For ESP and AH, we need to allocate an extra variable space in addition
to struct tdb_crypto. The fixed size of pool items may be larger than
an actual requisite size of a buffer, but still the performance
improvement by replacing malloc with pool wins.
An SP has a list of isr (ipsecrequest) that represents a sequence
of IPsec encryption/authentication processing. One isr corresponds
to one opencrypto processing. The lifetime of an isr follows its SP.
We pass an isr to a callback function of opencrypto to continue
to a next encryption/authentication processing. However nobody
guaranteed that the isr wasn't freed, i.e., its SP wasn't destroyed.
In order to avoid such unexpected destruction of isr, hold a reference
to its SP during opencrypto processing.
isr is a shared resource and using isr->sav as a temporal storage
for each packet processing is racy. And also having a reference from
isr to sav makes the lifetime of sav non-deterministic; such a reference
is removed when a packet is processed and isr->sav is overwritten by
new one. Let's have a sav locally for each packet processing instead of
using shared isr->sav.
However this change doesn't stop using isr->sav yet because there are
some users of isr->sav. isr->sav will be removed after the users find
a way to not use isr->sav.
It seems that PACKET_TAG_IPSEC_IN_CRYPTO_DONE is for network adapters
that have IPsec accelerators; a driver sets the mtag to a packet
when its device has already encrypted the packet.
Unfortunately no driver implements such offload features for long
years and seems unlikely to implement them soon. (Note that neither
FreeBSD nor Linux doesn't have such drivers.) Let's remove related
(unused) codes and simplify the IPsec code.
No functional changes (except for a debug printf).
Note that there remain some __FreeBSD__ for sysctl knobs which counerparts
to NetBSD don't exist. And ipsec_osdep.h isn't touched yet; tidying it up
requires actual code changes.
- Include "opt_*.h" only if _KERNEL_OPT is defined
- Allow encapinit to be called twice (by ifinit and ipe4_attach)
- ifinit didn't call encapinit if IPSEC is enabled (ipe4_attach called
it instead), however, on a rump kernel ipe4_attach may not be called
even if IPSEC is enabled. So we need to allow ifinit to call it anyway
- Setup sysctls in ipsec_attach explicitly instead of using SYSCTL_SETUP
- Call ip6flow_invalidate_all in key_spdadd only if in6_present
- It's possible that a rump kernel loads the ipsec library but not
the inet6 library
consttime_memequal is the same as the old consttime_bcmp.
explicit_memset is to memset as explicit_bzero was to bcmp.
Passes amd64 release and i386/ALL, but I'm sure I missed some spots,
so please let me know.
outdated pointers and pass ESP data to UPD-sockets.
While here, simplify the code and remove the IPSEC_NAT_T option; always
compile nat-traversal in so that it does not bitrot.
abd userland, as proposed on tech-security, with explicit_bzero using
a volatile function pointer as suggested by Alan Barrett.
Both do what the name says. For userland, both are prefixed by "__"
to keep them out of the user namespace.
Change some memset/memcmp uses to the new functions where it makes
sense -- these are just some examples, more to come.
to modify in the AH case) are writable/non-shared.
This addresses PR kern/33162 by Jeff Rizzo, and replaces the insufficient
patch from that time by a radical solution.
(The PR's problem had been worked around by rev.1.3 of xennetback_xenbus.c,
so it needs a network driver modification to reproduce it.)
Being here, clarify a bit of ipcomp -- uncompression is done in-place,
the header must be removed explicitly.
This is still somewhat experimental. Tested between 2 similar boxes
so far. There is much potential for performance improvement. For now,
I've changed the gmac code to accept any data alignment, as the "char *"
pointer suggests. As the code is practically used, 32-bit alignment
can be assumed, at the cost of data copies. I don't know whether
bytewise access or copies are worse performance-wise. For efficient
implementations using SSE2 instructions on x86, even stricter
alignment requirements might arise.
into "blocksize" and "IV size"
-add an "reinit" function pointer which, if set, means that the xform
does its IV handling itself and doesn't want the default CBC handling
by the framework (poor name, but left that way to avoid unecessary
differences)
This syncs with Open/FreeBSD, purpose is to allow non-CBC transforms.
Refer to ivsize instead of blocksize where appropriate.
(At this point, blocksize and ivsize are identical.)
crypto_{new.free}session() to be called with the "crypto_mtx"
spinlock held.
This doesn't change much for now because these functions acquire
the said mutex first on entry now, but at least it keeps the nasty
locks local to the opencrypto core.
-RFC2104 says that the block size of the hash algorithm must be used
for key/ipad/opad calculations. While formerly all ciphers used a block
length of 64, SHA384 and SHA512 use 128 bytes. So we can't use the
HMAC_BLOCK_LEN constant anymore. Add a new field to "struct auth_hash"
for the per-cipher blocksize.
-Due to this, there can't be a single "CRYPTO_SHA2_HMAC" external name
anymore. Replace this by 3 for the 3 different keysizes.
This was done by Open/FreeBSD before.
-Also fix the number of authenticator bits used tor ESP and AH to
conform to RFC4868, and remove uses of AH_HMAC_HASHLEN which did
assume a fixed authenticator size of 12 bytes.
FAST_IPSEC will not interoperate with KAME IPSEC anymore if sha2 is used,
because the latter doesn't implement these standards. It should
interoperate with at least modern Free/OpenBSD now.
(I've only tested with NetBSD-current/FAST_IPSEC on both ends.)
everywhere splsoftnet() was used before, to fix MP concurrency problems
-pull KERNEL_LOCK where ip(6)_output() is called, as this is what
the network stack (unfortunately) expects, in particular to avoid
races for packets in the interface send queues
From Wolfgang Stukenbrock per PR kern/44418, with the application
of KERNEL_LOCK to what I think are the essential points, tested
on a dual-core i386.