[patch 01/87] vfs: keep inodes with page cache off the inode shrinker LRU

[Andrew Morton <akpm@linux-foundation.org>]

From: Johannes Weiner <hannes@cmpxchg.org>
Subject: vfs: keep inodes with page cache off the inode shrinker LRU

Historically (pre-2.5), the inode shrinker used to reclaim only empty
inodes and skip over those that still contained page cache.  This caused
problems on highmem hosts: struct inode could put fill lowmem zones before
the cache was getting reclaimed in the highmem zones.

To address this, the inode shrinker started to strip page cache to
facilitate reclaiming lowmem.  However, this comes with its own set of
problems: the shrinkers may drop actively used page cache just because the
inodes are not currently open or dirty - think working with a large git
tree.  It further doesn't respect cgroup memory protection settings and
can cause priority inversions between containers.

Nowadays, the page cache also holds non-resident info for evicted cache
pages in order to detect refaults.  We've come to rely heavily on this
data inside reclaim for protecting the cache workingset and driving swap
behavior.  We also use it to quantify and report workload health through
psi.  The latter in turn is used for fleet health monitoring, as well as
driving automated memory sizing of workloads and containers, proactive
reclaim and memory offloading schemes.

The consequences of dropping page cache prematurely is that we're seeing
subtle and not-so-subtle failures in all of the above-mentioned scenarios,
with the workload generally entering unexpected thrashing states while
losing the ability to reliably detect it.

To fix this on non-highmem systems at least, going back to rotating inodes
on the LRU isn't feasible.  We've tried (commit a76cf1a474d7 ("mm: don't
reclaim inodes with many attached pages")) and failed (commit 69056ee6a8a3
("Revert "mm: don't reclaim inodes with many attached pages"")).  The
issue is mostly that shrinker pools attract pressure based on their size,
and when objects get skipped the shrinkers remember this as deferred
reclaim work.  This accumulates excessive pressure on the remaining
inodes, and we can quickly eat into heavily used ones, or dirty ones that
require IO to reclaim, when there potentially is plenty of cold, clean
cache around still.

Instead, this patch keeps populated inodes off the inode LRU in the first
place - just like an open file or dirty state would.  An otherwise clean
and unused inode then gets queued when the last cache entry disappears. 
This solves the problem without reintroducing the reclaim issues, and
generally is a bit more scalable than having to wade through potentially
hundreds of thousands of busy inodes.

Locking is a bit tricky because the locks protecting the inode state
(i_lock) and the inode LRU (lru_list.lock) don't nest inside the irq-safe
page cache lock (i_pages.xa_lock).  Page cache deletions are serialized
through i_lock, taken before the i_pages lock, to make sure depopulated
inodes are queued reliably.  Additions may race with deletions, but we'll
check again in the shrinker.  If additions race with the shrinker itself,
we're protected by the i_lock: if find_inode() or iput() win, the shrinker
will bail on the elevated i_count or I_REFERENCED; if the shrinker wins
and goes ahead with the inode, it will set I_FREEING and inhibit further
igets(), which will cause the other side to create a new instance of the
inode instead.

Link: https://lkml.kernel.org/r/20210614211904.14420-4-hannes@cmpxchg.org
Signed-off-by: Johannes Weiner <hannes@cmpxchg.org>
Cc: Roman Gushchin <guro@fb.com>
Cc: Tejun Heo <tj@kernel.org>
Cc: Dave Chinner <david@fromorbit.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
---

 fs/inode.c              |   46 ++++++++++++++++++++--------------
 fs/internal.h           |    1 
 include/linux/fs.h      |    1 
 include/linux/pagemap.h |   50 ++++++++++++++++++++++++++++++++++++++
 mm/filemap.c            |    8 ++++++
 mm/truncate.c           |   19 ++++++++++++--
 mm/vmscan.c             |    7 +++++
 mm/workingset.c         |   10 +++++++
 8 files changed, 120 insertions(+), 22 deletions(-)

--- a/fs/inode.c~vfs-keep-inodes-with-page-cache-off-the-inode-shrinker-lru
+++ a/fs/inode.c
@@ -428,11 +428,20 @@ void ihold(struct inode *inode)
 }
 EXPORT_SYMBOL(ihold);
 
-static void inode_lru_list_add(struct inode *inode)
+static void __inode_add_lru(struct inode *inode, bool rotate)
 {
+	if (inode->i_state & (I_DIRTY_ALL | I_SYNC | I_FREEING | I_WILL_FREE))
+		return;
+	if (atomic_read(&inode->i_count))
+		return;
+	if (!(inode->i_sb->s_flags & SB_ACTIVE))
+		return;
+	if (!mapping_shrinkable(&inode->i_data))
+		return;
+
 	if (list_lru_add(&inode->i_sb->s_inode_lru, &inode->i_lru))
 		this_cpu_inc(nr_unused);
-	else
+	else if (rotate)
 		inode->i_state |= I_REFERENCED;
 }
 
@@ -443,16 +452,11 @@ static void inode_lru_list_add(struct in
  */
 void inode_add_lru(struct inode *inode)
 {
-	if (!(inode->i_state & (I_DIRTY_ALL | I_SYNC |
-				I_FREEING | I_WILL_FREE)) &&
-	    !atomic_read(&inode->i_count) && inode->i_sb->s_flags & SB_ACTIVE)
-		inode_lru_list_add(inode);
+	__inode_add_lru(inode, false);
 }
 
-
 static void inode_lru_list_del(struct inode *inode)
 {
-
 	if (list_lru_del(&inode->i_sb->s_inode_lru, &inode->i_lru))
 		this_cpu_dec(nr_unused);
 }
@@ -728,10 +732,6 @@ again:
 /*
  * Isolate the inode from the LRU in preparation for freeing it.
  *
- * Any inodes which are pinned purely because of attached pagecache have their
- * pagecache removed.  If the inode has metadata buffers attached to
- * mapping->private_list then try to remove them.
- *
  * If the inode has the I_REFERENCED flag set, then it means that it has been
  * used recently - the flag is set in iput_final(). When we encounter such an
  * inode, clear the flag and move it to the back of the LRU so it gets another
@@ -747,31 +747,39 @@ static enum lru_status inode_lru_isolate
 	struct inode	*inode = container_of(item, struct inode, i_lru);
 
 	/*
-	 * we are inverting the lru lock/inode->i_lock here, so use a trylock.
-	 * If we fail to get the lock, just skip it.
+	 * We are inverting the lru lock/inode->i_lock here, so use a
+	 * trylock. If we fail to get the lock, just skip it.
 	 */
 	if (!spin_trylock(&inode->i_lock))
 		return LRU_SKIP;
 
 	/*
-	 * Referenced or dirty inodes are still in use. Give them another pass
-	 * through the LRU as we canot reclaim them now.
+	 * Inodes can get referenced, redirtied, or repopulated while
+	 * they're already on the LRU, and this can make them
+	 * unreclaimable for a while. Remove them lazily here; iput,
+	 * sync, or the last page cache deletion will requeue them.
 	 */
 	if (atomic_read(&inode->i_count) ||
-	    (inode->i_state & ~I_REFERENCED)) {
+	    (inode->i_state & ~I_REFERENCED) ||
+	    !mapping_shrinkable(&inode->i_data)) {
 		list_lru_isolate(lru, &inode->i_lru);
 		spin_unlock(&inode->i_lock);
 		this_cpu_dec(nr_unused);
 		return LRU_REMOVED;
 	}
 
-	/* recently referenced inodes get one more pass */
+	/* Recently referenced inodes get one more pass */
 	if (inode->i_state & I_REFERENCED) {
 		inode->i_state &= ~I_REFERENCED;
 		spin_unlock(&inode->i_lock);
 		return LRU_ROTATE;
 	}
 
+	/*
+	 * On highmem systems, mapping_shrinkable() permits dropping
+	 * page cache in order to free up struct inodes: lowmem might
+	 * be under pressure before the cache inside the highmem zone.
+	 */
 	if (inode_has_buffers(inode) || !mapping_empty(&inode->i_data)) {
 		__iget(inode);
 		spin_unlock(&inode->i_lock);
@@ -1638,7 +1646,7 @@ static void iput_final(struct inode *ino
 	if (!drop &&
 	    !(inode->i_state & I_DONTCACHE) &&
 	    (sb->s_flags & SB_ACTIVE)) {
-		inode_add_lru(inode);
+		__inode_add_lru(inode, true);
 		spin_unlock(&inode->i_lock);
 		return;
 	}
--- a/fs/internal.h~vfs-keep-inodes-with-page-cache-off-the-inode-shrinker-lru
+++ a/fs/internal.h
@@ -149,7 +149,6 @@ extern int vfs_open(const struct path *,
  * inode.c
  */
 extern long prune_icache_sb(struct super_block *sb, struct shrink_control *sc);
-extern void inode_add_lru(struct inode *inode);
 extern int dentry_needs_remove_privs(struct dentry *dentry);
 
 /*
--- a/include/linux/fs.h~vfs-keep-inodes-with-page-cache-off-the-inode-shrinker-lru
+++ a/include/linux/fs.h
@@ -3193,6 +3193,7 @@ static inline void remove_inode_hash(str
 }
 
 extern void inode_sb_list_add(struct inode *inode);
+extern void inode_add_lru(struct inode *inode);
 
 extern int sb_set_blocksize(struct super_block *, int);
 extern int sb_min_blocksize(struct super_block *, int);
--- a/include/linux/pagemap.h~vfs-keep-inodes-with-page-cache-off-the-inode-shrinker-lru
+++ a/include/linux/pagemap.h
@@ -24,6 +24,56 @@ static inline bool mapping_empty(struct
 }
 
 /*
+ * mapping_shrinkable - test if page cache state allows inode reclaim
+ * @mapping: the page cache mapping
+ *
+ * This checks the mapping's cache state for the pupose of inode
+ * reclaim and LRU management.
+ *
+ * The caller is expected to hold the i_lock, but is not required to
+ * hold the i_pages lock, which usually protects cache state. That's
+ * because the i_lock and the list_lru lock that protect the inode and
+ * its LRU state don't nest inside the irq-safe i_pages lock.
+ *
+ * Cache deletions are performed under the i_lock, which ensures that
+ * when an inode goes empty, it will reliably get queued on the LRU.
+ *
+ * Cache additions do not acquire the i_lock and may race with this
+ * check, in which case we'll report the inode as shrinkable when it
+ * has cache pages. This is okay: the shrinker also checks the
+ * refcount and the referenced bit, which will be elevated or set in
+ * the process of adding new cache pages to an inode.
+ */
+static inline bool mapping_shrinkable(struct address_space *mapping)
+{
+	void *head;
+
+	/*
+	 * On highmem systems, there could be lowmem pressure from the
+	 * inodes before there is highmem pressure from the page
+	 * cache. Make inodes shrinkable regardless of cache state.
+	 */
+	if (IS_ENABLED(CONFIG_HIGHMEM))
+		return true;
+
+	/* Cache completely empty? Shrink away. */
+	head = rcu_access_pointer(mapping->i_pages.xa_head);
+	if (!head)
+		return true;
+
+	/*
+	 * The xarray stores single offset-0 entries directly in the
+	 * head pointer, which allows non-resident page cache entries
+	 * to escape the shadow shrinker's list of xarray nodes. The
+	 * inode shrinker needs to pick them up under memory pressure.
+	 */
+	if (!xa_is_node(head) && xa_is_value(head))
+		return true;
+
+	return false;
+}
+
+/*
  * Bits in mapping->flags.
  */
 enum mapping_flags {
--- a/mm/filemap.c~vfs-keep-inodes-with-page-cache-off-the-inode-shrinker-lru
+++ a/mm/filemap.c
@@ -262,9 +262,13 @@ void delete_from_page_cache(struct page
 	struct address_space *mapping = page_mapping(page);
 
 	BUG_ON(!PageLocked(page));
+	spin_lock(&mapping->host->i_lock);
 	xa_lock_irq(&mapping->i_pages);
 	__delete_from_page_cache(page, NULL);
 	xa_unlock_irq(&mapping->i_pages);
+	if (mapping_shrinkable(mapping))
+		inode_add_lru(mapping->host);
+	spin_unlock(&mapping->host->i_lock);
 
 	page_cache_free_page(mapping, page);
 }
@@ -340,6 +344,7 @@ void delete_from_page_cache_batch(struct
 	if (!pagevec_count(pvec))
 		return;
 
+	spin_lock(&mapping->host->i_lock);
 	xa_lock_irq(&mapping->i_pages);
 	for (i = 0; i < pagevec_count(pvec); i++) {
 		trace_mm_filemap_delete_from_page_cache(pvec->pages[i]);
@@ -348,6 +353,9 @@ void delete_from_page_cache_batch(struct
 	}
 	page_cache_delete_batch(mapping, pvec);
 	xa_unlock_irq(&mapping->i_pages);
+	if (mapping_shrinkable(mapping))
+		inode_add_lru(mapping->host);
+	spin_unlock(&mapping->host->i_lock);
 
 	for (i = 0; i < pagevec_count(pvec); i++)
 		page_cache_free_page(mapping, pvec->pages[i]);
--- a/mm/truncate.c~vfs-keep-inodes-with-page-cache-off-the-inode-shrinker-lru
+++ a/mm/truncate.c
@@ -45,9 +45,13 @@ static inline void __clear_shadow_entry(
 static void clear_shadow_entry(struct address_space *mapping, pgoff_t index,
 			       void *entry)
 {
+	spin_lock(&mapping->host->i_lock);
 	xa_lock_irq(&mapping->i_pages);
 	__clear_shadow_entry(mapping, index, entry);
 	xa_unlock_irq(&mapping->i_pages);
+	if (mapping_shrinkable(mapping))
+		inode_add_lru(mapping->host);
+	spin_unlock(&mapping->host->i_lock);
 }
 
 /*
@@ -73,8 +77,10 @@ static void truncate_exceptional_pvec_en
 		return;
 
 	dax = dax_mapping(mapping);
-	if (!dax)
+	if (!dax) {
+		spin_lock(&mapping->host->i_lock);
 		xa_lock_irq(&mapping->i_pages);
+	}
 
 	for (i = j; i < pagevec_count(pvec); i++) {
 		struct page *page = pvec->pages[i];
@@ -93,8 +99,12 @@ static void truncate_exceptional_pvec_en
 		__clear_shadow_entry(mapping, index, page);
 	}
 
-	if (!dax)
+	if (!dax) {
 		xa_unlock_irq(&mapping->i_pages);
+		if (mapping_shrinkable(mapping))
+			inode_add_lru(mapping->host);
+		spin_unlock(&mapping->host->i_lock);
+	}
 	pvec->nr = j;
 }
 
@@ -567,6 +577,7 @@ invalidate_complete_page2(struct address
 	if (page_has_private(page) && !try_to_release_page(page, GFP_KERNEL))
 		return 0;
 
+	spin_lock(&mapping->host->i_lock);
 	xa_lock_irq(&mapping->i_pages);
 	if (PageDirty(page))
 		goto failed;
@@ -574,6 +585,9 @@ invalidate_complete_page2(struct address
 	BUG_ON(page_has_private(page));
 	__delete_from_page_cache(page, NULL);
 	xa_unlock_irq(&mapping->i_pages);
+	if (mapping_shrinkable(mapping))
+		inode_add_lru(mapping->host);
+	spin_unlock(&mapping->host->i_lock);
 
 	if (mapping->a_ops->freepage)
 		mapping->a_ops->freepage(page);
@@ -582,6 +596,7 @@ invalidate_complete_page2(struct address
 	return 1;
 failed:
 	xa_unlock_irq(&mapping->i_pages);
+	spin_unlock(&mapping->host->i_lock);
 	return 0;
 }
 
--- a/mm/vmscan.c~vfs-keep-inodes-with-page-cache-off-the-inode-shrinker-lru
+++ a/mm/vmscan.c
@@ -1190,6 +1190,8 @@ static int __remove_mapping(struct addre
 	BUG_ON(!PageLocked(page));
 	BUG_ON(mapping != page_mapping(page));
 
+	if (!PageSwapCache(page))
+		spin_lock(&mapping->host->i_lock);
 	xa_lock_irq(&mapping->i_pages);
 	/*
 	 * The non racy check for a busy page.
@@ -1258,6 +1260,9 @@ static int __remove_mapping(struct addre
 			shadow = workingset_eviction(page, target_memcg);
 		__delete_from_page_cache(page, shadow);
 		xa_unlock_irq(&mapping->i_pages);
+		if (mapping_shrinkable(mapping))
+			inode_add_lru(mapping->host);
+		spin_unlock(&mapping->host->i_lock);
 
 		if (freepage != NULL)
 			freepage(page);
@@ -1267,6 +1272,8 @@ static int __remove_mapping(struct addre
 
 cannot_free:
 	xa_unlock_irq(&mapping->i_pages);
+	if (!PageSwapCache(page))
+		spin_unlock(&mapping->host->i_lock);
 	return 0;
 }
 
--- a/mm/workingset.c~vfs-keep-inodes-with-page-cache-off-the-inode-shrinker-lru
+++ a/mm/workingset.c
@@ -543,6 +543,13 @@ static enum lru_status shadow_lru_isolat
 		goto out;
 	}
 
+	if (!spin_trylock(&mapping->host->i_lock)) {
+		xa_unlock(&mapping->i_pages);
+		spin_unlock_irq(lru_lock);
+		ret = LRU_RETRY;
+		goto out;
+	}
+
 	list_lru_isolate(lru, item);
 	__dec_lruvec_kmem_state(node, WORKINGSET_NODES);
 
@@ -562,6 +569,9 @@ static enum lru_status shadow_lru_isolat
 
 out_invalid:
 	xa_unlock_irq(&mapping->i_pages);
+	if (mapping_shrinkable(mapping))
+		inode_add_lru(mapping->host);
+	spin_unlock(&mapping->host->i_lock);
 	ret = LRU_REMOVED_RETRY;
 out:
 	cond_resched();
_