Received-SPF: pass (google.com: domain of linux-kernel-owner@vger.kernel.org designates 23.128.96.18 as permitted sender) client-ip=23.128.96.18;
From:   Gao Xiang <hsiangkao@aol.com>
To:     linux-erofs@lists.ozlabs.org, Chao Yu <yuchao0@huawei.com>
Cc:     Chao Yu <chao@kernel.org>, LKML <linux-kernel@vger.kernel.org>,
        Gao Xiang <hsiangkao@redhat.com>
Subject: [PATCH v2] erofs: force inplace I/O under low memory scenario
Date:   Wed,  9 Dec 2020 20:37:17 +0800
Message-Id: <20201209123717.12430-1-hsiangkao@aol.com>
In-Reply-To: <20201208054600.16302-1-hsiangkao@aol.com>
References: <20201208054600.16302-1-hsiangkao@aol.com>
MIME-Version: 1.0
Content-Transfer-Encoding: 8bit
Content-Length: 6467
Precedence: bulk

From: Gao Xiang <hsiangkao@redhat.com>

Try to forcely switch to inplace I/O under low memory scenario in
order to avoid direct memory reclaim due to cached page allocation.

Reviewed-by: Chao Yu <yuchao0@huawei.com>
Signed-off-by: Gao Xiang <hsiangkao@redhat.com>
---
v2:
 refine the gfp definition.

 fs/erofs/compress.h |  3 +++
 fs/erofs/zdata.c    | 48 +++++++++++++++++++++++++++++++++++++--------
 2 files changed, 43 insertions(+), 8 deletions(-)

diff --git a/fs/erofs/compress.h b/fs/erofs/compress.h
index 4dadde18cdf1..aea129ddda74 100644
--- a/fs/erofs/compress.h
+++ b/fs/erofs/compress.h
@@ -28,11 +28,13 @@ struct z_erofs_decompress_req {
 
 /* some special page->private (unsigned long, see below) */
 #define Z_EROFS_SHORTLIVED_PAGE		(-1UL << 2)
+#define Z_EROFS_PREALLOCATED_PAGE	(-2UL << 2)
 
 /*
  * For all pages in a pcluster, page->private should be one of
  * Type                         Last 2bits      page->private
  * short-lived page             00              Z_EROFS_SHORTLIVED_PAGE
+ * preallocated page (tryalloc) 00              Z_EROFS_PREALLOCATED_PAGE
  * cached/managed page          00              pointer to z_erofs_pcluster
  * online page (file-backed,    01/10/11        sub-index << 2 | count
  *              some pages can be used for inplace I/O)
@@ -40,6 +42,7 @@ struct z_erofs_decompress_req {
  * page->mapping should be one of
  * Type                 page->mapping
  * short-lived page     NULL
+ * preallocated page    NULL
  * cached/managed page  non-NULL or NULL (invalidated/truncated page)
  * online page          non-NULL
  *
diff --git a/fs/erofs/zdata.c b/fs/erofs/zdata.c
index 777790038bc9..6cb356c4217b 100644
--- a/fs/erofs/zdata.c
+++ b/fs/erofs/zdata.c
@@ -20,6 +20,11 @@
 enum z_erofs_cache_alloctype {
 	DONTALLOC,	/* don't allocate any cached pages */
 	DELAYEDALLOC,	/* delayed allocation (at the time of submitting io) */
+	/*
+	 * try to use cached I/O if page allocation succeeds or fallback
+	 * to in-place I/O instead to avoid any direct reclaim.
+	 */
+	TRYALLOC,
 };
 
 /*
@@ -154,13 +159,16 @@ static DEFINE_MUTEX(z_pagemap_global_lock);
 
 static void preload_compressed_pages(struct z_erofs_collector *clt,
 				     struct address_space *mc,
-				     enum z_erofs_cache_alloctype type)
+				     enum z_erofs_cache_alloctype type,
+				     struct list_head *pagepool)
 {
 	const struct z_erofs_pcluster *pcl = clt->pcl;
 	const unsigned int clusterpages = BIT(pcl->clusterbits);
 	struct page **pages = clt->compressedpages;
 	pgoff_t index = pcl->obj.index + (pages - pcl->compressed_pages);
 	bool standalone = true;
+	gfp_t gfp = (mapping_gfp_mask(mc) & ~__GFP_DIRECT_RECLAIM) |
+			__GFP_NOMEMALLOC | __GFP_NORETRY | __GFP_NOWARN;
 
 	if (clt->mode < COLLECT_PRIMARY_FOLLOWED)
 		return;
@@ -168,6 +176,7 @@ static void preload_compressed_pages(struct z_erofs_collector *clt,
 	for (; pages < pcl->compressed_pages + clusterpages; ++pages) {
 		struct page *page;
 		compressed_page_t t;
+		struct page *newpage = NULL;
 
 		/* the compressed page was loaded before */
 		if (READ_ONCE(*pages))
@@ -179,7 +188,15 @@ static void preload_compressed_pages(struct z_erofs_collector *clt,
 			t = tag_compressed_page_justfound(page);
 		} else if (type == DELAYEDALLOC) {
 			t = tagptr_init(compressed_page_t, PAGE_UNALLOCATED);
+		} else if (type == TRYALLOC) {
+			newpage = erofs_allocpage(pagepool, gfp);
+			if (!newpage)
+				goto dontalloc;
+
+			set_page_private(newpage, Z_EROFS_PREALLOCATED_PAGE);
+			t = tag_compressed_page_justfound(newpage);
 		} else {	/* DONTALLOC */
+dontalloc:
 			if (standalone)
 				clt->compressedpages = pages;
 			standalone = false;
@@ -189,8 +206,12 @@ static void preload_compressed_pages(struct z_erofs_collector *clt,
 		if (!cmpxchg_relaxed(pages, NULL, tagptr_cast_ptr(t)))
 			continue;
 
-		if (page)
+		if (page) {
 			put_page(page);
+		} else if (newpage) {
+			set_page_private(newpage, 0);
+			list_add(&newpage->lru, pagepool);
+		}
 	}
 
 	if (standalone)		/* downgrade to PRIMARY_FOLLOWED_NOINPLACE */
@@ -554,7 +575,7 @@ static bool should_alloc_managed_pages(struct z_erofs_decompress_frontend *fe,
 }
 
 static int z_erofs_do_read_page(struct z_erofs_decompress_frontend *fe,
-				struct page *page)
+				struct page *page, struct list_head *pagepool)
 {
 	struct inode *const inode = fe->inode;
 	struct erofs_sb_info *const sbi = EROFS_I_SB(inode);
@@ -607,11 +628,12 @@ static int z_erofs_do_read_page(struct z_erofs_decompress_frontend *fe,
 
 	/* preload all compressed pages (maybe downgrade role if necessary) */
 	if (should_alloc_managed_pages(fe, sbi->ctx.cache_strategy, map->m_la))
-		cache_strategy = DELAYEDALLOC;
+		cache_strategy = TRYALLOC;
 	else
 		cache_strategy = DONTALLOC;
 
-	preload_compressed_pages(clt, MNGD_MAPPING(sbi), cache_strategy);
+	preload_compressed_pages(clt, MNGD_MAPPING(sbi),
+				 cache_strategy, pagepool);
 
 hitted:
 	/*
@@ -1005,6 +1027,16 @@ static struct page *pickup_page_for_submission(struct z_erofs_pcluster *pcl,
 	justfound = tagptr_unfold_tags(t);
 	page = tagptr_unfold_ptr(t);
 
+	/*
+	 * preallocated cached pages, which is used to avoid direct reclaim
+	 * otherwise, it will go inplace I/O path instead.
+	 */
+	if (page->private == Z_EROFS_PREALLOCATED_PAGE) {
+		WRITE_ONCE(pcl->compressed_pages[nr], page);
+		set_page_private(page, 0);
+		tocache = true;
+		goto out_tocache;
+	}
 	mapping = READ_ONCE(page->mapping);
 
 	/*
@@ -1067,7 +1099,7 @@ static struct page *pickup_page_for_submission(struct z_erofs_pcluster *pcl,
 		cond_resched();
 		goto repeat;
 	}
-
+out_tocache:
 	if (!tocache || add_to_page_cache_lru(page, mc, index + nr, gfp)) {
 		/* turn into temporary page if fails (1 ref) */
 		set_page_private(page, Z_EROFS_SHORTLIVED_PAGE);
@@ -1278,7 +1310,7 @@ static int z_erofs_readpage(struct file *file, struct page *page)
 
 	f.headoffset = (erofs_off_t)page->index << PAGE_SHIFT;
 
-	err = z_erofs_do_read_page(&f, page);
+	err = z_erofs_do_read_page(&f, page, &pagepool);
 	(void)z_erofs_collector_end(&f.clt);
 
 	/* if some compressed cluster ready, need submit them anyway */
@@ -1332,7 +1364,7 @@ static void z_erofs_readahead(struct readahead_control *rac)
 		/* traversal in reverse order */
 		head = (void *)page_private(page);
 
-		err = z_erofs_do_read_page(&f, page);
+		err = z_erofs_do_read_page(&f, page, &pagepool);
 		if (err)
 			erofs_err(inode->i_sb,
 				  "readahead error at page %lu @ nid %llu",
-- 
2.24.0