User Details
- User Since
- Apr 13 2023, 7:00 PM (187 w, 6 d)
Today
Yesterday
Rebased on D61039 Diff 2: the new min-ttl 255 leg in fragmentation_pass:v6_refrag_hlim chooses the path with vnet_ip6_forward_path, like the rest of the tests here; no other change.
Give the hop back only below IPV6_MAXHLIM. When an outbound scrub (min-ttl 255) had raised the hop limit to 255, the give-back wrapped it to 0 and ip6_forward() dropped the refragmented packet with time exceeded (thanks, @pouria). Such a packet now leaves with 254, as it did before this change. fragmentation_pass:v6_refrag_hlim gets a min-ttl 255 leg over both paths; summary updated to match.
Rebased on D61038 Diff 3 (fwd.sh uses sysctl -j there); no other change.
nexthop_v4 now uses ifconfig -j, route -j and sysctl -j, same as nexthop_v6 in D61038.
The new test now uses ifconfig -j, route -j and sysctl -j instead of jexec; the change to ip6_fastfwd.c is unchanged
Tue, Nov 17
Rebased onto the two new fixes below it in the stack, and reworked the tests. The change to ip6_fastfwd.c and ip6_input.c is the same.
Rewrote the test; the kernel change is the same.
Mon, Nov 16
Rebased onto 1b9a8e1dcc69 (D61194 as committed), plus one fix from a final review pass: the bundle relabel tested the final SA's own mode, but an SA added without -m has mode "any" and still serves a transport request, so such a bundle kept the label of the tunnel SA beneath. It now tests the policy request's mode (sp->req[idx - 1]), as the encapsulation decision does.
rebased and drop an initializer
The six callbacks now call ipsec_crypto_retry(), which records the new session in the SA, updates xd->cryptoid and clears crp_etype. ipsec_updateid() has no other callers and becomes static. The callbacks still call crypto_dispatch() themselves, because the input callbacks have to release the sahtree lock first.
Sun, Nov 15
Fri, Nov 13
Thu, Nov 12
Thanks all, I am working on a new revision!
Wed, Nov 11
- Size the AAD buffer for the high-order half of the sequence number as well.
- Allocate struct esp_req from a dedicated UMA zone, created in esp_attach() and destroyed in esp_detach() after xform_detach(). M_ESP is gone.
- Pass the esp_req as crp_opaque, so esp_freereq() takes it directly and no longer needs __containerof().
Drop the ipsec_default_discard() helper and compare V_def_policy->policy with IPSEC_POLICY_DISCARD in the callers. Say in the comment on ipsec[46]_getpolicy() what a NULL return means.
The enum is now named enum pf_addr_type, and pf_hash_addr() switches on (enum pf_addr_type)addr->type with no default case. Leaving a type out is now a -Wswitch error in the kernel build (checked by dropping PF_ADDR_URPFFAILED).
Tue, Nov 10
Add a comment at the lookup saying why no reference is needed (@zlei's question). No code change. Test plan extended with the nexthop lifetime and failed try-ref runs.
Mon, Nov 9
Reworked to keep the change minimal, and to answer the ABI point properly:
fetch_getln() no longer returns a line the peer cut off before
its newline as if it were complete. A chunk size line cut after a
leading "0" used to end the body as complete, and a truncated status,
header or FTP reply line was parsed as whatever was left; it is EPROTO
now. An empty remainder still returns an empty line, so headers
without the final blank line end as before.
Updated: a read error while waiting for a chunk header is no longer
reported as EPROTO. http_new_chunk() sets errno to EPROTO itself for
a malformed size, and http_fillbuf() records errno, so a stall between
chunks fails as "transfer timed out" like a stall inside a chunk does.
Sun, Nov 8
Enable symbol versioning
