There's a race between rxrpc_sendmsg setting up a call, but then failing to
send anything on it due to an error, and recvmsg() seeing the call
completion occur and trying to return the state to the user.
An assertion fails in rxrpc_recvmsg() because the call has already been
released from the socket and is about to be released again as recvmsg deals
with it. (The recvmsg_q queue on the socket holds a ref, so there's no
problem with use-after-free.)
We also have to be careful not to end up reporting an error twice, in such
a way that both returns indicate to userspace that the user ID supplied
with the call is no longer in use - which could cause the client to
malfunction if it recycles the user ID fast enough.
Fix this by the following means:
(1) When sendmsg() creates a call after the point that the call has been
successfully added to the socket, don't return any errors through
sendmsg(), but rather complete the call and let recvmsg() retrieve
them. Make sendmsg() return 0 at this point. Further calls to
sendmsg() for that call will fail with ESHUTDOWN.
Note that at this point, we haven't send any packets yet, so the
server doesn't yet know about the call.
(2) If sendmsg() returns an error when it was expected to create a new
call, it means that the user ID wasn't used.
(3) Mark the call disconnected before marking it completed to prevent an
oops in rxrpc_release_call().
(4) recvmsg() will then retrieve the error and set MSG_EOR to indicate
that the user ID is no longer known by the kernel.
An oops like the following is produced:
kernel BUG at net/rxrpc/recvmsg.c:605!
...
RIP: 0010:rxrpc_recvmsg+0x256/0x5ae
...
Call Trace:
? __init_waitqueue_head+0x2f/0x2f
____sys_recvmsg+0x8a/0x148
? import_iovec+0x69/0x9c
? copy_msghdr_from_user+0x5c/0x86
___sys_recvmsg+0x72/0xaa
? __fget_files+0x22/0x57
? __fget_light+0x46/0x51
? fdget+0x9/0x1b
do_recvmmsg+0x15e/0x232
? _raw_spin_unlock+0xa/0xb
? vtime_delta+0xf/0x25
__x64_sys_recvmmsg+0x2c/0x2f
do_syscall_64+0x4c/0x78
entry_SYSCALL_64_after_hwframe+0x44/0xa9
Fixes: 357f5ef646 ("rxrpc: Call rxrpc_release_call() on error in rxrpc_new_client_call()")
Reported-by: syzbot+b54969381df354936d96@syzkaller.appspotmail.com
Signed-off-by: David Howells <dhowells@redhat.com>
Reviewed-by: Marc Dionne <marc.dionne@auristor.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
The UDP reuseport conflict was a little bit tricky.
The net-next code, via bpf-next, extracted the reuseport handling
into a helper so that the BPF sk lookup code could invoke it.
At the same time, the logic for reuseport handling of unconnected
sockets changed via commit efc6b6f6c3
which changed the logic to carry on the reuseport result into the
rest of the lookup loop if we do not return immediately.
This requires moving the reuseport_has_conns() logic into the callers.
While we are here, get rid of inline directives as they do not belong
in foo.c files.
The other changes were cases of more straightforward overlapping
modifications.
Signed-off-by: David S. Miller <davem@davemloft.net>
Rework the remaining setsockopt code to pass a sockptr_t instead of a
plain user pointer. This removes the last remaining set_fs(KERNEL_DS)
outside of architecture specific code.
Signed-off-by: Christoph Hellwig <hch@lst.de>
Acked-by: Stefan Schmidt <stefan@datenfreihafen.org> [ieee802154]
Acked-by: Matthieu Baerts <matthieu.baerts@tessares.net>
Signed-off-by: David S. Miller <davem@davemloft.net>
rxrpc_sendmsg() returns EPIPE if there's an outstanding error, such as if
rxrpc_recvmsg() indicating ENODATA if there's nothing for it to read.
Change rxrpc_recvmsg() to return EAGAIN instead if there's nothing to read
as this particular error doesn't get stored in ->sk_err by the networking
core.
Also change rxrpc_sendmsg() so that it doesn't fail with delayed receive
errors (there's no way for it to report which call, if any, the error was
caused by).
Fixes: 17926a7932 ("[AF_RXRPC]: Provide secure RxRPC sockets for use by userspace and kernel both")
Signed-off-by: David Howells <dhowells@redhat.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Commit 2ad6691d98, which moved the modification of the status annotation
for a packet in the Tx buffer prior to the retransmission moved the state
clearance, but managed to lose the bit that set it to UNACK.
Consequently, if a retransmission occurs, the packet is accidentally
changed to the ACK state (ie. 0) by masking it off, which means that the
packet isn't counted towards the tally of newly-ACK'd packets if it gets
hard-ACK'd. This then prevents the congestion control algorithm from
recovering properly.
Fix by reinstating the change of state to UNACK.
Spotted by the generic/460 xfstest.
Fixes: 2ad6691d98 ("rxrpc: Fix race between incoming ACK parser and retransmitter")
Signed-off-by: David Howells <dhowells@redhat.com>
The handling of the receive window size (rwind) from a received ACK packet
is not correct. The rxrpc_input_ackinfo() function currently checks the
current Tx window size against the rwind from the ACK to see if it has
changed, but then limits the rwind size before storing it in the tx_winsize
member and, if it increased, wake up the transmitting process. This means
that if rwind > RXRPC_RXTX_BUFF_SIZE - 1, this path will always be
followed.
Fix this by limiting rwind before we compare it to tx_winsize.
The effect of this can be seen by enabling the rxrpc_rx_rwind_change
tracepoint.
Fixes: 702f2ac87a ("rxrpc: Wake up the transmitter if Rx window size increases on the peer")
Signed-off-by: David Howells <dhowells@redhat.com>
There's a race between the retransmission code and the received ACK parser.
The problem is that the retransmission loop has to drop the lock under
which it is iterating through the transmission buffer in order to transmit
a packet, but whilst the lock is dropped, the ACK parser can crank the Tx
window round and discard the packets from the buffer.
The retransmission code then updated the annotations for the wrong packet
and a later retransmission thought it had to retransmit a packet that
wasn't there, leading to a NULL pointer dereference.
Fix this by:
(1) Moving the annotation change to before we drop the lock prior to
transmission. This means we can't vary the annotation depending on
the outcome of the transmission, but that's fine - we'll retransmit
again later if it failed now.
(2) Skipping the packet if the skb pointer is NULL.
The following oops was seen:
BUG: kernel NULL pointer dereference, address: 000000000000002d
Workqueue: krxrpcd rxrpc_process_call
RIP: 0010:rxrpc_get_skb+0x14/0x8a
...
Call Trace:
rxrpc_resend+0x331/0x41e
? get_vtime_delta+0x13/0x20
rxrpc_process_call+0x3c0/0x4ac
process_one_work+0x18f/0x27f
worker_thread+0x1a3/0x247
? create_worker+0x17d/0x17d
kthread+0xe6/0xeb
? kthread_delayed_work_timer_fn+0x83/0x83
ret_from_fork+0x1f/0x30
Fixes: 248f219cb8 ("rxrpc: Rewrite the data and ack handling code")
Signed-off-by: David Howells <dhowells@redhat.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
David Howells says:
====================
rxrpc: Fix hang due to missing notification
Here's a fix for AF_RXRPC. Occasionally calls hang because there are
circumstances in which rxrpc generate a notification when a call is
completed - primarily because initial packet transmission failed and the
call was killed off and an error returned. But the AFS filesystem driver
doesn't check this under all circumstances, expecting failure to be
delivered by asynchronous notification.
There are two patches: the first moves the problematic bits out-of-line and
the second contains the fix.
====================
Signed-off-by: David S. Miller <davem@davemloft.net>
Pull AFS updates from David Howells:
"There's some core VFS changes which affect a couple of filesystems:
- Make the inode hash table RCU safe and providing some RCU-safe
accessor functions. The search can then be done without taking the
inode_hash_lock. Care must be taken because the object may be being
deleted and no wait is made.
- Allow iunique() to avoid taking the inode_hash_lock.
- Allow AFS's callback processing to avoid taking the inode_hash_lock
when using the inode table to find an inode to notify.
- Improve Ext4's time updating. Konstantin Khlebnikov said "For now,
I've plugged this issue with try-lock in ext4 lazy time update.
This solution is much better."
Then there's a set of changes to make a number of improvements to the
AFS driver:
- Improve callback (ie. third party change notification) processing
by:
(a) Relying more on the fact we're doing this under RCU and by
using fewer locks. This makes use of the RCU-based inode
searching outlined above.
(b) Moving to keeping volumes in a tree indexed by volume ID
rather than a flat list.
(c) Making the server and volume records logically part of the
cell. This means that a server record now points directly at
the cell and the tree of volumes is there. This removes an N:M
mapping table, simplifying things.
- Improve keeping NAT or firewall channels open for the server
callbacks to reach the client by actively polling the fileserver on
a timed basis, instead of only doing it when we have an operation
to process.
- Improving detection of delayed or lost callbacks by including the
parent directory in the list of file IDs to be queried when doing a
bulk status fetch from lookup. We can then check to see if our copy
of the directory has changed under us without us getting notified.
- Determine aliasing of cells (such as a cell that is pointed to be a
DNS alias). This allows us to avoid having ambiguity due to
apparently different cells using the same volume and file servers.
- Improve the fileserver rotation to do more probing when it detects
that all of the addresses to a server are listed as non-responsive.
It's possible that an address that previously stopped responding
has become responsive again.
Beyond that, lay some foundations for making some calls asynchronous:
- Turn the fileserver cursor struct into a general operation struct
and hang the parameters off of that rather than keeping them in
local variables and hang results off of that rather than the call
struct.
- Implement some general operation handling code and simplify the
callers of operations that affect a volume or a volume component
(such as a file). Most of the operation is now done by core code.
- Operations are supplied with a table of operations to issue
different variants of RPCs and to manage the completion, where all
the required data is held in the operation object, thereby allowing
these to be called from a workqueue.
- Put the standard "if (begin), while(select), call op, end" sequence
into a canned function that just emulates the current behaviour for
now.
There are also some fixes interspersed:
- Don't let the EACCES from ICMP6 mapping reach the user as such,
since it's confusing as to whether it's a filesystem error. Convert
it to EHOSTUNREACH.
- Don't use the epoch value acquired through probing a server. If we
have two servers with the same UUID but in different cells, it's
hard to draw conclusions from them having different epoch values.
- Don't interpret the argument to the CB.ProbeUuid RPC as a
fileserver UUID and look up a fileserver from it.
- Deal with servers in different cells having the same UUIDs. In the
event that a CB.InitCallBackState3 RPC is received, we have to
break the callback promises for every server record matching that
UUID.
- Don't let afs_statfs return values that go below 0.
- Don't use running fileserver probe state to make server selection
and address selection decisions on. Only make decisions on final
state as the running state is cleared at the start of probing"
Acked-by: Al Viro <viro@zeniv.linux.org.uk> (fs/inode.c part)
* tag 'afs-next-20200604' of git://git.kernel.org/pub/scm/linux/kernel/git/dhowells/linux-fs: (27 commits)
afs: Adjust the fileserver rotation algorithm to reprobe/retry more quickly
afs: Show more a bit more server state in /proc/net/afs/servers
afs: Don't use probe running state to make decisions outside probe code
afs: Fix afs_statfs() to not let the values go below zero
afs: Fix the by-UUID server tree to allow servers with the same UUID
afs: Reorganise volume and server trees to be rooted on the cell
afs: Add a tracepoint to track the lifetime of the afs_volume struct
afs: Detect cell aliases 3 - YFS Cells with a canonical cell name op
afs: Detect cell aliases 2 - Cells with no root volumes
afs: Detect cell aliases 1 - Cells with root volumes
afs: Implement client support for the YFSVL.GetCellName RPC op
afs: Retain more of the VLDB record for alias detection
afs: Fix handling of CB.ProbeUuid cache manager op
afs: Don't get epoch from a server because it may be ambiguous
afs: Build an abstraction around an "operation" concept
afs: Rename struct afs_fs_cursor to afs_operation
afs: Remove the error argument from afs_protocol_error()
afs: Set error flag rather than return error from file status decode
afs: Make callback processing more efficient.
afs: Show more information in /proc/net/afs/servers
...
Under some circumstances, rxrpc will fail a transmit a packet through the
underlying UDP socket (ie. UDP sendmsg returns an error). This may result
in a call getting stuck.
In the instance being seen, where AFS tries to send a probe to the Volume
Location server, tracepoints show the UDP Tx failure (in this case returing
error 99 EADDRNOTAVAIL) and then nothing more:
afs_make_vl_call: c=0000015d VL.GetCapabilities
rxrpc_call: c=0000015d NWc u=1 sp=rxrpc_kernel_begin_call+0x106/0x170 [rxrpc] a=00000000dd89ee8a
rxrpc_call: c=0000015d Gus u=2 sp=rxrpc_new_client_call+0x14f/0x580 [rxrpc] a=00000000e20e4b08
rxrpc_call: c=0000015d SEE u=2 sp=rxrpc_activate_one_channel+0x7b/0x1c0 [rxrpc] a=00000000e20e4b08
rxrpc_call: c=0000015d CON u=2 sp=rxrpc_kernel_begin_call+0x106/0x170 [rxrpc] a=00000000e20e4b08
rxrpc_tx_fail: c=0000015d r=1 ret=-99 CallDataNofrag
The problem is that if the initial packet fails and the retransmission
timer hasn't been started, the call is set to completed and an error is
returned from rxrpc_send_data_packet() to rxrpc_queue_packet(). Though
rxrpc_instant_resend() is called, this does nothing because the call is
marked completed.
So rxrpc_notify_socket() isn't called and the error is passed back up to
rxrpc_send_data(), rxrpc_kernel_send_data() and thence to afs_make_call()
and afs_vl_get_capabilities() where it is simply ignored because it is
assumed that the result of a probe will be collected asynchronously.
Fileserver probing is similarly affected via afs_fs_get_capabilities().
Fix this by always issuing a notification in __rxrpc_set_call_completion()
if it shifts a call to the completed state, even if an error is also
returned to the caller through the function return value.
Also put in a little bit of optimisation to avoid taking the call
state_lock and disabling softirqs if the call is already in the completed
state and remove some now redundant rxrpc_notify_socket() calls.
Fixes: f5c17aaeb2 ("rxrpc: Calls should only have one terminal state")
Reported-by: Gerry Seidman <gerry@auristor.com>
Signed-off-by: David Howells <dhowells@redhat.com>
Reviewed-by: Marc Dionne <marc.dionne@auristor.com>
Move the handling of call completion out of line so that the next patch can
add more code in that area.
Signed-off-by: David Howells <dhowells@redhat.com>
Reviewed-by: Marc Dionne <marc.dionne@auristor.com>
The user ID value isn't actually much use - and leaks a kernel pointer or a
userspace value - so replace it with the call debug ID, which appears in trace
points.
Signed-off-by: David Howells <dhowells@redhat.com>
Map the EACCES error that is produced by some ICMP6 packets to EHOSTUNREACH
when we get them as EACCES has other meanings within a filesystem context.
Signed-off-by: David Howells <dhowells@redhat.com>
Add a helper to directly set the RXRPC_MIN_SECURITY_LEVEL sockopt from
kernel space without going through a fake uaccess.
Thanks to David Howells for the documentation updates.
Signed-off-by: Christoph Hellwig <hch@lst.de>
Acked-by: David Howells <dhowells@redhat.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Add a helper to directly set the IPV6_RECVERR sockopt from kernel space
without going through a fake uaccess.
Signed-off-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: David Howells <dhowells@redhat.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Add a helper to directly set the IP_MTU_DISCOVER sockopt from kernel
space without going through a fake uaccess.
Signed-off-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: David Howells <dhowells@redhat.com> [rxrpc bits]
Signed-off-by: David S. Miller <davem@davemloft.net>
Add a helper to directly set the IP_RECVERR sockopt from kernel space
without going through a fake uaccess.
Signed-off-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: David Howells <dhowells@redhat.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Add a helper to directly enable timestamps instead of setting the
SO_TIMESTAMP* sockopts from kernel space and going through a fake
uaccess.
Signed-off-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: David S. Miller <davem@davemloft.net>
The MSCC bug fix in 'net' had to be slightly adjusted because the
register accesses are done slightly differently in net-next.
Signed-off-by: David S. Miller <davem@davemloft.net>
A ticket was not released after a call of the function
"rxkad_decrypt_ticket" failed. Thus replace the jump target
"temporary_error_free_resp" by "temporary_error_free_ticket".
Fixes: 8c2f826dc3 ("rxrpc: Don't put crypto buffers on the stack")
Signed-off-by: Qiushi Wu <wu000273@umn.edu>
Signed-off-by: David Howells <dhowells@redhat.com>
cc: Markus Elfring <Markus.Elfring@web.de>
The Rx protocol has a "previousPacket" field in it that is not handled in
the same way by all protocol implementations. Sometimes it contains the
serial number of the last DATA packet received, sometimes the sequence
number of the last DATA packet received and sometimes the highest sequence
number so far received.
AF_RXRPC is using this to weed out ACKs that are out of date (it's possible
for ACK packets to get reordered on the wire), but this does not work with
OpenAFS which will just stick the sequence number of the last packet seen
into previousPacket.
The issue being seen is that big AFS FS.StoreData RPC (eg. of ~256MiB) are
timing out when partly sent. A trace was captured, with an additional
tracepoint to show ACKs being discarded in rxrpc_input_ack(). Here's an
excerpt showing the problem.
52873.203230: rxrpc_tx_data: c=000004ae DATA ed1a3584:00000002 0002449c q=00024499 fl=09
A DATA packet with sequence number 00024499 has been transmitted (the "q="
field).
...
52873.243296: rxrpc_rx_ack: c=000004ae 00012a2b DLY r=00024499 f=00024497 p=00024496 n=0
52873.243376: rxrpc_rx_ack: c=000004ae 00012a2c IDL r=0002449b f=00024499 p=00024498 n=0
52873.243383: rxrpc_rx_ack: c=000004ae 00012a2d OOS r=0002449d f=00024499 p=0002449a n=2
The Out-Of-Sequence ACK indicates that the server didn't see DATA sequence
number 00024499, but did see seq 0002449a (previousPacket, shown as "p=",
skipped the number, but firstPacket, "f=", which shows the bottom of the
window is set at that point).
52873.252663: rxrpc_retransmit: c=000004ae q=24499 a=02 xp=14581537
52873.252664: rxrpc_tx_data: c=000004ae DATA ed1a3584:00000002 000244bc q=00024499 fl=0b *RETRANS*
The packet has been retransmitted. Retransmission recurs until the peer
says it got the packet.
52873.271013: rxrpc_rx_ack: c=000004ae 00012a31 OOS r=000244a1 f=00024499 p=0002449e n=6
More OOS ACKs indicate that the other packets that are already in the
transmission pipeline are being received. The specific-ACK list is up to 6
ACKs and NAKs.
...
52873.284792: rxrpc_rx_ack: c=000004ae 00012a49 OOS r=000244b9 f=00024499 p=000244b6 n=30
52873.284802: rxrpc_retransmit: c=000004ae q=24499 a=0a xp=63505500
52873.284804: rxrpc_tx_data: c=000004ae DATA ed1a3584:00000002 000244c2 q=00024499 fl=0b *RETRANS*
52873.287468: rxrpc_rx_ack: c=000004ae 00012a4a OOS r=000244ba f=00024499 p=000244b7 n=31
52873.287478: rxrpc_rx_ack: c=000004ae 00012a4b OOS r=000244bb f=00024499 p=000244b8 n=32
At this point, the server's receive window is full (n=32) with presumably 1
NAK'd packet and 31 ACK'd packets. We can't transmit any more packets.
52873.287488: rxrpc_retransmit: c=000004ae q=24499 a=0a xp=61327980
52873.287489: rxrpc_tx_data: c=000004ae DATA ed1a3584:00000002 000244c3 q=00024499 fl=0b *RETRANS*
52873.293850: rxrpc_rx_ack: c=000004ae 00012a4c DLY r=000244bc f=000244a0 p=00024499 n=25
And now we've received an ACK indicating that a DATA retransmission was
received. 7 packets have been processed (the occupied part of the window
moved, as indicated by f= and n=).
52873.293853: rxrpc_rx_discard_ack: c=000004ae r=00012a4c 000244a0<00024499 00024499<000244b8
However, the DLY ACK gets discarded because its previousPacket has gone
backwards (from p=000244b8, in the ACK at 52873.287478 to p=00024499 in the
ACK at 52873.293850).
We then end up in a continuous cycle of retransmit/discard. kafs fails to
update its window because it's discarding the ACKs and can't transmit an
extra packet that would clear the issue because the window is full.
OpenAFS doesn't change the previousPacket value in the ACKs because no new
DATA packets are received with a different previousPacket number.
Fix this by altering the discard check to only discard an ACK based on
previousPacket if there was no advance in the firstPacket. This allows us
to transmit a new packet which will cause previousPacket to advance in the
next ACK.
The check, however, needs to allow for the possibility that previousPacket
may actually have had the serial number placed in it instead - in which
case it will go outside the window and we should ignore it.
Fixes: 1a2391c30c ("rxrpc: Fix detection of out of order acks")
Reported-by: Dave Botsch <botsch@cnf.cornell.edu>
Signed-off-by: David Howells <dhowells@redhat.com>
