2019-05-27 13:55:01 +07:00
|
|
|
// SPDX-License-Identifier: GPL-2.0-or-later
|
2007-04-27 05:57:07 +07:00
|
|
|
/* AFS security handling
|
|
|
|
*
|
afs: Overhaul permit caching
Overhaul permit caching in AFS by making it per-vnode and sharing permit
lists where possible.
When most of the fileserver operations are called, they return a status
structure indicating the (revised) details of the vnode or vnodes involved
in the operation. This includes the access mark derived from the ACL
(named CallerAccess in the protocol definition file). This is cacheable
and if the ACL changes, the server will tell us that it is breaking the
callback promise, at which point we can discard the currently cached
permits.
With this patch, the afs_permits structure has, at the end, an array of
{ key, CallerAccess } elements, sorted by key pointer. This is then cached
in a hash table so that it can be shared between vnodes with the same
access permits.
Permit lists can only be shared if they contain the exact same set of
key->CallerAccess mappings.
Note that that table is global rather than being per-net_ns. If the keys
in a permit list cross net_ns boundaries, there is no problem sharing the
cached permits, since the permits are just integer masks.
Since permit lists pin keys, the permit cache also makes it easier for a
future patch to find all occurrences of a key and remove them by means of
setting the afs_permits::invalidated flag and then clearing the appropriate
key pointer. In such an event, memory barriers will need adding.
Lastly, the permit caching is skipped if the server has sent either a
vnode-specific or an entire-server callback since the start of the
operation.
Signed-off-by: David Howells <dhowells@redhat.com>
2017-11-02 22:27:49 +07:00
|
|
|
* Copyright (C) 2007, 2017 Red Hat, Inc. All Rights Reserved.
|
2007-04-27 05:57:07 +07:00
|
|
|
* Written by David Howells (dhowells@redhat.com)
|
|
|
|
*/
|
|
|
|
|
|
|
|
#include <linux/init.h>
|
|
|
|
#include <linux/slab.h>
|
|
|
|
#include <linux/fs.h>
|
|
|
|
#include <linux/ctype.h>
|
Detach sched.h from mm.h
First thing mm.h does is including sched.h solely for can_do_mlock() inline
function which has "current" dereference inside. By dealing with can_do_mlock()
mm.h can be detached from sched.h which is good. See below, why.
This patch
a) removes unconditional inclusion of sched.h from mm.h
b) makes can_do_mlock() normal function in mm/mlock.c
c) exports can_do_mlock() to not break compilation
d) adds sched.h inclusions back to files that were getting it indirectly.
e) adds less bloated headers to some files (asm/signal.h, jiffies.h) that were
getting them indirectly
Net result is:
a) mm.h users would get less code to open, read, preprocess, parse, ... if
they don't need sched.h
b) sched.h stops being dependency for significant number of files:
on x86_64 allmodconfig touching sched.h results in recompile of 4083 files,
after patch it's only 3744 (-8.3%).
Cross-compile tested on
all arm defconfigs, all mips defconfigs, all powerpc defconfigs,
alpha alpha-up
arm
i386 i386-up i386-defconfig i386-allnoconfig
ia64 ia64-up
m68k
mips
parisc parisc-up
powerpc powerpc-up
s390 s390-up
sparc sparc-up
sparc64 sparc64-up
um-x86_64
x86_64 x86_64-up x86_64-defconfig x86_64-allnoconfig
as well as my two usual configs.
Signed-off-by: Alexey Dobriyan <adobriyan@gmail.com>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2007-05-21 04:22:52 +07:00
|
|
|
#include <linux/sched.h>
|
afs: Overhaul permit caching
Overhaul permit caching in AFS by making it per-vnode and sharing permit
lists where possible.
When most of the fileserver operations are called, they return a status
structure indicating the (revised) details of the vnode or vnodes involved
in the operation. This includes the access mark derived from the ACL
(named CallerAccess in the protocol definition file). This is cacheable
and if the ACL changes, the server will tell us that it is breaking the
callback promise, at which point we can discard the currently cached
permits.
With this patch, the afs_permits structure has, at the end, an array of
{ key, CallerAccess } elements, sorted by key pointer. This is then cached
in a hash table so that it can be shared between vnodes with the same
access permits.
Permit lists can only be shared if they contain the exact same set of
key->CallerAccess mappings.
Note that that table is global rather than being per-net_ns. If the keys
in a permit list cross net_ns boundaries, there is no problem sharing the
cached permits, since the permits are just integer masks.
Since permit lists pin keys, the permit cache also makes it easier for a
future patch to find all occurrences of a key and remove them by means of
setting the afs_permits::invalidated flag and then clearing the appropriate
key pointer. In such an event, memory barriers will need adding.
Lastly, the permit caching is skipped if the server has sent either a
vnode-specific or an entire-server callback since the start of the
operation.
Signed-off-by: David Howells <dhowells@redhat.com>
2017-11-02 22:27:49 +07:00
|
|
|
#include <linux/hashtable.h>
|
2007-04-27 05:57:07 +07:00
|
|
|
#include <keys/rxrpc-type.h>
|
|
|
|
#include "internal.h"
|
|
|
|
|
afs: Overhaul permit caching
Overhaul permit caching in AFS by making it per-vnode and sharing permit
lists where possible.
When most of the fileserver operations are called, they return a status
structure indicating the (revised) details of the vnode or vnodes involved
in the operation. This includes the access mark derived from the ACL
(named CallerAccess in the protocol definition file). This is cacheable
and if the ACL changes, the server will tell us that it is breaking the
callback promise, at which point we can discard the currently cached
permits.
With this patch, the afs_permits structure has, at the end, an array of
{ key, CallerAccess } elements, sorted by key pointer. This is then cached
in a hash table so that it can be shared between vnodes with the same
access permits.
Permit lists can only be shared if they contain the exact same set of
key->CallerAccess mappings.
Note that that table is global rather than being per-net_ns. If the keys
in a permit list cross net_ns boundaries, there is no problem sharing the
cached permits, since the permits are just integer masks.
Since permit lists pin keys, the permit cache also makes it easier for a
future patch to find all occurrences of a key and remove them by means of
setting the afs_permits::invalidated flag and then clearing the appropriate
key pointer. In such an event, memory barriers will need adding.
Lastly, the permit caching is skipped if the server has sent either a
vnode-specific or an entire-server callback since the start of the
operation.
Signed-off-by: David Howells <dhowells@redhat.com>
2017-11-02 22:27:49 +07:00
|
|
|
static DEFINE_HASHTABLE(afs_permits_cache, 10);
|
|
|
|
static DEFINE_SPINLOCK(afs_permits_lock);
|
|
|
|
|
2007-04-27 05:57:07 +07:00
|
|
|
/*
|
|
|
|
* get a key
|
|
|
|
*/
|
|
|
|
struct key *afs_request_key(struct afs_cell *cell)
|
|
|
|
{
|
|
|
|
struct key *key;
|
|
|
|
|
|
|
|
_enter("{%x}", key_serial(cell->anonymous_key));
|
|
|
|
|
|
|
|
_debug("key %s", cell->anonymous_key->description);
|
|
|
|
key = request_key(&key_type_rxrpc, cell->anonymous_key->description,
|
|
|
|
NULL);
|
|
|
|
if (IS_ERR(key)) {
|
|
|
|
if (PTR_ERR(key) != -ENOKEY) {
|
|
|
|
_leave(" = %ld", PTR_ERR(key));
|
|
|
|
return key;
|
|
|
|
}
|
|
|
|
|
|
|
|
/* act as anonymous user */
|
|
|
|
_leave(" = {%x} [anon]", key_serial(cell->anonymous_key));
|
|
|
|
return key_get(cell->anonymous_key);
|
|
|
|
} else {
|
|
|
|
/* act as authorised user */
|
|
|
|
_leave(" = {%x} [auth]", key_serial(key));
|
|
|
|
return key;
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
|
|
|
/*
|
afs: Overhaul permit caching
Overhaul permit caching in AFS by making it per-vnode and sharing permit
lists where possible.
When most of the fileserver operations are called, they return a status
structure indicating the (revised) details of the vnode or vnodes involved
in the operation. This includes the access mark derived from the ACL
(named CallerAccess in the protocol definition file). This is cacheable
and if the ACL changes, the server will tell us that it is breaking the
callback promise, at which point we can discard the currently cached
permits.
With this patch, the afs_permits structure has, at the end, an array of
{ key, CallerAccess } elements, sorted by key pointer. This is then cached
in a hash table so that it can be shared between vnodes with the same
access permits.
Permit lists can only be shared if they contain the exact same set of
key->CallerAccess mappings.
Note that that table is global rather than being per-net_ns. If the keys
in a permit list cross net_ns boundaries, there is no problem sharing the
cached permits, since the permits are just integer masks.
Since permit lists pin keys, the permit cache also makes it easier for a
future patch to find all occurrences of a key and remove them by means of
setting the afs_permits::invalidated flag and then clearing the appropriate
key pointer. In such an event, memory barriers will need adding.
Lastly, the permit caching is skipped if the server has sent either a
vnode-specific or an entire-server callback since the start of the
operation.
Signed-off-by: David Howells <dhowells@redhat.com>
2017-11-02 22:27:49 +07:00
|
|
|
* Dispose of a list of permits.
|
2007-04-27 05:57:07 +07:00
|
|
|
*/
|
afs: Overhaul permit caching
Overhaul permit caching in AFS by making it per-vnode and sharing permit
lists where possible.
When most of the fileserver operations are called, they return a status
structure indicating the (revised) details of the vnode or vnodes involved
in the operation. This includes the access mark derived from the ACL
(named CallerAccess in the protocol definition file). This is cacheable
and if the ACL changes, the server will tell us that it is breaking the
callback promise, at which point we can discard the currently cached
permits.
With this patch, the afs_permits structure has, at the end, an array of
{ key, CallerAccess } elements, sorted by key pointer. This is then cached
in a hash table so that it can be shared between vnodes with the same
access permits.
Permit lists can only be shared if they contain the exact same set of
key->CallerAccess mappings.
Note that that table is global rather than being per-net_ns. If the keys
in a permit list cross net_ns boundaries, there is no problem sharing the
cached permits, since the permits are just integer masks.
Since permit lists pin keys, the permit cache also makes it easier for a
future patch to find all occurrences of a key and remove them by means of
setting the afs_permits::invalidated flag and then clearing the appropriate
key pointer. In such an event, memory barriers will need adding.
Lastly, the permit caching is skipped if the server has sent either a
vnode-specific or an entire-server callback since the start of the
operation.
Signed-off-by: David Howells <dhowells@redhat.com>
2017-11-02 22:27:49 +07:00
|
|
|
static void afs_permits_rcu(struct rcu_head *rcu)
|
2007-04-27 05:57:07 +07:00
|
|
|
{
|
|
|
|
struct afs_permits *permits =
|
|
|
|
container_of(rcu, struct afs_permits, rcu);
|
afs: Overhaul permit caching
Overhaul permit caching in AFS by making it per-vnode and sharing permit
lists where possible.
When most of the fileserver operations are called, they return a status
structure indicating the (revised) details of the vnode or vnodes involved
in the operation. This includes the access mark derived from the ACL
(named CallerAccess in the protocol definition file). This is cacheable
and if the ACL changes, the server will tell us that it is breaking the
callback promise, at which point we can discard the currently cached
permits.
With this patch, the afs_permits structure has, at the end, an array of
{ key, CallerAccess } elements, sorted by key pointer. This is then cached
in a hash table so that it can be shared between vnodes with the same
access permits.
Permit lists can only be shared if they contain the exact same set of
key->CallerAccess mappings.
Note that that table is global rather than being per-net_ns. If the keys
in a permit list cross net_ns boundaries, there is no problem sharing the
cached permits, since the permits are just integer masks.
Since permit lists pin keys, the permit cache also makes it easier for a
future patch to find all occurrences of a key and remove them by means of
setting the afs_permits::invalidated flag and then clearing the appropriate
key pointer. In such an event, memory barriers will need adding.
Lastly, the permit caching is skipped if the server has sent either a
vnode-specific or an entire-server callback since the start of the
operation.
Signed-off-by: David Howells <dhowells@redhat.com>
2017-11-02 22:27:49 +07:00
|
|
|
int i;
|
2007-04-27 05:57:07 +07:00
|
|
|
|
afs: Overhaul permit caching
Overhaul permit caching in AFS by making it per-vnode and sharing permit
lists where possible.
When most of the fileserver operations are called, they return a status
structure indicating the (revised) details of the vnode or vnodes involved
in the operation. This includes the access mark derived from the ACL
(named CallerAccess in the protocol definition file). This is cacheable
and if the ACL changes, the server will tell us that it is breaking the
callback promise, at which point we can discard the currently cached
permits.
With this patch, the afs_permits structure has, at the end, an array of
{ key, CallerAccess } elements, sorted by key pointer. This is then cached
in a hash table so that it can be shared between vnodes with the same
access permits.
Permit lists can only be shared if they contain the exact same set of
key->CallerAccess mappings.
Note that that table is global rather than being per-net_ns. If the keys
in a permit list cross net_ns boundaries, there is no problem sharing the
cached permits, since the permits are just integer masks.
Since permit lists pin keys, the permit cache also makes it easier for a
future patch to find all occurrences of a key and remove them by means of
setting the afs_permits::invalidated flag and then clearing the appropriate
key pointer. In such an event, memory barriers will need adding.
Lastly, the permit caching is skipped if the server has sent either a
vnode-specific or an entire-server callback since the start of the
operation.
Signed-off-by: David Howells <dhowells@redhat.com>
2017-11-02 22:27:49 +07:00
|
|
|
for (i = 0; i < permits->nr_permits; i++)
|
|
|
|
key_put(permits->permits[i].key);
|
2007-04-27 05:57:07 +07:00
|
|
|
kfree(permits);
|
|
|
|
}
|
|
|
|
|
|
|
|
/*
|
afs: Overhaul permit caching
Overhaul permit caching in AFS by making it per-vnode and sharing permit
lists where possible.
When most of the fileserver operations are called, they return a status
structure indicating the (revised) details of the vnode or vnodes involved
in the operation. This includes the access mark derived from the ACL
(named CallerAccess in the protocol definition file). This is cacheable
and if the ACL changes, the server will tell us that it is breaking the
callback promise, at which point we can discard the currently cached
permits.
With this patch, the afs_permits structure has, at the end, an array of
{ key, CallerAccess } elements, sorted by key pointer. This is then cached
in a hash table so that it can be shared between vnodes with the same
access permits.
Permit lists can only be shared if they contain the exact same set of
key->CallerAccess mappings.
Note that that table is global rather than being per-net_ns. If the keys
in a permit list cross net_ns boundaries, there is no problem sharing the
cached permits, since the permits are just integer masks.
Since permit lists pin keys, the permit cache also makes it easier for a
future patch to find all occurrences of a key and remove them by means of
setting the afs_permits::invalidated flag and then clearing the appropriate
key pointer. In such an event, memory barriers will need adding.
Lastly, the permit caching is skipped if the server has sent either a
vnode-specific or an entire-server callback since the start of the
operation.
Signed-off-by: David Howells <dhowells@redhat.com>
2017-11-02 22:27:49 +07:00
|
|
|
* Discard a permission cache.
|
2007-04-27 05:57:07 +07:00
|
|
|
*/
|
afs: Overhaul permit caching
Overhaul permit caching in AFS by making it per-vnode and sharing permit
lists where possible.
When most of the fileserver operations are called, they return a status
structure indicating the (revised) details of the vnode or vnodes involved
in the operation. This includes the access mark derived from the ACL
(named CallerAccess in the protocol definition file). This is cacheable
and if the ACL changes, the server will tell us that it is breaking the
callback promise, at which point we can discard the currently cached
permits.
With this patch, the afs_permits structure has, at the end, an array of
{ key, CallerAccess } elements, sorted by key pointer. This is then cached
in a hash table so that it can be shared between vnodes with the same
access permits.
Permit lists can only be shared if they contain the exact same set of
key->CallerAccess mappings.
Note that that table is global rather than being per-net_ns. If the keys
in a permit list cross net_ns boundaries, there is no problem sharing the
cached permits, since the permits are just integer masks.
Since permit lists pin keys, the permit cache also makes it easier for a
future patch to find all occurrences of a key and remove them by means of
setting the afs_permits::invalidated flag and then clearing the appropriate
key pointer. In such an event, memory barriers will need adding.
Lastly, the permit caching is skipped if the server has sent either a
vnode-specific or an entire-server callback since the start of the
operation.
Signed-off-by: David Howells <dhowells@redhat.com>
2017-11-02 22:27:49 +07:00
|
|
|
void afs_put_permits(struct afs_permits *permits)
|
2007-04-27 05:57:07 +07:00
|
|
|
{
|
afs: Overhaul permit caching
Overhaul permit caching in AFS by making it per-vnode and sharing permit
lists where possible.
When most of the fileserver operations are called, they return a status
structure indicating the (revised) details of the vnode or vnodes involved
in the operation. This includes the access mark derived from the ACL
(named CallerAccess in the protocol definition file). This is cacheable
and if the ACL changes, the server will tell us that it is breaking the
callback promise, at which point we can discard the currently cached
permits.
With this patch, the afs_permits structure has, at the end, an array of
{ key, CallerAccess } elements, sorted by key pointer. This is then cached
in a hash table so that it can be shared between vnodes with the same
access permits.
Permit lists can only be shared if they contain the exact same set of
key->CallerAccess mappings.
Note that that table is global rather than being per-net_ns. If the keys
in a permit list cross net_ns boundaries, there is no problem sharing the
cached permits, since the permits are just integer masks.
Since permit lists pin keys, the permit cache also makes it easier for a
future patch to find all occurrences of a key and remove them by means of
setting the afs_permits::invalidated flag and then clearing the appropriate
key pointer. In such an event, memory barriers will need adding.
Lastly, the permit caching is skipped if the server has sent either a
vnode-specific or an entire-server callback since the start of the
operation.
Signed-off-by: David Howells <dhowells@redhat.com>
2017-11-02 22:27:49 +07:00
|
|
|
if (permits && refcount_dec_and_test(&permits->usage)) {
|
|
|
|
spin_lock(&afs_permits_lock);
|
|
|
|
hash_del_rcu(&permits->hash_node);
|
|
|
|
spin_unlock(&afs_permits_lock);
|
|
|
|
call_rcu(&permits->rcu, afs_permits_rcu);
|
|
|
|
}
|
2007-04-27 05:57:07 +07:00
|
|
|
}
|
|
|
|
|
|
|
|
/*
|
afs: Overhaul permit caching
Overhaul permit caching in AFS by making it per-vnode and sharing permit
lists where possible.
When most of the fileserver operations are called, they return a status
structure indicating the (revised) details of the vnode or vnodes involved
in the operation. This includes the access mark derived from the ACL
(named CallerAccess in the protocol definition file). This is cacheable
and if the ACL changes, the server will tell us that it is breaking the
callback promise, at which point we can discard the currently cached
permits.
With this patch, the afs_permits structure has, at the end, an array of
{ key, CallerAccess } elements, sorted by key pointer. This is then cached
in a hash table so that it can be shared between vnodes with the same
access permits.
Permit lists can only be shared if they contain the exact same set of
key->CallerAccess mappings.
Note that that table is global rather than being per-net_ns. If the keys
in a permit list cross net_ns boundaries, there is no problem sharing the
cached permits, since the permits are just integer masks.
Since permit lists pin keys, the permit cache also makes it easier for a
future patch to find all occurrences of a key and remove them by means of
setting the afs_permits::invalidated flag and then clearing the appropriate
key pointer. In such an event, memory barriers will need adding.
Lastly, the permit caching is skipped if the server has sent either a
vnode-specific or an entire-server callback since the start of the
operation.
Signed-off-by: David Howells <dhowells@redhat.com>
2017-11-02 22:27:49 +07:00
|
|
|
* Clear a permit cache on callback break.
|
2007-04-27 05:57:07 +07:00
|
|
|
*/
|
afs: Overhaul permit caching
Overhaul permit caching in AFS by making it per-vnode and sharing permit
lists where possible.
When most of the fileserver operations are called, they return a status
structure indicating the (revised) details of the vnode or vnodes involved
in the operation. This includes the access mark derived from the ACL
(named CallerAccess in the protocol definition file). This is cacheable
and if the ACL changes, the server will tell us that it is breaking the
callback promise, at which point we can discard the currently cached
permits.
With this patch, the afs_permits structure has, at the end, an array of
{ key, CallerAccess } elements, sorted by key pointer. This is then cached
in a hash table so that it can be shared between vnodes with the same
access permits.
Permit lists can only be shared if they contain the exact same set of
key->CallerAccess mappings.
Note that that table is global rather than being per-net_ns. If the keys
in a permit list cross net_ns boundaries, there is no problem sharing the
cached permits, since the permits are just integer masks.
Since permit lists pin keys, the permit cache also makes it easier for a
future patch to find all occurrences of a key and remove them by means of
setting the afs_permits::invalidated flag and then clearing the appropriate
key pointer. In such an event, memory barriers will need adding.
Lastly, the permit caching is skipped if the server has sent either a
vnode-specific or an entire-server callback since the start of the
operation.
Signed-off-by: David Howells <dhowells@redhat.com>
2017-11-02 22:27:49 +07:00
|
|
|
void afs_clear_permits(struct afs_vnode *vnode)
|
2007-04-27 05:57:07 +07:00
|
|
|
{
|
afs: Overhaul permit caching
Overhaul permit caching in AFS by making it per-vnode and sharing permit
lists where possible.
When most of the fileserver operations are called, they return a status
structure indicating the (revised) details of the vnode or vnodes involved
in the operation. This includes the access mark derived from the ACL
(named CallerAccess in the protocol definition file). This is cacheable
and if the ACL changes, the server will tell us that it is breaking the
callback promise, at which point we can discard the currently cached
permits.
With this patch, the afs_permits structure has, at the end, an array of
{ key, CallerAccess } elements, sorted by key pointer. This is then cached
in a hash table so that it can be shared between vnodes with the same
access permits.
Permit lists can only be shared if they contain the exact same set of
key->CallerAccess mappings.
Note that that table is global rather than being per-net_ns. If the keys
in a permit list cross net_ns boundaries, there is no problem sharing the
cached permits, since the permits are just integer masks.
Since permit lists pin keys, the permit cache also makes it easier for a
future patch to find all occurrences of a key and remove them by means of
setting the afs_permits::invalidated flag and then clearing the appropriate
key pointer. In such an event, memory barriers will need adding.
Lastly, the permit caching is skipped if the server has sent either a
vnode-specific or an entire-server callback since the start of the
operation.
Signed-off-by: David Howells <dhowells@redhat.com>
2017-11-02 22:27:49 +07:00
|
|
|
struct afs_permits *permits;
|
2007-04-27 05:57:07 +07:00
|
|
|
|
afs: Overhaul permit caching
Overhaul permit caching in AFS by making it per-vnode and sharing permit
lists where possible.
When most of the fileserver operations are called, they return a status
structure indicating the (revised) details of the vnode or vnodes involved
in the operation. This includes the access mark derived from the ACL
(named CallerAccess in the protocol definition file). This is cacheable
and if the ACL changes, the server will tell us that it is breaking the
callback promise, at which point we can discard the currently cached
permits.
With this patch, the afs_permits structure has, at the end, an array of
{ key, CallerAccess } elements, sorted by key pointer. This is then cached
in a hash table so that it can be shared between vnodes with the same
access permits.
Permit lists can only be shared if they contain the exact same set of
key->CallerAccess mappings.
Note that that table is global rather than being per-net_ns. If the keys
in a permit list cross net_ns boundaries, there is no problem sharing the
cached permits, since the permits are just integer masks.
Since permit lists pin keys, the permit cache also makes it easier for a
future patch to find all occurrences of a key and remove them by means of
setting the afs_permits::invalidated flag and then clearing the appropriate
key pointer. In such an event, memory barriers will need adding.
Lastly, the permit caching is skipped if the server has sent either a
vnode-specific or an entire-server callback since the start of the
operation.
Signed-off-by: David Howells <dhowells@redhat.com>
2017-11-02 22:27:49 +07:00
|
|
|
spin_lock(&vnode->lock);
|
|
|
|
permits = rcu_dereference_protected(vnode->permit_cache,
|
|
|
|
lockdep_is_held(&vnode->lock));
|
|
|
|
RCU_INIT_POINTER(vnode->permit_cache, NULL);
|
|
|
|
spin_unlock(&vnode->lock);
|
2007-04-27 05:57:07 +07:00
|
|
|
|
2019-05-11 05:14:41 +07:00
|
|
|
afs_put_permits(permits);
|
2007-04-27 05:57:07 +07:00
|
|
|
}
|
|
|
|
|
|
|
|
/*
|
afs: Overhaul permit caching
Overhaul permit caching in AFS by making it per-vnode and sharing permit
lists where possible.
When most of the fileserver operations are called, they return a status
structure indicating the (revised) details of the vnode or vnodes involved
in the operation. This includes the access mark derived from the ACL
(named CallerAccess in the protocol definition file). This is cacheable
and if the ACL changes, the server will tell us that it is breaking the
callback promise, at which point we can discard the currently cached
permits.
With this patch, the afs_permits structure has, at the end, an array of
{ key, CallerAccess } elements, sorted by key pointer. This is then cached
in a hash table so that it can be shared between vnodes with the same
access permits.
Permit lists can only be shared if they contain the exact same set of
key->CallerAccess mappings.
Note that that table is global rather than being per-net_ns. If the keys
in a permit list cross net_ns boundaries, there is no problem sharing the
cached permits, since the permits are just integer masks.
Since permit lists pin keys, the permit cache also makes it easier for a
future patch to find all occurrences of a key and remove them by means of
setting the afs_permits::invalidated flag and then clearing the appropriate
key pointer. In such an event, memory barriers will need adding.
Lastly, the permit caching is skipped if the server has sent either a
vnode-specific or an entire-server callback since the start of the
operation.
Signed-off-by: David Howells <dhowells@redhat.com>
2017-11-02 22:27:49 +07:00
|
|
|
* Hash a list of permits. Use simple addition to make it easy to add an extra
|
|
|
|
* one at an as-yet indeterminate position in the list.
|
2007-04-27 05:57:07 +07:00
|
|
|
*/
|
afs: Overhaul permit caching
Overhaul permit caching in AFS by making it per-vnode and sharing permit
lists where possible.
When most of the fileserver operations are called, they return a status
structure indicating the (revised) details of the vnode or vnodes involved
in the operation. This includes the access mark derived from the ACL
(named CallerAccess in the protocol definition file). This is cacheable
and if the ACL changes, the server will tell us that it is breaking the
callback promise, at which point we can discard the currently cached
permits.
With this patch, the afs_permits structure has, at the end, an array of
{ key, CallerAccess } elements, sorted by key pointer. This is then cached
in a hash table so that it can be shared between vnodes with the same
access permits.
Permit lists can only be shared if they contain the exact same set of
key->CallerAccess mappings.
Note that that table is global rather than being per-net_ns. If the keys
in a permit list cross net_ns boundaries, there is no problem sharing the
cached permits, since the permits are just integer masks.
Since permit lists pin keys, the permit cache also makes it easier for a
future patch to find all occurrences of a key and remove them by means of
setting the afs_permits::invalidated flag and then clearing the appropriate
key pointer. In such an event, memory barriers will need adding.
Lastly, the permit caching is skipped if the server has sent either a
vnode-specific or an entire-server callback since the start of the
operation.
Signed-off-by: David Howells <dhowells@redhat.com>
2017-11-02 22:27:49 +07:00
|
|
|
static void afs_hash_permits(struct afs_permits *permits)
|
2007-04-27 05:57:07 +07:00
|
|
|
{
|
afs: Overhaul permit caching
Overhaul permit caching in AFS by making it per-vnode and sharing permit
lists where possible.
When most of the fileserver operations are called, they return a status
structure indicating the (revised) details of the vnode or vnodes involved
in the operation. This includes the access mark derived from the ACL
(named CallerAccess in the protocol definition file). This is cacheable
and if the ACL changes, the server will tell us that it is breaking the
callback promise, at which point we can discard the currently cached
permits.
With this patch, the afs_permits structure has, at the end, an array of
{ key, CallerAccess } elements, sorted by key pointer. This is then cached
in a hash table so that it can be shared between vnodes with the same
access permits.
Permit lists can only be shared if they contain the exact same set of
key->CallerAccess mappings.
Note that that table is global rather than being per-net_ns. If the keys
in a permit list cross net_ns boundaries, there is no problem sharing the
cached permits, since the permits are just integer masks.
Since permit lists pin keys, the permit cache also makes it easier for a
future patch to find all occurrences of a key and remove them by means of
setting the afs_permits::invalidated flag and then clearing the appropriate
key pointer. In such an event, memory barriers will need adding.
Lastly, the permit caching is skipped if the server has sent either a
vnode-specific or an entire-server callback since the start of the
operation.
Signed-off-by: David Howells <dhowells@redhat.com>
2017-11-02 22:27:49 +07:00
|
|
|
unsigned long h = permits->nr_permits;
|
|
|
|
int i;
|
2007-04-27 05:57:07 +07:00
|
|
|
|
afs: Overhaul permit caching
Overhaul permit caching in AFS by making it per-vnode and sharing permit
lists where possible.
When most of the fileserver operations are called, they return a status
structure indicating the (revised) details of the vnode or vnodes involved
in the operation. This includes the access mark derived from the ACL
(named CallerAccess in the protocol definition file). This is cacheable
and if the ACL changes, the server will tell us that it is breaking the
callback promise, at which point we can discard the currently cached
permits.
With this patch, the afs_permits structure has, at the end, an array of
{ key, CallerAccess } elements, sorted by key pointer. This is then cached
in a hash table so that it can be shared between vnodes with the same
access permits.
Permit lists can only be shared if they contain the exact same set of
key->CallerAccess mappings.
Note that that table is global rather than being per-net_ns. If the keys
in a permit list cross net_ns boundaries, there is no problem sharing the
cached permits, since the permits are just integer masks.
Since permit lists pin keys, the permit cache also makes it easier for a
future patch to find all occurrences of a key and remove them by means of
setting the afs_permits::invalidated flag and then clearing the appropriate
key pointer. In such an event, memory barriers will need adding.
Lastly, the permit caching is skipped if the server has sent either a
vnode-specific or an entire-server callback since the start of the
operation.
Signed-off-by: David Howells <dhowells@redhat.com>
2017-11-02 22:27:49 +07:00
|
|
|
for (i = 0; i < permits->nr_permits; i++) {
|
|
|
|
h += (unsigned long)permits->permits[i].key / sizeof(void *);
|
|
|
|
h += permits->permits[i].access;
|
|
|
|
}
|
2007-04-27 05:57:07 +07:00
|
|
|
|
afs: Overhaul permit caching
Overhaul permit caching in AFS by making it per-vnode and sharing permit
lists where possible.
When most of the fileserver operations are called, they return a status
structure indicating the (revised) details of the vnode or vnodes involved
in the operation. This includes the access mark derived from the ACL
(named CallerAccess in the protocol definition file). This is cacheable
and if the ACL changes, the server will tell us that it is breaking the
callback promise, at which point we can discard the currently cached
permits.
With this patch, the afs_permits structure has, at the end, an array of
{ key, CallerAccess } elements, sorted by key pointer. This is then cached
in a hash table so that it can be shared between vnodes with the same
access permits.
Permit lists can only be shared if they contain the exact same set of
key->CallerAccess mappings.
Note that that table is global rather than being per-net_ns. If the keys
in a permit list cross net_ns boundaries, there is no problem sharing the
cached permits, since the permits are just integer masks.
Since permit lists pin keys, the permit cache also makes it easier for a
future patch to find all occurrences of a key and remove them by means of
setting the afs_permits::invalidated flag and then clearing the appropriate
key pointer. In such an event, memory barriers will need adding.
Lastly, the permit caching is skipped if the server has sent either a
vnode-specific or an entire-server callback since the start of the
operation.
Signed-off-by: David Howells <dhowells@redhat.com>
2017-11-02 22:27:49 +07:00
|
|
|
permits->h = h;
|
2007-04-27 05:57:07 +07:00
|
|
|
}
|
|
|
|
|
|
|
|
/*
|
afs: Overhaul permit caching
Overhaul permit caching in AFS by making it per-vnode and sharing permit
lists where possible.
When most of the fileserver operations are called, they return a status
structure indicating the (revised) details of the vnode or vnodes involved
in the operation. This includes the access mark derived from the ACL
(named CallerAccess in the protocol definition file). This is cacheable
and if the ACL changes, the server will tell us that it is breaking the
callback promise, at which point we can discard the currently cached
permits.
With this patch, the afs_permits structure has, at the end, an array of
{ key, CallerAccess } elements, sorted by key pointer. This is then cached
in a hash table so that it can be shared between vnodes with the same
access permits.
Permit lists can only be shared if they contain the exact same set of
key->CallerAccess mappings.
Note that that table is global rather than being per-net_ns. If the keys
in a permit list cross net_ns boundaries, there is no problem sharing the
cached permits, since the permits are just integer masks.
Since permit lists pin keys, the permit cache also makes it easier for a
future patch to find all occurrences of a key and remove them by means of
setting the afs_permits::invalidated flag and then clearing the appropriate
key pointer. In such an event, memory barriers will need adding.
Lastly, the permit caching is skipped if the server has sent either a
vnode-specific or an entire-server callback since the start of the
operation.
Signed-off-by: David Howells <dhowells@redhat.com>
2017-11-02 22:27:49 +07:00
|
|
|
* Cache the CallerAccess result obtained from doing a fileserver operation
|
|
|
|
* that returned a vnode status for a particular key. If a callback break
|
|
|
|
* occurs whilst the operation was in progress then we have to ditch the cache
|
|
|
|
* as the ACL *may* have changed.
|
2007-04-27 05:57:07 +07:00
|
|
|
*/
|
afs: Overhaul permit caching
Overhaul permit caching in AFS by making it per-vnode and sharing permit
lists where possible.
When most of the fileserver operations are called, they return a status
structure indicating the (revised) details of the vnode or vnodes involved
in the operation. This includes the access mark derived from the ACL
(named CallerAccess in the protocol definition file). This is cacheable
and if the ACL changes, the server will tell us that it is breaking the
callback promise, at which point we can discard the currently cached
permits.
With this patch, the afs_permits structure has, at the end, an array of
{ key, CallerAccess } elements, sorted by key pointer. This is then cached
in a hash table so that it can be shared between vnodes with the same
access permits.
Permit lists can only be shared if they contain the exact same set of
key->CallerAccess mappings.
Note that that table is global rather than being per-net_ns. If the keys
in a permit list cross net_ns boundaries, there is no problem sharing the
cached permits, since the permits are just integer masks.
Since permit lists pin keys, the permit cache also makes it easier for a
future patch to find all occurrences of a key and remove them by means of
setting the afs_permits::invalidated flag and then clearing the appropriate
key pointer. In such an event, memory barriers will need adding.
Lastly, the permit caching is skipped if the server has sent either a
vnode-specific or an entire-server callback since the start of the
operation.
Signed-off-by: David Howells <dhowells@redhat.com>
2017-11-02 22:27:49 +07:00
|
|
|
void afs_cache_permit(struct afs_vnode *vnode, struct key *key,
|
2019-05-09 21:16:10 +07:00
|
|
|
unsigned int cb_break, struct afs_status_cb *scb)
|
2007-04-27 05:57:07 +07:00
|
|
|
{
|
2017-12-01 18:40:43 +07:00
|
|
|
struct afs_permits *permits, *xpermits, *replacement, *zap, *new = NULL;
|
2019-05-09 21:16:10 +07:00
|
|
|
afs_access_t caller_access = scb->status.caller_access;
|
afs: Overhaul permit caching
Overhaul permit caching in AFS by making it per-vnode and sharing permit
lists where possible.
When most of the fileserver operations are called, they return a status
structure indicating the (revised) details of the vnode or vnodes involved
in the operation. This includes the access mark derived from the ACL
(named CallerAccess in the protocol definition file). This is cacheable
and if the ACL changes, the server will tell us that it is breaking the
callback promise, at which point we can discard the currently cached
permits.
With this patch, the afs_permits structure has, at the end, an array of
{ key, CallerAccess } elements, sorted by key pointer. This is then cached
in a hash table so that it can be shared between vnodes with the same
access permits.
Permit lists can only be shared if they contain the exact same set of
key->CallerAccess mappings.
Note that that table is global rather than being per-net_ns. If the keys
in a permit list cross net_ns boundaries, there is no problem sharing the
cached permits, since the permits are just integer masks.
Since permit lists pin keys, the permit cache also makes it easier for a
future patch to find all occurrences of a key and remove them by means of
setting the afs_permits::invalidated flag and then clearing the appropriate
key pointer. In such an event, memory barriers will need adding.
Lastly, the permit caching is skipped if the server has sent either a
vnode-specific or an entire-server callback since the start of the
operation.
Signed-off-by: David Howells <dhowells@redhat.com>
2017-11-02 22:27:49 +07:00
|
|
|
size_t size = 0;
|
|
|
|
bool changed = false;
|
|
|
|
int i, j;
|
|
|
|
|
2018-10-20 06:57:57 +07:00
|
|
|
_enter("{%llx:%llu},%x,%x",
|
afs: Overhaul permit caching
Overhaul permit caching in AFS by making it per-vnode and sharing permit
lists where possible.
When most of the fileserver operations are called, they return a status
structure indicating the (revised) details of the vnode or vnodes involved
in the operation. This includes the access mark derived from the ACL
(named CallerAccess in the protocol definition file). This is cacheable
and if the ACL changes, the server will tell us that it is breaking the
callback promise, at which point we can discard the currently cached
permits.
With this patch, the afs_permits structure has, at the end, an array of
{ key, CallerAccess } elements, sorted by key pointer. This is then cached
in a hash table so that it can be shared between vnodes with the same
access permits.
Permit lists can only be shared if they contain the exact same set of
key->CallerAccess mappings.
Note that that table is global rather than being per-net_ns. If the keys
in a permit list cross net_ns boundaries, there is no problem sharing the
cached permits, since the permits are just integer masks.
Since permit lists pin keys, the permit cache also makes it easier for a
future patch to find all occurrences of a key and remove them by means of
setting the afs_permits::invalidated flag and then clearing the appropriate
key pointer. In such an event, memory barriers will need adding.
Lastly, the permit caching is skipped if the server has sent either a
vnode-specific or an entire-server callback since the start of the
operation.
Signed-off-by: David Howells <dhowells@redhat.com>
2017-11-02 22:27:49 +07:00
|
|
|
vnode->fid.vid, vnode->fid.vnode, key_serial(key), caller_access);
|
|
|
|
|
|
|
|
rcu_read_lock();
|
|
|
|
|
|
|
|
/* Check for the common case first: We got back the same access as last
|
|
|
|
* time we tried and already have it recorded.
|
|
|
|
*/
|
|
|
|
permits = rcu_dereference(vnode->permit_cache);
|
|
|
|
if (permits) {
|
|
|
|
if (!permits->invalidated) {
|
|
|
|
for (i = 0; i < permits->nr_permits; i++) {
|
|
|
|
if (permits->permits[i].key < key)
|
|
|
|
continue;
|
|
|
|
if (permits->permits[i].key > key)
|
|
|
|
break;
|
|
|
|
if (permits->permits[i].access != caller_access) {
|
|
|
|
changed = true;
|
|
|
|
break;
|
|
|
|
}
|
2007-04-27 05:57:07 +07:00
|
|
|
|
2018-10-20 06:57:59 +07:00
|
|
|
if (afs_cb_is_broken(cb_break, vnode,
|
2019-05-13 22:14:32 +07:00
|
|
|
rcu_dereference(vnode->cb_interest))) {
|
afs: Overhaul permit caching
Overhaul permit caching in AFS by making it per-vnode and sharing permit
lists where possible.
When most of the fileserver operations are called, they return a status
structure indicating the (revised) details of the vnode or vnodes involved
in the operation. This includes the access mark derived from the ACL
(named CallerAccess in the protocol definition file). This is cacheable
and if the ACL changes, the server will tell us that it is breaking the
callback promise, at which point we can discard the currently cached
permits.
With this patch, the afs_permits structure has, at the end, an array of
{ key, CallerAccess } elements, sorted by key pointer. This is then cached
in a hash table so that it can be shared between vnodes with the same
access permits.
Permit lists can only be shared if they contain the exact same set of
key->CallerAccess mappings.
Note that that table is global rather than being per-net_ns. If the keys
in a permit list cross net_ns boundaries, there is no problem sharing the
cached permits, since the permits are just integer masks.
Since permit lists pin keys, the permit cache also makes it easier for a
future patch to find all occurrences of a key and remove them by means of
setting the afs_permits::invalidated flag and then clearing the appropriate
key pointer. In such an event, memory barriers will need adding.
Lastly, the permit caching is skipped if the server has sent either a
vnode-specific or an entire-server callback since the start of the
operation.
Signed-off-by: David Howells <dhowells@redhat.com>
2017-11-02 22:27:49 +07:00
|
|
|
changed = true;
|
|
|
|
break;
|
|
|
|
}
|
2007-04-27 05:57:07 +07:00
|
|
|
|
afs: Overhaul permit caching
Overhaul permit caching in AFS by making it per-vnode and sharing permit
lists where possible.
When most of the fileserver operations are called, they return a status
structure indicating the (revised) details of the vnode or vnodes involved
in the operation. This includes the access mark derived from the ACL
(named CallerAccess in the protocol definition file). This is cacheable
and if the ACL changes, the server will tell us that it is breaking the
callback promise, at which point we can discard the currently cached
permits.
With this patch, the afs_permits structure has, at the end, an array of
{ key, CallerAccess } elements, sorted by key pointer. This is then cached
in a hash table so that it can be shared between vnodes with the same
access permits.
Permit lists can only be shared if they contain the exact same set of
key->CallerAccess mappings.
Note that that table is global rather than being per-net_ns. If the keys
in a permit list cross net_ns boundaries, there is no problem sharing the
cached permits, since the permits are just integer masks.
Since permit lists pin keys, the permit cache also makes it easier for a
future patch to find all occurrences of a key and remove them by means of
setting the afs_permits::invalidated flag and then clearing the appropriate
key pointer. In such an event, memory barriers will need adding.
Lastly, the permit caching is skipped if the server has sent either a
vnode-specific or an entire-server callback since the start of the
operation.
Signed-off-by: David Howells <dhowells@redhat.com>
2017-11-02 22:27:49 +07:00
|
|
|
/* The cache is still good. */
|
|
|
|
rcu_read_unlock();
|
|
|
|
return;
|
|
|
|
}
|
|
|
|
}
|
2007-04-27 05:57:07 +07:00
|
|
|
|
afs: Overhaul permit caching
Overhaul permit caching in AFS by making it per-vnode and sharing permit
lists where possible.
When most of the fileserver operations are called, they return a status
structure indicating the (revised) details of the vnode or vnodes involved
in the operation. This includes the access mark derived from the ACL
(named CallerAccess in the protocol definition file). This is cacheable
and if the ACL changes, the server will tell us that it is breaking the
callback promise, at which point we can discard the currently cached
permits.
With this patch, the afs_permits structure has, at the end, an array of
{ key, CallerAccess } elements, sorted by key pointer. This is then cached
in a hash table so that it can be shared between vnodes with the same
access permits.
Permit lists can only be shared if they contain the exact same set of
key->CallerAccess mappings.
Note that that table is global rather than being per-net_ns. If the keys
in a permit list cross net_ns boundaries, there is no problem sharing the
cached permits, since the permits are just integer masks.
Since permit lists pin keys, the permit cache also makes it easier for a
future patch to find all occurrences of a key and remove them by means of
setting the afs_permits::invalidated flag and then clearing the appropriate
key pointer. In such an event, memory barriers will need adding.
Lastly, the permit caching is skipped if the server has sent either a
vnode-specific or an entire-server callback since the start of the
operation.
Signed-off-by: David Howells <dhowells@redhat.com>
2017-11-02 22:27:49 +07:00
|
|
|
changed |= permits->invalidated;
|
|
|
|
size = permits->nr_permits;
|
2007-04-27 05:57:07 +07:00
|
|
|
|
afs: Overhaul permit caching
Overhaul permit caching in AFS by making it per-vnode and sharing permit
lists where possible.
When most of the fileserver operations are called, they return a status
structure indicating the (revised) details of the vnode or vnodes involved
in the operation. This includes the access mark derived from the ACL
(named CallerAccess in the protocol definition file). This is cacheable
and if the ACL changes, the server will tell us that it is breaking the
callback promise, at which point we can discard the currently cached
permits.
With this patch, the afs_permits structure has, at the end, an array of
{ key, CallerAccess } elements, sorted by key pointer. This is then cached
in a hash table so that it can be shared between vnodes with the same
access permits.
Permit lists can only be shared if they contain the exact same set of
key->CallerAccess mappings.
Note that that table is global rather than being per-net_ns. If the keys
in a permit list cross net_ns boundaries, there is no problem sharing the
cached permits, since the permits are just integer masks.
Since permit lists pin keys, the permit cache also makes it easier for a
future patch to find all occurrences of a key and remove them by means of
setting the afs_permits::invalidated flag and then clearing the appropriate
key pointer. In such an event, memory barriers will need adding.
Lastly, the permit caching is skipped if the server has sent either a
vnode-specific or an entire-server callback since the start of the
operation.
Signed-off-by: David Howells <dhowells@redhat.com>
2017-11-02 22:27:49 +07:00
|
|
|
/* If this set of permits is now wrong, clear the permits
|
|
|
|
* pointer so that no one tries to use the stale information.
|
|
|
|
*/
|
|
|
|
if (changed) {
|
|
|
|
spin_lock(&vnode->lock);
|
|
|
|
if (permits != rcu_access_pointer(vnode->permit_cache))
|
|
|
|
goto someone_else_changed_it_unlock;
|
|
|
|
RCU_INIT_POINTER(vnode->permit_cache, NULL);
|
|
|
|
spin_unlock(&vnode->lock);
|
|
|
|
|
|
|
|
afs_put_permits(permits);
|
|
|
|
permits = NULL;
|
|
|
|
size = 0;
|
|
|
|
}
|
2007-04-27 05:57:07 +07:00
|
|
|
}
|
|
|
|
|
2019-05-13 22:14:32 +07:00
|
|
|
if (afs_cb_is_broken(cb_break, vnode, rcu_dereference(vnode->cb_interest)))
|
afs: Overhaul permit caching
Overhaul permit caching in AFS by making it per-vnode and sharing permit
lists where possible.
When most of the fileserver operations are called, they return a status
structure indicating the (revised) details of the vnode or vnodes involved
in the operation. This includes the access mark derived from the ACL
(named CallerAccess in the protocol definition file). This is cacheable
and if the ACL changes, the server will tell us that it is breaking the
callback promise, at which point we can discard the currently cached
permits.
With this patch, the afs_permits structure has, at the end, an array of
{ key, CallerAccess } elements, sorted by key pointer. This is then cached
in a hash table so that it can be shared between vnodes with the same
access permits.
Permit lists can only be shared if they contain the exact same set of
key->CallerAccess mappings.
Note that that table is global rather than being per-net_ns. If the keys
in a permit list cross net_ns boundaries, there is no problem sharing the
cached permits, since the permits are just integer masks.
Since permit lists pin keys, the permit cache also makes it easier for a
future patch to find all occurrences of a key and remove them by means of
setting the afs_permits::invalidated flag and then clearing the appropriate
key pointer. In such an event, memory barriers will need adding.
Lastly, the permit caching is skipped if the server has sent either a
vnode-specific or an entire-server callback since the start of the
operation.
Signed-off-by: David Howells <dhowells@redhat.com>
2017-11-02 22:27:49 +07:00
|
|
|
goto someone_else_changed_it;
|
2007-04-27 05:57:07 +07:00
|
|
|
|
afs: Overhaul permit caching
Overhaul permit caching in AFS by making it per-vnode and sharing permit
lists where possible.
When most of the fileserver operations are called, they return a status
structure indicating the (revised) details of the vnode or vnodes involved
in the operation. This includes the access mark derived from the ACL
(named CallerAccess in the protocol definition file). This is cacheable
and if the ACL changes, the server will tell us that it is breaking the
callback promise, at which point we can discard the currently cached
permits.
With this patch, the afs_permits structure has, at the end, an array of
{ key, CallerAccess } elements, sorted by key pointer. This is then cached
in a hash table so that it can be shared between vnodes with the same
access permits.
Permit lists can only be shared if they contain the exact same set of
key->CallerAccess mappings.
Note that that table is global rather than being per-net_ns. If the keys
in a permit list cross net_ns boundaries, there is no problem sharing the
cached permits, since the permits are just integer masks.
Since permit lists pin keys, the permit cache also makes it easier for a
future patch to find all occurrences of a key and remove them by means of
setting the afs_permits::invalidated flag and then clearing the appropriate
key pointer. In such an event, memory barriers will need adding.
Lastly, the permit caching is skipped if the server has sent either a
vnode-specific or an entire-server callback since the start of the
operation.
Signed-off-by: David Howells <dhowells@redhat.com>
2017-11-02 22:27:49 +07:00
|
|
|
/* We need a ref on any permits list we want to copy as we'll have to
|
|
|
|
* drop the lock to do memory allocation.
|
|
|
|
*/
|
2018-04-10 03:12:31 +07:00
|
|
|
if (permits && !refcount_inc_not_zero(&permits->usage))
|
afs: Overhaul permit caching
Overhaul permit caching in AFS by making it per-vnode and sharing permit
lists where possible.
When most of the fileserver operations are called, they return a status
structure indicating the (revised) details of the vnode or vnodes involved
in the operation. This includes the access mark derived from the ACL
(named CallerAccess in the protocol definition file). This is cacheable
and if the ACL changes, the server will tell us that it is breaking the
callback promise, at which point we can discard the currently cached
permits.
With this patch, the afs_permits structure has, at the end, an array of
{ key, CallerAccess } elements, sorted by key pointer. This is then cached
in a hash table so that it can be shared between vnodes with the same
access permits.
Permit lists can only be shared if they contain the exact same set of
key->CallerAccess mappings.
Note that that table is global rather than being per-net_ns. If the keys
in a permit list cross net_ns boundaries, there is no problem sharing the
cached permits, since the permits are just integer masks.
Since permit lists pin keys, the permit cache also makes it easier for a
future patch to find all occurrences of a key and remove them by means of
setting the afs_permits::invalidated flag and then clearing the appropriate
key pointer. In such an event, memory barriers will need adding.
Lastly, the permit caching is skipped if the server has sent either a
vnode-specific or an entire-server callback since the start of the
operation.
Signed-off-by: David Howells <dhowells@redhat.com>
2017-11-02 22:27:49 +07:00
|
|
|
goto someone_else_changed_it;
|
|
|
|
|
|
|
|
rcu_read_unlock();
|
|
|
|
|
|
|
|
/* Speculatively create a new list with the revised permission set. We
|
|
|
|
* discard this if we find an extant match already in the hash, but
|
|
|
|
* it's easier to compare with memcmp this way.
|
|
|
|
*
|
|
|
|
* We fill in the key pointers at this time, but we don't get the refs
|
|
|
|
* yet.
|
|
|
|
*/
|
|
|
|
size++;
|
|
|
|
new = kzalloc(sizeof(struct afs_permits) +
|
|
|
|
sizeof(struct afs_permit) * size, GFP_NOFS);
|
|
|
|
if (!new)
|
2017-12-01 18:40:43 +07:00
|
|
|
goto out_put;
|
afs: Overhaul permit caching
Overhaul permit caching in AFS by making it per-vnode and sharing permit
lists where possible.
When most of the fileserver operations are called, they return a status
structure indicating the (revised) details of the vnode or vnodes involved
in the operation. This includes the access mark derived from the ACL
(named CallerAccess in the protocol definition file). This is cacheable
and if the ACL changes, the server will tell us that it is breaking the
callback promise, at which point we can discard the currently cached
permits.
With this patch, the afs_permits structure has, at the end, an array of
{ key, CallerAccess } elements, sorted by key pointer. This is then cached
in a hash table so that it can be shared between vnodes with the same
access permits.
Permit lists can only be shared if they contain the exact same set of
key->CallerAccess mappings.
Note that that table is global rather than being per-net_ns. If the keys
in a permit list cross net_ns boundaries, there is no problem sharing the
cached permits, since the permits are just integer masks.
Since permit lists pin keys, the permit cache also makes it easier for a
future patch to find all occurrences of a key and remove them by means of
setting the afs_permits::invalidated flag and then clearing the appropriate
key pointer. In such an event, memory barriers will need adding.
Lastly, the permit caching is skipped if the server has sent either a
vnode-specific or an entire-server callback since the start of the
operation.
Signed-off-by: David Howells <dhowells@redhat.com>
2017-11-02 22:27:49 +07:00
|
|
|
|
|
|
|
refcount_set(&new->usage, 1);
|
|
|
|
new->nr_permits = size;
|
|
|
|
i = j = 0;
|
|
|
|
if (permits) {
|
|
|
|
for (i = 0; i < permits->nr_permits; i++) {
|
|
|
|
if (j == i && permits->permits[i].key > key) {
|
|
|
|
new->permits[j].key = key;
|
|
|
|
new->permits[j].access = caller_access;
|
|
|
|
j++;
|
2007-04-27 05:57:07 +07:00
|
|
|
}
|
afs: Overhaul permit caching
Overhaul permit caching in AFS by making it per-vnode and sharing permit
lists where possible.
When most of the fileserver operations are called, they return a status
structure indicating the (revised) details of the vnode or vnodes involved
in the operation. This includes the access mark derived from the ACL
(named CallerAccess in the protocol definition file). This is cacheable
and if the ACL changes, the server will tell us that it is breaking the
callback promise, at which point we can discard the currently cached
permits.
With this patch, the afs_permits structure has, at the end, an array of
{ key, CallerAccess } elements, sorted by key pointer. This is then cached
in a hash table so that it can be shared between vnodes with the same
access permits.
Permit lists can only be shared if they contain the exact same set of
key->CallerAccess mappings.
Note that that table is global rather than being per-net_ns. If the keys
in a permit list cross net_ns boundaries, there is no problem sharing the
cached permits, since the permits are just integer masks.
Since permit lists pin keys, the permit cache also makes it easier for a
future patch to find all occurrences of a key and remove them by means of
setting the afs_permits::invalidated flag and then clearing the appropriate
key pointer. In such an event, memory barriers will need adding.
Lastly, the permit caching is skipped if the server has sent either a
vnode-specific or an entire-server callback since the start of the
operation.
Signed-off-by: David Howells <dhowells@redhat.com>
2017-11-02 22:27:49 +07:00
|
|
|
new->permits[j].key = permits->permits[i].key;
|
|
|
|
new->permits[j].access = permits->permits[i].access;
|
|
|
|
j++;
|
2007-04-27 05:57:07 +07:00
|
|
|
}
|
|
|
|
}
|
|
|
|
|
afs: Overhaul permit caching
Overhaul permit caching in AFS by making it per-vnode and sharing permit
lists where possible.
When most of the fileserver operations are called, they return a status
structure indicating the (revised) details of the vnode or vnodes involved
in the operation. This includes the access mark derived from the ACL
(named CallerAccess in the protocol definition file). This is cacheable
and if the ACL changes, the server will tell us that it is breaking the
callback promise, at which point we can discard the currently cached
permits.
With this patch, the afs_permits structure has, at the end, an array of
{ key, CallerAccess } elements, sorted by key pointer. This is then cached
in a hash table so that it can be shared between vnodes with the same
access permits.
Permit lists can only be shared if they contain the exact same set of
key->CallerAccess mappings.
Note that that table is global rather than being per-net_ns. If the keys
in a permit list cross net_ns boundaries, there is no problem sharing the
cached permits, since the permits are just integer masks.
Since permit lists pin keys, the permit cache also makes it easier for a
future patch to find all occurrences of a key and remove them by means of
setting the afs_permits::invalidated flag and then clearing the appropriate
key pointer. In such an event, memory barriers will need adding.
Lastly, the permit caching is skipped if the server has sent either a
vnode-specific or an entire-server callback since the start of the
operation.
Signed-off-by: David Howells <dhowells@redhat.com>
2017-11-02 22:27:49 +07:00
|
|
|
if (j == i) {
|
|
|
|
new->permits[j].key = key;
|
|
|
|
new->permits[j].access = caller_access;
|
|
|
|
}
|
|
|
|
|
|
|
|
afs_hash_permits(new);
|
|
|
|
|
|
|
|
/* Now see if the permit list we want is actually already available */
|
|
|
|
spin_lock(&afs_permits_lock);
|
|
|
|
|
|
|
|
hash_for_each_possible(afs_permits_cache, xpermits, hash_node, new->h) {
|
|
|
|
if (xpermits->h != new->h ||
|
|
|
|
xpermits->invalidated ||
|
|
|
|
xpermits->nr_permits != new->nr_permits ||
|
|
|
|
memcmp(xpermits->permits, new->permits,
|
|
|
|
new->nr_permits * sizeof(struct afs_permit)) != 0)
|
|
|
|
continue;
|
|
|
|
|
|
|
|
if (refcount_inc_not_zero(&xpermits->usage)) {
|
|
|
|
replacement = xpermits;
|
|
|
|
goto found;
|
|
|
|
}
|
|
|
|
|
|
|
|
break;
|
|
|
|
}
|
|
|
|
|
|
|
|
for (i = 0; i < new->nr_permits; i++)
|
|
|
|
key_get(new->permits[i].key);
|
|
|
|
hash_add_rcu(afs_permits_cache, &new->hash_node, new->h);
|
|
|
|
replacement = new;
|
|
|
|
new = NULL;
|
|
|
|
|
|
|
|
found:
|
|
|
|
spin_unlock(&afs_permits_lock);
|
|
|
|
|
|
|
|
kfree(new);
|
|
|
|
|
2019-05-13 22:14:32 +07:00
|
|
|
rcu_read_lock();
|
afs: Overhaul permit caching
Overhaul permit caching in AFS by making it per-vnode and sharing permit
lists where possible.
When most of the fileserver operations are called, they return a status
structure indicating the (revised) details of the vnode or vnodes involved
in the operation. This includes the access mark derived from the ACL
(named CallerAccess in the protocol definition file). This is cacheable
and if the ACL changes, the server will tell us that it is breaking the
callback promise, at which point we can discard the currently cached
permits.
With this patch, the afs_permits structure has, at the end, an array of
{ key, CallerAccess } elements, sorted by key pointer. This is then cached
in a hash table so that it can be shared between vnodes with the same
access permits.
Permit lists can only be shared if they contain the exact same set of
key->CallerAccess mappings.
Note that that table is global rather than being per-net_ns. If the keys
in a permit list cross net_ns boundaries, there is no problem sharing the
cached permits, since the permits are just integer masks.
Since permit lists pin keys, the permit cache also makes it easier for a
future patch to find all occurrences of a key and remove them by means of
setting the afs_permits::invalidated flag and then clearing the appropriate
key pointer. In such an event, memory barriers will need adding.
Lastly, the permit caching is skipped if the server has sent either a
vnode-specific or an entire-server callback since the start of the
operation.
Signed-off-by: David Howells <dhowells@redhat.com>
2017-11-02 22:27:49 +07:00
|
|
|
spin_lock(&vnode->lock);
|
2017-12-01 18:40:43 +07:00
|
|
|
zap = rcu_access_pointer(vnode->permit_cache);
|
2019-05-13 22:14:32 +07:00
|
|
|
if (!afs_cb_is_broken(cb_break, vnode, rcu_dereference(vnode->cb_interest)) &&
|
2017-12-01 18:40:43 +07:00
|
|
|
zap == permits)
|
|
|
|
rcu_assign_pointer(vnode->permit_cache, replacement);
|
|
|
|
else
|
|
|
|
zap = replacement;
|
afs: Overhaul permit caching
Overhaul permit caching in AFS by making it per-vnode and sharing permit
lists where possible.
When most of the fileserver operations are called, they return a status
structure indicating the (revised) details of the vnode or vnodes involved
in the operation. This includes the access mark derived from the ACL
(named CallerAccess in the protocol definition file). This is cacheable
and if the ACL changes, the server will tell us that it is breaking the
callback promise, at which point we can discard the currently cached
permits.
With this patch, the afs_permits structure has, at the end, an array of
{ key, CallerAccess } elements, sorted by key pointer. This is then cached
in a hash table so that it can be shared between vnodes with the same
access permits.
Permit lists can only be shared if they contain the exact same set of
key->CallerAccess mappings.
Note that that table is global rather than being per-net_ns. If the keys
in a permit list cross net_ns boundaries, there is no problem sharing the
cached permits, since the permits are just integer masks.
Since permit lists pin keys, the permit cache also makes it easier for a
future patch to find all occurrences of a key and remove them by means of
setting the afs_permits::invalidated flag and then clearing the appropriate
key pointer. In such an event, memory barriers will need adding.
Lastly, the permit caching is skipped if the server has sent either a
vnode-specific or an entire-server callback since the start of the
operation.
Signed-off-by: David Howells <dhowells@redhat.com>
2017-11-02 22:27:49 +07:00
|
|
|
spin_unlock(&vnode->lock);
|
2019-05-13 22:14:32 +07:00
|
|
|
rcu_read_unlock();
|
2017-12-01 18:40:43 +07:00
|
|
|
afs_put_permits(zap);
|
|
|
|
out_put:
|
afs: Overhaul permit caching
Overhaul permit caching in AFS by making it per-vnode and sharing permit
lists where possible.
When most of the fileserver operations are called, they return a status
structure indicating the (revised) details of the vnode or vnodes involved
in the operation. This includes the access mark derived from the ACL
(named CallerAccess in the protocol definition file). This is cacheable
and if the ACL changes, the server will tell us that it is breaking the
callback promise, at which point we can discard the currently cached
permits.
With this patch, the afs_permits structure has, at the end, an array of
{ key, CallerAccess } elements, sorted by key pointer. This is then cached
in a hash table so that it can be shared between vnodes with the same
access permits.
Permit lists can only be shared if they contain the exact same set of
key->CallerAccess mappings.
Note that that table is global rather than being per-net_ns. If the keys
in a permit list cross net_ns boundaries, there is no problem sharing the
cached permits, since the permits are just integer masks.
Since permit lists pin keys, the permit cache also makes it easier for a
future patch to find all occurrences of a key and remove them by means of
setting the afs_permits::invalidated flag and then clearing the appropriate
key pointer. In such an event, memory barriers will need adding.
Lastly, the permit caching is skipped if the server has sent either a
vnode-specific or an entire-server callback since the start of the
operation.
Signed-off-by: David Howells <dhowells@redhat.com>
2017-11-02 22:27:49 +07:00
|
|
|
afs_put_permits(permits);
|
|
|
|
return;
|
|
|
|
|
|
|
|
someone_else_changed_it_unlock:
|
|
|
|
spin_unlock(&vnode->lock);
|
|
|
|
someone_else_changed_it:
|
|
|
|
/* Someone else changed the cache under us - don't recheck at this
|
|
|
|
* time.
|
|
|
|
*/
|
2018-04-10 03:12:31 +07:00
|
|
|
rcu_read_unlock();
|
afs: Overhaul permit caching
Overhaul permit caching in AFS by making it per-vnode and sharing permit
lists where possible.
When most of the fileserver operations are called, they return a status
structure indicating the (revised) details of the vnode or vnodes involved
in the operation. This includes the access mark derived from the ACL
(named CallerAccess in the protocol definition file). This is cacheable
and if the ACL changes, the server will tell us that it is breaking the
callback promise, at which point we can discard the currently cached
permits.
With this patch, the afs_permits structure has, at the end, an array of
{ key, CallerAccess } elements, sorted by key pointer. This is then cached
in a hash table so that it can be shared between vnodes with the same
access permits.
Permit lists can only be shared if they contain the exact same set of
key->CallerAccess mappings.
Note that that table is global rather than being per-net_ns. If the keys
in a permit list cross net_ns boundaries, there is no problem sharing the
cached permits, since the permits are just integer masks.
Since permit lists pin keys, the permit cache also makes it easier for a
future patch to find all occurrences of a key and remove them by means of
setting the afs_permits::invalidated flag and then clearing the appropriate
key pointer. In such an event, memory barriers will need adding.
Lastly, the permit caching is skipped if the server has sent either a
vnode-specific or an entire-server callback since the start of the
operation.
Signed-off-by: David Howells <dhowells@redhat.com>
2017-11-02 22:27:49 +07:00
|
|
|
return;
|
2007-04-27 05:57:07 +07:00
|
|
|
}
|
|
|
|
|
|
|
|
/*
|
|
|
|
* check with the fileserver to see if the directory or parent directory is
|
|
|
|
* permitted to be accessed with this authorisation, and if so, what access it
|
|
|
|
* is granted
|
|
|
|
*/
|
2017-11-13 23:59:50 +07:00
|
|
|
int afs_check_permit(struct afs_vnode *vnode, struct key *key,
|
|
|
|
afs_access_t *_access)
|
2007-04-27 05:57:07 +07:00
|
|
|
{
|
|
|
|
struct afs_permits *permits;
|
afs: Overhaul permit caching
Overhaul permit caching in AFS by making it per-vnode and sharing permit
lists where possible.
When most of the fileserver operations are called, they return a status
structure indicating the (revised) details of the vnode or vnodes involved
in the operation. This includes the access mark derived from the ACL
(named CallerAccess in the protocol definition file). This is cacheable
and if the ACL changes, the server will tell us that it is breaking the
callback promise, at which point we can discard the currently cached
permits.
With this patch, the afs_permits structure has, at the end, an array of
{ key, CallerAccess } elements, sorted by key pointer. This is then cached
in a hash table so that it can be shared between vnodes with the same
access permits.
Permit lists can only be shared if they contain the exact same set of
key->CallerAccess mappings.
Note that that table is global rather than being per-net_ns. If the keys
in a permit list cross net_ns boundaries, there is no problem sharing the
cached permits, since the permits are just integer masks.
Since permit lists pin keys, the permit cache also makes it easier for a
future patch to find all occurrences of a key and remove them by means of
setting the afs_permits::invalidated flag and then clearing the appropriate
key pointer. In such an event, memory barriers will need adding.
Lastly, the permit caching is skipped if the server has sent either a
vnode-specific or an entire-server callback since the start of the
operation.
Signed-off-by: David Howells <dhowells@redhat.com>
2017-11-02 22:27:49 +07:00
|
|
|
bool valid = false;
|
|
|
|
int i, ret;
|
2007-04-27 05:57:07 +07:00
|
|
|
|
2018-10-20 06:57:57 +07:00
|
|
|
_enter("{%llx:%llu},%x",
|
2007-05-09 16:33:45 +07:00
|
|
|
vnode->fid.vid, vnode->fid.vnode, key_serial(key));
|
2007-04-27 05:57:07 +07:00
|
|
|
|
|
|
|
/* check the permits to see if we've got one yet */
|
afs: Overhaul permit caching
Overhaul permit caching in AFS by making it per-vnode and sharing permit
lists where possible.
When most of the fileserver operations are called, they return a status
structure indicating the (revised) details of the vnode or vnodes involved
in the operation. This includes the access mark derived from the ACL
(named CallerAccess in the protocol definition file). This is cacheable
and if the ACL changes, the server will tell us that it is breaking the
callback promise, at which point we can discard the currently cached
permits.
With this patch, the afs_permits structure has, at the end, an array of
{ key, CallerAccess } elements, sorted by key pointer. This is then cached
in a hash table so that it can be shared between vnodes with the same
access permits.
Permit lists can only be shared if they contain the exact same set of
key->CallerAccess mappings.
Note that that table is global rather than being per-net_ns. If the keys
in a permit list cross net_ns boundaries, there is no problem sharing the
cached permits, since the permits are just integer masks.
Since permit lists pin keys, the permit cache also makes it easier for a
future patch to find all occurrences of a key and remove them by means of
setting the afs_permits::invalidated flag and then clearing the appropriate
key pointer. In such an event, memory barriers will need adding.
Lastly, the permit caching is skipped if the server has sent either a
vnode-specific or an entire-server callback since the start of the
operation.
Signed-off-by: David Howells <dhowells@redhat.com>
2017-11-02 22:27:49 +07:00
|
|
|
if (key == vnode->volume->cell->anonymous_key) {
|
2007-04-27 05:57:07 +07:00
|
|
|
_debug("anon");
|
afs: Overhaul permit caching
Overhaul permit caching in AFS by making it per-vnode and sharing permit
lists where possible.
When most of the fileserver operations are called, they return a status
structure indicating the (revised) details of the vnode or vnodes involved
in the operation. This includes the access mark derived from the ACL
(named CallerAccess in the protocol definition file). This is cacheable
and if the ACL changes, the server will tell us that it is breaking the
callback promise, at which point we can discard the currently cached
permits.
With this patch, the afs_permits structure has, at the end, an array of
{ key, CallerAccess } elements, sorted by key pointer. This is then cached
in a hash table so that it can be shared between vnodes with the same
access permits.
Permit lists can only be shared if they contain the exact same set of
key->CallerAccess mappings.
Note that that table is global rather than being per-net_ns. If the keys
in a permit list cross net_ns boundaries, there is no problem sharing the
cached permits, since the permits are just integer masks.
Since permit lists pin keys, the permit cache also makes it easier for a
future patch to find all occurrences of a key and remove them by means of
setting the afs_permits::invalidated flag and then clearing the appropriate
key pointer. In such an event, memory barriers will need adding.
Lastly, the permit caching is skipped if the server has sent either a
vnode-specific or an entire-server callback since the start of the
operation.
Signed-off-by: David Howells <dhowells@redhat.com>
2017-11-02 22:27:49 +07:00
|
|
|
*_access = vnode->status.anon_access;
|
2007-04-27 05:57:07 +07:00
|
|
|
valid = true;
|
|
|
|
} else {
|
|
|
|
rcu_read_lock();
|
afs: Overhaul permit caching
Overhaul permit caching in AFS by making it per-vnode and sharing permit
lists where possible.
When most of the fileserver operations are called, they return a status
structure indicating the (revised) details of the vnode or vnodes involved
in the operation. This includes the access mark derived from the ACL
(named CallerAccess in the protocol definition file). This is cacheable
and if the ACL changes, the server will tell us that it is breaking the
callback promise, at which point we can discard the currently cached
permits.
With this patch, the afs_permits structure has, at the end, an array of
{ key, CallerAccess } elements, sorted by key pointer. This is then cached
in a hash table so that it can be shared between vnodes with the same
access permits.
Permit lists can only be shared if they contain the exact same set of
key->CallerAccess mappings.
Note that that table is global rather than being per-net_ns. If the keys
in a permit list cross net_ns boundaries, there is no problem sharing the
cached permits, since the permits are just integer masks.
Since permit lists pin keys, the permit cache also makes it easier for a
future patch to find all occurrences of a key and remove them by means of
setting the afs_permits::invalidated flag and then clearing the appropriate
key pointer. In such an event, memory barriers will need adding.
Lastly, the permit caching is skipped if the server has sent either a
vnode-specific or an entire-server callback since the start of the
operation.
Signed-off-by: David Howells <dhowells@redhat.com>
2017-11-02 22:27:49 +07:00
|
|
|
permits = rcu_dereference(vnode->permit_cache);
|
2007-04-27 05:57:07 +07:00
|
|
|
if (permits) {
|
afs: Overhaul permit caching
Overhaul permit caching in AFS by making it per-vnode and sharing permit
lists where possible.
When most of the fileserver operations are called, they return a status
structure indicating the (revised) details of the vnode or vnodes involved
in the operation. This includes the access mark derived from the ACL
(named CallerAccess in the protocol definition file). This is cacheable
and if the ACL changes, the server will tell us that it is breaking the
callback promise, at which point we can discard the currently cached
permits.
With this patch, the afs_permits structure has, at the end, an array of
{ key, CallerAccess } elements, sorted by key pointer. This is then cached
in a hash table so that it can be shared between vnodes with the same
access permits.
Permit lists can only be shared if they contain the exact same set of
key->CallerAccess mappings.
Note that that table is global rather than being per-net_ns. If the keys
in a permit list cross net_ns boundaries, there is no problem sharing the
cached permits, since the permits are just integer masks.
Since permit lists pin keys, the permit cache also makes it easier for a
future patch to find all occurrences of a key and remove them by means of
setting the afs_permits::invalidated flag and then clearing the appropriate
key pointer. In such an event, memory barriers will need adding.
Lastly, the permit caching is skipped if the server has sent either a
vnode-specific or an entire-server callback since the start of the
operation.
Signed-off-by: David Howells <dhowells@redhat.com>
2017-11-02 22:27:49 +07:00
|
|
|
for (i = 0; i < permits->nr_permits; i++) {
|
|
|
|
if (permits->permits[i].key < key)
|
|
|
|
continue;
|
|
|
|
if (permits->permits[i].key > key)
|
2007-04-27 05:57:07 +07:00
|
|
|
break;
|
afs: Overhaul permit caching
Overhaul permit caching in AFS by making it per-vnode and sharing permit
lists where possible.
When most of the fileserver operations are called, they return a status
structure indicating the (revised) details of the vnode or vnodes involved
in the operation. This includes the access mark derived from the ACL
(named CallerAccess in the protocol definition file). This is cacheable
and if the ACL changes, the server will tell us that it is breaking the
callback promise, at which point we can discard the currently cached
permits.
With this patch, the afs_permits structure has, at the end, an array of
{ key, CallerAccess } elements, sorted by key pointer. This is then cached
in a hash table so that it can be shared between vnodes with the same
access permits.
Permit lists can only be shared if they contain the exact same set of
key->CallerAccess mappings.
Note that that table is global rather than being per-net_ns. If the keys
in a permit list cross net_ns boundaries, there is no problem sharing the
cached permits, since the permits are just integer masks.
Since permit lists pin keys, the permit cache also makes it easier for a
future patch to find all occurrences of a key and remove them by means of
setting the afs_permits::invalidated flag and then clearing the appropriate
key pointer. In such an event, memory barriers will need adding.
Lastly, the permit caching is skipped if the server has sent either a
vnode-specific or an entire-server callback since the start of the
operation.
Signed-off-by: David Howells <dhowells@redhat.com>
2017-11-02 22:27:49 +07:00
|
|
|
|
|
|
|
*_access = permits->permits[i].access;
|
|
|
|
valid = !permits->invalidated;
|
|
|
|
break;
|
2007-04-27 05:57:07 +07:00
|
|
|
}
|
|
|
|
}
|
|
|
|
rcu_read_unlock();
|
|
|
|
}
|
|
|
|
|
|
|
|
if (!valid) {
|
afs: Overhaul permit caching
Overhaul permit caching in AFS by making it per-vnode and sharing permit
lists where possible.
When most of the fileserver operations are called, they return a status
structure indicating the (revised) details of the vnode or vnodes involved
in the operation. This includes the access mark derived from the ACL
(named CallerAccess in the protocol definition file). This is cacheable
and if the ACL changes, the server will tell us that it is breaking the
callback promise, at which point we can discard the currently cached
permits.
With this patch, the afs_permits structure has, at the end, an array of
{ key, CallerAccess } elements, sorted by key pointer. This is then cached
in a hash table so that it can be shared between vnodes with the same
access permits.
Permit lists can only be shared if they contain the exact same set of
key->CallerAccess mappings.
Note that that table is global rather than being per-net_ns. If the keys
in a permit list cross net_ns boundaries, there is no problem sharing the
cached permits, since the permits are just integer masks.
Since permit lists pin keys, the permit cache also makes it easier for a
future patch to find all occurrences of a key and remove them by means of
setting the afs_permits::invalidated flag and then clearing the appropriate
key pointer. In such an event, memory barriers will need adding.
Lastly, the permit caching is skipped if the server has sent either a
vnode-specific or an entire-server callback since the start of the
operation.
Signed-off-by: David Howells <dhowells@redhat.com>
2017-11-02 22:27:49 +07:00
|
|
|
/* Check the status on the file we're actually interested in
|
|
|
|
* (the post-processing will cache the result).
|
|
|
|
*/
|
2007-04-27 05:57:07 +07:00
|
|
|
_debug("no valid permit");
|
|
|
|
|
2019-05-09 21:16:10 +07:00
|
|
|
ret = afs_fetch_status(vnode, key, false, _access);
|
2007-04-27 05:57:07 +07:00
|
|
|
if (ret < 0) {
|
|
|
|
*_access = 0;
|
|
|
|
_leave(" = %d", ret);
|
|
|
|
return ret;
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
|
|
|
_leave(" = 0 [access %x]", *_access);
|
|
|
|
return 0;
|
|
|
|
}
|
|
|
|
|
|
|
|
/*
|
|
|
|
* check the permissions on an AFS file
|
|
|
|
* - AFS ACLs are attached to directories only, and a file is controlled by its
|
|
|
|
* parent directory's ACL
|
|
|
|
*/
|
2011-06-21 06:28:19 +07:00
|
|
|
int afs_permission(struct inode *inode, int mask)
|
2007-04-27 05:57:07 +07:00
|
|
|
{
|
|
|
|
struct afs_vnode *vnode = AFS_FS_I(inode);
|
2008-02-08 19:20:53 +07:00
|
|
|
afs_access_t uninitialized_var(access);
|
2007-04-27 05:57:07 +07:00
|
|
|
struct key *key;
|
|
|
|
int ret;
|
|
|
|
|
2011-06-21 06:28:19 +07:00
|
|
|
if (mask & MAY_NOT_BLOCK)
|
2011-01-07 13:49:58 +07:00
|
|
|
return -ECHILD;
|
|
|
|
|
2018-10-20 06:57:57 +07:00
|
|
|
_enter("{{%llx:%llu},%lx},%x,",
|
2007-04-27 05:59:35 +07:00
|
|
|
vnode->fid.vid, vnode->fid.vnode, vnode->flags, mask);
|
2007-04-27 05:57:07 +07:00
|
|
|
|
|
|
|
key = afs_request_key(vnode->volume->cell);
|
|
|
|
if (IS_ERR(key)) {
|
|
|
|
_leave(" = %ld [key]", PTR_ERR(key));
|
|
|
|
return PTR_ERR(key);
|
|
|
|
}
|
|
|
|
|
2017-11-02 22:27:49 +07:00
|
|
|
ret = afs_validate(vnode, key);
|
|
|
|
if (ret < 0)
|
|
|
|
goto error;
|
2007-04-27 05:59:35 +07:00
|
|
|
|
2007-04-27 05:57:07 +07:00
|
|
|
/* check the permits to see if we've got one yet */
|
|
|
|
ret = afs_check_permit(vnode, key, &access);
|
2007-04-27 05:59:35 +07:00
|
|
|
if (ret < 0)
|
|
|
|
goto error;
|
2007-04-27 05:57:07 +07:00
|
|
|
|
|
|
|
/* interpret the access mask */
|
|
|
|
_debug("REQ %x ACC %x on %s",
|
|
|
|
mask, access, S_ISDIR(inode->i_mode) ? "dir" : "file");
|
|
|
|
|
|
|
|
if (S_ISDIR(inode->i_mode)) {
|
2018-05-17 03:25:46 +07:00
|
|
|
if (mask & (MAY_EXEC | MAY_READ | MAY_CHDIR)) {
|
2007-04-27 05:57:07 +07:00
|
|
|
if (!(access & AFS_ACE_LOOKUP))
|
|
|
|
goto permission_denied;
|
2018-05-17 03:25:46 +07:00
|
|
|
}
|
|
|
|
if (mask & MAY_WRITE) {
|
2007-04-27 05:57:07 +07:00
|
|
|
if (!(access & (AFS_ACE_DELETE | /* rmdir, unlink, rename from */
|
2017-07-06 21:50:18 +07:00
|
|
|
AFS_ACE_INSERT))) /* create, mkdir, symlink, rename to */
|
2007-04-27 05:57:07 +07:00
|
|
|
goto permission_denied;
|
|
|
|
}
|
|
|
|
} else {
|
|
|
|
if (!(access & AFS_ACE_LOOKUP))
|
|
|
|
goto permission_denied;
|
2017-03-16 23:27:44 +07:00
|
|
|
if ((mask & MAY_EXEC) && !(inode->i_mode & S_IXUSR))
|
|
|
|
goto permission_denied;
|
2007-04-27 05:57:07 +07:00
|
|
|
if (mask & (MAY_EXEC | MAY_READ)) {
|
|
|
|
if (!(access & AFS_ACE_READ))
|
|
|
|
goto permission_denied;
|
2017-03-16 23:27:44 +07:00
|
|
|
if (!(inode->i_mode & S_IRUSR))
|
|
|
|
goto permission_denied;
|
2007-04-27 05:57:07 +07:00
|
|
|
} else if (mask & MAY_WRITE) {
|
|
|
|
if (!(access & AFS_ACE_WRITE))
|
|
|
|
goto permission_denied;
|
2017-03-16 23:27:44 +07:00
|
|
|
if (!(inode->i_mode & S_IWUSR))
|
|
|
|
goto permission_denied;
|
2007-04-27 05:57:07 +07:00
|
|
|
}
|
|
|
|
}
|
|
|
|
|
|
|
|
key_put(key);
|
2007-04-27 05:59:35 +07:00
|
|
|
_leave(" = %d", ret);
|
|
|
|
return ret;
|
2007-04-27 05:57:07 +07:00
|
|
|
|
|
|
|
permission_denied:
|
2007-04-27 05:59:35 +07:00
|
|
|
ret = -EACCES;
|
|
|
|
error:
|
2007-04-27 05:57:07 +07:00
|
|
|
key_put(key);
|
2007-04-27 05:59:35 +07:00
|
|
|
_leave(" = %d", ret);
|
|
|
|
return ret;
|
2007-04-27 05:57:07 +07:00
|
|
|
}
|
afs: Overhaul permit caching
Overhaul permit caching in AFS by making it per-vnode and sharing permit
lists where possible.
When most of the fileserver operations are called, they return a status
structure indicating the (revised) details of the vnode or vnodes involved
in the operation. This includes the access mark derived from the ACL
(named CallerAccess in the protocol definition file). This is cacheable
and if the ACL changes, the server will tell us that it is breaking the
callback promise, at which point we can discard the currently cached
permits.
With this patch, the afs_permits structure has, at the end, an array of
{ key, CallerAccess } elements, sorted by key pointer. This is then cached
in a hash table so that it can be shared between vnodes with the same
access permits.
Permit lists can only be shared if they contain the exact same set of
key->CallerAccess mappings.
Note that that table is global rather than being per-net_ns. If the keys
in a permit list cross net_ns boundaries, there is no problem sharing the
cached permits, since the permits are just integer masks.
Since permit lists pin keys, the permit cache also makes it easier for a
future patch to find all occurrences of a key and remove them by means of
setting the afs_permits::invalidated flag and then clearing the appropriate
key pointer. In such an event, memory barriers will need adding.
Lastly, the permit caching is skipped if the server has sent either a
vnode-specific or an entire-server callback since the start of the
operation.
Signed-off-by: David Howells <dhowells@redhat.com>
2017-11-02 22:27:49 +07:00
|
|
|
|
|
|
|
void __exit afs_clean_up_permit_cache(void)
|
|
|
|
{
|
|
|
|
int i;
|
|
|
|
|
|
|
|
for (i = 0; i < HASH_SIZE(afs_permits_cache); i++)
|
|
|
|
WARN_ON_ONCE(!hlist_empty(&afs_permits_cache[i]));
|
|
|
|
|
|
|
|
}
|