When releasing one end of a pty pair, that end may just have written
to the other, which the input processing worker, flush_to_ldisc(), is
still working on but has not completed the copy to the other end's
read buffer. So input may not appear to be available to a waiting
reader but yet TTY_OTHER_CLOSED is now observed. The n_tty line
discipline has worked around this by waiting for input processing
to complete and then re-checking if input is available before
exiting with -EIO.
Since the tty/ldisc lock reordering, the wait for input processing
to complete can now occur during final close before setting
TTY_OTHER_CLOSED. In this way, a waiting reader is guaranteed to
see input available (if any) before observing TTY_OTHER_CLOSED.
Reviewed-by: Alan Cox <alan@linux.intel.com>
Signed-off-by: Peter Hurley <peter@hurleysoftware.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
With the revised tty lock order and lockdep annotation, claiming
the pty slave lock is now safe while still holding the pty master lock.
Signed-off-by: Peter Hurley <peter@hurleysoftware.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
Eliminate the requirement of specifying the tty lock nesting at
lock time; instead, set the lock subclass for slave ptys at pty
install (normal ttys and master ptys use subclass 0).
Signed-off-by: Peter Hurley <peter@hurleysoftware.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
Opening the slave BSD pty first already returns -EIO from the slave
pty_open(), which in turn causes the newly installed tty pair to be
released before returning from tty_open(). However, this can also
cause a parallel master BSD pty open to fail because the pty pair
destruction may already been taking place in tty_release().
Failing at driver->install() if the slave pty is opened first ensures
that a pty master open cannot fail, because the driver tables will
not have been updated so tty_driver_lookup_tty() won't find the
master pty (and attempt to "re-open" it).
In turn, this guarantees that any tty with a tty->count == 0 is
in final close (rather than never opened).
Reviewed-by: Alan Cox <alan@linux.intel.com>
Signed-off-by: Peter Hurley <peter@hurleysoftware.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
Updates to the packet mode enable require holding the ctrl_lock;
the serialization prevents corruption of adjacent fields.
Signed-off-by: Peter Hurley <peter@hurleysoftware.com>
Reviewed-by: Alan Cox <alan@linux.intel.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
Because pty_set_pktmode() does not claim the slave's ctrl_lock
to clear ->ctrl_status (to avoid unnecessary lock nesting),
pty_set_pktmode() may accidentally erase new ->ctrl_status updates.
For example,
CPU 0 | CPU 1
pty_set_pktmode() | pty_start()
spin_lock(master's ctrl_lock) |
tty->packet = 1 |
| if (tty->link->packet)
| spin_lock(slave's ctrl_lock)
| tty->ctrl_status = TIOCPKT_START
tty->link->ctrl_status = 0 |
Ensure the clear of ->ctrl_status occurs before packet mode is set
(and observable on another cpu).
Signed-off-by: Peter Hurley <peter@hurleysoftware.com>
Reviewed-by: Alan Cox <alan@linux.intel.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
The slave's ctrl_lock serializes updates to the ctrl_status field
only, whereas the master's ctrl_lock serializes updates to the
packet mode enable (ie., the master does not have ctrl_status and
the slave does not have packet mode). Thus, claiming the slave's
ctrl_lock to access ->packet is useless.
Unlocked reads of ->packet are already smp-safe.
Signed-off-by: Peter Hurley <peter@hurleysoftware.com>
Reviewed-by: Alan Cox <alan@linux.intel.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
Interrupts are enabled in the n_tty_read() loop, ioctl(TIOCPKT)
and pty driver flush_buffer() routine; no need to save and restore
local interrupt state.
Signed-off-by: Peter Hurley <peter@hurleysoftware.com>
Reviewed-by: Alan Cox <alan@linux.intel.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
The tty driver's set_termios() method is called with interrupts
enabled; there is no need to save and restore the local interrupt state.
Signed-off-by: Peter Hurley <peter@hurleysoftware.com>
Reviewed-by: Alan Cox <alan@linux.intel.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
Packet mode is unique to the pty driver; move the packet mode state
change code from the generic tty ioctl handler to the pty driver.
Signed-off-by: Peter Hurley <peter@hurleysoftware.com>
Reviewed-by: Alan Cox <alan@linux.intel.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
The pty master's termios should never be set; currently, all code
paths which call the driver's set_termios() method ensure that the
pty slave's termios is being set.
Signed-off-by: Peter Hurley <peter@hurleysoftware.com>
Reviewed-by: Alan Cox <alan@linux.intel.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
When a master pty is set to packet mode, flow control changes to
the slave pty cause notifications to the master pty via reads and
polls. However, these tests are occurring for all ttys, not
just ptys.
Implement flow control packet mode notifications in the pty driver.
Only the slave side implements the flow control handlers since
packet mode is asymmetric; the master pty receives notifications
for slave-side changes, but not vice versa.
Signed-off-by: Peter Hurley <peter@hurleysoftware.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
In 2c964a2f "drivers: tty: Merge alloc_tty_struct and
initialize_tty_struct", I messed up the refactorization of
pty_common_install, causing use-after-free and NULL pointer derefs on
various error paths. This should fix it.
Reported-by: Julia Lawall <julia.lawall@lip6.fr>
Reported-by: Dan Carpenter <dan.carpenter@oracle.com>
Signed-off-by: Rasmus Villemoes <linux@rasmusvillemoes.dk>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
The two functions alloc_tty_struct and initialize_tty_struct are
always called together. Merge them into alloc_tty_struct, updating its
prototype and the only two callers of these functions.
Signed-off-by: Rasmus Villemoes <linux@rasmusvillemoes.dk>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
Acquiring the write_wait queue spin lock now accounts for the largest
slice of cpu time on the tty write path. Two factors contribute to
this situation; a overly-pessimistic line discipline write loop which
_always_ sets up a wait loop even if i/o will immediately succeed, and
on ptys, a wakeup storm from reads and writes.
Writer wakeup does not need to be performed by the pty driver.
Firstly, since the actual i/o is performed within the write, the
line discipline write loop will continue while space remains in
the flip buffers. Secondly, when space becomes avail in the
line discipline receive buffer (and thus also in the flip buffers),
the pty unthrottle re-wakes the writer (non-flow-controlled line
disciplines unconditionally unthrottle the driver when data is
received). Thus, existing in-kernel i/o is guaranteed to advance.
Finally, writer wakeup occurs at the conclusion of the line discipline
write (in tty_write_unlock()). This guarantees that any user-space write
waiters are woken to continue additional i/o.
Signed-off-by: Peter Hurley <peter@hurleysoftware.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
Lockless flip buffers require atomically updating the bytes-in-use
watermark.
The pty driver also peeks at the watermark value to limit
memory consumption to a much lower value than the default; query
the watermark with new fn, tty_buffer_space_avail().
Signed-off-by: Peter Hurley <peter@hurleysoftware.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
Commit 699390354d
('pty: Ignore slave pty close() if never successfully opened')
introduced a bug with ptys whereby a write() in parallel with an
open() on an existing pty could mistakenly indicate an I/O error.
Only indicate an I/O error if the condition on open() actually exists.
Reported-by: Markus Trippelsdorf <markus@trippelsdorf.de>
Signed-off-by: Peter Hurley <peter@hurleysoftware.com>
Tested-by: Mikael Pettersson <mikpe@it.uu.se>
Cc: stable <stable@vger.kernel.org> # 3.9
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
We first tried to avoid updating atime/mtime entirely (commit
b0de59b573: "TTY: do not update atime/mtime on read/write"), and then
limited it to only update it occasionally (commit 37b7f3c765: "TTY:
fix atime/mtime regression"), but it turns out that this was both
insufficient and overkill.
It was insufficient because we let people attach to the shared ptmx node
to see activity without even reading atime/mtime, and it was overkill
because the "only once a minute" means that you can't really tell an
idle person from an active one with 'w'.
So this tries to fix the problem properly. It marks the shared ptmx
node as un-notifiable, and it lowers the "only once a minute" to a few
seconds instead - still long enough that you can't time individual
keystrokes, but short enough that you can tell whether somebody is
active or not.
Reported-by: Simon Kirby <sim@hostway.ca>
Acked-by: Jiri Slaby <jslaby@suse.cz>
Cc: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
Cc: stable@vger.kernel.org
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
When CONFIG_UNIX98_PTYS is unset, we see this warning in pty:
drivers/tty/pty.c:409:13: warning: ‘pty_unix98_shutdown’ defined but not used
Fix that by moving the function to a section which depends on that
config.
Signed-off-by: Jiri Slaby <jslaby@suse.cz>
Reported-by: Toralf Foerster <toralf.foerster@gmx.de>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
port->itty has already been reset by release_tty() before
pty_cleanup() is called.
Call stack:
release_tty()
tty_kref_put()
queue_release_one_tty()
release_one_tty() : workqueue
tty->ops->cleanup()
pty_cleanup()
Signed-off-by: Peter Hurley <peter@hurleysoftware.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
Multiple slave pty opens may be performed in parallel with the
master open. Of course, all the slave opens will fail because the
master pty is still locked but during this time the slave pty
count will be artificially greater than 1. This is should not
cause the master pty open to fail.
Signed-off-by: Peter Hurley <peter@hurleysoftware.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
If the master and slave ptys are opened in parallel, the slave open
fails because the pty is still locked. This is as designed.
However, pty_close() is still called for the slave pty which sets
TTY_OTHER_CLOSED in the master pty. This can cause the master open
to fail as well.
Use a common pattern in other tty drivers by setting TTY_IO_ERROR
until the open is successful and only closing the pty if not set.
Note: the master pty always closes regardless of whether the open
was successful, so that proper cleanup can occur.
Signed-off-by: Peter Hurley <peter@hurleysoftware.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
Commit bbb63c514a (drivers:tty:fix up
ENOIOCTLCMD error handling) changed the default return value from tty
ioctl to be ENOTTY and not EINVAL. This is appropriate.
But in case of TIOCGPTN for the old BSD ptys glibc started failing
because it expects EINVAL to be returned. Only then it continues to
obtain the pts name the other way around.
So fix this case by explicit return of EINVAL in this case.
Signed-off-by: Jiri Slaby <jslaby@suse.cz>
Reported-by: Florian Westphal <fw@strlen.de>
Cc: Alan Cox <alan@linux.intel.com>
Cc: stable <stable@vger.kernel.org> # 3.7+
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
Now that login from util-linux is forced to drop all references to a
TTY which it wants to hangup (to reach reference count 1) we are
seeing issues with telnet. When login closes its last reference to the
slave PTY, it also resets packet mode on the *master* side. And we
have a race here.
What telnet does is fork+exec of `login'. Then there are two
scenarios:
* `login' closes the slave TTY and resets thus master's packet mode,
but even now telnet properly sets the mode, or
* `telnetd' sets packet mode on the master, `login' closes the slave
TTY and resets master's packet mode.
The former case is OK. However the latter happens in much more cases,
by the order of magnitude to be precise. So when one tries to login to
such a messed telnet setup, they see the following:
inux login:
ogin incorrect
Note the missing first letters -- telnet thinks it is still in the
packet mode, so when it receives "linux login" from `login', it
considers "l" as the type of the packet and strips it.
SuS does not mention how the implementation should behave. Both BSDs I
checked (Free and Net) do not reset the flag upon the last close.
By this I am resurrecting an old bug, see References. We are hitting
it regularly now, i.e. with updated util-linux, ergo login.
Here, I am changing a behavior introduced back in 2.1 times. It would
better have a long time testing before goes upstream.
Signed-off-by: Jiri Slaby <jslaby@suse.cz>
Cc: Mauro Carvalho Chehab <mchehab@redhat.com>
Cc: Bryan Mason <bmason@redhat.com>
References: https://lkml.org/lkml/2009/11/11/223
References: https://bugzilla.redhat.com/show_bug.cgi?id=504703
References: https://bugzilla.novell.com/show_bug.cgi?id=797042
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
spaces are used for indent in 3 places of tty/pty.c, we change it to tab.
Signed-off-by: Cong Ding <dinggnu@gmail.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
the "\n" in panic message is excess, so we remove it in tty/pty.c as what it
is used in other places.
Signed-off-by: Cong Ding <dinggnu@gmail.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
Now, we start converting tty buffer functions to actually use
tty_port. This will allow us to get rid of the need of tty in many
call sites. Only tty_port will needed and hence no more
tty_port_tty_get in those paths.
Now, the one where most of tty_port_tty_get gets removed:
tty_flip_buffer_push.
IOW we also closed all the races in drivers not using tty_port_tty_get
at all yet.
Also we move tty_flip_buffer_push declaration from include/linux/tty.h
to include/linux/tty_flip.h to all others while we are changing it
anyway.
Signed-off-by: Jiri Slaby <jslaby@suse.cz>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
Now, we start converting tty buffer functions to actually use
tty_port. This will allow us to get rid of the need of tty in many
call sites. Only tty_port will needed and hence no more
tty_port_tty_get in those paths.
tty_insert_flip_string this time.
Signed-off-by: Jiri Slaby <jslaby@suse.cz>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
After commit "TTY: move tty buffers to tty_port", the tty buffers are
not freed in some drivers. This is because tty_port_destructor is not
called whenever a tty_port is freed. This was an assumption I counted
with but was unfortunately untrue. So fix the drivers to fulfil this
assumption.
PTY is one of those, here we just need to use tty_port_put instead of
kfree. (Assuming tty_port_destructor does not need port->ops to be set
which we change here too.)
Signed-off-by: Jiri Slaby <jslaby@suse.cz>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
For checkpoint/restore we need to know if tty has
exclusive or packet mode set, as well as if pty
is currently locked. Just to be able to restore
this characteristics.
For this sake the following ioctl codes are introduced
- TIOCGPKT to get packet mode state
- TIOCGPTLCK to get Pty locked state
- TIOCGEXCL to get Exclusive mode state
Note this ioctls are a bit unsafe in terms of data
obtained consistency. The tty characteristics might
be changed right after ioctl complete. Keep it in
mind and use this ioctl carefully.
v2:
- Use TIOC prefix for ioctl codes (by jslaby@)
Signed-off-by: Cyrill Gorcunov <gorcunov@openvz.org>
CC: Alan Cox <alan@lxorguk.ukuu.org.uk>
CC: "H. Peter Anvin" <hpa@zytor.com>
CC: Pavel Emelyanov <xemul@parallels.com>
CC: Jiri Slaby <jslaby@suse.cz>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
Since this ioctl is for pty devices only move it to pty.c.
v2:
- drop PTY_TYPE_MASTER test since it's master peer
ioctl anyway (by jslaby@)
Suggested-by: Alan Cox <alan@lxorguk.ukuu.org.uk>
Signed-off-by: Cyrill Gorcunov <gorcunov@openvz.org>
CC: "H. Peter Anvin" <hpa@zytor.com>
CC: Pavel Emelyanov <xemul@parallels.com>
CC: Jiri Slaby <jslaby@suse.cz>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
So this is it. The big step why we did all the work over the past
kernel releases. Now everything is prepared, so nothing protects us
from doing that big step.
| | \ \ nnnn/^l | |
| | \ / / | |
| '-,.__ => \/ ,-` => | '-,.__
| O __.´´) ( .` | O __.´´)
~~~ ~~ `` ~~~ ~~
The buffers are now in the tty_port structure and we can start
teaching the buffer helpers (insert char/string, flip etc.) to use
tty_port instead of tty_struct all around.
Signed-off-by: Jiri Slaby <jslaby@suse.cz>
Acked-by: Alan Cox <alan@linux.intel.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
For that purpose we have to temporarily introduce a second tty back
pointer into tty_port. It is because serial layer, and maybe others,
still do not use tty_port_tty_set/get. So that we cannot set the
tty_port->tty to NULL at will now.
Yes, the fix would be to convert whole serial layer and all its users
to tty_port_tty_set/get. However we are in the process of removing the
need of tty in most of the call sites, so this would lead to a
duplicated work.
Instead we have now tty_port->itty (internal tty) which will be used
only in flush_to_ldisc. For that one it is ensured that itty is valid
wherever the work is run. IOW, the work is synchronously cancelled
before we set itty to NULL and also before hangup is processed.
After we need only tty_port and not tty_struct in most code, this
shall be changed to tty_port_tty_set/get and itty removed completely.
Signed-off-by: Jiri Slaby <jslaby@suse.cz>
Acked-by: Alan Cox <alan@linux.intel.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
Now that we have control over tty->driver_data in pty, we can just
kill the /dev/pts/ in pty code too. Namely, in ->shutdown hook of
tty. For pty, this is called only once, for whichever end is closed
last. But we don't care, both driver_data are the inode as it used to
be till now.
Signed-off-by: Jiri Slaby <jslaby@suse.cz>
Acked-by: Alan Cox <alan@linux.intel.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
The goal is to stop setting and using tty->driver_data in devpts code.
It should be used solely by the driver's code, pty in this case.
Now driver_data are managed only in the pty driver. devpts_pty_new is
switched to accept what we used to dig out of tty_struct, i.e. device
node number and index.
This also removes a note about driver_data being set outside of the
driver.
Signed-off-by: Jiri Slaby <jslaby@suse.cz>
Acked-by: Alan Cox <alan@linux.intel.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
The goal is to stop setting and using tty->driver_data in devpts code.
It should be used solely by the driver's code, pty in this case.
For the cleanup of layering, we will need the inode created in
devpts_pty_new to be stored into slave's driver_data. So we convert
devpts_pty_new to return the inode or an ERR_PTR-encoded error in case
of failure.
The move of 'inode = new_inode(sb);' from declarators to the code is
only cosmetical, but it makes the code easier to read.
Signed-off-by: Jiri Slaby <jslaby@suse.cz>
Acked-by: Alan Cox <alan@linux.intel.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
The goal is to stop setting and using tty->driver_data in devpts code.
It should be used solely by the driver's code, pty in this case.
First, here we remove TTY from devpts_get_tty and rename it to
devpts_get_priv. Note we do not remove type safety, we just shift the
[implicit] (void *) cast one layer up.
index was unused in devpts_get_tty, so remove that from the prototype
too.
Signed-off-by: Jiri Slaby <jslaby@suse.cz>
Acked-by: Alan Cox <alan@linux.intel.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
We end up dropping the mutex twice on some errors. We don't want to do
that.
Reported-by: Fengguang Wu <fengguang.wu@intel.com>
Signed-off-by: Alan Cox <alan@linux.intel.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
We changed these from alloc_tty_driver() to tty_alloc_driver() so the
error handling needs to modified to check for IS_ERR() instead of NULL.
Signed-off-by: Dan Carpenter <dan.carpenter@oracle.com>
Acked-by: Jiri Slaby <jslaby@suse.cz>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
Switch to the new driver allocation interface, as this is one of the
special call-sites. Here, we need TTY_DRIVER_DYNAMIC_ALLOC to not
allocate tty_driver->ports, cdevs and potentially other structures
because we reserve too many lines in pty. Instead, it provides the
tty_port<->tty_struct link in tty->ops->install already.
Signed-off-by: Jiri Slaby <jslaby@suse.cz>
Acked-by: Alan Cox <alan@linux.intel.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
In case alloc_tty_struct fails in pty_common_install, we pass NULL to
free_tty_struct. This is invalid as the function is not ready to cope
with that. And even if it was, it is not nice to do that anyway.
Signed-off-by: Jiri Slaby <jslaby@suse.cz>
Reported-by: Dan Carpenter <dan.carpenter@oracle.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
The termios and other changes mean the other protections needed on the driver
tty arrays should be adequate. Turn it all back on.
This contains pieces folded in from the fixes made to the original patches
| From: Geert Uytterhoeven <geert@linux-m68k.org> (fix m68k)
| From: Paul Gortmaker <paul.gortmaker@windriver.com> (fix cris)
| From: Jiri Kosina <jkosina@suze.cz> (lockdep)
| From: Eric Dumazet <eric.dumazet@gmail.com> (lockdep)
Signed-off-by: Alan Cox <alan@linux.intel.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
Ian Abbott found that the tty layer would explode with the right set of
parallel open and close operations. This is because we race in the
handling of tty->drivers->termios[].
Correct this by
Making tty_ldisc_release behave like nromal code (takes the lock,
does stuff, drops the lock)
Drop the tty lock earlier in tty_ldisc_release
Taking the tty mutex around the driver->termios update in all cases
Adding a WARN_ON to catch future screwups.
I also forgot to clean up the pty resources properly. With a pty pair we
need to pull both halves out of the tables.
Signed-off-by: Alan Cox <alan@linux.intel.com>
Tested-by: Ian Abbott <abbotti@mev.co.uk>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
Now that we don't have tty->termios tied to drivers->tty we can untangle
the logic here. In addition we can push the removal logic out of the
destructor path.
At that point we can think about sorting out tty_port and console and all
the other ugly hangovers.
Signed-off-by: Alan Cox <alan@linux.intel.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
