This patch changes rhashtable_shrink to shrink to the smallest
size possible rather than halving the table. This is needed
because with multiple rehashing we will defer shrinking until
all other rehashing is done, meaning that when we do shrink
we may be able to shrink a lot.
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
Acked-by: Thomas Graf <tgraf@suug.ch>
Signed-off-by: David S. Miller <davem@davemloft.net>
Since every current rhashtable user uses jhash as their hash
function, the fact that jhash is an inline function causes each
user to generate a copy of its code.
This function provides a solution to this problem by allowing
hashfn to be unset. In which case rhashtable will automatically
set it to jhash. Furthermore, if the key length is a multiple
of 4, we will switch over to jhash2.
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
Acked-by: Thomas Graf <tgraf@suug.ch>
Signed-off-by: David S. Miller <davem@davemloft.net>
The walker is a lockless reader so it too needs an smp_rmb before
reading the future_tbl field in order to see any new tables that
may contain elements that we should have walked over.
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
Acked-by: Thomas Graf <tgraf@suug.ch>
Signed-off-by: David S. Miller <davem@davemloft.net>
Now that all rhashtable users have been converted over to the
inline interface, this patch removes the unused out-of-line
interface.
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
Signed-off-by: David S. Miller <davem@davemloft.net>
This patch deals with the complaint that we make indirect function
calls on the fast paths unnecessarily in rhashtable. We resolve
it by moving the fast paths into inline functions that take struct
rhashtable_param (which obviously must be the same set of parameters
supplied to rhashtable_init) as an argument.
The only remaining indirect call is to obj_hashfn (or key_hashfn it
obj_hashfn is unset) on the rehash as well as the insert-during-
rehash slow path.
This patch also extends the support of vairable-length keys to
include those where the key is fixed but scattered in the object.
For example, in netlink we want to key off the namespace and the
portid but they're not next to each other.
This patch does this by directly using the object hash function
as the indicator of whether the key is accessible or not. It
also adds a new function obj_cmpfn to compare a key against an
object. This means that the caller no longer needs to supply
explicit compare functions.
All this is done in a backwards compatible manner so no existing
users are affected until they convert to the new interface.
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
Signed-off-by: David S. Miller <davem@davemloft.net>
This patch marks the rhashtable_init params argument const as
there is no reason to modify it since we will always make a copy
of it in the rhashtable.
This patch also fixes a bug where we don't actually round up the
value of min_size unless it is less than HASH_MIN_SIZE.
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
Acked-by: Thomas Graf <tgraf@suug.ch>
Signed-off-by: David S. Miller <davem@davemloft.net>
Round up min_size respectively round down max_size to the next power
of two to make sure we always respect the limit specified by the
user. This is required because we compare the table size against the
limit before we expand or shrink.
Also fixes a minor bug where we modified min_size in the params
provided instead of the copy stored in struct rhashtable.
Signed-off-by: Thomas Graf <tgraf@suug.ch>
Acked-by: Herbert Xu <herbert@gondor.apana.org.au>
Signed-off-by: David S. Miller <davem@davemloft.net>
Now that nobody uses max_shift and min_shift, we can safely remove
them.
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
Signed-off-by: David S. Miller <davem@davemloft.net>
This patch adds the parameters max_size and min_size which are
meant to replace max_shift and min_shift.
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
Signed-off-by: David S. Miller <davem@davemloft.net>
Keeping both size and shift is silly. We only need one.
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
Signed-off-by: David S. Miller <davem@davemloft.net>
Caching the lock pointer avoids having to hash on the object
again to unlock the bucket locks.
Signed-off-by: Thomas Graf <tgraf@suug.ch>
Signed-off-by: David S. Miller <davem@davemloft.net>
Fixes the following sparse warnings:
lib/rhashtable.c:767:5: warning: context imbalance in 'rhashtable_walk_start' - wrong count at exit
lib/rhashtable.c:849:6: warning: context imbalance in 'rhashtable_walk_stop' - unexpected unlock
Fixes: f2dba9c6ff ("rhashtable: Introduce rhashtable_walk_*")
Signed-off-by: Thomas Graf <tgraf@suug.ch>
Signed-off-by: David S. Miller <davem@davemloft.net>
The commit 9d901bc051 ("rhashtable:
Free bucket tables asynchronously after rehash") causes gratuitous
failures in rhashtable_remove.
The reason is that it inadvertently introduced multiple rehashing
from the perspective of readers. IOW it is now possible to see
more than two tables during a single RCU critical section.
Fortunately the other reader rhashtable_lookup already deals with
this correctly thanks to c4db8848af
("rhashtable: rhashtable: Move future_tbl into struct bucket_table")
so only rhashtable_remove is broken by this change.
This patch fixes this by looping over every table from the first
one to the last or until we find the element that we were trying
to delete.
Incidentally the simple test for detecting rehashing to prevent
starting another shrinking no longer works. Since it isn't needed
anyway (the work queue and the mutex serves as a natural barrier
to unnecessary rehashes) I've simply killed the test.
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
Signed-off-by: David S. Miller <davem@davemloft.net>
The commit c4db8848af ("rhashtable:
Move future_tbl into struct bucket_table") introduced a use-after-
free bug in rhashtable_walk_stop because it dereferences tbl after
droping the RCU read lock.
This patch fixes it by moving the RCU read unlock down to the bottom
of rhashtable_walk_stop. In fact this was how I had it originally
but it got dropped while rearranging patches because this one
depended on the async freeing of bucket_table.
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
Signed-off-by: David S. Miller <davem@davemloft.net>
This patch moves future_tbl to open up the possibility of having
multiple rehashes on the same table.
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
Signed-off-by: David S. Miller <davem@davemloft.net>
This patch adds a rehash counter to bucket_table to indicate
the last bucket that has been rehashed. This serves two purposes:
1. Any bucket that has been rehashed can never gain a new object.
2. If the rehash counter reaches the size of the table, the table
will forever remain empty.
This patch also downsizes bucket_table->size to an unsigned int
since we do not support sizes greater than 32 bits yet.
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
Signed-off-by: David S. Miller <davem@davemloft.net>
There is in fact no need to wait for an RCU grace period in the
rehash function, since all insertions are guaranteed to go into
the new table through spin locks.
This patch uses call_rcu to free the old/rehashed table at our
leisure.
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
Signed-off-by: David S. Miller <davem@davemloft.net>
It seems that I have already made every rehash redo the random
seed even though my commit message indicated otherwise :)
Since we have already taken that step, this patch goes one step
further and moves the seed initialisation into bucket_table_alloc.
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
Signed-off-by: David S. Miller <davem@davemloft.net>
We only nest one level deep there is no need to roll our own
subclasses.
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
Signed-off-by: David S. Miller <davem@davemloft.net>
Previously whenever the walker encountered a resize it simply
snaps back to the beginning and starts again. However, this only
works if the rehash started and completed while the walker was
idle.
If the walker attempts to restart while the rehash is still ongoing,
we may miss objects that we shouldn't have.
This patch fixes this by making the walker walk the old table
followed by the new table just like all other readers. If a
rehash is detected we will still signal our caller of the fact
so they can prepare for duplicates but we will simply continue
the walk onto the new table after the old one is finished either
by us or by the rehasher.
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
Signed-off-by: David S. Miller <davem@davemloft.net>
This patch fixes a typo rhashtable_lookup_compare where we fail
to recompute the hash when looking up the new table. This causes
elements to be missed and potentially a crash during a resize.
Reported-by: Thomas Graf <tgraf@suug.ch>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
Signed-off-by: David S. Miller <davem@davemloft.net>
Commit c0c09bfdc4 ("rhashtable: avoid unnecessary wakeup for worker
queue") changed ht->shift to be atomic, which is actually unnecessary.
Instead of leaving the current shift in the core rhashtable structure,
it can be cached inside the individual bucket tables.
There, it will only be initialized once during a new table allocation
in the shrink/expansion slow path, and from then onward it stays immutable
for the rest of the bucket table liftime.
That allows shift to be non-atomic. The patch also moves hash_rnd
management into the table setup. The rhashtable structure now consumes
3 instead of 4 cachelines.
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
Cc: Ying Xue <ying.xue@windriver.com>
Acked-by: Thomas Graf <tgraf@suug.ch>
Signed-off-by: David S. Miller <davem@davemloft.net>
There is a potential race condition between readers and the rehasher.
In particular, the rehasher could have started a rehash while the
reader finishes a scan of the old table but fails to see the new
table pointer.
This patch closes this window by adding smp_wmb/smp_rmb.
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
Signed-off-by: David S. Miller <davem@davemloft.net>
Now that the only caller of obj_raw_hashfn is head_hashfn, we can
simply kill it and fold it into the latter.
This patch also moves the common shift from head_hashfn/key_hashfn
into rht_bucket_index.
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
Acked-by: Thomas Graf <tgraf@suug.ch>
Signed-off-by: David S. Miller <davem@davemloft.net>
key_hashfn has only one caller and it doesn't really need to supply
the key length as an extra parameter.
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
Acked-by: Thomas Graf <tgraf@suug.ch>
Signed-off-by: David S. Miller <davem@davemloft.net>
Now that we don't have cross-table hashes, we no longer need to
keep the entire hash value so all users of obj_raw_hashfn can
use head_hashfn instead.
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
Acked-by: Thomas Graf <tgraf@suug.ch>
Signed-off-by: David S. Miller <davem@davemloft.net>
This patch reverts commit c88455ce50
("rhashtable: key_hashfn() must return full hash value") because
the only user of it always masks the hash value.
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
Acked-by: Thomas Graf <tgraf@suug.ch>
Signed-off-by: David S. Miller <davem@davemloft.net>
Commit aa34a6cb04 ("rhashtable:
Add arbitrary rehash function") killed the annotation on the
nested lock which leads to bitching from lockdep.
Reported-by: Fengguang Wu <fengguang.wu@intel.com>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
Signed-off-by: David S. Miller <davem@davemloft.net>
This patch adds a rehash function that supports the use of any
hash function for the new table. This is needed to support changing
the random seed value during the lifetime of the hash table.
However for now the random seed value is still constant and the
rehash function is simply used to replace the existing expand/shrink
functions.
[ ASSERT_BUCKET_LOCK() and thus debug_dump_table() +
debug_dump_buckets() are not longer used, so delete them
entirely. -DaveM ]
Signed-off-by: Herbert Xu <herbert.xu@redhat.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Currently hash_rnd is a parameter that users can set. However,
no existing users set this parameter. It is also something that
people are unlikely to want to set directly since it's just a
random number.
In preparation for allowing the reseeding/rehashing of rhashtable,
this patch moves hash_rnd into bucket_table so that it's now an
internal state rather than a parameter.
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
Acked-by: Thomas Graf <tgraf@suug.ch>
Signed-off-by: David S. Miller <davem@davemloft.net>
If a hash table has 128 slots and 16384 elems, expand to 256 slots
takes more than one second. For larger sets, a soft lockup is detected.
Holding cpu for that long, even in a work queue is a show stopper
for non preemptable kernels.
cond_resched() at strategic points to allow process scheduler
to reschedule us.
Signed-off-by: Eric Dumazet <edumazet@google.com>
Acked-by: Daniel Borkmann <daniel@iogearbox.net>
Signed-off-by: David S. Miller <davem@davemloft.net>
Currently, all real users of rhashtable default their grow and shrink
decision functions to rht_grow_above_75() and rht_shrink_below_30(),
so that there's currently no need to have this explicitly selectable.
It can/should be generic and private inside rhashtable until a real
use case pops up. Since we can make this private, we'll save us this
additional indirection layer and can improve insertion/deletion time
as well.
Reference: http://patchwork.ozlabs.org/patch/443040/
Suggested-by: David S. Miller <davem@davemloft.net>
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
Acked-by: Thomas Graf <tgraf@suug.ch>
Signed-off-by: David S. Miller <davem@davemloft.net>
While commit c0c09bfdc4 ("rhashtable: avoid unnecessary wakeup for
worker queue") rightfully moved part of the decision making of
whether we should expand or shrink from the expand/shrink functions
themselves into insert/delete functions in order to avoid unnecessary
worker wake-ups, it however introduced a regression by doing so.
Before that change, if no max_shift was specified (= 0) on rhashtable
initialization, rhashtable_expand() would just grow unconditionally
and lets the available memory be the limiting factor. After that
change, if no max_shift was specified, there would be _no_ expansion
step at all.
Given that netlink and tipc have a max_shift specified, it was not
visible there, but Josh Hunt reported that if nft that starts out
with a default element hint of 3 if not otherwise provided, would
slow i.e. inserts down trememdously as it cannot grow larger to
relax table occupancy.
Given that the test case verifies shrinks/expands manually, we also
must remove pointer to the helper functions to explicitly avoid
parallel resizing on insertions/deletions. test_bucket_stats() and
test_rht_lookup() could also be wrapped around rhashtable mutex to
explicitly synchronize a walk from resizing, but I think that defeats
the actual test case which intended to have explicit test steps,
i.e. 1) inserts, 2) expands, 3) shrinks, 4) deletions, with object
verification after each stage.
Reported-by: Josh Hunt <johunt@akamai.com>
Fixes: c0c09bfdc4 ("rhashtable: avoid unnecessary wakeup for worker queue")
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
Cc: Ying Xue <ying.xue@windriver.com>
Cc: Josh Hunt <johunt@akamai.com>
Acked-by: Thomas Graf <tgraf@suug.ch>
Signed-off-by: David S. Miller <davem@davemloft.net>
Commit f2dba9c6ff ("rhashtable: Introduce rhashtable_walk_*") forgot to
initialize the members of struct rhashtable_walker after allocating it, which
caused an undefined value for 'resize' which is used later on.
Fixes: f2dba9c6ff ("rhashtable: Introduce rhashtable_walk_*")
Signed-off-by: Sasha Levin <sasha.levin@oracle.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
When trying to allocate future tables via bucket_table_alloc(), it seems
overkill on large table shifts that we probe for kzalloc() unconditionally
first, as it's likely to fail.
Only probe with kzalloc() for more reasonable table sizes and use vzalloc()
either as a fallback on failure or directly in case of large table sizes.
Fixes: 7e1e77636e ("lib: Resizable, Scalable, Concurrent Hash Table")
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
Acked-by: Thomas Graf <tgraf@suug.ch>
Signed-off-by: David S. Miller <davem@davemloft.net>
Restore pre 54c5b7d311 behaviour and only probe for expansions on inserts
and shrinks on deletes. Currently, it will happen that on initial inserts
into a sparse hash table, we may i.e. shrink it first simply because it's
not fully populated yet, only to later realize that we need to grow again.
This however is counter intuitive, e.g. an initial default size of 64
elements is already small enough, and in case an elements size hint is given
to the hash table by a user, we should avoid unnecessary expansion steps,
so a shrink is clearly unintended here.
Fixes: 54c5b7d311 ("rhashtable: introduce rhashtable_wakeup_worker helper function")
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
Cc: Ying Xue <ying.xue@windriver.com>
Acked-by: Thomas Graf <tgraf@suug.ch>
Signed-off-by: David S. Miller <davem@davemloft.net>
The remove logic properly searched the remaining chain for a matching
entry with an identical hash but it did this while searching from both
the old and new table. Instead in order to not leave stale references
behind we need to:
1. When growing and searching from the new table:
Search remaining chain for entry with same hash to avoid having
the new table directly point to a entry with a different hash.
2. When shrinking and searching from the old table:
Check if the element after the removed would create a cross
reference and avoid it if so.
These bugs were present from the beginning in nft_hash.
Also, both insert functions calculated the hash based on the mask of
the new table. This worked while growing. Wwhile shrinking, the mask
of the inew table is smaller than the mask of the old table. This lead
to a bit not being taken into account when selecting the bucket lock
and thus caused the wrong bucket to be locked eventually.
Fixes: 7e1e77636e ("lib: Resizable, Scalable, Concurrent Hash Table")
Fixes: 97defe1ecf ("rhashtable: Per bucket locks & deferred expansion/shrinking")
Reported-by: Ying Xue <ying.xue@windriver.com>
Signed-off-by: Thomas Graf <tgraf@suug.ch>
Signed-off-by: David S. Miller <davem@davemloft.net>
During a resize, when two buckets in the larger table map to
a single bucket in the smaller table and the new table has already
been (partially) linked to the old table. Removal of an element
may result the bucket in the larger table to point to entries
which all hash to a different value than the bucket index. Thus
causing two buckets to point to the same sub chain after unzipping.
This is not illegal *during* the resize phase but after it has
completed.
Keep the old table around until all of the unzipping is done to
allow the removal code to only search for matching hashed entries
during this special period.
Reported-by: Ying Xue <ying.xue@windriver.com>
Fixes: 97defe1ecf ("rhashtable: Per bucket locks & deferred expansion/shrinking")
Signed-off-by: Thomas Graf <tgraf@suug.ch>
Signed-off-by: David S. Miller <davem@davemloft.net>
Catch hash miscalculations which result in hard to track down race
conditions.
Signed-off-by: Thomas Graf <tgraf@suug.ch>
Signed-off-by: David S. Miller <davem@davemloft.net>
This simplifies debugging of locking violations if compiled with
CONFIG_PROVE_LOCKING.
Signed-off-by: Thomas Graf <tgraf@suug.ch>
Signed-off-by: David S. Miller <davem@davemloft.net>
We need to wait for all RCU readers to complete after the last bit of
unzipping has been completed. Otherwise the old table is freed up
prematurely.
Fixes: 7e1e77636e ("lib: Resizable, Scalable, Concurrent Hash Table")
Signed-off-by: Thomas Graf <tgraf@suug.ch>
Signed-off-by: David S. Miller <davem@davemloft.net>
rhashtable currently allows to use a bucket lock per bucket. This
requires multiple levels of complicated nested locking because when
resizing, a single bucket of the smaller table will map to two
buckets in the larger table. So far rhashtable has explicitly locked
both buckets in the larger table.
By excluding the highest bit of the hash from the bucket lock map and
thus only allowing locks to buckets in a ratio of 1:2, the locking
can be simplified a lot without losing the benefits of multiple locks.
Larger tables which benefit from multiple locks will not have a single
lock per bucket anyway.
Signed-off-by: Thomas Graf <tgraf@suug.ch>
Signed-off-by: David S. Miller <davem@davemloft.net>
The value computed by key_hashfn() is used by rhashtable_lookup_compare()
to traverse both tables during a resize. key_hashfn() must therefore
return the hash value without the buckets mask applied so it can be
masked to the size of each individual table.
Fixes: 97defe1ecf ("rhashtable: Per bucket locks & deferred expansion/shrinking")
Signed-off-by: Thomas Graf <tgraf@suug.ch>
Signed-off-by: David S. Miller <davem@davemloft.net>
Some existing rhashtable users get too intimate with it by walking
the buckets directly. This prevents us from easily changing the
internals of rhashtable.
This patch adds the helpers rhashtable_walk_init/exit/start/next/stop
which will replace these custom walkers.
They are meant to be usable for both procfs seq_file walks as well
as walking by a netlink dump. The iterator structure should fit
inside a netlink dump cb structure, with at least one element to
spare.
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
Signed-off-by: David S. Miller <davem@davemloft.net>
The current being_destroyed check in rhashtable_expand is not
enough since if we start a shrinking process after freeing all
elements in the table that's also going to crash.
This patch adds a being_destroyed check to the deferred worker
thread so that we bail out as soon as we take the lock.
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
Signed-off-by: David S. Miller <davem@davemloft.net>
Allow the selftest on the resizable hash table to be built modular, just
like all other tests that do not depend on DEBUG_KERNEL.
Signed-off-by: Geert Uytterhoeven <geert@linux-m68k.org>
Acked-by: Thomas Graf <tgraf@suug.ch>
Signed-off-by: David S. Miller <davem@davemloft.net>
As removals can occur during resizes, entries may be referred to from
both tbl and future_tbl when the removal is requested. Therefore
rhashtable_remove() must unlink the entry in both tables if this is
the case. The existing code did search both tables but stopped when it
hit the first match.
Failing to unlink in both tables resulted in use after free.
Fixes: 97defe1ecf ("rhashtable: Per bucket locks & deferred expansion/shrinking")
Reported-by: Ying Xue <ying.xue@windriver.com>
Signed-off-by: Thomas Graf <tgraf@suug.ch>
Signed-off-by: David S. Miller <davem@davemloft.net>
When we put our declared work task in the global workqueue with
schedule_delayed_work(), its delay parameter is always zero.
Therefore, we should define a regular work in rhashtable structure
instead of a delayed work.
By the way, we add a condition to check whether resizing functions
are NULL before cancelling the work, avoiding to cancel an
uninitialized work.
Lastly, while we wait for all work items we submitted before to run
to completion with cancel_delayed_work(), ht->mutex has been taken in
rhashtable_destroy(). Moreover, cancel_delayed_work() doesn't return
until all work items are accomplished, and when work items are
scheduled, the work's function - rht_deferred_worker() will be called.
However, as rht_deferred_worker() also needs to acquire the lock,
deadlock might happen at the moment as the lock is already held before.
So if the cancel work function is moved out of the lock covered scope,
this will avoid the deadlock.
Fixes: 97defe1 ("rhashtable: Per bucket locks & deferred expansion/shrinking")
Signed-off-by: Ying Xue <ying.xue@windriver.com>
Cc: Thomas Graf <tgraf@suug.ch>
Acked-by: Thomas Graf <tgraf@suug.ch>
Signed-off-by: David S. Miller <davem@davemloft.net>
Each per bucket lock covers a configurable number of buckets. While
shrinking, two buckets in the old table contain entries for a single
bucket in the new table. We need to lock down both while linking.
Check if they are protected by different locks to avoid a recursive
lock.
Fixes: 97defe1e ("rhashtable: Per bucket locks & deferred expansion/shrinking")
Reported-by: Fengguang Wu <fengguang.wu@intel.com>
Signed-off-by: Thomas Graf <tgraf@suug.ch>
Signed-off-by: David S. Miller <davem@davemloft.net>
Introduce a new function called rhashtable_lookup_compare_insert()
which is very similar to rhashtable_lookup_insert(). But the former
makes use of users' given compare function to look for an object,
and then inserts it into hash table if found. As the entire process
of search and insertion is under protection of per bucket lock, this
can help users to avoid the involvement of extra lock.
Signed-off-by: Ying Xue <ying.xue@windriver.com>
Cc: Thomas Graf <tgraf@suug.ch>
Acked-by: Thomas Graf <tgraf@suug.ch>
Signed-off-by: David S. Miller <davem@davemloft.net>
Signed-off-by: Ying Xue <ying.xue@windriver.com>
Cc: Thomas Graf <tgraf@suug.ch>
Acked-by: Thomas Graf <tgraf@suug.ch>
Signed-off-by: David S. Miller <davem@davemloft.net>
Move condition statements of verifying whether hash table size exceeds
its maximum threshold or reaches its minimum threshold from resizing
functions to resizing decision functions, avoiding unnecessary wakeup
for worker queue thread.
Signed-off-by: Ying Xue <ying.xue@windriver.com>
Cc: Thomas Graf <tgraf@suug.ch>
Acked-by: Thomas Graf <tgraf@suug.ch>
Signed-off-by: David S. Miller <davem@davemloft.net>
When remove an object from hash table, we currently only traverse old
bucket table to check whether the object exists. If the object is not
found in it, we will try again. But in the second search loop, we still
search the object from the old table instead of future table. As a
result, the object may be not removed from hash table especially when
resizing is currently in progress and the object is just saved in the
future table.
Signed-off-by: Ying Xue <ying.xue@windriver.com>
Cc: Thomas Graf <tgraf@suug.ch>
Acked-by: Thomas Graf <tgraf@suug.ch>
Signed-off-by: David S. Miller <davem@davemloft.net>
Involve a new function called rhashtable_lookup_insert() which makes
lookup and insertion atomic under bucket lock protection, helping us
avoid to introduce an extra lock when we search and insert an object
into hash table.
Signed-off-by: Ying Xue <ying.xue@windriver.com>
Signed-off-by: Thomas Graf <tgraf@suug.ch>
Acked-by: Thomas Graf <tgraf@suug.ch>
Signed-off-by: David S. Miller <davem@davemloft.net>
Introduce rhashtable_wakeup_worker() helper function to reduce
duplicated code where to wake up worker.
By the way, as long as the both "future_tbl" and "tbl" bucket table
pointers point to the same bucket array, we should try to wake up
the resizing worker thread, otherwise, it indicates the work of
resizing hash table is not finished yet. However, currently we will
wake up the worker thread only when the two pointers point to
different bucket array. Obviously this is wrong. So, the issue is
also fixed as well in the patch.
Signed-off-by: Ying Xue <ying.xue@windriver.com>
Cc: Thomas Graf <tgraf@suug.ch>
Acked-by: Thomas Graf <tgraf@suug.ch>
Signed-off-by: David S. Miller <davem@davemloft.net>
Define an internal compare function and relevant compare argument,
and then make use of rhashtable_lookup_compare() to lookup key in
hash table, reducing duplicated code between rhashtable_lookup()
and rhashtable_lookup_compare().
Signed-off-by: Ying Xue <ying.xue@windriver.com>
Cc: Thomas Graf <tgraf@suug.ch>
Acked-by: Thomas Graf <tgraf@suug.ch>
Signed-off-by: David S. Miller <davem@davemloft.net>
In order to allow for wider usage of rhashtable, use a special nulls
marker to terminate each chain. The reason for not using the existing
nulls_list is that the prev pointer usage would not be valid as entries
can be linked in two different buckets at the same time.
The 4 nulls base bits can be set through the rhashtable_params structure
like this:
struct rhashtable_params params = {
[...]
.nulls_base = (1U << RHT_BASE_SHIFT),
};
This reduces the hash length from 32 bits to 27 bits.
Signed-off-by: Thomas Graf <tgraf@suug.ch>
Signed-off-by: David S. Miller <davem@davemloft.net>
Introduces an array of spinlocks to protect bucket mutations. The number
of spinlocks per CPU is configurable and selected based on the hash of
the bucket. This allows for parallel insertions and removals of entries
which do not share a lock.
The patch also defers expansion and shrinking to a worker queue which
allows insertion and removal from atomic context. Insertions and
deletions may occur in parallel to it and are only held up briefly
while the particular bucket is linked or unzipped.
Mutations of the bucket table pointer is protected by a new mutex, read
access is RCU protected.
In the event of an expansion or shrinking, the new bucket table allocated
is exposed as a so called future table as soon as the resize process
starts. Lookups, deletions, and insertions will briefly use both tables.
The future table becomes the main table after an RCU grace period and
initial linking of the old to the new table was performed. Optimization
of the chains to make use of the new number of buckets follows only the
new table is in use.
The side effect of this is that during that RCU grace period, a bucket
traversal using any rht_for_each() variant on the main table will not see
any insertions performed during the RCU grace period which would at that
point land in the future table. The lookup will see them as it searches
both tables if needed.
Having multiple insertions and removals occur in parallel requires nelems
to become an atomic counter.
Signed-off-by: Thomas Graf <tgraf@suug.ch>
Signed-off-by: David S. Miller <davem@davemloft.net>
The removal function of nft_hash currently stores a reference to the
previous element during lookup which is used to optimize removal later
on. This was possible because a lock is held throughout calling
rhashtable_lookup() and rhashtable_remove().
With the introdution of deferred table resizing in parallel to lookups
and insertions, the nftables lock will no longer synchronize all
table mutations and the stored pprev may become invalid.
Removing this optimization makes removal slightly more expensive on
average but allows taking the resize cost out of the insert and
remove path.
Signed-off-by: Thomas Graf <tgraf@suug.ch>
Cc: netfilter-devel@vger.kernel.org
Signed-off-by: David S. Miller <davem@davemloft.net>
Subsequent patches will require access to the bucket tail. Access
to the tail is relatively cheap as the automatic resizing of the
table should keep the number of entries per bucket to no more
than 0.75 on average.
Signed-off-by: Thomas Graf <tgraf@suug.ch>
Signed-off-by: David S. Miller <davem@davemloft.net>
This patch is in preparation to introduce per bucket spinlocks. It
extends all iterator macros to take the bucket table and bucket
index. It also introduces a new rht_dereference_bucket() to
handle protected accesses to buckets.
It introduces a barrier() to the RCU iterators to the prevent
the compiler from caching the first element.
The lockdep verifier is introduced as stub which always succeeds
and properly implement in the next patch when the locks are
introduced.
Signed-off-by: Thomas Graf <tgraf@suug.ch>
Signed-off-by: David S. Miller <davem@davemloft.net>
Hash the key inside of rhashtable_lookup_compare() like
rhashtable_lookup() does. This allows to simplify the hashing
functions and keep them private.
Signed-off-by: Thomas Graf <tgraf@suug.ch>
Cc: netfilter-devel@vger.kernel.org
Signed-off-by: David S. Miller <davem@davemloft.net>
This patch effectively reverts commit 500f808726 ("net: ovs: use CRC32
accelerated flow hash if available"), and other remaining arch_fast_hash()
users such as from nfsd via commit 6282cd5655 ("NFSD: Don't hand out
delegations for 30 seconds after recalling them.") where it has been used
as a hash function for bloom filtering.
While we think that these users are actually not much of concern, it has
been requested to remove the arch_fast_hash() library bits that arose
from [1] entirely as per recent discussion [2]. The main argument is that
using it as a hash may introduce bias due to its linearity (see avalanche
criterion) and thus makes it less clear (though we tried to document that)
when this security/performance trade-off is actually acceptable for a
general purpose library function.
Lets therefore avoid any further confusion on this matter and remove it to
prevent any future accidental misuse of it. For the time being, this is
going to make hashing of flow keys a bit more expensive in the ovs case,
but future work could reevaluate a different hashing discipline.
[1] https://patchwork.ozlabs.org/patch/299369/
[2] https://patchwork.ozlabs.org/patch/418756/
Cc: Neil Brown <neilb@suse.de>
Cc: Francesco Fusco <fusco@ntop.org>
Cc: Jesse Gross <jesse@nicira.com>
Cc: Thomas Graf <tgraf@suug.ch>
Signed-off-by: Daniel Borkmann <dborkman@redhat.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Verify whether both the lock and RCU protected iterators see all
test entries before and after expanding and shrinking has been
performed. Also verify whether the number of entries in the hashtable
remains stable during expansion and shrinking.
Signed-off-by: Thomas Graf <tgraf@suug.ch>
Signed-off-by: David S. Miller <davem@davemloft.net>
Conflicts:
drivers/net/ethernet/chelsio/cxgb4vf/sge.c
drivers/net/ethernet/intel/ixgbe/ixgbe_phy.c
sge.c was overlapping two changes, one to use the new
__dev_alloc_page() in net-next, and one to use s->fl_pg_order in net.
ixgbe_phy.c was a set of overlapping whitespace changes.
Signed-off-by: David S. Miller <davem@davemloft.net>
Reallocation is only required for shrinking and expanding and both rely
on a mutex for synchronization and callers of rhashtable_init() are in
non atomic context. Therefore, no reason to continue passing allocation
hints through the API.
Instead, use GFP_KERNEL and add __GFP_NOWARN | __GFP_NORETRY to allow
for silent fall back to vzalloc() without the OOM killer jumping in as
pointed out by Eric Dumazet and Eric W. Biederman.
Signed-off-by: Thomas Graf <tgraf@suug.ch>
Signed-off-by: David S. Miller <davem@davemloft.net>
Currently mutex_is_held can only test locks in the that are global
since it takes no arguments. This prevents rhashtable from being
used in places where locks are lock, e.g., per-namespace locks.
This patch adds a parent field to mutex_is_held and rhashtable_params
so that local locks can be used (and tested).
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
Signed-off-by: David S. Miller <davem@davemloft.net>
The rhashtable function mutex_is_held is only used when PROVE_LOCKING
is enabled. This patch makes the mutex_is_held field in rhashtable
optional depending on PROVE_LOCKING.
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
Signed-off-by: David S. Miller <davem@davemloft.net>
My editor spewed garbage that looked like memory corruption on
my screen. It turns out that a number of occurences of "fi" got
turned into a ligature.
This patch replaces these ligatures with the ASCII letters "fi".
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
Cheers,
Acked-by: Thomas Graf <tgraf@suug.ch>
Signed-off-by: David S. Miller <davem@davemloft.net>
Pull networking updates from David Miller:
"Most notable changes in here:
1) By far the biggest accomplishment, thanks to a large range of
contributors, is the addition of multi-send for transmit. This is
the result of discussions back in Chicago, and the hard work of
several individuals.
Now, when the ->ndo_start_xmit() method of a driver sees
skb->xmit_more as true, it can choose to defer the doorbell
telling the driver to start processing the new TX queue entires.
skb->xmit_more means that the generic networking is guaranteed to
call the driver immediately with another SKB to send.
There is logic added to the qdisc layer to dequeue multiple
packets at a time, and the handling mis-predicted offloads in
software is now done with no locks held.
Finally, pktgen is extended to have a "burst" parameter that can
be used to test a multi-send implementation.
Several drivers have xmit_more support: i40e, igb, ixgbe, mlx4,
virtio_net
Adding support is almost trivial, so export more drivers to
support this optimization soon.
I want to thank, in no particular or implied order, Jesper
Dangaard Brouer, Eric Dumazet, Alexander Duyck, Tom Herbert, Jamal
Hadi Salim, John Fastabend, Florian Westphal, Daniel Borkmann,
David Tat, Hannes Frederic Sowa, and Rusty Russell.
2) PTP and timestamping support in bnx2x, from Michal Kalderon.
3) Allow adjusting the rx_copybreak threshold for a driver via
ethtool, and add rx_copybreak support to enic driver. From
Govindarajulu Varadarajan.
4) Significant enhancements to the generic PHY layer and the bcm7xxx
driver in particular (EEE support, auto power down, etc.) from
Florian Fainelli.
5) Allow raw buffers to be used for flow dissection, allowing drivers
to determine the optimal "linear pull" size for devices that DMA
into pools of pages. The objective is to get exactly the
necessary amount of headers into the linear SKB area pre-pulled,
but no more. The new interface drivers use is eth_get_headlen().
From WANG Cong, with driver conversions (several had their own
by-hand duplicated implementations) by Alexander Duyck and Eric
Dumazet.
6) Support checksumming more smoothly and efficiently for
encapsulations, and add "foo over UDP" facility. From Tom
Herbert.
7) Add Broadcom SF2 switch driver to DSA layer, from Florian
Fainelli.
8) eBPF now can load programs via a system call and has an extensive
testsuite. Alexei Starovoitov and Daniel Borkmann.
9) Major overhaul of the packet scheduler to use RCU in several major
areas such as the classifiers and rate estimators. From John
Fastabend.
10) Add driver for Intel FM10000 Ethernet Switch, from Alexander
Duyck.
11) Rearrange TCP_SKB_CB() to reduce cache line misses, from Eric
Dumazet.
12) Add Datacenter TCP congestion control algorithm support, From
Florian Westphal.
13) Reorganize sk_buff so that __copy_skb_header() is significantly
faster. From Eric Dumazet"
* git://git.kernel.org/pub/scm/linux/kernel/git/davem/net-next: (1558 commits)
netlabel: directly return netlbl_unlabel_genl_init()
net: add netdev_txq_bql_{enqueue, complete}_prefetchw() helpers
net: description of dma_cookie cause make xmldocs warning
cxgb4: clean up a type issue
cxgb4: potential shift wrapping bug
i40e: skb->xmit_more support
net: fs_enet: Add NAPI TX
net: fs_enet: Remove non NAPI RX
r8169:add support for RTL8168EP
net_sched: copy exts->type in tcf_exts_change()
wimax: convert printk to pr_foo()
af_unix: remove 0 assignment on static
ipv6: Do not warn for informational ICMP messages, regardless of type.
Update Intel Ethernet Driver maintainers list
bridge: Save frag_max_size between PRE_ROUTING and POST_ROUTING
tipc: fix bug in multicast congestion handling
net: better IFF_XMIT_DST_RELEASE support
net/mlx4_en: remove NETDEV_TX_BUSY
3c59x: fix bad split of cpu_to_le32(pci_map_single())
net: bcmgenet: fix Tx ring priority programming
...
Conflicts:
drivers/net/usb/r8152.c
net/netfilter/nfnetlink.c
Both r8152 and nfnetlink conflicts were simple overlapping changes.
Signed-off-by: David S. Miller <davem@davemloft.net>
Pablo Neira Ayuso says:
====================
nf pull request for net
This series contains netfilter fixes for net, they are:
1) Fix lockdep splat in nft_hash when releasing sets from the
rcu_callback context. We don't the mutex there anymore.
2) Remove unnecessary spinlock_bh in the destroy path of the nf_tables
rbtree set type from rcu_callback context.
3) Fix another lockdep splat in rhashtable. None of the callers hold
a mutex when calling rhashtable_destroy.
4) Fix duplicated error reporting from nfnetlink when aborting and
replaying a batch.
5) Fix a Kconfig issue reported by kbuild robot.
====================
Signed-off-by: David S. Miller <davem@davemloft.net>
Conflicts:
arch/mips/net/bpf_jit.c
drivers/net/can/flexcan.c
Both the flexcan and MIPS bpf_jit conflicts were cases of simple
overlapping changes.
Signed-off-by: David S. Miller <davem@davemloft.net>
Although rhashtable library allows user to specify a quiet big size
for user's created hash table, the table may be shrunk to a
very small size - HASH_MIN_SIZE(4) after object is removed from
the table at the first time. Subsequently, even if the total amount
of objects saved in the table is quite lower than user's initial
setting in a long time, the hash table size is still dynamically
adjusted by rhashtable_shrink() or rhashtable_expand() each time
object is inserted or removed from the table. However, as
synchronize_rcu() has to be called when table is shrunk or
expanded by the two functions, we should permit user to set the
minimum table size through configuring the minimum number of shifts
according to user specific requirement, avoiding these expensive
actions of shrinking or expanding because of calling synchronize_rcu().
Signed-off-by: Ying Xue <ying.xue@windriver.com>
Acked-by: Thomas Graf <tgraf@suug.ch>
Signed-off-by: David S. Miller <davem@davemloft.net>
No need for rht_dereference() from rhashtable_destroy() since the
existing callers don't hold the mutex when invoking this function
from:
1) Netlink, this is called in case of memory allocation errors in the
initialization path, no nl_sk_hash_lock is held.
2) Netfilter, this is called from the rcu callback, no nfnl_lock is
held either.
I think it's reasonable to assume that the caller has to make sure
that no hash resizing may happen before releasing the bucket array.
Therefore, the caller should be responsible for releasing this in a
safe way, document this to make people aware of it.
This resolves a rcu lockdep splat in nft_hash:
===============================
[ INFO: suspicious RCU usage. ]
3.16.0+ #178 Not tainted
-------------------------------
lib/rhashtable.c:596 suspicious rcu_dereference_protected() usage!
other info that might help us debug this:
rcu_scheduler_active = 1, debug_locks = 1
1 lock held by ksoftirqd/2/18:
#0: (rcu_callback){......}, at: [<ffffffff810918fd>] rcu_process_callbacks+0x27e/0x4c7
stack backtrace:
CPU: 2 PID: 18 Comm: ksoftirqd/2 Not tainted 3.16.0+ #178
Hardware name: LENOVO 23259H1/23259H1, BIOS G2ET32WW (1.12 ) 05/30/2012
0000000000000001 ffff88011706bb68 ffffffff8143debc 0000000000000000
ffff880117062610 ffff88011706bb98 ffffffff81077515 ffff8800ca041a50
0000000000000004 ffff8800ca386480 ffff8800ca041a00 ffff88011706bbb8
Call Trace:
[<ffffffff8143debc>] dump_stack+0x4e/0x68
[<ffffffff81077515>] lockdep_rcu_suspicious+0xfa/0x103
[<ffffffff81228b1b>] rhashtable_destroy+0x46/0x52
[<ffffffffa06f21a7>] nft_hash_destroy+0x73/0x82 [nft_hash]
Signed-off-by: Pablo Neira Ayuso <pablo@netfilter.org>
Acked-by: Thomas Graf <tgraf@suug.ch>
No need to export rht_obj(), all inner to outer object translations
occur internally. It was intended to be used with rht_for_each() which
now primarily serves as the iterator for rhashtable_remove_pprev() to
effectively flush and free the full table.
Signed-off-by: Thomas Graf <tgraf@suug.ch>
Signed-off-by: David S. Miller <davem@davemloft.net>
Properly annotate next pointers as access is RCU protected in
the lookup path.
Signed-off-by: Thomas Graf <tgraf@suug.ch>
Signed-off-by: David S. Miller <davem@davemloft.net>
Generic implementation of a resizable, scalable, concurrent hash table
based on [0]. The implementation supports both, fixed size keys specified
via an offset and length, or arbitrary keys via own hash and compare
functions.
Lookups are lockless and protected as RCU read side critical sections.
Automatic growing/shrinking based on user configurable watermarks is
available while allowing concurrent lookups to take place.
Objects to be hashed must include a struct rhash_head. The reason for not
using the existing struct hlist_head is that the expansion and shrinking
will have two buckets point to a single entry which would lead in obscure
reverse chaining behaviour.
Code includes a boot selftest if CONFIG_TEST_RHASHTABLE is defined.
[0] https://www.usenix.org/legacy/event/atc11/tech/final_files/Triplett.pdf
Signed-off-by: Thomas Graf <tgraf@suug.ch>
Reviewed-by: Nikolay Aleksandrov <nikolay@redhat.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
