From: Trond Myklebust <trond.myklebust@hammerspace.com>
Date: Mon, 21 Mar 2022 12:34:19 -0400
Subject: [PATCH] NFS: nfsiod should not block forever in mempool_alloc()
Git-commit: 515dcdcd48736576c6f5c197814da6f81c60a21e
Patch-mainline: v5.18
References: git-fixes

The concern is that since nfsiod is sometimes required to kick off a
commit, it can get locked up waiting forever in mempool_alloc() instead
of failing gracefully and leaving the commit until later.

Try to allocate from the slab first, with GFP_KERNEL | __GFP_NORETRY,
then fall back to a non-blocking attempt to allocate from the memory
pool.

Signed-off-by: Trond Myklebust <trond.myklebust@hammerspace.com>
Acked-by: NeilBrown <neilb@suse.com>

---
 fs/nfs/internal.h      |    7 +++++++
 fs/nfs/pnfs_nfs.c      |    8 ++++++--
 fs/nfs/write.c         |   24 +++++++++---------------
 include/linux/nfs_fs.h |    2 +-
 4 files changed, 23 insertions(+), 18 deletions(-)

--- a/fs/nfs/internal.h
+++ b/fs/nfs/internal.h
@@ -569,6 +569,13 @@ nfs_write_match_verf(const struct nfs_wr
 		!nfs_write_verifier_cmp(&req->wb_verf, &verf->verifier);
 }
 
+static inline gfp_t nfs_io_gfp_mask(void)
+{
+	if (current->flags & PF_WQ_WORKER)
+		return GFP_KERNEL | __GFP_NORETRY | __GFP_NOWARN;
+	return GFP_KERNEL;
+}
+
 /* unlink.c */
 extern struct rpc_task *
 nfs_async_rename(struct inode *old_dir, struct inode *new_dir,
--- a/fs/nfs/pnfs_nfs.c
+++ b/fs/nfs/pnfs_nfs.c
@@ -419,7 +419,7 @@ static struct nfs_commit_data *
 pnfs_bucket_fetch_commitdata(struct pnfs_commit_bucket *bucket,
 			     struct nfs_commit_info *cinfo)
 {
-	struct nfs_commit_data *data = nfs_commitdata_alloc(false);
+	struct nfs_commit_data *data = nfs_commitdata_alloc();
 
 	if (!data)
 		return NULL;
@@ -515,7 +515,11 @@ pnfs_generic_commit_pagelist(struct inod
 	unsigned int nreq = 0;
 
 	if (!list_empty(mds_pages)) {
-		data = nfs_commitdata_alloc(true);
+		data = nfs_commitdata_alloc();
+		if (!data) {
+			nfs_retry_commit(mds_pages, NULL, cinfo, -1);
+			return -ENOMEM;
+		}
 		data->ds_commit_index = -1;
 		list_splice_init(mds_pages, &data->pages);
 		list_add_tail(&data->list, &list);
--- a/fs/nfs/write.c
+++ b/fs/nfs/write.c
@@ -70,27 +70,17 @@ static mempool_t *nfs_wdata_mempool;
 static struct kmem_cache *nfs_cdata_cachep;
 static mempool_t *nfs_commit_mempool;
 
-struct nfs_commit_data *nfs_commitdata_alloc(bool never_fail)
+struct nfs_commit_data *nfs_commitdata_alloc(void)
 {
 	struct nfs_commit_data *p;
 
-	if (never_fail)
-		p = mempool_alloc(nfs_commit_mempool, GFP_NOIO);
-	else {
-		/* It is OK to do some reclaim, not no safe to wait
-		 * for anything to be returned to the pool.
-		 * mempool_alloc() cannot handle that particular combination,
-		 * so we need two separate attempts.
-		 */
+	p = kmem_cache_zalloc(nfs_cdata_cachep, nfs_io_gfp_mask());
+	if (!p) {
 		p = mempool_alloc(nfs_commit_mempool, GFP_NOWAIT);
 		if (!p)
-			p = kmem_cache_alloc(nfs_cdata_cachep, GFP_NOIO |
-					     __GFP_NOWARN | __GFP_NORETRY);
-		if (!p)
 			return NULL;
+		memset(p, 0, sizeof(*p));
 	}
-
-	memset(p, 0, sizeof(*p));
 	INIT_LIST_HEAD(&p->pages);
 	return p;
 }
@@ -1811,7 +1801,11 @@ nfs_commit_list(struct inode *inode, str
 	if (list_empty(head))
 		return 0;
 
-	data = nfs_commitdata_alloc(true);
+	data = nfs_commitdata_alloc();
+	if (!data) {
+		nfs_retry_commit(head, NULL, cinfo, -1);
+		return -ENOMEM;
+	}
 
 	/* Set up the argument struct */
 	nfs_init_commit(data, head, NULL, cinfo);
--- a/include/linux/nfs_fs.h
+++ b/include/linux/nfs_fs.h
@@ -577,7 +577,7 @@ extern int nfs_wb_all(struct inode *inod
 extern int nfs_wb_page(struct inode *inode, struct page *page);
 extern int nfs_wb_page_cancel(struct inode *inode, struct page* page);
 extern int  nfs_commit_inode(struct inode *, int);
-extern struct nfs_commit_data *nfs_commitdata_alloc(bool never_fail);
+extern struct nfs_commit_data *nfs_commitdata_alloc(void);
 extern void nfs_commit_free(struct nfs_commit_data *data);
 bool nfs_commit_end(struct nfs_mds_commit_info *cinfo);