Make add_entropy_word() be the
		non-inlined function which is used in non-timing critical
		places, in order to save space.

		(initialize_benchmark, begin_benchmark, end_benchmark): New
		functions defined when RANDOM_BENCHMARK is defined.  They
		allow us to benchmark the speed of the
		add_timer_randomness() call.

		(int_ln, rotate_left): Add two new inline functions with
		i386 optimized asm instructions.  This speeds up the
		critical add_entropy_word() and add_timer_randomness()
		functions, which are called from interrupt handlers.

Tue May  7 22:51:11 1996    <tytso@rsts-11.mit.edu>

	* random.c (add_timer_randomness): Limit the amount randomness
		that we estimate to 12 bits.  (An arbitrary amount).

		(extract_entropy): To make it harder to analyze the hash
		function, fold the hash function in half using XOR, and
		use the folded result as the value to emit to the user.
		Also, add timer randomness each pass through the
		exact_entropy call, to increase the amount of unknown
		values during the extraction process.

		(random_ioctl): Use IOR/IOW definitions to define the
		ioctl values used by the /dev/random driver.  Allow the
		old ioctl values to be used for backwards compatibility
		(for a limited amount of time).

Wed Apr 24 14:02:04 1996  Theodore Ts'o  <tytso@rsts-11.mit.edu>

	* random.c (add_timer_randomness): Use 2nd derivative as well to
		better estimate entropy.

		(rand_initialize): Explicitly initialize all the pointers
		to NULL.  (Clearing pointers using memset isn't portable.)
		Initialize the random pool with OS-dependent data.

		(random_write): Add sanity checking to the arguments to
		random_write(), so that bad arguments won't cause a kernel
		SEGV. 

		(random_read): Update the access time of the device inode
		when you return data to the user.

		(random_ioctl): Wake up the random_wait channel when there
		are only WAIT_INPUT_BITS available.  Add more paranoia
		checks to make sure entropy_count doesn't go beyond the
		bounds of (0, POOLSIZE).  Add a few missing verify_area
		checks.  Add support for the RNDCLEARPOOL ioctl, which
		zaps the random pool.

		(add_timer_randomness): Wake up the random_wait
		channel only when there are WAIT_INPUT_BITS available.

		(random_select): Allow a random refresh daemon process to
		select on /dev/random for writing; wake up the daemon when
		there are less than WAIT_OUTPUT_BITS bits of randomness
		available.

Tue Apr 23 22:56:07 1996    <tytso@rsts-11.mit.edu>

	* tty_io.c (init_dev): Change return code when user attempts to
		open master pty which is already open from EAGAIN to EIO,
		to match with BSD expectations.  EIO is more correct
		anyway, since EAGAIN implies that retrying will be
		successful --- which it might be.... Eventually!!

	* pty.c (pty_open, pty_close): Fix wait loop so that we don't
		busy loop while waiting for the master side to open.
		Fix tty opening/closing logic.  TTY_SLAVE_CLOSED was
		renamed to TTY_OTHER_CLOSED, so that the name is more
		descriptive.  Also fixed code so that the tty flag
		actually works correctly now....

Mon Apr  1 10:22:01 1996    <tytso@rsts-11.mit.edu>

	* serial.c (rs_close): Cleaned up modularization changes.
		Remove code which forced line discipline back to N_TTY
		this is done in the tty upper layers, and there's no
		reason to do it here.  (Making this change also
		removed the requirement that the serial module access
		the internal kernel symbol "ldiscs".)

	* tty_io.c (tty_init): Formally register a tty_driver entry for
		/dev/tty (device 4, 0) and /dev/console (device 5, 0).
		This guarantees that major device numbers 4 and 5 will be
		reserved for the tty subsystem (as they have to be because
		of /dev/tty and /dev/console).  Removed tty_regdev, as
		this interface is no longer necessary.

Sun Mar 17 20:42:47 GMT 1996 <ah@doc.ic.ac.uk>

	* serial.c : modularisation (changes in linux/fs/device.c allow
		kerneld to automatically load the serial module).

	* Makefile, Config.in : serial modularisation adds.

	* tty_io.c : tty_init_ctty used by to register "cua" driver just
		for the /dev/tty device (5,0).  Added tty_regdev.
	
	* serial.c (shutdown, rs_ioctl) : when port shuts down wakeup processes
	  waiting on delta_msr_wait. The TIOCMIWAIT ioctl returns EIO
	  if no change was done since the time of call.

Sat Mar 16 14:33:13 1996 <aeb@cwi.nl>

	* tty_io.c (disassociate_ctty): If disassociate_ctty is called by
		exit, do not perform an implicit vhangup on a pty.

Fri Feb  9 14:15:47 1996    <tytso@rsts-11.mit.edu>

	* serial.c (block_til_ready): Fixed another race condition which
		happens if a hangup happens during the open.

Wed Jan 10 10:08:00 1996    <tytso@rsts-11.mit.edu>

	* serial.c (block_til_ready): Remove race condition which happened
		if a hangup condition happened during the setup of the
		UART, before rs_open() called block_til_ready().  This
		caused the info->count counter to be erroneously
		decremented.

	* serial.c (startup, rs_open): Remove race condition that could
		cause a memory leak of one page.  (Fortunately, both race
		conditions were relatively rare in practice.)

Tue Dec  5 13:21:27 1995    <tytso@rsts-11.mit.edu>

	* serial.c (check_modem_status, rs_ioctl): Support the new
		ioctl()'s TIOCGICOUNT, TIOCMIWAIT.  These allow an
		application program to wait on a modem serial register
		status bit change, and to find out how many changes have
		taken place for the MSR bits.

		(rs_write): Eliminate a race condition which is introduced
		if it is necessary to wait for the semaphore.

Sat Nov  4 17:14:45 1995    <tytso@rsts-11.mit.edu>

	* tty_io.c (tty_init): Move registration of TTY_MAJOR and
		TTY_AUX_MAJOR to the end, so that /proc/devices looks
		prettier. 

	* pty.c (pty_init): Use new major numbers for PTY master and slave
		devices.  This allow us to have more than 64 pty's.  We
		register the old pty devices for backwards compatibility.
		Note that a system should either be using the old pty
		devices or the new pty devices --- in general, it should
		try to use both, since they map into the same pty table.
		The old pty devices are strictly for backwards compatibility.

Wed Oct 11 12:45:24 1995    <tytso@rsts-11.mit.edu>

	* tty_io.c (disassociate_ctty): If disassociate_ctty is called by
		exit, perform an implicit vhangup on the tty.

	* pty.c (pty_close): When the master pty is closed, send a hangup
		to the slave pty.
		(pty_open): Use the flag TTY_SLAVE_CLOSED to test to see
		if there are any open slave ptys, instead of using
		tty->link->count.  The old method got confused if there
		were processes that had hung-up file descriptors on the
		slave tty.

Tue May  2 00:53:25 1995    <tytso@rsx-11.mit.edu>

	* tty_io.c (tty_set_ldisc): Wait until the output buffer is
		drained before closing the old line discipline --- needed
		in only one case: XON/XOFF processing.

	* n_tty.c (n_tty_close): Don't bother waiting until the output
		driver is closed; in general, the line discipline
		shouldn't care if the hardware is finished
		transmitting before the line discipline terminates.

	* tty_io.c (release_dev): Shutdown the line discipline after
		decrementing the tty count variable; but set the
		TTY_CLOSING flag so that we know that this tty structure
		isn't long for this world.

	* tty_io.c (init_dev): Add sanity code to check to see if
		TTY_CLOSING is set on a tty structure; if so, something
		bad has happened (probably a line discipline close blocked
		when it shouldn't have; so do a kernel printk and then
		return an error).

Wed Apr 26 10:23:44 1995  Theodore Y. Ts'o  <tytso@localhost>

	* tty_io.c (release_dev): Try to shutdown the line discipline
		*before* decrementing the tty count variable; this removes
		a potential race condition which occurs when the line
		discipline close blocks, and another process then tries
		open the same serial port.

	* serial.c (rs_hangup): When hanging up, flush the output buffer
		before shutting down the UART.  Otherwise the line
		discipline close blocks waiting for the characters to get
		flushed, which never happens until the serial port gets reused.

Wed Apr 12 08:06:16 1995  Theodore Y. Ts'o  <tytso@localhost>

	* serial.c (do_serial_hangup, do_softint, check_modem_status,
		rs_init):  Hangups are now scheduled via a separate tqueue
		structure in the async_struct structure, tqueue_hangup.
		This task is pushed on to the tq_schedule queue, so that
		it is processed synchronously by the scheduler.

Sat Feb 18 12:13:51 1995  Theodore Y. Ts'o  (tytso@rt-11)

	* tty_io.c (disassociate_ctty, tty_open, tty_ioctl): Clear
		current->tty_old_pgrp field when a session leader
		acquires a controlling tty, and after a session leader
		has disassociated from a controlling tty.

Fri Feb 17 09:34:09 1995  Theodore Y. Ts'o  (tytso@rt-11)

	* serial.c (rs_interrupt_single, rs_interrupt, rs_interrupt_multi): 
		Change the the number of passes made from 64 to be 256,
		configurable with the #define RS_ISR_PASS_LIMIT.

	* serial.c (rs_init, set_serial_info, get_serial_info, rs_close):
		Remove support for closing_wait2.  Instead, set
		tty->closing and rely on the line discipline to prevent
		echo wars.

	* n_tty.c (n_tty_receive_char):  IEXTEN does not need to be
		enabled in order for IXANY to be active.

		If tty->closing is set, then only process XON and XOFF
                characters.

Sun Feb 12 23:57:48 1995  Theodore Y. Ts'o  (tytso@rt-11)

	* serial.c (rs_timer): Change the interrupt poll time from 60
		seconds to 10 seconds, configurable with the #define
		RS_STROBE_TIME.

	* serial.c (rs_interrupt_multi, startup, shutdown, rs_ioctl,
		set_multiport_struct, get_multiport_struct): Add
		provisions for a new type of interrupt service routine,
		which better supports multiple serial ports on a single
		IRQ.  

Sun Feb  5 19:35:11 1995  Theodore Y. Ts'o  (tytso@rt-11)

	* tty_ioctl.c (n_tty_ioctl, set_termios, tty_wait_until_sent): 
	* serial.c (rs_ioctl, rs_close): 
	* cyclades.c (cy_ioctl, cy_close): 
	* n_tty.c (n_tty_close):  Rename wait_until_sent to
		tty_wait_until_sent, so that it's a better name to export
		in ksyms.c.

Sat Feb  4 23:36:20 1995  Theodore Y. Ts'o  (tytso@rt-11)

	* serial.c (rs_close): Added missing check for closing_wait2 being
		ASYNC_CLOSING_WAIT_NONE.

Thu Jan 26 09:02:49 1995  Theodore Y. Ts'o  (tytso@rt-11)

	* serial.c (rs_init, set_serial_info, get_serial_info,
		rs_close): Support close_wait in the serial driver.
		This is helpful for slow devices (like serial
		plotters) so that their outputs don't get flushed upon
		device close.  This has to be configurable because
		normally we don't want ports to be hung up for long
		periods of time during a close when they are not
		connected to a device, or the device is powered off.

		The default is to wait 30 seconds; in the case of a
		very slow device, the close_wait timeout should be
		lengthened.  If it is set to 0, the kernel will wait
		forever for all of the data to be transmitted.

Thu Jan 17 01:17:20 1995  Theodore Y. Ts'o  (tytso@rt-11)

	* serial.c (startup, change_speed, rs_init): Add support to detect
		the StarTech 16650 chip.  Treat it as a 16450 for now,
		because of its FIFO bugs.

Thu Jan  5 21:21:57 1995  <dhinds@allegro.stanford.edu>

	* serial.c: (receive_char): Added counter to prevent infinite loop
		when a PCMCIA serial device is ejected.

Thu Dec 29 17:53:48 1994    <tytso@rsx-11.mit.edu>

	* tty_io.c (check_tty_count): New procedure which checks
		tty->count to make sure that it matches with the number of
		open file descriptors which point at the structure.  If
		the number doesn't match, it prints a warning message.

Wed Dec 28 15:41:51 1994    <tytso@rsx-11.mit.edu>

	* tty_io.c (do_tty_hangup, disassociate_ctty): At hangup time,
		save the tty's current foreground process group in the
		session leader's task structure.  When the session leader
		terminates, send a SIGHUP, SIGCONT to that process group.
		This is not required by POSIX, but it's not prohibited
		either, and it appears to be the least intrusive way
		to fix a problem that dialup servers have with
		orphaned process groups caused by modem hangups.

Thu Dec  8 14:52:11 1994    <tytso@rsx-11.mit.edu>

	* serial.c (rs_ioctl): Don't allow most ioctl's if the serial port
		isn't initialized.

	* serial.c (rs_close): Don't clear the IER if the serial port
		isn't initialized. 

	* serial.c (block_til_ready): Don't try to block on the dialin
		port if the serial port isn't initialized.

Wed Dec  7 10:48:30 1994  Si Park (si@wimpol.demon.co.uk)
	* tty_io.c (tty_register_driver): Fix bug when linking onto
		the tty_drivers list. We now test that there are elements
		already on the list before setting the back link from the
		first element to the new driver.

	* tty_io.c (tty_unregister_driver): Fix bug in unlinking the
		specified driver from the tty_drivers list. We were not
		setting the back link correctly. This used to result in
		a dangling back link pointer and cause panics on the next
		call to get_tty_driver().

Tue Nov 29 10:21:09 1994  Theodore Y. Ts'o  (tytso@rt-11)

	* tty_io.c (tty_unregister_driver): Fix bug in
		tty_unregister_driver where the pointer to the refcount is
		tested, instead of the refcount itself.  This caused
		tty_unregister_driver to always return EBUSY.

Sat Nov 26 11:59:24 1994  Theodore Y. Ts'o  (tytso@rt-11)

	* tty_io.c (tty_ioctl): Add support for the new ioctl
		TIOCTTYGSTRUCT, which allow a kernel debugging program
		direct read access to the tty and tty_driver structures.

Fri Nov 25 17:26:22 1994  Theodore Y. Ts'o  (tytso@rt-11)

	* serial.c (rs_set_termios): Don't wake up processes blocked in
		open when the CLOCAL flag changes, since a blocking
		open only samples the CLOCAL flag once when it blocks,
		and doesn't check it again.  (n.b.  FreeBSD has a
		different behavior for blocking opens; it's not clear
		whether Linux or FreeBSD's interpretation is correct.
		POSIX doesn't give clear guidance on this issue, so
		this may change in the future....)

	* serial.c (block_til_ready): Use the correct termios structure to
		check the CLOCAL flag.  If the cuaXX device is active,
		then check the saved termios for the ttySXX device.
		Otherwise, use the currently active termios structure.

Sun Nov  6 21:05:44 1994  Theodore Y. Ts'o  (tytso@rt-11)

	* serial.c (change_speed): Add support for direct access of
		57,600 and 115,200 bps.

Wed Nov  2 10:32:36 1994  Theodore Y. Ts'o  (tytso@rt-11)

	* n_tty.c (n_tty_receive_room): Only allow excess characters
		through if we are in ICANON mode *and* there are other no
		pending lines in the buffer.  Otherwise cut and paste over
		4k breaks.

Sat Oct 29 18:17:34 1994  Theodore Y. Ts'o  (tytso@rt-11)

	* serial.c (rs_ioctl, get_lsr_info): Added patch suggested by Arne
		Riiber so that user mode programs can tell when the
		transmitter shift register is empty.

Thu Oct 27 23:14:29 1994  Theodore Y. Ts'o  (tytso@rt-11)

	* tty_ioctl.c (wait_until_sent): Added debugging printk statements
		(under the #ifdef TTY_DEBUG_WAIT_UNTIL_SENT)  

	* serial.c (rs_interrupt, rs_interrupt_single, receive_chars,
		change_speed, rs_close): rs_close now disables receiver
		interrupts when closing the serial port.  This allows the
		serial port to close quickly when Linux and a modem (or a
		mouse) are engaged in an echo war; when closing the serial
		port, we now first stop listening to incoming characters,
		and *then* wait for the transmit buffer to drain.  

		In order to make this change, the info->read_status_mask
		is now used to control what bits of the line status
		register are looked at in the interrupt routine in all
		cases; previously it was only used in receive_chars to
		select a few of the status bits.

Mon Oct 24 23:36:21 1994  Theodore Y. Ts'o  (tytso@rt-11)

	* serial.c (rs_close): Add a timeout to the transmitter flush
		loop; this is just a sanity check in case we have flaky
		(or non-existent-but-configured-by-the-user) hardware.

Fri Oct 21 09:37:23 1994  Theodore Y. Ts'o  (tytso@rt-11)

	* tty_io.c (tty_fasync): When asynchronous I/O is enabled, if the
		process or process group has not be specified yet, set it
		to be the tty's process group, or if that is not yet set,
		to the current process's pid.

Thu Oct 20 23:17:28 1994  Theodore Y. Ts'o  (tytso@rt-11)

	* n_tty.c (n_tty_receive_room): If we are doing input
		canonicalization, let as many characters through as
		possible, so that the excess characters can be "beeped".

Tue Oct 18 10:02:43 1994  Theodore Y. Ts'o  (tytso@rt-11)

	* serial.c (rs_start): Removed an incorrect '!' that was
		preventing transmit interrupts from being re-enabled in
		rs_start().  Fortunately in most cases it would be
		re-enabled elsewhere, but this still should be fixed
		correctly.

Sun Oct  9 23:46:03 1994  Theodore Y. Ts'o  (tytso@rt-11)

	* tty_io.c (do_tty_hangup): If the tty driver flags
		TTY_DRIVER_RESET_TERMIOS is set, then reset the termios
		settings back to the driver's initial configuration.  This
		allows the termios settings to be reset even if a process
		has hung up file descriptors keeping a pty's termios from
		being freed and reset.

	* tty_io.c (release_dev): Fix memory leak.  The pty's other
		termios structure should also be freed.

	* serial.c (rs_close, shutdown): Change how we wait for the
		transmitter to completely drain before shutting down the
		serial port.  We now do it by scheduling in another
		process instead of busy looping with the interrupts turned
		on.  This may eliminate some race condition problems that
		some people seem to be reporting.

Sun Sep 25 14:18:14 1994  Theodore Y. Ts'o  (tytso@rt-11)

	* tty_io.c (release_dev): When freeing a tty make sure that both
		the tty and the o_tty (if present) aren't a process's
		controlling tty.  (Previously, we only checked the tty.)

	* serial.c (change_speed): Only enable the Modem Status
		Interrupt for a port if CLOCAL is not set or CRTSCTS
		is set.  If we're not checking the carrier detect and
		CTS line, there's no point in enabling the modem
		status interrupt.  This will save spurious interrupts
		from slowing down systems who have terminals that
		don't support either line.  (Of course, if you want
		only one of CD and CTS support, you will need a
		properly wired serial cable.)

Thu Sep 22 08:32:48 1994  Theodore Y. Ts'o  (tytso@rt-11)

	* tty_io.c (do_SAK): Return if tty is null.

	* tty_io.c (_tty_name): Return "NULL tty" if the passed in tty is
		NULL.

Sat Sep 17 13:19:25 1994  Theodore Y. Ts'o  (tytso@rt-11)

	* tty_ioctl.c (n_tty_ioctl): Fix TIOCGLCKTRMIOS and
		TIOCSLCKTRMIOS, which were totally broken.  Remove
		extra indirection from argument; it should be a struct
		termios *, not a struct termios **.
		&real_tty->termios_locked should have been
		real_tty->termios_locked.  This caused us to be
		reading and writing the termios_locked structure to
		random places in kernel memory.  

	* tty_io.c (release_dev): Oops!  Forgot to delete a critical kfree
		of the locked_termios.  This leaves the locked_termios
		structure pointed at a freed object.  

Fri Sep 16 08:13:25 1994  Theodore Y. Ts'o  (tytso@rt-11)

	* tty_io.c (tty_open): Don't check for an exclusive open until
		after the device specific open routine has been called.
		Otherwise, the serial device ref counting will be screwed
		up.

	* serial.c (rs_open, block_til_ready): Don't set termios structure
		until after block_til_ready has returned successfully.
		Modify block_til_ready to check the normal_termios
		structure directly, so it doesn't rely on termios being
		set before it's called.

Thu Sep 15 23:34:01 1994  Theodore Y. Ts'o  (tytso@rt-11)

	* serial.c (rs_close): Turn off interrupts during rs_close() to
		prevent a race condition with the hangup code (which
		runs during a software interrupt).

	* tty_io.c (release_dev): Don't free the locked_termios structure;
		its state must be retained across device opens.


	* tty_io.c (tty_unregister_driver): Added function to unregister a
		tty driver.  (For loadable device drivers.)