-----BEGIN PGP SIGNATURE-----
iHUEABYIAB0WIQTFp0I1jqZrAX+hPRXbK58LschIgwUCZPhEYwAKCRDbK58LschI
g6puAQCGLE9S2d5xw68Fq2KLrTjTXC5trB9N25NWvZWqgXHAeQEA3wXxTC8eF0dG
J24RwZC2vOC4hF3aNGeI0fv/6j7SKA0=
=0IU4
-----END PGP SIGNATURE-----
Merge tag 'for-netdev' of https://git.kernel.org/pub/scm/linux/kernel/git/bpf/bpf
Daniel Borkmann says:
====================
pull-request: bpf 2023-09-06
We've added 9 non-merge commits during the last 6 day(s) which contain
a total of 12 files changed, 189 insertions(+), 44 deletions(-).
The main changes are:
1) Fix bpf_sk_storage to address an invalid wait context lockdep
report and another one to address missing omem uncharge,
from Martin KaFai Lau.
2) Two BPF recursion detection related fixes,
from Sebastian Andrzej Siewior.
3) Fix tailcall limit enforcement in trampolines for s390 JIT,
from Ilya Leoshkevich.
4) Fix a sockmap refcount race where skbs in sk_psock_backlog can
be referenced after user space side has already skb_consumed them,
from John Fastabend.
5) Fix BPF CI flake/race wrt sockmap vsock write test where
the transport endpoint is not connected, from Xu Kuohai.
6) Follow-up doc fix to address a cross-link warning,
from Eduard Zingerman.
* tag 'for-netdev' of https://git.kernel.org/pub/scm/linux/kernel/git/bpf/bpf:
selftests/bpf: Check bpf_sk_storage has uncharged sk_omem_alloc
bpf: bpf_sk_storage: Fix the missing uncharge in sk_omem_alloc
bpf: bpf_sk_storage: Fix invalid wait context lockdep report
s390/bpf: Pass through tail call counter in trampolines
bpf: Assign bpf_tramp_run_ctx::saved_run_ctx before recursion check.
bpf: Invoke __bpf_prog_exit_sleepable_recur() on recursion in kern_sys_bpf().
bpf, sockmap: Fix skb refcnt race after locking changes
docs/bpf: Fix "file doesn't exist" warnings in {llvm_reloc,btf}.rst
selftests/bpf: Fix a CI failure caused by vsock write
====================
Link: https://lore.kernel.org/r/20230906095117.16941-1-daniel@iogearbox.net
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
The commit c83597fa5d ("bpf: Refactor some inode/task/sk storage functions
for reuse"), refactored the bpf_{sk,task,inode}_storage_free() into
bpf_local_storage_unlink_nolock() which then later renamed to
bpf_local_storage_destroy(). The commit accidentally passed the
"bool uncharge_mem = false" argument to bpf_selem_unlink_storage_nolock()
which then stopped the uncharge from happening to the sk->sk_omem_alloc.
This missing uncharge only happens when the sk is going away (during
__sk_destruct).
This patch fixes it by always passing "uncharge_mem = true". It is a
noop to the task/inode/cgroup storage because they do not have the
map_local_storage_(un)charge enabled in the map_ops. A followup patch
will be done in bpf-next to remove the uncharge_mem argument.
A selftest is added in the next patch.
Fixes: c83597fa5d ("bpf: Refactor some inode/task/sk storage functions for reuse")
Signed-off-by: Martin KaFai Lau <martin.lau@kernel.org>
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
Link: https://lore.kernel.org/bpf/20230901231129.578493-3-martin.lau@linux.dev
'./test_progs -t test_local_storage' reported a splat:
[ 27.137569] =============================
[ 27.138122] [ BUG: Invalid wait context ]
[ 27.138650] 6.5.0-03980-gd11ae1b16b0a #247 Tainted: G O
[ 27.139542] -----------------------------
[ 27.140106] test_progs/1729 is trying to lock:
[ 27.140713] ffff8883ef047b88 (stock_lock){-.-.}-{3:3}, at: local_lock_acquire+0x9/0x130
[ 27.141834] other info that might help us debug this:
[ 27.142437] context-{5:5}
[ 27.142856] 2 locks held by test_progs/1729:
[ 27.143352] #0: ffffffff84bcd9c0 (rcu_read_lock){....}-{1:3}, at: rcu_lock_acquire+0x4/0x40
[ 27.144492] #1: ffff888107deb2c0 (&storage->lock){..-.}-{2:2}, at: bpf_local_storage_update+0x39e/0x8e0
[ 27.145855] stack backtrace:
[ 27.146274] CPU: 0 PID: 1729 Comm: test_progs Tainted: G O 6.5.0-03980-gd11ae1b16b0a #247
[ 27.147550] Hardware name: QEMU Standard PC (i440FX + PIIX, 1996), BIOS rel-1.14.0-0-g155821a1990b-prebuilt.qemu.org 04/01/2014
[ 27.149127] Call Trace:
[ 27.149490] <TASK>
[ 27.149867] dump_stack_lvl+0x130/0x1d0
[ 27.152609] dump_stack+0x14/0x20
[ 27.153131] __lock_acquire+0x1657/0x2220
[ 27.153677] lock_acquire+0x1b8/0x510
[ 27.157908] local_lock_acquire+0x29/0x130
[ 27.159048] obj_cgroup_charge+0xf4/0x3c0
[ 27.160794] slab_pre_alloc_hook+0x28e/0x2b0
[ 27.161931] __kmem_cache_alloc_node+0x51/0x210
[ 27.163557] __kmalloc+0xaa/0x210
[ 27.164593] bpf_map_kzalloc+0xbc/0x170
[ 27.165147] bpf_selem_alloc+0x130/0x510
[ 27.166295] bpf_local_storage_update+0x5aa/0x8e0
[ 27.167042] bpf_fd_sk_storage_update_elem+0xdb/0x1a0
[ 27.169199] bpf_map_update_value+0x415/0x4f0
[ 27.169871] map_update_elem+0x413/0x550
[ 27.170330] __sys_bpf+0x5e9/0x640
[ 27.174065] __x64_sys_bpf+0x80/0x90
[ 27.174568] do_syscall_64+0x48/0xa0
[ 27.175201] entry_SYSCALL_64_after_hwframe+0x6e/0xd8
[ 27.175932] RIP: 0033:0x7effb40e41ad
[ 27.176357] Code: ff c3 66 2e 0f 1f 84 00 00 00 00 00 90 f3 0f 1e fa 48 89 f8 48 89 f7 48 89 d6 48 89 ca 4d 89 c2 4d 89 c8 4c 8b 4c 24 08 0f 05 <48> 3d 01 f0 ff ff 73 01 c3 48 8b 0d8
[ 27.179028] RSP: 002b:00007ffe64c21fc8 EFLAGS: 00000202 ORIG_RAX: 0000000000000141
[ 27.180088] RAX: ffffffffffffffda RBX: 00007ffe64c22768 RCX: 00007effb40e41ad
[ 27.181082] RDX: 0000000000000020 RSI: 00007ffe64c22008 RDI: 0000000000000002
[ 27.182030] RBP: 00007ffe64c21ff0 R08: 0000000000000000 R09: 00007ffe64c22788
[ 27.183038] R10: 0000000000000064 R11: 0000000000000202 R12: 0000000000000000
[ 27.184006] R13: 00007ffe64c22788 R14: 00007effb42a1000 R15: 0000000000000000
[ 27.184958] </TASK>
It complains about acquiring a local_lock while holding a raw_spin_lock.
It means it should not allocate memory while holding a raw_spin_lock
since it is not safe for RT.
raw_spin_lock is needed because bpf_local_storage supports tracing
context. In particular for task local storage, it is easy to
get a "current" task PTR_TO_BTF_ID in tracing bpf prog.
However, task (and cgroup) local storage has already been moved to
bpf mem allocator which can be used after raw_spin_lock.
The splat is for the sk storage. For sk (and inode) storage,
it has not been moved to bpf mem allocator. Using raw_spin_lock or not,
kzalloc(GFP_ATOMIC) could theoretically be unsafe in tracing context.
However, the local storage helper requires a verifier accepted
sk pointer (PTR_TO_BTF_ID), it is hypothetical if that (mean running
a bpf prog in a kzalloc unsafe context and also able to hold a verifier
accepted sk pointer) could happen.
This patch avoids kzalloc after raw_spin_lock to silent the splat.
There is an existing kzalloc before the raw_spin_lock. At that point,
a kzalloc is very likely required because a lookup has just been done
before. Thus, this patch always does the kzalloc before acquiring
the raw_spin_lock and remove the later kzalloc usage after the
raw_spin_lock. After this change, it will have a charge and then
uncharge during the syscall bpf_map_update_elem() code path.
This patch opts for simplicity and not continue the old
optimization to save one charge and uncharge.
This issue is dated back to the very first commit of bpf_sk_storage
which had been refactored multiple times to create task, inode, and
cgroup storage. This patch uses a Fixes tag with a more recent
commit that should be easier to do backport.
Fixes: b00fa38a9c ("bpf: Enable non-atomic allocations in local storage")
Signed-off-by: Martin KaFai Lau <martin.lau@kernel.org>
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
Link: https://lore.kernel.org/bpf/20230901231129.578493-2-martin.lau@linux.dev
s390x eBPF programs use the following extension to the s390x calling
convention: tail call counter is passed on stack at offset
STK_OFF_TCCNT, which callees otherwise use as scratch space.
Currently trampoline does not respect this and clobbers tail call
counter. This breaks enforcing tail call limits in eBPF programs, which
have trampolines attached to them.
Fix by forwarding a copy of the tail call counter to the original eBPF
program in the trampoline (for fexit), and by restoring it at the end
of the trampoline (for fentry).
Fixes: 528eb2cb87 ("s390/bpf: Implement arch_prepare_bpf_trampoline()")
Reported-by: Leon Hwang <hffilwlqm@gmail.com>
Signed-off-by: Ilya Leoshkevich <iii@linux.ibm.com>
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
Link: https://lore.kernel.org/bpf/20230906004448.111674-1-iii@linux.ibm.com
__bpf_prog_enter_recur() assigns bpf_tramp_run_ctx::saved_run_ctx before
performing the recursion check which means in case of a recursion
__bpf_prog_exit_recur() uses the previously set bpf_tramp_run_ctx::saved_run_ctx
value.
__bpf_prog_enter_sleepable_recur() assigns bpf_tramp_run_ctx::saved_run_ctx
after the recursion check which means in case of a recursion
__bpf_prog_exit_sleepable_recur() uses an uninitialized value. This does not
look right. If I read the entry trampoline code right, then bpf_tramp_run_ctx
isn't initialized upfront.
Align __bpf_prog_enter_sleepable_recur() with __bpf_prog_enter_recur() and
set bpf_tramp_run_ctx::saved_run_ctx before the recursion check is made.
Remove the assignment of saved_run_ctx in kern_sys_bpf() since it happens
a few cycles later.
Fixes: e384c7b7b4 ("bpf, x86: Create bpf_tramp_run_ctx on the caller thread's stack")
Signed-off-by: Sebastian Andrzej Siewior <bigeasy@linutronix.de>
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
Acked-by: Jiri Olsa <jolsa@kernel.org>
Link: https://lore.kernel.org/bpf/20230830080405.251926-3-bigeasy@linutronix.de
If __bpf_prog_enter_sleepable_recur() detects recursion then it returns
0 without undoing rcu_read_lock_trace(), migrate_disable() or
decrementing the recursion counter. This is fine in the JIT case because
the JIT code will jump in the 0 case to the end and invoke the matching
exit trampoline (__bpf_prog_exit_sleepable_recur()).
This is not the case in kern_sys_bpf() which returns directly to the
caller with an error code.
Add __bpf_prog_exit_sleepable_recur() as clean up in the recursion case.
Fixes: b1d18a7574 ("bpf: Extend sys_bpf commands for bpf_syscall programs.")
Signed-off-by: Sebastian Andrzej Siewior <bigeasy@linutronix.de>
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
Acked-by: Jiri Olsa <jolsa@kernel.org>
Link: https://lore.kernel.org/bpf/20230830080405.251926-2-bigeasy@linutronix.de
sphinx complains about the use of "%PHYLINK_PCS_NEG_*":
Documentation/networking/kapi:144: ./include/linux/phylink.h:601: WARNING: Inline literal start-string without end-string.
Documentation/networking/kapi:144: ./include/linux/phylink.h:633: WARNING: Inline literal start-string without end-string.
These are not valid symbols so drop the '%' prefix.
Alternatively we could use %PHYLINK_PCS_NEG_\* (escape the *)
or use normal literal ``PHYLINK_PCS_NEG_*`` but there is already
a handful of un-adorned DEFINE_* in this file.
Fixes: f99d471afa ("net: phylink: add PCS negotiation mode")
Reported-by: Stephen Rothwell <sfr@canb.auug.org.au>
Link: https://lore.kernel.org/all/20230626162908.2f149f98@canb.auug.org.au/
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
Reviewed-by: Bagas Sanjaya <bagasdotme@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Vladimir Oltean says:
====================
tc-cbs offload fixes for SJA1105 DSA
Yanan Yang has pointed out to me that certain tc-cbs offloaded
configurations do not appear to do any shaping on the LS1021A-TSN board
(SJA1105T).
This is due to an apparent documentation error that also made its way
into the driver, which patch 1/3 now fixes.
While investigating and then testing, I've found 2 more bugs, which are
patches 2/3 and 3/3.
====================
Signed-off-by: David S. Miller <davem@davemloft.net>
The blamed commit left this delta behind:
struct sja1105_cbs_entry {
- u64 port;
- u64 prio;
+ u64 port; /* Not used for SJA1110 */
+ u64 prio; /* Not used for SJA1110 */
u64 credit_hi;
u64 credit_lo;
u64 send_slope;
u64 idle_slope;
};
but did not actually implement tc-cbs offload fully for the new switch.
The offload is accepted, but it doesn't work.
The difference compared to earlier switch generations is that now, the
table of CBS shapers is sparse, because there are many more shapers, so
the mapping between a {port, prio} and a table index is static, rather
than requiring us to store the port and prio into the sja1105_cbs_entry.
So, the problem is that the code programs the CBS shaper parameters at a
dynamic table index which is incorrect.
All that needs to be done for SJA1110 CBS shapers to work is to bypass
the logic which allocates shapers in a dense manner, as for SJA1105, and
use the fixed mapping instead.
Fixes: 3e77e59bf8 ("net: dsa: sja1105: add support for the SJA1110 switch family")
Signed-off-by: Vladimir Oltean <vladimir.oltean@nxp.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
After running command [2] too many times in a row:
[1] $ tc qdisc add dev sw2p0 root handle 1: mqprio num_tc 8 \
map 0 1 2 3 4 5 6 7 queues 1@0 1@1 1@2 1@3 1@4 1@5 1@6 1@7 hw 0
[2] $ tc qdisc replace dev sw2p0 parent 1:1 cbs offload 1 \
idleslope 120000 sendslope -880000 locredit -1320 hicredit 180
(aka more than priv->info->num_cbs_shapers times)
we start seeing the following error message:
Error: Specified device failed to setup cbs hardware offload.
This comes from the fact that ndo_setup_tc(TC_SETUP_QDISC_CBS) presents
the same API for the qdisc create and replace cases, and the sja1105
driver fails to distinguish between the 2. Thus, it always thinks that
it must allocate the same shaper for a {port, queue} pair, when it may
instead have to replace an existing one.
Fixes: 4d7525085a ("net: dsa: sja1105: offload the Credit-Based Shaper qdisc")
Signed-off-by: Vladimir Oltean <vladimir.oltean@nxp.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
More careful measurement of the tc-cbs bandwidth shows that the stream
bandwidth (effectively idleslope) increases, there is a larger and
larger discrepancy between the rate limit obtained by the software
Qdisc, and the rate limit obtained by its offloaded counterpart.
The discrepancy becomes so large, that e.g. at an idleslope of 40000
(40Mbps), the offloaded cbs does not actually rate limit anything, and
traffic will pass at line rate through a 100 Mbps port.
The reason for the discrepancy is that the hardware documentation I've
been following is incorrect. UM11040.pdf (for SJA1105P/Q/R/S) states
about IDLE_SLOPE that it is "the rate (in unit of bytes/sec) at which
the credit counter is increased".
Cross-checking with UM10944.pdf (for SJA1105E/T) and UM11107.pdf
(for SJA1110), the wording is different: "This field specifies the
value, in bytes per second times link speed, by which the credit counter
is increased".
So there's an extra scaling for link speed that the driver is currently
not accounting for, and apparently (empirically), that link speed is
expressed in Kbps.
I've pondered whether to pollute the sja1105_mac_link_up()
implementation with CBS shaper reprogramming, but I don't think it is
worth it. IMO, the UAPI exposed by tc-cbs requires user space to
recalculate the sendslope anyway, since the formula for that depends on
port_transmit_rate (see man tc-cbs), which is not an invariant from tc's
perspective.
So we use the offload->sendslope and offload->idleslope to deduce the
original port_transmit_rate from the CBS formula, and use that value to
scale the offload->sendslope and offload->idleslope to values that the
hardware understands.
Some numerical data points:
40Mbps stream, max interfering frame size 1500, port speed 100M
---------------------------------------------------------------
tc-cbs parameters:
idleslope 40000 sendslope -60000 locredit -900 hicredit 600
which result in hardware values:
Before (doesn't work) After (works)
credit_hi 600 600
credit_lo 900 900
send_slope 7500000 75
idle_slope 5000000 50
40Mbps stream, max interfering frame size 1500, port speed 1G
-------------------------------------------------------------
tc-cbs parameters:
idleslope 40000 sendslope -960000 locredit -1440 hicredit 60
which result in hardware values:
Before (doesn't work) After (works)
credit_hi 60 60
credit_lo 1440 1440
send_slope 120000000 120
idle_slope 5000000 5
5.12Mbps stream, max interfering frame size 1522, port speed 100M
-----------------------------------------------------------------
tc-cbs parameters:
idleslope 5120 sendslope -94880 locredit -1444 hicredit 77
which result in hardware values:
Before (doesn't work) After (works)
credit_hi 77 77
credit_lo 1444 1444
send_slope 11860000 118
idle_slope 640000 6
Tested on SJA1105T, SJA1105S and SJA1110A, at 1Gbps and 100Mbps.
Fixes: 4d7525085a ("net: dsa: sja1105: offload the Credit-Based Shaper qdisc")
Reported-by: Yanan Yang <yanan.yang@nxp.com>
Signed-off-by: Vladimir Oltean <vladimir.oltean@nxp.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Tony Nguyen says:
====================
Change MIN_TXD and MIN_RXD to allow set rx/tx value between 64 and 80
Olga Zaborska says:
Change the minimum value of RX/TX descriptors to 64 to enable setting the rx/tx
value between 64 and 80. All igb, igbvf and igc devices can use as low as 64
descriptors.
====================
Signed-off-by: David S. Miller <davem@davemloft.net>
ACL flow table is required in switchdev mode when metadata is enabled,
driver creates such table when loading each vport. However, not every
vport is loaded in switchdev mode. Such as ECPF if it's the eswitch manager.
In this case, ACL flow table is still needed.
To make it modularized, create ACL flow table for eswitch manager as
default and skip such operations when loading manager vport.
Also, there is no need to load the eswitch manager vport in switchdev mode.
This means there is no need to load it on regular connect-x HCAs where
the PF is the eswitch manager. This will avoid creating duplicate ACL
flow table for host PF vport.
Fixes: 29bcb6e4fe ("net/mlx5e: E-Switch, Use metadata for vport matching in send-to-vport rules")
Fixes: eb8e9fae0a ("mlx5/core: E-Switch, Allocate ECPF vport if it's an eswitch manager")
Fixes: 5019833d66 ("net/mlx5: E-switch, Introduce helper function to enable/disable vports")
Signed-off-by: Bodong Wang <bodong@nvidia.com>
Reviewed-by: Mark Bloch <mbloch@nvidia.com>
Signed-off-by: Saeed Mahameed <saeedm@nvidia.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
In the cited commit, the mirred devices are recorded and checked while
parsing the actions. In order to avoid system crash, the duplicate
action in a single rule is not allowed.
But the rule is actually break down into several FTEs in different
tables, for either mirroring, or the specified types of actions which
use post action infrastructure.
It will reject certain action list by mistake, for example:
actions:enp8s0f0_1,set(ipv4(ttl=63)),enp8s0f0_0,enp8s0f0_1.
Here the rule is split to two FTEs because of pedit action.
To fix this issue, when parsing the rule actions, reset if_count to
clear the mirred devices array if the rule is split to multiple
FTEs, and then the duplicate checking is restarted.
Fixes: 554fe75c1b ("net/mlx5e: Avoid duplicating rule destinations")
Signed-off-by: Jianbo Liu <jianbol@nvidia.com>
Reviewed-by: Vlad Buslov <vladbu@nvidia.com>
Signed-off-by: Saeed Mahameed <saeedm@nvidia.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
team interface has used a dynamic lockdep key to avoid false-positive
lockdep deadlock detection. Virtual interfaces such as team usually
have their own lock for protecting private data.
These interfaces can be nested.
team0
|
team1
Each interface's lock is actually different(team0->lock and team1->lock).
So,
mutex_lock(&team0->lock);
mutex_lock(&team1->lock);
mutex_unlock(&team1->lock);
mutex_unlock(&team0->lock);
The above case is absolutely safe. But lockdep warns about deadlock.
Because the lockdep understands these two locks are same. This is a
false-positive lockdep warning.
So, in order to avoid this problem, the team interfaces started to use
dynamic lockdep key. The false-positive problem was fixed, but it
introduced a new problem.
When the new team virtual interface is created, it registers a dynamic
lockdep key(creates dynamic lockdep key) and uses it. But there is the
limitation of the number of lockdep keys.
So, If so many team interfaces are created, it consumes all lockdep keys.
Then, the lockdep stops to work and warns about it.
In order to fix this problem, team interfaces use the subclass instead
of the dynamic key. So, when a new team interface is created, it doesn't
register(create) a new lockdep, but uses existed subclass key instead.
It is already used by the bonding interface for a similar case.
As the bonding interface does, the subclass variable is the same as
the 'dev->nested_level'. This variable indicates the depth in the stacked
interface graph.
The 'dev->nested_level' is protected by RTNL and RCU.
So, 'mutex_lock_nested()' for 'team->lock' requires RTNL or RCU.
In the current code, 'team->lock' is usually acquired under RTNL, there is
no problem with using 'dev->nested_level'.
The 'team_nl_team_get()' and The 'lb_stats_refresh()' functions acquire
'team->lock' without RTNL.
But these don't iterate their own ports nested so they don't need nested
lock.
Reproducer:
for i in {0..1000}
do
ip link add team$i type team
ip link add dummy$i master team$i type dummy
ip link set dummy$i up
ip link set team$i up
done
Splat looks like:
BUG: MAX_LOCKDEP_ENTRIES too low!
turning off the locking correctness validator.
Please attach the output of /proc/lock_stat to the bug report
CPU: 0 PID: 4104 Comm: ip Not tainted 6.5.0-rc7+ #45
Call Trace:
<TASK>
dump_stack_lvl+0x64/0xb0
add_lock_to_list+0x30d/0x5e0
check_prev_add+0x73a/0x23a0
...
sock_def_readable+0xfe/0x4f0
netlink_broadcast+0x76b/0xac0
nlmsg_notify+0x69/0x1d0
dev_open+0xed/0x130
...
Reported-by: syzbot+9bbbacfbf1e04d5221f7@syzkaller.appspotmail.com
Fixes: 369f61bee0 ("team: fix nested locking lockdep warning")
Signed-off-by: Taehee Yoo <ap420073@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
__skb_get_hash_symmetric() was added to compute a symmetric hash over
the protocol, addresses and transport ports, by commit eb70db8756
("packet: Use symmetric hash for PACKET_FANOUT_HASH."). It uses
flow_keys_dissector_symmetric_keys as the flow_dissector to incorporate
IPv4 addresses, IPv6 addresses and ports. However, it should not specify
the flag as FLOW_DISSECTOR_F_STOP_AT_FLOW_LABEL, which stops further
dissection when an IPv6 flow label is encountered, making transport
ports not being incorporated in such case.
As a consequence, the symmetric hash is based on 5-tuple for IPv4 but
3-tuple for IPv6 when flow label is present. It caused a few problems,
e.g. when nft symhash and openvswitch l4_sym rely on the symmetric hash
to perform load balancing as different L4 flows between two given IPv6
addresses would always get the same symmetric hash, leading to uneven
traffic distribution.
Removing the use of FLOW_DISSECTOR_F_STOP_AT_FLOW_LABEL makes sure the
symmetric hash is based on 5-tuple for both IPv4 and IPv6 consistently.
Fixes: eb70db8756 ("packet: Use symmetric hash for PACKET_FANOUT_HASH.")
Reported-by: Lars Ekman <uablrek@gmail.com>
Closes: https://github.com/antrea-io/antrea/issues/5457
Signed-off-by: Quan Tian <qtian@vmware.com>
Reviewed-by: Eric Dumazet <edumazet@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Change the minimum value of RX/TX descriptors to 64 to enable setting the rx/tx
value between 64 and 80. All igb devices can use as low as 64 descriptors.
This change will unify igb with other drivers.
Based on commit 7b1be1987c ("e1000e: lower ring minimum size to 64")
Fixes: 9d5c824399 ("igb: PCI-Express 82575 Gigabit Ethernet driver")
Signed-off-by: Olga Zaborska <olga.zaborska@intel.com>
Tested-by: Pucha Himasekhar Reddy <himasekharx.reddy.pucha@intel.com> (A Contingent worker at Intel)
Signed-off-by: Tony Nguyen <anthony.l.nguyen@intel.com>
Change the minimum value of RX/TX descriptors to 64 to enable setting the rx/tx
value between 64 and 80. All igbvf devices can use as low as 64 descriptors.
This change will unify igbvf with other drivers.
Based on commit 7b1be1987c ("e1000e: lower ring minimum size to 64")
Fixes: d4e0fe01a3 ("igbvf: add new driver to support 82576 virtual functions")
Signed-off-by: Olga Zaborska <olga.zaborska@intel.com>
Tested-by: Rafal Romanowski <rafal.romanowski@intel.com>
Signed-off-by: Tony Nguyen <anthony.l.nguyen@intel.com>
Change the minimum value of RX/TX descriptors to 64 to enable setting the rx/tx
value between 64 and 80. All igc devices can use as low as 64 descriptors.
This change will unify igc with other drivers.
Based on commit 7b1be1987c ("e1000e: lower ring minimum size to 64")
Fixes: 0507ef8a03 ("igc: Add transmit and receive fastpath and interrupt handlers")
Signed-off-by: Olga Zaborska <olga.zaborska@intel.com>
Tested-by: Naama Meir <naamax.meir@linux.intel.com>
Signed-off-by: Tony Nguyen <anthony.l.nguyen@intel.com>
The smq value used in the CN10K NIX AQ instruction enqueue mailbox
handler was truncated to 9-bit value from 10-bit value because of
typecasting the CN10K mbox request structure to the CN9K structure.
Though this hasn't caused any problems when programming the NIX SQ
context to the HW because the context structure is the same size.
However, this causes a problem when accessing the structure parameters.
This patch reads the right smq value for each platform.
Fixes: 30077d210c ("octeontx2-af: cn10k: Update NIX/NPA context structure")
Signed-off-by: Geetha sowjanya <gakula@marvell.com>
Signed-off-by: Sunil Kovvuri Goutham <sgoutham@marvell.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
This is a follow up of commit 915d975b2f ("net: deal with integer
overflows in kmalloc_reserve()") based on David Laight feedback.
Back in 2010, I failed to realize malicious users could set dev->mtu
to arbitrary values. This mtu has been since limited to 0x7fffffff but
regardless of how big dev->mtu is, it makes no sense for igmpv3_newpack()
to allocate more than IP_MAX_MTU and risk various skb fields overflows.
Fixes: 57e1ab6ead ("igmp: refine skb allocations")
Link: https://lore.kernel.org/netdev/d273628df80f45428e739274ab9ecb72@AcuMS.aculab.com/
Signed-off-by: Eric Dumazet <edumazet@google.com>
Reported-by: David Laight <David.Laight@ACULAB.COM>
Cc: Kyle Zeng <zengyhkyle@gmail.com>
Reviewed-by: Simon Horman <horms@kernel.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
kcm_exit_net() should call mutex_destroy() on knet->mutex. This is especially
needed if CONFIG_DEBUG_MUTEXES is enabled.
Fixes: ab7ac4eb98 ("kcm: Kernel Connection Multiplexor module")
Signed-off-by: Shigeru Yoshida <syoshida@redhat.com>
Link: https://lore.kernel.org/r/20230902170708.1727999-1-syoshida@redhat.com
Signed-off-by: Paolo Abeni <pabeni@redhat.com>
Kuniyuki Iwashima says:
====================
af_unix: Fix four data-races.
While running syzkaller, KCSAN reported 3 data-races with
systemd-coredump using AF_UNIX sockets.
This series fixes the three and another one inspiered by
one of the reports.
====================
Signed-off-by: David S. Miller <davem@davemloft.net>
As with sk->sk_shutdown shown in the previous patch, sk->sk_err can be
read locklessly by unix_dgram_sendmsg().
Let's use READ_ONCE() for sk_err as well.
Note that the writer side is marked by commit cc04410af7 ("af_unix:
annotate lockless accesses to sk->sk_err").
Fixes: 1da177e4c3 ("Linux-2.6.12-rc2")
Signed-off-by: Kuniyuki Iwashima <kuniyu@amazon.com>
Reviewed-by: Eric Dumazet <edumazet@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Heiko Carstens reported that SCM_PIDFD does not work with MSG_CMSG_COMPAT
because scm_pidfd_recv() always checks msg_controllen against sizeof(struct
cmsghdr).
We need to use sizeof(struct compat_cmsghdr) for the compat case.
Fixes: 5e2ff6704a ("scm: add SO_PASSPIDFD and SCM_PIDFD")
Reported-by: Heiko Carstens <hca@linux.ibm.com>
Closes: https://lore.kernel.org/netdev/20230901200517.8742-A-hca@linux.ibm.com/
Signed-off-by: Kuniyuki Iwashima <kuniyu@amazon.com>
Tested-by: Heiko Carstens <hca@linux.ibm.com>
Reviewed-by: Alexander Mikhalitsyn <aleksandr.mikhalitsyn@canonical.com>
Reviewed-by: Michal Swiatkowski <michal.swiatkowski@linux.intel.com>
Acked-by: Christian Brauner <brauner@kernel.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
Some corporate proxies block our current NIPA URLs because
they use a free / shady DNS domain. As suggested by Jesse
we got a new DNS entry from Konstantin - netdev.bots.linux.dev,
use it.
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
Reviewed-by: Simon Horman <horms@kernel.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
The patchwork states are largely self-explanatory but small
ambiguities may still come up. Document how we interpret
the states in networking.
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
There is a race where skb's from the sk_psock_backlog can be referenced
after userspace side has already skb_consumed() the sk_buff and its refcnt
dropped to zer0 causing use after free.
The flow is the following:
while ((skb = skb_peek(&psock->ingress_skb))
sk_psock_handle_Skb(psock, skb, ..., ingress)
if (!ingress) ...
sk_psock_skb_ingress
sk_psock_skb_ingress_enqueue(skb)
msg->skb = skb
sk_psock_queue_msg(psock, msg)
skb_dequeue(&psock->ingress_skb)
The sk_psock_queue_msg() puts the msg on the ingress_msg queue. This is
what the application reads when recvmsg() is called. An application can
read this anytime after the msg is placed on the queue. The recvmsg hook
will also read msg->skb and then after user space reads the msg will call
consume_skb(skb) on it effectively free'ing it.
But, the race is in above where backlog queue still has a reference to
the skb and calls skb_dequeue(). If the skb_dequeue happens after the
user reads and free's the skb we have a use after free.
The !ingress case does not suffer from this problem because it uses
sendmsg_*(sk, msg) which does not pass the sk_buff further down the
stack.
The following splat was observed with 'test_progs -t sockmap_listen':
[ 1022.710250][ T2556] general protection fault, ...
[...]
[ 1022.712830][ T2556] Workqueue: events sk_psock_backlog
[ 1022.713262][ T2556] RIP: 0010:skb_dequeue+0x4c/0x80
[ 1022.713653][ T2556] Code: ...
[...]
[ 1022.720699][ T2556] Call Trace:
[ 1022.720984][ T2556] <TASK>
[ 1022.721254][ T2556] ? die_addr+0x32/0x80^M
[ 1022.721589][ T2556] ? exc_general_protection+0x25a/0x4b0
[ 1022.722026][ T2556] ? asm_exc_general_protection+0x22/0x30
[ 1022.722489][ T2556] ? skb_dequeue+0x4c/0x80
[ 1022.722854][ T2556] sk_psock_backlog+0x27a/0x300
[ 1022.723243][ T2556] process_one_work+0x2a7/0x5b0
[ 1022.723633][ T2556] worker_thread+0x4f/0x3a0
[ 1022.723998][ T2556] ? __pfx_worker_thread+0x10/0x10
[ 1022.724386][ T2556] kthread+0xfd/0x130
[ 1022.724709][ T2556] ? __pfx_kthread+0x10/0x10
[ 1022.725066][ T2556] ret_from_fork+0x2d/0x50
[ 1022.725409][ T2556] ? __pfx_kthread+0x10/0x10
[ 1022.725799][ T2556] ret_from_fork_asm+0x1b/0x30
[ 1022.726201][ T2556] </TASK>
To fix we add an skb_get() before passing the skb to be enqueued in the
engress queue. This bumps the skb->users refcnt so that consume_skb()
and kfree_skb will not immediately free the sk_buff. With this we can
be sure the skb is still around when we do the dequeue. Then we just
need to decrement the refcnt or free the skb in the backlog case which
we do by calling kfree_skb() on the ingress case as well as the sendmsg
case.
Before locking change from fixes tag we had the sock locked so we
couldn't race with user and there was no issue here.
Fixes: 799aa7f98d ("skmsg: Avoid lock_sock() in sk_psock_backlog()")
Reported-by: Jiri Olsa <jolsa@kernel.org>
Signed-off-by: John Fastabend <john.fastabend@gmail.com>
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
Tested-by: Xu Kuohai <xukuohai@huawei.com>
Tested-by: Jiri Olsa <jolsa@kernel.org>
Link: https://lore.kernel.org/bpf/20230901202137.214666-1-john.fastabend@gmail.com
Previously, the defines for phy_device flags in the Micrel driver were
ambiguous in their representation. They were intended to be bit masks
but were mistakenly defined as bit positions. This led to the following
issues:
- MICREL_KSZ8_P1_ERRATA, designated for KSZ88xx switches, overlapped
with MICREL_PHY_FXEN and MICREL_PHY_50MHZ_CLK.
- Due to this overlap, the code path for MICREL_PHY_FXEN, tailored for
the KSZ8041 PHY, was not executed for KSZ88xx PHYs.
- Similarly, the code associated with MICREL_PHY_50MHZ_CLK wasn't
triggered for KSZ88xx.
To rectify this, all three flags have now been explicitly converted to
use the `BIT()` macro, ensuring they are defined as bit masks and
preventing potential overlaps in the future.
Fixes: 49011e0c15 ("net: phy: micrel: ksz886x/ksz8081: add cabletest support")
Signed-off-by: Oleksij Rempel <o.rempel@pengutronix.de>
Reviewed-by: Russell King (Oracle) <rmk+kernel@armlinux.org.uk>
Signed-off-by: David S. Miller <davem@davemloft.net>
The existing code incorrectly casted a negative value (the result of a
subtraction) to an unsigned value without checking. For example, if
/proc/sys/net/ipv6/conf/*/temp_prefered_lft was set to 1, the preferred
lifetime would jump to 4 billion seconds. On my machine and network the
shortest lifetime that avoided underflow was 3 seconds.
Fixes: 76506a986d ("IPv6: fix DESYNC_FACTOR")
Signed-off-by: Alex Henrie <alexhenrie24@gmail.com>
Reviewed-by: David Ahern <dsahern@kernel.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
The veth_xmit function returns NETDEV_TX_OK even when packets are dropped.
This behavior leads to incorrect calculations of statistics counts, as
well as things like txq->trans_start updates.
Fixes: e314dbdc1c ("[NET]: Virtual ethernet device driver.")
Signed-off-by: Liang Chen <liangchen.linux@gmail.com>
Reviewed-by: Eric Dumazet <edumazet@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
gve_rx_append_frags() is able to build skbs chained with frag_list,
like GRO engine.
Problem is that shinfo->frag_list should only be used
for the head of the chain.
All other links should use skb->next pointer.
Otherwise, built skbs are not valid and can cause crashes.
Equivalent code in GRO (skb_gro_receive()) is:
if (NAPI_GRO_CB(p)->last == p)
skb_shinfo(p)->frag_list = skb;
else
NAPI_GRO_CB(p)->last->next = skb;
NAPI_GRO_CB(p)->last = skb;
Fixes: 9b8dd5e5ea ("gve: DQO: Add RX path")
Signed-off-by: Eric Dumazet <edumazet@google.com>
Cc: Bailey Forrest <bcf@google.com>
Cc: Willem de Bruijn <willemb@google.com>
Cc: Catherine Sullivan <csully@google.com>
Reviewed-by: David Ahern <dsahern@kernel.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
Blamed commit changed:
ptr = kmalloc(size);
if (ptr)
size = ksize(ptr);
to:
size = kmalloc_size_roundup(size);
ptr = kmalloc(size);
This allowed various crash as reported by syzbot [1]
and Kyle Zeng.
Problem is that if @size is bigger than 0x80000001,
kmalloc_size_roundup(size) returns 2^32.
kmalloc_reserve() uses a 32bit variable (obj_size),
so 2^32 is truncated to 0.
kmalloc(0) returns ZERO_SIZE_PTR which is not handled by
skb allocations.
Following trace can be triggered if a netdev->mtu is set
close to 0x7fffffff
We might in the future limit netdev->mtu to more sensible
limit (like KMALLOC_MAX_SIZE).
This patch is based on a syzbot report, and also a report
and tentative fix from Kyle Zeng.
[1]
BUG: KASAN: user-memory-access in __build_skb_around net/core/skbuff.c:294 [inline]
BUG: KASAN: user-memory-access in __alloc_skb+0x3c4/0x6e8 net/core/skbuff.c:527
Write of size 32 at addr 00000000fffffd10 by task syz-executor.4/22554
CPU: 1 PID: 22554 Comm: syz-executor.4 Not tainted 6.1.39-syzkaller #0
Hardware name: Google Google Compute Engine/Google Compute Engine, BIOS Google 07/03/2023
Call trace:
dump_backtrace+0x1c8/0x1f4 arch/arm64/kernel/stacktrace.c:279
show_stack+0x2c/0x3c arch/arm64/kernel/stacktrace.c:286
__dump_stack lib/dump_stack.c:88 [inline]
dump_stack_lvl+0x120/0x1a0 lib/dump_stack.c:106
print_report+0xe4/0x4b4 mm/kasan/report.c:398
kasan_report+0x150/0x1ac mm/kasan/report.c:495
kasan_check_range+0x264/0x2a4 mm/kasan/generic.c:189
memset+0x40/0x70 mm/kasan/shadow.c:44
__build_skb_around net/core/skbuff.c:294 [inline]
__alloc_skb+0x3c4/0x6e8 net/core/skbuff.c:527
alloc_skb include/linux/skbuff.h:1316 [inline]
igmpv3_newpack+0x104/0x1088 net/ipv4/igmp.c:359
add_grec+0x81c/0x1124 net/ipv4/igmp.c:534
igmpv3_send_cr net/ipv4/igmp.c:667 [inline]
igmp_ifc_timer_expire+0x1b0/0x1008 net/ipv4/igmp.c:810
call_timer_fn+0x1c0/0x9f0 kernel/time/timer.c:1474
expire_timers kernel/time/timer.c:1519 [inline]
__run_timers+0x54c/0x710 kernel/time/timer.c:1790
run_timer_softirq+0x28/0x4c kernel/time/timer.c:1803
_stext+0x380/0xfbc
____do_softirq+0x14/0x20 arch/arm64/kernel/irq.c:79
call_on_irq_stack+0x24/0x4c arch/arm64/kernel/entry.S:891
do_softirq_own_stack+0x20/0x2c arch/arm64/kernel/irq.c:84
invoke_softirq kernel/softirq.c:437 [inline]
__irq_exit_rcu+0x1c0/0x4cc kernel/softirq.c:683
irq_exit_rcu+0x14/0x78 kernel/softirq.c:695
el0_interrupt+0x7c/0x2e0 arch/arm64/kernel/entry-common.c:717
__el0_irq_handler_common+0x18/0x24 arch/arm64/kernel/entry-common.c:724
el0t_64_irq_handler+0x10/0x1c arch/arm64/kernel/entry-common.c:729
el0t_64_irq+0x1a0/0x1a4 arch/arm64/kernel/entry.S:584
Fixes: 12d6c1d3a2 ("skbuff: Proactively round up to kmalloc bucket size")
Reported-by: syzbot <syzkaller@googlegroups.com>
Reported-by: Kyle Zeng <zengyhkyle@gmail.com>
Signed-off-by: Eric Dumazet <edumazet@google.com>
Cc: Kees Cook <keescook@chromium.org>
Cc: Vlastimil Babka <vbabka@suse.cz>
Signed-off-by: David S. Miller <davem@davemloft.net>
Disable virtualization features on 82580 just as on i210/i211.
This avoids that virt functions are acidentally called on 82850.
Fixes: 55cac248ca ("igb: Add full support for 82580 devices")
Signed-off-by: Corinna Vinschen <vinschen@redhat.com>
Reviewed-by: Simon Horman <horms@kernel.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
scripts/documentation-file-ref-check reports warnings for (valid) cross-links
of form:
:ref:`Documentation/bpf/btf <BTF_Ext_Section>`
Adding extension to the file name helps to avoid the warning, e.g:
:ref:`Documentation/bpf/btf.rst <BTF_Ext_Section>`
Fixes: be4033d360 ("docs/bpf: Add description for CO-RE relocations")
Reported-by: kernel test robot <lkp@intel.com>
Signed-off-by: Eduard Zingerman <eddyz87@gmail.com>
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
Acked-by: Jiri Olsa <jolsa@kernel.org>
Closes: https://lore.kernel.org/oe-kbuild-all/202309010804.G3MpXo59-lkp@intel.com
Link: https://lore.kernel.org/bpf/20230901125935.487972-1-eddyz87@gmail.com
While commit 90f0074cd9 ("selftests/bpf: fix a CI failure caused by vsock sockmap test")
fixes a receive failure of vsock sockmap test, there is still a write failure:
Error: #211/79 sockmap_listen/sockmap VSOCK test_vsock_redir
Error: #211/79 sockmap_listen/sockmap VSOCK test_vsock_redir
./test_progs:vsock_unix_redir_connectible:1501: egress: write: Transport endpoint is not connected
vsock_unix_redir_connectible:FAIL:1501
./test_progs:vsock_unix_redir_connectible:1501: ingress: write: Transport endpoint is not connected
vsock_unix_redir_connectible:FAIL:1501
./test_progs:vsock_unix_redir_connectible:1501: egress: write: Transport endpoint is not connected
vsock_unix_redir_connectible:FAIL:1501
The reason is that the vsock connection in the test is set to ESTABLISHED state
by function virtio_transport_recv_pkt, which is executed in a workqueue thread,
so when the user space test thread runs before the workqueue thread, this
problem occurs.
To fix it, before writing the connection, wait for it to be connected.
Fixes: d61bd8c1fd ("selftests/bpf: add a test case for vsock sockmap")
Signed-off-by: Xu Kuohai <xukuohai@huawei.com>
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
Link: https://lore.kernel.org/bpf/20230901031037.3314007-1-xukuohai@huaweicloud.com
When EF10 RXDP firmware is operating in cut-through mode, packet length
is not known at the time the RX prefix is generated, so it is left as
zero and RX event merging is inhibited to ensure that the length is
available in the RX event. However, it has been found that in certain
circumstances the RX events for these packets still get merged,
meaning the driver cannot read the length from the RX event, and tries
to use the length from the prefix.
The resulting zero-length SKBs cause crashes in GRO since commit
1d11fa6967 ("net-gro: remove GRO_DROP"), so add a check to the driver
to detect these zero-length RX events and discard the packet.
Signed-off-by: Edward Cree <ecree.xilinx@gmail.com>
Reviewed-by: Eric Dumazet <edumazet@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Sriram Yagnaraman says:
====================
Avoid TCP resets when using ECMP for load-balancing between multiple servers.
All packets in the same flow (L3/L4 depending on multipath hash policy)
should be directed to the same target, but after [0]/[1] we see stray
packets directed towards other targets. This, for instance, causes RST
to be sent on TCP connections.
The first two patches solve the problem by ignoring route hints for
destinations that are part of multipath group, by using new SKB flags
for IPv4 and IPv6. The third patch is a selftest that tests the
scenario.
Thanks to Ido, for reviewing and suggesting a way forward in [2] and
also suggesting how to write a selftest for this.
v4->v5:
- Fixed review comments from Ido
v3->v4:
- Remove single path test
- Rebase to latest
v2->v3:
- Add NULL check for skb in fib6_select_path (Ido Schimmel)
- Use fib_tests.sh for selftest instead of the forwarding suite (Ido
Schimmel)
v1->v2:
- Update to commit messages describing the solution (Ido Schimmel)
- Use perf stat to count fib table lookups in selftest (Ido Schimmel)
====================
Signed-off-by: David S. Miller <davem@davemloft.net>
The test uses perf stat to count the number of fib:fib_table_lookup
tracepoint hits for IPv4 and the number of fib6:fib6_table_lookup for
IPv6. The measured count is checked to be within 5% of the total number
of packets sent via veth1.
Signed-off-by: Sriram Yagnaraman <sriram.yagnaraman@est.tech>
Signed-off-by: David S. Miller <davem@davemloft.net>
Route hints when the nexthop is part of a multipath group causes packets
in the same receive batch to be sent to the same nexthop irrespective of
the multipath hash of the packet. So, do not extract route hint for
packets whose destination is part of a multipath group.
A new SKB flag IP6SKB_MULTIPATH is introduced for this purpose, set the
flag when route is looked up in fib6_select_path() and use it in
ip6_can_use_hint() to check for the existence of the flag.
Fixes: 197dbf24e3 ("ipv6: introduce and uses route look hints for list input.")
Signed-off-by: Sriram Yagnaraman <sriram.yagnaraman@est.tech>
Reviewed-by: Ido Schimmel <idosch@nvidia.com>
Reviewed-by: David Ahern <dsahern@kernel.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
Route hints when the nexthop is part of a multipath group causes packets
in the same receive batch to be sent to the same nexthop irrespective of
the multipath hash of the packet. So, do not extract route hint for
packets whose destination is part of a multipath group.
A new SKB flag IPSKB_MULTIPATH is introduced for this purpose, set the
flag when route is looked up in ip_mkroute_input() and use it in
ip_extract_route_hint() to check for the existence of the flag.
Fixes: 02b2494161 ("ipv4: use dst hint for ipv4 list receive")
Signed-off-by: Sriram Yagnaraman <sriram.yagnaraman@est.tech>
Reviewed-by: Ido Schimmel <idosch@nvidia.com>
Reviewed-by: David Ahern <dsahern@kernel.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
Commit bf5c25d608 ("skbuff: in skb_segment, call zerocopy functions
once per nskb") added the call to zero copy functions in skb_segment().
The change introduced a bug in skb_segment() because skb_orphan_frags()
may possibly change the number of fragments or allocate new fragments
altogether leaving nrfrags and frag to point to the old values. This can
cause a panic with stacktrace like the one below.
[ 193.894380] BUG: kernel NULL pointer dereference, address: 00000000000000bc
[ 193.895273] CPU: 13 PID: 18164 Comm: vh-net-17428 Kdump: loaded Tainted: G O 5.15.123+ #26
[ 193.903919] RIP: 0010:skb_segment+0xb0e/0x12f0
[ 194.021892] Call Trace:
[ 194.027422] <TASK>
[ 194.072861] tcp_gso_segment+0x107/0x540
[ 194.082031] inet_gso_segment+0x15c/0x3d0
[ 194.090783] skb_mac_gso_segment+0x9f/0x110
[ 194.095016] __skb_gso_segment+0xc1/0x190
[ 194.103131] netem_enqueue+0x290/0xb10 [sch_netem]
[ 194.107071] dev_qdisc_enqueue+0x16/0x70
[ 194.110884] __dev_queue_xmit+0x63b/0xb30
[ 194.121670] bond_start_xmit+0x159/0x380 [bonding]
[ 194.128506] dev_hard_start_xmit+0xc3/0x1e0
[ 194.131787] __dev_queue_xmit+0x8a0/0xb30
[ 194.138225] macvlan_start_xmit+0x4f/0x100 [macvlan]
[ 194.141477] dev_hard_start_xmit+0xc3/0x1e0
[ 194.144622] sch_direct_xmit+0xe3/0x280
[ 194.147748] __dev_queue_xmit+0x54a/0xb30
[ 194.154131] tap_get_user+0x2a8/0x9c0 [tap]
[ 194.157358] tap_sendmsg+0x52/0x8e0 [tap]
[ 194.167049] handle_tx_zerocopy+0x14e/0x4c0 [vhost_net]
[ 194.173631] handle_tx+0xcd/0xe0 [vhost_net]
[ 194.176959] vhost_worker+0x76/0xb0 [vhost]
[ 194.183667] kthread+0x118/0x140
[ 194.190358] ret_from_fork+0x1f/0x30
[ 194.193670] </TASK>
In this case calling skb_orphan_frags() updated nr_frags leaving nrfrags
local variable in skb_segment() stale. This resulted in the code hitting
i >= nrfrags prematurely and trying to move to next frag_skb using
list_skb pointer, which was NULL, and caused kernel panic. Move the call
to zero copy functions before using frags and nr_frags.
Fixes: bf5c25d608 ("skbuff: in skb_segment, call zerocopy functions once per nskb")
Signed-off-by: Mohamed Khalfella <mkhalfella@purestorage.com>
Reported-by: Amit Goyal <agoyal@purestorage.com>
Cc: stable@vger.kernel.org
Reviewed-by: Eric Dumazet <edumazet@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Eric Dumazet says:
====================
net: another round of data-race annotations
Series inspired by some syzbot reports, taking care
of 4 socket fields that can be read locklessly.
====================
Signed-off-by: David S. Miller <davem@davemloft.net>
