Merge branch 'writeback'

fs/nfs/callback_proc.c

@@ -51,7 +51,7 @@ __be32 nfs4_callback_getattr(void *argp, void *resp,
 		goto out_iput;
 	res->size = i_size_read(inode);
 	res->change_attr = delegation->change_attr;
-	if (nfsi->nrequests != 0)
+	if (nfs_have_writebacks(inode))
 		res->change_attr++;
 	res->ctime = inode->i_ctime;
 	res->mtime = inode->i_mtime;

fs/nfs/delegation.c

@@ -1089,7 +1089,7 @@ bool nfs4_delegation_flush_on_close(const struct inode *inode)
 	delegation = rcu_dereference(nfsi->delegation);
 	if (delegation == NULL || !(delegation->type & FMODE_WRITE))
 		goto out;
-	if (nfsi->nrequests < delegation->pagemod_limit)
+	if (atomic_long_read(&nfsi->nrequests) < delegation->pagemod_limit)
 		ret = false;
 out:
 	rcu_read_unlock();

fs/nfs/direct.c

@@ -616,13 +616,13 @@ nfs_direct_write_scan_commit_list(struct inode *inode,
 				  struct list_head *list,
 				  struct nfs_commit_info *cinfo)
 {
-	spin_lock(&cinfo->inode->i_lock);
+	mutex_lock(&NFS_I(cinfo->inode)->commit_mutex);
 #ifdef CONFIG_NFS_V4_1
 	if (cinfo->ds != NULL && cinfo->ds->nwritten != 0)
 		NFS_SERVER(inode)->pnfs_curr_ld->recover_commit_reqs(list, cinfo);
 #endif
 	nfs_scan_commit_list(&cinfo->mds->list, list, cinfo, 0);
-	spin_unlock(&cinfo->inode->i_lock);
+	mutex_unlock(&NFS_I(cinfo->inode)->commit_mutex);
 }
 
 static void nfs_direct_write_reschedule(struct nfs_direct_req *dreq)

fs/nfs/inode.c

@@ -1285,7 +1285,6 @@ static bool nfs_file_has_buffered_writers(struct nfs_inode *nfsi)
 
 static unsigned long nfs_wcc_update_inode(struct inode *inode, struct nfs_fattr *fattr)
 {
-	struct nfs_inode *nfsi = NFS_I(inode);
 	unsigned long ret = 0;
 
 	if ((fattr->valid & NFS_ATTR_FATTR_PRECHANGE)
@@ -1315,7 +1314,7 @@ static unsigned long nfs_wcc_update_inode(struct inode *inode, struct nfs_fattr
 	if ((fattr->valid & NFS_ATTR_FATTR_PRESIZE)
 			&& (fattr->valid & NFS_ATTR_FATTR_SIZE)
 			&& i_size_read(inode) == nfs_size_to_loff_t(fattr->pre_size)
-			&& nfsi->nrequests == 0) {
+			&& !nfs_have_writebacks(inode)) {
 		i_size_write(inode, nfs_size_to_loff_t(fattr->size));
 		ret |= NFS_INO_INVALID_ATTR;
 	}
@@ -1823,7 +1822,7 @@ static int nfs_update_inode(struct inode *inode, struct nfs_fattr *fattr)
 		if (new_isize != cur_isize) {
 			/* Do we perhaps have any outstanding writes, or has
 			 * the file grown beyond our last write? */
-			if (nfsi->nrequests == 0 || new_isize > cur_isize) {
+			if (!nfs_have_writebacks(inode) || new_isize > cur_isize) {
 				i_size_write(inode, new_isize);
 				if (!have_writers)
 					invalid |= NFS_INO_INVALID_ATTR|NFS_INO_INVALID_DATA;
@@ -2012,10 +2011,11 @@ static void init_once(void *foo)
 	INIT_LIST_HEAD(&nfsi->access_cache_entry_lru);
 	INIT_LIST_HEAD(&nfsi->access_cache_inode_lru);
 	INIT_LIST_HEAD(&nfsi->commit_info.list);
-	nfsi->nrequests = 0;
-	nfsi->commit_info.ncommit = 0;
+	atomic_long_set(&nfsi->nrequests, 0);
+	atomic_long_set(&nfsi->commit_info.ncommit, 0);
 	atomic_set(&nfsi->commit_info.rpcs_out, 0);
 	init_rwsem(&nfsi->rmdir_sem);
+	mutex_init(&nfsi->commit_mutex);
 	nfs4_init_once(nfsi);
 }
 

fs/nfs/pagelist.c

@@ -134,19 +134,14 @@ EXPORT_SYMBOL_GPL(nfs_async_iocounter_wait);
 /*
  * nfs_page_group_lock - lock the head of the page group
  * @req - request in group that is to be locked
- * @nonblock - if true don't block waiting for lock
  *
- * this lock must be held if modifying the page group list
+ * this lock must be held when traversing or modifying the page
+ * group list
  *
- * return 0 on success, < 0 on error: -EDELAY if nonblocking or the
- * result from wait_on_bit_lock
- *
- * NOTE: calling with nonblock=false should always have set the
- *       lock bit (see fs/buffer.c and other uses of wait_on_bit_lock
- *       with TASK_UNINTERRUPTIBLE), so there is no need to check the result.
+ * return 0 on success, < 0 on error
  */
 int
-nfs_page_group_lock(struct nfs_page *req, bool nonblock)
+nfs_page_group_lock(struct nfs_page *req)
 {
 	struct nfs_page *head = req->wb_head;
 
@@ -155,35 +150,10 @@ nfs_page_group_lock(struct nfs_page *req, bool nonblock)
 	if (!test_and_set_bit(PG_HEADLOCK, &head->wb_flags))
 		return 0;
 
-	if (!nonblock) {
 	set_bit(PG_CONTENDED1, &head->wb_flags);
 	smp_mb__after_atomic();
 	return wait_on_bit_lock(&head->wb_flags, PG_HEADLOCK,
 				TASK_UNINTERRUPTIBLE);
-	}
-
-	return -EAGAIN;
-}
-
-/*
- * nfs_page_group_lock_wait - wait for the lock to clear, but don't grab it
- * @req - a request in the group
- *
- * This is a blocking call to wait for the group lock to be cleared.
- */
-void
-nfs_page_group_lock_wait(struct nfs_page *req)
-{
-	struct nfs_page *head = req->wb_head;
-
-	WARN_ON_ONCE(head != head->wb_head);
-
-	if (!test_bit(PG_HEADLOCK, &head->wb_flags))
-		return;
-	set_bit(PG_CONTENDED1, &head->wb_flags);
-	smp_mb__after_atomic();
-	wait_on_bit(&head->wb_flags, PG_HEADLOCK,
-		TASK_UNINTERRUPTIBLE);
 }
 
 /*
@@ -246,7 +216,7 @@ bool nfs_page_group_sync_on_bit(struct nfs_page *req, unsigned int bit)
 {
 	bool ret;
 
-	nfs_page_group_lock(req, false);
+	nfs_page_group_lock(req);
 	ret = nfs_page_group_sync_on_bit_locked(req, bit);
 	nfs_page_group_unlock(req);
 
@@ -288,9 +258,7 @@ nfs_page_group_init(struct nfs_page *req, struct nfs_page *prev)
 			inode = page_file_mapping(req->wb_page)->host;
 			set_bit(PG_INODE_REF, &req->wb_flags);
 			kref_get(&req->wb_kref);
-			spin_lock(&inode->i_lock);
-			NFS_I(inode)->nrequests++;
-			spin_unlock(&inode->i_lock);
+			atomic_long_inc(&NFS_I(inode)->nrequests);
 		}
 	}
 }
@@ -306,14 +274,11 @@ static void
 nfs_page_group_destroy(struct kref *kref)
 {
 	struct nfs_page *req = container_of(kref, struct nfs_page, wb_kref);
+	struct nfs_page *head = req->wb_head;
 	struct nfs_page *tmp, *next;
 
-	/* subrequests must release the ref on the head request */
-	if (req->wb_head != req)
-		nfs_release_request(req->wb_head);
-
 	if (!nfs_page_group_sync_on_bit(req, PG_TEARDOWN))
-		return;
+		goto out;
 
 	tmp = req;
 	do {
@@ -324,6 +289,10 @@ nfs_page_group_destroy(struct kref *kref)
 		nfs_free_request(tmp);
 		tmp = next;
 	} while (tmp != req);
+out:
+	/* subrequests must release the ref on the head request */
+	if (head != req)
+		nfs_release_request(head);
 }
 
 /**
@@ -465,6 +434,7 @@ void nfs_release_request(struct nfs_page *req)
 {
 	kref_put(&req->wb_kref, nfs_page_group_destroy);
 }
+EXPORT_SYMBOL_GPL(nfs_release_request);
 
 /**
  * nfs_wait_on_request - Wait for a request to complete.
@@ -483,6 +453,7 @@ nfs_wait_on_request(struct nfs_page *req)
 	return wait_on_bit_io(&req->wb_flags, PG_BUSY,
 			      TASK_UNINTERRUPTIBLE);
 }
+EXPORT_SYMBOL_GPL(nfs_wait_on_request);
 
 /*
  * nfs_generic_pg_test - determine if requests can be coalesced
@@ -1036,7 +1007,7 @@ static int __nfs_pageio_add_request(struct nfs_pageio_descriptor *desc,
 	unsigned int bytes_left = 0;
 	unsigned int offset, pgbase;
 
-	nfs_page_group_lock(req, false);
+	nfs_page_group_lock(req);
 
 	subreq = req;
 	bytes_left = subreq->wb_bytes;
@@ -1058,7 +1029,7 @@ static int __nfs_pageio_add_request(struct nfs_pageio_descriptor *desc,
 			if (mirror->pg_recoalesce)
 				return 0;
 			/* retry add_request for this subreq */
-			nfs_page_group_lock(req, false);
+			nfs_page_group_lock(req);
 			continue;
 		}
 
@@ -1155,7 +1126,7 @@ int nfs_pageio_add_request(struct nfs_pageio_descriptor *desc,
 
 	for (midx = 0; midx < desc->pg_mirror_count; midx++) {
 		if (midx) {
-			nfs_page_group_lock(req, false);
+			nfs_page_group_lock(req);
 
 			/* find the last request */
 			for (lastreq = req->wb_head;

fs/nfs/pnfs.c

@@ -529,47 +529,6 @@ pnfs_put_lseg(struct pnfs_layout_segment *lseg)
 }
 EXPORT_SYMBOL_GPL(pnfs_put_lseg);
 
-static void pnfs_free_lseg_async_work(struct work_struct *work)
-{
-	struct pnfs_layout_segment *lseg;
-	struct pnfs_layout_hdr *lo;
-
-	lseg = container_of(work, struct pnfs_layout_segment, pls_work);
-	lo = lseg->pls_layout;
-
-	pnfs_free_lseg(lseg);
-	pnfs_put_layout_hdr(lo);
-}
-
-static void pnfs_free_lseg_async(struct pnfs_layout_segment *lseg)
-{
-	INIT_WORK(&lseg->pls_work, pnfs_free_lseg_async_work);
-	schedule_work(&lseg->pls_work);
-}
-
-void
-pnfs_put_lseg_locked(struct pnfs_layout_segment *lseg)
-{
-	if (!lseg)
-		return;
-
-	assert_spin_locked(&lseg->pls_layout->plh_inode->i_lock);
-
-	dprintk("%s: lseg %p ref %d valid %d\n", __func__, lseg,
-		atomic_read(&lseg->pls_refcount),
-		test_bit(NFS_LSEG_VALID, &lseg->pls_flags));
-	if (atomic_dec_and_test(&lseg->pls_refcount)) {
-		struct pnfs_layout_hdr *lo = lseg->pls_layout;
-		if (test_bit(NFS_LSEG_VALID, &lseg->pls_flags))
-			return;
-		pnfs_layout_remove_lseg(lo, lseg);
-		if (!pnfs_cache_lseg_for_layoutreturn(lo, lseg)) {
-			pnfs_get_layout_hdr(lo);
-			pnfs_free_lseg_async(lseg);
-		}
-	}
-}
-
 /*
  * is l2 fully contained in l1?
  *   start1                             end1

fs/nfs/pnfs.h

@@ -67,7 +67,6 @@ struct pnfs_layout_segment {
 	u32 pls_seq;
 	unsigned long pls_flags;
 	struct pnfs_layout_hdr *pls_layout;
-	struct work_struct pls_work;
 };
 
 enum pnfs_try_status {
@@ -230,7 +229,6 @@ extern int nfs4_proc_layoutreturn(struct nfs4_layoutreturn *lrp, bool sync);
 /* pnfs.c */
 void pnfs_get_layout_hdr(struct pnfs_layout_hdr *lo);
 void pnfs_put_lseg(struct pnfs_layout_segment *lseg);
-void pnfs_put_lseg_locked(struct pnfs_layout_segment *lseg);
 
 void set_pnfs_layoutdriver(struct nfs_server *, const struct nfs_fh *, struct nfs_fsinfo *);
 void unset_pnfs_layoutdriver(struct nfs_server *);

fs/nfs/pnfs_nfs.c

@@ -83,7 +83,7 @@ pnfs_generic_clear_request_commit(struct nfs_page *req,
 	}
 out:
 	nfs_request_remove_commit_list(req, cinfo);
-	pnfs_put_lseg_locked(freeme);
+	pnfs_put_lseg(freeme);
 }
 EXPORT_SYMBOL_GPL(pnfs_generic_clear_request_commit);
 
@@ -91,21 +91,30 @@ static int
 pnfs_generic_transfer_commit_list(struct list_head *src, struct list_head *dst,
 				  struct nfs_commit_info *cinfo, int max)
 {
-	struct nfs_page *req, *tmp;
+	struct nfs_page *req;
 	int ret = 0;
 
-	list_for_each_entry_safe(req, tmp, src, wb_list) {
-		if (!nfs_lock_request(req))
-			continue;
+	while(!list_empty(src)) {
+		req = list_first_entry(src, struct nfs_page, wb_list);
+
 		kref_get(&req->wb_kref);
-		if (cond_resched_lock(&cinfo->inode->i_lock))
-			list_safe_reset_next(req, tmp, wb_list);
+		if (!nfs_lock_request(req)) {
+			int status;
+			mutex_unlock(&NFS_I(cinfo->inode)->commit_mutex);
+			status = nfs_wait_on_request(req);
+			nfs_release_request(req);
+			mutex_lock(&NFS_I(cinfo->inode)->commit_mutex);
+			if (status < 0)
+				break;
+			continue;
+		}
 		nfs_request_remove_commit_list(req, cinfo);
 		clear_bit(PG_COMMIT_TO_DS, &req->wb_flags);
 		nfs_list_add_request(req, dst);
 		ret++;
 		if ((ret == max) && !cinfo->dreq)
 			break;
+		cond_resched();
 	}
 	return ret;
 }
@@ -119,7 +128,7 @@ pnfs_generic_scan_ds_commit_list(struct pnfs_commit_bucket *bucket,
 	struct list_head *dst = &bucket->committing;
 	int ret;
 
-	lockdep_assert_held(&cinfo->inode->i_lock);
+	lockdep_assert_held(&NFS_I(cinfo->inode)->commit_mutex);
 	ret = pnfs_generic_transfer_commit_list(src, dst, cinfo, max);
 	if (ret) {
 		cinfo->ds->nwritten -= ret;
@@ -127,7 +136,7 @@ pnfs_generic_scan_ds_commit_list(struct pnfs_commit_bucket *bucket,
 		if (bucket->clseg == NULL)
 			bucket->clseg = pnfs_get_lseg(bucket->wlseg);
 		if (list_empty(src)) {
-			pnfs_put_lseg_locked(bucket->wlseg);
+			pnfs_put_lseg(bucket->wlseg);
 			bucket->wlseg = NULL;
 		}
 	}
@@ -142,7 +151,7 @@ int pnfs_generic_scan_commit_lists(struct nfs_commit_info *cinfo,
 {
 	int i, rv = 0, cnt;
 
-	lockdep_assert_held(&cinfo->inode->i_lock);
+	lockdep_assert_held(&NFS_I(cinfo->inode)->commit_mutex);
 	for (i = 0; i < cinfo->ds->nbuckets && max != 0; i++) {
 		cnt = pnfs_generic_scan_ds_commit_list(&cinfo->ds->buckets[i],
 						       cinfo, max);
@@ -162,7 +171,7 @@ void pnfs_generic_recover_commit_reqs(struct list_head *dst,
 	int nwritten;
 	int i;
 
-	lockdep_assert_held(&cinfo->inode->i_lock);
+	lockdep_assert_held(&NFS_I(cinfo->inode)->commit_mutex);
 restart:
 	for (i = 0, b = cinfo->ds->buckets; i < cinfo->ds->nbuckets; i++, b++) {
 		nwritten = pnfs_generic_transfer_commit_list(&b->written,
@@ -953,12 +962,12 @@ pnfs_layout_mark_request_commit(struct nfs_page *req,
 	struct list_head *list;
 	struct pnfs_commit_bucket *buckets;
 
-	spin_lock(&cinfo->inode->i_lock);
+	mutex_lock(&NFS_I(cinfo->inode)->commit_mutex);
 	buckets = cinfo->ds->buckets;
 	list = &buckets[ds_commit_idx].written;
 	if (list_empty(list)) {
 		if (!pnfs_is_valid_lseg(lseg)) {
-			spin_unlock(&cinfo->inode->i_lock);
+			mutex_unlock(&NFS_I(cinfo->inode)->commit_mutex);
 			cinfo->completion_ops->resched_write(cinfo, req);
 			return;
 		}
@@ -975,7 +984,7 @@ pnfs_layout_mark_request_commit(struct nfs_page *req,
 	cinfo->ds->nwritten++;
 
 	nfs_request_add_commit_list_locked(req, list, cinfo);
-	spin_unlock(&cinfo->inode->i_lock);
+	mutex_unlock(&NFS_I(cinfo->inode)->commit_mutex);
 	nfs_mark_page_unstable(req->wb_page, cinfo);
 }
 EXPORT_SYMBOL_GPL(pnfs_layout_mark_request_commit);

include/linux/nfs_fs.h

@@ -154,7 +154,7 @@ struct nfs_inode {
 	 */
 	__be32			cookieverf[2];
 
-	unsigned long		nrequests;
+	atomic_long_t		nrequests;
 	struct nfs_mds_commit_info commit_info;
 
 	/* Open contexts for shared mmap writes */
@@ -163,6 +163,7 @@ struct nfs_inode {
 	/* Readers: in-flight sillydelete RPC calls */
 	/* Writers: rmdir */
 	struct rw_semaphore	rmdir_sem;
+	struct mutex		commit_mutex;
 
 #if IS_ENABLED(CONFIG_NFS_V4)
 	struct nfs4_cached_acl	*nfs4_acl;
@@ -510,7 +511,7 @@ extern void nfs_commit_free(struct nfs_commit_data *data);
 static inline int
 nfs_have_writebacks(struct inode *inode)
 {
-	return NFS_I(inode)->nrequests != 0;
+	return atomic_long_read(&NFS_I(inode)->nrequests) != 0;
 }
 
 /*

include/linux/nfs_page.h

@@ -139,8 +139,7 @@ extern size_t nfs_generic_pg_test(struct nfs_pageio_descriptor *desc,
 extern  int nfs_wait_on_request(struct nfs_page *);
 extern	void nfs_unlock_request(struct nfs_page *req);
 extern	void nfs_unlock_and_release_request(struct nfs_page *);
-extern int nfs_page_group_lock(struct nfs_page *, bool);
-extern void nfs_page_group_lock_wait(struct nfs_page *);
+extern int nfs_page_group_lock(struct nfs_page *);
 extern void nfs_page_group_unlock(struct nfs_page *);
 extern bool nfs_page_group_sync_on_bit(struct nfs_page *, unsigned int);
 extern bool nfs_async_iocounter_wait(struct rpc_task *, struct nfs_lock_context *);

include/linux/nfs_xdr.h

@@ -1476,7 +1476,7 @@ struct nfs_pgio_header {
 
 struct nfs_mds_commit_info {
 	atomic_t rpcs_out;
-	unsigned long		ncommit;
+	atomic_long_t		ncommit;
 	struct list_head	list;
 };
 

include/linux/sunrpc/sched.h

@@ -139,6 +139,8 @@ struct rpc_task_setup {
 #define RPC_TASK_RUNNING	0
 #define RPC_TASK_QUEUED		1
 #define RPC_TASK_ACTIVE		2
+#define RPC_TASK_MSG_RECV	3
+#define RPC_TASK_MSG_RECV_WAIT	4
 
 #define RPC_IS_RUNNING(t)	test_bit(RPC_TASK_RUNNING, &(t)->tk_runstate)
 #define rpc_set_running(t)	set_bit(RPC_TASK_RUNNING, &(t)->tk_runstate)

include/linux/sunrpc/xprt.h

@@ -232,6 +232,7 @@ struct rpc_xprt {
 	 */
 	spinlock_t		transport_lock;	/* lock transport info */
 	spinlock_t		reserve_lock;	/* lock slot table */
+	spinlock_t		recv_lock;	/* lock receive list */
 	u32			xid;		/* Next XID value to use */
 	struct rpc_task *	snd_task;	/* Task blocked in send */
 	struct svc_xprt		*bc_xprt;	/* NFSv4.1 backchannel */
@@ -372,6 +373,8 @@ void			xprt_write_space(struct rpc_xprt *xprt);
 void			xprt_adjust_cwnd(struct rpc_xprt *xprt, struct rpc_task *task, int result);
 struct rpc_rqst *	xprt_lookup_rqst(struct rpc_xprt *xprt, __be32 xid);
 void			xprt_complete_rqst(struct rpc_task *task, int copied);
+void			xprt_pin_rqst(struct rpc_rqst *req);
+void			xprt_unpin_rqst(struct rpc_rqst *req);
 void			xprt_release_rqst_cong(struct rpc_task *task);
 void			xprt_disconnect_done(struct rpc_xprt *xprt);
 void			xprt_force_disconnect(struct rpc_xprt *xprt);

net/sunrpc/backchannel_rqst.c

@@ -171,10 +171,10 @@ int xprt_setup_bc(struct rpc_xprt *xprt, unsigned int min_reqs)
 	/*
 	 * Add the temporary list to the backchannel preallocation list
 	 */
-	spin_lock_bh(&xprt->bc_pa_lock);
+	spin_lock(&xprt->bc_pa_lock);
 	list_splice(&tmp_list, &xprt->bc_pa_list);
 	xprt_inc_alloc_count(xprt, min_reqs);
-	spin_unlock_bh(&xprt->bc_pa_lock);
+	spin_unlock(&xprt->bc_pa_lock);
 
 	dprintk("RPC:       setup backchannel transport done\n");
 	return 0;

net/sunrpc/svcsock.c

@@ -1001,7 +1001,7 @@ static int receive_cb_reply(struct svc_sock *svsk, struct svc_rqst *rqstp)
 
 	if (!bc_xprt)
 		return -EAGAIN;
-	spin_lock_bh(&bc_xprt->transport_lock);
+	spin_lock(&bc_xprt->recv_lock);
 	req = xprt_lookup_rqst(bc_xprt, xid);
 	if (!req)
 		goto unlock_notfound;
@@ -1019,7 +1019,7 @@ static int receive_cb_reply(struct svc_sock *svsk, struct svc_rqst *rqstp)
 	memcpy(dst->iov_base, src->iov_base, src->iov_len);
 	xprt_complete_rqst(req->rq_task, rqstp->rq_arg.len);
 	rqstp->rq_arg.len = 0;
-	spin_unlock_bh(&bc_xprt->transport_lock);
+	spin_unlock(&bc_xprt->recv_lock);
 	return 0;
 unlock_notfound:
 	printk(KERN_NOTICE
@@ -1028,7 +1028,7 @@ static int receive_cb_reply(struct svc_sock *svsk, struct svc_rqst *rqstp)
 		__func__, ntohl(calldir),
 		bc_xprt, ntohl(xid));
 unlock_eagain:
-	spin_unlock_bh(&bc_xprt->transport_lock);
+	spin_unlock(&bc_xprt->recv_lock);
 	return -EAGAIN;
 }
 

net/sunrpc/xprt.c

@@ -844,6 +844,48 @@ struct rpc_rqst *xprt_lookup_rqst(struct rpc_xprt *xprt, __be32 xid)
 }
 EXPORT_SYMBOL_GPL(xprt_lookup_rqst);
 
+/**
+ * xprt_pin_rqst - Pin a request on the transport receive list
+ * @req: Request to pin
+ *
+ * Caller must ensure this is atomic with the call to xprt_lookup_rqst()
+ * so should be holding the xprt transport lock.
+ */
+void xprt_pin_rqst(struct rpc_rqst *req)
+{
+	set_bit(RPC_TASK_MSG_RECV, &req->rq_task->tk_runstate);
+}
+
+/**
+ * xprt_unpin_rqst - Unpin a request on the transport receive list
+ * @req: Request to pin
+ *
+ * Caller should be holding the xprt transport lock.
+ */
+void xprt_unpin_rqst(struct rpc_rqst *req)
+{
+	struct rpc_task *task = req->rq_task;
+
+	clear_bit(RPC_TASK_MSG_RECV, &task->tk_runstate);
+	if (test_bit(RPC_TASK_MSG_RECV_WAIT, &task->tk_runstate))
+		wake_up_bit(&task->tk_runstate, RPC_TASK_MSG_RECV);
+}
+
+static void xprt_wait_on_pinned_rqst(struct rpc_rqst *req)
+__must_hold(&req->rq_xprt->recv_lock)
+{
+	struct rpc_task *task = req->rq_task;
+	
+	if (task && test_bit(RPC_TASK_MSG_RECV, &task->tk_runstate)) {
+		spin_unlock(&req->rq_xprt->recv_lock);
+		set_bit(RPC_TASK_MSG_RECV_WAIT, &task->tk_runstate);
+		wait_on_bit(&task->tk_runstate, RPC_TASK_MSG_RECV,
+				TASK_UNINTERRUPTIBLE);
+		clear_bit(RPC_TASK_MSG_RECV_WAIT, &task->tk_runstate);
+		spin_lock(&req->rq_xprt->recv_lock);
+	}
+}
+
 static void xprt_update_rtt(struct rpc_task *task)
 {
 	struct rpc_rqst *req = task->tk_rqstp;
@@ -966,13 +1008,13 @@ void xprt_transmit(struct rpc_task *task)
 			/*
 			 * Add to the list only if we're expecting a reply
 			 */
-			spin_lock_bh(&xprt->transport_lock);
 			/* Update the softirq receive buffer */
 			memcpy(&req->rq_private_buf, &req->rq_rcv_buf,
 					sizeof(req->rq_private_buf));
 			/* Add request to the receive list */
+			spin_lock(&xprt->recv_lock);
 			list_add_tail(&req->rq_list, &xprt->recv);
-			spin_unlock_bh(&xprt->transport_lock);
+			spin_unlock(&xprt->recv_lock);
 			xprt_reset_majortimeo(req);
 			/* Turn off autodisconnect */
 			del_singleshot_timer_sync(&xprt->timer);
@@ -1287,12 +1329,16 @@ void xprt_release(struct rpc_task *task)
 		task->tk_ops->rpc_count_stats(task, task->tk_calldata);
 	else if (task->tk_client)
 		rpc_count_iostats(task, task->tk_client->cl_metrics);
+	spin_lock(&xprt->recv_lock);
+	if (!list_empty(&req->rq_list)) {
+		list_del(&req->rq_list);
+		xprt_wait_on_pinned_rqst(req);
+	}
+	spin_unlock(&xprt->recv_lock);
 	spin_lock_bh(&xprt->transport_lock);
 	xprt->ops->release_xprt(xprt, task);
 	if (xprt->ops->release_request)
 		xprt->ops->release_request(task);
-	if (!list_empty(&req->rq_list))
-		list_del(&req->rq_list);
 	xprt->last_used = jiffies;
 	xprt_schedule_autodisconnect(xprt);
 	spin_unlock_bh(&xprt->transport_lock);
@@ -1318,6 +1364,7 @@ static void xprt_init(struct rpc_xprt *xprt, struct net *net)
 
 	spin_lock_init(&xprt->transport_lock);
 	spin_lock_init(&xprt->reserve_lock);
+	spin_lock_init(&xprt->recv_lock);
 
 	INIT_LIST_HEAD(&xprt->free);
 	INIT_LIST_HEAD(&xprt->recv);

net/sunrpc/xprtrdma/rpc_rdma.c

@@ -1051,7 +1051,7 @@ rpcrdma_reply_handler(struct work_struct *work)
 	 * RPC completion while holding the transport lock to ensure
 	 * the rep, rqst, and rq_task pointers remain stable.
 	 */
-	spin_lock_bh(&xprt->transport_lock);
+	spin_lock(&xprt->recv_lock);
 	rqst = xprt_lookup_rqst(xprt, headerp->rm_xid);
 	if (!rqst)
 		goto out_norqst;
@@ -1136,7 +1136,7 @@ rpcrdma_reply_handler(struct work_struct *work)
 		xprt_release_rqst_cong(rqst->rq_task);
 
 	xprt_complete_rqst(rqst->rq_task, status);
-	spin_unlock_bh(&xprt->transport_lock);
+	spin_unlock(&xprt->recv_lock);
 	dprintk("RPC:       %s: xprt_complete_rqst(0x%p, 0x%p, %d)\n",
 		__func__, xprt, rqst, status);
 	return;
@@ -1187,12 +1187,12 @@ rpcrdma_reply_handler(struct work_struct *work)
 	r_xprt->rx_stats.bad_reply_count++;
 	goto out;
 
-/* The req was still available, but by the time the transport_lock
+/* The req was still available, but by the time the recv_lock
  * was acquired, the rqst and task had been released. Thus the RPC
  * has already been terminated.
  */
 out_norqst:
-	spin_unlock_bh(&xprt->transport_lock);
+	spin_unlock(&xprt->recv_lock);
 	rpcrdma_buffer_put(req);
 	dprintk("RPC:       %s: race, no rqst left for req %p\n",
 		__func__, req);

net/sunrpc/xprtrdma/svc_rdma_backchannel.c

@@ -52,7 +52,7 @@ int svc_rdma_handle_bc_reply(struct rpc_xprt *xprt, __be32 *rdma_resp,
 	if (src->iov_len < 24)
 		goto out_shortreply;
 
-	spin_lock_bh(&xprt->transport_lock);
+	spin_lock(&xprt->recv_lock);
 	req = xprt_lookup_rqst(xprt, xid);
 	if (!req)
 		goto out_notfound;
@@ -69,17 +69,20 @@ int svc_rdma_handle_bc_reply(struct rpc_xprt *xprt, __be32 *rdma_resp,
 	else if (credits > r_xprt->rx_buf.rb_bc_max_requests)
 		credits = r_xprt->rx_buf.rb_bc_max_requests;
 
+	spin_lock_bh(&xprt->transport_lock);
 	cwnd = xprt->cwnd;
 	xprt->cwnd = credits << RPC_CWNDSHIFT;
 	if (xprt->cwnd > cwnd)
 		xprt_release_rqst_cong(req->rq_task);
+	spin_unlock_bh(&xprt->transport_lock);
+
 
 	ret = 0;
 	xprt_complete_rqst(req->rq_task, rcvbuf->len);
 	rcvbuf->len = 0;
 
 out_unlock:
-	spin_unlock_bh(&xprt->transport_lock);
+	spin_unlock(&xprt->recv_lock);
 out:
 	return ret;
 

net/sunrpc/xprtsock.c

@@ -969,10 +969,12 @@ static void xs_local_data_read_skb(struct rpc_xprt *xprt,
 		return;
 
 	/* Look up and lock the request corresponding to the given XID */
-	spin_lock_bh(&xprt->transport_lock);
+	spin_lock(&xprt->recv_lock);
 	rovr = xprt_lookup_rqst(xprt, *xp);
 	if (!rovr)
 		goto out_unlock;
+	xprt_pin_rqst(rovr);
+	spin_unlock(&xprt->recv_lock);
 	task = rovr->rq_task;
 
 	copied = rovr->rq_private_buf.buflen;
@@ -981,13 +983,16 @@ static void xs_local_data_read_skb(struct rpc_xprt *xprt,
 
 	if (xs_local_copy_to_xdr(&rovr->rq_private_buf, skb)) {
 		dprintk("RPC:       sk_buff copy failed\n");
-		goto out_unlock;
+		spin_lock(&xprt->recv_lock);
+		goto out_unpin;
 	}
 
+	spin_lock(&xprt->recv_lock);
 	xprt_complete_rqst(task, copied);
-
+out_unpin:
+	xprt_unpin_rqst(rovr);
  out_unlock:
-	spin_unlock_bh(&xprt->transport_lock);
+	spin_unlock(&xprt->recv_lock);
 }
 
 static void xs_local_data_receive(struct sock_xprt *transport)
@@ -1050,10 +1055,12 @@ static void xs_udp_data_read_skb(struct rpc_xprt *xprt,
 		return;
 
 	/* Look up and lock the request corresponding to the given XID */
-	spin_lock_bh(&xprt->transport_lock);
+	spin_lock(&xprt->recv_lock);
 	rovr = xprt_lookup_rqst(xprt, *xp);
 	if (!rovr)
 		goto out_unlock;
+	xprt_pin_rqst(rovr);
+	spin_unlock(&xprt->recv_lock);
 	task = rovr->rq_task;
 
 	if ((copied = rovr->rq_private_buf.buflen) > repsize)
@@ -1062,16 +1069,21 @@ static void xs_udp_data_read_skb(struct rpc_xprt *xprt,
 	/* Suck it into the iovec, verify checksum if not done by hw. */
 	if (csum_partial_copy_to_xdr(&rovr->rq_private_buf, skb)) {
 		__UDPX_INC_STATS(sk, UDP_MIB_INERRORS);
-		goto out_unlock;
+		spin_lock(&xprt->recv_lock);
+		goto out_unpin;
 	}
 
 	__UDPX_INC_STATS(sk, UDP_MIB_INDATAGRAMS);
 
+	spin_lock_bh(&xprt->transport_lock);
 	xprt_adjust_cwnd(xprt, task, copied);
+	spin_unlock_bh(&xprt->transport_lock);
+	spin_lock(&xprt->recv_lock);
 	xprt_complete_rqst(task, copied);
-
+out_unpin:
+	xprt_unpin_rqst(rovr);
  out_unlock:
-	spin_unlock_bh(&xprt->transport_lock);
+	spin_unlock(&xprt->recv_lock);
 }
 
 static void xs_udp_data_receive(struct sock_xprt *transport)
@@ -1277,25 +1289,12 @@ static inline void xs_tcp_read_common(struct rpc_xprt *xprt,
 	}
 
 	len = desc->count;
-	if (len > transport->tcp_reclen - transport->tcp_offset) {
-		struct xdr_skb_reader my_desc;
-
-		len = transport->tcp_reclen - transport->tcp_offset;
-		memcpy(&my_desc, desc, sizeof(my_desc));
-		my_desc.count = len;
-		r = xdr_partial_copy_from_skb(rcvbuf, transport->tcp_copied,
-					  &my_desc, xdr_skb_read_bits);
-		desc->count -= r;
-		desc->offset += r;
-	} else
+	if (len > transport->tcp_reclen - transport->tcp_offset)
+		desc->count = transport->tcp_reclen - transport->tcp_offset;
 	r = xdr_partial_copy_from_skb(rcvbuf, transport->tcp_copied,
 					  desc, xdr_skb_read_bits);
 
-	if (r > 0) {
-		transport->tcp_copied += r;
-		transport->tcp_offset += r;
-	}
-	if (r != len) {
+	if (desc->count) {
 		/* Error when copying to the receive buffer,
 		 * usually because we weren't able to allocate
 		 * additional buffer pages. All we can do now
@@ -1315,6 +1314,10 @@ static inline void xs_tcp_read_common(struct rpc_xprt *xprt,
 		return;
 	}
 
+	transport->tcp_copied += r;
+	transport->tcp_offset += r;
+	desc->count = len - r;
+
 	dprintk("RPC:       XID %08x read %zd bytes\n",
 			ntohl(transport->tcp_xid), r);
 	dprintk("RPC:       xprt = %p, tcp_copied = %lu, tcp_offset = %u, "
@@ -1343,21 +1346,24 @@ static inline int xs_tcp_read_reply(struct rpc_xprt *xprt,
 	dprintk("RPC:       read reply XID %08x\n", ntohl(transport->tcp_xid));
 
 	/* Find and lock the request corresponding to this xid */
-	spin_lock_bh(&xprt->transport_lock);
+	spin_lock(&xprt->recv_lock);
 	req = xprt_lookup_rqst(xprt, transport->tcp_xid);
 	if (!req) {
 		dprintk("RPC:       XID %08x request not found!\n",
 				ntohl(transport->tcp_xid));
-		spin_unlock_bh(&xprt->transport_lock);
+		spin_unlock(&xprt->recv_lock);
 		return -1;
 	}
+	xprt_pin_rqst(req);
+	spin_unlock(&xprt->recv_lock);
 
 	xs_tcp_read_common(xprt, desc, req);
 
+	spin_lock(&xprt->recv_lock);
 	if (!(transport->tcp_flags & TCP_RCV_COPY_DATA))
 		xprt_complete_rqst(req->rq_task, transport->tcp_copied);
-
-	spin_unlock_bh(&xprt->transport_lock);
+	xprt_unpin_rqst(req);
+	spin_unlock(&xprt->recv_lock);
 	return 0;
 }
 
@@ -1376,11 +1382,9 @@ static int xs_tcp_read_callback(struct rpc_xprt *xprt,
 				container_of(xprt, struct sock_xprt, xprt);
 	struct rpc_rqst *req;
 
-	/* Look up and lock the request corresponding to the given XID */
-	spin_lock_bh(&xprt->transport_lock);
+	/* Look up the request corresponding to the given XID */
 	req = xprt_lookup_bc_request(xprt, transport->tcp_xid);
 	if (req == NULL) {
-		spin_unlock_bh(&xprt->transport_lock);
 		printk(KERN_WARNING "Callback slot table overflowed\n");
 		xprt_force_disconnect(xprt);
 		return -1;
@@ -1391,7 +1395,6 @@ static int xs_tcp_read_callback(struct rpc_xprt *xprt,
 
 	if (!(transport->tcp_flags & TCP_RCV_COPY_DATA))
 		xprt_complete_bc_request(req, transport->tcp_copied);
-	spin_unlock_bh(&xprt->transport_lock);
 
 	return 0;
 }
@@ -1516,6 +1519,7 @@ static void xs_tcp_data_receive(struct sock_xprt *transport)
 		.arg.data = xprt,
 	};
 	unsigned long total = 0;
+	int loop;
 	int read = 0;
 
 	mutex_lock(&transport->recv_mutex);
@@ -1524,20 +1528,20 @@ static void xs_tcp_data_receive(struct sock_xprt *transport)
 		goto out;
 
 	/* We use rd_desc to pass struct xprt to xs_tcp_data_recv */
-	for (;;) {
+	for (loop = 0; loop < 64; loop++) {
 		lock_sock(sk);
 		read = tcp_read_sock(sk, &rd_desc, xs_tcp_data_recv);
 		if (read <= 0) {
 			clear_bit(XPRT_SOCK_DATA_READY, &transport->sock_state);
 			release_sock(sk);
-			if (!test_bit(XPRT_SOCK_DATA_READY, &transport->sock_state))
 			break;
-		} else {
+		}
 		release_sock(sk);
 		total += read;
-		}
 		rd_desc.count = 65536;
 	}
+	if (test_bit(XPRT_SOCK_DATA_READY, &transport->sock_state))
+		queue_work(xprtiod_workqueue, &transport->recv_worker);
 out:
 	mutex_unlock(&transport->recv_mutex);
 	trace_xs_tcp_data_ready(xprt, read, total);