In commit (d56c0d45f0e27 "proc: decouple proc from VFS with "struct proc_ops"")
[1], proc_create_data no longer takes a file_operations structure, but instead
takes a struct proc_ops in order to conserve memory in the kernel.
This change is necessary for DAHDI to work with kernels >= 5.6
[1] https://git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git/commit/?id=d56c0d45f0e27f814e87a1676b6bd
Signed-off-by: Shaun Ruffell <sruffell@sruffell.net>
Quiets warnings like:
drivers/dahdi/dahdi-base.c:8494:6: warning: this statement may fall through [-Wimplicit-fallthrough=]
if (chan->txstate != DAHDI_TXSTATE_OFFHOOK) break;
^
Needed since Linux 5.4 commit (a035d552a93bb9ef60487, "Makefile: Globally enable
fall-33bb9f2a0dc857ff869through warning") [1]
[1] https://git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git/commit/?id=a035d552a93bb9ef6048733bb9f2a0dc857
Signed-off-by: Shaun Ruffell <sruffell@sruffell.net>
Ideally we want to standardize on storing all timestamps derivied from the
system clock as ktime_t values.
Signed-off-by: Shaun Ruffell <sruffell@sruffell.net>
Since this is only used internally, and ktime is the basis for
timekeeping in the kernel, this allows this interface to be validated,
before converting the other internal timekeeping to it.
This is part of the changes necessary to remove the use of 'struct
timeval' from the driver suite for compatibility with Linux 5.0, which
is updating the internal timekeeping interfaces to fix the 2038 problem.
Signed-off-by: Shaun Ruffell <sruffell@sruffell.net>
There are not any major distributions that are still supporting kernels
older than 2.6.27 so we can remove many typedefs. The primary motivator
for this change is that kernel 5.0 is dropping support for timeval and
it would be ideal if the in-kernel time representation can
standardize on ktime_t, but 2.6.18 did not support the ktime
interface that was needed.
Signed-off-by: Shaun Ruffell <sruffell@sruffell.net>
Upstream kernel 4.14, in commit (686fef928bba6b "timer: Prepare to change timer
callback argument type") [1], introduced the timer_setup interface to replace
the init_timer/setup_timer interfaces. The primary change is that the timer
callback functions now follow the standard kernel pattern where the structure
the callback sits in is passed to the callback instead of storing a pointer to
an unassociated data type.
The setup_timer functions were removed in upstream kernel v4.15, and therefore
this change is needed in order to compile DAHDI for kernels >= 4.15.
This change follows the same strategy that was done in the kernel to while the
existing users of setup_timer were migrated to the new interface.
[1] https://git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git/commit/?id=686fef928bba6b
This fixes an error and quiets the following warning pointed out by gcc 7.3.0:
warning: ‘memset’ used with length equal to number of elements without
multiplication by element size [-Wmemset-elt-size] memset(chan->conflast, 0,
DAHDI_MAX_CHUNKSIZE);
Previously only the first half of the conference buffers were cleared out.
In case a user has a file descriptor that is used for non-blocking I/O
and the span to which it belongs gets unassigned, its reads and writes
will still yield -EAGAIN.
This commit adds a test for such a case and returns the proper -ENODEV.
Signed-off-by: Tzafrir Cohen <tzafrir.cohen@xorcom.com>
Upstream kernel 4.11, in commit (bd0f9b356d00aa241ced36fb075a07041c28d3b8
"sched/headers: fix up header file dependency on <linux/sched/signal.h>"), now
requires users of signal_pending to include the new linux/sched/signal.h file.
Signed-off-by: Shaun Ruffell <sruffell@digium.com>
Kernel version 3.16+, since upstream commit (7f7f25e82d54870d "replace checking
for ->read/->aio_read presence with check in ->f_mode" )[1], does not like it
when dahdi changes the set of allowed file operations on a file descriptor
outside of the context of an open() system call.
DAHDI changes the available file operations when a channel is opened by first
opening /dev/dahdi/channel and then calling the DAHDI_SPECIFY ioctl to bind it
to a particular DAHDI channel. Until DAHDI_SPECIFY is called there weren't any
read()/write() callbacks implemented and therefore after the initial open, the
kernel was setting not setting FMODE_CAN_{WRITE,READ} on the file descriptor
indicating that those operations were not allowed.
Now define empty shell functions on the general dahdi_fops so the vfs layer will
not mark a file descriptor as unwritteable or unreadable on open.
[1] https://git.kernel.org/cgit/linux/kernel/git/torvalds/linux.git/commit/?id=7f7f25e82d54870df24d415a7007fbd327da027b
Internal-Issue-ID: DAHLIN-340
Reported-and-tested-by: Thomas B. Clark
Signed-off-by: Shaun Ruffell <sruffell@digium.com>
Signed-off-by: Russ Meyerriecks <rmeyerriecks@digium.com>
If a channel is currently playing a tone when the tone zone is updated, the
existing tone zone could be freed while the channel keeps a reference to the
current tone (curtone) that points into the freed zone.
If the newly freed tone is then modified, there was a window where it was
possible to corrupt 'struct dahdi_chan' (by overrunning swritechunk[])
resulting in a "BUG: unable to handle kernel paging request at virtual address"
panic in the context of __dahdi_transmit_chunk().
Reported-and-Tested-by: Matt Behrens <matt@zigg.com>
Signed-off-by: Shaun Ruffell <sruffell@digium.com>
Signed-off-by: Russ Meyerriecks <rmeyerriecks@digium.com>
This makes the open symmetrical with the close. It is also considered good
practice to not call through callbacks with spinlocks held.
I modified all the drivers where I could not tell whether it was necessary to
hold the chan->lock with interrupts disabled to simply take the lock inside the
callback.
Signed-off-by: Shaun Ruffell <sruffell@digium.com>
Signed-off-by: Russ Meyerriecks <rmeyerriecks@digium.com>
This is never accessed or modified in interrupt context. This closes a potential
race if the echocan is being changed on a channel while enabling disabling is
hapening on another thread.
Signed-off-by: Shaun Ruffell <sruffell@digium.com>
Signed-off-by: Russ Meyerriecks <rmeyerriecks@digium.com>
This closes a reference and memory leak when multiple CPUs are enabling echocan
on a single channel in parallel.
The essential problem is that the call to try_module_get() is not serialized.
Two separate threads can come into ioctl_echocan() on the same channel, they
coordinate via the dahdi_chan.lock to release any current echocan, but then both
create a new echocan state, bump the reference on the module, and the last one
through will actually attach the new state to the channel. The earlier reference
/ memory is leaked.
I tried to conceive of a way to fix this leak without adding a new lock, but the
choices where calling throught the function pointers with dahdi_chan.lock.
Otherwise I needed to change the semantics of echocan_create /free which would
ripple through the hardware echocan modules.
Signed-off-by: Shaun Ruffell <sruffell@digium.com>
Signed-off-by: Russ Meyerriecks <rmeyerriecks@digium.com>
Some architectures, like arm, do not automatically pull in the definitions for
kzalloc and friends. This allows DAHDI to build on those platforms.
Originally reported to the asterisk-users mailing list here
http://lists.digium.com/pipermail/asterisk-users/2014-February/282338.html
Signed-off-by: Shaun Ruffell <sruffell@digium.com>
Acked-by: Tzafrir Cohen <tzafrir.cohen@xorcom.com>
Signed-off-by: Russ Meyerriecks <rmeyerriecks@digium.com>
Commit 74e949c33a exported the module
variable dahdi.auto_assigned_spans. However this is not a good name.
This commit reverts the export and replaces it with a new function to
get the value of the variable.
Signed-off-by: Tzafrir Cohen <tzafrir.cohen@xorcom.com>
Acked-by: Russ Meyerriecks <rmeyerriecks@digium.com>
A fix to 03b3ce1a10: use ktime_get_ts
instead of getnstimeofday to better handle system time changes.
(Shaun Ruffell)
Signed-off-by: Tzafrir Cohen <tzafrir.cohen@xorcom.com>
Add a new sysfs attribute to dahdi_device: registration_time
* Records the time of the device's registration with DAHDI.
* Used by dahdi_auto_assign_compat to assign spans by device
registration order when backward compatibility needed.
Signed-off-by: Tzafrir Cohen <tzafrir.cohen@xorcom.com>
Acked-by: Shaun Ruffell <sruffell@digium.com>
Instead, use the inverse of dahdi.auto_assign_spans value.
Signed-off-by: Tzafrir Cohen <tzafrir.cohen@xorcom.com>
Acked-by: Russ Meyerriecks <rmeyerriecks@digium.com>
The infrastructure has been put in place in 2.8.0 for fully dynamic device and
span configuration. This will be the default mode in DAHDI-Linux 2.9.0.
Signed-off-by: Shaun Ruffell <sruffell@digium.com>
Acked-by: Tzafrir Cohen <tzafrir.cohen@xorcom.com>
Acked-by: Russ Meyerriecks <rmeyerriecks@digium.com>
Acked-by: Oron Peled <oron.peled@xorcom.com>
Previously this was in __dahdi_init_span(). The problem was that
__dahdi_init_span() was only called when a spans' line mode was being changed.
Therefore it was possible to unassign and resassign an analog span and leave it
stuck in the 'NOP' alarm state.
It also make the setting / clearing of DAHDI_ALARM_NOTOPEN symetrical about span
unassignment / assignment in addition to updating the alarm states on all the
channels on the span via the dahdi_alarm_notify() function.
This is a better version of commit 496f817773.
Signed-off-by: Shaun Ruffell <sruffell@digium.com>
Fixes the following warning when loading the driver:
dahdi: Warning: Span DYN/eth/eth1/00:50:c2:97:92:1d/1 didn't specify a spantype. Please fix driver!
The spantype is intended to be used by the auto configuration tools
(dahdi_genconf, pinned spans, etc..) but dynamic spans, since they are created
directly by dahdi_cfg, never take part in the pre-registration auto
configuration, therefore the span type was never used.
Signed-off-by: Shaun Ruffell <sruffell@digium.com>
* This changeset adds master_span as an attribute of the span's driver:
/sys/bus/dahdi_spans/drivers/generic_lowlevel/master_span
* This is mainly used for debugging.
* Reading from it the master span number (or 0 if there is no master
span).
* Writing a number to it, force specified span to be master
* Existing Alternatives:
- grep "MASTER" from /proc/dahdi/*
- cat /sys/bus/dahdi_spans/devices/span-*/is_sync_master
(Note: commit d8fe2af23d is also Acked-By
Tzafrir Cohen <tzafrir.cohen@xorcom.com>)
Signed-off-by: Tzafrir Cohen <tzafrir.cohen@xorcom.com>
Acked-by: Russ Meyerriecks <rmeyerriecks@digium.com>
* No functional changes.
* Rename the span 'master' to 'master_span' to avoid ambiguity with the
'master' channel.
Signed-off-by: Tzafrir Cohen <tzafrir.cohen@xorcom.com>
Acked-by: Russ Meyerriecks <rmeyerriecks@digium.com>
This eliminates the need for board drivers to always re-report their alarm
states when spans are unassigned and then reassigned.
Signed-off-by: Shaun Ruffell <sruffell@digium.com>
This fixes a regression introduced in 2.7.0 in commit
(da8b96d725 "dahdi: Remove unused 'rxbufpolicy'
and 'rxdisable' from dahdi_chan.") when CONFIG_DAHDI_NET is defined in
include/dahdi/dahdi_config.h.
rxbufpolicy was always hardcoded to immediate policy and was removed from the
channel structure. There is no longer any need to set it in dahdi_net_open.
Reported-by: Dave Fullerton <dfullertasterisk@shorelinecontainer.com>
Signed-off-by: Shaun Ruffell <sruffell@digium.com>
Signed-off-by: Russ Meyerriecks <rmeyerriecks@digium.com>
create_proc_entry() was deprecated and replace with proc_create_data() since
it's open to a race condition where the proc entry is visible before the file
operations have been set for it.
The PDE() macro also is no longer available as of Linux 3.10 and is replaced
with PDE_DATA() to get the data member from a proc entry. This is due to the
fact that 'struct proc_dir_entry' is now private to the proc_fs.
This commit changes the core of DAHDI and also introduces proc_create_data() and
PDE_DATA() for older kernels.
Signed-off-by: Shaun Ruffell <sruffell@digium.com>
Acked-by: Tzafrir Cohen <tzafrir.cohen@xorcom.com>
Cc: Russ Meyerriecks <rmeyerriecks@digium.com>
Cc: Oron Peled <oron.peled@xorcom.com>
When spans are unassigned, dahdi_span_ops.shutdown was called, but the RUNNING
flag was never cleared. Now make sure all calls to the shutdown span ops
callback are the same.
Signed-off-by: Shaun Ruffell <sruffell@digium.com>
Signed-off-by: Russ Meyerriecks <rmeyerriecks@digium.com>
If you have a multiple span system configured for t1 mode, and you try to change
only the first span from t1 to e1 via sysfs, you could get an error in the
kernel log about trying to create duplicate channel.
The problem is that the check for whether there was enough room for the all the
channels of the switched span was wrong.
Reported-by: James Brown <jbrown@digium.com>
Signed-off-by: Shaun Ruffell <sruffell@digium.com>
Signed-off-by: Russ Meyerriecks <rmeyerriecks@digium.com>
Some of the card callbacks needed maintstat set when processing loopup codes
from the remote side. This change is a modification of commit: (6c02c3c
"dahdi-base: Minor maint mode error")
Without this change cetain cards would not properly process the loopup /
loopdown codes (I know I should have more specific information here but the
details escape me at the moment).
Signed-off-by: Shaun Ruffell <sruffell@digium.com>
An unset conditional compile flag was triggering the unused variable compile
warning. Added the condition around the variable define.
Signed-off-by: Russ Meyerriecks <rmeyerriecks@digium.com>
Otherwise, it's possible for a link to remain in use if a process quits without
unlinking the conferences. Now when all the channels are removed from the
conference, any links are also cleaned up.
Signed-off-by: Shaun Ruffell <sruffell@digium.com>
This is mostly a revert of commit r9463. If you need to use DAHDI_CONFLINK
ioctl, make sure to define CONFIG_DAHDI_CONFLINK in
include/dahdi/dahdi_config.h. Apparently there were some users of CONFLINK out
there still.
It's a compile time option now since most users won't need to run the test for
conflinks in the hot-path that is the process_masterspan function.
Signed-off-by: Shaun Ruffell <sruffell@digium.com>
Tested-by: Ted Gerold <ted@twg.org>
Trades the memory of a file_operations structure to eliminate some tests in the
timer operations.
Internal-Issue-ID: ABE-2904
Signed-off-by: Shaun Ruffell <sruffell@digium.com>
On SMP systems there isn't a need for all dahdi timers to
synchronize on dahdi_timer_lock. Instead give each timer it's own
lock to synchronize with process_timers and use dahdi_timer_lock to
protect the global list of locks.
Also, the dahdi_timers list now only contains timers that are set to
fire. This eliminates a test for each timer in the hot path of the
process_masterspan context.
Reduces system load on many-core systems which are heavy users of
DAHDI timers.
Internal-Issue-ID: ABE-2904
Signed-off-by: Shaun Ruffell <sruffell@digium.com>
The Big Kernel Lock removal is no longer experimental so we can eliminate the
call to lock_kernel when calling unlock_ioctl.
Internal-Issue-ID: ABE-2904
Signed-off-by: Shaun Ruffell <sruffell@digium.com>
This allows timingslips to be reset along with the other counters and clarifies
the intended use.
This came up when Doug Bailey asked why he couldn't use dahdi_maint to clear
timing slips in addition to the other counters.
Signed-off-by: Shaun Ruffell <sruffell@digium.com>
This resolves an issue with a Glenayre GL3000 Paging Terminal with v8.000
software. The Glenayre would change both A and B bits to signal on/off hook
states, but there were a couple of milliseconds between when those bits changed.
This resulted in DAHDI generating an on-hook event to Asterisk before generating
the off-hook event and therefore Asterisk terminated the call prematurely.
Looking the A and B bits before this patch:
Asterisk (AB) Glenayre (AB)
00 00 Both sides on-hook
11 00 Asterisk goes off hook to sieze line.
11 01 Glenayre starts going-offhook. On-hook event
sent to Asterisk from drivers since bits
changed but ABIT was still 0.
11 11 Glenayre finishes going off-hook. Off-hook
event sent to Asterisk since ABIT changed
from 0 to 1.
00 11 Asterisk, processes on-hook event and goes
on-hook itself to release line.
00 00 Glenayre releases line. Call fails.
After this patch:
Asterisk (AB) Glenayre (AB)
00 00 Both sides on-hook
11 00 Asterisk initiates call
11 01 Glenayre starts handling call. No event is
sent to Asterisk since only ABIT is checked.
11 11 Glanayre finishes going off-hook. Call
proceeds normally.
Internal-Issue-ID: DAHDI-1009
Signed-off-by: Shaun Ruffell <sruffell@digium.com>
Signed-off-by: Russ Meyerriecks <rmeyerriecks@digium.com>
Since r5021 [1], first released in DAHDI-Linux 2.2.0, it's been impossible for
user space to change the rxbufpolicy. This hasn't caused any problems and it's
safe to remove a few more of the vestiges of the rxbufpolicy from the driver.
This streamlines the code path in a few places and saves 8 bytes from the size
of struct dahdi_chan.
The user visible parts are maintained and will indicate
DAHDI_POLICY_IMMEDIATE, like it has since 2.2.0.
[1] http://svnview.digium.com/svn/dahdi?view=revision&revision=5021
Signed-off-by: Shaun Ruffell <sruffell@digium.com>
Signed-off-by: Russ Meyerriecks <rmeyerriecks@digium.com>
git-svn-id: http://svn.asterisk.org/svn/dahdi/linux/trunk@10730 a0bf4364-ded3-4de4-8d8a-66a801d63aff
Internally in DAHDI there is always a hardware echocan factory registered and
available. Having this factory always registered allows for DAHDI to work in a
backward compatible fashion. Namely, by default DAHDI will always use a hardware
echocan if one is available unless 'hwec_overrides_swec' dahdi module parameter
is set to 0 on load. However, if there were no real hardware echocans available
in the system dahdi would still report "Echo Canceller(s): HWEC" in the
dahdi_cfg -v output since the hwec factory is always there.
After this change dahdi_cfg will no longer report HWEC as one of the available
echocans if there isn't a physical span present that actually has a hardware
echocan.
Internal-Issue-ID: DAHLIN-300
Signed-off-by: Shaun Ruffell <sruffell@digium.com>
Signed-off-by: Russ Meyerriecks <rmeyerriecks@digium.com>
git-svn-id: http://svn.asterisk.org/svn/dahdi/linux/trunk@10727 a0bf4364-ded3-4de4-8d8a-66a801d63aff
The previous maint state was saved regardless if the base driver returned an
error or not. This caused strange behavior in dahdi tools. Moved the maint
state save to after the switch case to reflect this.
https://issues.asterisk.org/jira/browse/DAHDI-984
Signed-off-by: Russ Meyerriecks <rmeyerriecks@digium.com>
git-svn-id: http://svn.asterisk.org/svn/dahdi/linux/trunk@10718 a0bf4364-ded3-4de4-8d8a-66a801d63aff
