Remove rxrpc_get_reply_time() as that is no longer used now that the call
issue time is used instead of the reply time.
Signed-off-by: David Howells <dhowells@redhat.com>
If the local processor work item for the rxrpc local endpoint gets requeued
by an event (such as an incoming packet) between it getting scheduled for
destruction and the UDP socket being closed, the rxrpc_local_destroyer()
function can get run twice. The second time it can hang because it can end
up waiting for cleanup events that will never happen.
Signed-off-by: David Howells <dhowells@redhat.com>
rxkad_verify_packet_2() has a small stack-allocated sglist of 4 elements,
but if that isn't sufficient for the number of fragments in the socket
buffer, we try to allocate an sglist large enough to hold all the
fragments.
However, for large packets with a lot of fragments, this isn't sufficient
and we need at least one additional fragment.
The problem manifests as skb_to_sgvec() returning -EMSGSIZE and this then
getting returned by userspace. Most of the time, this isn't a problem as
rxrpc sets a limit of 5692, big enough for 4 jumbo subpackets to be glued
together; occasionally, however, the server will ignore the reported limit
and give a packet that's a lot bigger - say 19852 bytes with ->nr_frags
being 7. skb_to_sgvec() then tries to return a "zeroth" fragment that
seems to occur before the fragments counted by ->nr_frags and we hit the
end of the sglist too early.
Note that __skb_to_sgvec() also has an skb_walk_frags() loop that is
recursive up to 24 deep. I'm not sure if I need to take account of that
too - or if there's an easy way of counting those frags too.
Fix this by counting an extra frag and allocating a larger sglist based on
that.
Fixes: d0d5c0cd1e ("rxrpc: Use skb_unshare() rather than skb_cow_data()")
Reported-by: Marc Dionne <marc.dionne@auristor.com>
Signed-off-by: David Howells <dhowells@redhat.com>
cc: linux-afs@lists.infradead.org
Because rxrpc pretends to be a tunnel on top of a UDP/UDP6 socket, allowing
it to siphon off UDP packets early in the handling of received UDP packets
thereby avoiding the packet going through the UDP receive queue, it doesn't
get ICMP packets through the UDP ->sk_error_report() callback. In fact, it
doesn't appear that there's any usable option for getting hold of ICMP
packets.
Fix this by adding a new UDP encap hook to distribute error messages for
UDP tunnels. If the hook is set, then the tunnel driver will be able to
see ICMP packets. The hook provides the offset into the packet of the UDP
header of the original packet that caused the notification.
An alternative would be to call the ->error_handler() hook - but that
requires that the skbuff be cloned (as ip_icmp_error() or ipv6_cmp_error()
do, though isn't really necessary or desirable in rxrpc's case is we want
to parse them there and then, not queue them).
Changes
=======
ver #3)
- Fixed an uninitialised variable.
ver #2)
- Fixed some missing CONFIG_AF_RXRPC_IPV6 conditionals.
Fixes: 5271953cad ("rxrpc: Use the UDP encap_rcv hook")
Signed-off-by: David Howells <dhowells@redhat.com>
Fix three bugs in the rxrpc's sendmsg implementation:
(1) rxrpc_new_client_call() should release the socket lock when returning
an error from rxrpc_get_call_slot().
(2) rxrpc_wait_for_tx_window_intr() will return without the call mutex
held in the event that we're interrupted by a signal whilst waiting
for tx space on the socket or relocking the call mutex afterwards.
Fix this by: (a) moving the unlock/lock of the call mutex up to
rxrpc_send_data() such that the lock is not held around all of
rxrpc_wait_for_tx_window*() and (b) indicating to higher callers
whether we're return with the lock dropped. Note that this means
recvmsg() will not block on this call whilst we're waiting.
(3) After dropping and regaining the call mutex, rxrpc_send_data() needs
to go and recheck the state of the tx_pending buffer and the
tx_total_len check in case we raced with another sendmsg() on the same
call.
Thinking on this some more, it might make sense to have different locks for
sendmsg() and recvmsg(). There's probably no need to make recvmsg() wait
for sendmsg(). It does mean that recvmsg() can return MSG_EOR indicating
that a call is dead before a sendmsg() to that call returns - but that can
currently happen anyway.
Without fix (2), something like the following can be induced:
WARNING: bad unlock balance detected!
5.16.0-rc6-syzkaller #0 Not tainted
-------------------------------------
syz-executor011/3597 is trying to release lock (&call->user_mutex) at:
[<ffffffff885163a3>] rxrpc_do_sendmsg+0xc13/0x1350 net/rxrpc/sendmsg.c:748
but there are no more locks to release!
other info that might help us debug this:
no locks held by syz-executor011/3597.
...
Call Trace:
<TASK>
__dump_stack lib/dump_stack.c:88 [inline]
dump_stack_lvl+0xcd/0x134 lib/dump_stack.c:106
print_unlock_imbalance_bug include/trace/events/lock.h:58 [inline]
__lock_release kernel/locking/lockdep.c:5306 [inline]
lock_release.cold+0x49/0x4e kernel/locking/lockdep.c:5657
__mutex_unlock_slowpath+0x99/0x5e0 kernel/locking/mutex.c:900
rxrpc_do_sendmsg+0xc13/0x1350 net/rxrpc/sendmsg.c:748
rxrpc_sendmsg+0x420/0x630 net/rxrpc/af_rxrpc.c:561
sock_sendmsg_nosec net/socket.c:704 [inline]
sock_sendmsg+0xcf/0x120 net/socket.c:724
____sys_sendmsg+0x6e8/0x810 net/socket.c:2409
___sys_sendmsg+0xf3/0x170 net/socket.c:2463
__sys_sendmsg+0xe5/0x1b0 net/socket.c:2492
do_syscall_x64 arch/x86/entry/common.c:50 [inline]
do_syscall_64+0x35/0xb0 arch/x86/entry/common.c:80
entry_SYSCALL_64_after_hwframe+0x44/0xae
[Thanks to Hawkins Jiawei and Khalid Masum for their attempts to fix this]
Fixes: bc5e3a546d ("rxrpc: Use MSG_WAITALL to tell sendmsg() to temporarily ignore signals")
Reported-by: syzbot+7f0483225d0c94cb3441@syzkaller.appspotmail.com
Signed-off-by: David Howells <dhowells@redhat.com>
Reviewed-by: Marc Dionne <marc.dionne@auristor.com>
Tested-by: syzbot+7f0483225d0c94cb3441@syzkaller.appspotmail.com
cc: Hawkins Jiawei <yin31149@gmail.com>
cc: Khalid Masum <khalid.masum.92@gmail.com>
cc: Dan Carpenter <dan.carpenter@oracle.com>
cc: linux-afs@lists.infradead.org
Link: https://lore.kernel.org/r/166135894583.600315.7170979436768124075.stgit@warthog.procyon.org.uk
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
When building with Clang we encounter this warning:
| net/rxrpc/rxkad.c:434:33: error: format specifies type 'unsigned short'
| but the argument has type 'u32' (aka 'unsigned int') [-Werror,-Wformat]
| _leave(" = %d [set %hx]", ret, y);
y is a u32 but the format specifier is `%hx`. Going from unsigned int to
short int results in a loss of data. This is surely not intended
behavior. If it is intended, the warning should be suppressed through
other means.
This patch should get us closer to the goal of enabling the -Wformat
flag for Clang builds.
Link: https://github.com/ClangBuiltLinux/linux/issues/378
Signed-off-by: Justin Stitt <justinstitt@google.com>
Reviewed-by: Nathan Chancellor <nathan@kernel.org>
Acked-by: David Howells <dhowells@redhat.com>
Link: https://lore.kernel.org/r/20220707182052.769989-1-justinstitt@google.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
Fix the decision on when to generate an IDLE ACK by keeping a count of the
number of packets we've received, but not yet soft-ACK'd, and the number of
packets we've processed, but not yet hard-ACK'd, rather than trying to keep
track of which DATA sequence numbers correspond to those points.
We then generate an ACK when either counter exceeds 2. The counters are
both cleared when we transcribe the information into any sort of ACK packet
for transmission. IDLE and DELAY ACKs are skipped if both counters are 0
(ie. no change).
Fixes: 805b21b929 ("rxrpc: Send an ACK after every few DATA packets we receive")
Signed-off-by: David Howells <dhowells@redhat.com>
cc: Marc Dionne <marc.dionne@auristor.com>
cc: linux-afs@lists.infradead.org
Signed-off-by: David S. Miller <davem@davemloft.net>
The previousPacket field in the rx ACK packet should never go backwards -
it's now the highest DATA sequence number received, not the last on
received (it used to be used for out of sequence detection).
Fixes: 248f219cb8 ("rxrpc: Rewrite the data and ack handling code")
Signed-off-by: David Howells <dhowells@redhat.com>
cc: Marc Dionne <marc.dionne@auristor.com>
cc: linux-afs@lists.infradead.org
Signed-off-by: David S. Miller <davem@davemloft.net>
Fix accidental overlapping of Rx-phase ACK accounting with Tx-phase ACK
accounting through variables shared between the two. call->acks_* members
refer to ACKs received in the Tx phase and call->ackr_* members to ACKs
sent/to be sent during the Rx phase.
Fixes: 1a2391c30c ("rxrpc: Fix detection of out of order acks")
Signed-off-by: David Howells <dhowells@redhat.com>
cc: Jeffrey Altman <jaltman@auristor.com>
cc: Marc Dionne <marc.dionne@auristor.com>
cc: linux-afs@lists.infradead.org
Signed-off-by: David S. Miller <davem@davemloft.net>
rxrpc has a timer to trigger resending of unacked data packets in a call.
This is not cancelled when a client call switches to the receive phase on
the basis that most calls don't last long enough for it to ever expire.
However, if it *does* expire after we've started to receive the reply, we
shouldn't then go into trying to retransmit or pinging the server to find
out if an ack got lost.
Fix this by skipping the resend code if we're into receiving the reply to a
client call.
Fixes: 17926a7932 ("[AF_RXRPC]: Provide secure RxRPC sockets for use by userspace and kernel both")
Signed-off-by: David Howells <dhowells@redhat.com>
cc: linux-afs@lists.infradead.org
Signed-off-by: David S. Miller <davem@davemloft.net>
AF_RXRPC's listen() handler lets you set the backlog up to 32 (if you bump
up the sysctl), but whilst the preallocation circular buffers have 32 slots
in them, one of them has to be a dead slot because we're using CIRC_CNT().
This means that listen(rxrpc_sock, 32) will cause an oops when the socket
is closed because rxrpc_service_prealloc_one() allocated one too many calls
and rxrpc_discard_prealloc() won't then be able to get rid of them because
it'll think the ring is empty. rxrpc_release_calls_on_socket() then tries
to abort them, but oopses because call->peer isn't yet set.
Fix this by setting the maximum backlog to RXRPC_BACKLOG_MAX - 1 to match
the ring capacity.
BUG: kernel NULL pointer dereference, address: 0000000000000086
...
RIP: 0010:rxrpc_send_abort_packet+0x73/0x240 [rxrpc]
Call Trace:
<TASK>
? __wake_up_common_lock+0x7a/0x90
? rxrpc_notify_socket+0x8e/0x140 [rxrpc]
? rxrpc_abort_call+0x4c/0x60 [rxrpc]
rxrpc_release_calls_on_socket+0x107/0x1a0 [rxrpc]
rxrpc_release+0xc9/0x1c0 [rxrpc]
__sock_release+0x37/0xa0
sock_close+0x11/0x20
__fput+0x89/0x240
task_work_run+0x59/0x90
do_exit+0x319/0xaa0
Fixes: 00e907127e ("rxrpc: Preallocate peers, conns and calls for incoming service requests")
Reported-by: Marc Dionne <marc.dionne@auristor.com>
Signed-off-by: David Howells <dhowells@redhat.com>
cc: linux-afs@lists.infradead.org
Link: https://lists.infradead.org/pipermail/linux-afs/2022-March/005079.html
Signed-off-by: David S. Miller <davem@davemloft.net>
If a client's address changes, say if it is NAT'd, this can disrupt an in
progress operation. For most operations, this is not much of a problem,
but StoreData can be different as some servers modify the target file as
the data comes in, so if a store request is disrupted, the file can get
corrupted on the server.
The problem is that the server doesn't recognise packets that come after
the change of address as belonging to the original client and will bounce
them, either by sending an OUT_OF_SEQUENCE ACK to the apparent new call if
the packet number falls within the initial sequence number window of a call
or by sending an EXCEEDS_WINDOW ACK if it falls outside and then aborting
it. In both cases, firstPacket will be 1 and previousPacket will be 0 in
the ACK information.
Fix this by the following means:
(1) If a client call receives an EXCEEDS_WINDOW ACK with firstPacket as 1
and previousPacket as 0, assume this indicates that the server saw the
incoming packets from a different peer and thus as a different call.
Fail the call with error -ENETRESET.
(2) Also fail the call if a similar OUT_OF_SEQUENCE ACK occurs if the
first packet has been hard-ACK'd. If it hasn't been hard-ACK'd, the
ACK packet will cause it to get retransmitted, so the call will just
be repeated.
(3) Make afs_select_fileserver() treat -ENETRESET as a straight fail of
the operation.
(4) Prioritise the error code over things like -ECONNRESET as the server
did actually respond.
(5) Make writeback treat -ENETRESET as a retryable error and make it
redirty all the pages involved in a write so that the VM will retry.
Note that there is still a circumstance that I can't easily deal with: if
the operation is fully received and processed by the server, but the reply
is lost due to address change. There's no way to know if the op happened.
We can examine the server, but a conflicting change could have been made by
a third party - and we can't tell the difference. In such a case, a
message like:
kAFS: vnode modified {100058:146266} b7->b8 YFS.StoreData64 (op=2646a)
will be logged to dmesg on the next op to touch the file and the client
will reset the inode state, including invalidating clean parts of the
pagecache.
Reported-by: Marc Dionne <marc.dionne@auristor.com>
Signed-off-by: David Howells <dhowells@redhat.com>
cc: linux-afs@lists.infradead.org
Link: http://lists.infradead.org/pipermail/linux-afs/2021-December/004811.html # v1
Signed-off-by: David S. Miller <davem@davemloft.net>
The RX_USER_ABORT code should really only be used to indicate that the user
of the rxrpc service (ie. userspace) implicitly caused a call to be aborted
- for instance if the AF_RXRPC socket is closed whilst the call was in
progress. (The user may also explicitly abort a call and specify the abort
code to use).
Change some of the points of generation to use other abort codes instead:
(1) Abort the call with RXGEN_SS_UNMARSHAL or RXGEN_CC_UNMARSHAL if we see
ENOMEM and EFAULT during received data delivery and abort with
RX_CALL_DEAD in the default case.
(2) Abort with RXGEN_SS_MARSHAL if we get ENOMEM whilst trying to send a
reply.
(3) Abort with RX_CALL_DEAD if we stop hearing from the peer if we had
heard from the peer and abort with RX_CALL_TIMEOUT if we hadn't.
(4) Abort with RX_CALL_DEAD if we try to disconnect a call that's not
completed successfully or been aborted.
Reported-by: Jeffrey Altman <jaltman@auristor.com>
Signed-off-by: David Howells <dhowells@redhat.com>
cc: Marc Dionne <marc.dionne@auristor.com>
cc: linux-afs@lists.infradead.org
Signed-off-by: David S. Miller <davem@davemloft.net>
If at the end of rxrpc sendmsg() or rxrpc_kernel_send_data() the call that
was being given data was aborted remotely or otherwise failed, return an
error rather than returning the amount of data buffered for transmission.
The call (presumably) did not complete, so there's not much point
continuing with it. AF_RXRPC considers it "complete" and so will be
unwilling to do anything else with it - and won't send a notification for
it, deeming the return from sendmsg sufficient.
Not returning an error causes afs to incorrectly handle a StoreData
operation that gets interrupted by a change of address due to NAT
reconfiguration.
This doesn't normally affect most operations since their request parameters
tend to fit into a single UDP packet and afs_make_call() returns before the
server responds; StoreData is different as it involves transmission of a
lot of data.
This can be triggered on a client by doing something like:
dd if=/dev/zero of=/afs/example.com/foo bs=1M count=512
at one prompt, and then changing the network address at another prompt,
e.g.:
ifconfig enp6s0 inet 192.168.6.2 && route add 192.168.6.1 dev enp6s0
Tracing packets on an Auristor fileserver looks something like:
192.168.6.1 -> 192.168.6.3 RX 107 ACK Idle Seq: 0 Call: 4 Source Port: 7000 Destination Port: 7001
192.168.6.3 -> 192.168.6.1 AFS (RX) 1482 FS Request: Unknown(64538) (64538)
192.168.6.3 -> 192.168.6.1 AFS (RX) 1482 FS Request: Unknown(64538) (64538)
192.168.6.1 -> 192.168.6.3 RX 107 ACK Idle Seq: 0 Call: 4 Source Port: 7000 Destination Port: 7001
<ARP exchange for 192.168.6.2>
192.168.6.2 -> 192.168.6.1 AFS (RX) 1482 FS Request: Unknown(0) (0)
192.168.6.2 -> 192.168.6.1 AFS (RX) 1482 FS Request: Unknown(0) (0)
192.168.6.1 -> 192.168.6.2 RX 107 ACK Exceeds Window Seq: 0 Call: 4 Source Port: 7000 Destination Port: 7001
192.168.6.1 -> 192.168.6.2 RX 74 ABORT Seq: 0 Call: 4 Source Port: 7000 Destination Port: 7001
192.168.6.1 -> 192.168.6.2 RX 74 ABORT Seq: 29321 Call: 4 Source Port: 7000 Destination Port: 7001
The Auristor fileserver logs code -453 (RXGEN_SS_UNMARSHAL), but the abort
code received by kafs is -5 (RX_PROTOCOL_ERROR) as the rx layer sees the
condition and generates an abort first and the unmarshal error is a
consequence of that at the application layer.
Reported-by: Marc Dionne <marc.dionne@auristor.com>
Signed-off-by: David Howells <dhowells@redhat.com>
cc: linux-afs@lists.infradead.org
Link: http://lists.infradead.org/pipermail/linux-afs/2021-December/004810.html # v1
Signed-off-by: David S. Miller <davem@davemloft.net>
There's a locking issue with the per-netns list of calls in rxrpc. The
pieces of code that add and remove a call from the list use write_lock()
and the calls procfile uses read_lock() to access it. However, the timer
callback function may trigger a removal by trying to queue a call for
processing and finding that it's already queued - at which point it has a
spare refcount that it has to do something with. Unfortunately, if it puts
the call and this reduces the refcount to 0, the call will be removed from
the list. Unfortunately, since the _bh variants of the locking functions
aren't used, this can deadlock.
================================
WARNING: inconsistent lock state
5.18.0-rc3-build4+ #10 Not tainted
--------------------------------
inconsistent {SOFTIRQ-ON-W} -> {IN-SOFTIRQ-W} usage.
ksoftirqd/2/25 [HC0[0]:SC1[1]:HE1:SE0] takes:
ffff888107ac4038 (&rxnet->call_lock){+.?.}-{2:2}, at: rxrpc_put_call+0x103/0x14b
{SOFTIRQ-ON-W} state was registered at:
...
Possible unsafe locking scenario:
CPU0
----
lock(&rxnet->call_lock);
<Interrupt>
lock(&rxnet->call_lock);
*** DEADLOCK ***
1 lock held by ksoftirqd/2/25:
#0: ffff8881008ffdb0 ((&call->timer)){+.-.}-{0:0}, at: call_timer_fn+0x5/0x23d
Changes
=======
ver #2)
- Changed to using list_next_rcu() rather than rcu_dereference() directly.
Fixes: 17926a7932 ("[AF_RXRPC]: Provide secure RxRPC sockets for use by userspace and kernel both")
Signed-off-by: David Howells <dhowells@redhat.com>
cc: Marc Dionne <marc.dionne@auristor.com>
cc: linux-afs@lists.infradead.org
Signed-off-by: David S. Miller <davem@davemloft.net>
Allow the list of in-use local UDP endpoints in the current network
namespace to be viewed in /proc.
To aid with this, the endpoint list is converted to an hlist and RCU-safe
manipulation is used so that the list can be read with only the RCU
read lock held.
Signed-off-by: David Howells <dhowells@redhat.com>
cc: Marc Dionne <marc.dionne@auristor.com>
cc: linux-afs@lists.infradead.org
Signed-off-by: David S. Miller <davem@davemloft.net>
AF_RXRPC doesn't currently enable IPv6 UDP Tx checksums on the transport
socket it opens and the checksums in the packets it generates end up 0.
It probably should also enable IPv6 UDP Rx checksums and IPv4 UDP
checksums. The latter only seem to be applied if the socket family is
AF_INET and don't seem to apply if it's AF_INET6. IPv4 packets from an
IPv6 socket seem to have checksums anyway.
What seems to have happened is that the inet_inv_convert_csum() call didn't
get converted to the appropriate udp_port_cfg parameters - and
udp_sock_create() disables checksums unless explicitly told not too.
Fix this by enabling the three udp_port_cfg checksum options.
Fixes: 1a9b86c9fd ("rxrpc: use udp tunnel APIs instead of open code in rxrpc_open_socket")
Reported-by: Marc Dionne <marc.dionne@auristor.com>
Signed-off-by: David Howells <dhowells@redhat.com>
Reviewed-by: Xin Long <lucien.xin@gmail.com>
Reviewed-by: Marc Dionne <marc.dionne@auristor.com>
cc: Vadim Fedorenko <vfedorenko@novek.ru>
cc: David S. Miller <davem@davemloft.net>
cc: linux-afs@lists.infradead.org
Signed-off-by: David S. Miller <davem@davemloft.net>
A recent patch[1] from Eric Dumazet flipped the order in which the
keepalive timer and the keepalive worker were cancelled in order to fix a
syzbot reported issue[2]. Unfortunately, this enables the mirror image bug
whereby the timer races with rxrpc_exit_net(), restarting the worker after
it has been cancelled:
CPU 1 CPU 2
=============== =====================
if (rxnet->live)
<INTERRUPT>
rxnet->live = false;
cancel_work_sync(&rxnet->peer_keepalive_work);
rxrpc_queue_work(&rxnet->peer_keepalive_work);
del_timer_sync(&rxnet->peer_keepalive_timer);
Fix this by restoring the removed del_timer_sync() so that we try to remove
the timer twice. If the timer runs again, it should see ->live == false
and not restart the worker.
Fixes: 1946014ca3 ("rxrpc: fix a race in rxrpc_exit_net()")
Signed-off-by: David Howells <dhowells@redhat.com>
cc: Eric Dumazet <edumazet@google.com>
cc: Marc Dionne <marc.dionne@auristor.com>
cc: linux-afs@lists.infradead.org
Link: https://lore.kernel.org/r/20220404183439.3537837-1-eric.dumazet@gmail.com/ [1]
Link: https://syzkaller.appspot.com/bug?extid=724378c4bb58f703b09a [2]
Reviewed-by: Eric Dumazet <edumazet@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Gaosheng Cui
David Howells
William Dean
Justin Stitt
Jakub Kicinski
Eric Dumazet
Xiaolong Huang