safe/jmp/linux-2.6
17 years ago[CRYPTO] digest: Added user API for new hash type
Herbert Xu [Sat, 19 Aug 2006 12:24:23 +0000 (22:24 +1000)]
[CRYPTO] digest: Added user API for new hash type

The existing digest user interface is inadequate for support asynchronous
operations.  For one it doesn't return a value to indicate success or
failure, nor does it take a per-operation descriptor which is essential
for the issuing of requests while other requests are still outstanding.

This patch is the first in a series of steps to remodel the interface
for asynchronous operations.

For the ease of transition the new interface will be known as "hash"
while the old one will remain as "digest".

This patch also changes sg_next to allow chaining.

Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
17 years ago[CRYPTO] api: Mark parts of cipher interface as deprecated
Herbert Xu [Mon, 21 Aug 2006 11:40:49 +0000 (21:40 +1000)]
[CRYPTO] api: Mark parts of cipher interface as deprecated

Mark the parts of the cipher interface that have been replaced by
block ciphers as deprecated.  Thanks to Andrew Morton for suggesting
doing this before removing them completely.

Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
17 years ago[PATCH] scatterlist: Add const to sg_set_buf/sg_init_one pointer argument
Herbert Xu [Mon, 14 Aug 2006 13:11:53 +0000 (23:11 +1000)]
[PATCH] scatterlist: Add const to sg_set_buf/sg_init_one pointer argument

This patch adds a const modifier to the buf argument of sg_set_buf and
sg_init_one.  This lets people call it with pointers that are const.

Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
17 years ago[CRYPTO] drivers: Remove obsolete block cipher operations
Herbert Xu [Sat, 5 Aug 2006 06:28:19 +0000 (16:28 +1000)]
[CRYPTO] drivers: Remove obsolete block cipher operations

This patch removes obsolete block operations of the simple cipher type
from drivers.  These were preserved so that existing users can make a
smooth transition.  Now that the transition is complete, they are no
longer needed.

Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
17 years ago[CRYPTO] users: Use block ciphers where applicable
Herbert Xu [Tue, 22 Aug 2006 10:36:13 +0000 (20:36 +1000)]
[CRYPTO] users: Use block ciphers where applicable

This patch converts all remaining users to use the new block cipher type
where applicable.  It also changes all simple cipher operations to use
the new encrypt_one/decrypt_one interface.

Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
17 years ago[SUNRPC] GSS: Use block ciphers where applicable
Herbert Xu [Tue, 22 Aug 2006 10:33:54 +0000 (20:33 +1000)]
[SUNRPC] GSS: Use block ciphers where applicable

This patch converts SUNRPC/GSS to use the new block cipher type where
applicable.

Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
17 years ago[IPSEC] ESP: Use block ciphers where applicable
Herbert Xu [Sun, 30 Jul 2006 05:41:01 +0000 (15:41 +1000)]
[IPSEC] ESP: Use block ciphers where applicable

This patch converts IPSec/ESP to use the new block cipher type where
applicable.  Similar to the HMAC conversion, existing algorithm names
have been kept for compatibility.

Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
17 years ago[IPSEC]: Add compatibility algorithm name support
Herbert Xu [Sat, 12 Aug 2006 22:50:00 +0000 (08:50 +1000)]
[IPSEC]: Add compatibility algorithm name support

This patch adds a compatibility name field for each IPsec algorithm.  This
is needed when parameterised algorithms are used.  For example, "md5" will
become "hmac(md5)", and "aes" will become "cbc(aes)".

Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
17 years ago[BLOCK] dm-crypt: Use block ciphers where applicable
Herbert Xu [Tue, 22 Aug 2006 10:29:17 +0000 (20:29 +1000)]
[BLOCK] dm-crypt: Use block ciphers where applicable

This patch converts dm-crypt to use the new block cipher type where
applicable.  It also changes simple cipher operations to use the new
encrypt_one/decrypt_one interface.

Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
17 years ago[BLOCK] cryptoloop: Use block ciphers where applicable
Herbert Xu [Thu, 21 Sep 2006 01:45:53 +0000 (11:45 +1000)]
[BLOCK] cryptoloop: Use block ciphers where applicable

This patch converts cryptoloop to use the new block cipher type where
applicable.  As a result the ECB-specific and CBC-specific transfer
functions have been merged.

Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
17 years ago[CRYPTO] tcrypt: Use block ciphers where applicable
Herbert Xu [Sat, 12 Aug 2006 22:26:09 +0000 (08:26 +1000)]
[CRYPTO] tcrypt: Use block ciphers where applicable

This patch converts tcrypt to use the new block cipher type where
applicable.

Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
17 years ago[CRYPTO] s390: Added block cipher versions of CBC/ECB
Herbert Xu [Mon, 21 Aug 2006 11:39:24 +0000 (21:39 +1000)]
[CRYPTO] s390: Added block cipher versions of CBC/ECB

This patch adds block cipher algorithms for S390.  Once all users of the
old cipher type have been converted the existing CBC/ECB non-block cipher
operations will be removed.

Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
17 years ago[CRYPTO] padlock: Added block cipher versions of CBC/ECB
Herbert Xu [Mon, 21 Aug 2006 11:38:42 +0000 (21:38 +1000)]
[CRYPTO] padlock: Added block cipher versions of CBC/ECB

This patch adds block cipher algorithms for cbc(aes) and ecb(aes) for
the PadLock device.  Once all users to the old cipher type have been
converted the old cbc/ecb PadLock operations will be removed.

Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
17 years ago[CRYPTO] cipher: Added block ciphers for CBC/ECB
Herbert Xu [Thu, 21 Sep 2006 01:44:08 +0000 (11:44 +1000)]
[CRYPTO] cipher: Added block ciphers for CBC/ECB

This patch adds two block cipher algorithms, CBC and ECB.  These
are implemented as templates on top of existing single-block cipher
algorithms.  They invoke the single-block cipher through the new
encrypt_one/decrypt_one interface.

This also optimises the in-place encryption and decryption to remove
the cost of an IV copy each round.

Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
17 years ago[CRYPTO] cipher: Added block cipher type
Herbert Xu [Mon, 21 Aug 2006 14:07:53 +0000 (00:07 +1000)]
[CRYPTO] cipher: Added block cipher type

This patch adds the new type of block ciphers.  Unlike current cipher
algorithms which operate on a single block at a time, block ciphers
operate on an arbitrarily long linear area of data.  As it is block-based,
it will skip any data remaining at the end which cannot form a block.

The block cipher has one major difference when compared to the existing
block cipher implementation.  The sg walking is now performed by the
algorithm rather than the cipher mid-layer.  This is needed for drivers
that directly support sg lists.  It also improves performance for all
algorithms as it reduces the total number of indirect calls by one.

In future the existing cipher algorithm will be converted to only have
a single-block interface.  This will be done after all existing users
have switched over to the new block cipher type.

Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
17 years ago[CRYPTO] scatterwalk: Prepare for block ciphers
Herbert Xu [Sat, 12 Aug 2006 11:56:17 +0000 (21:56 +1000)]
[CRYPTO] scatterwalk: Prepare for block ciphers

This patch prepares the scatterwalk code for use by the new block cipher
type.

Firstly it halves the size of scatter_walk on 32-bit platforms.  This
is important as we allocate at least two of these objects on the stack
for each block cipher operation.

It also exports the symbols since the block cipher code can be built as
a module.

Finally there is a hack in scatterwalk_unmap that relies on progress
being made.  Unfortunately, for hardware crypto we can't guarantee
progress to be made since the hardware can fail.

So this also gets rid of the hack by not advancing the address returned
by scatterwalk_map.

Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
17 years ago[CRYPTO] cipher: Added encrypt_one/decrypt_one
Herbert Xu [Sun, 13 Aug 2006 10:58:18 +0000 (20:58 +1000)]
[CRYPTO] cipher: Added encrypt_one/decrypt_one

This patch adds two new operations for the simple cipher that encrypts or
decrypts a single block at a time.  This will be the main interface after
the existing block operations have moved over to the new block ciphers.

It also adds the crypto_cipher type which is currently only used on the
new operations but will be extended to setkey as well once existing users
have been converted to use block ciphers where applicable.

Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
17 years ago[CRYPTO] api: Added crypto_type support
Herbert Xu [Mon, 21 Aug 2006 14:06:54 +0000 (00:06 +1000)]
[CRYPTO] api: Added crypto_type support

This patch adds the crypto_type structure which will be used for all new
crypto algorithm types, beginning with block ciphers.

The primary purpose of this abstraction is to allow different crypto_type
objects for crypto algorithms of the same type, in particular, there will
be a different crypto_type objects for asynchronous algorithms.

Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
17 years ago[CRYPTO] api: Feed flag directly to crypto_yield
Herbert Xu [Sun, 30 Jul 2006 01:53:45 +0000 (11:53 +1000)]
[CRYPTO] api: Feed flag directly to crypto_yield

The sleeping flag used to determine whether crypto_yield can actually
yield is really a per-operation flag rather than a per-tfm flag.  This
patch changes crypto_yield to take a flag directly so that we can start
using a per-operation flag instead the tfm flag.

Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
17 years ago[CRYPTO] api: Added crypto_alloc_base
Herbert Xu [Sun, 30 Jul 2006 01:53:01 +0000 (11:53 +1000)]
[CRYPTO] api: Added crypto_alloc_base

Up until now all crypto transforms have been of the same type, struct
crypto_tfm, regardless of whether they are ciphers, digests, or other
types.  As a result of that, we check the types at run-time before
each crypto operation.

This is rather cumbersome.  We could instead use different C types for
each crypto type to ensure that the correct types are used at compile
time.  That is, we would have crypto_cipher/crypto_digest instead of
just crypto_tfm.  The appropriate type would then be required for the
actual operations such as crypto_digest_digest.

Now that we have the type/mask fields when looking up algorithms, it
is easy to request for an algorithm of the precise type that the user
wants.  However, crypto_alloc_tfm currently does not expose these new
attributes.

This patch introduces the function crypto_alloc_base which will carry
these new parameters.  It will be renamed to crypto_alloc_tfm once
all existing users have been converted.

Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
17 years ago[CRYPTO] s390: Added missing driver name and priority
Herbert Xu [Mon, 21 Aug 2006 11:18:50 +0000 (21:18 +1000)]
[CRYPTO] s390: Added missing driver name and priority

Accelerated versions of crypto algorithms must carry a distinct driver name
and priority in order to distinguish themselves from their generic counter-
part.

Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
17 years ago[CRYPTO] api: Added asynchronous flag
Herbert Xu [Sun, 6 Aug 2006 13:12:59 +0000 (23:12 +1000)]
[CRYPTO] api: Added asynchronous flag

This patch adds the asynchronous flag and changes all existing users to
only look up algorithms that are synchronous.

Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
17 years ago[CRYPTO] api: Add common instance initialisation code
Herbert Xu [Sun, 6 Aug 2006 13:10:45 +0000 (23:10 +1000)]
[CRYPTO] api: Add common instance initialisation code

This patch adds the helpers crypto_get_attr_alg and crypto_alloc_instance
which can be used by simple one-argument templates like hmac to process
input parameters and allocate instances.

Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
17 years ago[CRYPTO] cipher: Removed special IV checks for ECB
Herbert Xu [Fri, 14 Jul 2006 00:42:27 +0000 (10:42 +1000)]
[CRYPTO] cipher: Removed special IV checks for ECB

This patch makes IV operations on ECB fail through nocrypt_iv rather than
calling BUG().  This is needed to generalise CBC/ECB using the template
mechanism.

Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
17 years ago[CRYPTO] tcrypt: Use test_hash for crc32c
Herbert Xu [Mon, 21 Aug 2006 12:04:03 +0000 (22:04 +1000)]
[CRYPTO] tcrypt: Use test_hash for crc32c

Now that crc32c has been fixed to conform with standard digest semantics,
we can use test_hash for it.  I've turned the last test into a chunky
test.

Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
17 years ago[CRYPTO] digest: Store temporary digest in tfm
Herbert Xu [Sun, 9 Jul 2006 04:49:42 +0000 (14:49 +1000)]
[CRYPTO] digest: Store temporary digest in tfm

When the final result location is unaligned, we store the digest in a
temporary buffer before copying it to the final location.  Currently
that buffer sits on the stack.  This patch moves it to an area in the
tfm, just like the CBC IV buffer.

Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
17 years ago[CRYPTO] api: Get rid of flags argument to setkey
Herbert Xu [Sun, 13 Aug 2006 04:16:39 +0000 (14:16 +1000)]
[CRYPTO] api: Get rid of flags argument to setkey

Now that the tfm is passed directly to setkey instead of the ctx, we no
longer need to pass the &tfm->crt_flags pointer.

This patch also gets rid of a few unnecessary checks on the key length
for ciphers as the cipher layer guarantees that the key length is within
the bounds specified by the algorithm.

Rather than testing dia_setkey every time, this patch does it only once
during crypto_alloc_tfm.  The redundant check from crypto_digest_setkey
is also removed.

Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
17 years ago[CRYPTO] crc32c: Fix unconventional setkey usage
Herbert Xu [Sun, 6 Aug 2006 13:03:08 +0000 (23:03 +1000)]
[CRYPTO] crc32c: Fix unconventional setkey usage

The convention for setkey is that once it is set it should not change,
in particular, init must not wipe out the key set by it.  In fact, init
should always be used after setkey before any digestion is performed.

The only user of crc32c that sets the key is tcrypt.  This patch adds
the necessary init calls there.

Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
17 years ago[CRYPTO] padlock-sha: TFMs don't need to be static
Michal Ludvig [Sun, 16 Jul 2006 22:14:58 +0000 (08:14 +1000)]
[CRYPTO] padlock-sha: TFMs don't need to be static

TFMs are local variables. No need to declare them
static. After all one is enough.

Signed-off-by: Michal Ludvig <michal@logix.cz>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
17 years ago[CRYPTO] padlock: Helper module padlock.ko
Michal Ludvig [Sun, 6 Aug 2006 12:50:30 +0000 (22:50 +1000)]
[CRYPTO] padlock: Helper module padlock.ko

Compile a helper module padlock.ko that will try
to autoload all configured padlock algorithms.

This also provides backward compatibility with
the ancient times before padlock.ko was renamed
to padlock-aes.ko

Signed-off-by: Michal Ludvig <michal@logix.cz>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
17 years ago[CRYPTO] padlock-sha: Make 2 functions static
Adrian Bunk [Sat, 15 Jul 2006 01:31:25 +0000 (11:31 +1000)]
[CRYPTO] padlock-sha: Make 2 functions static

This patch makes two needlessly global functions static.

Signed-off-by: Adrian Bunk <bunk@stusta.de>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
17 years ago[CRYPTO] padlock: Driver for SHA1 / SHA256 algorithms
Michal Ludvig [Wed, 12 Jul 2006 02:29:38 +0000 (12:29 +1000)]
[CRYPTO] padlock: Driver for SHA1 / SHA256 algorithms

Support for SHA1 / SHA256 algorithms in VIA C7 processors.

Signed-off-by: Michal Ludvig <michal@logix.cz>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
17 years ago[CRYPTO] padlock: Update private header file
Michal Ludvig [Sat, 15 Jul 2006 00:23:49 +0000 (10:23 +1000)]
[CRYPTO] padlock: Update private header file

PADLOCK_CRA_PRIORITY is shared between padlock-aes and padlock-sha
so it should be in the header.

On the other hand "struct cword" is only used in padlock-aes.c
so it's unnecessary to have it in padlock.h

Signed-off-by: Michal Ludvig <michal@logix.cz>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
17 years ago[CRYPTO] padlock: Add compatibility alias after rename
Herbert Xu [Sun, 9 Jul 2006 00:35:49 +0000 (10:35 +1000)]
[CRYPTO] padlock: Add compatibility alias after rename

Whenever we rename modules we should add an alias to ensure that existing
users can still locate the new module.

This patch also gets rid of the now unused module function prototypes from
padlock.h.

Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
17 years ago[CRYPTO] padlock: Get rid of padlock-generic.c
Michal Ludvig [Sun, 6 Aug 2006 12:46:20 +0000 (22:46 +1000)]
[CRYPTO] padlock: Get rid of padlock-generic.c

Merge padlock-generic.c into padlock-aes.c and compile
AES as a standalone module. We won't make a monolithic
padlock.ko with all supported algorithms, instead we'll
compile each driver into its own module.

Signed-off-by: Michal Ludvig <michal@logix.cz>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
17 years ago[CRYPTO] api: Add missing accessors for new crypto_alg fields
Michal Ludvig [Sat, 8 Jul 2006 23:02:24 +0000 (09:02 +1000)]
[CRYPTO] api: Add missing accessors for new crypto_alg fields

Add missing accessors for cra_driver_name and cra_priority.

Signed-off-by: Michal Ludvig <michal@logix.cz>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
17 years ago[CRYPTO] sha: Add module aliases for sha1 / sha256
Michal Ludvig [Sat, 8 Jul 2006 22:59:38 +0000 (08:59 +1000)]
[CRYPTO] sha: Add module aliases for sha1 / sha256

Crypto modules should be loadable by their .cra_driver_name, so
we should make MODULE_ALIAS()es with these names. This patch adds
aliases for SHA1 and SHA256 only as that's what we need for
PadLock-SHA driver.

Signed-off-by: Michal Ludvig <michal@logix.cz>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
17 years ago[CRYPTO] api: Added spawns
Herbert Xu [Thu, 21 Sep 2006 01:39:29 +0000 (11:39 +1000)]
[CRYPTO] api: Added spawns

Spawns lock a specific crypto algorithm in place.  They can then be used
with crypto_spawn_tfm to allocate a tfm for that algorithm.  When the base
algorithm of a spawn is deregistered, all its spawns will be automatically
removed.

Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
Signed-off-by: David S. Miller <davem@davemloft.net>
17 years ago[CRYPTO] api: Allow algorithm lookup by type
Herbert Xu [Thu, 21 Sep 2006 01:35:17 +0000 (11:35 +1000)]
[CRYPTO] api: Allow algorithm lookup by type

This patch also adds the infrastructure to pick an algorithm based on
their type.  For example, this allows you to select the encryption
algorithm "aes", instead of any algorithm registered under the name
"aes".  For now this is only accessible internally.  Eventually it
will be made available through crypto_alloc_tfm.

Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
Signed-off-by: David S. Miller <davem@davemloft.net>
17 years ago[CRYPTO] api: Add cryptomgr
Herbert Xu [Thu, 21 Sep 2006 01:31:44 +0000 (11:31 +1000)]
[CRYPTO] api: Add cryptomgr

The cryptomgr module is a simple manager of crypto algorithm instances.
It ensures that parameterised algorithms of the type tmpl(alg) (e.g.,
cbc(aes)) are always created.

This is meant to satisfy the needs for most users.  For more complex
cases such as deeper combinations or multiple parameters, a netlink
module will be created which allows arbitrary expressions to be parsed
in user-space.

Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
Signed-off-by: David S. Miller <davem@davemloft.net>
17 years ago[CRYPTO] api: Added event notification
Herbert Xu [Sun, 6 Aug 2006 11:23:26 +0000 (21:23 +1000)]
[CRYPTO] api: Added event notification

This patch adds a notifier chain for algorithm/template registration events.
This will be used to register compound algorithms such as cbc(aes).  In
future this will also be passed onto user-space through netlink.

Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
Signed-off-by: David S. Miller <davem@davemloft.net>
17 years ago[CRYPTO] api: Add template registration
Herbert Xu [Sun, 6 Aug 2006 11:16:34 +0000 (21:16 +1000)]
[CRYPTO] api: Add template registration

A crypto_template generates a crypto_alg object when given a set of
parameters.  this patch adds the basic data structure fo templates
and code to handle their registration/deregistration.

Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
Signed-off-by: David S. Miller <davem@davemloft.net>
17 years ago[CRYPTO] api: Split out low-level API
Herbert Xu [Mon, 21 Aug 2006 11:08:13 +0000 (21:08 +1000)]
[CRYPTO] api: Split out low-level API

The crypto API is made up of the part facing users such as IPsec and the
low-level part which is used by cryptographic entities such as algorithms.
This patch splits out the latter so that the two APIs are more clearly
delineated.  As a bonus the low-level API can now be modularised if all
algorithms are built as modules.

Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
17 years ago[IPSEC]: Move linux/crypto.h inclusion out of net/xfrm.h
Herbert Xu [Sun, 6 Aug 2006 09:49:12 +0000 (19:49 +1000)]
[IPSEC]: Move linux/crypto.h inclusion out of net/xfrm.h

The header file linux/crypto.h is only needed by a few files so including
it in net/xfrm.h (which is included by half of the networking stack) is a
waste.  This patch moves it out of net/xfrm.h and into the specific header
files that actually need it.

Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
17 years ago[CRYPTO] api: Add crypto_alg reference counting
Herbert Xu [Sun, 6 Aug 2006 10:28:44 +0000 (20:28 +1000)]
[CRYPTO] api: Add crypto_alg reference counting

Up until now we've relied on module reference counting to ensure that the
crypto_alg structures don't disappear from under us.  This was good enough
as long as each crypto_alg came from exactly one module.

However, with parameterised crypto algorithms a crypto_alg object may need
two or more modules to operate.  This means that we need to count the
references to the crypto_alg object directly.

Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
Signed-off-by: David S. Miller <davem@davemloft.net>
17 years ago[CRYPTO] api: Rename crypto_alg_get to crypto_mod_get
Herbert Xu [Sat, 27 May 2006 23:05:24 +0000 (09:05 +1000)]
[CRYPTO] api: Rename crypto_alg_get to crypto_mod_get

The functions crypto_alg_get and crypto_alg_put operates on the crypto
modules rather than the algorithms.  Therefore it makes sense to call
them crypto_mod_get and crypto_alg_put respectively.

This is needed because we need to have real algorithm reference counters
for parameterised algorithms as they can be unregistered from below by
when their parameter algorithms are themselves unregistered.

Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
Signed-off-by: David S. Miller <davem@davemloft.net>
17 years ago[CRYPTO] twofish: x86-64 assembly version
Joachim Fritschi [Tue, 20 Jun 2006 11:12:02 +0000 (21:12 +1000)]
[CRYPTO] twofish: x86-64 assembly version

The patch passed the trycpt tests and automated filesystem tests.
This rewrite resulted in some nice perfomance increase over my last patch.

Short summary of the tcrypt benchmarks:

Twofish Assembler vs. Twofish C (256bit 8kb block CBC)
encrypt: -27% Cycles
decrypt: -23% Cycles

Twofish Assembler vs. AES Assembler (128bit 8kb block CBC)
encrypt: +18%  Cycles
decrypt: +15% Cycles

Twofish Assembler vs. AES Assembler (256bit 8kb block CBC)
encrypt: -9% Cycles
decrypt: -8% Cycles

Full Output:
http://homepages.tu-darmstadt.de/~fritschi/twofish/tcrypt-speed-twofish-c-x86_64.txt
http://homepages.tu-darmstadt.de/~fritschi/twofish/tcrypt-speed-twofish-asm-x86_64.txt
http://homepages.tu-darmstadt.de/~fritschi/twofish/tcrypt-speed-aes-asm-x86_64.txt

Here is another bonnie++ benchmark with encrypted filesystems. Most runs maxed
out the hd. It should give some idea what the module can do for encrypted filesystem
performance even though you can't see the full numbers.

http://homepages.tu-darmstadt.de/~fritschi/twofish/output_20060610_130806_x86_64.html

Signed-off-by: Joachim Fritschi <jfritschi@freenet.de>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
17 years ago[CRYPTO] twofish: i586 assembly version
Joachim Fritschi [Tue, 20 Jun 2006 10:59:16 +0000 (20:59 +1000)]
[CRYPTO] twofish: i586 assembly version

The patch passed the trycpt tests and automated filesystem tests.
This rewrite resulted in some nice perfomance increase over my last patch.

Short summary of the tcrypt benchmarks:

Twofish Assembler vs. Twofish C (256bit 8kb block CBC)
encrypt: -33% Cycles
decrypt: -45% Cycles

Twofish Assembler vs. AES Assembler (128bit 8kb block CBC)
encrypt: +3%  Cycles
decrypt: -22% Cycles

Twofish Assembler vs. AES Assembler (256bit 8kb block CBC)
encrypt: -20% Cycles
decrypt: -36% Cycles

Full Output:
http://homepages.tu-darmstadt.de/~fritschi/twofish/tcrypt-speed-twofish-asm-i586.txt
http://homepages.tu-darmstadt.de/~fritschi/twofish/tcrypt-speed-twofish-c-i586.txt
http://homepages.tu-darmstadt.de/~fritschi/twofish/tcrypt-speed-aes-asm-i586.txt

Here is another bonnie++ benchmark with encrypted filesystems. All runs with
the twofish assembler modules max out the drivespeed. It should give some
idea what the module can do for encrypted filesystem performance even though
you can't see the full numbers.

http://homepages.tu-darmstadt.de/~fritschi/twofish/output_20060611_205432_x86.html

Signed-off-by: Joachim Fritschi <jfritschi@freenet.de>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
17 years ago[CRYPTO] twofish: Fix the priority
Joachim Fritschi [Tue, 20 Jun 2006 10:39:29 +0000 (20:39 +1000)]
[CRYPTO] twofish: Fix the priority

This patch adds a proper driver name and priority to the generic c
implemtation to allow coexistance of c and assembler modules.

Signed-off-by: Joachim Fritschi <jfritschi@freenet.de>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
17 years ago[CRYPTO] twofish: Split out common c code
Joachim Fritschi [Tue, 20 Jun 2006 10:37:23 +0000 (20:37 +1000)]
[CRYPTO] twofish: Split out common c code

This patch splits up the twofish crypto routine into a common part ( key
setup  ) which will be uses by all twofish crypto modules ( generic-c , i586
assembler and x86_64 assembler ) and generic-c part. It also creates a new
header file which will be used by all 3 modules.

This eliminates all code duplication.

Correctness was verified with the tcrypt module and automated test scripts.

Signed-off-by: Joachim Fritschi <jfritschi@freenet.de>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
17 years ago[CRYPTO] api: Fixed crypto_tfm context alignment
Herbert Xu [Mon, 21 Aug 2006 11:03:52 +0000 (21:03 +1000)]
[CRYPTO] api: Fixed crypto_tfm context alignment

Previously the __aligned__ attribute was added to the crypto_tfm context
member to ensure it is alinged correctly on architectures such as arm.
Unfortunately kmalloc does not use the same minimum alignment rules as
gcc so this is useless.

This patch changes it to use kmalloc's minimum alignment.

Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
17 years agoLinux v2.6.18. Arrr! v2.6.18
Linus Torvalds [Wed, 20 Sep 2006 03:42:06 +0000 (20:42 -0700)]
Linux v2.6.18. Arrr!

Ahoy, all land-lubbers, test me out right smartly!

Signed-off-by: Linus Torvalds <torvalds@osdl.org>
17 years agoMerge master.kernel.org:/pub/scm/linux/kernel/git/davem/net-2.6
Linus Torvalds [Wed, 20 Sep 2006 03:36:22 +0000 (20:36 -0700)]
Merge /pub/scm/linux/kernel/git/davem/net-2.6

* master.kernel.org:/pub/scm/linux/kernel/git/davem/net-2.6:
  [IPV4] fib_trie: missing ntohl() when calling fib_semantic_match()
  [NETFILTER]: xt_quota: add missing module aliases
  [ATM]: [he] don't hold the device lock when upcalling

17 years ago[IPV4] fib_trie: missing ntohl() when calling fib_semantic_match()
Al Viro [Tue, 19 Sep 2006 20:42:46 +0000 (13:42 -0700)]
[IPV4] fib_trie: missing ntohl() when calling fib_semantic_match()

fib_trie.c::check_leaf() passes host-endian where fib_semantic_match()
expects (and stores into) net-endian.

Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
Signed-off-by: David S. Miller <davem@davemloft.net>
17 years ago[NETFILTER]: xt_quota: add missing module aliases
Patrick McHardy [Tue, 19 Sep 2006 20:00:57 +0000 (13:00 -0700)]
[NETFILTER]: xt_quota: add missing module aliases

Add missing aliases for ipt_quota and ip6t_quota to make autoload
work.

Signed-off-by: Patrick McHardy <kaber@trash.net>
Signed-off-by: David S. Miller <davem@davemloft.net>
17 years ago[ATM]: [he] don't hold the device lock when upcalling
Chas Williams [Tue, 19 Sep 2006 19:59:11 +0000 (12:59 -0700)]
[ATM]: [he] don't hold the device lock when upcalling

This can create a deadlock/lock ordering problem with other layers
that want to use the transmit (or other) path of the card at that
time.

Signed-off-by: Chas Williams <chas@cmf.nrl.navy.mil>
Signed-off-by: David S. Miller <davem@davemloft.net>
17 years agoMerge branch 'fixes' of git://git.linux-nfs.org/pub/linux/nfs-2.6
Linus Torvalds [Tue, 19 Sep 2006 16:46:53 +0000 (09:46 -0700)]
Merge branch 'fixes' of git://git.linux-nfs.org/pub/linux/nfs-2.6

* 'fixes' of git://git.linux-nfs.org/pub/linux/nfs-2.6:
  NFS: Fix nfs_page use after free issues in fs/nfs/write.c
  NFSv4: Fix incorrect semaphore release in _nfs4_do_open()
  NFS: Fix Oopsable condition in nfs_readpage_sync()

17 years agoNFS: Fix nfs_page use after free issues in fs/nfs/write.c
Trond Myklebust [Tue, 19 Sep 2006 03:20:35 +0000 (23:20 -0400)]
NFS: Fix nfs_page use after free issues in fs/nfs/write.c

Signed-off-by: Trond Myklebust <Trond.Myklebust@netapp.com>
17 years agoNFSv4: Fix incorrect semaphore release in _nfs4_do_open()
Trond Myklebust [Fri, 15 Sep 2006 12:11:51 +0000 (08:11 -0400)]
NFSv4: Fix incorrect semaphore release in _nfs4_do_open()

Signed-off-by: Trond Myklebust <Trond.Myklebust@netapp.com>
17 years agoNFS: Fix Oopsable condition in nfs_readpage_sync()
Trond Myklebust [Fri, 15 Sep 2006 20:03:45 +0000 (16:03 -0400)]
NFS: Fix Oopsable condition in nfs_readpage_sync()

Signed-off-by: Trond Myklebust <Trond.Myklebust@netapp.com>
17 years agoRevert mmiocfg heuristics and blacklist changes
Linus Torvalds [Tue, 19 Sep 2006 15:15:22 +0000 (08:15 -0700)]
Revert mmiocfg heuristics and blacklist changes

This reverts commits 11012d419cfc0e0f78ca356aca03674217910124 and
40dd2d20f220eda1cd0da8ea3f0f9db8971ba237, which allowed us to use the
MMIO accesses for PCI config cycles even without the area being marked
reserved in the e820 memory tables.

Those changes were needed for EFI-environment Intel macs, but broke some
newer Intel 965 boards, so for now it's better to revert to our old
2.6.17 behaviour and at least avoid introducing any new breakage.

Andi Kleen has a set of patches that work with both EFI and the broken
Intel 965 boards, which will be applied once they get wider testing.

Cc: Arjan van de Ven <arjan@infradead.org>
Cc: Edgar Hucek <hostmaster@ed-soft.at>
Cc: Andi Kleen <ak@suse.de>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
17 years agoMerge git://git.infradead.org/mtd-2.6
Linus Torvalds [Tue, 19 Sep 2006 15:01:58 +0000 (08:01 -0700)]
Merge git://git.infradead.org/mtd-2.6

* git://git.infradead.org/mtd-2.6:
  [MTD] Use SEEK_{SET,CUR,END} instead of hardcoded values in mtdchar lseek()
  MTD: Fix bug in fixup_convert_atmel_pri
  [JFFS2][SUMMARY] Fix a summary collecting bug.
  [PATCH] [MTD] DEVICES: Fill more device IDs in the structure of m25p80
  MTD: Add lock/unlock operations for Atmel AT49BV6416
  MTD: Convert Atmel PRI information to AMD format
  fs/jffs2/xattr.c: remove dead code
  [PATCH] [MTD] Maps: Add dependency on alternate probe methods to physmap
  [PATCH] MTD: Add Macronix MX29F040 to JEDEC
  [MTD] Fixes of performance and stability issues in CFI driver.
  block2mtd.c: Make kernel boot command line arguments work (try 4)
  [MTD NAND] Fix lookup error in nand_get_flash_type()
  remove #error on !PCI from pmc551.c
  MTD: [NAND] Fix the sharpsl driver after breakage from a core conversion
  [MTD] NAND: OOB buffer offset fixups
  make fs/jffs2/nodelist.c:jffs2_obsolete_node_frag() static
  [PATCH] [MTD] NAND: fix dead URL in Kconfig

17 years ago[PATCH] EXT2: Remove superblock lock contention in ext2_statfs
Dave Kleikamp [Tue, 19 Sep 2006 03:12:33 +0000 (20:12 -0700)]
[PATCH] EXT2: Remove superblock lock contention in ext2_statfs

Fix a performance degradation introduced in 2.6.17.  (30% degradation
running dbench with 16 threads)

Commit 21730eed11de42f22afcbd43f450a1872a0b5ea1, which claims to make
EXT2_DEBUG work again, moves the taking of the kernel lock out of
debug-only code in ext2_count_free_inodes and ext2_count_free_blocks and
into ext2_statfs.

The same problem was fixed in ext3 by removing the lock completely (commit
5b11687924e40790deb0d5f959247ade82196665)

Signed-off-by: Dave Kleikamp <shaggy@austin.ibm.com>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
17 years ago[PATCH] headers_check: Clean up asm-parisc/page.h for user headers
Matthew Wilcox [Tue, 19 Sep 2006 03:12:27 +0000 (20:12 -0700)]
[PATCH] headers_check: Clean up asm-parisc/page.h for user headers

Remove definitions of PAGE_* from the user view
Delete unnecessary comments referring to the size of pages
Only include <asm-generic> if we're in __KERNEL__

Signed-off-by: David Woodhouse <dwmw2@infradead.org>
Signed-off-by: Matthew Wilcox <matthew@wil.cx>
Signed-off-by: Kyle McMartin <kyle@parisc-linux.org>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
17 years ago[PATCH] Fix 'make headers_check' on ia64
Geert Uytterhoeven [Tue, 19 Sep 2006 03:12:23 +0000 (20:12 -0700)]
[PATCH] Fix 'make headers_check' on ia64

Fix 'make headers_check' on m68k

Signed-off-by: Geert Uytterhoeven <geert@linux-m68k.org>
Cc: Roman Zippel <zippel@linux-m68k.org>
Cc: David Woodhouse <dwmw2@infradead.org>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
17 years ago[PATCH] Add headers_check' target to output of 'make help'
David Woodhouse [Tue, 19 Sep 2006 03:12:19 +0000 (20:12 -0700)]
[PATCH] Add headers_check' target to output of 'make help'

Signed-off-by: David Woodhouse <dwmw2@infradead.org>
Cc: Sam Ravnborg <sam@ravnborg.org>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
17 years ago[PATCH] genirq core: fix handle_level_irq()
Ingo Molnar [Tue, 19 Sep 2006 09:14:34 +0000 (11:14 +0200)]
[PATCH] genirq core: fix handle_level_irq()

while porting the -rt tree to 2.6.18-rc7 i noticed the following
screaming-IRQ scenario on an SMP system:

 2274  0Dn.:1 0.001ms: do_IRQ+0xc/0x103  <= (ret_from_intr+0x0/0xf)
 2274  0Dn.:1 0.010ms: do_IRQ+0xc/0x103  <= (ret_from_intr+0x0/0xf)
 2274  0Dn.:1 0.020ms: do_IRQ+0xc/0x103  <= (ret_from_intr+0x0/0xf)
 2274  0Dn.:1 0.029ms: do_IRQ+0xc/0x103  <= (ret_from_intr+0x0/0xf)
 2274  0Dn.:1 0.039ms: do_IRQ+0xc/0x103  <= (ret_from_intr+0x0/0xf)
 2274  0Dn.:1 0.048ms: do_IRQ+0xc/0x103  <= (ret_from_intr+0x0/0xf)
 2274  0Dn.:1 0.058ms: do_IRQ+0xc/0x103  <= (ret_from_intr+0x0/0xf)
 2274  0Dn.:1 0.068ms: do_IRQ+0xc/0x103  <= (ret_from_intr+0x0/0xf)
 2274  0Dn.:1 0.077ms: do_IRQ+0xc/0x103  <= (ret_from_intr+0x0/0xf)
 2274  0Dn.:1 0.087ms: do_IRQ+0xc/0x103  <= (ret_from_intr+0x0/0xf)
 2274  0Dn.:1 0.097ms: do_IRQ+0xc/0x103  <= (ret_from_intr+0x0/0xf)

as it turns out, the bug is caused by handle_level_irq(), which if it
races with another CPU already handling this IRQ, it _unmasks_ the IRQ
line on the way out. This is not how 2.6.17 works, and we introduced
this bug in one of the early genirq cleanups right before it went into
-mm. (the bug was not in the genirq patchset for a long time, and we
didnt notice the bug due to the lack of -rt rebase to the new genirq
code. -rt, and hardirq-preemption in particular opens up such races much
wider than anything else.)

Signed-off-by: Ingo Molnar <mingo@elte.hu>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Acked-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
17 years agox86: save/restore eflags in context switch
Linus Torvalds [Mon, 18 Sep 2006 23:20:40 +0000 (16:20 -0700)]
x86: save/restore eflags in context switch

(And reset it on new thread creation)

It turns out that eflags is important to save and restore not just
because of iopl, but due to the magic bits like the NT bit, which we
don't want leaking between different threads.

Tested-by: Mike Galbraith <efault@gmx.de>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
17 years agoMerge master.kernel.org:/pub/scm/linux/kernel/git/davem/net-2.6
Linus Torvalds [Mon, 18 Sep 2006 23:16:05 +0000 (16:16 -0700)]
Merge /pub/scm/linux/kernel/git/davem/net-2.6

* master.kernel.org:/pub/scm/linux/kernel/git/davem/net-2.6:
  [ATM] CLIP: Do not refer freed skbuff in clip_mkip().
  [NET]: Drop tx lock in dev_watchdog_up
  [PACKET]: Don't truncate non-linear skbs with mmaped IO
  [NET]: Mark frame diverter for future removal.
  [NETFILTER]: Add secmark headers to header-y
  [ATM]: linux-atm-general mailing list is subscribers only
  [ATM]: [he] when transmit fails, unmap the dma regions
  [TCP] tcp-lp: update information to MAINTAINERS
  [TCP] tcp-lp: bug fix for oops in 2.6.18-rc6
  [BRIDGE]: random extra bytes on STP TCN packet
  [IPV6]: Accept -1 for IPV6_TCLASS
  [IPV6]: Fix tclass setting for raw sockets.
  [IPVS]: remove the debug option go ip_vs_ftp
  [IPVS]: Make sure ip_vs_ftp ports are valid
  [IPVS]: auto-help for ip_vs_ftp
  [IPVS]: Document the ports option to ip_vs_ftp in kernel-parameters.txt
  [TCP]: Turn ABC off.
  [NEIGH]: neigh_table_clear() doesn't free stats

17 years agoMerge master.kernel.org:/pub/scm/linux/kernel/git/davem/sparc-2.6
Linus Torvalds [Mon, 18 Sep 2006 23:15:40 +0000 (16:15 -0700)]
Merge /pub/scm/linux/kernel/git/davem/sparc-2.6

* master.kernel.org:/pub/scm/linux/kernel/git/davem/sparc-2.6:
  [SPARC]: Fix regression in sys_getdomainname()
  [OPENPROMIO]: Handle current_node being NULL correctly.

17 years agoMerge master.kernel.org:/home/rmk/linux-2.6-arm
Linus Torvalds [Mon, 18 Sep 2006 23:15:16 +0000 (16:15 -0700)]
Merge master.kernel.org:/home/rmk/linux-2.6-arm

* master.kernel.org:/home/rmk/linux-2.6-arm:
  [ARM] 3815/1: headers_install support for ARM
  [ARM] 3794/1: S3C24XX: do not defined set_irq_wake when no CONFIG_PM
  [ARM] 3793/1: S3C2412: fix wrong serial info struct
  [ARM] 3780/1: Fix iop321 cpuid
  [ARM] 3786/1: pnx4008: update defconfig
  [ARM] 3785/1: S3C2412: Fix idle code as default uses wrong clocks
  [ARM] 3784/1: S3C2413: fix config for MACH_S3C2413/MACH_SMDK2413

17 years ago[ARM] 3815/1: headers_install support for ARM
Ralph Siemsen [Mon, 18 Sep 2006 15:28:50 +0000 (16:28 +0100)]
[ARM] 3815/1: headers_install support for ARM

Move kernel-only #includes into #ifdef __KERNEL__, so that
headers_install target can be used on ARM.

Signed-off-by: Ralph Siemsen <ralphs@netwinder.org>
Signed-off-by: Russell King <rmk+kernel@arm.linux.org.uk>
17 years ago[SPARC]: Fix regression in sys_getdomainname()
Andy Walker [Mon, 18 Sep 2006 14:11:36 +0000 (07:11 -0700)]
[SPARC]: Fix regression in sys_getdomainname()

This patch corrects the buffer length checking in the
sys_getdomainname() implementation for sparc/sparc64.

Signed-off-by: Andy Walker <andy@puszczka.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
17 years ago[ATM] CLIP: Do not refer freed skbuff in clip_mkip().
YOSHIFUJI Hideaki [Mon, 18 Sep 2006 13:37:58 +0000 (06:37 -0700)]
[ATM] CLIP: Do not refer freed skbuff in clip_mkip().

In clip_mkip(), skb->dev is dereferenced after clip_push(),
which frees up skb.

Advisory: AD_LAB-06009 (<adlab@venustech.com.cn>).

Signed-off-by: YOSHIFUJI Hideaki <yoshfuji@linux-ipv6.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
17 years ago[ARM] 3794/1: S3C24XX: do not defined set_irq_wake when no CONFIG_PM
Ben Dooks [Mon, 18 Sep 2006 12:30:20 +0000 (13:30 +0100)]
[ARM] 3794/1: S3C24XX: do not defined set_irq_wake when no CONFIG_PM

Patch from Ben Dooks

Do not define set_irq_wake as a real function if
the CONFIG_PM option is not set.

Fixes bug reported by Thomas Gleixner.

Signed-off-by: Ben Dooks <ben-linux@fluff.org>
Signed-off-by: Russell King <rmk+kernel@arm.linux.org.uk>
17 years ago[ARM] 3793/1: S3C2412: fix wrong serial info struct
Ben Dooks [Mon, 18 Sep 2006 12:30:17 +0000 (13:30 +0100)]
[ARM] 3793/1: S3C2412: fix wrong serial info struct

Patch from Ben Dooks

The S3C2440 serial info struct is being passed
through the S3C2412 serial info struct probe
routine.

Signed-off-by: Ben Dooks <ben-linux@fluff.org>
Signed-off-by: Thomas Glexiner <tglx@linutronix.de>
Signed-off-by: Russell King <rmk+kernel@arm.linux.org.uk>
17 years ago[OPENPROMIO]: Handle current_node being NULL correctly.
David S. Miller [Mon, 18 Sep 2006 08:47:13 +0000 (01:47 -0700)]
[OPENPROMIO]: Handle current_node being NULL correctly.

If the user tries to traverse to the next node of the
last node, we get NULL in current_node and a zero phandle
returned.  That's fine, but if the user tries to obtain
properties in that state, we try to dereference a NULL
pointer in the downcall to the of_*() routines.

So protect against that.

Signed-off-by: David S. Miller <davem@davemloft.net>
17 years ago[NET]: Drop tx lock in dev_watchdog_up
Herbert Xu [Mon, 18 Sep 2006 07:22:30 +0000 (00:22 -0700)]
[NET]: Drop tx lock in dev_watchdog_up

Fix lockdep warning with GRE, iptables and Speedtouch ADSL, PPP over ATM.

On Sat, Sep 02, 2006 at 08:39:28PM +0000, Krzysztof Halasa wrote:
>
> =======================================================
> [ INFO: possible circular locking dependency detected ]
> -------------------------------------------------------
> swapper/0 is trying to acquire lock:
>  (&dev->queue_lock){-+..}, at: [<c02c8c46>] dev_queue_xmit+0x56/0x290
>
> but task is already holding lock:
>  (&dev->_xmit_lock){-+..}, at: [<c02c8e14>] dev_queue_xmit+0x224/0x290
>
> which lock already depends on the new lock.

This turns out to be a genuine bug.  The queue lock and xmit lock are
intentionally taken out of order.  Two things are supposed to prevent
dead-locks from occuring:

1) When we hold the queue_lock we're supposed to only do try_lock on the
tx_lock.

2) We always drop the queue_lock after taking the tx_lock and before doing
anything else.

>
> the existing dependency chain (in reverse order) is:
>
> -> #1 (&dev->_xmit_lock){-+..}:
>        [<c012e7b6>] lock_acquire+0x76/0xa0
>        [<c0336241>] _spin_lock_bh+0x31/0x40
>        [<c02d25a9>] dev_activate+0x69/0x120

This path obviously breaks assumption 1) and therefore can lead to ABBA
dead-locks.

I've looked at the history and there seems to be no reason for the lock
to be held at all in dev_watchdog_up.  The lock appeared in day one and
even there it was unnecessary.  In fact, people added __dev_watchdog_up
precisely in order to get around the tx lock there.

The function dev_watchdog_up is already serialised by rtnl_lock since
its only caller dev_activate is always called under it.

So here is a simple patch to remove the tx lock from dev_watchdog_up.
In 2.6.19 we can eliminate the unnecessary __dev_watchdog_up and
replace it with dev_watchdog_up.

Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
Signed-off-by: David S. Miller <davem@davemloft.net>
17 years ago[PACKET]: Don't truncate non-linear skbs with mmaped IO
Patrick McHardy [Mon, 18 Sep 2006 06:59:57 +0000 (23:59 -0700)]
[PACKET]: Don't truncate non-linear skbs with mmaped IO

Non-linear skbs are truncated to their linear part with mmaped IO.
Fix by using skb_copy_bits instead of memcpy.

Signed-off-by: Patrick McHardy <kaber@trash.net>
Acked-by: Alexey Kuznetsov <kuznet@ms2.inr.ac.ru>
Signed-off-by: David S. Miller <davem@davemloft.net>
17 years ago[NET]: Mark frame diverter for future removal.
Stephen Hemminger [Thu, 14 Sep 2006 04:13:54 +0000 (21:13 -0700)]
[NET]: Mark frame diverter for future removal.

The code for frame diverter is unmaintained and has bitrotted.
The number of users is very small and the code has lots of problems.
If anyone is using it, they maybe exposing themselves to bad packet attacks.

Signed-off-by: Stephen Hemminger <shemminger@osdl.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
17 years ago[NETFILTER]: Add secmark headers to header-y
James Morris [Thu, 14 Sep 2006 04:04:55 +0000 (21:04 -0700)]
[NETFILTER]: Add secmark headers to header-y

This patch includes xt_SECMARK.h and xt_CONNSECMARK.h to the kernel
headers which are exported via 'make headers_install'.  This is needed to
allow userland code to be built correctly with these features.

Please apply, and consider for inclusion with 2.6.18 as a bugfix.

Signed-off-by: James Morris <jmorris@redhat.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
17 years ago[ATM]: linux-atm-general mailing list is subscribers only
Roland Dreier [Thu, 14 Sep 2006 03:39:33 +0000 (20:39 -0700)]
[ATM]: linux-atm-general mailing list is subscribers only

As the automated reply I got to my last ATM patch shows, the
linux-atm-general mailing list is subscribers-only.

Signed-off-by: Roland Dreier <roland@digitalvampire.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
17 years ago[ATM]: [he] when transmit fails, unmap the dma regions
Chas Williams [Thu, 14 Sep 2006 03:33:40 +0000 (20:33 -0700)]
[ATM]: [he] when transmit fails, unmap the dma regions

Signed-off-by: Chas Williams <chas@cmf.nrl.navy.mil>
Signed-off-by: David S. Miller <davem@davemloft.net>
17 years ago[TCP] tcp-lp: update information to MAINTAINERS
Wong Hoi Sing Edison [Thu, 14 Sep 2006 03:31:13 +0000 (20:31 -0700)]
[TCP] tcp-lp: update information to MAINTAINERS

Signed-off-by: Wong Hoi Sing Edison <hswong3i@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
17 years ago[TCP] tcp-lp: bug fix for oops in 2.6.18-rc6
Wong Hoi Sing Edison [Thu, 14 Sep 2006 03:30:30 +0000 (20:30 -0700)]
[TCP] tcp-lp: bug fix for oops in 2.6.18-rc6

Sorry that the patch submited yesterday still contain a small bug.
This version have already been test for hours with BT connections. The
oops is now difficult to reproduce.

Signed-off-by: Wong Hoi Sing Edison <hswong3i@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
17 years ago[BRIDGE]: random extra bytes on STP TCN packet
Stephen Hemminger [Thu, 14 Sep 2006 03:12:40 +0000 (20:12 -0700)]
[BRIDGE]: random extra bytes on STP TCN packet

We seem to send 3 extra bytes in a TCN, which will be whatever happens
to be on the stack. Thanks to Aji_Srinivas@emc.com for seeing.

Signed-off-by: Stephen Hemminger <shemminger@osdl.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
17 years ago[IPV6]: Accept -1 for IPV6_TCLASS
Remi Denis-Courmont [Thu, 14 Sep 2006 03:08:07 +0000 (20:08 -0700)]
[IPV6]: Accept -1 for IPV6_TCLASS

This patch should add support for -1 as "default" IPv6 traffic class,
as specified in IETF RFC3542 ยง6.5. Within the kernel, it seems tclass
< 0 is already handled, but setsockopt, getsockopt and recvmsg calls
won't accept it from userland.

Signed-off-by: Remi Denis-Courmont <rdenis@simphalempin.com>
Acked-by: YOSHIFUJI Hideaki <yoshfuji@linux-ipv6.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
17 years ago[IPV6]: Fix tclass setting for raw sockets.
YOSHIFUJI Hideaki [Thu, 14 Sep 2006 03:01:28 +0000 (20:01 -0700)]
[IPV6]: Fix tclass setting for raw sockets.

np->cork.tclass is used only in cork'ed context.
Otherwise, np->tclass should be used.

Bug#7096 reported by Remi Denis-Courmont <rdenis@simphalempin.com>.

Signed-off-by: YOSHIFUJI Hideaki <yoshfuji@linux-ipv6.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
17 years ago[IPVS]: remove the debug option go ip_vs_ftp
Simon Horman [Thu, 14 Sep 2006 02:59:23 +0000 (19:59 -0700)]
[IPVS]: remove the debug option go ip_vs_ftp

This patch makes the debuging behaviour of this code more consistent
with the rest of IPVS.

Signed-Off-By: Simon Horman <horms@verge.net.au>
Signed-off-by: David S. Miller <davem@davemloft.net>
17 years ago[IPVS]: Make sure ip_vs_ftp ports are valid
Simon Horman [Thu, 14 Sep 2006 02:58:44 +0000 (19:58 -0700)]
[IPVS]: Make sure ip_vs_ftp ports are valid

I'm not entirely sure what happens in the case of a valid port,
at best it'll be silently ignored. This patch ignores them a little
more verbosely.

Signed-Off-By: Simon Horman <horms@verge.net.au>
Signed-off-by: David S. Miller <davem@davemloft.net>
17 years ago[IPVS]: auto-help for ip_vs_ftp
Simon Horman [Thu, 14 Sep 2006 02:57:54 +0000 (19:57 -0700)]
[IPVS]: auto-help for ip_vs_ftp

Fill in a help message for the ports option to ip_vs_ftp

Signed-Off-By: Simon Horman <horms@verge.net.au>
Signed-off-by: David S. Miller <davem@davemloft.net>
17 years ago[IPVS]: Document the ports option to ip_vs_ftp in kernel-parameters.txt
Simon Horman [Thu, 14 Sep 2006 02:57:18 +0000 (19:57 -0700)]
[IPVS]: Document the ports option to ip_vs_ftp in kernel-parameters.txt

I'm not sure if documenting this here is appropriate, but
if it is, here is some text to put there.

Signed-Off-By: Simon Horman <horms@verge.net.au>
Signed-off-by: David S. Miller <davem@davemloft.net>
17 years ago[TCP]: Turn ABC off.
Stephen Hemminger [Thu, 14 Sep 2006 02:51:02 +0000 (19:51 -0700)]
[TCP]: Turn ABC off.

Turn Appropriate Byte Count off by default because it unfairly
penalizes applications that do small writes.  Add better documentation
to describe what it is so users will understand why they might want to
turn it on.

Signed-off-by: Stephen Hemminger <shemminger@osdl.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
17 years ago[NEIGH]: neigh_table_clear() doesn't free stats
Kirill Korotaev [Fri, 1 Sep 2006 08:34:10 +0000 (01:34 -0700)]
[NEIGH]: neigh_table_clear() doesn't free stats

neigh_table_clear() doesn't free tbl->stats.
Found by Alexey Kuznetsov. Though Alexey considers this
leak minor for mainstream, I still believe that cleanup
code should not forget to free some of the resources :)

At least, this is critical for OpenVZ with virtualized
neighbour tables.

Signed-Off-By: Kirill Korotaev <dev@openvz.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
17 years ago[MTD] Use SEEK_{SET,CUR,END} instead of hardcoded values in mtdchar lseek()
Josef 'Jeff' Sipek [Sun, 17 Sep 2006 01:09:29 +0000 (21:09 -0400)]
[MTD] Use SEEK_{SET,CUR,END} instead of hardcoded values in mtdchar lseek()

Signed-off-by: Josef 'Jeff' Sipek <jeffpc@josefsipek.net>
Signed-off-by: David Woodhouse <dwmw2@infradead.org>
17 years agoMerge branch 'master' into upstream-fixes
Jeff Garzik [Sun, 17 Sep 2006 05:05:43 +0000 (01:05 -0400)]
Merge branch 'master' into upstream-fixes

17 years ago[PATCH] s390: qeth driver fixes [6/6]
Frank Pavlic [Fri, 15 Sep 2006 14:27:02 +0000 (16:27 +0200)]
[PATCH] s390: qeth driver fixes [6/6]

[PATCH 9/9] s390: qeth driver fixes [6/6]

From: Frank Pavlic <fpavlic@de.ibm.com>
- Hipersockets has no IPV6 support, thus prevent issueing
  SETRTG_IPV6 control commands on Hipersockets devices.
- fixed error handling in qeth_sysfs_(un)register

Signed-off-by: Frank Pavlic <fpavlic@de.ibm.com>
Signed-off-by: Jeff Garzik <jeff@garzik.org>
17 years ago[PATCH] s390: qeth driver fixes [5/6]
Frank Pavlic [Fri, 15 Sep 2006 14:26:52 +0000 (16:26 +0200)]
[PATCH] s390: qeth driver fixes [5/6]

[PATCH 8/9] s390: qeth driver fixes [5/6]

From: Frank Pavlic <fpavlic@de.ibm.com>
fix kernel panic in qdio queue handling.
qeth_qdio_clear_card() could be invoked by 2 CPUs
simultaneously (for example reboot event and recovery).

Signed-off-by: Frank Pavlic <fpavlic@de.ibm.com>
Signed-off-by: Jeff Garzik <jeff@garzik.org>
17 years ago[PATCH] s390: qeth driver fixes [4/6]
Frank Pavlic [Fri, 15 Sep 2006 14:26:34 +0000 (16:26 +0200)]
[PATCH] s390: qeth driver fixes [4/6]

[PATCH 7/9] s390: qeth driver fixes [4/6]

From: Frank Pavlic <fpavlic@de.ibm.com>
- fix kernel crash due to race,
  set card->state to SOFTSETUP after
  card and card->dev are initialized properly.
- remove CONFIG_QETH_PERF_STATS, use sysfs attribute instead,
  as we want to have the ability to turn on/off the
  statistics at runtime.

Signed-off-by: Frank Pavlic <fpavlic@de.ibm.com>
Signed-off-by: Jeff Garzik <jeff@garzik.org>
17 years ago[PATCH] s390: qeth driver fixes [3/6]
Frank Pavlic [Fri, 15 Sep 2006 14:26:19 +0000 (16:26 +0200)]
[PATCH] s390: qeth driver fixes [3/6]

[PATCH 6/9] s390: qeth driver fixes [3/6]

From: Frank Pavlic <fpavlic@de.ibm.com>
        fixed kernel panic caused by qeth driver:
        Using a bonding device qeth driver will realloc
        headroom for every skb coming from the bond device.
        Once this happens qeth frees the original skb and
        set the skb pointer to the new realloced skb.
        Under heavy transmit workload (e.g.UDP streams) through bond
        network device the qdio output queue might get full.
        In this case we return with EBUSY from qeth_send_packet.
        Returning to qeth_hard_start_xmit routine
        the skb address on the stack still points to the old address,
        which has been freed before.
        Returning from qeth_hard_start_xmit with EBUSY results in
        requeuing the skb. In this case it corrupts the qdisc queue
        and results in kernel panic.

Signed-off-by: Frank Pavlic <fpavlic@de.ibm.com>
Signed-off-by: Jeff Garzik <jeff@garzik.org>