Linux EXT4 FS development
 help / color / mirror / Atom feed
From: "Darrick J. Wong" <djwong@kernel.org>
To: tytso@mit.edu
Cc: linux-ext4@vger.kernel.org
Subject: [PATCH 19/23] cache: implement automatic shrinking
Date: Thu, 06 Nov 2025 14:48:07 -0800	[thread overview]
Message-ID: <176246795908.2864310.4023008384897404874.stgit@frogsfrogsfrogs> (raw)
In-Reply-To: <176246795459.2864310.10641701647593035148.stgit@frogsfrogsfrogs>

From: Darrick J. Wong <djwong@kernel.org>

Shrink the cache whenever maxcount has been expanded beyond its initial
value, we release a cached object to one of the mru lists and the number
of objects sitting on the mru is enough to drop the cache count down a
level.  This enables a cache to reduce its memory consumption after a
spike in which reclamation wasn't possible.

Signed-off-by: "Darrick J. Wong" <djwong@kernel.org>
---
 lib/support/cache.h |   20 +++++++--
 lib/support/cache.c |  119 ++++++++++++++++++++++++++++++++++++++++++++++++---
 2 files changed, 129 insertions(+), 10 deletions(-)


diff --git a/lib/support/cache.h b/lib/support/cache.h
index 32b99b5fe733e3..c7c8298c115d50 100644
--- a/lib/support/cache.h
+++ b/lib/support/cache.h
@@ -16,7 +16,11 @@
  */
 #define CACHE_MISCOMPARE_PURGE	(1 << 0)
 
-#define CACHE_FLAGS_ALL		(CACHE_MISCOMPARE_PURGE)
+/* Automatically shrink the cache's max_count when possible. */
+#define CACHE_AUTO_SHRINK	(1 << 1)
+
+#define CACHE_FLAGS_ALL		(CACHE_MISCOMPARE_PURGE | \
+				 CACHE_AUTO_SHRINK)
 
 /*
  * cache object campare return values
@@ -69,12 +73,18 @@ typedef unsigned int (*cache_bulk_relse_t)(struct cache *, struct list_head *);
 typedef int (*cache_node_get_t)(struct cache *c, struct cache_node *cn);
 typedef void (*cache_node_put_t)(struct cache *c, struct cache_node *cn);
 typedef unsigned int (*cache_node_resize_t)(const struct cache *c,
-					    unsigned int curr_size);
+					    unsigned int curr_size,
+					    int dir);
 
 static inline unsigned int cache_gradual_resize(const struct cache *cache,
-						unsigned int curr_size)
+						unsigned int curr_size,
+						int dir)
 {
-	return curr_size * 5 / 4;
+	if (dir < 0)
+		return curr_size * 9 / 10;
+	else if (dir > 0)
+		return curr_size * 5 / 4;
+	return curr_size;
 }
 
 struct cache_operations {
@@ -113,6 +123,7 @@ struct cache_node {
 
 struct cache {
 	int			c_flags;	/* behavioural flags */
+	unsigned int		c_orig_max;	/* original max cache nodes */
 	unsigned int		c_maxcount;	/* max cache nodes */
 	unsigned int		c_count;	/* count of nodes */
 	pthread_mutex_t		c_mutex;	/* node count mutex */
@@ -145,6 +156,7 @@ void cache_destroy(struct cache *cache);
 void cache_walk(struct cache *cache, cache_walk_t fn, void *data);
 void cache_purge(struct cache *);
 bool cache_flush(struct cache *cache);
+void cache_shrink(struct cache *cache);
 
 void cache_set_maxcount(struct cache *cache, unsigned int maxcount);
 int cache_set_flag(struct cache *cache, int flags);
diff --git a/lib/support/cache.c b/lib/support/cache.c
index 99044248b85d38..3a9e276f11af72 100644
--- a/lib/support/cache.c
+++ b/lib/support/cache.c
@@ -53,6 +53,7 @@ cache_init(
 	cache->c_hits = 0;
 	cache->c_misses = 0;
 	cache->c_maxcount = maxcount;
+	cache->c_orig_max = maxcount;
 	cache->hash = cache_operations->hash;
 	cache->alloc = cache_operations->alloc;
 	cache->flush = cache_operations->flush;
@@ -93,6 +94,7 @@ cache_set_maxcount(
 	unsigned int		maxcount)
 {
 	pthread_mutex_lock(&cache->c_mutex);
+	cache->c_orig_max = maxcount;
 	cache->c_maxcount = maxcount;
 	pthread_mutex_unlock(&cache->c_mutex);
 }
@@ -123,7 +125,7 @@ cache_expand(
 
 	pthread_mutex_lock(&cache->c_mutex);
 	if (cache->resize)
-		new_size = cache->resize(cache, cache->c_maxcount);
+		new_size = cache->resize(cache, cache->c_maxcount, 1);
 	if (new_size <= cache->c_maxcount)
 		new_size = cache->c_maxcount * 2;
 #ifdef CACHE_DEBUG
@@ -254,7 +256,8 @@ static unsigned int
 cache_shake(
 	struct cache *		cache,
 	unsigned int		priority,
-	bool			purge)
+	bool			purge,
+	unsigned int		nr_to_shake)
 {
 	struct cache_mru	*mru;
 	struct cache_hash	*hash;
@@ -302,7 +305,7 @@ cache_shake(
 		pthread_mutex_unlock(&node->cn_mutex);
 
 		count++;
-		if (!purge && count == CACHE_SHAKE_COUNT)
+		if (!purge && count == nr_to_shake)
 			break;
 	}
 	pthread_mutex_unlock(&mru->cm_mutex);
@@ -315,7 +318,7 @@ cache_shake(
 		pthread_mutex_unlock(&cache->c_mutex);
 	}
 
-	return (count == CACHE_SHAKE_COUNT) ? priority : ++priority;
+	return (count == nr_to_shake) ? priority : ++priority;
 }
 
 /*
@@ -505,7 +508,7 @@ cache_node_get(
 		node = cache_node_allocate(cache, key);
 		if (node)
 			break;
-		priority = cache_shake(cache, priority, false);
+		priority = cache_shake(cache, priority, false, CACHE_SHAKE_COUNT);
 		/*
 		 * We start at 0; if we free CACHE_SHAKE_COUNT we get
 		 * back the same priority, if not we get back priority+1.
@@ -535,12 +538,112 @@ cache_node_get(
 	return 1;
 }
 
+static unsigned int cache_mru_count(const struct cache *cache)
+{
+	const struct cache_mru	*mru = cache->c_mrus;
+	unsigned int		mru_count = 0;
+	unsigned int		i;
+
+	for (i = 0; i < CACHE_NR_PRIORITIES; i++, mru++)
+		mru_count += mru->cm_count;
+
+	return mru_count;
+}
+
+
+void cache_shrink(struct cache *cache)
+{
+	unsigned int		mru_count = 0;
+	unsigned int		threshold = 0;
+	unsigned int		priority = 0;
+	unsigned int		new_size;
+
+	pthread_mutex_lock(&cache->c_mutex);
+	/* Don't shrink below the original cache size */
+	if (cache->c_maxcount <= cache->c_orig_max)
+		goto out_unlock;
+
+	mru_count = cache_mru_count(cache);
+
+	/*
+	 * If there's not even a batch of nodes on the MRU to try to free,
+	 * don't bother with the rest.
+	 */
+	if (mru_count < CACHE_SHAKE_COUNT)
+		goto out_unlock;
+
+	/*
+	 * Figure out the next step down in size, but don't go below the
+	 * original size.
+	 */
+	if (cache->resize)
+		new_size = cache->resize(cache, cache->c_maxcount, -1);
+	else
+		new_size = cache->c_maxcount / 2;
+	if (new_size >= cache->c_maxcount)
+		goto out_unlock;
+	if (new_size < cache->c_orig_max)
+		new_size = cache->c_orig_max;
+
+	/*
+	 * If we can't purge enough nodes to get the node count below new_size,
+	 * don't resize the cache.
+	 */
+	if (cache->c_count - mru_count >= new_size)
+		goto out_unlock;
+
+#ifdef CACHE_DEBUG
+	fprintf(stderr, "decreasing cache max size from %u to %u (currently %u)\n",
+		cache->c_maxcount, new_size, cache->c_count);
+#endif
+	cache->c_maxcount = new_size;
+
+	/* Try to reduce the number of cached objects. */
+	do {
+		unsigned int new_priority;
+
+		/*
+		 * The threshold is the amount we need to purge to get c_count
+		 * below the new maxcount.  Try to free some objects off the
+		 * MRU.  Drop c_mutex because cache_shake will take it.
+		 */
+		threshold = cache->c_count - new_size;
+		pthread_mutex_unlock(&cache->c_mutex);
+
+		new_priority = cache_shake(cache, priority, false, threshold);
+
+		/* Either we made no progress or we ran out of MRU levels */
+		if (new_priority == priority ||
+		    new_priority > CACHE_MAX_PRIORITY)
+			return;
+		priority = new_priority;
+
+		pthread_mutex_lock(&cache->c_mutex);
+		/*
+		 * Someone could have walked in and changed the cache maxsize
+		 * again while we had the lock dropped.  If that happened, stop
+		 * clearing.
+		 */
+		if (cache->c_maxcount != new_size)
+			goto out_unlock;
+
+		mru_count = cache_mru_count(cache);
+		if (cache->c_count - mru_count >= new_size)
+			goto out_unlock;
+	} while (1);
+
+out_unlock:
+	pthread_mutex_unlock(&cache->c_mutex);
+	return;
+}
+
 void
 cache_node_put(
 	struct cache *		cache,
 	struct cache_node *	node)
 {
 	struct cache_mru *	mru;
+	bool was_put = false;
 
 	pthread_mutex_lock(&node->cn_mutex);
 #ifdef CACHE_DEBUG
@@ -556,6 +659,7 @@ cache_node_put(
 	}
 #endif
 	node->cn_count--;
+	was_put = (node->cn_count == 0);
 
 	if (node->cn_count == 0 && cache->put)
 		cache->put(cache, node);
@@ -569,6 +673,9 @@ cache_node_put(
 	}
 
 	pthread_mutex_unlock(&node->cn_mutex);
+
+	if (was_put && (cache->c_flags & CACHE_AUTO_SHRINK))
+		cache_shrink(cache);
 }
 
 void
@@ -660,7 +767,7 @@ cache_purge(
 	int			i;
 
 	for (i = 0; i <= CACHE_DIRTY_PRIORITY; i++)
-		cache_shake(cache, i, true);
+		cache_shake(cache, i, true, CACHE_SHAKE_COUNT);
 
 #ifdef CACHE_DEBUG
 	if (cache->c_count != 0) {


  parent reply	other threads:[~2025-11-06 22:48 UTC|newest]

Thread overview: 84+ messages / expand[flat|nested]  mbox.gz  Atom feed  top
2025-11-06 22:14 [PATCHBOMB 1.48] fuse2fs: new features, new server Darrick J. Wong
2025-11-06 22:27 ` [PATCHSET 1/9] fuse2fs: fix locking problems Darrick J. Wong
2025-11-06 22:30   ` [PATCH 1/4] libext2fs: add POSIX advisory locking to the unix IO manager Darrick J. Wong
2025-11-06 22:30   ` [PATCH 2/4] fuse2fs: try to lock filesystem image files before using them Darrick J. Wong
2025-11-06 22:30   ` [PATCH 3/4] fuse2fs: quiet down write-protect warning Darrick J. Wong
2025-11-06 22:31   ` [PATCH 4/4] fuse2fs: try to grab block device O_EXCL repeatedly Darrick J. Wong
2025-11-06 22:28 ` [PATCHSET 2/9] fuse2fs: add some easy new features Darrick J. Wong
2025-11-06 22:31   ` [PATCH 01/19] libext2fs: initialize htree when expanding directory Darrick J. Wong
2025-11-06 22:31   ` [PATCH 02/19] libext2fs: create link count adjustment helpers for dir_nlink Darrick J. Wong
2025-11-06 22:31   ` [PATCH 03/19] libext2fs: fix ext2fs_mmp_update Darrick J. Wong
2025-11-06 22:32   ` [PATCH 04/19] libext2fs: refactor aligned MMP buffer allocation Darrick J. Wong
2025-11-06 22:32   ` [PATCH 05/19] libext2fs: always use ext2fs_mmp_get_mem to allocate fs->mmp_buf Darrick J. Wong
2025-11-06 22:32   ` [PATCH 06/19] fuse2fs: check root directory while mounting Darrick J. Wong
2025-11-06 22:32   ` [PATCH 07/19] fuse2fs: read bitmaps asynchronously during initialization Darrick J. Wong
2025-11-06 22:33   ` [PATCH 08/19] fuse2fs: use file handles when possible Darrick J. Wong
2025-11-06 22:33   ` [PATCH 09/19] fuse2fs: implement dir seeking Darrick J. Wong
2025-11-06 22:33   ` [PATCH 10/19] fuse2fs: implement readdirplus Darrick J. Wong
2025-11-06 22:34   ` [PATCH 11/19] fuse2fs: implement dirsync mode Darrick J. Wong
2025-11-06 22:34   ` [PATCH 12/19] fuse2fs: only flush O_SYNC files on close Darrick J. Wong
2025-11-06 22:34   ` [PATCH 13/19] fuse2fs: improve want_extra_isize handling Darrick J. Wong
2025-11-06 22:34   ` [PATCH 14/19] fuse2fs: cache symlink targets in the kernel Darrick J. Wong
2025-11-06 22:35   ` [PATCH 15/19] fuse2fs: constrain worker thread count Darrick J. Wong
2025-11-06 22:35   ` [PATCH 16/19] fuse2fs: improve error handling behaviors Darrick J. Wong
2025-11-06 22:35   ` [PATCH 17/19] fuse2fs: fix link count overflows on dir_nlink filesystems Darrick J. Wong
2025-11-06 22:35   ` [PATCH 18/19] libsupport: add background thread manager Darrick J. Wong
2025-11-06 22:36   ` [PATCH 19/19] fuse2fs: implement MMP updates Darrick J. Wong
2025-11-06 22:28 ` [PATCHSET 3/9] fuse2fs: clean up operation startup Darrick J. Wong
2025-11-06 22:36   ` [PATCH 1/9] fuse2fs: rework FUSE2FS_CHECK_CONTEXT not to rely on global_fs Darrick J. Wong
2025-11-06 22:36   ` [PATCH 2/9] fuse2fs: rework checking file handles Darrick J. Wong
2025-11-06 22:36   ` [PATCH 3/9] fuse2fs: rework fallocate file handle extraction Darrick J. Wong
2025-11-06 22:37   ` [PATCH 4/9] fuse2fs: consolidate file handle checking in op_ioctl Darrick J. Wong
2025-11-06 22:37   ` [PATCH 5/9] fuse2fs: move fs assignment closer to locking the bfl Darrick J. Wong
2025-11-06 22:37   ` [PATCH 6/9] fuse2fs: clean up operation startup Darrick J. Wong
2025-11-06 22:37   ` [PATCH 7/9] fuse2fs: clean up operation completion Darrick J. Wong
2025-11-06 22:38   ` [PATCH 8/9] fuse2fs: clean up more boilerplate Darrick J. Wong
2025-11-06 22:38   ` [PATCH 9/9] fuse2fs: collect runtime of various operations Darrick J. Wong
2025-11-06 22:28 ` [PATCHSET 4/9] fuse2fs: refactor unmount code Darrick J. Wong
2025-11-06 22:38   ` [PATCH 1/3] fuse2fs: get rid of the global_fs variable Darrick J. Wong
2025-11-06 22:39   ` [PATCH 2/3] fuse2fs: hoist lockfile code Darrick J. Wong
2025-11-06 22:39   ` [PATCH 3/3] fuse2fs: hoist unmount code from main Darrick J. Wong
2025-11-06 22:28 ` [PATCHSET 5/9] fuse2fs: refactor mount code Darrick J. Wong
2025-11-06 22:39   ` [PATCH 1/3] fuse2fs: split filesystem mounting into helper functions Darrick J. Wong
2025-11-06 22:39   ` [PATCH 2/3] fuse2fs: register as an IO flusher thread Darrick J. Wong
2025-11-06 22:40   ` [PATCH 3/3] fuse2fs: adjust OOM killer score if possible Darrick J. Wong
2025-11-06 22:29 ` [PATCHSET 6/9] fuse2fs: improve operation tracing Darrick J. Wong
2025-11-06 22:40   ` [PATCH 1/4] fuse2fs: hook library error message printing Darrick J. Wong
2025-11-06 22:40   ` [PATCH 2/4] fuse2fs: print the function name in error messages, not the file name Darrick J. Wong
2025-11-06 22:40   ` [PATCH 3/4] fuse2fs: improve tracing for file range operations Darrick J. Wong
2025-11-06 22:41   ` [PATCH 4/4] fuse2fs: record thread id in debug trace data Darrick J. Wong
2025-11-06 22:29 ` [PATCHSET 7/9] fuse2fs: better tracking of writable state Darrick J. Wong
2025-11-06 22:41   ` [PATCH 1/3] fuse2fs: pass a struct fuse2fs to fs_writeable Darrick J. Wong
2025-11-06 22:41   ` [PATCH 2/3] fuse2fs: track our own writable state Darrick J. Wong
2025-11-06 22:41   ` [PATCH 3/3] fuse2fs: enable the shutdown ioctl Darrick J. Wong
2025-11-06 22:29 ` [PATCHSET 8/9] fuse2fs: upgrade to libfuse 3.17 Darrick J. Wong
2025-11-06 22:42   ` [PATCH 1/4] fuse2fs: bump library version Darrick J. Wong
2025-11-06 22:42   ` [PATCH 2/4] fuse2fs: wrap the fuse_set_feature_flag helper for older libfuse Darrick J. Wong
2025-11-06 22:42   ` [PATCH 3/4] fuse2fs: disable nfs exports Darrick J. Wong
2025-11-06 22:43   ` [PATCH 4/4] fuse2fs: drop fuse 2.x support code Darrick J. Wong
2025-11-06 22:30 ` [PATCHSET 9/9] fuse4fs: fork a low level fuse server Darrick J. Wong
2025-11-06 22:43   ` [PATCH 01/23] fuse2fs: separate libfuse3 and fuse2fs detection in configure Darrick J. Wong
2025-11-06 22:43   ` [PATCH 02/23] fuse2fs: start porting fuse2fs to lowlevel libfuse API Darrick J. Wong
2025-11-06 22:43   ` [PATCH 03/23] debian: create new package for fuse4fs Darrick J. Wong
2025-11-06 22:44   ` [PATCH 04/23] fuse4fs: namespace some helpers Darrick J. Wong
2025-11-07  8:09     ` Amir Goldstein
2025-11-08  0:25       ` Darrick J. Wong
2025-11-06 22:44   ` [PATCH 05/23] fuse4fs: convert to low level API Darrick J. Wong
2025-11-06 22:44   ` [PATCH 06/23] libsupport: port the kernel list.h to libsupport Darrick J. Wong
2025-11-06 22:44   ` [PATCH 07/23] libsupport: add a cache Darrick J. Wong
2025-11-06 22:45   ` [PATCH 08/23] cache: disable debugging Darrick J. Wong
2025-11-06 22:45   ` [PATCH 09/23] cache: use modern list iterator macros Darrick J. Wong
2025-11-06 22:45   ` [PATCH 10/23] cache: embed struct cache in the owner Darrick J. Wong
2025-11-06 22:45   ` [PATCH 11/23] cache: pass cache pointer to callbacks Darrick J. Wong
2025-11-06 22:46   ` [PATCH 12/23] cache: pass a private data pointer through cache_walk Darrick J. Wong
2025-11-06 22:46   ` [PATCH 13/23] cache: add a helper to grab a new refcount for a cache_node Darrick J. Wong
2025-11-06 22:46   ` [PATCH 14/23] cache: return results of a cache flush Darrick J. Wong
2025-11-06 22:47   ` [PATCH 15/23] cache: add a "get only if incore" flag to cache_node_get Darrick J. Wong
2025-11-06 22:47   ` [PATCH 16/23] cache: support gradual expansion Darrick J. Wong
2025-11-06 22:47   ` [PATCH 17/23] cache: support updating maxcount and flags Darrick J. Wong
2025-11-06 22:47   ` [PATCH 18/23] cache: support channging flags Darrick J. Wong
2025-11-06 22:48   ` Darrick J. Wong [this message]
2025-11-06 22:48   ` [PATCH 20/23] fuse4fs: add cache to track open files Darrick J. Wong
2025-11-06 22:48   ` [PATCH 21/23] fuse4fs: use the orphaned inode list Darrick J. Wong
2025-11-06 22:48   ` [PATCH 22/23] fuse4fs: implement FUSE_TMPFILE Darrick J. Wong
2025-11-06 22:49   ` [PATCH 23/23] fuse4fs: create incore reverse orphan list Darrick J. Wong

Reply instructions:

You may reply publicly to this message via plain-text email
using any one of the following methods:

* Save the following mbox file, import it into your mail client,
  and reply-to-all from there: mbox

  Avoid top-posting and favor interleaved quoting:
  https://en.wikipedia.org/wiki/Posting_style#Interleaved_style

* Reply using the --to, --cc, and --in-reply-to
  switches of git-send-email(1):

  git send-email \
    --in-reply-to=176246795908.2864310.4023008384897404874.stgit@frogsfrogsfrogs \
    --to=djwong@kernel.org \
    --cc=linux-ext4@vger.kernel.org \
    --cc=tytso@mit.edu \
    /path/to/YOUR_REPLY

  https://kernel.org/pub/software/scm/git/docs/git-send-email.html

* If your mail client supports setting the In-Reply-To header
  via mailto: links, try the mailto: link
Be sure your reply has a Subject: header at the top and a blank line before the message body.
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox