Skip to content

cloud-kernel
cloud-kernel

openanolis / cloud-kernel
大约 1 年 前同步成功

通知 156
Star 36
Fork 7
cloud-kernel
cloud-kernel

体验新版 GitCode，发现更多精彩内容 >>

切换分支/标签
  1. 29 1月, 2016 3 次提交
    • P
      n_tty: Fix stuck write wakeup · 7bccc365
      由 Peter Hurley 提交于 
      If signal-driven i/o is disabled while write wakeup is pending (ie.,
n_tty_write() has set TTY_DO_WRITE_WAKEUP but then signal-driven i/o
is disabled), the TTY_DO_WRITE_WAKEUP bit will never be cleared and
will cause tty_wakeup() to always call n_tty_write_wakeup.

Unconditionally clear the write wakeup, and since kill_fasync()
already checks if the fasync ptr is null, call kill_fasync()
unconditionally as well.
Signed-off-by: NPeter Hurley <peter@hurleysoftware.com>
Signed-off-by: NGreg Kroah-Hartman <gregkh@linuxfoundation.org>
      7bccc365
    • P
      tty, n_tty: Remove fasync() ldisc notification · bee6741c
      由 Peter Hurley 提交于 
      Only the N_TTY line discipline implements the signal-driven i/o
notification enabled/disabled by fcntl(F_SETFL, O_ASYNC). The ldisc
fasync() notification is sent to the ldisc when the enable state has
changed (the tty core is notified via the fasync() VFS file operation).

The N_TTY line discipline used the enable state to change the wakeup
condition (minimum_to_wake = 1) for notifying the signal handler i/o is
available. However, just the presence of data is sufficient and necessary
to signal i/o is available, so changing minimum_to_wake is unnecessary
(and creates a race condition with read() and poll() which may be
concurrently updating minimum_to_wake).

Furthermore, since the kill_fasync() VFS helper performs no action if
the fasync list is empty, calling unconditionally is preferred; if
signal driven i/o just has been disabled, no signal will be sent by
kill_fasync() anyway so notification of the change via the ldisc
fasync() method is superfluous.
Signed-off-by: NPeter Hurley <peter@hurleysoftware.com>
Signed-off-by: NGreg Kroah-Hartman <gregkh@linuxfoundation.org>
      bee6741c
    • P
      n_tty: Always wake up read()/poll() if new input · 33d71363
      由 Peter Hurley 提交于 
      A read() in non-canonical mode when VMIN > 0 and VTIME == 0 does not
complete until at least VMIN chars have been read (or the user buffer is
full). In this infrequent read mode, n_tty_read() attempts to reduce
wakeups by computing the amount of data still necessary to complete the
read (minimum_to_wake) and only waking the read()/poll() when that much
unread data has been processed. This is the only read mode for which
new data does not necessarily generate a wakeup.

However, this optimization is broken and commonly leads to hung reads
even though the necessary amount of data has been received. Since the
optimization is of marginal value anyway, just remove the whole
thing. This also remedies a race between a concurrent poll() and
read() in this mode, where the poll() can reset the minimum_to_wake
of the read() (and vice versa).
Signed-off-by: NPeter Hurley <peter@hurleysoftware.com>
Signed-off-by: NGreg Kroah-Hartman <gregkh@linuxfoundation.org>
      33d71363
  3. 28 1月, 2016 6 次提交
  5. 27 1月, 2016 1 次提交
  7. 14 12月, 2015 4 次提交
  9. 13 12月, 2015 1 次提交
    • P
      n_tty: Fix poll() after buffer-limited eof push read · ac8f3bf8
      由 Peter Hurley 提交于 
      commit 40d5e090 ("n_tty: Fix EOF push handling") fixed EOF push
for reads. However, that approach still allows a condition mismatch
between poll() and read(), where poll() returns POLLIN but read()
blocks. This state can happen when a previous read() returned because
the user buffer was full and the next character was an EOF not at the
beginning of the line. While the next read() will properly identify
the condition and advance the read buffer tail without improperly
indicating an EOF file condition (ie., read() will not mistakenly
return 0), poll() will mistakenly indicate POLLIN.

Although a possible solution would be to peek at the input buffer
in n_tty_poll(), the better solution in this patch is to eat the
EOF during the previous read() (ie., fix the problem by eliminating
the condition).

The current canon line buffer copy limits the scan for next end-of-line
to the smaller of either,
   a. the remaining user buffer size
   b. completed lines in the input buffer
When the remaining user buffer size is exactly one less than the
end-of-line marked by EOF push, the EOF is not scanned nor skipped
but left for subsequent reads. In the example below, the scan
index 'eol' has stopped at the EOF because it is past the scan
limit of 5 (not because it has found the next set bit in read_flags)

   user buffer [*nr = 5]    _ _ _ _ _

   read_flags               0 0 0 0 0   1
   input buffer             h e l l o [EOF]
                            ^           ^
                           /           /
                         tail        eol

   result: found = 0, tail += 5, *nr += 5

Instead, allow the scan to peek ahead 1 byte (while still limiting the
scan to completed lines in the input buffer). For the example above,

   result: found = 1, tail += 6, *nr += 5

Because the scan limit is now bumped +1 byte, when the scan is
completed, the tail advance and the user buffer copy limit is
re-clamped to *nr when EOF is _not_ found.

Fixes: 40d5e090 ("n_tty: Fix EOF push handling")
Cc: <stable@vger.kernel.org> # 3.12+
Signed-off-by: NPeter Hurley <peter@hurleysoftware.com>
Signed-off-by: NGreg Kroah-Hartman <gregkh@linuxfoundation.org>
      ac8f3bf8
  11. 21 11月, 2015 1 次提交
  13. 18 10月, 2015 3 次提交
  15. 05 10月, 2015 1 次提交
    • K
      tty: fix stall caused by missing memory barrier in drivers/tty/n_tty.c · e81107d4
      由 Kosuke Tatsukawa 提交于 
      My colleague ran into a program stall on a x86_64 server, where
n_tty_read() was waiting for data even if there was data in the buffer
in the pty.  kernel stack for the stuck process looks like below.
 #0 [ffff88303d107b58] __schedule at ffffffff815c4b20
 #1 [ffff88303d107bd0] schedule at ffffffff815c513e
 #2 [ffff88303d107bf0] schedule_timeout at ffffffff815c7818
 #3 [ffff88303d107ca0] wait_woken at ffffffff81096bd2
 #4 [ffff88303d107ce0] n_tty_read at ffffffff8136fa23
 #5 [ffff88303d107dd0] tty_read at ffffffff81368013
 #6 [ffff88303d107e20] __vfs_read at ffffffff811a3704
 #7 [ffff88303d107ec0] vfs_read at ffffffff811a3a57
 #8 [ffff88303d107f00] sys_read at ffffffff811a4306
 #9 [ffff88303d107f50] entry_SYSCALL_64_fastpath at ffffffff815c86d7

There seems to be two problems causing this issue.

First, in drivers/tty/n_tty.c, __receive_buf() stores the data and
updates ldata->commit_head using smp_store_release() and then checks
the wait queue using waitqueue_active().  However, since there is no
memory barrier, __receive_buf() could return without calling
wake_up_interactive_poll(), and at the same time, n_tty_read() could
start to wait in wait_woken() as in the following chart.

        __receive_buf()                         n_tty_read()
------------------------------------------------------------------------
if (waitqueue_active(&tty->read_wait))
/* Memory operations issued after the
   RELEASE may be completed before the
   RELEASE operation has completed */
                                        add_wait_queue(&tty->read_wait, &wait);
                                        ...
                                        if (!input_available_p(tty, 0)) {
smp_store_release(&ldata->commit_head,
                  ldata->read_head);
                                        ...
                                        timeout = wait_woken(&wait,
                                          TASK_INTERRUPTIBLE, timeout);
------------------------------------------------------------------------

The second problem is that n_tty_read() also lacks a memory barrier
call and could also cause __receive_buf() to return without calling
wake_up_interactive_poll(), and n_tty_read() to wait in wait_woken()
as in the chart below.

        __receive_buf()                         n_tty_read()
------------------------------------------------------------------------
                                        spin_lock_irqsave(&q->lock, flags);
                                        /* from add_wait_queue() */
                                        ...
                                        if (!input_available_p(tty, 0)) {
                                        /* Memory operations issued after the
                                           RELEASE may be completed before the
                                           RELEASE operation has completed */
smp_store_release(&ldata->commit_head,
                  ldata->read_head);
if (waitqueue_active(&tty->read_wait))
                                        __add_wait_queue(q, wait);
                                        spin_unlock_irqrestore(&q->lock,flags);
                                        /* from add_wait_queue() */
                                        ...
                                        timeout = wait_woken(&wait,
                                          TASK_INTERRUPTIBLE, timeout);
------------------------------------------------------------------------

There are also other places in drivers/tty/n_tty.c which have similar
calls to waitqueue_active(), so instead of adding many memory barrier
calls, this patch simply removes the call to waitqueue_active(),
leaving just wake_up*() behind.

This fixes both problems because, even though the memory access before
or after the spinlocks in both wake_up*() and add_wait_queue() can
sneak into the critical section, it cannot go past it and the critical
section assures that they will be serialized (please see "INTER-CPU
ACQUIRING BARRIER EFFECTS" in Documentation/memory-barriers.txt for a
better explanation).  Moreover, the resulting code is much simpler.

Latency measurement using a ping-pong test over a pty doesn't show any
visible performance drop.
Signed-off-by: NKosuke Tatsukawa <tatsu@ab.jp.nec.com>
Cc: stable@vger.kernel.org
Signed-off-by: NGreg Kroah-Hartman <gregkh@linuxfoundation.org>
      e81107d4
  17. 24 7月, 2015 2 次提交
  19. 01 6月, 2015 1 次提交
  21. 25 5月, 2015 1 次提交
  23. 11 5月, 2015 1 次提交
    • P
      pty: Fix input race when closing · 1a48632f
      由 Peter Hurley 提交于 
      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: 52bce7f8 ("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: NAndy Whitcroft <apw@canonical.com>
Reported-by: NH.J. Lu <hjl.tools@gmail.com>
Signed-off-by: NPeter Hurley <peter@hurleysoftware.com>
Signed-off-by: NGreg Kroah-Hartman <gregkh@linuxfoundation.org>
      1a48632f
  25. 07 5月, 2015 1 次提交
  27. 03 2月, 2015 7 次提交
    • P
      n_tty: Fix signal handling flushes · d2b6f447
      由 Peter Hurley 提交于 
      BRKINT and ISIG requires input and output flush when a signal char
is received. However, the order of operations is significant since
parallel i/o may be ongoing.

Merge the signal handling for BRKINT with ISIG handling.

Process the signal first. This ensures any ongoing i/o is aborted;
without this, a waiting writer may continue writing after the flush
occurs and after the signal char has been echoed.

Write lock the termios_rwsem, which excludes parallel writers from
pushing new i/o until after the output buffers are flushed; claiming
the write lock is necessary anyway to exclude parallel readers while
the read buffer is flushed.

Subclass the termios_rwsem for ptys since the slave pty performing
the flush may appear to reorder the termios_rwsem->tty buffer lock
lock order; adding annotation clarifies that
  slave tty_buffer lock-> slave termios_rwsem -> master tty_buffer lock
is a valid lock order.

Flush the echo buffer. In this context, the echo buffer is 'output'.
Otherwise, the output will appear discontinuous because the output buffer
was cleared which contains older output than the echo buffer.

Open-code the read buffer flush since the input worker does not need
kicking (this is the input worker).
Signed-off-by: NPeter Hurley <peter@hurleysoftware.com>
Signed-off-by: NGreg Kroah-Hartman <gregkh@linuxfoundation.org>
      d2b6f447
    • P
      n_tty: Fix read buffer overwrite when no newline · fb5ef9e7
      由 Peter Hurley 提交于 
      In canon mode, the read buffer head will advance over the buffer tail
if the input > 4095 bytes without receiving a line termination char.

Discard additional input until a line termination is received.
Before evaluating for overflow, the 'room' value is normalized for
I_PARMRK and 1 byte is reserved for line termination (even in !icanon
mode, in case the mode is switched). The following table shows the
transform:

 actual buffer |  'room' value before overflow calc
  space avail  |    !I_PARMRK    |    I_PARMRK
 --------------------------------------------------
      0        |       -1        |       -1
      1        |        0        |        0
      2        |        1        |        0
      3        |        2        |        0
      4+       |        3        |        1

When !icanon or when icanon and the read buffer contains newlines,
normalized 'room' values of -1 and 0 are clamped to 0, and
'overflow' is 0, so read_head is not adjusted and the input i/o loop
exits (setting no_room if called from flush_to_ldisc()). No input
is discarded since the reader does have input available to read
which ensures forward progress.

When icanon and the read buffer does not contain newlines and the
normalized 'room' value is 0, then overflow and room are reset to 1,
so that the i/o loop will process the next input char normally
(except for parity errors which are ignored). Thus, erasures, signalling
chars, 7-bit mode, etc. will continue to be handled properly.

If the input char processed was not a line termination char, then
the canon_head index will not have advanced, so the normalized 'room'
value will now be -1 and 'overflow' will be set, which indicates the
read_head can safely be reset, effectively erasing the last char
processed.

If the input char processed was a line termination, then the
canon_head index will have advanced, so 'overflow' is cleared to 0,
the read_head is not reset, and 'room' is cleared to 0, which exits
the i/o loop (because the reader now have input available to read
which ensures forward progress).

Note that it is possible for a line termination to be received, and
for the reader to copy the line to the user buffer before the
input i/o loop is ready to process the next input char. This is
why the i/o loop recomputes the room/overflow state with every
input char while handling overflow.

Finally, if the input data was processed without receiving
a line termination (so that overflow is still set), the pty
driver must receive a write wakeup. A pty writer may be waiting
to write more data in n_tty_write() but without unthrottling
here that wakeup will not arrive, and forward progress will halt.
(Normally, the pty writer is woken when the reader reads data out
of the buffer and more space become available).
Signed-off-by: NPeter Hurley <peter@hurleysoftware.com>
Signed-off-by: NGreg Kroah-Hartman <gregkh@linuxfoundation.org>
      fb5ef9e7
    • P
      n_tty: Fix PARMRK over-throttling · 06c49f9f
      由 Peter Hurley 提交于 
      If PARMRK is enabled, the available read buffer space computation is
overly-pessimistic, which results in severely throttled i/o, even
in the absence of parity errors. For example, if the 4k read buffer
contains 1k processed data, the input worker will compute available
space of 333 bytes, despite 3k being available. At 1365 chars of
processed data, 0 space available is computed.

*Divide remaining space* by 3, truncating down (if left == 2, left = 0).
Reported-by: NChristian Riesch <christian.riesch@omicron.at>

Conflicts:
	drivers/tty/n_tty.c
Signed-off-by: NPeter Hurley <peter@hurleysoftware.com>
Signed-off-by: NGreg Kroah-Hartman <gregkh@linuxfoundation.org>
      06c49f9f
    • P
      n_tty: Fix unordered accesses to lockless read buffer · 70aca71f
      由 Peter Hurley 提交于 
      Add commit_head buffer index, which the producer-side publishes
after input processing in non-canon mode. This ensures the consumer-side
observes correctly-ordered writes in non-canonical mode (ie., the buffer
data is written before the buffer index is advanced). Fix consumer-side
uses of read_cnt() to use commit_head instead.

Add required memory barriers to the tail index to guarantee
the consumer-side has completed the loads before the producer-side
begins writing new data. Open-code the producer-side receive_room()
into the i/o loop.

Remove no-longer-referenced receive_room().

Based on work by Christian Riesch <christian.riesch@omicron.at>

Cc: Christian Riesch <christian.riesch@omicron.at>
Signed-off-by: NPeter Hurley <peter@hurleysoftware.com>
Signed-off-by: NGreg Kroah-Hartman <gregkh@linuxfoundation.org>
      70aca71f
    • P
      n_tty: Simplify throttle threshold calculation · 5e28cca1
      由 Peter Hurley 提交于 
      The adjustments performed by receive_room() are to ensure a line
termination can always be written to the read buffer. However,
these adjustments are irrelevant to the throttle threshold (because
the threshold < buffer limit).
Signed-off-by: NPeter Hurley <peter@hurleysoftware.com>
Signed-off-by: NGreg Kroah-Hartman <gregkh@linuxfoundation.org>
      5e28cca1
    • P
      n_tty: Fix throttle for canon lines > 3967 chars · a342846f
      由 Peter Hurley 提交于 
      The tty driver will be mistakenly throttled if a line termination
has not been received, and the line exceeds 3967 chars. Thus, it is
possible for the driver to stop sending when it has not yet sent
the newline. This does not apply to the pty driver.

Don't throttle until at least one line termination has been
received.
Signed-off-by: NPeter Hurley <peter@hurleysoftware.com>
Signed-off-by: NGreg Kroah-Hartman <gregkh@linuxfoundation.org>
      a342846f
    • P
      n_tty: Eliminate receive_room() from consumer/exclusive paths · 2c5dc464
      由 Peter Hurley 提交于 
      The input worker never reschedules itself; it only processes input until
either there is no more input or the read buffer is full. So the reader
is responsible for restarting the input worker only if the read buffer
was previously full (no_room == 1) _and_ space is now available to process
more input because the reader has consumed data from the read buffer.

However, computing the actual space available is not required to determine
if the reader has consumed data from the read buffer. This condition is
evaluated in 5 situations, each of which the space avail is already known:
1. n_tty_flush_buffer() - the read buffer is empty; kick the worker
2. n_tty_set_termios() - no data has been consumed; do not kick the worker
       (although it may have kicked the reader so data _will be_ consumed)
3. n_tty_check_unthrottle - avail space > 3968; kick the worker
4. n_tty_read, before leaving - only kick the worker if the reader has
       moved the tail. This prevents unnecessarily kicking the worker
       when timeout-style reading is used.
5. n_tty_read, before sleeping - although it is possible for the read
       buffer to be full and input_available_p() to be false, this can
       only happen when the input worker is racing the reader, in which
       case the reader will have been woken and won't sleep.

Rename n_tty_set_room() to n_tty_kick_worker() to reflect what the
function actually does.
Signed-off-by: NPeter Hurley <peter@hurleysoftware.com>
Signed-off-by: NGreg Kroah-Hartman <gregkh@linuxfoundation.org>
      2c5dc464
  29. 10 1月, 2015 1 次提交
  31. 27 11月, 2014 1 次提交
    • C
      n_tty: Fix read_buf race condition, increment read_head after pushing data · 8bfbe2de
      由 Christian Riesch 提交于 
      Commit 19e2ad6a ("n_tty: Remove overflow
tests from receive_buf() path") moved the increment of read_head into
the arguments list of read_buf_addr(). Function calls represent a
sequence point in C. Therefore read_head is incremented before the
character c is placed in the buffer. Since the circular read buffer is
a lock-less design since commit 6d76bd26
("n_tty: Make N_TTY ldisc receive path lockless"), this creates a race
condition that leads to communication errors.

This patch modifies the code to increment read_head _after_ the data
is placed in the buffer and thus fixes the race for non-SMP machines.
To fix the problem for SMP machines, memory barriers must be added in
a separate patch.
Signed-off-by: NChristian Riesch <christian.riesch@omicron.at>
Cc: <stable@vger.kernel.org>
Signed-off-by: NGreg Kroah-Hartman <gregkh@linuxfoundation.org>
      8bfbe2de
  33. 07 11月, 2014 1 次提交
    • F
      tty: Fix pty master poll() after slave closes v2 · c4dc3046
      由 Francesco Ruggeri 提交于 
      Commit f95499c3 ("n_tty: Don't wait for buffer work in read() loop")
introduces a race window where a pty master can be signalled that the pty
slave was closed before all the data that the slave wrote is delivered.
Commit f8747d4a ("tty: Fix pty master read() after slave closes") fixed the
problem in case of n_tty_read, but the problem still exists for n_tty_poll.
This can be seen by running 'for ((i=0; i<100;i++));do ./test.py ;done'
where test.py is:

import os, select, pty

(pid, pty_fd) = pty.fork()

if pid == 0:
   os.write(1, 'This string should be received by parent')
else:
   poller = select.epoll()
   poller.register( pty_fd, select.EPOLLIN )
   ready = poller.poll( 1 * 1000 )
   for fd, events in ready:
      if not events & select.EPOLLIN:
         print 'missed POLLIN event'
      else:
         print os.read(fd, 100)
   poller.close()

The string from the slave is missed several times.
This patch takes the same approach as the fix for read and special cases
this condition for poll.
Tested on 3.16.
Signed-off-by: NFrancesco Ruggeri <fruggeri@arista.com>
Signed-off-by: NGreg Kroah-Hartman <gregkh@linuxfoundation.org>
      c4dc3046
  35. 06 11月, 2014 4 次提交