Reduce the initial permissions on new keys to grant the possessor everything,
view permission only to the user (so the keys can be seen in /proc/keys) and
nothing else.

This gives the creator a chance to adjust the permissions mask before other
processes can access the new key or create a link to it.

To aid with this, keyring_alloc() now takes a permission argument rather than
setting the permissions itself.

The following permissions are now set:

 (1) The user and user-session keyrings grant the user that owns them full
     permissions and grant a possessor everything bar SETATTR.

 (2) The process and thread keyrings grant the possessor full permissions but
     only grant the user VIEW.  This permits the user to see them in
     /proc/keys, but not to do anything with them.

 (3) Anonymous session keyrings grant the possessor full permissions, but only
     grant the user VIEW and READ.  This means that the user can see them in
     /proc/keys and can list them, but nothing else.  Possibly READ shouldn't
     be provided either.

 (4) Named session keyrings grant everything an anonymous session keyring does,
     plus they grant the user LINK permission.  The whole point of named
     session keyrings is that others can also subscribe to them.  Possibly this
     should be a separate permission to LINK.

 (5) The temporary session keyring created by call_sbin_request_key() gets the
     same permissions as an anonymous session keyring.

 (6) Keys created by add_key() get VIEW, SEARCH, LINK and SETATTR for the
     possessor, plus READ and/or WRITE if the key type supports them.  The used
     only gets VIEW now.

 (7) Keys created by request_key() now get the same as those created by
     add_key().
Reported-by: NLennart Poettering <lennart@poettering.net>
Reported-by: NStef Walter <stefw@redhat.com>
Signed-off-by: NDavid Howells <dhowells@redhat.com>

Make the session keyring per-thread rather than per-process, but still
inherited from the parent thread to solve a problem with PAM and gdm.

The problem is that join_session_keyring() will reject attempts to change the
session keyring of a multithreaded program but gdm is now multithreaded before
it gets to the point of starting PAM and running pam_keyinit to create the
session keyring.  See:

	https://bugs.freedesktop.org/show_bug.cgi?id=49211

The reason that join_session_keyring() will only change the session keyring
under a single-threaded environment is that it's hard to alter the other
thread's credentials to effect the change in a multi-threaded program.  The
problems are such as:

 (1) How to prevent two threads both running join_session_keyring() from
     racing.

 (2) Another thread's credentials may not be modified directly by this process.

 (3) The number of threads is uncertain whilst we're not holding the
     appropriate spinlock, making preallocation slightly tricky.

 (4) We could use TIF_NOTIFY_RESUME and key_replace_session_keyring() to get
     another thread to replace its keyring, but that means preallocating for
     each thread.

A reasonable way around this is to make the session keyring per-thread rather
than per-process and just document that if you want a common session keyring,
you must get it before you spawn any threads - which is the current situation
anyway.

Whilst we're at it, we can the process keyring behave in the same way.  This
means we can clean up some of the ickyness in the creds code.

Basically, after this patch, the session, process and thread keyrings are about
inheritance rules only and not about sharing changes of keyring.
Reported-by: NMantas M. <grawity@gmail.com>
Signed-off-by: NDavid Howells <dhowells@redhat.com>
Tested-by: NRay Strode <rstrode@redhat.com>

On an error iov may still have been reallocated and need freeing
Signed-off-by: NAlan Cox <alan@linux.intel.com>
Signed-off-by: NDavid Howells <dhowells@redhat.com>

We set ret to NULL then test it. Remove the bogus test
Signed-off-by: NAlan Cox <alan@linux.intel.com>
Signed-off-by: NDavid Howells <dhowells@redhat.com>

Move the tpm_get_random api from the trusted keys code into the TPM
device driver itself so that other callers can make use of it. Also,
change the api slightly so that the number of bytes read is returned in
the call, since the TPM command can potentially return fewer bytes than
requested.
Acked-by: NDavid Safford <safford@linux.vnet.ibm.com>
Reviewed-by: NH. Peter Anvin <hpa@linux.intel.com>
Signed-off-by: NKent Yoder <key@linux.vnet.ibm.com>

Signed-off-by: NAl Viro <viro@zeniv.linux.org.uk>

task_work and rcu_head are identical now; merge them (calling the result
struct callback_head, rcu_head #define'd to it), kill separate allocation
in security/keys since we can just use cred->rcu now.
Signed-off-by: NAl Viro <viro@zeniv.linux.org.uk>

get rid of the only user of ->data; this is _not_ the final variant - in the
end we'll have task_work and rcu_head identical and just use cred->rcu,
at which point the separate allocation will be gone completely.
Signed-off-by: NAl Viro <viro@zeniv.linux.org.uk>

A cleanup of rw_copy_check_uvector and compat_rw_copy_check_uvector after
changes made to support CMA in an earlier patch.

Rather than having an additional check_access parameter to these
functions, the first paramater type is overloaded to allow the caller to
specify CHECK_IOVEC_ONLY which means check that the contents of the iovec
are valid, but do not check the memory that they point to.  This is used
by process_vm_readv/writev where we need to validate that a iovec passed
to the syscall is valid but do not want to check the memory that it points
to at this point because it refers to an address space in another process.
Signed-off-by: NChris Yeoh <yeohc@au1.ibm.com>
Reviewed-by: NOleg Nesterov <oleg@redhat.com>
Signed-off-by: NAndrew Morton <akpm@linux-foundation.org>
Signed-off-by: NLinus Torvalds <torvalds@linux-foundation.org>

Both kernel/sys.c && security/keys/request_key.c where inlining the exact
same code as call_usermodehelper_fns(); So simply convert these sites to
directly use call_usermodehelper_fns().
Signed-off-by: NBoaz Harrosh <bharrosh@panasas.com>
Cc: Oleg Nesterov <oleg@redhat.com>
Cc: Tetsuo Handa <penguin-kernel@I-love.SAKURA.ne.jp>
Cc: Ingo Molnar <mingo@elte.hu>
Cc: Peter Zijlstra <a.p.zijlstra@chello.nl>
Signed-off-by: NAndrew Morton <akpm@linux-foundation.org>
Signed-off-by: NLinus Torvalds <torvalds@linux-foundation.org>

This allocation may be large.  The code is probing to see if it will
succeed and if not, it falls back to vmalloc().  We should suppress any
page-allocation failure messages when the fallback happens.
Reported-by: NDave Jones <davej@redhat.com>
Acked-by: NDavid Howells <dhowells@redhat.com>
Cc: James Morris <jmorris@namei.org>
Signed-off-by: NAndrew Morton <akpm@linux-foundation.org>
Signed-off-by: NLinus Torvalds <torvalds@linux-foundation.org>

Fix some sparse warnings in the keyrings code:

 (1) compat_keyctl_instantiate_key_iov() should be static.

 (2) There were a couple of places where a pointer was being compared against
     integer 0 rather than NULL.

 (3) keyctl_instantiate_key_common() should not take a __user-labelled iovec
     pointer as the caller must have copied the iovec to kernel space.

 (4) __key_link_begin() takes and __key_link_end() releases
     keyring_serialise_link_sem under some circumstances and so this should be
     declared.

     Note that adding __acquires() and __releases() for this doesn't help cure
     the warnings messages - something only commenting out both helps.
Signed-off-by: NDavid Howells <dhowells@redhat.com>
Signed-off-by: NJames Morris <james.l.morris@oracle.com>

Change keyctl_session_to_parent() to use task_work_add() and move
key_replace_session_keyring() logic into task_work->func().

Note that we do task_work_cancel() before task_work_add() to ensure that
only one work can be pending at any time.  This is important, we must not
allow user-space to abuse the parent's ->task_works list.

The callback, replace_session_keyring(), checks PF_EXITING.  I guess this
is not really needed but looks better.

As a side effect, this fixes the (unlikely) race.  The callers of
key_replace_session_keyring() and keyctl_session_to_parent() lack the
necessary barriers, the parent can miss the request.

Now we can remove task_struct->replacement_session_keyring and related
code.
Signed-off-by: NOleg Nesterov <oleg@redhat.com>
Acked-by: NDavid Howells <dhowells@redhat.com>
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: Richard Kuo <rkuo@codeaurora.org>
Cc: Linus Torvalds <torvalds@linux-foundation.org>
Cc: Alexander Gordeev <agordeev@redhat.com>
Cc: Chris Zankel <chris@zankel.net>
Cc: David Smith <dsmith@redhat.com>
Cc: "Frank Ch. Eigler" <fche@redhat.com>
Cc: Geert Uytterhoeven <geert@linux-m68k.org>
Cc: Larry Woodman <lwoodman@redhat.com>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Tejun Heo <tj@kernel.org>
Cc: Ingo Molnar <mingo@elte.hu>
Signed-off-by: NAndrew Morton <akpm@linux-foundation.org>
Signed-off-by: NAl Viro <viro@zeniv.linux.org.uk>

Signed-off-by: NAl Viro <viro@zeniv.linux.org.uk>

Don't bother checking for NULL key pointer in key_validate() as all of the
places that call it will crash anyway if the relevant key pointer is NULL by
the time they call key_validate().  Therefore, the checking must be done prior
to calling here.

Whilst we're at it, simplify the key_validate() function a bit and mark its
argument const.
Reported-by: NDan Carpenter <dan.carpenter@oracle.com>
Signed-off-by: NDavid Howells <dhowells@redhat.com>
cc: Dan Carpenter <dan.carpenter@oracle.com>
Signed-off-by: NJames Morris <james.l.morris@oracle.com>

Add support for invalidating a key - which renders it immediately invisible to
further searches and causes the garbage collector to immediately wake up,
remove it from keyrings and then destroy it when it's no longer referenced.

It's better not to do this with keyctl_revoke() as that marks the key to start
returning -EKEYREVOKED to searches when what is actually desired is to have the
key refetched.

To invalidate a key the caller must be granted SEARCH permission by the key.
This may be too strict.  It may be better to also permit invalidation if the
caller has any of READ, WRITE or SETATTR permission.

The primary use for this is to evict keys that are cached in special keyrings,
such as the DNS resolver or an ID mapper.
Signed-off-by: NDavid Howells <dhowells@redhat.com>

Do an LRU discard in keyrings that are full rather than returning ENFILE.  To
perform this, a time_t is added to the key struct and updated by the creation
of a link to a key and by a key being found as the result of a search.  At the
completion of a successful search, the keyrings in the path between the root of
the search and the first found link to it also have their last-used times
updated.

Note that discarding a link to a key from a keyring does not necessarily
destroy the key as there may be references held by other places.

An alternate discard method that might suffice is to perform FIFO discard from
the keyring, using the spare 2-byte hole in the keylist header as the index of
the next link to be discarded.

This is useful when using a keyring as a cache for DNS results or foreign
filesystem IDs.


This can be tested by the following.  As root do:

	echo 1000 >/proc/sys/kernel/keys/root_maxkeys

	kr=`keyctl newring foo @s`
	for ((i=0; i<2000; i++)); do keyctl add user a$i a $kr; done

Without this patch ENFILE should be reported when the keyring fills up.  With
this patch, the keyring discards keys in an LRU fashion.  Note that the stored
LRU time has a granularity of 1s.

After doing this, /proc/key-users can be observed and should show that most of
the 2000 keys have been discarded:

	[root@andromeda ~]# cat /proc/key-users
	    0:   517 516/516 513/1000 5249/20000

The "513/1000" here is the number of quota-accounted keys present for this user
out of the maximum permitted.

In /proc/keys, the keyring shows the number of keys it has and the number of
slots it has allocated:

	[root@andromeda ~]# grep foo /proc/keys
	200c64c4 I--Q--     1 perm 3b3f0000     0     0 keyring   foo: 509/509

The maximum is (PAGE_SIZE - header) / key pointer size.  That's typically 509
on a 64-bit system and 1020 on a 32-bit system.
Signed-off-by: NDavid Howells <dhowells@redhat.com>

Make use of the previous patch that makes the garbage collector perform RCU
synchronisation before destroying defunct keys.  Key pointers can now be
replaced in-place without creating a new keyring payload and replacing the
whole thing as the discarded keys will not be destroyed until all currently
held RCU read locks are released.

If the keyring payload space needs to be expanded or contracted, then a
replacement will still need allocating, and the original will still have to be
freed by RCU.
Signed-off-by: NDavid Howells <dhowells@redhat.com>

Make the keys garbage collector invoke synchronize_rcu() prior to destroying
keys with a zero usage count.  This means that a key can be examined under the
RCU read lock in the safe knowledge that it won't get deallocated until after
the lock is released - even if its usage count becomes zero whilst we're
looking at it.

This is useful in keyring search vs key link.  Consider a keyring containing a
link to a key.  That link can be replaced in-place in the keyring without
requiring an RCU copy-and-replace on the keyring contents without breaking a
search underway on that keyring when the displaced key is released, provided
the key is actually destroyed only after the RCU read lock held by the search
algorithm is released.

This permits __key_link() to replace a key without having to reallocate the key
payload.  A key gets replaced if a new key being linked into a keyring has the
same type and description.
Signed-off-by: NDavid Howells <dhowells@redhat.com>
Acked-by: NJeff Layton <jlayton@redhat.com>

Announce the (un)registration of a key type in the core key code rather than
in the callers.
Signed-off-by: NDavid Howells <dhowells@redhat.com>
Acked-by: NMimi Zohar <zohar@us.ibm.com>

Reorganise the keys directory Makefile to put all the core bits together and
the type-specific bits after.
Signed-off-by: NDavid Howells <dhowells@redhat.com>
Acked-by: NMimi Zohar <zohar@us.ibm.com>

Move the key config into security/keys/Kconfig as there are going to be a lot
of key-related options.
Signed-off-by: NDavid Howells <dhowells@redhat.com>
Acked-by: NMimi Zohar <zohar@us.ibm.com>

As a first step to converting struct cred to be all kuid_t and kgid_t
values convert the group values stored in group_info to always be
kgid_t values.   Unless user namespaces are used this change should
have no effect.
Acked-by: NSerge Hallyn <serge.hallyn@canonical.com>
Signed-off-by: NEric W. Biederman <ebiederm@xmission.com>

struct user_struct will shortly loose it's user_ns reference
so make the cred user_ns reference a proper reference complete
with reference counting.
Acked-by: NSerge Hallyn <serge.hallyn@canonical.com>
Signed-off-by: NEric W. Biederman <ebiederm@xmission.com>

Optimize performance and prepare for the removal of the user_ns reference
from user_struct.  Remove the slow long walk through cred->user->user_ns and
instead go straight to cred->user_ns.
Acked-by: NSerge Hallyn <serge.hallyn@canonical.com>
Signed-off-by: NEric W. Biederman <ebiederm@xmission.com>

No functional changes.  It is not sane to use UMH_KILLABLE with enum
umh_wait, but obviously we do not want another argument in
call_usermodehelper_* helpers.  Kill this enum, use the plain int.
Signed-off-by: NOleg Nesterov <oleg@redhat.com>
Cc: Tetsuo Handa <penguin-kernel@I-love.SAKURA.ne.jp>
Cc: Rusty Russell <rusty@rustcorp.com.au>
Cc: Tejun Heo <tj@kernel.org>
Cc: David Rientjes <rientjes@google.com>
Signed-off-by: NAndrew Morton <akpm@linux-foundation.org>
Signed-off-by: NLinus Torvalds <torvalds@linux-foundation.org>

The test for "if (cred->request_key_auth->flags & KEY_FLAG_REVOKED) {"
should actually testing that the (1 << KEY_FLAG_REVOKED) bit is set.
The current code actually checks for KEY_FLAG_DEAD.
Signed-off-by: NDan Carpenter <dan.carpenter@oracle.com>
Signed-off-by: NDavid Howells <dhowells@redhat.com>
Signed-off-by: NJames Morris <james.l.morris@oracle.com>

The keyctl_set_timeout function isn't exported to other parts of the
kernel, but I want to use it for the NFS idmapper.  I already have the
key, but I wanted a generic way to set the timeout.
Signed-off-by: NBryan Schumaker <bjschuma@netapp.com>
Acked-by: NDavid Howells <dhowells@redhat.com>
Signed-off-by: NTrond Myklebust <Trond.Myklebust@netapp.com>

Replace the rcu_assign_pointer() calls with rcu_assign_keypointer().
Signed-off-by: NMimi Zohar <zohar@us.ibm.com>
Signed-off-by: NDavid Howells <dhowells@redhat.com>
Signed-off-by: NJames Morris <jmorris@namei.org>

The kernel contains some special internal keyrings, for instance the DNS
resolver keyring :

2a93faf1 I-----     1 perm 1f030000     0     0 keyring   .dns_resolver: empty

It would occasionally be useful to allow the contents of such keyrings to be
flushed by root (cache invalidation).

Allow a flag to be set on a keyring to mark that someone possessing the
sysadmin capability can clear the keyring, even without normal write access to
the keyring.

Set this flag on the special keyrings created by the DNS resolver, the NFS
identity mapper and the CIFS identity mapper.
Signed-off-by: NDavid Howells <dhowells@redhat.com>
Acked-by: NJeff Layton <jlayton@redhat.com>
Acked-by: NSteve Dickson <steved@redhat.com>
Signed-off-by: NJames Morris <jmorris@namei.org>

For CIFS, we want to be able to store NTLM credentials (aka username
and password) in the keyring. We do not, however want to allow users
to fetch those keys back out of the keyring since that would be a
security risk.

Unfortunately, due to the nuances of key permission bits, it's not
possible to do this. We need to grant search permissions so the kernel
can find these keys, but that also implies permissions to read the
payload.

Resolve this by adding a new key_type. This key type is essentially
the same as key_type_user, but does not define a .read op. This
prevents the payload from ever being visible from userspace. This
key type also vets the description to ensure that it's "qualified"
by checking to ensure that it has a ':' in it that is preceded by
other characters.
Acked-by: NDavid Howells <dhowells@redhat.com>
Signed-off-by: NJeff Layton <jlayton@redhat.com>
Signed-off-by: NSteve French <smfrench@gmail.com>

Enabling CONFIG_PROVE_RCU and CONFIG_SPARSE_RCU_POINTER resulted in
"suspicious rcu_dereference_check() usage!" and "incompatible types
in comparison expression (different address spaces)" messages.

Access the masterkey directly when holding the rwsem.

Changelog v1:
- Use either rcu_read_lock()/rcu_derefence_key()/rcu_read_unlock()
or remove the unnecessary rcu_derefence() - David Howells
Reported-by: NDmitry Kasatkin <dmitry.kasatkin@intel.com>
Signed-off-by: NMimi Zohar <zohar@us.ibm.com>
Signed-off-by: NDavid Howells <dhowells@redhat.com>
Signed-off-by: NJames Morris <jmorris@namei.org>

Define rcu_assign_keypointer(), which uses the key payload.rcudata instead
of payload.data, to resolve the CONFIG_SPARSE_RCU_POINTER message:
"incompatible types in comparison expression (different address spaces)"

Replace the rcu_assign_pointer() calls in encrypted/trusted keys with
rcu_assign_keypointer().
Signed-off-by: NMimi Zohar <zohar@us.ibm.com>
Signed-off-by: NDavid Howells <dhowells@redhat.com>
Signed-off-by: NJames Morris <jmorris@namei.org>

Add missing smp_rmb() primitives to the keyring search code.

When keyring payloads are appended to without replacement (thus using up spare
slots in the key pointer array), an smp_wmb() is issued between the pointer
assignment and the increment of the key count (nkeys).

There should be corresponding read barriers between the read of nkeys and
dereferences of keys[n] when n is dependent on the value of nkeys.
Signed-off-by: NDavid Howells <dhowells@redhat.com>
Reviewed-by: NPaul E. McKenney <paulmck@linux.vnet.ibm.com>
Signed-off-by: NJames Morris <jmorris@namei.org>

Give keys their own lockdep class to differentiate them from each other in case
a key of one type has to refer to a key of another type.
Signed-off-by: NDavid Howells <dhowells@redhat.com>
Acked-by: NMimi Zohar <zohar@us.ibm.com>
Signed-off-by: NJames Morris <jmorris@namei.org>

Encrypted keys are encrypted/decrypted using either a trusted or
user-defined key type, which is referred to as the 'master' key.
The master key may be of type trusted iff the trusted key is
builtin or both the trusted key and encrypted keys are built as
modules.  This patch resolves the build dependency problem.

- Use "masterkey-$(CONFIG_TRUSTED_KEYS)-$(CONFIG_ENCRYPTED_KEYS)" construct
to encapsulate the above logic. (Suggested by Dimtry Kasatkin.)
- Fixing the encrypted-keys Makefile, results in a module name change
from encrypted.ko to encrypted-keys.ko.
- Add module dependency for request_trusted_key() definition
Signed-off-by: NMimi Zohar <zohar@us.ibm.com>

Fix request_master_key() error return code.
Signed-off-by: NMimi Zohar <zohar@us.ibm.com>

Fix a NULL pointer deref in the user-defined key type whereby updating a
negative key into a fully instantiated key will cause an oops to occur
when the code attempts to free the non-existent old payload.

This results in an oops that looks something like the following:

  BUG: unable to handle kernel NULL pointer dereference at 0000000000000008
  IP: [<ffffffff81085fa1>] __call_rcu+0x11/0x13e
  PGD 3391d067 PUD 3894a067 PMD 0
  Oops: 0002 [#1] SMP
  CPU 1
  Pid: 4354, comm: keyctl Not tainted 3.1.0-fsdevel+ #1140                  /DG965RY
  RIP: 0010:[<ffffffff81085fa1>]  [<ffffffff81085fa1>] __call_rcu+0x11/0x13e
  RSP: 0018:ffff88003d591df8  EFLAGS: 00010246
  RAX: 0000000000000000 RBX: 0000000000000000 RCX: 000000000000006e
  RDX: ffffffff8161d0c0 RSI: 0000000000000000 RDI: 0000000000000000
  RBP: ffff88003d591e18 R08: 0000000000000000 R09: ffffffff8152fa6c
  R10: 0000000000000000 R11: 0000000000000300 R12: ffff88003b8f9538
  R13: ffffffff8161d0c0 R14: ffff88003b8f9d50 R15: ffff88003c69f908
  FS:  00007f97eb18c720(0000) GS:ffff88003bd00000(0000) knlGS:0000000000000000
  CS:  0010 DS: 0000 ES: 0000 CR0: 0000000080050033
  CR2: 0000000000000008 CR3: 000000003d47a000 CR4: 00000000000006e0
  DR0: 0000000000000000 DR1: 0000000000000000 DR2: 0000000000000000
  DR3: 0000000000000000 DR6: 00000000ffff0ff0 DR7: 0000000000000400
  Process keyctl (pid: 4354, threadinfo ffff88003d590000, task ffff88003c78a040)
  Stack:
   ffff88003e0ffde0 ffff88003b8f9538 0000000000000001 ffff88003b8f9d50
   ffff88003d591e28 ffffffff810860f0 ffff88003d591e68 ffffffff8117bfea
   ffff88003d591e68 ffffffff00000000 ffff88003e0ffde1 ffff88003e0ffde0
  Call Trace:
   [<ffffffff810860f0>] call_rcu_sched+0x10/0x12
   [<ffffffff8117bfea>] user_update+0x8d/0xa2
   [<ffffffff8117723a>] key_create_or_update+0x236/0x270
   [<ffffffff811789b1>] sys_add_key+0x123/0x17e
   [<ffffffff813b84bb>] system_call_fastpath+0x16/0x1b
Signed-off-by: NDavid Howells <dhowells@redhat.com>
Acked-by: NJeff Layton <jlayton@redhat.com>
Acked-by: NNeil Horman <nhorman@redhat.com>
Acked-by: NSteve Dickson <steved@redhat.com>
Acked-by: NJames Morris <jmorris@namei.org>
Cc: stable@kernel.org
Signed-off-by: NLinus Torvalds <torvalds@linux-foundation.org>

There is no functional change.
Signed-off-by: NAndy Shevchenko <andriy.shevchenko@linux.intel.com>
Cc: Mimi Zohar <zohar@us.ibm.com>
Cc: James Morris <jmorris@namei.org>
Signed-off-by: NAndrew Morton <akpm@linux-foundation.org>
Signed-off-by: NLinus Torvalds <torvalds@linux-foundation.org>

The basic idea behind cross memory attach is to allow MPI programs doing
intra-node communication to do a single copy of the message rather than a
double copy of the message via shared memory.

The following patch attempts to achieve this by allowing a destination
process, given an address and size from a source process, to copy memory
directly from the source process into its own address space via a system
call.  There is also a symmetrical ability to copy from the current
process's address space into a destination process's address space.

- Use of /proc/pid/mem has been considered, but there are issues with
  using it:
  - Does not allow for specifying iovecs for both src and dest, assuming
    preadv or pwritev was implemented either the area read from or
  written to would need to be contiguous.
  - Currently mem_read allows only processes who are currently
  ptrace'ing the target and are still able to ptrace the target to read
  from the target. This check could possibly be moved to the open call,
  but its not clear exactly what race this restriction is stopping
  (reason  appears to have been lost)
  - Having to send the fd of /proc/self/mem via SCM_RIGHTS on unix
  domain socket is a bit ugly from a userspace point of view,
  especially when you may have hundreds if not (eventually) thousands
  of processes  that all need to do this with each other
  - Doesn't allow for some future use of the interface we would like to
  consider adding in the future (see below)
  - Interestingly reading from /proc/pid/mem currently actually
  involves two copies! (But this could be fixed pretty easily)

As mentioned previously use of vmsplice instead was considered, but has
problems.  Since you need the reader and writer working co-operatively if
the pipe is not drained then you block.  Which requires some wrapping to
do non blocking on the send side or polling on the receive.  In all to all
communication it requires ordering otherwise you can deadlock.  And in the
example of many MPI tasks writing to one MPI task vmsplice serialises the
copying.

There are some cases of MPI collectives where even a single copy interface
does not get us the performance gain we could.  For example in an
MPI_Reduce rather than copy the data from the source we would like to
instead use it directly in a mathops (say the reduce is doing a sum) as
this would save us doing a copy.  We don't need to keep a copy of the data
from the source.  I haven't implemented this, but I think this interface
could in the future do all this through the use of the flags - eg could
specify the math operation and type and the kernel rather than just
copying the data would apply the specified operation between the source
and destination and store it in the destination.

Although we don't have a "second user" of the interface (though I've had
some nibbles from people who may be interested in using it for intra
process messaging which is not MPI).  This interface is something which
hardware vendors are already doing for their custom drivers to implement
fast local communication.  And so in addition to this being useful for
OpenMPI it would mean the driver maintainers don't have to fix things up
when the mm changes.

There was some discussion about how much faster a true zero copy would
go. Here's a link back to the email with some testing I did on that:

http://marc.info/?l=linux-mm&m=130105930902915&w=2

There is a basic man page for the proposed interface here:

http://ozlabs.org/~cyeoh/cma/process_vm_readv.txt

This has been implemented for x86 and powerpc, other architecture should
mainly (I think) just need to add syscall numbers for the process_vm_readv
and process_vm_writev. There are 32 bit compatibility versions for
64-bit kernels.

For arch maintainers there are some simple tests to be able to quickly
verify that the syscalls are working correctly here:

http://ozlabs.org/~cyeoh/cma/cma-test-20110718.tgzSigned-off-by: NChris Yeoh <yeohc@au1.ibm.com>
Cc: Ingo Molnar <mingo@elte.hu>
Cc: "H. Peter Anvin" <hpa@zytor.com>
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: Arnd Bergmann <arnd@arndb.de>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Benjamin Herrenschmidt <benh@kernel.crashing.org>
Cc: David Howells <dhowells@redhat.com>
Cc: James Morris <jmorris@namei.org>
Cc: <linux-man@vger.kernel.org>
Cc: <linux-arch@vger.kernel.org>
Signed-off-by: NAndrew Morton <akpm@linux-foundation.org>
Signed-off-by: NLinus Torvalds <torvalds@linux-foundation.org>