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.tgz
Signed-off-by: Chris 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: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
Make request_key() and co. return an error for a negative or rejected key. If
the key was simply negated, then return ENOKEY, otherwise return the error
with which it was rejected.
Without this patch, the following command returns a key number (with the latest
keyutils):
[root@andromeda ~]# keyctl request2 user debug:foo rejected @s
586569904
Trying to print the key merely gets you a permission denied error:
[root@andromeda ~]# keyctl print 586569904
keyctl_read_alloc: Permission denied
Doing another request_key() call does get you the error, as long as it hasn't
expired yet:
[root@andromeda ~]# keyctl request user debug:foo
request_key: Key was rejected by service
Signed-off-by: David Howells <dhowells@redhat.com>
Signed-off-by: James Morris <jmorris@namei.org>
Add a keyctl op (KEYCTL_INSTANTIATE_IOV) that is like KEYCTL_INSTANTIATE, but
takes an iovec array and concatenates the data in-kernel into one buffer.
Since the KEYCTL_INSTANTIATE copies the data anyway, this isn't too much of a
problem.
Signed-off-by: David Howells <dhowells@redhat.com>
Signed-off-by: James Morris <jmorris@namei.org>
Add a new keyctl op to reject a key with a specified error code. This works
much the same as negating a key, and so keyctl_negate_key() is made a special
case of keyctl_reject_key(). The difference is that keyctl_negate_key()
selects ENOKEY as the error to be reported.
Typically the key would be rejected with EKEYEXPIRED, EKEYREVOKED or
EKEYREJECTED, but this is not mandatory.
Signed-off-by: David Howells <dhowells@redhat.com>
Signed-off-by: James Morris <jmorris@namei.org>
Do a bit of a style clean up in the key management code. No functional
changes.
Done using:
perl -p -i -e 's!^/[*]*/\n!!' security/keys/*.c
perl -p -i -e 's!} /[*] end [a-z0-9_]*[(][)] [*]/\n!}\n!' security/keys/*.c
sed -i -s -e ": next" -e N -e 's/^\n[}]$/}/' -e t -e P -e 's/^.*\n//' -e "b next" security/keys/*.c
To remove /*****/ lines, remove comments on the closing brace of a
function to name the function and remove blank lines before the closing
brace of a function.
Signed-off-by: David Howells <dhowells@redhat.com>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
Fix a bug in keyctl_session_to_parent() whereby it tries to check the ownership
of the parent process's session keyring whether or not the parent has a session
keyring [CVE-2010-2960].
This results in the following oops:
BUG: unable to handle kernel NULL pointer dereference at 00000000000000a0
IP: [<ffffffff811ae4dd>] keyctl_session_to_parent+0x251/0x443
...
Call Trace:
[<ffffffff811ae2f3>] ? keyctl_session_to_parent+0x67/0x443
[<ffffffff8109d286>] ? __do_fault+0x24b/0x3d0
[<ffffffff811af98c>] sys_keyctl+0xb4/0xb8
[<ffffffff81001eab>] system_call_fastpath+0x16/0x1b
if the parent process has no session keyring.
If the system is using pam_keyinit then it mostly protected against this as all
processes derived from a login will have inherited the session keyring created
by pam_keyinit during the log in procedure.
To test this, pam_keyinit calls need to be commented out in /etc/pam.d/.
Reported-by: Tavis Ormandy <taviso@cmpxchg8b.com>
Signed-off-by: David Howells <dhowells@redhat.com>
Acked-by: Tavis Ormandy <taviso@cmpxchg8b.com>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
There's an protected access to the parent process's credentials in the middle
of keyctl_session_to_parent(). This results in the following RCU warning:
===================================================
[ INFO: suspicious rcu_dereference_check() usage. ]
---------------------------------------------------
security/keys/keyctl.c:1291 invoked rcu_dereference_check() without protection!
other info that might help us debug this:
rcu_scheduler_active = 1, debug_locks = 0
1 lock held by keyctl-session-/2137:
#0: (tasklist_lock){.+.+..}, at: [<ffffffff811ae2ec>] keyctl_session_to_parent+0x60/0x236
stack backtrace:
Pid: 2137, comm: keyctl-session- Not tainted 2.6.36-rc2-cachefs+ #1
Call Trace:
[<ffffffff8105606a>] lockdep_rcu_dereference+0xaa/0xb3
[<ffffffff811ae379>] keyctl_session_to_parent+0xed/0x236
[<ffffffff811af77e>] sys_keyctl+0xb4/0xb6
[<ffffffff81001eab>] system_call_fastpath+0x16/0x1b
The code should take the RCU read lock to make sure the parents credentials
don't go away, even though it's holding a spinlock and has IRQ disabled.
Signed-off-by: David Howells <dhowells@redhat.com>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
keyctl_describe_key() turns the key reference it gets into a usable key pointer
and assigns that to a variable called 'key', which it then ignores in favour of
recomputing the key pointer each time it needs it. Make it use the precomputed
pointer instead.
Without this patch, gcc 4.6 reports that the variable key is set but not used:
building with gcc 4.6 I'm getting a warning message:
CC security/keys/keyctl.o
security/keys/keyctl.c: In function 'keyctl_describe_key':
security/keys/keyctl.c:472:14: warning: variable 'key' set but not used
Reported-by: Justin P. Mattock <justinmattock@gmail.com>
Signed-off-by: David Howells <dhowells@redhat.com>
Signed-off-by: James Morris <jmorris@namei.org>
Authorise a process to perform keyctl_set_timeout() on an uninstantiated key if
that process has the authorisation key for it.
This allows the instantiator to set the timeout on a key it is instantiating -
provided it does it before instantiating the key.
For instance, the test upcall script provided with the keyutils package could
be modified to set the expiry to an hour hence before instantiating the key:
[/usr/share/keyutils/request-key-debug.sh]
if [ "$3" != "neg" ]
then
+ keyctl timeout $1 3600
keyctl instantiate $1 "Debug $3" $4 || exit 1
else
Signed-off-by: David Howells <dhowells@redhat.com>
Signed-off-by: James Morris <jmorris@namei.org>
This is from a Smatch check I'm writing.
strncpy_from_user() returns -EFAULT on error so the first change just
silences a warning but doesn't change how the code works.
The other change is a bug fix because install_thread_keyring_to_cred()
can return a variety of errors such as -EINVAL, -EEXIST, -ENOMEM or
-EKEYREVOKED.
Signed-off-by: Dan Carpenter <error27@gmail.com>
Signed-off-by: David Howells <dhowells@redhat.com>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
No functional changes.
keyctl_session_to_parent() is the only user of signal->count which needs
the correct value. Change it to use thread_group_empty() instead, this
must be strictly equivalent under tasklist, and imho looks better.
Signed-off-by: Oleg Nesterov <oleg@redhat.com>
Acked-by: David Howells <dhowells@redhat.com>
Cc: Peter Zijlstra <peterz@infradead.org>
Acked-by: Roland McGrath <roland@redhat.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
As of commit ee18d64c1f ("KEYS: Add a keyctl to
install a process's session keyring on its parent [try #6]"), CONFIG_KEYS=y
fails to build on architectures that haven't implemented TIF_NOTIFY_RESUME yet:
security/keys/keyctl.c: In function 'keyctl_session_to_parent':
security/keys/keyctl.c:1312: error: 'TIF_NOTIFY_RESUME' undeclared (first use in this function)
security/keys/keyctl.c:1312: error: (Each undeclared identifier is reported only once
security/keys/keyctl.c:1312: error: for each function it appears in.)
Make KEYCTL_SESSION_TO_PARENT depend on TIF_NOTIFY_RESUME until
m68k, and xtensa have implemented it.
Signed-off-by: Geert Uytterhoeven <geert@linux-m68k.org>
Signed-off-by: James Morris <jmorris@namei.org>
Acked-by: Mike Frysinger <vapier@gentoo.org>
The destination keyring specified to request_key() and co. is made available to
the process that instantiates the key (the slave process started by
/sbin/request-key typically). This is passed in the request_key_auth struct as
the dest_keyring member.
keyctl_instantiate_key and keyctl_negate_key() call get_instantiation_keyring()
to get the keyring to attach the newly constructed key to at the end of
instantiation. This may be given a specific keyring into which a link will be
made later, or it may be asked to find the keyring passed to request_key(). In
the former case, it returns a keyring with the refcount incremented by
lookup_user_key(); in the latter case, it returns the keyring from the
request_key_auth struct - and does _not_ increment the refcount.
The latter case will eventually result in an oops when the keyring prematurely
runs out of references and gets destroyed. The effect may take some time to
show up as the key is destroyed lazily.
To fix this, the keyring returned by get_instantiation_keyring() must always
have its refcount incremented, no matter where it comes from.
This can be tested by setting /etc/request-key.conf to:
#OP TYPE DESCRIPTION CALLOUT INFO PROGRAM ARG1 ARG2 ARG3 ...
#====== ======= =============== =============== ===============================
create * test:* * |/bin/false %u %g %d %{user:_display}
negate * * * /bin/keyctl negate %k 10 @u
and then doing:
keyctl add user _display aaaaaaaa @u
while keyctl request2 user test:x test:x @u &&
keyctl list @u;
do
keyctl request2 user test:x test:x @u;
sleep 31;
keyctl list @u;
done
which will oops eventually. Changing the negate line to have @u rather than
%S at the end is important as that forces the latter case by passing a special
keyring ID rather than an actual keyring ID.
Reported-by: Alexander Zangerl <az@bond.edu.au>
Signed-off-by: David Howells <dhowells@redhat.com>
Tested-by: Alexander Zangerl <az@bond.edu.au>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
Fix a number of problems with the new key garbage collector:
(1) A rogue semicolon in keyring_gc() was causing the initial count of dead
keys to be miscalculated.
(2) A missing return in keyring_gc() meant that under certain circumstances,
the keyring semaphore would be unlocked twice.
(3) The key serial tree iterator (key_garbage_collector()) part of the garbage
collector has been modified to:
(a) Complete each scan of the keyrings before setting the new timer.
(b) Only set the new timer for keys that have yet to expire. This means
that the new timer is now calculated correctly, and the gc doesn't
get into a loop continually scanning for keys that have expired, and
preventing other things from happening, like RCU cleaning up the old
keyring contents.
(c) Perform an extra scan if any keys were garbage collected in this one
as a key might become garbage during a scan, and (b) could mean we
don't set the timer again.
(4) Made key_schedule_gc() take the time at which to do a collection run,
rather than the time at which the key expires. This means the collection
of dead keys (key type unregistered) can happen immediately.
Signed-off-by: David Howells <dhowells@redhat.com>
Signed-off-by: James Morris <jmorris@namei.org>
When we exit early from keyctl_session_to_parent because of permissions or
because the session keyring is the same as the parent, we need to unlock the
tasklist.
The missing unlock causes the system to hang completely when using
keyctl(KEYCTL_SESSION_TO_PARENT) with a keyring shared with the parent.
Signed-off-by: Marc Dionne <marc.c.dionne@gmail.com>
Signed-off-by: David Howells <dhowells@redhat.com>
Signed-off-by: James Morris <jmorris@namei.org>
Add a keyctl to install a process's session keyring onto its parent. This
replaces the parent's session keyring. Because the COW credential code does
not permit one process to change another process's credentials directly, the
change is deferred until userspace next starts executing again. Normally this
will be after a wait*() syscall.
To support this, three new security hooks have been provided:
cred_alloc_blank() to allocate unset security creds, cred_transfer() to fill in
the blank security creds and key_session_to_parent() - which asks the LSM if
the process may replace its parent's session keyring.
The replacement may only happen if the process has the same ownership details
as its parent, and the process has LINK permission on the session keyring, and
the session keyring is owned by the process, and the LSM permits it.
Note that this requires alteration to each architecture's notify_resume path.
This has been done for all arches barring blackfin, m68k* and xtensa, all of
which need assembly alteration to support TIF_NOTIFY_RESUME. This allows the
replacement to be performed at the point the parent process resumes userspace
execution.
This allows the userspace AFS pioctl emulation to fully emulate newpag() and
the VIOCSETTOK and VIOCSETTOK2 pioctls, all of which require the ability to
alter the parent process's PAG membership. However, since kAFS doesn't use
PAGs per se, but rather dumps the keys into the session keyring, the session
keyring of the parent must be replaced if, for example, VIOCSETTOK is passed
the newpag flag.
This can be tested with the following program:
#include <stdio.h>
#include <stdlib.h>
#include <keyutils.h>
#define KEYCTL_SESSION_TO_PARENT 18
#define OSERROR(X, S) do { if ((long)(X) == -1) { perror(S); exit(1); } } while(0)
int main(int argc, char **argv)
{
key_serial_t keyring, key;
long ret;
keyring = keyctl_join_session_keyring(argv[1]);
OSERROR(keyring, "keyctl_join_session_keyring");
key = add_key("user", "a", "b", 1, keyring);
OSERROR(key, "add_key");
ret = keyctl(KEYCTL_SESSION_TO_PARENT);
OSERROR(ret, "KEYCTL_SESSION_TO_PARENT");
return 0;
}
Compiled and linked with -lkeyutils, you should see something like:
[dhowells@andromeda ~]$ keyctl show
Session Keyring
-3 --alswrv 4043 4043 keyring: _ses
355907932 --alswrv 4043 -1 \_ keyring: _uid.4043
[dhowells@andromeda ~]$ /tmp/newpag
[dhowells@andromeda ~]$ keyctl show
Session Keyring
-3 --alswrv 4043 4043 keyring: _ses
1055658746 --alswrv 4043 4043 \_ user: a
[dhowells@andromeda ~]$ /tmp/newpag hello
[dhowells@andromeda ~]$ keyctl show
Session Keyring
-3 --alswrv 4043 4043 keyring: hello
340417692 --alswrv 4043 4043 \_ user: a
Where the test program creates a new session keyring, sticks a user key named
'a' into it and then installs it on its parent.
Signed-off-by: David Howells <dhowells@redhat.com>
Signed-off-by: James Morris <jmorris@namei.org>
Add garbage collection for dead, revoked and expired keys. This involved
erasing all links to such keys from keyrings that point to them. At that
point, the key will be deleted in the normal manner.
Keyrings from which garbage collection occurs are shrunk and their quota
consumption reduced as appropriate.
Dead keys (for which the key type has been removed) will be garbage collected
immediately.
Revoked and expired keys will hang around for a number of seconds, as set in
/proc/sys/kernel/keys/gc_delay before being automatically removed. The default
is 5 minutes.
Signed-off-by: David Howells <dhowells@redhat.com>
Signed-off-by: James Morris <jmorris@namei.org>
Allow keyctl_revoke() to operate on keys that have SETATTR but not WRITE
permission, rather than only on keys that have WRITE permission.
Signed-off-by: David Howells <dhowells@redhat.com>
Acked-by: Serge Hallyn <serue@us.ibm.com>
Signed-off-by: James Morris <jmorris@namei.org>
Allow keys for which the key type has been removed to be unlinked. Currently
dead-type keys can only be disposed of by completely clearing the keyrings
that point to them.
Signed-off-by: David Howells <dhowells@redhat.com>
Acked-by: Serge Hallyn <serue@us.ibm.com>
Signed-off-by: James Morris <jmorris@namei.org>
per-uid keys were looked by uid only. Use the user namespace
to distinguish the same uid in different namespaces.
This does not address key_permission. So a task can for instance
try to join a keyring owned by the same uid in another namespace.
That will be handled by a separate patch.
Signed-off-by: Serge E. Hallyn <serue@us.ibm.com>
Acked-by: David Howells <dhowells@redhat.com>
Signed-off-by: James Morris <jmorris@namei.org>
Christopher Yeoh
David Howells
Dan Carpenter
Oleg Nesterov
Justin P. Mattock
Eric Paris
Geert Uytterhoeven
Roel Kluin
Marc Dionne
Serge E. Hallyn
Vegard Nossum