The locking in mesh_{mpath,mpp}_table_grow not only
has an rcu_read_unlock() missing, it's also racy
(though really only technically since it's invoked
from a single function only) since it obtains the
new size of the table without any locking, so two
invocations of the function could attempt the same
resize.
Additionally, it uses synchronize_rcu() which is
rather expensive and can be avoided trivially here.
Modify the functions to only use the table lock
and use call_rcu() instead of synchronize_rcu().
Signed-off-by: Johannes Berg <johannes.berg@intel.com>
Signed-off-by: John W. Linville <linville@tuxdriver.com>
Mesh paths are deleted via mesh_path_del() which properly
deactivates the timer associated to a mesh path. But if paths were
deleted by mesh_table_free(..., true) timers would not be deactivated.
This fixes this case.
Reported-by: Johannes Berg <johannes@sipsolutions.net>
Signed-off-by: Javier Cardona <javier@cozybit.com>
Signed-off-by: John W. Linville <linville@tuxdriver.com>
There's no need to have table functions in one
file and all users in another, move the functions
to the right file and make them static. Also move
a static variable to the beginning of the file to
make it easier to find.
Signed-off-by: Johannes Berg <johannes.berg@intel.com>
Signed-off-by: John W. Linville <linville@tuxdriver.com>
The mesh and mpp path tables are accessed from softirq and workqueue
context so non-irq locking cannot be used. Or at least that's what
PROVE_RCU seems to tell us here:
[ 431.240946] =================================
[ 431.241061] [ INFO: inconsistent lock state ]
[ 431.241061] 2.6.39-rc3-wl+ #354
[ 431.241061] ---------------------------------
[ 431.241061] inconsistent {IN-SOFTIRQ-W} -> {SOFTIRQ-ON-W} usage.
[ 431.241061] kworker/u:1/1423 [HC0[0]:SC0[0]:HE1:SE1] takes:
[ 431.241061] (&(&newtbl->hashwlock[i])->rlock){+.?...}, at:
[<c14671bf>] mesh_path_add+0x167/0x257
Signed-off-by: Javier Cardona <javier@cozybit.com>
Signed-off-by: John W. Linville <linville@tuxdriver.com>
Not sure if I'm chasing a ghost here, seems like the
mesh_path->size_order needs to be inside an RCU-read section to prevent
that value from changing between table allocation and copying. We have
observed crashes that might be caused by this.
Signed-off-by: Javier Cardona <javier@cozybit.com>
Signed-off-by: John W. Linville <linville@tuxdriver.com>
The TTL in path selection information elements is different from
the mesh ttl used in mesh data frames. Version 7.03 of the 11s
draft calls this ttl 'Element TTL'.
Signed-off-by: Johannes Berg <johannes.berg@intel.com>
Signed-off-by: John W. Linville <linville@tuxdriver.com>
IBSS, managed and mesh modes all have their
own work struct, and in the future we want
to also use it in other modes to process
frames from the now common skb queue.
This also makes the skb queue and work safe
to use from other interface types.
Signed-off-by: Johannes Berg <johannes.berg@intel.com>
Signed-off-by: John W. Linville <linville@tuxdriver.com>
percpu.h is included by sched.h and module.h and thus ends up being
included when building most .c files. percpu.h includes slab.h which
in turn includes gfp.h making everything defined by the two files
universally available and complicating inclusion dependencies.
percpu.h -> slab.h dependency is about to be removed. Prepare for
this change by updating users of gfp and slab facilities include those
headers directly instead of assuming availability. As this conversion
needs to touch large number of source files, the following script is
used as the basis of conversion.
http://userweb.kernel.org/~tj/misc/slabh-sweep.py
The script does the followings.
* Scan files for gfp and slab usages and update includes such that
only the necessary includes are there. ie. if only gfp is used,
gfp.h, if slab is used, slab.h.
* When the script inserts a new include, it looks at the include
blocks and try to put the new include such that its order conforms
to its surrounding. It's put in the include block which contains
core kernel includes, in the same order that the rest are ordered -
alphabetical, Christmas tree, rev-Xmas-tree or at the end if there
doesn't seem to be any matching order.
* If the script can't find a place to put a new include (mostly
because the file doesn't have fitting include block), it prints out
an error message indicating which .h file needs to be added to the
file.
The conversion was done in the following steps.
1. The initial automatic conversion of all .c files updated slightly
over 4000 files, deleting around 700 includes and adding ~480 gfp.h
and ~3000 slab.h inclusions. The script emitted errors for ~400
files.
2. Each error was manually checked. Some didn't need the inclusion,
some needed manual addition while adding it to implementation .h or
embedding .c file was more appropriate for others. This step added
inclusions to around 150 files.
3. The script was run again and the output was compared to the edits
from #2 to make sure no file was left behind.
4. Several build tests were done and a couple of problems were fixed.
e.g. lib/decompress_*.c used malloc/free() wrappers around slab
APIs requiring slab.h to be added manually.
5. The script was run on all .h files but without automatically
editing them as sprinkling gfp.h and slab.h inclusions around .h
files could easily lead to inclusion dependency hell. Most gfp.h
inclusion directives were ignored as stuff from gfp.h was usually
wildly available and often used in preprocessor macros. Each
slab.h inclusion directive was examined and added manually as
necessary.
6. percpu.h was updated not to include slab.h.
7. Build test were done on the following configurations and failures
were fixed. CONFIG_GCOV_KERNEL was turned off for all tests (as my
distributed build env didn't work with gcov compiles) and a few
more options had to be turned off depending on archs to make things
build (like ipr on powerpc/64 which failed due to missing writeq).
* x86 and x86_64 UP and SMP allmodconfig and a custom test config.
* powerpc and powerpc64 SMP allmodconfig
* sparc and sparc64 SMP allmodconfig
* ia64 SMP allmodconfig
* s390 SMP allmodconfig
* alpha SMP allmodconfig
* um on x86_64 SMP allmodconfig
8. percpu.h modifications were reverted so that it could be applied as
a separate patch and serve as bisection point.
Given the fact that I had only a couple of failures from tests on step
6, I'm fairly confident about the coverage of this conversion patch.
If there is a breakage, it's likely to be something in one of the arch
headers which should be easily discoverable easily on most builds of
the specific arch.
Signed-off-by: Tejun Heo <tj@kernel.org>
Guess-its-ok-by: Christoph Lameter <cl@linux-foundation.org>
Cc: Ingo Molnar <mingo@redhat.com>
Cc: Lee Schermerhorn <Lee.Schermerhorn@hp.com>
For bluetooth 3, we will most likely not have
a netdev for a virtual interface (sdata), so
prepare for that by reducing the reliance on
having a netdev. This patch moves the name
and address fields into the sdata struct and
uses them from there all over. Some work is
needed to keep them sync'ed, but that's not
a lot of work and in slow paths anyway.
In doing so, this also reduces the number of
pointer dereferences in many places, because
of things like sdata->dev->dev_addr becoming
sdata->vif.addr.
Signed-off-by: Johannes Berg <johannes@sipsolutions.net>
Signed-off-by: John W. Linville <linville@tuxdriver.com>
The netdev broadcast address cannot change from
all-ones so there's no need to use it; we can
instead hard-code it. Since we already have an
instance in tkip.c, which will be shared if it
is marked static const, doing this reduces text
size at no data/bss cost.
The real motivation for this is, of course, the
desire to get rid of almost all uses of netdevs
in mac80211 so that auditing their use becomes
easier.
Signed-off-by: Johannes Berg <johannes@sipsolutions.net>
Signed-off-by: John W. Linville <linville@tuxdriver.com>
Signed-off-by: Rui Paulo <rpaulo@gmail.com>
Signed-off-by: Javier Cardona <javier@cozybit.com>
Reviewed-by: Andrey Yurovsky <andrey@cozybit.com>
Tested-by: Brian Cavagnolo <brian@cozybit.com>
Signed-off-by: John W. Linville <linville@tuxdriver.com>
Resulting object files have the same MD5 as before.
Signed-off-by: Rui Paulo <rpaulo@gmail.com>
Signed-off-by: Javier Cardona <javier@cozybit.com>
Reviewed-by: Andrey Yurovsky <andrey@cozybit.com>
Tested-by: Brian Cavagnolo <brian@cozybit.com>
Signed-off-by: John W. Linville <linville@tuxdriver.com>
Update the PERR IE frame format according to latest draft (3.03).
Signed-off-by: Rui Paulo <rpaulo@gmail.com>
Signed-off-by: Javier Cardona <javier@cozybit.com>
Reviewed-by: Andrey Yurovsky <andrey@cozybit.com>
Tested-by: Brian Cavagnolo <brian@cozybit.com>
Signed-off-by: John W. Linville <linville@tuxdriver.com>
This prevents calling rcu_synchronize from within the tx path by moving the
table growth code to the mesh workqueue.
Move mesh_table_free and mesh_table_grow from mesh.c to mesh_pathtbl.c and
declare them static.
Also, re-enable mesh in Kconfig and update the configuration description.
Signed-off-by: Javier Cardona <javier@cozybit.com>
Tested-by: Andrey Yurovsky <andrey@cozybit.com>
Signed-off-by: John W. Linville <linville@tuxdriver.com>
In order for userspace to be able to figure out whether
it obtained a consistent snapshot of data or not when
using netlink dumps, we need to have a generation number
in each dump message that indicates whether the list has
changed or not -- its value is arbitrary.
This patch adds such a number to all dumps, this needs
some mac80211 involvement to keep track of a generation
number to start with when adding/removing mesh paths or
stations.
The wiphy and netdev lists can be fully handled within
cfg80211, of course, but generation numbers need to be
stored there as well.
Signed-off-by: Johannes Berg <johannes@sipsolutions.net>
Signed-off-by: John W. Linville <linville@tuxdriver.com>
The removal of the master netdev broke the mesh forwarding path. This patch
fixes it by using the new internal 'pending' queue.
As a result of this change, mesh forwarding no longer does the inefficient
802.11 -> 802.3 -> 802.11 conversion that was done before.
[Changes since v1]
Suggested by Johannes:
- Select queue before adding to mpath queue
- ieee80211_add_pending_skb -> ieee80211_add_pending_skbs
- Remove unnecessary header wme.h
Signed-off-by: Javier Cardona <javier@cozybit.com>
Signed-off-by: Andrey Yurovsky <andrey@cozybit.com>
Reviewed-by: Johannes Berg <johannes@sipsolutions.net>
Signed-off-by: John W. Linville <linville@tuxdriver.com>
Assign next hop address to pending mesh frames once the path is resolved.
Regression. Frames transmitted when a mesh path was wating to be resolved were
being transmitted with an invalid Receiver Address.
[Changes since v1]
Suggested by Johannes:
- Improved frame_queue traversal
- Narower RCU scope
Signed-off-by: Javier Cardona <javier@cozybit.com>
Signed-off-by: Andrey Yurovsky <andrey@cozybit.com>
Signed-off-by: John W. Linville <linville@tuxdriver.com>
For forwarded frames, we save the precursor address in addr1 in case it
needs to be used to send a Path Error. mesh_path_discard_frame,
however, was using addr2 instead of addr1 to send Path Error frames, so
correct that and also make the comment regarding this more clear.
Signed-off-by: Andrey Yurovsky <andrey@cozybit.com>
Signed-off-by: John W. Linville <linville@tuxdriver.com>
My kvm instance was complaining a lot about sleeping
in atomic contexts in the mesh code, and it turns out
that both mesh_path_add() and mpp_path_add() need to
be able to sleep (they even use synchronize_rcu()!).
I put in a might_sleep() to annotate that, but I see
no way, at least right now, of actually making sure
those functions are only called from process context
since they are both called during TX and RX and the
mesh code itself even calls them with rcu_read_lock()
"held".
Therefore, let's disable it completely for now.
It's possible that I'm only seeing this because the
hwsim's beaconing is broken and thus the peers aren't
discovered right away, but it is possible that this
happens even if beaconing is working, for a peer that
doesn't exist or so.
It should be possible to solve this by deferring the
freeing of the tables to call_rcu() instead of using
synchronize_rcu(), and also using atomic allocations,
but maybe it makes more sense to rework the code to
not call these from atomic contexts and defer more of
the work to the workqueue. Right now, I can't work on
either of those solutions though.
Signed-off-by: Johannes Berg <johannes@sipsolutions.net>
Signed-off-by: John W. Linville <linville@tuxdriver.com>
Currently the mesh code doesn't support bridging mesh point interfaces
with wired ethernet or AP to construct an MPP or MAP. This patch adds
code to support the "6 address frame format packet" functionality to
mesh point interfaces. Now the mesh network can be used as backhaul
for end to end communication.
Signed-off-by: Li YanBo <dreamfly281@gmail.com>
Signed-off-by: John W. Linville <linville@tuxdriver.com>
This patch splits off mesh handling from the STA/IBSS.
Unfortunately it increases mesh code size a bit, but I
think it makes things clearer. The patch also reduces
per-interface run-time memory usage.
Also clean up a few places where ifdef is not required.
Signed-off-by: Johannes Berg <johannes@sipsolutions.net>
Signed-off-by: John W. Linville <linville@tuxdriver.com>
This patch replaces net_device arguments to mac80211 internal functions
with ieee80211_{local,sub_if_data} as appropriate.
It also does the same for many 802.11s mesh functions, and changes the
mesh path table to be indexed on sub_if_data rather than net_device.
If the mesh part needs to be a separate patch let me know, but since
mesh uses a lot of mac80211 functions which were being converted anyway,
the changes go hand-in-hand somewhat.
This patch probably does not convert all the functions which could be
converted, but it is a large chunk and followup patches will be
provided.
Signed-off-by: Jasper Bryant-Greene <jasper@amiton.co.nz>
Signed-off-by: John W. Linville <linville@tuxdriver.com>
Now we deal with mesh forwarding before the 802.11->802.3 conversion, thus
eliminating a few unnecessary steps. The next hop lookup is called from
ieee80211_master_start_xmit() instead of subif_start_xmit(). Until the next hop
is found, RA in the frame will be all zeroes for frames originating from the
device. For forwarded frames, RA will contain the TA of the received frame,
which will be necessary to send a path error if a next hop is not found.
Signed-off-by: Luis Carlos Cobo <luisca@cozybit.com>
Acked-by: Johannes Berg <johannes@sipsolutions.net>
Signed-off-by: John W. Linville <linville@tuxdriver.com>
The old infrastructure was:
- the default algorithm is built into mac80211
- other algorithms get into their own modules
The implementation of this complicated scheme was horrible
(just look at net/mac80211/Makefile), and anyone adding a new
algorithm would most likely not get it right at his first attempt.
This patch therefore builds all enabled algorithms into the mac80211
module.
The user interface for the rate control algorithms changes as follows:
- first the user can choose which algorithms to enable (currently only
MAC80211_RC_PID is available)
- if more than one algorithm is enabled (currently not possible since
only one algorithm is present) the user then chooses the default one
Note:
- MAC80211_RC_PID is always enables for CONFIG_EMBEDDED=n
Technical changes:
- all selected algorithms get into the mac80211 module
- net/mac80211/Makefile can now become much less complicated
- support for rc80211_pid_algo.c being modular is no longer required
- this includes unexporting mesh_plink_broken
Signed-off-by: Adrian Bunk <bunk@kernel.org>
Signed-off-by: John W. Linville <linville@tuxdriver.com>
In case the hash grow failed, it is not fair to return error -
the new node _was_ _actually_ added in this case.
Besides, after my previous patch, this grow is more likely
to fail on large hashes.
Signed-off-by: Pavel Emelyanov <xemul@openvz.org>
Signed-off-by: John W. Linville <linville@tuxdriver.com>
The mesh_path_node_copy() can be called like this:
mesh_path_add
`- write_lock(&pathtbl_resize_lock); /* ! */
`- mesh_table_grow
`- ->copy_node
`- mesh_path_node_copy
thus, the GFP_KERNEL is not suitable here.
The acceptable fix, I suppose, is make this allocation GPF_ATOMIC -
the mpath_node being allocated is 4 pointers, i.e. this allocation
is small enough to survive even under a moderate memory pressure.
Signed-off-by: Pavel Emelyanov <xemul@openvz.org>
Signed-off-by: John W. Linville <linville@tuxdriver.com>
Now - return the -ENOMEM in case kmalloc fails.
Signed-off-by: Pavel Emelyanov <xemul@openvz.org>
Signed-off-by: John W. Linville <linville@tuxdriver.com>
The mesh_path_node_copy() performs kmalloc() and thus - may fail
(well, it does not now, but I'm fixing this right now). Its caller -
the mesh_table_grow() - isn't prepared for such a trick yet.
This preparation is just flush the new hash and make copy_node()
return an int value.
Signed-off-by: Pavel Emelyanov <xemul@openvz.org>
Signed-off-by: John W. Linville <linville@tuxdriver.com>
(This set applies OK without the previous one of 4 patches,
but with some fuzz in the 7th one)
The mesh_path_node_free() does so under hashwlock.
But, this one is called
1. from mesh_path_add() after an old hash is hidden and
synchronize_rcu() is calld
2. mesh_pathtbl_unregister(), when the module is being
unloaded and no devices exist to mess with this hash.
So, it seems to me, that simply removing the call is OK.
Signed-off-by: Pavel Emelyanov <xemul@openvz.org>
Signed-off-by: John W. Linville <linville@tuxdriver.com>
There are already tree paths, that do incremental rollbacks, so
merge them together, rename labels and format the code to look a
bit nicer.
(I do not mind dropping/delaying this patch however).
Signed-off-by: Pavel Emelyanov <xemul@openvz.org>
Signed-off-by: John W. Linville <linville@tuxdriver.com>
Calling synchronize_rcu() under write-lock-ed pathtbl_resize_lock may
result in this warning (and other side effects).
It looks safe just dropping this lock before calling synchronize_rcu.
Signed-off-by: Pavel Emelyanov <xemul@openvz.org>
Signed-off-by: John W. Linville <linville@tuxdriver.com>
The new_node kmallocation is not checked for success, so add
this check.
BTW, it also happens under the read_lock.
Signed-off-by: Pavel Emelyanov <xemul@openvz.org>
Signed-off-by: John W. Linville <linville@tuxdriver.com>
The mesh_path_add() read-locks the pathtbl_resize_lock and calls
kmalloc with GFP_KERNEL mask.
Fix it and move the endadd2 label lower. It should be _before_ the
if() beyond, but it makes no sense for it being there, so I move it
right after this if().
Signed-off-by: Pavel Emelyanov <xemul@openvz.org>
Signed-off-by: John W. Linville <linville@tuxdriver.com>
Skip properly entries whose dev does not match.
Signed-off-by: Luis Carlos Cobo <luisca@cozybit.com>
Acked-by: Johannes Berg <johannes@sipsolutions.net>
Signed-off-by: John W. Linville <linville@tuxdriver.com>
Signed-off-by: Luis Carlos Cobo <luisca@cozybit.com>
Acked-by: Johannes Berg <johannes@sipsolutions.net>
Signed-off-by: John W. Linville <linville@tuxdriver.com>
Postponing the deletion is not really useful anymore.
Signed-off-by: Luis Carlos Cobo <luisca@cozybit.com>
Acked-by: Johannes Berg <johannes@sipsolutions.net>
Signed-off-by: John W. Linville <linville@tuxdriver.com>
This avoids dereferencing a no longer existing struct mesh_path.
Signed-off-by: Luis Carlos Cobo <luisca@cozybit.com>
Acked-by: Johannes Berg <johannes@sipsolutions.net>
Signed-off-by: John W. Linville <linville@tuxdriver.com>
Luis pointed out that this path is going to be freed right
away anyway so there's no point in assigning next_hop.
Signed-off-by: Johannes Berg <johannes@sipsolutions.net>
Cc: Luis Carlos Cobo <luisca@cozybit.com>
Signed-off-by: John W. Linville <linville@tuxdriver.com>
This makes access to the STA hash table/list use RCU to protect
against freeing of items. However, it's not a true RCU, the
copy step is missing: whenever somebody changes a STA item it
is simply updated. This is an existing race condition that is
now somewhat understandable.
This patch also fixes the race key freeing vs. STA destruction
by making sure that sta_info_destroy() is always called under
RTNL and frees the key.
Signed-off-by: Johannes Berg <johannes@sipsolutions.net>
Signed-off-by: John W. Linville <linville@tuxdriver.com>
This needs to be exported because rate control algorithms
can be modular.
Signed-off-by: Johannes Berg <johannes@sipsolutions.net>
Signed-off-by: John W. Linville <linville@tuxdriver.com>
The mesh path table associates destinations with the next hop to reach them. The
table is a hash of linked lists protected by rcu mechanisms. Every mesh path
contains a lock to protect the mesh path state.
Each outgoing mesh frame requires a look up into this table. Therefore, the
table it has been designed so it is not necessary to hold any lock to find the
appropriate next hop.
If the path is determined to be active within a rcu context we can safely
dereference mpath->next_hop->addr, since it holds a reference to the sta
next_hop. After a mesh path has been set active for the first time it next_hop
must always point to a valid sta. If this is not possible the mpath must be
deleted or replaced in a RCU safe fashion.
Signed-off-by: Luis Carlos Cobo <luisca@cozybit.com>
Signed-off-by: Johannes Berg <johannes@sipsolutions.net>
Signed-off-by: John W. Linville <linville@tuxdriver.com>
Johannes Berg
Javier Cardona
John W. Linville
Lucas De Marchi
Tejun Heo
David S. Miller
Johannes Berg
Linus Torvalds
André Goddard Rosa
Rui Paulo
YanBo
Jasper Bryant-Greene
Luis Carlos Cobo
Adrian Bunk
Pavel Emelyanov