If we come to node we'd already marked as seen and it's not a part of path
(i.e. we don't have a loop right there), we already know that it isn't a
part of any loop, so we don't need to revisit it.
That speeds the things up if some chain is refered to from several places
and kills O(exp(table size)) worst-case behaviour (without sleeping,
at that, so if you manage to self-LART that way, you are SOL for a long
time)...
Signed-off-by: Al Viro <[email protected]>
Signed-off-by: Patrick McHardy <[email protected]>
Signed-off-by: David S. Miller <[email protected]>
We can check newinfo->hook_entry[...] instead.
Kill unused argument.
Signed-off-by: Al Viro <[email protected]>
Signed-off-by: David S. Miller <[email protected]>
Check for valid_hooks is redundant (newinfo->hook_entry[i] will
be NULL if bit i is not set). Kill it, kill unused arguments.
Signed-off-by: Al Viro <[email protected]>
Signed-off-by: David S. Miller <[email protected]>
Since newinfo->hook_table[] already has been set up, we can switch to using
it instead of repl->{hook_info,valid_hooks}.
Signed-off-by: Al Viro <[email protected]>
Signed-off-by: David S. Miller <[email protected]>
Take intialization of ->hook_entry[...], ->entries_size and ->nentries
over there, pull the check for empty chains into the end of that sucker.
Now it's self-contained, so we can move it up in the very beginning of
translate_table() *and* we can rely on ->hook_entry[] being properly
transliterated after it.
Signed-off-by: Al Viro <[email protected]>
Signed-off-by: David S. Miller <[email protected]>
It's easier to expand the iterator here *and* we'll be able to move all
uses of ebt_replace from translate_table() into this one.
Signed-off-by: Al Viro <[email protected]>
Signed-off-by: David S. Miller <[email protected]>
Split ebt_check_entry_size_and_hooks() in two parts - one that does
sanity checks on pointers (basically, checks that we can safely
use iterator from now on) and the rest of it (looking into details
of entry).
The loop applying ebt_check_entry_size_and_hooks() is split in two.
Populating newinfo->hook_entry[] is done in the first part.
Unused arguments killed.
Signed-off-by: Al Viro <[email protected]>
Signed-off-by: David S. Miller <[email protected]>
No need to revisit a chain we'd already finished with during
the check for current hook. It's either instant loop (which
we'd just detected) or a duplicate work.
Signed-off-by: Al Viro <[email protected]>
Signed-off-by: David S. Miller <[email protected]>
We need that for iterator to work; existing check had been too weak.
Signed-off-by: Al Viro <[email protected]>
Signed-off-by: David S. Miller <[email protected]>
We need to verify that
a) we are not too close to the end of buffer to dereference
b) next entry we'll be checking won't be _before_ our
While we are at it, don't subtract unrelated pointers...
Signed-off-by: Al Viro <[email protected]>
Signed-off-by: David S. Miller <[email protected]>
... into anonymous union of __wsum and __u32 (csum and csum_offset resp.)
Signed-off-by: Al Viro <[email protected]>
Signed-off-by: David S. Miller <[email protected]>
... and switch the damn checksum update to something saner
Signed-off-by: Al Viro <[email protected]>
Signed-off-by: David S. Miller <[email protected]>
... switched to taking and returning pointers to net-endian
sctp_addr resp. Together, since the only user of sctp_find_unmatch_addr()
just passes its value to sctp_make_asconf_update_ip().
sctp_make_asconf_update_ip() is actually endian-agnostic.
Signed-off-by: Al Viro <[email protected]>
Signed-off-by: David S. Miller <[email protected]>
both are done in one go since almost always we have result of
the latter immediately passed to the former. Possibly non-obvious
note: sctp_process_param() is endian-agnostic
Signed-off-by: Al Viro <[email protected]>
Signed-off-by: David S. Miller <[email protected]>
Along with it, statics in input.c that end up calling it
(__sctp_lookup_association, sctp_lookup_association,
__sctp_rcv_init_lookup, __sctp_rcv_lookup). Callers
are adjusted.
Signed-off-by: Al Viro <[email protected]>
Signed-off-by: David S. Miller <[email protected]>
The only caller (__sctp_rcv_lookup_endpoint()) also switched,
its caller adjusted
Signed-off-by: Al Viro <[email protected]>
Signed-off-by: David S. Miller <[email protected]>
Its only use happens on the same host, when it gets quoted back to
us. So we are free to flip to net-endian and avoid extra PITA.
Signed-off-by: Al Viro <[email protected]>
Signed-off-by: David S. Miller <[email protected]>
switched to taking a pointer to net-endian sctp_addr
and a net-endian port number. Instances and callers
adjusted; interestingly enough, the only calls are
direct calls of specific instances - the method is not
used at all.
Signed-off-by: Al Viro <[email protected]>
Signed-off-by: David S. Miller <[email protected]>
instances of ->cmp_addr() are fine with switching both arguments
to net-endian; callers other than in sctp_cmp_addr_exact() (both
as ->cmp_addr(...) and direct calls of instances) adjusted;
sctp_cmp_addr_exact() switched to net-endian itself and adjustment
is done in its callers
Signed-off-by: Al Viro <[email protected]>
Signed-off-by: David S. Miller <[email protected]>
No actual modifications of method instances are needed -
they don't look at port numbers. Switch callers...
Signed-off-by: Al Viro <[email protected]>
Signed-off-by: David S. Miller <[email protected]>
Add sctp_chunk->source, sctp_sockaddr_entry->a, sctp_transport->ipaddr
and sctp_transport->saddr, maintain them as net-endian mirrors of
their host-endian counterparts.
Signed-off-by: Al Viro <[email protected]>
Signed-off-by: David S. Miller <[email protected]>
Part 1: rename sctp_chunk->source, sctp_sockaddr_entry->a,
sctp_transport->ipaddr and sctp_transport->saddr (to ..._h)
The next patch will reintroduce these fields and keep them as
net-endian mirrors of the original (renamed) ones. Split in
two patches to make sure that we hadn't forgotten any instanes.
Later in the series we'll eliminate uses of host-endian variants
(basically switching users to net-endian counterparts as we
progress through that mess). Then host-endian ones will die.
Other embedded host-endian sctp_addr will be easier to switch
directly, so we leave them alone for now.
Signed-off-by: Al Viro <[email protected]>
Signed-off-by: David S. Miller <[email protected]>
Again, invalid sockaddr passed to userland - host-endiand sin_port.
Potential leak, again, but less dramatic than in previous case.
Signed-off-by: Al Viro <[email protected]>
Signed-off-by: David S. Miller <[email protected]>
a) struct sockaddr_storage * passed to sctp_ulpevent_make_peer_addr_change()
actually points at union sctp_addr field in a structure. Then that sucker
gets copied to userland, with whatever junk we might have there.
b) it's actually having host-endian sin_port.
Signed-off-by: Al Viro <[email protected]>
Signed-off-by: David S. Miller <[email protected]>
It expects (and gets) laddr with net-endian sin_port. And then it calls
sctp_bind_addr_match(), which *does* care about port numbers in case of
ipv6 and expects them to be host-endian.
Signed-off-by: Al Viro <[email protected]>
Signed-off-by: David S. Miller <[email protected]>
That's going to be a long series. Introduced temporary helpers
doing copy-and-convert for sctp_addr; they are used to kill
flip-in-place in global data structures and will be used
to gradually push host-endian uses of sctp_addr out of existence.
Signed-off-by: Al Viro <[email protected]>
Signed-off-by: David S. Miller <[email protected]>
also always get __be16 protocol error; switch to SCTP_PERR()
Signed-off-by: Al Viro <[email protected]>
Signed-off-by: David S. Miller <[email protected]>
argument stored for SCTP_CMD_INIT_FAILED is always __be16
(protocol error). Introduced new field and accessor for
it (SCTP_PERR()); switched to their use (from SCTP_U32() and
.u32)
Signed-off-by: Al Viro <[email protected]>
Signed-off-by: David S. Miller <[email protected]>
Al Viro