Commit 323ebb61e3 ("net: use listified RX for handling GRO_NORMAL
skbs") introduces batching of GRO_NORMAL packets in napi_frags_finish,
and commit 6570bc79c0 ("net: core: use listified Rx for GRO_NORMAL in
napi_gro_receive()") adds the same to napi_skb_finish. However,
dev_gro_receive (that is called just before napi_{frags,skb}_finish) can
also pass skbs to the networking stack: e.g., when the GRO session is
flushed, napi_gro_complete is called, which passes pp directly to
netif_receive_skb_internal, skipping napi->rx_list. It means that the
packet stored in pp will be handled by the stack earlier than the
packets that arrived before, but are still waiting in napi->rx_list. It
leads to TCP reorderings that can be observed in the TCPOFOQueue counter
in netstat.
This commit fixes the reordering issue by making napi_gro_complete also
use napi->rx_list, so that all packets going through GRO will keep their
order. In order to keep napi_gro_flush working properly, gro_normal_list
calls are moved after the flush to clear napi->rx_list.
iwlwifi calls napi_gro_flush directly and does the same thing that is
done by gro_normal_list, so the same change is applied there:
napi_gro_flush is moved to be before the flush of napi->rx_list.
A few other drivers also use napi_gro_flush (brocade/bna/bnad.c,
cortina/gemini.c, hisilicon/hns3/hns3_enet.c). The first two also use
napi_complete_done afterwards, which performs the gro_normal_list flush,
so they are fine. The latter calls napi_gro_receive right after
napi_gro_flush, so it can end up with non-empty napi->rx_list anyway.
Fixes: 323ebb61e3 ("net: use listified RX for handling GRO_NORMAL skbs")
Signed-off-by: Maxim Mikityanskiy <maximmi@mellanox.com>
Cc: Alexander Lobakin <alobakin@dlink.ru>
Cc: Edward Cree <ecree@solarflare.com>
Acked-by: Alexander Lobakin <alobakin@dlink.ru>
Acked-by: Saeed Mahameed <saeedm@mellanox.com>
Acked-by: Edward Cree <ecree@solarflare.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Netdev_register_kobject is calling device_initialize. In case of error
reference taken by device_initialize is not given up.
Drivers are supposed to call free_netdev in case of error. In non-error
case the last reference is given up there and device release sequence
is triggered. In error case this reference is kept and the release
sequence is never started.
Fix this by setting reg_state as NETREG_UNREGISTERED if registering
fails.
This is the rootcause for couple of memory leaks reported by Syzkaller:
BUG: memory leak unreferenced object 0xffff8880675ca008 (size 256):
comm "netdev_register", pid 281, jiffies 4294696663 (age 6.808s)
hex dump (first 32 bytes):
00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 ................
00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 ................
backtrace:
[<0000000058ca4711>] kmem_cache_alloc_trace+0x167/0x280
[<000000002340019b>] device_add+0x882/0x1750
[<000000001d588c3a>] netdev_register_kobject+0x128/0x380
[<0000000011ef5535>] register_netdevice+0xa1b/0xf00
[<000000007fcf1c99>] __tun_chr_ioctl+0x20d5/0x3dd0
[<000000006a5b7b2b>] tun_chr_ioctl+0x2f/0x40
[<00000000f30f834a>] do_vfs_ioctl+0x1c7/0x1510
[<00000000fba062ea>] ksys_ioctl+0x99/0xb0
[<00000000b1c1b8d2>] __x64_sys_ioctl+0x78/0xb0
[<00000000984cabb9>] do_syscall_64+0x16f/0x580
[<000000000bde033d>] entry_SYSCALL_64_after_hwframe+0x44/0xa9
[<00000000e6ca2d9f>] 0xffffffffffffffff
BUG: memory leak
unreferenced object 0xffff8880668ba588 (size 8):
comm "kobject_set_nam", pid 286, jiffies 4294725297 (age 9.871s)
hex dump (first 8 bytes):
6e 72 30 00 cc be df 2b nr0....+
backtrace:
[<00000000a322332a>] __kmalloc_track_caller+0x16e/0x290
[<00000000236fd26b>] kstrdup+0x3e/0x70
[<00000000dd4a2815>] kstrdup_const+0x3e/0x50
[<0000000049a377fc>] kvasprintf_const+0x10e/0x160
[<00000000627fc711>] kobject_set_name_vargs+0x5b/0x140
[<0000000019eeab06>] dev_set_name+0xc0/0xf0
[<0000000069cb12bc>] netdev_register_kobject+0xc8/0x320
[<00000000f2e83732>] register_netdevice+0xa1b/0xf00
[<000000009e1f57cc>] __tun_chr_ioctl+0x20d5/0x3dd0
[<000000009c560784>] tun_chr_ioctl+0x2f/0x40
[<000000000d759e02>] do_vfs_ioctl+0x1c7/0x1510
[<00000000351d7c31>] ksys_ioctl+0x99/0xb0
[<000000008390040a>] __x64_sys_ioctl+0x78/0xb0
[<0000000052d196b7>] do_syscall_64+0x16f/0x580
[<0000000019af9236>] entry_SYSCALL_64_after_hwframe+0x44/0xa9
[<00000000bc384531>] 0xffffffffffffffff
v3 -> v4:
Set reg_state to NETREG_UNREGISTERED if registering fails
v2 -> v3:
* Replaced BUG_ON with WARN_ON in free_netdev and netdev_release
v1 -> v2:
* Relying on driver calling free_netdev rather than calling
put_device directly in error path
Reported-by: syzbot+ad8ca40ecd77896d51e2@syzkaller.appspotmail.com
Cc: David Miller <davem@davemloft.net>
Cc: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
Cc: Lukas Bulwahn <lukas.bulwahn@gmail.com>
Signed-off-by: Jouni Hogander <jouni.hogander@unikie.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
syzbot reported some bogus lockdep warnings, for example bad unlock
balance in sch_direct_xmit(). They are due to a race condition between
slow path and fast path, that is qdisc_xmit_lock_key gets re-registered
in netdev_update_lockdep_key() on slow path, while we could still
acquire the queue->_xmit_lock on fast path in this small window:
CPU A CPU B
__netif_tx_lock();
lockdep_unregister_key(qdisc_xmit_lock_key);
__netif_tx_unlock();
lockdep_register_key(qdisc_xmit_lock_key);
In fact, unlike the addr_list_lock which has to be reordered when
the master/slave device relationship changes, queue->_xmit_lock is
only acquired on fast path and only when NETIF_F_LLTX is not set,
so there is likely no nested locking for it.
Therefore, we can just get rid of re-registration of
qdisc_xmit_lock_key.
Reported-by: syzbot+4ec99438ed7450da6272@syzkaller.appspotmail.com
Fixes: ab92d68fc2 ("net: core: add generic lockdep keys")
Cc: Taehee Yoo <ap420073@gmail.com>
Signed-off-by: Cong Wang <xiyou.wangcong@gmail.com>
Acked-by: Taehee Yoo <ap420073@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Replace all the occurrences of FIELD_SIZEOF() with sizeof_field() except
at places where these are defined. Later patches will remove the unused
definition of FIELD_SIZEOF().
This patch is generated using following script:
EXCLUDE_FILES="include/linux/stddef.h|include/linux/kernel.h"
git grep -l -e "\bFIELD_SIZEOF\b" | while read file;
do
if [[ "$file" =~ $EXCLUDE_FILES ]]; then
continue
fi
sed -i -e 's/\bFIELD_SIZEOF\b/sizeof_field/g' $file;
done
Signed-off-by: Pankaj Bharadiya <pankaj.laxminarayan.bharadiya@intel.com>
Link: https://lore.kernel.org/r/20190924105839.110713-3-pankaj.laxminarayan.bharadiya@intel.com
Co-developed-by: Kees Cook <keescook@chromium.org>
Signed-off-by: Kees Cook <keescook@chromium.org>
Acked-by: David Miller <davem@davemloft.net> # for net
Pull RCU updates from Ingo Molnar:
"The main changes in this cycle were:
- Dynamic tick (nohz) updates, perhaps most notably changes to force
the tick on when needed due to lengthy in-kernel execution on CPUs
on which RCU is waiting.
- Linux-kernel memory consistency model updates.
- Replace rcu_swap_protected() with rcu_prepace_pointer().
- Torture-test updates.
- Documentation updates.
- Miscellaneous fixes"
* 'core-rcu-for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/tip/tip: (51 commits)
security/safesetid: Replace rcu_swap_protected() with rcu_replace_pointer()
net/sched: Replace rcu_swap_protected() with rcu_replace_pointer()
net/netfilter: Replace rcu_swap_protected() with rcu_replace_pointer()
net/core: Replace rcu_swap_protected() with rcu_replace_pointer()
bpf/cgroup: Replace rcu_swap_protected() with rcu_replace_pointer()
fs/afs: Replace rcu_swap_protected() with rcu_replace_pointer()
drivers/scsi: Replace rcu_swap_protected() with rcu_replace_pointer()
drm/i915: Replace rcu_swap_protected() with rcu_replace_pointer()
x86/kvm/pmu: Replace rcu_swap_protected() with rcu_replace_pointer()
rcu: Upgrade rcu_swap_protected() to rcu_replace_pointer()
rcu: Suppress levelspread uninitialized messages
rcu: Fix uninitialized variable in nocb_gp_wait()
rcu: Update descriptions for rcu_future_grace_period tracepoint
rcu: Update descriptions for rcu_nocb_wake tracepoint
rcu: Remove obsolete descriptions for rcu_barrier tracepoint
rcu: Ensure that ->rcu_urgent_qs is set before resched IPI
workqueue: Convert for_each_wq to use built-in list check
rcu: Several rcu_segcblist functions can be static
rcu: Remove unused function hlist_bl_del_init_rcu()
Documentation: Rename rcu_node_context_switch() to rcu_note_context_switch()
...
Commit 78d3fd0b7d ("gro: Only use skb_gro_header for completely
non-linear packets") back in May'09 (v2.6.31-rc1) has changed the
original condition '!skb_headlen(skb)' to
'skb->mac_header == skb->tail' in gro_reset_offset() saying: "Since
the drivers that need this optimisation all provide completely
non-linear packets" (note that this condition has become the current
'skb_mac_header(skb) == skb_tail_pointer(skb)' later with commmit
ced14f6804 ("net: Correct comparisons and calculations using
skb->tail and skb-transport_header") without any functional changes).
For now, we have the following rough statistics for v5.4-rc7:
1) napi_gro_frags: 14
2) napi_gro_receive with skb->head containing (most of) payload: 83
3) napi_gro_receive with skb->head containing all the headers: 20
4) napi_gro_receive with skb->head containing only Ethernet header: 2
With the current condition, fast GRO with the usage of
NAPI_GRO_CB(skb)->frag0 is available only in the [1] case.
Packets pushed by [2] and [3] go through the 'slow' path, but
it's not a problem for them as they already contain all the needed
headers in skb->head, so pskb_may_pull() only moves skb->data.
The layout of skbs in the fourth [4] case at the moment of
dev_gro_receive() is identical to skbs that have come through [1],
as napi_frags_skb() pulls Ethernet header to skb->head. The only
difference is that the mentioned condition is always false for them,
because skb_put() and friends irreversibly alter the tail pointer.
They also go through the 'slow' path, but now every single
pskb_may_pull() in every single .gro_receive() will call the *really*
slow __pskb_pull_tail() to pull headers to head. This significantly
decreases the overall performance for no visible reasons.
The only two users of method [4] is:
* drivers/staging/qlge
* drivers/net/wireless/iwlwifi (all three variants: dvm, mvm, mvm-mq)
Note that in case with wireless drivers we can't use [1]
(napi_gro_frags()) at least for now and mac80211 stack always
performs pushes and pulls anyways, so performance hit is inavoidable.
At the moment of v2.6.31 the mentioned change was necessary (that's
why I don't add the "Fixes:" tag), but it became obsolete since
skb_gro_mac_header() has gone in commit a50e233c50 ("net-gro:
restore frag0 optimization"), so we can simply revert the condition
in gro_reset_offset() to allow skbs from [4] go through the 'fast'
path just like in case [1].
This was tested on a 600 MHz MIPS CPU and a custom driver and this
patch gave boosts up to 40 Mbps to method [4] in both directions
comparing to net-next, which made overall performance relatively
close to [1] (without it, [4] is the slowest).
v2:
- Add more references and explanations to commit message
- Fix some typos ibid
- No functional changes
Signed-off-by: Alexander Lobakin <alobakin@dlink.ru>
Signed-off-by: David S. Miller <davem@davemloft.net>
The only slightly tricky merge conflict was the netdevsim because the
mutex locking fix overlapped a lot of driver reload reorganization.
The rest were (relatively) trivial in nature.
Signed-off-by: David S. Miller <davem@davemloft.net>
When netdevice with offloaded BPF programs is destroyed
the programs are orphaned and removed from the program
IDA - their IDs get released (the programs may remain
accessible via existing open file descriptors and pinned
files). After IDs are released they are set to 0.
This confuses dev_change_xdp_fd() because it compares
the __dev_xdp_query() result where 0 means no program
with prog->aux->id where 0 means orphaned.
dev_change_xdp_fd() would have incorrectly returned success
even though it had not installed the program.
Since drivers already catch this case via bpf_offload_dev_match()
let them handle this case. The error message drivers produce in
this case ("program loaded for a different device") is in fact
correct as the orphaned program must had to be loaded for a
different device.
Fixes: c14a9f633d ("net: Don't call XDP_SETUP_PROG when nothing is changed")
Signed-off-by: Jakub Kicinski <jakub.kicinski@netronome.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
In rtnl_net_notifyid(), we certainly can't pass a null GFP flag to
rtnl_notify(). A GFP_KERNEL flag would be fine in most circumstances,
but there are a few paths calling rtnl_net_notifyid() from atomic
context or from RCU critical sections. The later also precludes the use
of gfp_any() as it wouldn't detect the RCU case. Also, the nlmsg_new()
call is wrong too, as it uses GFP_KERNEL unconditionally.
Therefore, we need to pass the GFP flags as parameter and propagate it
through function calls until the proper flags can be determined.
In most cases, GFP_KERNEL is fine. The exceptions are:
* openvswitch: ovs_vport_cmd_get() and ovs_vport_cmd_dump()
indirectly call rtnl_net_notifyid() from RCU critical section,
* rtnetlink: rtmsg_ifinfo_build_skb() already receives GFP flags as
parameter.
Also, in ovs_vport_cmd_build_info(), let's change the GFP flags used
by nlmsg_new(). The function is allowed to sleep, so better make the
flags consistent with the ones used in the following
ovs_vport_cmd_fill_info() call.
Found by code inspection.
Fixes: 9a9634545c ("netns: notify netns id events")
Signed-off-by: Guillaume Nault <gnault@redhat.com>
Acked-by: Nicolas Dichtel <nicolas.dichtel@6wind.com>
Acked-by: Pravin B Shelar <pshelar@ovn.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
This patch removes variables and callback these are related to the nested
device structure.
devices that can be nested have their own nest_level variable that
represents the depth of nested devices.
In the previous patch, new {lower/upper}_level variables are added and
they replace old private nest_level variable.
So, this patch removes all 'nest_level' variables.
In order to avoid lockdep warning, ->ndo_get_lock_subclass() was added
to get lockdep subclass value, which is actually lower nested depth value.
But now, they use the dynamic lockdep key to avoid lockdep warning instead
of the subclass.
So, this patch removes ->ndo_get_lock_subclass() callback.
Signed-off-by: Taehee Yoo <ap420073@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
In order to link an adjacent node, netdev_upper_dev_link() is used
and in order to unlink an adjacent node, netdev_upper_dev_unlink() is used.
unlink operation does not fail, but link operation can fail.
In order to exchange adjacent nodes, we should unlink an old adjacent
node first. then, link a new adjacent node.
If link operation is failed, we should link an old adjacent node again.
But this link operation can fail too.
It eventually breaks the adjacent link relationship.
This patch adds an ignore flag into the netdev_adjacent structure.
If this flag is set, netdev_upper_dev_link() ignores an old adjacent
node for a moment.
This patch also adds new functions for other modules.
netdev_adjacent_change_prepare()
netdev_adjacent_change_commit()
netdev_adjacent_change_abort()
netdev_adjacent_change_prepare() inserts new device into adjacent list
but new device is not allowed to use immediately.
If netdev_adjacent_change_prepare() fails, it internally rollbacks
adjacent list so that we don't need any other action.
netdev_adjacent_change_commit() deletes old device in the adjacent list
and allows new device to use.
netdev_adjacent_change_abort() rollbacks adjacent list.
Signed-off-by: Taehee Yoo <ap420073@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Some interface types could be nested.
(VLAN, BONDING, TEAM, MACSEC, MACVLAN, IPVLAN, VIRT_WIFI, VXLAN, etc..)
These interface types should set lockdep class because, without lockdep
class key, lockdep always warn about unexisting circular locking.
In the current code, these interfaces have their own lockdep class keys and
these manage itself. So that there are so many duplicate code around the
/driver/net and /net/.
This patch adds new generic lockdep keys and some helper functions for it.
This patch does below changes.
a) Add lockdep class keys in struct net_device
- qdisc_running, xmit, addr_list, qdisc_busylock
- these keys are used as dynamic lockdep key.
b) When net_device is being allocated, lockdep keys are registered.
- alloc_netdev_mqs()
c) When net_device is being free'd llockdep keys are unregistered.
- free_netdev()
d) Add generic lockdep key helper function
- netdev_register_lockdep_key()
- netdev_unregister_lockdep_key()
- netdev_update_lockdep_key()
e) Remove unnecessary generic lockdep macro and functions
f) Remove unnecessary lockdep code of each interfaces.
After this patch, each interface modules don't need to maintain
their lockdep keys.
Signed-off-by: Taehee Yoo <ap420073@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Current code doesn't limit the number of nested devices.
Nested devices would be handled recursively and this needs huge stack
memory. So, unlimited nested devices could make stack overflow.
This patch adds upper_level and lower_level, they are common variables
and represent maximum lower/upper depth.
When upper/lower device is attached or dettached,
{lower/upper}_level are updated. and if maximum depth is bigger than 8,
attach routine fails and returns -EMLINK.
In addition, this patch converts recursive routine of
netdev_walk_all_{lower/upper} to iterator routine.
Test commands:
ip link add dummy0 type dummy
ip link add link dummy0 name vlan1 type vlan id 1
ip link set vlan1 up
for i in {2..55}
do
let A=$i-1
ip link add vlan$i link vlan$A type vlan id $i
done
ip link del dummy0
Splat looks like:
[ 155.513226][ T908] BUG: KASAN: use-after-free in __unwind_start+0x71/0x850
[ 155.514162][ T908] Write of size 88 at addr ffff8880608a6cc0 by task ip/908
[ 155.515048][ T908]
[ 155.515333][ T908] CPU: 0 PID: 908 Comm: ip Not tainted 5.4.0-rc3+ #96
[ 155.516147][ T908] Hardware name: innotek GmbH VirtualBox/VirtualBox, BIOS VirtualBox 12/01/2006
[ 155.517233][ T908] Call Trace:
[ 155.517627][ T908]
[ 155.517918][ T908] Allocated by task 0:
[ 155.518412][ T908] (stack is not available)
[ 155.518955][ T908]
[ 155.519228][ T908] Freed by task 0:
[ 155.519885][ T908] (stack is not available)
[ 155.520452][ T908]
[ 155.520729][ T908] The buggy address belongs to the object at ffff8880608a6ac0
[ 155.520729][ T908] which belongs to the cache names_cache of size 4096
[ 155.522387][ T908] The buggy address is located 512 bytes inside of
[ 155.522387][ T908] 4096-byte region [ffff8880608a6ac0, ffff8880608a7ac0)
[ 155.523920][ T908] The buggy address belongs to the page:
[ 155.524552][ T908] page:ffffea0001822800 refcount:1 mapcount:0 mapping:ffff88806c657cc0 index:0x0 compound_mapcount:0
[ 155.525836][ T908] flags: 0x100000000010200(slab|head)
[ 155.526445][ T908] raw: 0100000000010200 ffffea0001813808 ffffea0001a26c08 ffff88806c657cc0
[ 155.527424][ T908] raw: 0000000000000000 0000000000070007 00000001ffffffff 0000000000000000
[ 155.528429][ T908] page dumped because: kasan: bad access detected
[ 155.529158][ T908]
[ 155.529410][ T908] Memory state around the buggy address:
[ 155.530060][ T908] ffff8880608a6b80: fb fb fb fb fb fb fb fb fb fb fb fb fb fb fb fb
[ 155.530971][ T908] ffff8880608a6c00: fb fb fb fb fb f1 f1 f1 f1 00 f2 f2 f2 f3 f3 f3
[ 155.531889][ T908] >ffff8880608a6c80: f3 fb fb fb fb fb fb fb fb fb fb fb fb fb fb fb
[ 155.532806][ T908] ^
[ 155.533509][ T908] ffff8880608a6d00: fb fb fb fb fb fb fb fb fb f1 f1 f1 f1 00 00 00
[ 155.534436][ T908] ffff8880608a6d80: f2 f3 f3 f3 f3 fb fb fb 00 00 00 00 00 00 00 00
[ ... ]
Signed-off-by: Taehee Yoo <ap420073@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Commit 323ebb61e3 ("net: use listified RX for handling GRO_NORMAL
skbs") made use of listified skb processing for the users of
napi_gro_frags().
The same technique can be used in a way more common napi_gro_receive()
to speed up non-merged (GRO_NORMAL) skbs for a wide range of drivers
including gro_cells and mac80211 users.
This slightly changes the return value in cases where skb is being
dropped by the core stack, but it seems to have no impact on related
drivers' functionality.
gro_normal_batch is left untouched as it's very individual for every
single system configuration and might be tuned in manual order to
achieve an optimal performance.
Signed-off-by: Alexander Lobakin <alobakin@dlink.ru>
Acked-by: Edward Cree <ecree@solarflare.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
This reverts commit 0ad646c81b.
As noticed by Jakub, this is no longer needed after
commit 11fc7d5a0a ("tun: fix memory leak in error path")
This no longer exports dev_get_valid_name() for the exclusive
use of tun driver.
Suggested-by: Jakub Kicinski <jakub.kicinski@netronome.com>
Signed-off-by: Eric Dumazet <edumazet@google.com>
Signed-off-by: Jakub Kicinski <jakub.kicinski@netronome.com>
Use helper skb_ensure_writable in two more places to simplify the code.
Signed-off-by: Heiner Kallweit <hkallweit1@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Often the code for example in drivers is interested in getting notifier
call only from certain network namespace. In addition to the existing
global netdevice notifier chain introduce per-netns chains and allow
users to register to that. Eventually this would eliminate unnecessary
overhead in case there are many netdevices in many network namespaces.
Signed-off-by: Jiri Pirko <jiri@mellanox.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Push iterations over net namespaces and netdevices from
register_netdevice_notifier() and unregister_netdevice_notifier()
into helper functions. Along with that introduce continue_reverse macros
to make the code a bit nicer allowing to get rid of "last" marks.
Signed-off-by: Jiri Pirko <jiri@mellanox.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
