Merge tag 'nfsd-4.15' of git://linux-nfs.org/~bfields/linux

Pull nfsd updates from Bruce Fields:
 "Lots of good bugfixes, including:

   -  fix a number of races in the NFSv4+ state code

   -  fix some shutdown crashes in multiple-network-namespace cases

   -  relax our 4.1 session limits; if you've an artificially low limit
      to the number of 4.1 clients that can mount simultaneously, try
      upgrading"

* tag 'nfsd-4.15' of git://linux-nfs.org/~bfields/linux: (22 commits)
  SUNRPC: Improve ordering of transport processing
  nfsd: deal with revoked delegations appropriately
  svcrdma: Enqueue after setting XPT_CLOSE in completion handlers
  nfsd: use nfs->ns.inum as net ID
  rpc: remove some BUG()s
  svcrdma: Preserve CB send buffer across retransmits
  nfds: avoid gettimeofday for nfssvc_boot time
  fs, nfsd: convert nfs4_file.fi_ref from atomic_t to refcount_t
  fs, nfsd: convert nfs4_cntl_odstate.co_odcount from atomic_t to refcount_t
  fs, nfsd: convert nfs4_stid.sc_count from atomic_t to refcount_t
  lockd: double unregister of inetaddr notifiers
  nfsd4: catch some false session retries
  nfsd4: fix cached replies to solo SEQUENCE compounds
  sunrcp: make function _svc_create_xprt static
  SUNRPC: Fix tracepoint storage issues with svc_recv and svc_rqst_status
  nfsd: use ARRAY_SIZE
  nfsd: give out fewer session slots as limit approaches
  nfsd: increase DRC cache limit
  nfsd: remove unnecessary nofilehandle checks
  nfs_common: convert int to bool
  ...

fs/lockd/svc.c

@@ -369,6 +369,7 @@ static int lockd_start_svc(struct svc_serv *serv)
 		printk(KERN_WARNING
 			"lockd_up: svc_rqst allocation failed, error=%d\n",
 			error);
+		lockd_unregister_notifiers();
 		goto out_rqst;
 	}
 
@@ -459,13 +460,16 @@ int lockd_up(struct net *net)
 	}
 
 	error = lockd_up_net(serv, net);
-	if (error < 0)
-		goto err_net;
+	if (error < 0) {
+		lockd_unregister_notifiers();
+		goto err_put;
+	}
 
 	error = lockd_start_svc(serv);
-	if (error < 0)
-		goto err_start;
-
+	if (error < 0) {
+		lockd_down_net(serv, net);
+		goto err_put;
+	}
 	nlmsvc_users++;
 	/*
 	 * Note: svc_serv structures have an initial use count of 1,
@@ -476,12 +480,6 @@ int lockd_up(struct net *net)
 err_create:
 	mutex_unlock(&nlmsvc_mutex);
 	return error;
-
-err_start:
-	lockd_down_net(serv, net);
-err_net:
-	lockd_unregister_notifiers();
-	goto err_put;
 }
 EXPORT_SYMBOL_GPL(lockd_up);
 

fs/nfs_common/grace.c

@@ -55,14 +55,7 @@ locks_end_grace(struct lock_manager *lm)
 }
 EXPORT_SYMBOL_GPL(locks_end_grace);
 
-/**
- * locks_in_grace
- *
- * Lock managers call this function to determine when it is OK for them
- * to answer ordinary lock requests, and when they should accept only
- * lock reclaims.
- */
-int
+static bool
 __state_in_grace(struct net *net, bool open)
 {
 	struct list_head *grace_list = net_generic(net, grace_net_id);
@@ -78,15 +71,22 @@ __state_in_grace(struct net *net, bool open)
 	return false;
 }
 
-int locks_in_grace(struct net *net)
+/**
+ * locks_in_grace
+ *
+ * Lock managers call this function to determine when it is OK for them
+ * to answer ordinary lock requests, and when they should accept only
+ * lock reclaims.
+ */
+bool locks_in_grace(struct net *net)
 {
-	return __state_in_grace(net, 0);
+	return __state_in_grace(net, false);
 }
 EXPORT_SYMBOL_GPL(locks_in_grace);
 
-int opens_in_grace(struct net *net)
+bool opens_in_grace(struct net *net)
 {
-	return __state_in_grace(net, 1);
+	return __state_in_grace(net, true);
 }
 EXPORT_SYMBOL_GPL(opens_in_grace);
 

fs/nfsd/fault_inject.c

@@ -12,6 +12,7 @@
 #include <linux/nsproxy.h>
 #include <linux/sunrpc/addr.h>
 #include <linux/uaccess.h>
+#include <linux/kernel.h>
 
 #include "state.h"
 #include "netns.h"
@@ -126,8 +127,6 @@ static struct nfsd_fault_inject_op inject_ops[] = {
 	},
 };
 
-#define NUM_INJECT_OPS (sizeof(inject_ops)/sizeof(struct nfsd_fault_inject_op))
-
 int nfsd_fault_inject_init(void)
 {
 	unsigned int i;
@@ -138,7 +137,7 @@ int nfsd_fault_inject_init(void)
 	if (!debug_dir)
 		goto fail;
 
-	for (i = 0; i < NUM_INJECT_OPS; i++) {
+	for (i = 0; i < ARRAY_SIZE(inject_ops); i++) {
 		op = &inject_ops[i];
 		if (!debugfs_create_file(op->file, mode, debug_dir, op, &fops_nfsd))
 			goto fail;

fs/nfsd/netns.h

@@ -107,7 +107,7 @@ struct nfsd_net {
 	bool lockd_up;
 
 	/* Time of server startup */
-	struct timeval nfssvc_boot;
+	struct timespec64 nfssvc_boot;
 
 	/*
 	 * Max number of connections this nfsd container will allow. Defaults

fs/nfsd/nfs3xdr.c

@@ -748,8 +748,9 @@ nfs3svc_encode_writeres(struct svc_rqst *rqstp, __be32 *p)
 	if (resp->status == 0) {
 		*p++ = htonl(resp->count);
 		*p++ = htonl(resp->committed);
-		*p++ = htonl(nn->nfssvc_boot.tv_sec);
-		*p++ = htonl(nn->nfssvc_boot.tv_usec);
+		/* unique identifier, y2038 overflow can be ignored */
+		*p++ = htonl((u32)nn->nfssvc_boot.tv_sec);
+		*p++ = htonl(nn->nfssvc_boot.tv_nsec);
 	}
 	return xdr_ressize_check(rqstp, p);
 }
@@ -1119,8 +1120,9 @@ nfs3svc_encode_commitres(struct svc_rqst *rqstp, __be32 *p)
 	p = encode_wcc_data(rqstp, p, &resp->fh);
 	/* Write verifier */
 	if (resp->status == 0) {
-		*p++ = htonl(nn->nfssvc_boot.tv_sec);
-		*p++ = htonl(nn->nfssvc_boot.tv_usec);
+		/* unique identifier, y2038 overflow can be ignored */
+		*p++ = htonl((u32)nn->nfssvc_boot.tv_sec);
+		*p++ = htonl(nn->nfssvc_boot.tv_nsec);
 	}
 	return xdr_ressize_check(rqstp, p);
 }

fs/nfsd/nfs4layouts.c

@@ -336,7 +336,7 @@ nfsd4_recall_file_layout(struct nfs4_layout_stateid *ls)
 
 	trace_layout_recall(&ls->ls_stid.sc_stateid);
 
-	atomic_inc(&ls->ls_stid.sc_count);
+	refcount_inc(&ls->ls_stid.sc_count);
 	nfsd4_run_cb(&ls->ls_recall);
 
 out_unlock:
@@ -441,7 +441,7 @@ nfsd4_insert_layout(struct nfsd4_layoutget *lgp, struct nfs4_layout_stateid *ls)
 			goto done;
 	}
 
-	atomic_inc(&ls->ls_stid.sc_count);
+	refcount_inc(&ls->ls_stid.sc_count);
 	list_add_tail(&new->lo_perstate, &ls->ls_layouts);
 	new = NULL;
 done:

fs/nfsd/nfs4proc.c

@@ -485,9 +485,6 @@ static __be32
 nfsd4_getfh(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
 	    union nfsd4_op_u *u)
 {
-	if (!cstate->current_fh.fh_dentry)
-		return nfserr_nofilehandle;
-
 	u->getfh = &cstate->current_fh;
 	return nfs_ok;
 }
@@ -535,9 +532,6 @@ static __be32
 nfsd4_savefh(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
 	     union nfsd4_op_u *u)
 {
-	if (!cstate->current_fh.fh_dentry)
-		return nfserr_nofilehandle;
-
 	fh_dup2(&cstate->save_fh, &cstate->current_fh);
 	if (HAS_STATE_ID(cstate, CURRENT_STATE_ID_FLAG)) {
 		memcpy(&cstate->save_stateid, &cstate->current_stateid, sizeof(stateid_t));
@@ -570,10 +564,11 @@ static void gen_boot_verifier(nfs4_verifier *verifier, struct net *net)
 
 	/*
 	 * This is opaque to client, so no need to byte-swap. Use
-	 * __force to keep sparse happy
+	 * __force to keep sparse happy. y2038 time_t overflow is
+	 * irrelevant in this usage.
 	 */
 	verf[0] = (__force __be32)nn->nfssvc_boot.tv_sec;
-	verf[1] = (__force __be32)nn->nfssvc_boot.tv_usec;
+	verf[1] = (__force __be32)nn->nfssvc_boot.tv_nsec;
 	memcpy(verifier->data, verf, sizeof(verifier->data));
 }
 
@@ -703,10 +698,8 @@ nfsd4_link(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
 	   union nfsd4_op_u *u)
 {
 	struct nfsd4_link *link = &u->link;
-	__be32 status = nfserr_nofilehandle;
+	__be32 status;
 
-	if (!cstate->save_fh.fh_dentry)
-		return status;
 	status = nfsd_link(rqstp, &cstate->current_fh,
 			   link->li_name, link->li_namelen, &cstate->save_fh);
 	if (!status)
@@ -850,10 +843,8 @@ nfsd4_rename(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
 	     union nfsd4_op_u *u)
 {
 	struct nfsd4_rename *rename = &u->rename;
-	__be32 status = nfserr_nofilehandle;
+	__be32 status;
 
-	if (!cstate->save_fh.fh_dentry)
-		return status;
 	if (opens_in_grace(SVC_NET(rqstp)) &&
 		!(cstate->save_fh.fh_export->ex_flags & NFSEXP_NOSUBTREECHECK))
 		return nfserr_grace;

fs/nfsd/nfssvc.c

@@ -447,7 +447,7 @@ void nfsd_reset_versions(void)
  */
 static void set_max_drc(void)
 {
-	#define NFSD_DRC_SIZE_SHIFT	10
+	#define NFSD_DRC_SIZE_SHIFT	7
 	nfsd_drc_max_mem = (nr_free_buffer_pages()
 					>> NFSD_DRC_SIZE_SHIFT) * PAGE_SIZE;
 	nfsd_drc_mem_used = 0;
@@ -517,7 +517,7 @@ int nfsd_create_serv(struct net *net)
 		register_inet6addr_notifier(&nfsd_inet6addr_notifier);
 #endif
 	}
-	do_gettimeofday(&nn->nfssvc_boot);		/* record boot time */
+	ktime_get_real_ts64(&nn->nfssvc_boot); /* record boot time */
 	return 0;
 }
 

fs/nfsd/state.h

@@ -36,6 +36,7 @@
 #define _NFSD4_STATE_H
 
 #include <linux/idr.h>
+#include <linux/refcount.h>
 #include <linux/sunrpc/svc_xprt.h>
 #include "nfsfh.h"
 
@@ -83,7 +84,7 @@ struct nfsd4_callback_ops {
  * fields that are of general use to any stateid.
  */
 struct nfs4_stid {
-	atomic_t		sc_count;
+	refcount_t		sc_count;
 #define NFS4_OPEN_STID 1
 #define NFS4_LOCK_STID 2
 #define NFS4_DELEG_STID 4
@@ -169,11 +170,13 @@ static inline struct nfs4_delegation *delegstateid(struct nfs4_stid *s)
 struct nfsd4_slot {
 	u32	sl_seqid;
 	__be32	sl_status;
+	struct svc_cred sl_cred;
 	u32	sl_datalen;
 	u16	sl_opcnt;
 #define NFSD4_SLOT_INUSE	(1 << 0)
 #define NFSD4_SLOT_CACHETHIS	(1 << 1)
 #define NFSD4_SLOT_INITIALIZED	(1 << 2)
+#define NFSD4_SLOT_CACHED	(1 << 3)
 	u8	sl_flags;
 	char	sl_data[];
 };
@@ -465,7 +468,7 @@ struct nfs4_clnt_odstate {
 	struct nfs4_client	*co_client;
 	struct nfs4_file	*co_file;
 	struct list_head	co_perfile;
-	atomic_t		co_odcount;
+	refcount_t		co_odcount;
 };
 
 /*
@@ -481,7 +484,7 @@ struct nfs4_clnt_odstate {
  * the global state_lock spinlock.
  */
 struct nfs4_file {
-	atomic_t		fi_ref;
+	refcount_t		fi_ref;
 	spinlock_t		fi_lock;
 	struct hlist_node       fi_hash;	/* hash on fi_fhandle */
 	struct list_head        fi_stateids;
@@ -634,7 +637,7 @@ struct nfs4_file *find_file(struct knfsd_fh *fh);
 void put_nfs4_file(struct nfs4_file *fi);
 static inline void get_nfs4_file(struct nfs4_file *fi)
 {
-	atomic_inc(&fi->fi_ref);
+	refcount_inc(&fi->fi_ref);
 }
 struct file *find_any_file(struct nfs4_file *f);
 

fs/nfsd/xdr4.h

@@ -649,9 +649,18 @@ static inline bool nfsd4_is_solo_sequence(struct nfsd4_compoundres *resp)
 	return resp->opcnt == 1 && args->ops[0].opnum == OP_SEQUENCE;
 }
 
-static inline bool nfsd4_not_cached(struct nfsd4_compoundres *resp)
+/*
+ * The session reply cache only needs to cache replies that the client
+ * actually asked us to.  But it's almost free for us to cache compounds
+ * consisting of only a SEQUENCE op, so we may as well cache those too.
+ * Also, the protocol doesn't give us a convenient response in the case
+ * of a replay of a solo SEQUENCE op that wasn't cached
+ * (RETRY_UNCACHED_REP can only be returned in the second op of a
+ * compound).
+ */
+static inline bool nfsd4_cache_this(struct nfsd4_compoundres *resp)
 {
-	return !(resp->cstate.slot->sl_flags & NFSD4_SLOT_CACHETHIS)
+	return (resp->cstate.slot->sl_flags & NFSD4_SLOT_CACHETHIS)
 		|| nfsd4_is_solo_sequence(resp);
 }
 

include/linux/fs.h

@@ -971,8 +971,8 @@ struct lock_manager {
 struct net;
 void locks_start_grace(struct net *, struct lock_manager *);
 void locks_end_grace(struct lock_manager *);
-int locks_in_grace(struct net *);
-int opens_in_grace(struct net *);
+bool locks_in_grace(struct net *);
+bool opens_in_grace(struct net *);
 
 /* that will die - we need it for nfs_lock_info */
 #include <linux/nfs_fs_i.h>

include/linux/sunrpc/svc.h

@@ -47,6 +47,7 @@ struct svc_pool {
 	struct svc_pool_stats	sp_stats;	/* statistics on pool operation */
 #define	SP_TASK_PENDING		(0)		/* still work to do even if no
 						 * xprt is queued. */
+#define SP_CONGESTED		(1)
 	unsigned long		sp_flags;
 } ____cacheline_aligned_in_smp;
 

include/trace/events/sunrpc.h

@@ -486,20 +486,22 @@ TRACE_EVENT(svc_recv,
 	TP_ARGS(rqst, status),
 
 	TP_STRUCT__entry(
-		__field(struct sockaddr *, addr)
 		__field(u32, xid)
 		__field(int, status)
 		__field(unsigned long, flags)
+		__dynamic_array(unsigned char, addr, rqst->rq_addrlen)
 	),
 
 	TP_fast_assign(
-		__entry->addr = (struct sockaddr *)&rqst->rq_addr;
 		__entry->xid = status > 0 ? be32_to_cpu(rqst->rq_xid) : 0;
 		__entry->status = status;
 		__entry->flags = rqst->rq_flags;
+		memcpy(__get_dynamic_array(addr),
+			&rqst->rq_addr, rqst->rq_addrlen);
 	),
 
-	TP_printk("addr=%pIScp xid=0x%08x status=%d flags=%s", __entry->addr,
+	TP_printk("addr=%pIScp xid=0x%08x status=%d flags=%s",
+			(struct sockaddr *)__get_dynamic_array(addr),
 			__entry->xid, __entry->status,
 			show_rqstp_flags(__entry->flags))
 );
@@ -544,22 +546,23 @@ DECLARE_EVENT_CLASS(svc_rqst_status,
 	TP_ARGS(rqst, status),
 
 	TP_STRUCT__entry(
-		__field(struct sockaddr *, addr)
 		__field(u32, xid)
-		__field(int, dropme)
 		__field(int, status)
 		__field(unsigned long, flags)
+		__dynamic_array(unsigned char, addr, rqst->rq_addrlen)
 	),
 
 	TP_fast_assign(
-		__entry->addr = (struct sockaddr *)&rqst->rq_addr;
 		__entry->xid = be32_to_cpu(rqst->rq_xid);
 		__entry->status = status;
 		__entry->flags = rqst->rq_flags;
+		memcpy(__get_dynamic_array(addr),
+			&rqst->rq_addr, rqst->rq_addrlen);
 	),
 
 	TP_printk("addr=%pIScp rq_xid=0x%08x status=%d flags=%s",
-		__entry->addr, __entry->xid,
+		(struct sockaddr *)__get_dynamic_array(addr),
+		__entry->xid,
 		__entry->status, show_rqstp_flags(__entry->flags))
 );
 

net/sunrpc/auth_gss/svcauth_gss.c

@@ -855,11 +855,13 @@ unwrap_integ_data(struct svc_rqst *rqstp, struct xdr_buf *buf, u32 seq, struct g
 		return stat;
 	if (integ_len > buf->len)
 		return stat;
-	if (xdr_buf_subsegment(buf, &integ_buf, 0, integ_len))
-		BUG();
+	if (xdr_buf_subsegment(buf, &integ_buf, 0, integ_len)) {
+		WARN_ON_ONCE(1);
+		return stat;
+	}
 	/* copy out mic... */
 	if (read_u32_from_xdr_buf(buf, integ_len, &mic.len))
-		BUG();
+		return stat;
 	if (mic.len > RPC_MAX_AUTH_SIZE)
 		return stat;
 	mic.data = kmalloc(mic.len, GFP_KERNEL);
@@ -1611,8 +1613,10 @@ svcauth_gss_wrap_resp_integ(struct svc_rqst *rqstp)
 	BUG_ON(integ_len % 4);
 	*p++ = htonl(integ_len);
 	*p++ = htonl(gc->gc_seq);
-	if (xdr_buf_subsegment(resbuf, &integ_buf, integ_offset, integ_len))
-		BUG();
+	if (xdr_buf_subsegment(resbuf, &integ_buf, integ_offset, integ_len)) {
+		WARN_ON_ONCE(1);
+		goto out_err;
+	}
 	if (resbuf->tail[0].iov_base == NULL) {
 		if (resbuf->head[0].iov_len + RPC_MAX_AUTH_SIZE > PAGE_SIZE)
 			goto out_err;

net/sunrpc/svc_xprt.c

@@ -250,7 +250,7 @@ void svc_add_new_perm_xprt(struct svc_serv *serv, struct svc_xprt *new)
 	svc_xprt_received(new);
 }
 
-int _svc_create_xprt(struct svc_serv *serv, const char *xprt_name,
+static int _svc_create_xprt(struct svc_serv *serv, const char *xprt_name,
 			    struct net *net, const int family,
 			    const unsigned short port, int flags)
 {
@@ -380,7 +380,6 @@ void svc_xprt_do_enqueue(struct svc_xprt *xprt)
 	struct svc_pool *pool;
 	struct svc_rqst	*rqstp = NULL;
 	int cpu;
-	bool queued = false;
 
 	if (!svc_xprt_has_something_to_do(xprt))
 		goto out;
@@ -401,58 +400,25 @@ void svc_xprt_do_enqueue(struct svc_xprt *xprt)
 
 	atomic_long_inc(&pool->sp_stats.packets);
 
-redo_search:
+	dprintk("svc: transport %p put into queue\n", xprt);
+	spin_lock_bh(&pool->sp_lock);
+	list_add_tail(&xprt->xpt_ready, &pool->sp_sockets);
+	pool->sp_stats.sockets_queued++;
+	spin_unlock_bh(&pool->sp_lock);
+
 	/* find a thread for this xprt */
 	rcu_read_lock();
 	list_for_each_entry_rcu(rqstp, &pool->sp_all_threads, rq_all) {
-		/* Do a lockless check first */
-		if (test_bit(RQ_BUSY, &rqstp->rq_flags))
+		if (test_and_set_bit(RQ_BUSY, &rqstp->rq_flags))
 			continue;
-
-		/*
-		 * Once the xprt has been queued, it can only be dequeued by
-		 * the task that intends to service it. All we can do at that
-		 * point is to try to wake this thread back up so that it can
-		 * do so.
-		 */
-		if (!queued) {
-			spin_lock_bh(&rqstp->rq_lock);
-			if (test_and_set_bit(RQ_BUSY, &rqstp->rq_flags)) {
-				/* already busy, move on... */
-				spin_unlock_bh(&rqstp->rq_lock);
-				continue;
-			}
-
-			/* this one will do */
-			rqstp->rq_xprt = xprt;
-			svc_xprt_get(xprt);
-			spin_unlock_bh(&rqstp->rq_lock);
-		}
-		rcu_read_unlock();
-
 		atomic_long_inc(&pool->sp_stats.threads_woken);
 		wake_up_process(rqstp->rq_task);
-		put_cpu();
-		goto out;
-	}
-	rcu_read_unlock();
-
-	/*
-	 * We didn't find an idle thread to use, so we need to queue the xprt.
-	 * Do so and then search again. If we find one, we can't hook this one
-	 * up to it directly but we can wake the thread up in the hopes that it
-	 * will pick it up once it searches for a xprt to service.
-	 */
-	if (!queued) {
-		queued = true;
-		dprintk("svc: transport %p put into queue\n", xprt);
-		spin_lock_bh(&pool->sp_lock);
-		list_add_tail(&xprt->xpt_ready, &pool->sp_sockets);
-		pool->sp_stats.sockets_queued++;
-		spin_unlock_bh(&pool->sp_lock);
-		goto redo_search;
+		goto out_unlock;
 	}
+	set_bit(SP_CONGESTED, &pool->sp_flags);
 	rqstp = NULL;
+out_unlock:
+	rcu_read_unlock();
 	put_cpu();
 out:
 	trace_svc_xprt_do_enqueue(xprt, rqstp);
@@ -721,38 +687,25 @@ rqst_should_sleep(struct svc_rqst *rqstp)
 
 static struct svc_xprt *svc_get_next_xprt(struct svc_rqst *rqstp, long timeout)
 {
-	struct svc_xprt *xprt;
 	struct svc_pool		*pool = rqstp->rq_pool;
 	long			time_left = 0;
 
 	/* rq_xprt should be clear on entry */
 	WARN_ON_ONCE(rqstp->rq_xprt);
 
-	/* Normally we will wait up to 5 seconds for any required
-	 * cache information to be provided.
-	 */
-	rqstp->rq_chandle.thread_wait = 5*HZ;
-
-	xprt = svc_xprt_dequeue(pool);
-	if (xprt) {
-		rqstp->rq_xprt = xprt;
-
-		/* As there is a shortage of threads and this request
-		 * had to be queued, don't allow the thread to wait so
-		 * long for cache updates.
-		 */
-		rqstp->rq_chandle.thread_wait = 1*HZ;
-		clear_bit(SP_TASK_PENDING, &pool->sp_flags);
-		return xprt;
-	}
+	rqstp->rq_xprt = svc_xprt_dequeue(pool);
+	if (rqstp->rq_xprt)
+		goto out_found;
 
 	/*
 	 * We have to be able to interrupt this wait
 	 * to bring down the daemons ...
 	 */
 	set_current_state(TASK_INTERRUPTIBLE);
+	smp_mb__before_atomic();
+	clear_bit(SP_CONGESTED, &pool->sp_flags);
 	clear_bit(RQ_BUSY, &rqstp->rq_flags);
-	smp_mb();
+	smp_mb__after_atomic();
 
 	if (likely(rqst_should_sleep(rqstp)))
 		time_left = schedule_timeout(timeout);
@@ -761,13 +714,11 @@ static struct svc_xprt *svc_get_next_xprt(struct svc_rqst *rqstp, long timeout)
 
 	try_to_freeze();
 
-	spin_lock_bh(&rqstp->rq_lock);
 	set_bit(RQ_BUSY, &rqstp->rq_flags);
-	spin_unlock_bh(&rqstp->rq_lock);
-
-	xprt = rqstp->rq_xprt;
-	if (xprt != NULL)
-		return xprt;
+	smp_mb__after_atomic();
+	rqstp->rq_xprt = svc_xprt_dequeue(pool);
+	if (rqstp->rq_xprt)
+		goto out_found;
 
 	if (!time_left)
 		atomic_long_inc(&pool->sp_stats.threads_timedout);
@@ -775,6 +726,15 @@ static struct svc_xprt *svc_get_next_xprt(struct svc_rqst *rqstp, long timeout)
 	if (signalled() || kthread_should_stop())
 		return ERR_PTR(-EINTR);
 	return ERR_PTR(-EAGAIN);
+out_found:
+	/* Normally we will wait up to 5 seconds for any required
+	 * cache information to be provided.
+	 */
+	if (!test_bit(SP_CONGESTED, &pool->sp_flags))
+		rqstp->rq_chandle.thread_wait = 5*HZ;
+	else
+		rqstp->rq_chandle.thread_wait = 1*HZ;
+	return rqstp->rq_xprt;
 }
 
 static void svc_add_new_temp_xprt(struct svc_serv *serv, struct svc_xprt *newxpt)

net/sunrpc/xprtrdma/svc_rdma_backchannel.c

@@ -133,6 +133,10 @@ static int svc_rdma_bc_sendto(struct svcxprt_rdma *rdma,
 	if (ret)
 		goto out_err;
 
+	/* Bump page refcnt so Send completion doesn't release
+	 * the rq_buffer before all retransmits are complete.
+	 */
+	get_page(virt_to_page(rqst->rq_buffer));
 	ret = svc_rdma_post_send_wr(rdma, ctxt, 1, 0);
 	if (ret)
 		goto out_unmap;
@@ -165,7 +169,6 @@ xprt_rdma_bc_allocate(struct rpc_task *task)
 		return -EINVAL;
 	}
 
-	/* svc_rdma_sendto releases this page */
 	page = alloc_page(RPCRDMA_DEF_GFP);
 	if (!page)
 		return -ENOMEM;
@@ -184,6 +187,7 @@ xprt_rdma_bc_free(struct rpc_task *task)
 {
 	struct rpc_rqst *rqst = task->tk_rqstp;
 
+	put_page(virt_to_page(rqst->rq_buffer));
 	kfree(rqst->rq_rbuffer);
 }
 

net/sunrpc/xprtrdma/svc_rdma_transport.c

@@ -290,6 +290,7 @@ static void qp_event_handler(struct ib_event *event, void *context)
 			ib_event_msg(event->event), event->event,
 			event->element.qp);
 		set_bit(XPT_CLOSE, &xprt->xpt_flags);
+		svc_xprt_enqueue(xprt);
 		break;
 	}
 }
@@ -322,8 +323,7 @@ static void svc_rdma_wc_receive(struct ib_cq *cq, struct ib_wc *wc)
 	set_bit(XPT_DATA, &xprt->sc_xprt.xpt_flags);
 	if (test_bit(RDMAXPRT_CONN_PENDING, &xprt->sc_flags))
 		goto out;
-	svc_xprt_enqueue(&xprt->sc_xprt);
-	goto out;
+	goto out_enqueue;
 
 flushed:
 	if (wc->status != IB_WC_WR_FLUSH_ERR)
@@ -333,6 +333,8 @@ static void svc_rdma_wc_receive(struct ib_cq *cq, struct ib_wc *wc)
 	set_bit(XPT_CLOSE, &xprt->sc_xprt.xpt_flags);
 	svc_rdma_put_context(ctxt, 1);
 
+out_enqueue:
+	svc_xprt_enqueue(&xprt->sc_xprt);
 out:
 	svc_xprt_put(&xprt->sc_xprt);
 }
@@ -358,6 +360,7 @@ void svc_rdma_wc_send(struct ib_cq *cq, struct ib_wc *wc)
 
 	if (unlikely(wc->status != IB_WC_SUCCESS)) {
 		set_bit(XPT_CLOSE, &xprt->sc_xprt.xpt_flags);
+		svc_xprt_enqueue(&xprt->sc_xprt);
 		if (wc->status != IB_WC_WR_FLUSH_ERR)
 			pr_err("svcrdma: Send: %s (%u/0x%x)\n",
 			       ib_wc_status_msg(wc->status),
@@ -569,8 +572,10 @@ static int rdma_listen_handler(struct rdma_cm_id *cma_id,
 	case RDMA_CM_EVENT_DEVICE_REMOVAL:
 		dprintk("svcrdma: Device removal xprt=%p, cm_id=%p\n",
 			xprt, cma_id);
-		if (xprt)
+		if (xprt) {
 			set_bit(XPT_CLOSE, &xprt->sc_xprt.xpt_flags);
+			svc_xprt_enqueue(&xprt->sc_xprt);
+		}
 		break;
 
 	default: