Previously, page cache radix tree nodes were freed after reclaim emptied
out their page pointers.  But now reclaim stores shadow entries in their
place, which are only reclaimed when the inodes themselves are
reclaimed.  This is problematic for bigger files that are still in use
after they have a significant amount of their cache reclaimed, without
any of those pages actually refaulting.  The shadow entries will just
sit there and waste memory.  In the worst case, the shadow entries will
accumulate until the machine runs out of memory.

To get this under control, the VM will track radix tree nodes
exclusively containing shadow entries on a per-NUMA node list.  Per-NUMA
rather than global because we expect the radix tree nodes themselves to
be allocated node-locally and we want to reduce cross-node references of
otherwise independent cache workloads.  A simple shrinker will then
reclaim these nodes on memory pressure.

A few things need to be stored in the radix tree node to implement the
shadow node LRU and allow tree deletions coming from the list:

1. There is no index available that would describe the reverse path
   from the node up to the tree root, which is needed to perform a
   deletion.  To solve this, encode in each node its offset inside the
   parent.  This can be stored in the unused upper bits of the same
   member that stores the node's height at no extra space cost.

2. The number of shadow entries needs to be counted in addition to the
   regular entries, to quickly detect when the node is ready to go to
   the shadow node LRU list.  The current entry count is an unsigned
   int but the maximum number of entries is 64, so a shadow counter
   can easily be stored in the unused upper bits.

3. Tree modification needs tree lock and tree root, which are located
   in the address space, so store an address_space backpointer in the
   node.  The parent pointer of the node is in a union with the 2-word
   rcu_head, so the backpointer comes at no extra cost as well.

4. The node needs to be linked to an LRU list, which requires a list
   head inside the node.  This does increase the size of the node, but
   it does not change the number of objects that fit into a slab page.

[akpm@linux-foundation.org: export the right function]
Signed-off-by: NJohannes Weiner <hannes@cmpxchg.org>
Reviewed-by: NRik van Riel <riel@redhat.com>
Reviewed-by: NMinchan Kim <minchan@kernel.org>
Cc: Andrea Arcangeli <aarcange@redhat.com>
Cc: Bob Liu <bob.liu@oracle.com>
Cc: Christoph Hellwig <hch@infradead.org>
Cc: Dave Chinner <david@fromorbit.com>
Cc: Greg Thelen <gthelen@google.com>
Cc: Hugh Dickins <hughd@google.com>
Cc: Jan Kara <jack@suse.cz>
Cc: KOSAKI Motohiro <kosaki.motohiro@jp.fujitsu.com>
Cc: Luigi Semenzato <semenzato@google.com>
Cc: Mel Gorman <mgorman@suse.de>
Cc: Metin Doslu <metin@citusdata.com>
Cc: Michel Lespinasse <walken@google.com>
Cc: Ozgun Erdogan <ozgun@citusdata.com>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Roman Gushchin <klamm@yandex-team.ru>
Cc: Ryan Mallon <rmallon@gmail.com>
Cc: Tejun Heo <tj@kernel.org>
Cc: Vlastimil Babka <vbabka@suse.cz>
Signed-off-by: NAndrew Morton <akpm@linux-foundation.org>
Signed-off-by: NLinus Torvalds <torvalds@linux-foundation.org>

I've seen a fair number of issues with kswapd and other processes
appearing to get stuck in v3.12-rc.  Using sysrq-p many times seems to
indicate that it gets stuck somewhere in list_lru_walk_node(), called
from prune_icache_sb() and super_cache_scan().

I never seem to be able to trigger a calltrace for functions above that
point.

So I decided to add the following to super_cache_scan():

    @@ -81,10 +81,14 @@ static unsigned long super_cache_scan(struct shrinker *shrink,
            inodes = list_lru_count_node(&sb->s_inode_lru, sc->nid);
            dentries = list_lru_count_node(&sb->s_dentry_lru, sc->nid);
            total_objects = dentries + inodes + fs_objects + 1;
    +printk("%s:%u: %s: dentries %lu inodes %lu total %lu\n", current->comm, current->pid, __func__, dentries, inodes, total_objects);

            /* proportion the scan between the caches */
            dentries = mult_frac(sc->nr_to_scan, dentries, total_objects);
            inodes = mult_frac(sc->nr_to_scan, inodes, total_objects);
    +printk("%s:%u: %s: dentries %lu inodes %lu\n", current->comm, current->pid, __func__, dentries, inodes);
    +BUG_ON(dentries == 0);
    +BUG_ON(inodes == 0);

            /*
             * prune the dcache first as the icache is pinned by it, then
    @@ -99,7 +103,7 @@ static unsigned long super_cache_scan(struct shrinker *shrink,
                    freed += sb->s_op->free_cached_objects(sb, fs_objects,
                                                           sc->nid);
            }
    -
    +printk("%s:%u: %s: dentries %lu inodes %lu freed %lu\n", current->comm, current->pid, __func__, dentries, inodes, freed);
            drop_super(sb);
            return freed;
     }

and shortly thereafter, having applied some pressure, I got this:

    update-apt-xapi:1616: super_cache_scan: dentries 25632 inodes 2 total 25635
    update-apt-xapi:1616: super_cache_scan: dentries 1023 inodes 0
    ------------[ cut here ]------------
    Kernel BUG at c0101994 [verbose debug info unavailable]
    Internal error: Oops - BUG: 0 [#3] SMP ARM
    Modules linked in: fuse rfcomm bnep bluetooth hid_cypress
    CPU: 0 PID: 1616 Comm: update-apt-xapi Tainted: G      D      3.12.0-rc7+ #154
    task: daea1200 ti: c3bf8000 task.ti: c3bf8000
    PC is at super_cache_scan+0x1c0/0x278
    LR is at trace_hardirqs_on+0x14/0x18
    Process update-apt-xapi (pid: 1616, stack limit = 0xc3bf8240)
    ...
    Backtrace:
      (super_cache_scan) from [<c00cd69c>] (shrink_slab+0x254/0x4c8)
      (shrink_slab) from [<c00d09a0>] (try_to_free_pages+0x3a0/0x5e0)
      (try_to_free_pages) from [<c00c59cc>] (__alloc_pages_nodemask+0x5)
      (__alloc_pages_nodemask) from [<c00e07c0>] (__pte_alloc+0x2c/0x13)
      (__pte_alloc) from [<c00e3a70>] (handle_mm_fault+0x84c/0x914)
      (handle_mm_fault) from [<c001a4cc>] (do_page_fault+0x1f0/0x3bc)
      (do_page_fault) from [<c001a7b0>] (do_translation_fault+0xac/0xb8)
      (do_translation_fault) from [<c000840c>] (do_DataAbort+0x38/0xa0)
      (do_DataAbort) from [<c00133f8>] (__dabt_usr+0x38/0x40)

Notice that we had a very low number of inodes, which were reduced to
zero my mult_frac().

Now, prune_icache_sb() calls list_lru_walk_node() passing that number of
inodes (0) into that as the number of objects to scan:

    long prune_icache_sb(struct super_block *sb, unsigned long nr_to_scan,
                         int nid)
    {
            LIST_HEAD(freeable);
            long freed;

            freed = list_lru_walk_node(&sb->s_inode_lru, nid, inode_lru_isolate,
                                           &freeable, &nr_to_scan);

which does:

    unsigned long
    list_lru_walk_node(struct list_lru *lru, int nid, list_lru_walk_cb isolate,
                       void *cb_arg, unsigned long *nr_to_walk)
    {

            struct list_lru_node    *nlru = &lru->node[nid];
            struct list_head *item, *n;
            unsigned long isolated = 0;

            spin_lock(&nlru->lock);
    restart:
            list_for_each_safe(item, n, &nlru->list) {
                    enum lru_status ret;

                    /*
                     * decrement nr_to_walk first so that we don't livelock if we
                     * get stuck on large numbesr of LRU_RETRY items
                     */
                    if (--(*nr_to_walk) == 0)
                            break;

So, if *nr_to_walk was zero when this function was entered, that means
we're wanting to operate on (~0UL)+1 objects - which might as well be
infinite.

Clearly this is not correct behaviour.  If we think about the behaviour
of this function when *nr_to_walk is 1, then clearly it's wrong - we
decrement first and then test for zero - which results in us doing
nothing at all.  A post-decrement would give the desired behaviour -
we'd try to walk one object and one object only if *nr_to_walk were one.

It also gives the correct behaviour for zero - we exit at this point.

Fixes: 5cedf721 ("list_lru: fix broken LRU_RETRY behaviour")
Signed-off-by: NRussell King <rmk+kernel@arm.linux.org.uk>
Cc: Dave Chinner <dchinner@redhat.com>
Cc: Al Viro <viro@zeniv.linux.org.uk>
Cc: Andrew Morton <akpm@linux-foundation.org>
[ Modified to make sure we never underflow the count: this function gets
  called in a loop, so the 0 -> ~0ul transition is dangerous  - Linus ]
Signed-off-by: NLinus Torvalds <torvalds@linux-foundation.org>

We currently use a compile-time constant to size the node array for the
list_lru structure.  Due to this, we don't need to allocate any memory at
initialization time.  But as a consequence, the structures that contain
embedded list_lru lists can become way too big (the superblock for
instance contains two of them).

This patch aims at ameliorating this situation by dynamically allocating
the node arrays with the firmware provided nr_node_ids.
Signed-off-by: NGlauber Costa <glommer@openvz.org>
Cc: Dave Chinner <dchinner@redhat.com>
Cc: Mel Gorman <mgorman@suse.de>
Cc: "Theodore Ts'o" <tytso@mit.edu>
Cc: Adrian Hunter <adrian.hunter@intel.com>
Cc: Al Viro <viro@zeniv.linux.org.uk>
Cc: Artem Bityutskiy <artem.bityutskiy@linux.intel.com>
Cc: Arve Hjønnevåg <arve@android.com>
Cc: Carlos Maiolino <cmaiolino@redhat.com>
Cc: Christoph Hellwig <hch@lst.de>
Cc: Chuck Lever <chuck.lever@oracle.com>
Cc: Daniel Vetter <daniel.vetter@ffwll.ch>
Cc: David Rientjes <rientjes@google.com>
Cc: Gleb Natapov <gleb@redhat.com>
Cc: Greg Thelen <gthelen@google.com>
Cc: J. Bruce Fields <bfields@redhat.com>
Cc: Jan Kara <jack@suse.cz>
Cc: Jerome Glisse <jglisse@redhat.com>
Cc: John Stultz <john.stultz@linaro.org>
Cc: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Cc: Kent Overstreet <koverstreet@google.com>
Cc: Kirill A. Shutemov <kirill.shutemov@linux.intel.com>
Cc: Marcelo Tosatti <mtosatti@redhat.com>
Cc: Mel Gorman <mgorman@suse.de>
Cc: Steven Whitehouse <swhiteho@redhat.com>
Cc: Thomas Hellstrom <thellstrom@vmware.com>
Cc: Trond Myklebust <Trond.Myklebust@netapp.com>
Signed-off-by: NAndrew Morton <akpm@linux-foundation.org>
Signed-off-by: NAl Viro <viro@zeniv.linux.org.uk>

The list_lru implementation has one function, list_lru_dispose_all, with
only one user (the dentry code).  At first, such function appears to make
sense because we are really not interested in the result of isolating each
dentry separately - all of them are going away anyway.  However, it's
implementation is buggy in the following way:

When we call list_lru_dispose_all in fs/dcache.c, we scan all dentries
marking them with DCACHE_SHRINK_LIST.  However, this is done without the
nlru->lock taken.  The imediate result of that is that someone else may
add or remove the dentry from the LRU at the same time.  When list_lru_del
happens in that scenario we will see an element that is not yet marked
with DCACHE_SHRINK_LIST (even though it will be in the future) and
obviously remove it from an lru where the element no longer is.  Since
list_lru_dispose_all will in effect count down nlru's nr_items and
list_lru_del will do the same, this will lead to an imbalance.

The solution for this would not be so simple: we can obviously just keep
the lru_lock taken, but then we have no guarantees that we will be able to
acquire the dentry lock (dentry->d_lock).  To properly solve this, we need
a communication mechanism between the lru and dentry code, so they can
coordinate this with each other.

Such mechanism already exists in the form of the list_lru_walk_cb
callback.  So it is possible to construct a dcache-side prune function
that does the right thing only by calling list_lru_walk in a loop until no
more dentries are available.

With only one user, plus the fact that a sane solution for the problem
would involve boucing between dcache and list_lru anyway, I see little
justification to keep the special case list_lru_dispose_all in tree.
Signed-off-by: NGlauber Costa <glommer@openvz.org>
Cc: Michal Hocko <mhocko@suse.cz>
Acked-by: NDave Chinner <dchinner@redhat.com>
Signed-off-by: NAndrew Morton <akpm@linux-foundation.org>
Signed-off-by: NAl Viro <viro@zeniv.linux.org.uk>

This patch adapts the list_lru API to accept an optional node argument, to
be used by NUMA aware shrinking functions.  Code that does not care about
the NUMA placement of objects can still call into the very same functions
as before.  They will simply iterate over all nodes.
Signed-off-by: NGlauber Costa <glommer@openvz.org>
Cc: Dave Chinner <dchinner@redhat.com>
Cc: Mel Gorman <mgorman@suse.de>
Cc: "Theodore Ts'o" <tytso@mit.edu>
Cc: Adrian Hunter <adrian.hunter@intel.com>
Cc: Al Viro <viro@zeniv.linux.org.uk>
Cc: Artem Bityutskiy <artem.bityutskiy@linux.intel.com>
Cc: Arve Hjønnevåg <arve@android.com>
Cc: Carlos Maiolino <cmaiolino@redhat.com>
Cc: Christoph Hellwig <hch@lst.de>
Cc: Chuck Lever <chuck.lever@oracle.com>
Cc: Daniel Vetter <daniel.vetter@ffwll.ch>
Cc: David Rientjes <rientjes@google.com>
Cc: Gleb Natapov <gleb@redhat.com>
Cc: Greg Thelen <gthelen@google.com>
Cc: J. Bruce Fields <bfields@redhat.com>
Cc: Jan Kara <jack@suse.cz>
Cc: Jerome Glisse <jglisse@redhat.com>
Cc: John Stultz <john.stultz@linaro.org>
Cc: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Cc: Kent Overstreet <koverstreet@google.com>
Cc: Kirill A. Shutemov <kirill.shutemov@linux.intel.com>
Cc: Marcelo Tosatti <mtosatti@redhat.com>
Cc: Mel Gorman <mgorman@suse.de>
Cc: Steven Whitehouse <swhiteho@redhat.com>
Cc: Thomas Hellstrom <thellstrom@vmware.com>
Cc: Trond Myklebust <Trond.Myklebust@netapp.com>
Signed-off-by: NAndrew Morton <akpm@linux-foundation.org>
Signed-off-by: NAl Viro <viro@zeniv.linux.org.uk>

The LRU_RETRY code assumes that the list traversal status after we have
dropped and regained the list lock.  Unfortunately, this is not a valid
assumption, and that can lead to racing traversals isolating objects that
the other traversal expects to be the next item on the list.

This is causing problems with the inode cache shrinker isolation, with
races resulting in an inode on a dispose list being "isolated" because a
racing traversal still thinks it is on the LRU.  The inode is then never
reclaimed and that causes hangs if a subsequent lookup on that inode
occurs.

Fix it by always restarting the list walk on a LRU_RETRY return from the
isolate callback.  Avoid the possibility of livelocks the current code was
trying to avoid by always decrementing the nr_to_walk counter on retries
so that even if we keep hitting the same item on the list we'll eventually
stop trying to walk and exit out of the situation causing the problem.
Reported-by: NMichal Hocko <mhocko@suse.cz>
Signed-off-by: NDave Chinner <dchinner@redhat.com>
Cc: Glauber Costa <glommer@gmail.com>
Signed-off-by: NAndrew Morton <akpm@linux-foundation.org>
Signed-off-by: NAl Viro <viro@zeniv.linux.org.uk>

Now that we have an LRU list API, we can start to enhance the
implementation.  This splits the single LRU list into per-node lists and
locks to enhance scalability.  Items are placed on lists according to the
node the memory belongs to.  To make scanning the lists efficient, also
track whether the per-node lists have entries in them in a active
nodemask.

Note: We use a fixed-size array for the node LRU, this struct can be very
big if MAX_NUMNODES is big.  If this becomes a problem this is fixable by
turning this into a pointer and dynamically allocating this to
nr_node_ids.  This quantity is firwmare-provided, and still would provide
room for all nodes at the cost of a pointer lookup and an extra
allocation.  Because that allocation will most likely come from a may very
well fail.

[glommer@openvz.org: fix warnings, added note about node lru]
Signed-off-by: NDave Chinner <dchinner@redhat.com>
Signed-off-by: NGlauber Costa <glommer@openvz.org>
Reviewed-by: NGreg Thelen <gthelen@google.com>
Acked-by: NMel Gorman <mgorman@suse.de>
Cc: "Theodore Ts'o" <tytso@mit.edu>
Cc: Adrian Hunter <adrian.hunter@intel.com>
Cc: Al Viro <viro@zeniv.linux.org.uk>
Cc: Artem Bityutskiy <artem.bityutskiy@linux.intel.com>
Cc: Arve Hjønnevåg <arve@android.com>
Cc: Carlos Maiolino <cmaiolino@redhat.com>
Cc: Christoph Hellwig <hch@lst.de>
Cc: Chuck Lever <chuck.lever@oracle.com>
Cc: Daniel Vetter <daniel.vetter@ffwll.ch>
Cc: David Rientjes <rientjes@google.com>
Cc: Gleb Natapov <gleb@redhat.com>
Cc: Greg Thelen <gthelen@google.com>
Cc: J. Bruce Fields <bfields@redhat.com>
Cc: Jan Kara <jack@suse.cz>
Cc: Jerome Glisse <jglisse@redhat.com>
Cc: John Stultz <john.stultz@linaro.org>
Cc: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Cc: Kent Overstreet <koverstreet@google.com>
Cc: Kirill A. Shutemov <kirill.shutemov@linux.intel.com>
Cc: Marcelo Tosatti <mtosatti@redhat.com>
Cc: Mel Gorman <mgorman@suse.de>
Cc: Steven Whitehouse <swhiteho@redhat.com>
Cc: Thomas Hellstrom <thellstrom@vmware.com>
Cc: Trond Myklebust <Trond.Myklebust@netapp.com>
Signed-off-by: NAndrew Morton <akpm@linux-foundation.org>
Signed-off-by: NAl Viro <viro@zeniv.linux.org.uk>

Several subsystems use the same construct for LRU lists - a list head, a
spin lock and and item count.  They also use exactly the same code for
adding and removing items from the LRU.  Create a generic type for these
LRU lists.

This is the beginning of generic, node aware LRUs for shrinkers to work
with.

[glommer@openvz.org: enum defined constants for lru. Suggested by gthelen, don't relock over retry]
Signed-off-by: NDave Chinner <dchinner@redhat.com>
Signed-off-by: NGlauber Costa <glommer@openvz.org>
Reviewed-by: NGreg Thelen <gthelen@google.com>
Acked-by: NMel Gorman <mgorman@suse.de>
Cc: "Theodore Ts'o" <tytso@mit.edu>
Cc: Adrian Hunter <adrian.hunter@intel.com>
Cc: Al Viro <viro@zeniv.linux.org.uk>
Cc: Artem Bityutskiy <artem.bityutskiy@linux.intel.com>
Cc: Arve Hjønnevåg <arve@android.com>
Cc: Carlos Maiolino <cmaiolino@redhat.com>
Cc: Christoph Hellwig <hch@lst.de>
Cc: Chuck Lever <chuck.lever@oracle.com>
Cc: Daniel Vetter <daniel.vetter@ffwll.ch>
Cc: David Rientjes <rientjes@google.com>
Cc: Gleb Natapov <gleb@redhat.com>
Cc: Greg Thelen <gthelen@google.com>
Cc: J. Bruce Fields <bfields@redhat.com>
Cc: Jan Kara <jack@suse.cz>
Cc: Jerome Glisse <jglisse@redhat.com>
Cc: John Stultz <john.stultz@linaro.org>
Cc: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Cc: Kent Overstreet <koverstreet@google.com>
Cc: Kirill A. Shutemov <kirill.shutemov@linux.intel.com>
Cc: Marcelo Tosatti <mtosatti@redhat.com>
Cc: Mel Gorman <mgorman@suse.de>
Cc: Steven Whitehouse <swhiteho@redhat.com>
Cc: Thomas Hellstrom <thellstrom@vmware.com>
Cc: Trond Myklebust <Trond.Myklebust@netapp.com>
Signed-off-by: NAndrew Morton <akpm@linux-foundation.org>
Signed-off-by: NAl Viro <viro@zeniv.linux.org.uk>