Alexei Starovoitov says:
====================
pull-request: bpf-next 2020-01-22
The following pull-request contains BPF updates for your *net-next* tree.
We've added 92 non-merge commits during the last 16 day(s) which contain
a total of 320 files changed, 7532 insertions(+), 1448 deletions(-).
The main changes are:
1) function by function verification and program extensions from Alexei.
2) massive cleanup of selftests/bpf from Toke and Andrii.
3) batched bpf map operations from Brian and Yonghong.
4) tcp congestion control in bpf from Martin.
5) bulking for non-map xdp_redirect form Toke.
6) bpf_send_signal_thread helper from Yonghong.
====================
Signed-off-by: David S. Miller <davem@davemloft.net>
XDP sockets use the default implementation of struct sock's
sk_data_ready callback, which is sock_def_readable(). This function
is called in the XDP socket fast-path, and involves a retpoline. By
letting sock_def_readable() have external linkage, and being called
directly, the retpoline can be avoided.
Signed-off-by: Björn Töpel <bjorn.topel@intel.com>
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
Link: https://lore.kernel.org/bpf/20200120092917.13949-1-bjorn.topel@gmail.com
Otherwise we will find stray/unexpected/old extensions value on next
iteration.
On tcp_write_xmit() we can end-up splitting an already queued skb in two
parts, via tso_fragment(). The newly created skb can be allocated via
the tx cache and an upper layer will not be aware of it, so that upper
layer cannot set the ext properly.
Resetting the ext on recycle ensures that stale data is not propagated
in to packet headers or elsewhere.
An alternative would be add an additional hook in tso_fragment() or in
sk_stream_alloc_skb() to init the ext for upper layers that need it.
Co-developed-by: Florian Westphal <fw@strlen.de>
Signed-off-by: Florian Westphal <fw@strlen.de>
Signed-off-by: Paolo Abeni <pabeni@redhat.com>
Reviewed-by: Eric Dumazet <edumazet@google.com>
Signed-off-by: Mat Martineau <mathew.j.martineau@linux.intel.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Match the 16-bit width of skbuff->protocol. Fills an 8-bit hole so
sizeof(struct sock) does not change.
Also take care of BPF field access for sk_type/sk_protocol. Both of them
are now outside the bitfield, so we can use load instructions without
further shifting/masking.
v5 -> v6:
- update eBPF accessors, too (Intel's kbuild test robot)
v2 -> v3:
- keep 'sk_type' 2 bytes aligned (Eric)
v1 -> v2:
- preserve sk_pacing_shift as bit field (Eric)
Cc: Alexei Starovoitov <ast@kernel.org>
Cc: Daniel Borkmann <daniel@iogearbox.net>
Cc: bpf@vger.kernel.org
Co-developed-by: Paolo Abeni <pabeni@redhat.com>
Signed-off-by: Paolo Abeni <pabeni@redhat.com>
Co-developed-by: Matthieu Baerts <matthieu.baerts@tessares.net>
Signed-off-by: Matthieu Baerts <matthieu.baerts@tessares.net>
Signed-off-by: Mat Martineau <mathew.j.martineau@linux.intel.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
SK_PROTOCOL_MAX is only used in two places, for DECNet and AX.25. The
limits have more to do with the those protocol definitions than they do
with the data type of sk_protocol, so remove SK_PROTOCOL_MAX and use
U8_MAX directly.
Reviewed-by: Eric Dumazet <edumazet@google.com>
Signed-off-by: Mat Martineau <mathew.j.martineau@linux.intel.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Pull networking fixes from David Miller:
1) Several nf_flow_table_offload fixes from Pablo Neira Ayuso,
including adding a missing ipv6 match description.
2) Several heap overflow fixes in mwifiex from qize wang and Ganapathi
Bhat.
3) Fix uninit value in bond_neigh_init(), from Eric Dumazet.
4) Fix non-ACPI probing of nxp-nci, from Stephan Gerhold.
5) Fix use after free in tipc_disc_rcv(), from Tuong Lien.
6) Enforce limit of 33 tail calls in mips and riscv JIT, from Paul
Chaignon.
7) Multicast MAC limit test is off by one in qede, from Manish Chopra.
8) Fix established socket lookup race when socket goes from
TCP_ESTABLISHED to TCP_LISTEN, because there lacks an intervening
RCU grace period. From Eric Dumazet.
9) Don't send empty SKBs from tcp_write_xmit(), also from Eric Dumazet.
10) Fix active backup transition after link failure in bonding, from
Mahesh Bandewar.
11) Avoid zero sized hash table in gtp driver, from Taehee Yoo.
12) Fix wrong interface passed to ->mac_link_up(), from Russell King.
13) Fix DSA egress flooding settings in b53, from Florian Fainelli.
14) Memory leak in gmac_setup_txqs(), from Navid Emamdoost.
15) Fix double free in dpaa2-ptp code, from Ioana Ciornei.
16) Reject invalid MTU values in stmmac, from Jose Abreu.
17) Fix refcount leak in error path of u32 classifier, from Davide
Caratti.
18) Fix regression causing iwlwifi firmware crashes on boot, from Anders
Kaseorg.
19) Fix inverted return value logic in llc2 code, from Chan Shu Tak.
20) Disable hardware GRO when XDP is attached to qede, frm Manish
Chopra.
21) Since we encode state in the low pointer bits, dst metrics must be
at least 4 byte aligned, which is not necessarily true on m68k. Add
annotations to fix this, from Geert Uytterhoeven.
* git://git.kernel.org/pub/scm/linux/kernel/git/netdev/net: (160 commits)
sfc: Include XDP packet headroom in buffer step size.
sfc: fix channel allocation with brute force
net: dst: Force 4-byte alignment of dst_metrics
selftests: pmtu: fix init mtu value in description
hv_netvsc: Fix unwanted rx_table reset
net: phy: ensure that phy IDs are correctly typed
mod_devicetable: fix PHY module format
qede: Disable hardware gro when xdp prog is installed
net: ena: fix issues in setting interrupt moderation params in ethtool
net: ena: fix default tx interrupt moderation interval
net/smc: unregister ib devices in reboot_event
net: stmmac: platform: Fix MDIO init for platforms without PHY
llc2: Fix return statement of llc_stat_ev_rx_null_dsap_xid_c (and _test_c)
net: hisilicon: Fix a BUG trigered by wrong bytes_compl
net: dsa: ksz: use common define for tag len
s390/qeth: don't return -ENOTSUPP to userspace
s390/qeth: fix promiscuous mode after reset
s390/qeth: handle error due to unsupported transport mode
cxgb4: fix refcount init for TC-MQPRIO offload
tc-testing: initial tdc selftests for cls_u32
...
sk->sk_pacing_shift can be read and written without lock
synchronization. This patch adds annotations to
document this fact and avoid future syzbot complains.
This might also avoid unexpected false sharing
in sk_pacing_shift_update(), as the compiler
could remove the conditional check and always
write over sk->sk_pacing_shift :
if (sk->sk_pacing_shift != val)
sk->sk_pacing_shift = val;
Signed-off-by: Eric Dumazet <edumazet@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Michal Kubecek and Firo Yang did a very nice analysis of crashes
happening in __inet_lookup_established().
Since a TCP socket can go from TCP_ESTABLISH to TCP_LISTEN
(via a close()/socket()/listen() cycle) without a RCU grace period,
I should not have changed listeners linkage in their hash table.
They must use the nulls protocol (Documentation/RCU/rculist_nulls.txt),
so that a lookup can detect a socket in a hash list was moved in
another one.
Since we added code in commit d296ba60d8 ("soreuseport: Resolve
merge conflict for v4/v6 ordering fix"), we have to add
hlist_nulls_add_tail_rcu() helper.
Fixes: 3b24d854cb ("tcp/dccp: do not touch listener sk_refcnt under synflood")
Signed-off-by: Eric Dumazet <edumazet@google.com>
Reported-by: Michal Kubecek <mkubecek@suse.cz>
Reported-by: Firo Yang <firo.yang@suse.com>
Reviewed-by: Michal Kubecek <mkubecek@suse.cz>
Link: https://lore.kernel.org/netdev/20191120083919.GH27852@unicorn.suse.cz/
Signed-off-by: Jakub Kicinski <jakub.kicinski@netronome.com>
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 locking updates from Ingo Molnar:
"The main changes in this cycle were:
- A comprehensive rewrite of the robust/PI futex code's exit handling
to fix various exit races. (Thomas Gleixner et al)
- Rework the generic REFCOUNT_FULL implementation using
atomic_fetch_* operations so that the performance impact of the
cmpxchg() loops is mitigated for common refcount operations.
With these performance improvements the generic implementation of
refcount_t should be good enough for everybody - and this got
confirmed by performance testing, so remove ARCH_HAS_REFCOUNT and
REFCOUNT_FULL entirely, leaving the generic implementation enabled
unconditionally. (Will Deacon)
- Other misc changes, fixes, cleanups"
* 'locking-core-for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/tip/tip: (27 commits)
lkdtm: Remove references to CONFIG_REFCOUNT_FULL
locking/refcount: Remove unused 'refcount_error_report()' function
locking/refcount: Consolidate implementations of refcount_t
locking/refcount: Consolidate REFCOUNT_{MAX,SATURATED} definitions
locking/refcount: Move saturation warnings out of line
locking/refcount: Improve performance of generic REFCOUNT_FULL code
locking/refcount: Move the bulk of the REFCOUNT_FULL implementation into the <linux/refcount.h> header
locking/refcount: Remove unused refcount_*_checked() variants
locking/refcount: Ensure integer operands are treated as signed
locking/refcount: Define constants for saturation and max refcount values
futex: Prevent exit livelock
futex: Provide distinct return value when owner is exiting
futex: Add mutex around futex exit
futex: Provide state handling for exec() as well
futex: Sanitize exit state handling
futex: Mark the begin of futex exit explicitly
futex: Set task::futex_state to DEAD right after handling futex exit
futex: Split futex_mm_release() for exit/exec
exit/exec: Seperate mm_release()
futex: Replace PF_EXITPIDONE with a state
...
One conflict in the BPF samples Makefile, some fixes in 'net' whilst
we were converting over to Makefile.target rules in 'net-next'.
Signed-off-by: David S. Miller <davem@davemloft.net>
sk->sk_max_ack_backlog can be read without any lock being held
at least in TCP/DCCP cases.
We need to use READ_ONCE()/WRITE_ONCE() to avoid load/store tearing
and/or potential KCSAN warnings.
Signed-off-by: Eric Dumazet <edumazet@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
sk->sk_ack_backlog can be read without any lock being held.
We need to use READ_ONCE()/WRITE_ONCE() to avoid load/store tearing
and/or potential KCSAN warnings.
Signed-off-by: Eric Dumazet <edumazet@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
There are common instances of the following construct :
if (n->confirmed != now)
n->confirmed = now;
A C compiler could legally remove the conditional.
Use READ_ONCE()/WRITE_ONCE() to avoid this problem.
Signed-off-by: Eric Dumazet <edumazet@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Add a couple of READ_ONCE() and WRITE_ONCE() to prevent
load-tearing and store-tearing in sock_read_timestamp()
and sock_write_timestamp()
This might prevent another KCSAN report.
Fixes: 3a0ed3e961 ("sock: Make sock->sk_stamp thread-safe")
Signed-off-by: Eric Dumazet <edumazet@google.com>
Cc: Deepa Dinamani <deepa.kernel@gmail.com>
Acked-by: Deepa Dinamani <deepa.kernel@gmail.com>
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>
The only smc-related reference in net/sock.h is struct smc_hashinfo.
But just its address is refered to. Thus there is no need for the
include of net/smc.h. Remove it.
Suggested-by: Jakub Kicinski <jakub.kicinski@netronome.com>
Reviewed by: Karsten Graul <kgraul@linux.ibm.com>
Signed-off-by: Ursula Braun <ubraun@linux.ibm.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
sk_page_frag() optimizes skb_frag allocations by using per-task
skb_frag cache when it knows it's the only user. The condition is
determined by seeing whether the socket allocation mask allows
blocking - if the allocation may block, it obviously owns the task's
context and ergo exclusively owns current->task_frag.
Unfortunately, this misses recursion through memory reclaim path.
Please take a look at the following backtrace.
[2] RIP: 0010:tcp_sendmsg_locked+0xccf/0xe10
...
tcp_sendmsg+0x27/0x40
sock_sendmsg+0x30/0x40
sock_xmit.isra.24+0xa1/0x170 [nbd]
nbd_send_cmd+0x1d2/0x690 [nbd]
nbd_queue_rq+0x1b5/0x3b0 [nbd]
__blk_mq_try_issue_directly+0x108/0x1b0
blk_mq_request_issue_directly+0xbd/0xe0
blk_mq_try_issue_list_directly+0x41/0xb0
blk_mq_sched_insert_requests+0xa2/0xe0
blk_mq_flush_plug_list+0x205/0x2a0
blk_flush_plug_list+0xc3/0xf0
[1] blk_finish_plug+0x21/0x2e
_xfs_buf_ioapply+0x313/0x460
__xfs_buf_submit+0x67/0x220
xfs_buf_read_map+0x113/0x1a0
xfs_trans_read_buf_map+0xbf/0x330
xfs_btree_read_buf_block.constprop.42+0x95/0xd0
xfs_btree_lookup_get_block+0x95/0x170
xfs_btree_lookup+0xcc/0x470
xfs_bmap_del_extent_real+0x254/0x9a0
__xfs_bunmapi+0x45c/0xab0
xfs_bunmapi+0x15/0x30
xfs_itruncate_extents_flags+0xca/0x250
xfs_free_eofblocks+0x181/0x1e0
xfs_fs_destroy_inode+0xa8/0x1b0
destroy_inode+0x38/0x70
dispose_list+0x35/0x50
prune_icache_sb+0x52/0x70
super_cache_scan+0x120/0x1a0
do_shrink_slab+0x120/0x290
shrink_slab+0x216/0x2b0
shrink_node+0x1b6/0x4a0
do_try_to_free_pages+0xc6/0x370
try_to_free_mem_cgroup_pages+0xe3/0x1e0
try_charge+0x29e/0x790
mem_cgroup_charge_skmem+0x6a/0x100
__sk_mem_raise_allocated+0x18e/0x390
__sk_mem_schedule+0x2a/0x40
[0] tcp_sendmsg_locked+0x8eb/0xe10
tcp_sendmsg+0x27/0x40
sock_sendmsg+0x30/0x40
___sys_sendmsg+0x26d/0x2b0
__sys_sendmsg+0x57/0xa0
do_syscall_64+0x42/0x100
entry_SYSCALL_64_after_hwframe+0x44/0xa9
In [0], tcp_send_msg_locked() was using current->page_frag when it
called sk_wmem_schedule(). It already calculated how many bytes can
be fit into current->page_frag. Due to memory pressure,
sk_wmem_schedule() called into memory reclaim path which called into
xfs and then IO issue path. Because the filesystem in question is
backed by nbd, the control goes back into the tcp layer - back into
tcp_sendmsg_locked().
nbd sets sk_allocation to (GFP_NOIO | __GFP_MEMALLOC) which makes
sense - it's in the process of freeing memory and wants to be able to,
e.g., drop clean pages to make forward progress. However, this
confused sk_page_frag() called from [2]. Because it only tests
whether the allocation allows blocking which it does, it now thinks
current->page_frag can be used again although it already was being
used in [0].
After [2] used current->page_frag, the offset would be increased by
the used amount. When the control returns to [0],
current->page_frag's offset is increased and the previously calculated
number of bytes now may overrun the end of allocated memory leading to
silent memory corruptions.
Fix it by adding gfpflags_normal_context() which tests sleepable &&
!reclaim and use it to determine whether to use current->task_frag.
v2: Eric didn't like gfp flags being tested twice. Introduce a new
helper gfpflags_normal_context() and combine the two tests.
Signed-off-by: Tejun Heo <tj@kernel.org>
Cc: Josef Bacik <josef@toxicpanda.com>
Cc: Eric Dumazet <eric.dumazet@gmail.com>
Cc: stable@vger.kernel.org
Signed-off-by: David S. Miller <davem@davemloft.net>
For the sake of tcp_poll(), there are few places where we fetch
sk->sk_wmem_queued while this field can change from IRQ or other cpu.
We need to add READ_ONCE() annotations, and also make sure write
sides use corresponding WRITE_ONCE() to avoid store-tearing.
sk_wmem_queued_add() helper is added so that we can in
the future convert to ADD_ONCE() or equivalent if/when
available.
Signed-off-by: Eric Dumazet <edumazet@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
For the sake of tcp_poll(), there are few places where we fetch
sk->sk_sndbuf while this field can change from IRQ or other cpu.
We need to add READ_ONCE() annotations, and also make sure write
sides use corresponding WRITE_ONCE() to avoid store-tearing.
Note that other transports probably need similar fixes.
Signed-off-by: Eric Dumazet <edumazet@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
sock_rcvlowat() or int_sk_rcvlowat() might be called without the socket
lock for example from tcp_poll().
Use READ_ONCE() to document the fact that other cpus might change
sk->sk_rcvlowat under us and avoid KCSAN splats.
Use WRITE_ONCE() on write sides too.
Signed-off-by: Eric Dumazet <edumazet@google.com>
Signed-off-by: Jakub Kicinski <jakub.kicinski@netronome.com>
sk_validate_xmit_skb() and drivers depend on the sk member of
struct sk_buff to identify segments requiring encryption.
Any operation which removes or does not preserve the original TLS
socket such as skb_orphan() or skb_clone() will cause clear text
leaks.
Make the TCP socket underlying an offloaded TLS connection
mark all skbs as decrypted, if TLS TX is in offload mode.
Then in sk_validate_xmit_skb() catch skbs which have no socket
(or a socket with no validation) and decrypted flag set.
Note that CONFIG_SOCK_VALIDATE_XMIT, CONFIG_TLS_DEVICE and
sk->sk_validate_xmit_skb are slightly interchangeable right now,
they all imply TLS offload. The new checks are guarded by
CONFIG_TLS_DEVICE because that's the option guarding the
sk_buff->decrypted member.
Second, smaller issue with orphaning is that it breaks
the guarantee that packets will be delivered to device
queues in-order. All TLS offload drivers depend on that
scheduling property. This means skb_orphan_partial()'s
trick of preserving partial socket references will cause
issues in the drivers. We need a full orphan, and as a
result netem delay/throttling will cause all TLS offload
skbs to be dropped.
Reusing the sk_buff->decrypted flag also protects from
leaking clear text when incoming, decrypted skb is redirected
(e.g. by TC).
See commit 0608c69c9a ("bpf: sk_msg, sock{map|hash} redirect
through ULP") for justification why the internal flag is safe.
The only location which could leak the flag in is tcp_bpf_sendmsg(),
which is taken care of by clearing the previously unused bit.
v2:
- remove superfluous decrypted mark copy (Willem);
- remove the stale doc entry (Boris);
- rely entirely on EOR marking to prevent coalescing (Boris);
- use an internal sendpages flag instead of marking the socket
(Boris).
v3 (Willem):
- reorganize the can_skb_orphan_partial() condition;
- fix the flag leak-in through tcp_bpf_sendmsg.
Signed-off-by: Jakub Kicinski <jakub.kicinski@netronome.com>
Acked-by: Willem de Bruijn <willemb@google.com>
Reviewed-by: Boris Pismenny <borisp@mellanox.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
socket->wq is assign-once, set when we are initializing both
struct socket it's in and struct socket_wq it points to. As the
matter of fact, the only reason for separate allocation was the
ability to RCU-delay freeing of socket_wq. RCU-delaying the
freeing of socket itself gets rid of that need, so we can just
fold struct socket_wq into the end of struct socket and simplify
the life both for sock_alloc_inode() (one allocation instead of
two) and for tun/tap oddballs, where we used to embed struct socket
and struct socket_wq into the same structure (now - embedding just
the struct socket).
Note that reference to struct socket_wq in struct sock does remain
a reference - that's unchanged.
Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
Signed-off-by: David S. Miller <davem@davemloft.net>
>From linux-3.7, (commit 5640f76858 "net: use a per task frag
allocator") TCP sendmsg() has preferred using order-3 allocations.
While it gives good results for most cases, we had reports
that heavy uses of TCP over loopback were hitting a spinlock
contention in page allocations/freeing.
This commits adds a sysctl so that admins can opt-in
for order-0 allocations. Hopefully mm layer might optimize
order-3 allocations in the future since it could give us
a nice boost (see 8 lines of following benchmark)
The following benchmark shows a win when more than 8 TCP_STREAM
threads are running (56 x86 cores server in my tests)
for thr in {1..30}
do
sysctl -wq net.core.high_order_alloc_disable=0
T0=`./super_netperf $thr -H 127.0.0.1 -l 15`
sysctl -wq net.core.high_order_alloc_disable=1
T1=`./super_netperf $thr -H 127.0.0.1 -l 15`
echo $thr:$T0:$T1
done
1: 49979: 37267
2: 98745: 76286
3: 141088: 110051
4: 177414: 144772
5: 197587: 173563
6: 215377: 208448
7: 241061: 234087
8: 267155: 263373
9: 295069: 297402
10: 312393: 335213
11: 340462: 368778
12: 371366: 403954
13: 412344: 443713
14: 426617: 473580
15: 474418: 507861
16: 503261: 538539
17: 522331: 563096
18: 532409: 567084
19: 550824: 605240
20: 525493: 641988
21: 564574: 665843
22: 567349: 690868
23: 583846: 710917
24: 588715: 736306
25: 603212: 763494
26: 604083: 792654
27: 602241: 796450
28: 604291: 797993
29: 611610: 833249
30: 577356: 841062
Signed-off-by: Eric Dumazet <edumazet@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Feng Tang reported a performance regression after introduction
of per TCP socket tx/rx caches, for TCP over loopback (netperf)
There is high chance the regression is caused by a change on
how well the 32 KB per-thread page (current->task_frag) can
be recycled, and lack of pcp caches for order-3 pages.
I could not reproduce the regression myself, cpus all being
spinning on the mm spinlocks for page allocs/freeing, regardless
of enabling or disabling the per tcp socket caches.
It seems best to disable the feature by default, and let
admins enabling it.
MM layer either needs to provide scalable order-3 pages
allocations, or could attempt a trylock on zone->lock if
the caller only attempts to get a high-order page and is
able to fallback to order-0 ones in case of pressure.
Tests run on a 56 cores host (112 hyper threads)
- 35.49% netperf [kernel.vmlinux] [k] queued_spin_lock_slowpath
- 35.49% queued_spin_lock_slowpath
- 18.18% get_page_from_freelist
- __alloc_pages_nodemask
- 18.18% alloc_pages_current
skb_page_frag_refill
sk_page_frag_refill
tcp_sendmsg_locked
tcp_sendmsg
inet_sendmsg
sock_sendmsg
__sys_sendto
__x64_sys_sendto
do_syscall_64
entry_SYSCALL_64_after_hwframe
__libc_send
+ 17.31% __free_pages_ok
+ 31.43% swapper [kernel.vmlinux] [k] intel_idle
+ 9.12% netperf [kernel.vmlinux] [k] copy_user_enhanced_fast_string
+ 6.53% netserver [kernel.vmlinux] [k] copy_user_enhanced_fast_string
+ 0.69% netserver [kernel.vmlinux] [k] queued_spin_lock_slowpath
+ 0.68% netperf [kernel.vmlinux] [k] skb_release_data
+ 0.52% netperf [kernel.vmlinux] [k] tcp_sendmsg_locked
0.46% netperf [kernel.vmlinux] [k] _raw_spin_lock_irqsave
Fixes: 472c2e07ee ("tcp: add one skb cache for tx")
Signed-off-by: Eric Dumazet <edumazet@google.com>
Reported-by: Feng Tang <feng.tang@intel.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Instead of relying on rps_needed, it is safer to use a separate
static key, since we do not want to enable TCP rx_skb_cache
by default. This feature can cause huge increase of memory
usage on hosts with millions of sockets.
Signed-off-by: Eric Dumazet <edumazet@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Based on 1 normalized pattern(s):
this program is free software you can redistribute it and or modify
it under the terms of the gnu general public license as published by
the free software foundation either version 2 of the license or at
your option any later version
extracted by the scancode license scanner the SPDX license identifier
GPL-2.0-or-later
has been chosen to replace the boilerplate/reference in 3029 file(s).
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Reviewed-by: Allison Randal <allison@lohutok.net>
Cc: linux-spdx@vger.kernel.org
Link: https://lkml.kernel.org/r/20190527070032.746973796@linutronix.de
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
It is illegal to change arbitrary fields in skb_shared_info if the
skb is cloned.
Before calling skb_zcopy_clear() we need to ensure this rule,
therefore we need to move the test from sk_stream_alloc_skb()
to sk_wmem_free_skb()
Fixes: 4f661542a4 ("tcp: fix zerocopy and notsent_lowat issues")
Signed-off-by: Eric Dumazet <edumazet@google.com>
Diagnosed-by: Willem de Bruijn <willemb@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
After allowing a bpf prog to
- directly read the skb->sk ptr
- get the fullsock bpf_sock by "bpf_sk_fullsock()"
- get the bpf_tcp_sock by "bpf_tcp_sock()"
- get the listener sock by "bpf_get_listener_sock()"
- avoid duplicating the fields of "(bpf_)sock" and "(bpf_)tcp_sock"
into different bpf running context.
this patch is another effort to make bpf's network programming
more intuitive to do (together with memory and performance benefit).
When bpf prog needs to store data for a sk, the current practice is to
define a map with the usual 4-tuples (src/dst ip/port) as the key.
If multiple bpf progs require to store different sk data, multiple maps
have to be defined. Hence, wasting memory to store the duplicated
keys (i.e. 4 tuples here) in each of the bpf map.
[ The smallest key could be the sk pointer itself which requires
some enhancement in the verifier and it is a separate topic. ]
Also, the bpf prog needs to clean up the elem when sk is freed.
Otherwise, the bpf map will become full and un-usable quickly.
The sk-free tracking currently could be done during sk state
transition (e.g. BPF_SOCK_OPS_STATE_CB).
The size of the map needs to be predefined which then usually ended-up
with an over-provisioned map in production. Even the map was re-sizable,
while the sk naturally come and go away already, this potential re-size
operation is arguably redundant if the data can be directly connected
to the sk itself instead of proxy-ing through a bpf map.
This patch introduces sk->sk_bpf_storage to provide local storage space
at sk for bpf prog to use. The space will be allocated when the first bpf
prog has created data for this particular sk.
The design optimizes the bpf prog's lookup (and then optionally followed by
an inline update). bpf_spin_lock should be used if the inline update needs
to be protected.
BPF_MAP_TYPE_SK_STORAGE:
-----------------------
To define a bpf "sk-local-storage", a BPF_MAP_TYPE_SK_STORAGE map (new in
this patch) needs to be created. Multiple BPF_MAP_TYPE_SK_STORAGE maps can
be created to fit different bpf progs' needs. The map enforces
BTF to allow printing the sk-local-storage during a system-wise
sk dump (e.g. "ss -ta") in the future.
The purpose of a BPF_MAP_TYPE_SK_STORAGE map is not for lookup/update/delete
a "sk-local-storage" data from a particular sk.
Think of the map as a meta-data (or "type") of a "sk-local-storage". This
particular "type" of "sk-local-storage" data can then be stored in any sk.
The main purposes of this map are mostly:
1. Define the size of a "sk-local-storage" type.
2. Provide a similar syscall userspace API as the map (e.g. lookup/update,
map-id, map-btf...etc.)
3. Keep track of all sk's storages of this "type" and clean them up
when the map is freed.
sk->sk_bpf_storage:
------------------
The main lookup/update/delete is done on sk->sk_bpf_storage (which
is a "struct bpf_sk_storage"). When doing a lookup,
the "map" pointer is now used as the "key" to search on the
sk_storage->list. The "map" pointer is actually serving
as the "type" of the "sk-local-storage" that is being
requested.
To allow very fast lookup, it should be as fast as looking up an
array at a stable-offset. At the same time, it is not ideal to
set a hard limit on the number of sk-local-storage "type" that the
system can have. Hence, this patch takes a cache approach.
The last search result from sk_storage->list is cached in
sk_storage->cache[] which is a stable sized array. Each
"sk-local-storage" type has a stable offset to the cache[] array.
In the future, a map's flag could be introduced to do cache
opt-out/enforcement if it became necessary.
The cache size is 16 (i.e. 16 types of "sk-local-storage").
Programs can share map. On the program side, having a few bpf_progs
running in the networking hotpath is already a lot. The bpf_prog
should have already consolidated the existing sock-key-ed map usage
to minimize the map lookup penalty. 16 has enough runway to grow.
All sk-local-storage data will be removed from sk->sk_bpf_storage
during sk destruction.
bpf_sk_storage_get() and bpf_sk_storage_delete():
------------------------------------------------
Instead of using bpf_map_(lookup|update|delete)_elem(),
the bpf prog needs to use the new helper bpf_sk_storage_get() and
bpf_sk_storage_delete(). The verifier can then enforce the
ARG_PTR_TO_SOCKET argument. The bpf_sk_storage_get() also allows to
"create" new elem if one does not exist in the sk. It is done by
the new BPF_SK_STORAGE_GET_F_CREATE flag. An optional value can also be
provided as the initial value during BPF_SK_STORAGE_GET_F_CREATE.
The BPF_MAP_TYPE_SK_STORAGE also supports bpf_spin_lock. Together,
it has eliminated the potential use cases for an equivalent
bpf_map_update_elem() API (for bpf_prog) in this patch.
Misc notes:
----------
1. map_get_next_key is not supported. From the userspace syscall
perspective, the map has the socket fd as the key while the map
can be shared by pinned-file or map-id.
Since btf is enforced, the existing "ss" could be enhanced to pretty
print the local-storage.
Supporting a kernel defined btf with 4 tuples as the return key could
be explored later also.
2. The sk->sk_lock cannot be acquired. Atomic operations is used instead.
e.g. cmpxchg is done on the sk->sk_bpf_storage ptr.
Please refer to the source code comments for the details in
synchronization cases and considerations.
3. The mem is charged to the sk->sk_omem_alloc as the sk filter does.
Benchmark:
---------
Here is the benchmark data collected by turning on
the "kernel.bpf_stats_enabled" sysctl.
Two bpf progs are tested:
One bpf prog with the usual bpf hashmap (max_entries = 8192) with the
sk ptr as the key. (verifier is modified to support sk ptr as the key
That should have shortened the key lookup time.)
Another bpf prog is with the new BPF_MAP_TYPE_SK_STORAGE.
Both are storing a "u32 cnt", do a lookup on "egress_skb/cgroup" for
each egress skb and then bump the cnt. netperf is used to drive
data with 4096 connected UDP sockets.
BPF_MAP_TYPE_HASH with a modifier verifier (152ns per bpf run)
27: cgroup_skb name egress_sk_map tag 74f56e832918070b run_time_ns 58280107540 run_cnt 381347633
loaded_at 2019-04-15T13:46:39-0700 uid 0
xlated 344B jited 258B memlock 4096B map_ids 16
btf_id 5
BPF_MAP_TYPE_SK_STORAGE in this patch (66ns per bpf run)
30: cgroup_skb name egress_sk_stora tag d4aa70984cc7bbf6 run_time_ns 25617093319 run_cnt 390989739
loaded_at 2019-04-15T13:47:54-0700 uid 0
xlated 168B jited 156B memlock 4096B map_ids 17
btf_id 6
Here is a high-level picture on how are the objects organized:
sk
┌──────┐
│ │
│ │
│ │
│*sk_bpf_storage─────▶ bpf_sk_storage
└──────┘ ┌───────┐
┌───────────┤ list │
│ │ │
│ │ │
│ │ │
│ └───────┘
│
│ elem
│ ┌────────┐
├─▶│ snode │
│ ├────────┤
│ │ data │ bpf_map
│ ├────────┤ ┌─────────┐
│ │map_node│◀─┬─────┤ list │
│ └────────┘ │ │ │
│ │ │ │
│ elem │ │ │
│ ┌────────┐ │ └─────────┘
└─▶│ snode │ │
├────────┤ │
bpf_map │ data │ │
┌─────────┐ ├────────┤ │
│ list ├───────▶│map_node│ │
│ │ └────────┘ │
│ │ │
│ │ elem │
└─────────┘ ┌────────┐ │
┌─▶│ snode │ │
│ ├────────┤ │
│ │ data │ │
│ ├────────┤ │
│ │map_node│◀─┘
│ └────────┘
│
│
│ ┌───────┐
sk └──────────│ list │
┌──────┐ │ │
│ │ │ │
│ │ │ │
│ │ └───────┘
│*sk_bpf_storage───────▶bpf_sk_storage
└──────┘
Signed-off-by: Martin KaFai Lau <kafai@fb.com>
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
The SIOCGSTAMP/SIOCGSTAMPNS ioctl commands are implemented by many
socket protocol handlers, and all of those end up calling the same
sock_get_timestamp()/sock_get_timestampns() helper functions, which
results in a lot of duplicate code.
With the introduction of 64-bit time_t on 32-bit architectures, this
gets worse, as we then need four different ioctl commands in each
socket protocol implementation.
To simplify that, let's add a new .gettstamp() operation in
struct proto_ops, and move ioctl implementation into the common
sock_ioctl()/compat_sock_ioctl_trans() functions that these all go
through.
We can reuse the sock_get_timestamp() implementation, but generalize
it so it can deal with both native and compat mode, as well as
timeval and timespec structures.
Acked-by: Stefan Schmidt <stefan@datenfreihafen.org>
Acked-by: Neil Horman <nhorman@tuxdriver.com>
Acked-by: Marc Kleine-Budde <mkl@pengutronix.de>
Link: https://lore.kernel.org/lkml/CAK8P3a038aDQQotzua_QtKGhq8O9n+rdiz2=WDCp82ys8eUT+A@mail.gmail.com/
Signed-off-by: Arnd Bergmann <arnd@arndb.de>
Acked-by: Willem de Bruijn <willemb@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
fcntl(fd, F_SETOWN, getpid()) selects the recipient of SIGURG signals
that are delivered when out-of-band data arrives on socket fd.
If an SMC socket program makes use of such an fcntl() call, it fails
in case of fallback to TCP-mode. In case of fallback the traffic is
processed with the internal TCP socket. Propagating field "file" from the
SMC socket to the internal TCP socket fixes the issue.
Reviewed-by: Karsten Graul <kgraul@linux.ibm.com>
Signed-off-by: Ursula Braun <ubraun@linux.ibm.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
My recent patch had at least three problems :
1) TX zerocopy wants notification when skb is acknowledged,
thus we need to call skb_zcopy_clear() if the skb is
cached into sk->sk_tx_skb_cache
2) Some applications might expect precise EPOLLOUT
notifications, so we need to update sk->sk_wmem_queued
and call sk_mem_uncharge() from sk_wmem_free_skb()
in all cases. The SOCK_QUEUE_SHRUNK flag must also be set.
3) Reuse of saved skb should have used skb_cloned() instead
of simply checking if the fast clone has been freed.
Fixes: 472c2e07ee ("tcp: add one skb cache for tx")
Signed-off-by: Eric Dumazet <edumazet@google.com>
Cc: Willem de Bruijn <willemb@google.com>
Cc: Soheil Hassas Yeganeh <soheil@google.com>
Acked-by: Soheil Hassas Yeganeh <soheil@google.com>
Tested-by: Holger Hoffstätte <holger@applied-asynchrony.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Often times, recvmsg() system calls and BH handling for a particular
TCP socket are done on different cpus.
This means the incoming skb had to be allocated on a cpu,
but freed on another.
This incurs a high spinlock contention in slab layer for small rpc,
but also a high number of cache line ping pongs for larger packets.
A full size GRO packet might use 45 page fragments, meaning
that up to 45 put_page() can be involved.
More over performing the __kfree_skb() in the recvmsg() context
adds a latency for user applications, and increase probability
of trapping them in backlog processing, since the BH handler
might found the socket owned by the user.
This patch, combined with the prior one increases the rpc
performance by about 10 % on servers with large number of cores.
(tcp_rr workload with 10,000 flows and 112 threads reach 9 Mpps
instead of 8 Mpps)
This also increases single bulk flow performance on 40Gbit+ links,
since in this case there are often two cpus working in tandem :
- CPU handling the NIC rx interrupts, feeding the receive queue,
and (after this patch) freeing the skbs that were consumed.
- CPU in recvmsg() system call, essentially 100 % busy copying out
data to user space.
Having at most one skb in a per-socket cache has very little risk
of memory exhaustion, and since it is protected by socket lock,
its management is essentially free.
Note that if rps/rfs is used, we do not enable this feature, because
there is high chance that the same cpu is handling both the recvmsg()
system call and the TCP rx path, but that another cpu did the skb
allocations in the device driver right before the RPS/RFS logic.
To properly handle this case, it seems we would need to record
on which cpu skb was allocated, and use a different channel
to give skbs back to this cpu.
Signed-off-by: Eric Dumazet <edumazet@google.com>
Acked-by: Soheil Hassas Yeganeh <soheil@google.com>
Acked-by: Willem de Bruijn <willemb@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
On hosts with a lot of cores, RPC workloads suffer from heavy contention on slab spinlocks.
20.69% [kernel] [k] queued_spin_lock_slowpath
5.64% [kernel] [k] _raw_spin_lock
3.83% [kernel] [k] syscall_return_via_sysret
3.48% [kernel] [k] __entry_text_start
1.76% [kernel] [k] __netif_receive_skb_core
1.64% [kernel] [k] __fget
For each sendmsg(), we allocate one skb, and free it at the time ACK packet comes.
In many cases, ACK packets are handled by another cpus, and this unfortunately
incurs heavy costs for slab layer.
This patch uses an extra pointer in socket structure, so that we try to reuse
the same skb and avoid these expensive costs.
We cache at most one skb per socket so this should be safe as far as
memory pressure is concerned.
Signed-off-by: Eric Dumazet <edumazet@google.com>
Acked-by: Soheil Hassas Yeganeh <soheil@google.com>
Acked-by: Willem de Bruijn <willemb@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
We prefer static_branch_unlikely() over static_key_false() these days.
Signed-off-by: Eric Dumazet <edumazet@google.com>
Acked-by: Soheil Hassas Yeganeh <soheil@google.com>
Acked-by: Willem de Bruijn <willemb@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
When using fanouts with AF_PACKET, the demux functions such as
fanout_demux_cpu will return an index in the fanout socket array, which
corresponds to the selected socket.
The ordering of this array depends on the order the sockets were added
to a given fanout group, so for FANOUT_CPU this means sockets are bound
to cpus in the order they are configured, which is OK.
However, when stopping then restarting the interface these sockets are
bound to, the sockets are reassigned to the fanout group in the reverse
order, due to the fact that they were inserted at the head of the
interface's AF_PACKET socket list.
This means that traffic that was directed to the first socket in the
fanout group is now directed to the last one after an interface restart.
In the case of FANOUT_CPU, traffic from CPU0 will be directed to the
socket that used to receive traffic from the last CPU after an interface
restart.
This commit introduces a helper to add a socket at the tail of a list,
then uses it to register AF_PACKET sockets.
Note that this changes the order in which sockets are listed in /proc and
with sock_diag.
Fixes: dc99f60069 ("packet: Add fanout support")
Signed-off-by: Maxime Chevallier <maxime.chevallier@bootlin.com>
Acked-by: Willem de Bruijn <willemb@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
The netfilter conflicts were rather simple overlapping
changes.
However, the cls_tcindex.c stuff was a bit more complex.
On the 'net' side, Cong is fixing several races and memory
leaks. Whilst on the 'net-next' side we have Vlad adding
the rtnl-ness support.
What I've decided to do, in order to resolve this, is revert the
conversion over to using a workqueue that Cong did, bringing us back
to pure RCU. I did it this way because I believe that either Cong's
races don't apply with have Vlad did things, or Cong will have to
implement the race fix slightly differently.
Signed-off-by: David S. Miller <davem@davemloft.net>
With many active TCP sockets, fat TCP sockets could fool
__sk_mem_raise_allocated() thanks to an overflow.
They would increase their share of the memory, instead
of decreasing it.
Signed-off-by: Eric Dumazet <edumazet@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Al Viro mentioned (Message-ID
<20170626041334.GZ10672@ZenIV.linux.org.uk>)
that there is probably a race condition
lurking in accesses of sk_stamp on 32-bit machines.
sock->sk_stamp is of type ktime_t which is always an s64.
On a 32 bit architecture, we might run into situations of
unsafe access as the access to the field becomes non atomic.
Use seqlocks for synchronization.
This allows us to avoid using spinlocks for readers as
readers do not need mutual exclusion.
Another approach to solve this is to require sk_lock for all
modifications of the timestamps. The current approach allows
for timestamps to have their own lock: sk_stamp_lock.
This allows for the patch to not compete with already
existing critical sections, and side effects are limited
to the paths in the patch.
The addition of the new field maintains the data locality
optimizations from
commit 9115e8cd2a ("net: reorganize struct sock for better data
locality")
Note that all the instances of the sk_stamp accesses
are either through the ioctl or the syscall recvmsg.
Signed-off-by: Deepa Dinamani <deepa.kernel@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Lots of conflicts, by happily all cases of overlapping
changes, parallel adds, things of that nature.
Thanks to Stephen Rothwell, Saeed Mahameed, and others
for their guidance in these resolutions.
Signed-off-by: David S. Miller <davem@davemloft.net>
SOF_TIMESTAMPING_OPT_ID is supported on TCP, UDP and RAW sockets.
But it was missing on RAW with IPPROTO_IP, PF_PACKET and CAN.
Add skb_setup_tx_timestamp that configures both tx_flags and tskey
for these paths that do not need corking or use bytestream keys.
Fixes: 09c2d251b7 ("net-timestamp: add key to disambiguate concurrent datagrams")
Signed-off-by: Willem de Bruijn <willemb@google.com>
Acked-by: Soheil Hassas Yeganeh <soheil@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
TCP_NOTSENT_LOWAT socket option or sysctl was added in linux-3.12
as a step to enable bigger tcp sndbuf limits.
It works reasonably well, but the following happens :
Once the limit is reached, TCP stack generates
an [E]POLLOUT event for every incoming ACK packet.
This causes a high number of context switches.
This patch implements the strategy David Miller added
in sock_def_write_space() :
- If TCP socket has a notsent_lowat constraint of X bytes,
allow sendmsg() to fill up to X bytes, but send [E]POLLOUT
only if number of notsent bytes is below X/2
This considerably reduces TCP_NOTSENT_LOWAT overhead,
while allowing to keep the pipe full.
Tested:
100 ms RTT netem testbed between A and B, 100 concurrent TCP_STREAM
A:/# cat /proc/sys/net/ipv4/tcp_wmem
4096 262144 64000000
A:/# super_netperf 100 -H B -l 1000 -- -K bbr &
A:/# grep TCP /proc/net/sockstat
TCP: inuse 203 orphan 0 tw 19 alloc 414 mem 1364904 # This is about 54 MB of memory per flow :/
A:/# vmstat 5 5
procs -----------memory---------- ---swap-- -----io---- -system-- ------cpu-----
r b swpd free buff cache si so bi bo in cs us sy id wa st
0 0 0 256220672 13532 694976 0 0 10 0 28 14 0 1 99 0 0
2 0 0 256320016 13532 698480 0 0 512 0 715901 5927 0 10 90 0 0
0 0 0 256197232 13532 700992 0 0 735 13 771161 5849 0 11 89 0 0
1 0 0 256233824 13532 703320 0 0 512 23 719650 6635 0 11 89 0 0
2 0 0 256226880 13532 705780 0 0 642 4 775650 6009 0 12 88 0 0
A:/# echo 2097152 >/proc/sys/net/ipv4/tcp_notsent_lowat
A:/# grep TCP /proc/net/sockstat
TCP: inuse 203 orphan 0 tw 19 alloc 414 mem 86411 # 3.5 MB per flow
A:/# vmstat 5 5 # check that context switches have not inflated too much.
procs -----------memory---------- ---swap-- -----io---- -system-- ------cpu-----
r b swpd free buff cache si so bi bo in cs us sy id wa st
2 0 0 260386512 13592 662148 0 0 10 0 17 14 0 1 99 0 0
0 0 0 260519680 13592 604184 0 0 512 13 726843 12424 0 10 90 0 0
1 1 0 260435424 13592 598360 0 0 512 25 764645 12925 0 10 90 0 0
1 0 0 260855392 13592 578380 0 0 512 7 722943 13624 0 11 88 0 0
1 0 0 260445008 13592 601176 0 0 614 34 772288 14317 0 10 90 0 0
Signed-off-by: Eric Dumazet <edumazet@google.com>
Acked-by: Soheil Hassas Yeganeh <soheil@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
This reverts commit dd979b4df8.
This broke tcp_poll for SMC fallback: An AF_SMC socket establishes an
internal TCP socket for the initial handshake with the remote peer.
Whenever the SMC connection can not be established this TCP socket is
used as a fallback. All socket operations on the SMC socket are then
forwarded to the TCP socket. In case of poll, the file->private_data
pointer references the SMC socket because the TCP socket has no file
assigned. This causes tcp_poll to wait on the wrong socket.
Signed-off-by: Karsten Graul <kgraul@linux.ibm.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
David S. Miller