linux.patch

diff --git a/include/linux/mm_types.h b/include/linux/mm_types.h
index 3086502ac..426abad79 100644
--- a/include/linux/mm_types.h
+++ b/include/linux/mm_types.h
@@ -1389,6 +1389,12 @@ struct cache_ext_admission_ctx {
 	u64 ino;
 	u64 offset;
 	u64 size;
+
+	/**
+	 * 0 means invalid
+	 */
+	u64 victim_ino;
+	u64 victim_page_offset;
 };
 
 // TODO: How can I make only some fields cache_ext_eviction_ctx writeable?
@@ -1400,6 +1406,7 @@ struct cache_ext_ops {
 	void (*folio_accessed)(struct folio *folio);
 	void (*folio_evicted)(struct folio *folio);
 	bool (*admit_folio)(struct cache_ext_admission_ctx *ctx);
+	bool (*filter_inode)(u64 ino);
 	// TODO: Add name?
 };
 
diff --git a/mm/cache_ext_ds.c b/mm/cache_ext_ds.c
index cde0fca84..920f716b8 100644
--- a/mm/cache_ext_ds.c
+++ b/mm/cache_ext_ds.c
@@ -66,7 +66,7 @@ int __cache_ext_list_add_impl(struct cache_ext_list *list, struct folio *folio,
 	struct valid_folio *valid_folio = valid_folios_lookup(folio);
 	if (!valid_folio) {
 		spin_unlock(bucket_lock);
-		return -1;
+		return -2;
 	}
 
 	// TODO: Make sure the cache_ext_list still exists.
@@ -79,7 +79,7 @@ int __cache_ext_list_add_impl(struct cache_ext_list *list, struct folio *folio,
 	if (!list_empty(&node->node)) {
 		cache_ext_ds_registry_write_unlock(folio);
 		spin_unlock(bucket_lock);
-		return -1;
+		return -3;
 	}
 
 	if (tail)
@@ -111,7 +111,7 @@ int cache_ext_list_move(struct cache_ext_list *list, struct folio *folio,
 	struct valid_folio *valid_folio = valid_folios_lookup(folio);
 	if (!valid_folio) {
 		spin_unlock(bucket_lock);
-		return -1;
+		return -2;
 	}
 
 	// Get the global list lock
diff --git a/mm/filemap.c b/mm/filemap.c
index b84763395..af27883d2 100644
--- a/mm/filemap.c
+++ b/mm/filemap.c
@@ -2595,28 +2595,32 @@ static int __cache_ext_dio(struct file *file, struct address_space *mapping,
 	for (size_t i = index; i < last_index; ++i) {
 		folio = filemap_alloc_folio(mapping_gfp_mask(mapping), 0);
 		if (!folio) {
-			if (i == index)
+			if (i == index) {
 				return -ENOMEM;
-			// We added some folios that we want to copy and then
-			// free, so this could fix our OOM
+			}
 			return 1;
 		}
 
 		filemap_invalidate_lock_shared(mapping);
 		__folio_set_locked(folio);
 
-		folio_ref_add(folio, folio_nr_pages(folio));
 		folio->mapping = mapping;
 		folio->index = i;
 		error = filemap_read_folio_cache_ext(
 			file, mapping->a_ops->read_folio, folio);
 		filemap_invalidate_unlock_shared(mapping);
 		if (error) {
-			if (i == index)
+			folio->mapping = NULL;
+			folio_put(folio);
+			if (i == index) {
 				return error;
+			}
+
 			return 1;
 		}
 
+		folio->mapping = NULL;
+
 		// Don't overflow fbatch
 		if (!folio_batch_add(fbatch, folio))
 			break;
@@ -2653,15 +2657,41 @@ static int filemap_get_pages(struct kiocb *iocb, size_t count,
 
 		memcg = mem_cgroup_from_task(current);
 		cache_ext_ops = get_cache_ext_ops(memcg);
+		if (cache_ext_ops && cache_ext_ops->filter_inode) {
+			if (!cache_ext_ops->filter_inode(filp->f_inode->i_ino)) {
+				// If the inode is not relevant, admit it
+				ret = false;
+				goto out_cache_ext;
+			}
+		}
+
 		if (cache_ext_ops && cache_ext_ops->admit_folio) {
 			struct cache_ext_admission_ctx ctx = {
 				.ino = filp->f_inode->i_ino,
 				.offset = iocb->ki_pos,
 				.size = count,
+				.victim_ino = 0,
+				.victim_page_offset = 0,
+			};
+
+			/* Get a victim candidate to compare against */
+			struct cache_ext_eviction_ctx evict_ctx = {
+				.request_nr_folios_to_evict = 1,
 			};
+			if (cache_ext_ops->evict_folios) {
+				cache_ext_ops->evict_folios(&evict_ctx, memcg);
+				if (evict_ctx.nr_folios_to_evict > 0) {
+					struct folio *victim = evict_ctx.folios_to_evict[0];
+					ctx.victim_ino = victim->mapping->host->i_ino;
+					ctx.victim_page_offset = victim->index;
+				}
+			}
+
 			ret = cache_ext_ops->admit_folio(&ctx);
 		}
 
+out_cache_ext:
+
 		rcu_read_unlock();
 
 		/* Admission hook decided not to add to page cache */
@@ -2767,10 +2797,12 @@ ssize_t filemap_read(struct kiocb *iocb, struct iov_iter *iter,
 			break;
 
 		error = filemap_get_pages(iocb, iter->count, &fbatch, false);
-		if (error < 0)
+		if (error < 0) {
 			break;
-		else if (error == 1)
+		}
+		else if (error == 1) {
 			cache_ext_flag = 1;
+		}
 
 		/*
 		 * i_size must be checked after we know the pages are Uptodate.
@@ -2797,7 +2829,8 @@ ssize_t filemap_read(struct kiocb *iocb, struct iov_iter *iter,
 		 */
 		if (!pos_same_folio(iocb->ki_pos, last_pos - 1,
 				    fbatch.folios[0]))
-			folio_mark_accessed(fbatch.folios[0]);
+			if (!cache_ext_flag)
+				folio_mark_accessed(fbatch.folios[0]);
 
 		for (i = 0; i < folio_batch_count(&fbatch); i++) {
 			struct folio *folio = fbatch.folios[i];
@@ -2816,8 +2849,6 @@ ssize_t filemap_read(struct kiocb *iocb, struct iov_iter *iter,
 			 * virtual addresses, take care of potential aliasing
 			 * before reading the folio on the kernel side.
 			 */
-
-			// TODO: investigate
 			if (writably_mapped)
 				flush_dcache_folio(folio);
 
@@ -2831,15 +2862,11 @@ ssize_t filemap_read(struct kiocb *iocb, struct iov_iter *iter,
 				error = -EFAULT;
 				break;
 			}
-
-			// TODO: drop reference count before freeing?
-			if (cache_ext_flag)
-				filemap_free_folio(mapping, folio);
 		}
 put_folios:
-		if (!cache_ext_flag)
-			for (i = 0; i < folio_batch_count(&fbatch); i++)
-				folio_put(fbatch.folios[i]);
+		for (i = 0; i < folio_batch_count(&fbatch); i++) {
+			folio_put(fbatch.folios[i]);
+		}
 
 		cache_ext_flag = 0;
 		folio_batch_init(&fbatch);
@@ -3044,9 +3071,11 @@ ssize_t filemap_splice_read(struct file *in, loff_t *ppos,
 			break;
 
 		iocb.ki_pos = *ppos;
+
 		error = filemap_get_pages(&iocb, len, &fbatch, true);
-		if (error < 0)
+		if (error < 0) {
 			break;
+		}
 
 		/*
 		 * i_size must be checked after we know the pages are Uptodate.