a socket from a regular socket to a listening socket able to accept new
connections. As part of this state transition, solisten() calls into the
protocol to update protocol-layer state. There were several bugs in this
implementation that could result in a race wherein a TCP SYN received
in the interval between the protocol state transition and the shortly
following socket layer transition would result in a panic in the TCP code,
as the socket would be in the TCPS_LISTEN state, but the socket would not
have the SO_ACCEPTCONN flag set.
This change does the following:
- Pushes the socket state transition from the socket layer solisten() to
to socket "library" routines called from the protocol. This permits
the socket routines to be called while holding the protocol mutexes,
preventing a race exposing the incomplete socket state transition to TCP
after the TCP state transition has completed. The check for a socket
layer state transition is performed by solisten_proto_check(), and the
actual transition is performed by solisten_proto().
- Holds the socket lock for the duration of the socket state test and set,
and over the protocol layer state transition, which is now possible as
the socket lock is acquired by the protocol layer, rather than vice
versa. This prevents additional state related races in the socket
layer.
This permits the dual transition of socket layer and protocol layer state
to occur while holding locks for both layers, making the two changes
atomic with respect to one another. Similar changes are likely require
elsewhere in the socket/protocol code.
Reported by: Peter Holm <peter@holm.cc>
Review and fixes from: emax, Antoine Brodin <antoine.brodin@laposte.net>
Philosophical head nod: gnn
structures in IPX/SPX -- primarily, sequence numbering, PCB lists,
and PCBs for IPX raw sockets, IPX datagram sockets, and IPX/SPX.
As such, remove remove NET_NEEDS_GIANT() for IPX, and remove the
assertion of Giant in the ipxintr() IPX input path.
Note that IPX/SPX is not fully MPSAFE, and that there are some
problems with IPX/SPX locking that will require some further work.
However, it is now safe enough to run in general without the Giant
lock.
MFC after: 4 weeks
portion of IPX/SPX:
- Protect IPX PCB lists with the IPX PCB list mutex, in particular
when calling PCB and PCB list manipulation routines in ipx_pcb.c.
- Protect both IPX PCB state and SPX PCB state using the IPX PCB
mutex.
- Generally annotate locking, as well as adding liberal use of lock
assertions to document locking requirements.
- Where possible, use unlocked reads when reading integer or smaller
sized socket options on SPX sockets.
- De-spl throughout.
Notes:
- spx_input() expects both the list mutex and PCB mutex to be held
on entry, but will release both on return. Because sonewconn() is
called from spx_input(), it may actually drop one PCB lock and
acquire another during generation of a new connection, meaning the
caller is not in a position to unlock the PCB mutex.
MFC after: 3 weeks
were derived from more complex TCP versions of the same:
- spx_close(), spx_disconnect(), spx_drop(), and spx_usrclosed() all
always free's the spxpcb invalidating the argument, so a return
value is not required to indicate if it has.
- Annotate that the cb arguments to each of these functions is
invalidated via a comment.
- When tearing down a pcb due to sonewconn() having failed, mark the
cb as NULL; later, when deciding whether to store trace information
due to SO_DEBUG, check that cb is not NULL before dereferencing or
a NULL pointer dereference may occur.
MFC after: 3 weeks
When processing socket options against IPX PCBs, generally protect
PCB fields using the IPX PCB mutex. Where possible, use unlocked
reads on integer values to avoid locking overhead.
MFC after: 3 weeks
protocol methods relating to IPX. Conditionally acquire the PCB list
lock in the send operation only if the socket requires binding in order
to use the requested address.
Remove spl's generally no longer required during these accesses.
MFC after: 3 weeks
the IPX-related PCB routines. In general, the list lock is required
to iterate the PCB list, either for read or write; the PCB lock is
required to access or modify a PCB. To change the binding of a PCB,
both locks must be held.
MFC after: 3 weeks
IPX PCB lists. Add macros to initialize, destroy, lock, unlock,
and assert the mutex. Initialize the mutex when IPX is started.
Add per-IPX PCB mutexes, ipxp_mtx in struct ipxpcb, to protect
per-PCB IPX/SPX state. Add macros to initialize, destroy, lock,
unlock, and assert the mutex. Initialize the mutex when a new
PCB is allocated; destroy it when the PCB is free'd.
MFC after: 2 weeks
spx_reass() to increase atomicity across multiple operations on the
socket buffer when iterating over the SPX fragment reassembly list
for the ipxpcb, as well a to reduce the number of locking operations.
record loop for ACK'd data, rather than relying on lokcing in
sbdroprecord() and sowwakeup(), reducing the number of lock operations
as well as eliminating a possible race against the head of the send
buffer mbuf chain. Use the _locked variants of sbdroprecord() and
sowwakeup().
the peer address by using M_WAITOK in ipx_setpeeraddr() to prevent
allocation failure. The socket reference used to reach these calls
will prevent the ipxpcb from being released prematurely.
properly handle the case where a connection is disconnected. The
queue(9)-enabled version of this code broke from the inner but not
outer loop, and so potentially frobbed an ipxpcb flag after the ipxpcb
was free'd, which might be picked up later by the malloc debugging
code. Properly break from the loop context and avoid touching the
cb/ipxpcb after free.
expects a locked route reference. This removes a panic that occurs
when connected ipxpcb is closed and its route free'd, and may have been
present since the route locking took place.
MFC after: 2 weeks
the netperf branch but for some reason didn't trigger a build failure
locally when I merged to CVS and omitted it. Presumably driver error.
Pointed out by: cperciva, tinderbox
- ipx_pcbnotify(), which is never called.
- ipx_rtchange(), which is never called, is incomplete inplemented, and
also #ifdef notdef.
- spx_fixmtu(), which is never called, is incompletely implemented, and
also #ifdef notdef.
(sorele()/sotryfree()):
- This permits the caller to acquire the accept mutex before the socket
mutex, avoiding sofree() having to drop the socket mutex and re-order,
which could lead to races permitting more than one thread to enter
sofree() after a socket is ready to be free'd.
- This also covers clearing of the so_pcb weak socket reference from
the protocol to the socket, preventing races in clearing and
evaluation of the reference such that sofree() might be called more
than once on the same socket.
This appears to close a race I was able to easily trigger by repeatedly
opening and resetting TCP connections to a host, in which the
tcp_close() code called as a result of the RST raced with the close()
of the accepted socket in the user process resulting in simultaneous
attempts to de-allocate the same socket. The new locking increases
the overhead for operations that may potentially free the socket, so we
will want to revise the synchronization strategy here as we normalize
the reference counting model for sockets. The use of the accept mutex
in freeing of sockets that are not listen sockets is primarily
motivated by the potential need to remove the socket from the
incomplete connection queue on its parent (listen) socket, so cleaning
up the reference model here may allow us to substantially weaken the
synchronization requirements.
RELENG_5_3 candidate.
MFC after: 3 days
Reviewed by: dwhite
Discussed with: gnn, dwhite, green
Reported by: Marc UBM Bocklet <ubm at u-boot-man dot de>
Reported by: Vlad <marchenko at gmail dot com>
operation using NET_NEEDS_GIANT(). This will result in a boot-time
restoration of Giant-enabled network operation, or run-time warning on
dynamic load (applicable only to the Netgraph component). Additional
components will likely need to be marked with this in the future.
socket from its accept queue when aborting it during a new inbound
connection. Update spx_input() to acquire the accept lock, assert
the condition of the socket on its parent queue, and approriately
disconnect it from the queue before calling soabort() on it.
flags relating to several aspects of socket functionality. This change
breaks out several bits relating to send and receive operation into a
new per-socket buffer field, sb_state, in order to facilitate locking.
This is required because, in order to provide more granular locking of
sockets, different state fields have different locking properties. The
following fields are moved to sb_state:
SS_CANTRCVMORE (so_state)
SS_CANTSENDMORE (so_state)
SS_RCVATMARK (so_state)
Rename respectively to:
SBS_CANTRCVMORE (so_rcv.sb_state)
SBS_CANTSENDMORE (so_snd.sb_state)
SBS_RCVATMARK (so_rcv.sb_state)
This facilitates locking by isolating fields to be located with other
identically locked fields, and permits greater granularity in socket
locking by avoiding storing fields with different locking semantics in
the same short (avoiding locking conflicts). In the future, we may
wish to coallesce sb_state and sb_flags; for the time being I leave
them separate and there is no additional memory overhead due to the
packing/alignment of shorts in the socket buffer structure.
