Skip to content

cloud-kernel
cloud-kernel

openanolis / cloud-kernel
1 年多 前同步成功

通知 160
Star 36
Fork 7
cloud-kernel
cloud-kernel
切换分支/标签
  1. 08 7月, 2018 5 次提交
  3. 05 7月, 2018 4 次提交
  5. 04 7月, 2018 1 次提交
  7. 03 7月, 2018 8 次提交
    • A
      Merge branch 'af_xdp-fixes' · 39d393cf
      由 Alexei Starovoitov 提交于 
      Magnus Karlsson says:

====================
This patch set fixes three bugs in the SKB TX path of AF_XDP.
Details in the individual commits.

The structure of the patch set is as follows:

Patch 1: Fix for lost completion message
Patch 2-3: Fix for possible multiple completions of single packet
Patch 4: Fix potential race during error

Changes from v1:

* Added explanation of race in commit message of patch 4.
====================
Signed-off-by: NAlexei Starovoitov <ast@kernel.org>
      39d393cf
    • M
      xsk: fix potential race in SKB TX completion code · a9744f7c
      由 Magnus Karlsson 提交于 
      There is a potential race in the TX completion code for the SKB
case. One process enters the sendmsg code of an AF_XDP socket in order
to send a frame. The execution eventually trickles down to the driver
that is told to send the packet. However, it decides to drop the
packet due to some error condition (e.g., rings full) and frees the
SKB. This will trigger the SKB destructor and a completion will be
sent to the AF_XDP user space through its
single-producer/single-consumer queues.

At the same time a TX interrupt has fired on another core and it
dispatches the TX completion code in the driver. It does its HW
specific things and ends up freeing the SKB associated with the
transmitted packet. This will trigger the SKB destructor and a
completion will be sent to the AF_XDP user space through its
single-producer/single-consumer queues. With a pseudo call stack, it
would look like this:

Core 1:
sendmsg() being called in the application
  netdev_start_xmit()
    Driver entered through ndo_start_xmit
      Driver decides to free the SKB for some reason (e.g., rings full)
        Destructor of SKB called
          xskq_produce_addr() is called to signal completion to user space

Core 2:
TX completion irq
  NAPI loop
    Driver irq handler for TX completions
      Frees the SKB
        Destructor of SKB called
          xskq_produce_addr() is called to signal completion to user space

We now have a violation of the single-producer/single-consumer
principle for our queues as there are two threads trying to produce at
the same time on the same queue.

Fixed by introducing a spin_lock in the destructor. In regards to the
performance, I get around 1.74 Mpps for txonly before and after the
introduction of the spinlock. There is of course some impact due to
the spin lock but it is in the less significant digits that are too
noisy for me to measure. But let us say that the version without the
spin lock got 1.745 Mpps in the best case and the version with 1.735
Mpps in the worst case, then that would mean a maximum drop in
performance of 0.5%.

Fixes: 35fcde7f ("xsk: support for Tx")
Signed-off-by: NMagnus Karlsson <magnus.karlsson@intel.com>
Signed-off-by: NAlexei Starovoitov <ast@kernel.org>
      a9744f7c
    • M
      samples/bpf: deal with EBUSY return code from sendmsg in xdpsock sample · c03079c9
      由 Magnus Karlsson 提交于 
      Sendmsg in the SKB path of AF_XDP can now return EBUSY when a packet
was discarded and completed by the driver. Just ignore this message
in the sample application.

Fixes: b4b8faa1 ("samples/bpf: sample application and documentation for AF_XDP sockets")
Signed-off-by: NMagnus Karlsson <magnus.karlsson@intel.com>
Reported-by: NPavel Odintsov <pavel@fastnetmon.com>
Signed-off-by: NAlexei Starovoitov <ast@kernel.org>
      c03079c9
    • M
      xsk: frame could be completed more than once in SKB path · fe588685
      由 Magnus Karlsson 提交于 
      Fixed a bug in which a frame could be completed more than once
when an error was returned from dev_direct_xmit(). The code
erroneously retried sending the message leading to multiple
calls to the SKB destructor and therefore multiple completions
of the same buffer to user space.

The error code in this case has been changed from EAGAIN to EBUSY
in order to tell user space that the sending of the packet failed
and the buffer has been return to user space through the completion
queue.

Fixes: 35fcde7f ("xsk: support for Tx")
Signed-off-by: NMagnus Karlsson <magnus.karlsson@intel.com>
Reported-by: NPavel Odintsov <pavel@fastnetmon.com>
Signed-off-by: NAlexei Starovoitov <ast@kernel.org>
      fe588685
    • M
      xsk: fix potential lost completion message in SKB path · 20b52a75
      由 Magnus Karlsson 提交于 
      The code in xskq_produce_addr erroneously checked if there
was up to LAZY_UPDATE_THRESHOLD amount of space in the completion
queue. It only needs to check if there is one slot left in the
queue. This bug could under some circumstances lead to a WARN_ON_ONCE
being triggered and the completion message to user space being lost.

Fixes: 35fcde7f ("xsk: support for Tx")
Signed-off-by: NMagnus Karlsson <magnus.karlsson@intel.com>
Reported-by: NPavel Odintsov <pavel@fastnetmon.com>
Signed-off-by: NAlexei Starovoitov <ast@kernel.org>
      20b52a75
    • L
      Merge branch 'for-next' of git://git.kernel.org/pub/scm/linux/kernel/git/shli/md · d0fbad0a
      由 Linus Torvalds 提交于 
      Pull MD fixes from Shaohua Li:
 "Two small fixes for MD:

   - an error handling fix from me

   - a recover bug fix for raid10 from BingJing"

* 'for-next' of git://git.kernel.org/pub/scm/linux/kernel/git/shli/md:
  md/raid10: fix that replacement cannot complete recovery after reassemble
  MD: cleanup resources in failure
      d0fbad0a
    • L
      Merge tag 'for-linus' of git://github.com/stffrdhrn/linux · 8d2b6f6b
      由 Linus Torvalds 提交于 
      Pull OpenRISC fixes from Stafford Horne:
 "Two fixes for issues which were breaking OpenRISC boot:

   - Fix bug in __pte_free_tlb() exposed in 4.18 by Matthew Wilcox's
     page table flag addition.

   - Fix issue booting on real hardware if delay slot detection
     emulation is disabled"

* tag 'for-linus' of git://github.com/stffrdhrn/linux:
  openrisc: entry: Fix delay slot exception detection
  openrisc: Call destructor during __pte_free_tlb
      8d2b6f6b
    • L
      Merge git://git.kernel.org/pub/scm/linux/kernel/git/davem/net · 4e33d7d4
      由 Linus Torvalds 提交于 
      Pull networking fixes from David Miller:

 1) Verify netlink attributes properly in nf_queue, from Eric Dumazet.

 2) Need to bump memory lock rlimit for test_sockmap bpf test, from
    Yonghong Song.

 3) Fix VLAN handling in lan78xx driver, from Dave Stevenson.

 4) Fix uninitialized read in nf_log, from Jann Horn.

 5) Fix raw command length parsing in mlx5, from Alex Vesker.

 6) Cleanup loopback RDS connections upon netns deletion, from Sowmini
    Varadhan.

 7) Fix regressions in FIB rule matching during create, from Jason A.
    Donenfeld and Roopa Prabhu.

 8) Fix mpls ether type detection in nfp, from Pieter Jansen van Vuuren.

 9) More bpfilter build fixes/adjustments from Masahiro Yamada.

10) Fix XDP_{TX,REDIRECT} flushing in various drivers, from Jesper
    Dangaard Brouer.

11) fib_tests.sh file permissions were broken, from Shuah Khan.

12) Make sure BH/preemption is disabled in data path of mac80211, from
    Denis Kenzior.

13) Don't ignore nla_parse_nested() return values in nl80211, from
    Johannes berg.

14) Properly account sock objects ot kmemcg, from Shakeel Butt.

15) Adjustments to setting bpf program permissions to read-only, from
    Daniel Borkmann.

16) TCP Fast Open key endianness was broken, it always took on the host
    endiannness. Whoops. Explicitly make it little endian. From Yuching
    Cheng.

17) Fix prefix route setting for link local addresses in ipv6, from
    David Ahern.

18) Potential Spectre v1 in zatm driver, from Gustavo A. R. Silva.

19) Various bpf sockmap fixes, from John Fastabend.

20) Use after free for GRO with ESP, from Sabrina Dubroca.

21) Passing bogus flags to crypto_alloc_shash() in ipv6 SR code, from
    Eric Biggers.

* git://git.kernel.org/pub/scm/linux/kernel/git/davem/net: (87 commits)
  qede: Adverstise software timestamp caps when PHC is not available.
  qed: Fix use of incorrect size in memcpy call.
  qed: Fix setting of incorrect eswitch mode.
  qed: Limit msix vectors in kdump kernel to the minimum required count.
  ipvlan: call dev_change_flags when ipvlan mode is reset
  ipv6: sr: fix passing wrong flags to crypto_alloc_shash()
  net: fix use-after-free in GRO with ESP
  tcp: prevent bogus FRTO undos with non-SACK flows
  bpf: sockhash, add release routine
  bpf: sockhash fix omitted bucket lock in sock_close
  bpf: sockmap, fix smap_list_map_remove when psock is in many maps
  bpf: sockmap, fix crash when ipv6 sock is added
  net: fib_rules: bring back rule_exists to match rule during add
  hv_netvsc: split sub-channel setup into async and sync
  net: use dev_change_tx_queue_len() for SIOCSIFTXQLEN
  atm: zatm: Fix potential Spectre v1
  s390/qeth: consistently re-enable device features
  s390/qeth: don't clobber buffer on async TX completion
  s390/qeth: avoid using is_multicast_ether_addr_64bits on (u8 *)[6]
  s390/qeth: fix race when setting MAC address
  ...
      4e33d7d4
  9. 02 7月, 2018 15 次提交
  11. 01 7月, 2018 7 次提交
    • I
      tcp: prevent bogus FRTO undos with non-SACK flows · 1236f22f
      由 Ilpo Järvinen 提交于 
      If SACK is not enabled and the first cumulative ACK after the RTO
retransmission covers more than the retransmitted skb, a spurious
FRTO undo will trigger (assuming FRTO is enabled for that RTO).
The reason is that any non-retransmitted segment acknowledged will
set FLAG_ORIG_SACK_ACKED in tcp_clean_rtx_queue even if there is
no indication that it would have been delivered for real (the
scoreboard is not kept with TCPCB_SACKED_ACKED bits in the non-SACK
case so the check for that bit won't help like it does with SACK).
Having FLAG_ORIG_SACK_ACKED set results in the spurious FRTO undo
in tcp_process_loss.

We need to use more strict condition for non-SACK case and check
that none of the cumulatively ACKed segments were retransmitted
to prove that progress is due to original transmissions. Only then
keep FLAG_ORIG_SACK_ACKED set, allowing FRTO undo to proceed in
non-SACK case.

(FLAG_ORIG_SACK_ACKED is planned to be renamed to FLAG_ORIG_PROGRESS
to better indicate its purpose but to keep this change minimal, it
will be done in another patch).

Besides burstiness and congestion control violations, this problem
can result in RTO loop: When the loss recovery is prematurely
undoed, only new data will be transmitted (if available) and
the next retransmission can occur only after a new RTO which in case
of multiple losses (that are not for consecutive packets) requires
one RTO per loss to recover.
Signed-off-by: NIlpo Järvinen <ilpo.jarvinen@helsinki.fi>
Tested-by: NNeal Cardwell <ncardwell@google.com>
Acked-by: NNeal Cardwell <ncardwell@google.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>
      1236f22f
    • S
      openrisc: entry: Fix delay slot exception detection · ae15a41a
      由 Stafford Horne 提交于 
      Originally in patch e6d20c55 ("openrisc: entry: Fix delay slot
detection") I fixed delay slot detection, but only for QEMU.  We missed
that hardware delay slot detection using delay slot exception flag (DSX)
was still broken.  This was because QEMU set the DSX flag in both
pre-exception supervision register (ESR) and supervision register (SR)
register, but on real hardware the DSX flag is only set on the SR
register during exceptions.

Fix this by carrying the DSX flag into the SR register during exception.
We also update the DSX flag read locations to read the value from the SR
register not the pt_regs SR register which represents ESR.  The ESR
should never have the DSX flag set.

In the process I updated/removed a few comments to match the current
state.  Including removing a comment saying that the DSX detection logic
was inefficient and needed to be rewritten.

I have tested this on QEMU with a patch ensuring it matches the hardware
specification.

Link: https://lists.gnu.org/archive/html/qemu-devel/2018-07/msg00000.html
Fixes: e6d20c55 ("openrisc: entry: Fix delay slot detection")
Signed-off-by: NStafford Horne <shorne@gmail.com>
      ae15a41a
    • D
      Merge git://git.kernel.org/pub/scm/linux/kernel/git/bpf/bpf · 271b955e
      由 David S. Miller 提交于 
      Daniel Borkmann says:

====================
pull-request: bpf 2018-07-01

The following pull-request contains BPF updates for your *net* tree.

The main changes are:

1) A bpf_fib_lookup() helper fix to change the API before freeze to
   return an encoding of the FIB lookup result and return the nexthop
   device index in the params struct (instead of device index as return
   code that we had before), from David.

2) Various BPF JIT fixes to address syzkaller fallout, that is, do not
   reject progs when set_memory_*() fails since it could still be RO.
   Also arm32 JIT was not using bpf_jit_binary_lock_ro() API which was
   an issue, and a memory leak in s390 JIT found during review, from
   Daniel.

3) Multiple fixes for sockmap/hash to address most of the syzkaller
   triggered bugs. Usage with IPv6 was crashing, a GPF in bpf_tcp_close(),
   a missing sock_map_release() routine to hook up to callbacks, and a
   fix for an omitted bucket lock in sock_close(), from John.

4) Two bpftool fixes to remove duplicated error message on program load,
   and another one to close the libbpf object after program load. One
   additional fix for nfp driver's BPF offload to avoid stopping offload
   completely if replace of program failed, from Jakub.

5) Couple of BPF selftest fixes that bail out in some of the test
   scripts if the user does not have the right privileges, from Jeffrin.

6) Fixes in test_bpf for s390 when CONFIG_BPF_JIT_ALWAYS_ON is set
   where we need to set the flag that some of the test cases are expected
   to fail, from Kleber.

7) Fix to detangle BPF_LIRC_MODE2 dependency from CONFIG_CGROUP_BPF
   since it has no relation to it and lirc2 users often have configs
   without cgroups enabled and thus would not be able to use it, from Sean.

8) Fix a selftest failure in sockmap by removing a useless setrlimit()
   call that would set a too low limit where at the same time we are
   already including bpf_rlimit.h that does the job, from Yonghong.

9) Fix BPF selftest config with missing missing NET_SCHED, from Anders.
====================
Signed-off-by: NDavid S. Miller <davem@davemloft.net>
      271b955e
    • D
      Merge branch 'bpf-sockmap-fixes' · bf2b866a
      由 Daniel Borkmann 提交于 
      John Fastabend says:

====================
This addresses two syzbot issues that lead to identifying (by Eric and
Wei) a class of bugs where we don't correctly check for IPv4/v6
sockets and their associated state. The second issue was a locking
omission in sockhash.

The first patch addresses IPv6 socks and fixing an error where
sockhash would overwrite the prot pointer with IPv4 prot. To fix
this build similar solution to TLS ULP. Although we continue to
allow socks in all states not just ESTABLISH in this patch set
because as Martin points out there should be no issue with this
on the sockmap ULP because we don't use the ctx in this code. Once
multiple ULPs coexist we may need to revisit this. However we
can do this in *next trees.

The other issue syzbot found that the tcp_close() handler missed
locking the hash bucket lock which could result in corrupting the
sockhash bucket list if delete and close ran at the same time.
And also the smap_list_remove() routine was not working correctly
at all. This was not caught in my testing because in general my
tests (to date at least lets add some more robust selftest in
bpf-next) do things in the "expected" order, create map, add socks,
delete socks, then tear down maps. The tests we have that do the
ops out of this order where only working on single maps not multi-
maps so we never saw the issue. Thanks syzbot. The fix is to
restructure the tcp_close() lock handling. And fix the obvious
bug in smap_list_remove().

Finally, during review I noticed the release handler was omitted
from the upstream code (patch 4) due to an incorrect merge conflict
fix when I ported the code to latest bpf-next before submitting.
This would leave references to the map around if the user never
closes the map.

v3: rework patches, dropping ESTABLISH check and adding rcu
    annotation along with the smap_list_remove fix

v4: missed one more case where maps was being accessed without
    the sk_callback_lock, spoted by Martin as well.

v5: changed to use a specific lock for maps and reduced callback
    lock so that it is only used to gaurd sk callbacks. I think
    this makes the logic a bit cleaner and avoids confusion
    ovoer what each lock is doing.

Also big thanks to Martin for thorough review he caught at least
one case where I missed a rcu_call().
====================
Signed-off-by: NDaniel Borkmann <daniel@iogearbox.net>
      bf2b866a
    • J
      bpf: sockhash, add release routine · caac76a5
      由 John Fastabend 提交于 
      Add map_release_uref pointer to hashmap ops. This was dropped when
original sockhash code was ported into bpf-next before initial
commit.

Fixes: 81110384 ("bpf: sockmap, add hash map support")
Acked-by: NMartin KaFai Lau <kafai@fb.com>
Signed-off-by: NJohn Fastabend <john.fastabend@gmail.com>
Signed-off-by: NDaniel Borkmann <daniel@iogearbox.net>
      caac76a5
    • J
      bpf: sockhash fix omitted bucket lock in sock_close · e9db4ef6
      由 John Fastabend 提交于 
      First the sk_callback_lock() was being used to protect both the
sock callback hooks and the psock->maps list. This got overly
convoluted after the addition of sockhash (in sockmap it made
some sense because masp and callbacks were tightly coupled) so
lets split out a specific lock for maps and only use the callback
lock for its intended purpose. This fixes a couple cases where
we missed using maps lock when it was in fact needed. Also this
makes it easier to follow the code because now we can put the
locking closer to the actual code its serializing.

Next, in sock_hash_delete_elem() the pattern was as follows,

  sock_hash_delete_elem()
     [...]
     spin_lock(bucket_lock)
     l = lookup_elem_raw()
     if (l)
        hlist_del_rcu()
        write_lock(sk_callback_lock)
         .... destroy psock ...
        write_unlock(sk_callback_lock)
     spin_unlock(bucket_lock)

The ordering is necessary because we only know the {p}sock after
dereferencing the hash table which we can't do unless we have the
bucket lock held. Once we have the bucket lock and the psock element
it is deleted from the hashmap to ensure any other path doing a lookup
will fail. Finally, the refcnt is decremented and if zero the psock
is destroyed.

In parallel with the above (or free'ing the map) a tcp close event
may trigger tcp_close(). Which at the moment omits the bucket lock
altogether (oops!) where the flow looks like this,

  bpf_tcp_close()
     [...]
     write_lock(sk_callback_lock)
     for each psock->maps // list of maps this sock is part of
         hlist_del_rcu(ref_hash_node);
         .... destroy psock ...
     write_unlock(sk_callback_lock)

Obviously, and demonstrated by syzbot, this is broken because
we can have multiple threads deleting entries via hlist_del_rcu().

To fix this we might be tempted to wrap the hlist operation in a
bucket lock but that would create a lock inversion problem. In
summary to follow locking rules the psocks maps list needs the
sk_callback_lock (after this patch maps_lock) but we need the bucket
lock to do the hlist_del_rcu.

To resolve the lock inversion problem pop the head of the maps list
repeatedly and remove the reference until no more are left. If a
delete happens in parallel from the BPF API that is OK as well because
it will do a similar action, lookup the lock in the map/hash, delete
it from the map/hash, and dec the refcnt. We check for this case
before doing a destroy on the psock to ensure we don't have two
threads tearing down a psock. The new logic is as follows,

  bpf_tcp_close()
  e = psock_map_pop(psock->maps) // done with map lock
  bucket_lock() // lock hash list bucket
  l = lookup_elem_raw(head, hash, key, key_size);
  if (l) {
     //only get here if elmnt was not already removed
     hlist_del_rcu()
     ... destroy psock...
  }
  bucket_unlock()

And finally for all the above to work add missing locking around  map
operations per above. Then add RCU annotations and use
rcu_dereference/rcu_assign_pointer to manage values relying on RCU so
that the object is not free'd from sock_hash_free() while it is being
referenced in bpf_tcp_close().

Reported-by: syzbot+0ce137753c78f7b6acc1@syzkaller.appspotmail.com
Fixes: 81110384 ("bpf: sockmap, add hash map support")
Signed-off-by: NJohn Fastabend <john.fastabend@gmail.com>
Signed-off-by: NDaniel Borkmann <daniel@iogearbox.net>
      e9db4ef6
    • J
      bpf: sockmap, fix smap_list_map_remove when psock is in many maps · 54fedb42
      由 John Fastabend 提交于 
      If a hashmap is free'd with open socks it removes the reference to
the hash entry from the psock. If that is the last reference to the
psock then it will also be free'd by the reference counting logic.
However the current logic that removes the hash reference from the
list of references is broken. In smap_list_remove() we first check
if the sockmap entry matches and then check if the hashmap entry
matches. But, the sockmap entry sill always match because its NULL in
this case which causes the first entry to be removed from the list.
If this is always the "right" entry (because the user adds/removes
entries in order) then everything is OK but otherwise a subsequent
bpf_tcp_close() may reference a free'd object.

To fix this create two list handlers one for sockmap and one for
sockhash.

Reported-by: syzbot+0ce137753c78f7b6acc1@syzkaller.appspotmail.com
Fixes: 81110384 ("bpf: sockmap, add hash map support")
Acked-by: NMartin KaFai Lau <kafai@fb.com>
Signed-off-by: NJohn Fastabend <john.fastabend@gmail.com>
Signed-off-by: NDaniel Borkmann <daniel@iogearbox.net>
      54fedb42