OpenSSH expects the (non-blocking) read() of pty master to return
EAGAIN only if it has received all of the slave-side output after
it has received SIGCHLD. This used to work on pre-3.12 kernels.
This fix effectively forces non-blocking read() and poll() to
block for parallel i/o to complete for all ttys. It also unwinds
these changes:
1) f8747d4a46
tty: Fix pty master read() after slave closes
2) 52bce7f8d4
pty, n_tty: Simplify input processing on final close
3) 1a48632ffe
pty: Fix input race when closing
Inspired by analysis and patch from Marc Aurele La France <tsi@tuyoix.net>
Reported-by: Volth <openssh@volth.com>
Reported-by: Marc Aurele La France <tsi@tuyoix.net>
BugLink: https://bugzilla.mindrot.org/show_bug.cgi?id=52
BugLink: https://bugzilla.mindrot.org/show_bug.cgi?id=2492
Signed-off-by: Brian Bloniarz <brian.bloniarz@gmail.com>
Reviewed-by: Peter Hurley <peter@hurleysoftware.com>
Cc: stable <stable@vger.kernel.org>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
Instead of two distinct code branches for receive_buf() handling,
use tty_ldisc_receive_buf() as the single code path.
Signed-off-by: Peter Hurley <peter@hurleysoftware.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
The commonly accepted wisdom that scheduling work on the same cpu
that handled interrupt i/o benefits from cache-locality is only
true if the cpu is idle (since bound kworkers are often the highest
vruntime and thus the lowest priority).
Measurements of scheduling via the unbound queue show lowered
worst-case latency responses of up to 5x over bound workqueue, without
increase in average latency or throughput.
pty i/o test measurements show >3x (!) reduced total running time; tests
previously taking ~8s now complete in <2.5s.
Signed-off-by: Peter Hurley <peter@hurleysoftware.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
Introduce API functions to restart and cancel tty buffer work, rather
than manipulate buffer work directly.
Signed-off-by: Peter Hurley <peter@hurleysoftware.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
Race on buffer data happens when newly committed data is
picked up by an old flush work in the following scenario:
__tty_buffer_request_room does a plain write of tail->commit,
no barriers were executed before that.
At this point flush_to_ldisc reads this new value of commit,
and reads buffer data, no barriers in between.
The committed buffer data is not necessary visible to flush_to_ldisc.
Similar bug happens when tty_schedule_flip commits data.
Update commit with smp_store_release and read commit with
smp_load_acquire, as it is commit that signals data readiness.
This is orthogonal to the existing synchronization on tty_buffer.next,
which is required to not dismiss a buffer with unconsumed data.
The data race was found with KernelThreadSanitizer (KTSAN).
Signed-off-by: Dmitry Vyukov <dvyukov@google.com>
Reviewed-by: Peter Hurley <peter@hurleysoftware.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
tty_buffer_flush frees not acquired buffers.
As the result, for example, read of b->size in tty_buffer_free
can return garbage value which will lead to a huge buffer
hanging in the freelist. This is just the benignest
manifestation of freeing of a not acquired object.
If the object is passed to kfree, heap can be corrupted.
Acquire visibility over the buffer before freeing it.
The data race was found with KernelThreadSanitizer (KTSAN).
Signed-off-by: Dmitry Vyukov <dvyukov@google.com>
Reviewed-by: Peter Hurley <peter@hurleysoftware.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
flush_to_ldisc reads port->itty and checks that it is not NULL,
concurrently release_tty sets port->itty to NULL. It is possible
that flush_to_ldisc loads port->itty once, ensures that it is
not NULL, but then reloads it again and uses. The second load
can already return NULL, which will cause a crash.
Use READ_ONCE to read port->itty.
The data race was found with KernelThreadSanitizer (KTSAN).
Signed-off-by: Dmitry Vyukov <dvyukov@google.com>
Reviewed-by: Peter Hurley <peter@hurleysoftware.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
The advance of the 'read' buffer index belongs in the outer
flip buffer consume loop, with the other buffer index arithmetic.
No functional change.
Signed-off-by: Peter Hurley <peter@hurleysoftware.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
Clarify flip buffer producer/consumer operation; the use of
smp_load_acquire() and smp_store_release() more clearly indicates
which memory access requires a barrier.
Signed-off-by: Peter Hurley <peter@hurleysoftware.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
A read() from a pty master may mistakenly indicate EOF (errno == -EIO)
after the pty slave has closed, even though input data remains to be read.
For example,
pty slave | input worker | pty master
| |
| | n_tty_read()
pty_write() | | input avail? no
add data | | sleep
schedule worker --->| | .
|---> flush_to_ldisc() | .
pty_close() | fill read buffer | .
wait for worker | wakeup reader --->| .
| read buffer full? |---> input avail ? yes
|<--- yes - exit worker | copy 4096 bytes to user
TTY_OTHER_CLOSED <---| |<--- kick worker
| |
**** New read() before worker starts ****
| | n_tty_read()
| | input avail? no
| | TTY_OTHER_CLOSED? yes
| | return -EIO
Several conditions are required to trigger this race:
1. the ldisc read buffer must become full so the input worker exits
2. the read() count parameter must be >= 4096 so the ldisc read buffer
is empty
3. the subsequent read() occurs before the kicked worker has processed
more input
However, the underlying cause of the race is that data is pipelined, while
tty state is not; ie., data already written by the pty slave end is not
yet visible to the pty master end, but state changes by the pty slave end
are visible to the pty master end immediately.
Pipeline the TTY_OTHER_CLOSED state through input worker to the reader.
1. Introduce TTY_OTHER_DONE which is set by the input worker when
TTY_OTHER_CLOSED is set and either the input buffers are flushed or
input processing has completed. Readers/polls are woken when
TTY_OTHER_DONE is set.
2. Reader/poll checks TTY_OTHER_DONE instead of TTY_OTHER_CLOSED.
3. A new input worker is started from pty_close() after setting
TTY_OTHER_CLOSED, which ensures the TTY_OTHER_DONE state will be
set if the last input worker is already finished (or just about to
exit).
Remove tty_flush_to_ldisc(); no in-tree callers.
Fixes: 52bce7f8d4 ("pty, n_tty: Simplify input processing on final close")
Bugzilla: https://bugzilla.kernel.org/show_bug.cgi?id=96311
BugLink: http://bugs.launchpad.net/bugs/1429756
Cc: <stable@vger.kernel.org> # 3.19+
Reported-by: Andy Whitcroft <apw@canonical.com>
Reported-by: H.J. Lu <hjl.tools@gmail.com>
Signed-off-by: Peter Hurley <peter@hurleysoftware.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
We should not be doing assignments within an if () block
so fix up the code to not do this.
change was created using Coccinelle.
CC: Jiri Slaby <jslaby@suse.cz>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
The pty driver does not clear its write buffer when commanded.
This is to avoid an apparent deadlock between parallel flushes from
both pty ends; specifically when handling either BRK or INTR input.
However, parallel flushes from this source is not possible since
the pty master can never be set to BRKINT or ISIG. Parallel flushes
from other sources are possible but these do not threaten deadlocks.
Annotate the tty buffer mutex for lockdep to represent the nested
tty_buffer locking which occurs when the pty slave is processing input
(its buffer mutex held) and receives INTR or BRK and acquires the
linked tty buffer mutex via tty_buffer_flush().
Signed-off-by: Peter Hurley <peter@hurleysoftware.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
tty_ldisc_flush() first clears the line discipline input buffer,
then clears the tty flip buffers. However, this allows for existing
data in the tty flip buffers to be added after the ldisc input
buffer has been cleared, but before the flip buffers have been cleared.
Add an optional ldisc parameter to tty_buffer_flush() to allow
tty_ldisc_flush() to pass the ldisc to clear.
NB: Initially, the plan was to do this automatically in
tty_buffer_flush(). However, an audit of the behavior of existing
line disciplines showed that performing a ldisc buffer flush on
ioctl(TCFLSH) was not always the outcome. For example, some line
disciplines have flush_buffer() methods but not ioctl() methods,
so a ->flush_buffer() command would be unexpected.
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>
This function is largely a duplicate of paste_selection() in
drivers/tty/vt/selection.c, but with its own selection state. The
speakup selection mechanism should really be merged with vt.
For now, apply the changes from 'TTY: vt, fix paste_selection ldisc
handling', 'tty: Make ldisc input flow control concurrency-friendly',
and 'tty: Fix unsafe vt paste_selection()'.
References: https://bugs.debian.org/735202
References: https://bugs.debian.org/744015
Reported-by: Paul Gevers <elbrus@debian.org>
Reported-and-tested-by: Jarek Czekalski <jarekczek@poczta.onet.pl>
Signed-off-by: Ben Hutchings <ben@decadent.org.uk>
Cc: <stable@vger.kernel.org> # v3.8 but needs backporting for < 3.12
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
Commit 6a20dbd6ca,
"tty: Fix race condition between __tty_buffer_request_room and flush_to_ldisc"
correctly identifies an unsafe race condition between
__tty_buffer_request_room() and flush_to_ldisc(), where the consumer
flush_to_ldisc() prematurely advances the head before consuming the
last of the data committed. For example:
CPU 0 | CPU 1
__tty_buffer_request_room | flush_to_ldisc
... | ...
| count = head->commit - head->read
n = tty_buffer_alloc() |
b->commit = b->used |
b->next = n |
| if (!count) /* T */
| if (head->next == NULL) /* F */
| buf->head = head->next
In this case, buf->head has been advanced but head->commit may have
been updated with a new value.
Instead of reintroducing an unnecessary lock, fix the race locklessly.
Read the commit-next pair in the reverse order of writing, which guarantees
the commit value read is the latest value written if the head is
advancing.
Reported-by: Manfred Schlaegl <manfred.schlaegl@gmx.at>
Cc: <stable@vger.kernel.org> # 3.12.x+
Signed-off-by: Peter Hurley <peter@hurleysoftware.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
This reverts commit 6a20dbd6ca.
Although the commit correctly identifies an unsafe race condition
between __tty_buffer_request_room() and flush_to_ldisc(), the commit
fixes the race with an unnecessary spinlock in a lockless algorithm.
The follow-on commit, "tty: Fix lockless tty buffer race" fixes
the race locklessly.
Signed-off-by: Peter Hurley <peter@hurleysoftware.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
The race was introduced while development of linux-3.11 by
e8437d7ecb and
e9975fdec0.
Originally it was found and reproduced on linux-3.12.15 and
linux-3.12.15-rt25, by sending 500 byte blocks with 115kbaud to the
target uart in a loop with 100 milliseconds delay.
In short:
1. The consumer flush_to_ldisc is on to remove the head tty_buffer.
2. The producer adds a number of bytes, so that a new tty_buffer must
be allocated and added by __tty_buffer_request_room.
3. The consumer removes the head tty_buffer element, without handling
newly committed data.
Detailed example:
* Initial buffer:
* Head, Tail -> 0: used=250; commit=250; read=240; next=NULL
* Consumer: ''flush_to_ldisc''
* consumed 10 Byte
* buffer:
* Head, Tail -> 0: used=250; commit=250; read=250; next=NULL
{{{
count = head->commit - head->read; // count = 0
if (!count) { // enter
// INTERRUPTED BY PRODUCER ->
if (head->next == NULL)
break;
buf->head = head->next;
tty_buffer_free(port, head);
continue;
}
}}}
* Producer: tty_insert_flip_... 10 bytes + tty_flip_buffer_push
* buffer:
* Head, Tail -> 0: used=250; commit=250; read=250; next=NULL
* added 6 bytes: head-element filled to maximum.
* buffer:
* Head, Tail -> 0: used=256; commit=250; read=250; next=NULL
* added 4 bytes: __tty_buffer_request_room is called
* buffer:
* Head -> 0: used=256; commit=256; read=250; next=1
* Tail -> 1: used=4; commit=0; read=250 next=NULL
* push (tty_flip_buffer_push)
* buffer:
* Head -> 0: used=256; commit=256; read=250; next=1
* Tail -> 1: used=4; commit=4; read=250 next=NULL
* Consumer
{{{
count = head->commit - head->read;
if (!count) {
// INTERRUPTED BY PRODUCER <-
if (head->next == NULL) // -> no break
break;
buf->head = head->next;
tty_buffer_free(port, head);
// ERROR: tty_buffer head freed -> 6 bytes lost
continue;
}
}}}
This patch reintroduces a spin_lock to protect this case. Perhaps later
a lock-less solution could be found.
Signed-off-by: Manfred Schlaegl <manfred.schlaegl@gmx.at>
Cc: stable <stable@vger.kernel.org> # 3.11
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
The user-settable knob, low_latency, has been the source of
several BUG reports which stem from flush_to_ldisc() running
in interrupt context. Since 3.12, which added several sleeping
locks (termios_rwsem and buf->lock) to the input processing path,
the frequency of these BUG reports has increased.
Note that changes in 3.12 did not introduce this regression;
sleeping locks were first added to the input processing path
with the removal of the BKL from N_TTY in commit
a88a69c912,
'n_tty: Fix loss of echoed characters and remove bkl from n_tty'
and later in commit 38db89799b,
'tty: throttling race fix'. Since those changes, executing
flush_to_ldisc() in interrupt_context (ie, low_latency set), is unsafe.
However, since most devices do not validate if the low_latency
setting is appropriate for the context (process or interrupt) in
which they receive data, some reports are due to misconfiguration.
Further, serial dma devices for which dma fails, resort to
interrupt receiving as a backup without resetting low_latency.
Historically, low_latency was used to force wake-up the reading
process rather than wait for the next scheduler tick. The
effect was to trim multiple milliseconds of latency from
when the process would receive new data.
Recent tests [1] have shown that the reading process now receives
data with only 10's of microseconds latency without low_latency set.
Remove the low_latency rx steering from tty_flip_buffer_push();
however, leave the knob as an optional hint to drivers that can
tune their rx fifos and such like. Cleanup stale code comments
regarding low_latency.
[1] https://lkml.org/lkml/2014/2/20/434
"Yay.. thats an annoying historical pain in the butt gone."
-- Alan Cox
Reported-by: Beat Bolli <bbolli@ewanet.ch>
Reported-by: Pavel Roskin <proski@gnu.org>
Acked-by: David Sterba <dsterba@suse.cz>
Cc: Grant Edwards <grant.b.edwards@gmail.com>
Cc: Stanislaw Gruszka <sgruszka@redhat.com>
Cc: Hal Murray <murray+fedora@ip-64-139-1-69.sjc.megapath.net>
Cc: <stable@vger.kernel.org> # 3.12.x+
Signed-off-by: Peter Hurley <peter@hurleysoftware.com>
Signed-off-by: Alan Cox <alan@linux.intel.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
None of these files are actually using any __init type directives
and hence don't need to include <linux/init.h>. Most are just a
left over from __devinit and __cpuinit removal, or simply due to
code getting copied from one driver to the next.
Signed-off-by: Paul Gortmaker <paul.gortmaker@windriver.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
tty flip buffers use GFP_ATOMIC allocations for received data
which is to be processed by the line discipline. For each byte
received, an extra byte is used to indicate the error status of
that byte.
Instead, if the received data is error-free, encode the entire
buffer without status bytes.
Signed-off-by: Peter Hurley <peter@hurleysoftware.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
Commit d7a68be4f2,
'tty: Only perform flip buffer flush from tty_buffer_flush()',
removed buffer flushing from flush_to_ldisc().
Fix function header comment which describes the former behavior.
Signed-off-by: Peter Hurley <peter@hurleysoftware.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
There is no in-tree user of tty_prepare_flip_string_flags(); remove.
Signed-off-by: Peter Hurley <peter@hurleysoftware.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
Trim up the memory_used field name to mem_used.
Signed-off-by: Peter Hurley <peter@hurleysoftware.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
Allow driver to configure its maximum flip buffer memory
consumption/limit. This is necessary for very-high speed line
rates (in excess of 10MB/sec) because the flip buffers can
be saturated before the line discipline has a chance to
throttle the input.
Signed-off-by: Peter Hurley <peter@hurleysoftware.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
TTY_BUFFER_PAGE is only used within drivers/tty/tty_buffer.c;
relocate to that file scope.
Signed-off-by: Peter Hurley <peter@hurleysoftware.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
Convert the tty_buffer_flush() exclusion mechanism to a
public interface - tty_buffer_lock/unlock_exclusive() - and use
the interface to safely write the paste selection to the line
discipline.
Signed-off-by: Peter Hurley <peter@hurleysoftware.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
__tty_flush_buffer() is now only called by tty_flush_buffer();
merge functions.
Signed-off-by: Peter Hurley <peter@hurleysoftware.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
Atomic bit ops are no longer required to indicate a flip buffer
flush is pending, as the flush_mutex is sufficient barrier.
Remove the unnecessary port .iflags field and localize flip buffer
state to struct tty_bufhead.
Signed-off-by: Peter Hurley <peter@hurleysoftware.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
Now that dropping the buffer lock is not necessary (as result of
converting the spin lock to a mutex), the flip buffer flush no
longer needs to be handled by the buffer work.
Simply signal a flush is required; the buffer work will exit the
i/o loop, which allows tty_buffer_flush() to proceed.
Signed-off-by: Peter Hurley <peter@hurleysoftware.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
The buffer work may race with parallel tty_buffer_flush. Use a
mutex to guarantee exclusive modify access to the head flip
buffer.
Remove the unneeded spin lock.
Signed-off-by: Peter Hurley <peter@hurleysoftware.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
Driver-side flip buffer input is already single-threaded; 'publish'
the .next link as the last operation on the tail buffer so the
'consumer' sees the already-completed flip buffer.
The commit buffer index is already 'published' by driver-side functions.
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>
Use a 0-sized sentinel to avoid assigning the head ptr from
the driver side thread. This also eliminates testing head/tail
for NULL.
When the sentinel is first 'consumed' by the buffer work
(or by tty_buffer_flush()), it is detached from the list but not
freed nor added to the free list. Both buffer work and
tty_buffer_flush() continue to preserve at least 1 flip buffer
to which head & tail is pointed.
Signed-off-by: Peter Hurley <peter@hurleysoftware.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
In preparation for lockless flip buffers, make the flip buffer
free list lockless.
NB: using llist is not the optimal solution, as the driver and
buffer work may contend over the llist head unnecessarily. However,
test measurements indicate this contention is low.
Signed-off-by: Peter Hurley <peter@hurleysoftware.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
tty_buffer_find() implements a simple free list lookaside cache.
Merge this functionality into tty_buffer_alloc() to reflect the
more traditional alloc/free symmetry.
Signed-off-by: Peter Hurley <peter@hurleysoftware.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
Since flip buffers are size-aligned to 256 bytes and all flip
buffers 512-bytes or larger are not added to the free list, the
free list only contains 256-byte flip buffers.
Remove the list search when allocating a new flip buffer.
Signed-off-by: Peter Hurley <peter@hurleysoftware.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
The char_buf_ptr and flag_buf_ptr values are trivially derived from
the .data field offset; compute values as needed.
Fixes a long-standing type-mismatch with the char and flag ptrs.
Signed-off-by: Peter Hurley <peter@hurleysoftware.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
Although line discipline receiving is single-producer/single-consumer,
using tty->receive_room to manage flow control creates unnecessary
critical regions requiring additional lock use.
Instead, introduce the optional .receive_buf2() ldisc method which
returns the # of bytes actually received. Serialization is guaranteed
by the caller.
In turn, the line discipline should schedule the buffer work item
whenever space becomes available; ie., when there is room to receive
data and receive_room() previously returned 0 (the buffer work
item stops processing if receive_buf2() returns 0). Note the
'no room' state need not be atomic despite concurrent use by two
threads because only the buffer work thread can set the state and
only the read() thread can clear the state.
Add n_tty_receive_buf2() as the receive_buf2() method for N_TTY.
Provide a public helper function, tty_ldisc_receive_buf(), to use
when directly accessing the receive_buf() methods.
Line disciplines not using input flow control can continue to set
tty->receive_room to a fixed value and only provide the receive_buf()
method.
Signed-off-by: Peter Hurley <peter@hurleysoftware.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
Ldisc interface functions must be called with interrupts enabled.
Separating the ldisc calls into a helper function simplies the
eventual removal of the spinlock.
Note that access to the buf->head ptr outside the spinlock is
safe here because;
* __tty_buffer_flush() is prevented from running while buffer work
performs i/o,
* tty_buffer_find() only assigns buf->head if the flip buffer list
is empty (which is never the case in flush_to_ldisc() since at
least one buffer is always left in the list after use)
Access to the read index outside the spinlock is safe here for the
same reasons.
Update the buffer's read index _after_ the data has been received
by the ldisc.
Signed-off-by: Peter Hurley <peter@hurleysoftware.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
Line discipline locking was performed with a combination of
a mutex, a status bit, a count, and a waitqueue -- basically,
a rw semaphore.
Replace the existing combination with an ld_semaphore.
Fixes:
1) the 'reference acquire after ldisc locked' bug
2) the over-complicated halt mechanism
3) lock order wrt. tty_lock()
4) dropping locks while changing ldisc
5) previously unidentified deadlock while locking ldisc from
both linked ttys concurrently
6) previously unidentified recursive deadlocks
Adds much-needed lockdep diagnostics.
Signed-off-by: Peter Hurley <peter@hurleysoftware.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
As Ilya Zykov identified in his patch 'PROBLEM: Race condition in
tty buffer's function flush_to_ldisc()', a race condition exists
which allows a parallel flush_to_ldisc() to flush and free the tty
flip buffers while those buffers are in-use. For example,
CPU 0 | CPU 1 | CPU 2
| flush_to_ldisc() |
| grab spin lock |
tty_buffer_flush() | | flush_to_ldisc()
wait for spin lock | | wait for spin lock
| if (!test_and_set_bit(TTYP_FLUSHING)) |
| while (next flip buffer) |
| ... |
| drop spin lock |
grab spin lock | |
if (test_bit(TTYP_FLUSHING)) | |
set_bit(TTYP_FLUSHPENDING) | receive_buf() |
drop spin lock | |
| | grab spin lock
| | if (!test_and_set_bit(TTYP_FLUSHING))
| | if (test_bit(TTYP_FLUSHPENDING))
| | __tty_buffer_flush()
CPU 2 has just flushed and freed all tty flip buffers while CPU 1 is
transferring data from the head flip buffer.
The original patch was rejected under the assumption that parallel
flush_to_ldisc() was not possible. Because of necessary changes to
the workqueue api, work items can execute in parallel on SMP.
This patch differs slightly from the original patch by testing for
a pending flush _after_ each receive_buf(), since TTYP_FLUSHPENDING
can only be set while the lock is dropped around receive_buf().
Reported-by: Ilya Zykov <linux@izyk.ru>
Signed-off-by: Peter Hurley <peter@hurleysoftware.com>
Acked-by: Ilya Zykov <linux@izyk.ru>
Cc: Jiri Slaby <jslaby@suse.cz>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
We added a warning to flush_to_ldisc to report cases when it is called
with a NULL tty. It was for debugging purposes and it lead to a
patchset from Peter Hurley. The patchset however did not make it to
3.9, so disable the warning now to not disturb people.
We can re-add it when the series is in and we are hunting for another
bugs.
Reported-by: David Miller <davem@davemloft.net>
Cc: stable <stable@vger.kernel.org> # 3.8
Signed-off-by: Jiri Slaby <jslaby@suse.cz>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
The PPS (Pulse-Per-Second) line discipline has developed a number of
unhealthy attachments to core tty data and functions, ultimately leading
to its breakage.
The previous patches fixed the crashing. This one reduces coupling further
by eliminating the timestamp parameter from the dcd_change ldisc method.
This reduces header file linkage and makes the extension more generic,
and the timestamp read is delayed only slightly, from just before the
ldisc->ops->dcd_change method call to just after.
Fix attendant build breakage in
drivers/tty/n_tty.c
drivers/tty/tty_buffer.c
drivers/staging/speakup/selection.c
drivers/staging/dgrp/dgrp_*.c
Cc: William Hubbs <w.d.hubbs@gmail.com>
Cc: Chris Brannon <chris@the-brannons.com>
Cc: Kirk Reiser <kirk@braille.uwo.ca>
Cc: Samuel Thibault <samuel.thibault@ens-lyon.org>
Signed-off-by: Peter Hurley <peter@hurleysoftware.com>
Signed-off-by: George Spelvin <linux@horizon.com>
Acked-by: Rodolfo Giometti <giometti@enneenne.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
The root of problem is carelessly zeroing pointer(in function __tty_buffer_flush()),
when another thread can use it. It can be cause of "NULL pointer dereference".
Main idea of the patch, this is never free last (struct tty_buffer) in the active buffer.
Only flush the data for ldisc(buf->head->read = buf->head->commit).
At that moment driver can collect(write) data in buffer without conflict.
It is repeat behavior of flush_to_ldisc(), only without feeding data to ldisc.
Also revert:
commit c56a00a165
tty: hold lock across tty buffer finding and buffer filling
In order to delete the unneeded locks any more.
Signed-off-by: Ilya Zykov <ilya@ilyx.ru>
CC: Alan Cox <alan@lxorguk.ukuu.org.uk>
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.
This is the last one: tty_schedule_flip
Signed-off-by: Jiri Slaby <jslaby@suse.cz>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
Brian Bloniarz