git: 4fee353bcb45 - stable/13 - nfscl: Add hash lists for the NFSv4 opens

[Rick Macklem]

The branch stable/13 has been updated by rmacklem:

URL: https://cgit.FreeBSD.org/src/commit/?id=4fee353bcb452022ae8a049eca7b32ab4b816b90

commit 4fee353bcb452022ae8a049eca7b32ab4b816b90
Author:     Rick Macklem <rmacklem at FreeBSD.org>
AuthorDate: 2021-05-22 21:51:38 +0000
Commit:     Rick Macklem <rmacklem at FreeBSD.org>
CommitDate: 2021-06-05 01:16:10 +0000

    nfscl: Add hash lists for the NFSv4 opens
    
    A problem was reported via email, where a large (130000+) accumulation
    of NFSv4 opens on an NFSv4 mount caused significant lock contention
    on the mutex used to protect the client mount's open/lock state.
    Although the root cause for the accumulation of opens was not
    resolved, it is obvious that the NFSv4 client is not designed to
    handle 100000+ opens efficiently.  When searching for an open,
    usually for a match by file handle, a linear search of all opens
    is done.
    
    This patch adds a table of hash lists for the opens, hashed on
    file handle.  This table will be used by future commits to
    search for an open based on file handle more efficiently.
    
    (cherry picked from commit 3f7e14ad93454476bb11b4b8de5b41930d13312e)
---
 sys/fs/nfs/nfsclstate.h        |  7 +++++++
 sys/fs/nfsclient/nfs_clstate.c | 15 ++++++++++++++-
 2 files changed, 21 insertions(+), 1 deletion(-)

diff --git a/sys/fs/nfs/nfsclstate.h b/sys/fs/nfs/nfsclstate.h
index e17be74c5581..898a7de391dc 100644
--- a/sys/fs/nfs/nfsclstate.h
+++ b/sys/fs/nfs/nfsclstate.h
@@ -42,6 +42,7 @@ LIST_HEAD(nfsclhead, nfsclclient);
 LIST_HEAD(nfsclownerhead, nfsclowner);
 TAILQ_HEAD(nfscldeleghead, nfscldeleg);
 LIST_HEAD(nfscldeleghash, nfscldeleg);
+LIST_HEAD(nfsclopenhash, nfsclopen);
 TAILQ_HEAD(nfscllayouthead, nfscllayout);
 LIST_HEAD(nfscllayouthash, nfscllayout);
 LIST_HEAD(nfsclflayouthead, nfsclflayout);
@@ -50,6 +51,10 @@ LIST_HEAD(nfsclrecalllayouthead, nfsclrecalllayout);
 #define	NFSCLDELEGHASHSIZE	256
 #define	NFSCLDELEGHASH(c, f, l)							\
 	(&((c)->nfsc_deleghash[ncl_hash((f), (l)) % NFSCLDELEGHASHSIZE]))
+#define	NFSCLOPENHASHSIZE	256
+#define	NFSCLOPENHASHFUNC(f, l) (ncl_hash((f), (l)) % NFSCLOPENHASHSIZE)
+#define	NFSCLOPENHASH(c, f, l)							\
+	(&((c)->nfsc_openhash[NFSCLOPENHASHFUNC((f), (l))]))
 #define	NFSCLLAYOUTHASHSIZE	256
 #define	NFSCLLAYOUTHASH(c, f, l)						\
 	(&((c)->nfsc_layouthash[ncl_hash((f), (l)) % NFSCLLAYOUTHASHSIZE]))
@@ -104,6 +109,7 @@ struct nfsclclient {
 	struct nfsclownerhead	nfsc_owner;
 	struct nfscldeleghead	nfsc_deleg;
 	struct nfscldeleghash	nfsc_deleghash[NFSCLDELEGHASHSIZE];
+	struct nfsclopenhash	nfsc_openhash[NFSCLOPENHASHSIZE];
 	struct nfscllayouthead	nfsc_layout;
 	struct nfscllayouthash	nfsc_layouthash[NFSCLLAYOUTHASHSIZE];
 	struct nfscldevinfohead	nfsc_devinfo;
@@ -183,6 +189,7 @@ struct nfscldeleg {
  */
 struct nfsclopen {
 	LIST_ENTRY(nfsclopen)	nfso_list;
+	LIST_ENTRY(nfsclopen)	nfso_hash;
 	struct nfscllockownerhead nfso_lock;
 	nfsv4stateid_t		nfso_stateid;
 	struct nfsclowner	*nfso_own;
diff --git a/sys/fs/nfsclient/nfs_clstate.c b/sys/fs/nfsclient/nfs_clstate.c
index 1ed3630ce6e7..a8eace2ffd0b 100644
--- a/sys/fs/nfsclient/nfs_clstate.c
+++ b/sys/fs/nfsclient/nfs_clstate.c
@@ -240,9 +240,11 @@ nfscl_open(vnode_t vp, u_int8_t *nfhp, int fhlen, u_int32_t amode, int usedeleg,
 	 */
 	nowp = malloc(sizeof (struct nfsclowner),
 	    M_NFSCLOWNER, M_WAITOK);
-	if (nfhp != NULL)
+	if (nfhp != NULL) {
 	    nop = malloc(sizeof (struct nfsclopen) +
 		fhlen - 1, M_NFSCLOPEN, M_WAITOK);
+	    nop->nfso_hash.le_prev = NULL;
+	}
 	ret = nfscl_getcl(vp->v_mount, cred, p, 1, &clp);
 	if (ret != 0) {
 		free(nowp, M_NFSCLOWNER);
@@ -412,6 +414,8 @@ nfscl_newopen(struct nfsclclient *clp, struct nfscldeleg *dp,
 				dp->nfsdl_timestamp = NFSD_MONOSEC + 120;
 				nfsstatsv1.cllocalopens++;
 			} else {
+				LIST_INSERT_HEAD(NFSCLOPENHASH(clp, fhp, fhlen),
+				    nop, nfso_hash);
 				nfsstatsv1.clopens++;
 			}
 			LIST_INSERT_HEAD(&owp->nfsow_open, nop, nfso_list);
@@ -837,6 +841,8 @@ nfscl_getcl(struct mount *mp, struct ucred *cred, NFSPROC_T *p,
 		LIST_INIT(&clp->nfsc_devinfo);
 		for (i = 0; i < NFSCLDELEGHASHSIZE; i++)
 			LIST_INIT(&clp->nfsc_deleghash[i]);
+		for (i = 0; i < NFSCLOPENHASHSIZE; i++)
+			LIST_INIT(&clp->nfsc_openhash[i]);
 		for (i = 0; i < NFSCLLAYOUTHASHSIZE; i++)
 			LIST_INIT(&clp->nfsc_layouthash[i]);
 		clp->nfsc_flags = NFSCLFLAGS_INITED;
@@ -1475,6 +1481,8 @@ nfscl_freeopen(struct nfsclopen *op, int local)
 {
 
 	LIST_REMOVE(op, nfso_list);
+	if (op->nfso_hash.le_prev != NULL)
+		LIST_REMOVE(op, nfso_hash);
 	nfscl_freealllocks(&op->nfso_lock, local);
 	free(op, M_NFSCLOPEN);
 	if (local)
@@ -1706,6 +1714,8 @@ nfscl_expireclient(struct nfsclclient *clp, struct nfsmount *nmp,
 			    LIST_REMOVE(op, nfso_list);
 			    op->nfso_own = towp;
 			    LIST_INSERT_HEAD(&towp->nfsow_open, op, nfso_list);
+			    LIST_INSERT_HEAD(NFSCLOPENHASH(clp, op->nfso_fh,
+				op->nfso_fhlen), op, nfso_hash);
 			    nfsstatsv1.cllocalopens--;
 			    nfsstatsv1.clopens++;
 			}
@@ -1714,6 +1724,8 @@ nfscl_expireclient(struct nfsclclient *clp, struct nfsmount *nmp,
 			LIST_REMOVE(owp, nfsow_list);
 			owp->nfsow_clp = clp;
 			LIST_INSERT_HEAD(&clp->nfsc_owner, owp, nfsow_list);
+			LIST_INSERT_HEAD(NFSCLOPENHASH(clp, op->nfso_fh,
+			    op->nfso_fhlen), op, nfso_hash);
 			nfsstatsv1.cllocalopenowners--;
 			nfsstatsv1.clopenowners++;
 			nfsstatsv1.cllocalopens--;
@@ -4198,6 +4210,7 @@ nfscl_moveopen(vnode_t vp, struct nfsclclient *clp, struct nfsmount *nmp,
 	np = VTONFS(vp);
 	nop = malloc(sizeof (struct nfsclopen) +
 	    lop->nfso_fhlen - 1, M_NFSCLOPEN, M_WAITOK);
+	nop->nfso_hash.le_prev = NULL;
 	newone = 0;
 	nfscl_newopen(clp, NULL, &owp, NULL, &op, &nop, owp->nfsow_owner,
 	    lop->nfso_fh, lop->nfso_fhlen, cred, &newone);