Merge tag 'nfs-for-4.3-1' of git://git.linux-nfs.org/projects/trondmy/linux-nfs

Pull NFS client updates from Trond Myklebust:
"Highlights include:

Stable patches:
- Fix atomicity of pNFS commit list updates
- Fix NFSv4 handling of open(O_CREAT|O_EXCL|O_RDONLY)
- nfs_set_pgio_error sometimes misses errors
- Fix a thinko in xs_connect()
- Fix borkage in _same_data_server_addrs_locked()
- Fix a NULL pointer dereference of migration recovery ops for v4.2
client
- Don't let the ctime override attribute barriers.
- Revert "NFSv4: Remove incorrect check in can_open_delegated()"
- Ensure flexfiles pNFS driver updates the inode after write finishes
- flexfiles must not pollute the attribute cache with attrbutes from
the DS
- Fix a protocol error in layoutreturn
- Fix a protocol issue with NFSv4.1 CLOSE stateids

Bugfixes + cleanups
- pNFS blocks bugfixes from Christoph
- Various cleanups from Anna
- More fixes for delegation corner cases
- Don't fsync twice for O_SYNC/IS_SYNC files
- Fix pNFS and flexfiles layoutstats bugs
- pnfs/flexfiles: avoid duplicate tracking of mirror data
- pnfs: Fix layoutget/layoutreturn/return-on-close serialisation
issues
- pnfs/flexfiles: error handling retries a layoutget before fallback
to MDS

Features:
- Full support for the OPEN NFS4_CREATE_EXCLUSIVE4_1 mode from
Kinglong
- More RDMA client transport improvements from Chuck
- Removal of the deprecated ib_reg_phys_mr() and ib_rereg_phys_mr()
verbs from the SUNRPC, Lustre and core infiniband tree.
- Optimise away the close-to-open getattr if there is no cached data"

* tag 'nfs-for-4.3-1' of git://git.linux-nfs.org/projects/trondmy/linux-nfs: (108 commits)
NFSv4: Respect the server imposed limit on how many changes we may cache
NFSv4: Express delegation limit in units of pages
Revert "NFS: Make close(2) asynchronous when closing NFS O_DIRECT files"
NFS: Optimise away the close-to-open getattr if there is no cached data
NFSv4.1/flexfiles: Clean up ff_layout_write_done_cb/ff_layout_commit_done_cb
NFSv4.1/flexfiles: Mark the layout for return in ff_layout_io_track_ds_error()
nfs: Remove unneeded checking of the return value from scnprintf
nfs: Fix truncated client owner id without proto type
NFSv4.1/flexfiles: Mark layout for return if the mirrors are invalid
NFSv4.1/flexfiles: RW layouts are valid only if all mirrors are valid
NFSv4.1/flexfiles: Fix incorrect usage of pnfs_generic_mark_devid_invalid()
NFSv4.1/flexfiles: Fix freeing of mirrors
NFSv4.1/pNFS: Don't request a minimal read layout beyond the end of file
NFSv4.1/pnfs: Handle LAYOUTGET return values correctly
NFSv4.1/pnfs: Don't ask for a read layout for an empty file.
NFSv4.1: Fix a protocol issue with CLOSE stateids
NFSv4.1/flexfiles: Don't mark the entire deviceid as bad for file errors
SUNRPC: Prevent SYN+SYNACK+RST storms
SUNRPC: xs_reset_transport must mark the connection as disconnected
NFSv4.1/pnfs: Ensure layoutreturn reserves space for the opaque payload
...

Documentation/kernel-parameters.txt

···
 			The default parameter value of '0' causes the kernel
 			not to attempt recovery of lost locks.
 
+	nfs4.layoutstats_timer =
+			[NFSv4.2] Change the rate at which the kernel sends
+			layoutstats to the pNFS metadata server.
+
+			Setting this to value to 0 causes the kernel to use
+			whatever value is the default set by the layout
+			driver. A non-zero value sets the minimum interval
+			in seconds between layoutstats transmissions.
+
 	nfsd.nfs4_disable_idmapping=
 			[NFSv4] When set to the default of '1', the NFSv4
 			server will return only numeric uids and gids to

drivers/infiniband/core/verbs.c

···
 }
 EXPORT_SYMBOL(ib_get_dma_mr);
 
-struct ib_mr *ib_reg_phys_mr(struct ib_pd *pd,
-			     struct ib_phys_buf *phys_buf_array,
-			     int num_phys_buf,
-			     int mr_access_flags,
-			     u64 *iova_start)
-{
-	struct ib_mr *mr;
-	int err;
-
-	err = ib_check_mr_access(mr_access_flags);
-	if (err)
-		return ERR_PTR(err);
-
-	if (!pd->device->reg_phys_mr)
-		return ERR_PTR(-ENOSYS);
-
-	mr = pd->device->reg_phys_mr(pd, phys_buf_array, num_phys_buf,
-				     mr_access_flags, iova_start);
-
-	if (!IS_ERR(mr)) {
-		mr->device  = pd->device;
-		mr->pd      = pd;
-		mr->uobject = NULL;
-		atomic_inc(&pd->usecnt);
-		atomic_set(&mr->usecnt, 0);
-	}
-
-	return mr;
-}
-EXPORT_SYMBOL(ib_reg_phys_mr);
-
-int ib_rereg_phys_mr(struct ib_mr *mr,
-		     int mr_rereg_mask,
-		     struct ib_pd *pd,
-		     struct ib_phys_buf *phys_buf_array,
-		     int num_phys_buf,
-		     int mr_access_flags,
-		     u64 *iova_start)
-{
-	struct ib_pd *old_pd;
-	int ret;
-
-	ret = ib_check_mr_access(mr_access_flags);
-	if (ret)
-		return ret;
-
-	if (!mr->device->rereg_phys_mr)
-		return -ENOSYS;
-
-	if (atomic_read(&mr->usecnt))
-		return -EBUSY;
-
-	old_pd = mr->pd;
-
-	ret = mr->device->rereg_phys_mr(mr, mr_rereg_mask, pd,
-					phys_buf_array, num_phys_buf,
-					mr_access_flags, iova_start);
-
-	if (!ret && (mr_rereg_mask & IB_MR_REREG_PD)) {
-		atomic_dec(&old_pd->usecnt);
-		atomic_inc(&pd->usecnt);
-	}
-
-	return ret;
-}
-EXPORT_SYMBOL(ib_rereg_phys_mr);
-
 int ib_query_mr(struct ib_mr *mr, struct ib_mr_attr *mr_attr)
 {
 	return mr->device->query_mr ?

fs/nfs/blocklayout/blocklayout.h

···
 
 struct pnfs_block_dev;
 
-enum pnfs_block_volume_type {
-	PNFS_BLOCK_VOLUME_SIMPLE	= 0,
-	PNFS_BLOCK_VOLUME_SLICE		= 1,
-	PNFS_BLOCK_VOLUME_CONCAT	= 2,
-	PNFS_BLOCK_VOLUME_STRIPE	= 3,
-};
-
 #define PNFS_BLOCK_MAX_UUIDS	4
 #define PNFS_BLOCK_MAX_DEVICES	64
 
···
 			struct pnfs_block_dev_map *map);
 };
 
-enum exstate4 {
-	PNFS_BLOCK_READWRITE_DATA	= 0,
-	PNFS_BLOCK_READ_DATA		= 1,
-	PNFS_BLOCK_INVALID_DATA		= 2, /* mapped, but data is invalid */
-	PNFS_BLOCK_NONE_DATA		= 3  /* unmapped, it's a hole */
-};
-
 /* sector_t fields are all in 512-byte sectors */
 struct pnfs_block_extent {
 	union {
···
 	sector_t	be_f_offset;	/* the starting offset in the file */
 	sector_t	be_length;	/* the size of the extent */
 	sector_t	be_v_offset;	/* the starting offset in the volume */
-	enum exstate4	be_state;	/* the state of this extent */
+	enum pnfs_block_extent_state be_state;	/* the state of this extent */
 #define EXTENT_WRITTEN		1
 #define EXTENT_COMMITTING	2
 	unsigned int	be_tag;
 };
-
-/* on the wire size of the extent */
-#define BL_EXTENT_SIZE	(7 * sizeof(__be32) + NFS4_DEVICEID4_SIZE)
 
 struct pnfs_block_layout {
 	struct pnfs_layout_hdr	bl_layout;

fs/nfs/blocklayout/dev.c

···
 		kfree(dev->children);
 	} else {
 		if (dev->bdev)
-			blkdev_put(dev->bdev, FMODE_READ);
+			blkdev_put(dev->bdev, FMODE_READ | FMODE_WRITE);
 	}
 }
 
···
 				return -EIO;
 			p = xdr_decode_hyper(p, &b->simple.sigs[i].offset);
 			b->simple.sigs[i].sig_len = be32_to_cpup(p++);
+			if (b->simple.sigs[i].sig_len > PNFS_BLOCK_UUID_LEN) {
+				pr_info("signature too long: %d\n",
+					b->simple.sigs[i].sig_len);
+				return -EIO;
+			}
 
 			p = xdr_inline_decode(xdr, b->simple.sigs[i].sig_len);
 			if (!p)
···
 	if (!dev)
 		return -EIO;
 
-	d->bdev = blkdev_get_by_dev(dev, FMODE_READ, NULL);
+	d->bdev = blkdev_get_by_dev(dev, FMODE_READ | FMODE_WRITE, NULL);
 	if (IS_ERR(d->bdev)) {
 		printk(KERN_WARNING "pNFS: failed to open device %d:%d (%ld)\n",
 			MAJOR(dev), MINOR(dev), PTR_ERR(d->bdev));

fs/nfs/blocklayout/extent_tree.c

···
 	return err;
 }
 
+static size_t ext_tree_layoutupdate_size(size_t count)
+{
+	return sizeof(__be32) /* number of entries */ +
+		PNFS_BLOCK_EXTENT_SIZE * count;
+}
+
 static void ext_tree_free_commitdata(struct nfs4_layoutcommit_args *arg,
 		size_t buffer_size)
 {
···
 			continue;
 
 		(*count)++;
-		if (*count * BL_EXTENT_SIZE > buffer_size) {
+		if (ext_tree_layoutupdate_size(*count) > buffer_size) {
 			/* keep counting.. */
 			ret = -ENOSPC;
 			continue;
···
 	if (unlikely(ret)) {
 		ext_tree_free_commitdata(arg, buffer_size);
 
-		buffer_size = sizeof(__be32) + BL_EXTENT_SIZE * count;
+		buffer_size = ext_tree_layoutupdate_size(count);
 		count = 0;
 
 		arg->layoutupdate_pages =
···
 	}
 
 	*start_p = cpu_to_be32(count);
-	arg->layoutupdate_len = sizeof(__be32) + BL_EXTENT_SIZE * count;
+	arg->layoutupdate_len = ext_tree_layoutupdate_size(count);
 
 	if (unlikely(arg->layoutupdate_pages != &arg->layoutupdate_page)) {
-		__be32 *p = start_p;
+		void *p = start_p, *end = p + arg->layoutupdate_len;
 		int i = 0;
 
-		for (p = start_p;
-		     p < start_p + arg->layoutupdate_len;
-		     p += PAGE_SIZE) {
+		for ( ; p < end; p += PAGE_SIZE)
 			arg->layoutupdate_pages[i++] = vmalloc_to_page(p);
-		}
 	}
 
 	dprintk("%s found %zu ranges\n", __func__, count);

fs/nfs/callback.c

···
 	spin_lock_init(&serv->sv_cb_lock);
 	init_waitqueue_head(&serv->sv_cb_waitq);
 	rqstp = svc_prepare_thread(serv, &serv->sv_pools[0], NUMA_NO_NODE);
-	if (IS_ERR(rqstp)) {
-		svc_xprt_put(serv->sv_bc_xprt);
-		serv->sv_bc_xprt = NULL;
-	}
 	dprintk("--> %s return %d\n", __func__, PTR_ERR_OR_ZERO(rqstp));
 	return rqstp;
 }

fs/nfs/callback_proc.c

···
 		rpc_peeraddr2str(cps->clp->cl_rpcclient, RPC_DISPLAY_ADDR));
 
 	inode = nfs_delegation_find_inode(cps->clp, &args->fh);
-	if (inode == NULL)
+	if (inode == NULL) {
+		trace_nfs4_cb_getattr(cps->clp, &args->fh, NULL,
+				-ntohl(res->status));
 		goto out;
+	}
 	nfsi = NFS_I(inode);
 	rcu_read_lock();
 	delegation = rcu_dereference(nfsi->delegation);
···
 	res->status = 0;
 out_iput:
 	rcu_read_unlock();
+	trace_nfs4_cb_getattr(cps->clp, &args->fh, inode, -ntohl(res->status));
 	iput(inode);
 out:
 	dprintk("%s: exit with status = %d\n", __func__, ntohl(res->status));
···
 	spin_unlock(&ino->i_lock);
 	pnfs_free_lseg_list(&free_me_list);
 	pnfs_put_layout_hdr(lo);
+	trace_nfs4_cb_layoutrecall_inode(clp, &args->cbl_fh, ino, -rv);
 	iput(ino);
 out:
 	return rv;
···
 	status = htonl(NFS4_OK);
 
 	nfs41_set_target_slotid(fc_tbl, args->crsa_target_highest_slotid);
-	nfs41_server_notify_target_slotid_update(cps->clp);
+	nfs41_notify_server(cps->clp);
 out:
 	dprintk("%s: exit with status = %d\n", __func__, ntohl(status));
 	return status;

fs/nfs/client.c

···
 #include <linux/stat.h>
 #include <linux/errno.h>
 #include <linux/unistd.h>
+#include <linux/sunrpc/addr.h>
 #include <linux/sunrpc/clnt.h>
 #include <linux/sunrpc/stats.h>
 #include <linux/sunrpc/metrics.h>
···
 }
 EXPORT_SYMBOL_GPL(nfs_put_client);
 
-#if defined(CONFIG_IPV6) || defined(CONFIG_IPV6_MODULE)
-/*
- * Test if two ip6 socket addresses refer to the same socket by
- * comparing relevant fields. The padding bytes specifically, are not
- * compared. sin6_flowinfo is not compared because it only affects QoS
- * and sin6_scope_id is only compared if the address is "link local"
- * because "link local" addresses need only be unique to a specific
- * link. Conversely, ordinary unicast addresses might have different
- * sin6_scope_id.
- *
- * The caller should ensure both socket addresses are AF_INET6.
- */
-static int nfs_sockaddr_match_ipaddr6(const struct sockaddr *sa1,
-				      const struct sockaddr *sa2)
-{
-	const struct sockaddr_in6 *sin1 = (const struct sockaddr_in6 *)sa1;
-	const struct sockaddr_in6 *sin2 = (const struct sockaddr_in6 *)sa2;
-
-	if (!ipv6_addr_equal(&sin1->sin6_addr, &sin2->sin6_addr))
-		return 0;
-	else if (ipv6_addr_type(&sin1->sin6_addr) & IPV6_ADDR_LINKLOCAL)
-		return sin1->sin6_scope_id == sin2->sin6_scope_id;
-
-	return 1;
-}
-#else	/* !defined(CONFIG_IPV6) && !defined(CONFIG_IPV6_MODULE) */
-static int nfs_sockaddr_match_ipaddr6(const struct sockaddr *sa1,
-				      const struct sockaddr *sa2)
-{
-	return 0;
-}
-#endif
-
-/*
- * Test if two ip4 socket addresses refer to the same socket, by
- * comparing relevant fields. The padding bytes specifically, are
- * not compared.
- *
- * The caller should ensure both socket addresses are AF_INET.
- */
-static int nfs_sockaddr_match_ipaddr4(const struct sockaddr *sa1,
-				      const struct sockaddr *sa2)
-{
-	const struct sockaddr_in *sin1 = (const struct sockaddr_in *)sa1;
-	const struct sockaddr_in *sin2 = (const struct sockaddr_in *)sa2;
-
-	return sin1->sin_addr.s_addr == sin2->sin_addr.s_addr;
-}
-
-static int nfs_sockaddr_cmp_ip6(const struct sockaddr *sa1,
-				const struct sockaddr *sa2)
-{
-	const struct sockaddr_in6 *sin1 = (const struct sockaddr_in6 *)sa1;
-	const struct sockaddr_in6 *sin2 = (const struct sockaddr_in6 *)sa2;
-
-	return nfs_sockaddr_match_ipaddr6(sa1, sa2) &&
-		(sin1->sin6_port == sin2->sin6_port);
-}
-
-static int nfs_sockaddr_cmp_ip4(const struct sockaddr *sa1,
-				const struct sockaddr *sa2)
-{
-	const struct sockaddr_in *sin1 = (const struct sockaddr_in *)sa1;
-	const struct sockaddr_in *sin2 = (const struct sockaddr_in *)sa2;
-
-	return nfs_sockaddr_match_ipaddr4(sa1, sa2) &&
-		(sin1->sin_port == sin2->sin_port);
-}
-
-#if defined(CONFIG_NFS_V4_1)
-/*
- * Test if two socket addresses represent the same actual socket,
- * by comparing (only) relevant fields, excluding the port number.
- */
-int nfs_sockaddr_match_ipaddr(const struct sockaddr *sa1,
-			      const struct sockaddr *sa2)
-{
-	if (sa1->sa_family != sa2->sa_family)
-		return 0;
-
-	switch (sa1->sa_family) {
-	case AF_INET:
-		return nfs_sockaddr_match_ipaddr4(sa1, sa2);
-	case AF_INET6:
-		return nfs_sockaddr_match_ipaddr6(sa1, sa2);
-	}
-	return 0;
-}
-EXPORT_SYMBOL_GPL(nfs_sockaddr_match_ipaddr);
-#endif /* CONFIG_NFS_V4_1 */
-
-/*
- * Test if two socket addresses represent the same actual socket,
- * by comparing (only) relevant fields, including the port number.
- */
-static int nfs_sockaddr_cmp(const struct sockaddr *sa1,
-			    const struct sockaddr *sa2)
-{
-	if (sa1->sa_family != sa2->sa_family)
-		return 0;
-
-	switch (sa1->sa_family) {
-	case AF_INET:
-		return nfs_sockaddr_cmp_ip4(sa1, sa2);
-	case AF_INET6:
-		return nfs_sockaddr_cmp_ip6(sa1, sa2);
-	}
-	return 0;
-}
-
 /*
  * Find an nfs_client on the list that matches the initialisation data
  * that is supplied.
···
 		if (clp->cl_minorversion != data->minorversion)
 			continue;
 		/* Match the full socket address */
-		if (!nfs_sockaddr_cmp(sap, clap))
+		if (!rpc_cmp_addr_port(sap, clap))
 			continue;
 
 		atomic_inc(&clp->cl_count);

fs/nfs/delegation.c

···
 		if (delegation->inode != NULL) {
 			nfs4_stateid_copy(&delegation->stateid, &res->delegation);
 			delegation->type = res->delegation_type;
-			delegation->maxsize = res->maxsize;
+			delegation->pagemod_limit = res->pagemod_limit;
 			oldcred = delegation->cred;
 			delegation->cred = get_rpccred(cred);
 			clear_bit(NFS_DELEGATION_NEED_RECLAIM,
···
 		return -ENOMEM;
 	nfs4_stateid_copy(&delegation->stateid, &res->delegation);
 	delegation->type = res->delegation_type;
-	delegation->maxsize = res->maxsize;
+	delegation->pagemod_limit = res->pagemod_limit;
 	delegation->change_attr = inode->i_version;
 	delegation->cred = get_rpccred(cred);
 	delegation->inode = inode;
···
 		nfs4_stateid_copy(dst, &delegation->stateid);
 		nfs_mark_delegation_referenced(delegation);
 	}
+	rcu_read_unlock();
+	return ret;
+}
+
+/**
+ * nfs4_delegation_flush_on_close - Check if we must flush file on close
+ * @inode: inode to check
+ *
+ * This function checks the number of outstanding writes to the file
+ * against the delegation 'space_limit' field to see if
+ * the spec requires us to flush the file on close.
+ */
+bool nfs4_delegation_flush_on_close(const struct inode *inode)
+{
+	struct nfs_inode *nfsi = NFS_I(inode);
+	struct nfs_delegation *delegation;
+	bool ret = true;
+
+	rcu_read_lock();
+	delegation = rcu_dereference(nfsi->delegation);
+	if (delegation == NULL || !(delegation->type & FMODE_WRITE))
+		goto out;
+	if (nfsi->nrequests < delegation->pagemod_limit)
+		ret = false;
+out:
 	rcu_read_unlock();
 	return ret;
 }

fs/nfs/delegation.h

···
 	struct inode *inode;
 	nfs4_stateid stateid;
 	fmode_t type;
-	loff_t maxsize;
+	unsigned long pagemod_limit;
 	__u64 change_attr;
 	unsigned long flags;
 	spinlock_t lock;
···
 void nfs_mark_delegation_referenced(struct nfs_delegation *delegation);
 int nfs4_have_delegation(struct inode *inode, fmode_t flags);
 int nfs4_check_delegation(struct inode *inode, fmode_t flags);
+bool nfs4_delegation_flush_on_close(const struct inode *inode);
 
 #endif
 

fs/nfs/dir.c

···
 }
 
 static
-void nfs_readdir_free_pagearray(struct page **pages, unsigned int npages)
+void nfs_readdir_free_pages(struct page **pages, unsigned int npages)
 {
 	unsigned int i;
 	for (i = 0; i < npages; i++)
 		put_page(pages[i]);
 }
 
-static
-void nfs_readdir_free_large_page(void *ptr, struct page **pages,
-		unsigned int npages)
-{
-	nfs_readdir_free_pagearray(pages, npages);
-}
-
 /*
  * nfs_readdir_large_page will allocate pages that must be freed with a call
- * to nfs_readdir_free_large_page
+ * to nfs_readdir_free_pagearray
  */
 static
-int nfs_readdir_large_page(struct page **pages, unsigned int npages)
+int nfs_readdir_alloc_pages(struct page **pages, unsigned int npages)
 {
 	unsigned int i;
 
···
 	return 0;
 
 out_freepages:
-	nfs_readdir_free_pagearray(pages, i);
+	nfs_readdir_free_pages(pages, i);
 	return -ENOMEM;
 }
 
···
 int nfs_readdir_xdr_to_array(nfs_readdir_descriptor_t *desc, struct page *page, struct inode *inode)
 {
 	struct page *pages[NFS_MAX_READDIR_PAGES];
-	void *pages_ptr = NULL;
 	struct nfs_entry entry;
 	struct file	*file = desc->file;
 	struct nfs_cache_array *array;
···
 	memset(array, 0, sizeof(struct nfs_cache_array));
 	array->eof_index = -1;
 
-	status = nfs_readdir_large_page(pages, array_size);
+	status = nfs_readdir_alloc_pages(pages, array_size);
 	if (status < 0)
 		goto out_release_array;
 	do {
···
 		}
 	} while (array->eof_index < 0);
 
-	nfs_readdir_free_large_page(pages_ptr, pages, array_size);
+	nfs_readdir_free_pages(pages, array_size);
 out_release_array:
 	nfs_readdir_release_array(page);
 out_label_free:

fs/nfs/file.c

···
 	dprintk("NFS: release(%pD2)\n", filp);
 
 	nfs_inc_stats(inode, NFSIOS_VFSRELEASE);
-	return nfs_release(inode, filp);
+	nfs_file_clear_open_context(filp);
+	return 0;
 }
 EXPORT_SYMBOL_GPL(nfs_file_release);
 
···
 /*
  * Flush all dirty pages, and check for write errors.
  */
-int
+static int
 nfs_file_flush(struct file *file, fl_owner_t id)
 {
 	struct inode	*inode = file_inode(file);
···
 	if ((file->f_mode & FMODE_WRITE) == 0)
 		return 0;
 
-	/*
-	 * If we're holding a write delegation, then just start the i/o
-	 * but don't wait for completion (or send a commit).
-	 */
-	if (NFS_PROTO(inode)->have_delegation(inode, FMODE_WRITE))
-		return filemap_fdatawrite(file->f_mapping);
-
 	/* Flush writes to the server and return any errors */
 	return vfs_fsync(file, 0);
 }
-EXPORT_SYMBOL_GPL(nfs_file_flush);
 
 ssize_t
 nfs_file_read(struct kiocb *iocb, struct iov_iter *to)
···
 	.page_mkwrite = nfs_vm_page_mkwrite,
 };
 
-static int nfs_need_sync_write(struct file *filp, struct inode *inode)
+static int nfs_need_check_write(struct file *filp, struct inode *inode)
 {
 	struct nfs_open_context *ctx;
 
-	if (IS_SYNC(inode) || (filp->f_flags & O_DSYNC))
-		return 1;
 	ctx = nfs_file_open_context(filp);
 	if (test_bit(NFS_CONTEXT_ERROR_WRITE, &ctx->flags) ||
 	    nfs_ctx_key_to_expire(ctx))
···
 	if (result > 0)
 		written = result;
 
-	/* Return error values for O_DSYNC and IS_SYNC() */
-	if (result >= 0 && nfs_need_sync_write(file, inode)) {
+	/* Return error values */
+	if (result >= 0 && nfs_need_check_write(file, inode)) {
 		int err = vfs_fsync(file, 0);
 		if (err < 0)
 			result = err;

fs/nfs/flexfilelayout/flexfilelayout.c

···
 	ffl = kzalloc(sizeof(*ffl), gfp_flags);
 	if (ffl) {
 		INIT_LIST_HEAD(&ffl->error_list);
+		INIT_LIST_HEAD(&ffl->mirrors);
 		return &ffl->generic_hdr;
 	} else
 		return NULL;
···
 	return 0;
 }
 
+static bool ff_mirror_match_fh(const struct nfs4_ff_layout_mirror *m1,
+		const struct nfs4_ff_layout_mirror *m2)
+{
+	int i, j;
+
+	if (m1->fh_versions_cnt != m2->fh_versions_cnt)
+		return false;
+	for (i = 0; i < m1->fh_versions_cnt; i++) {
+		bool found_fh = false;
+		for (j = 0; j < m2->fh_versions_cnt; i++) {
+			if (nfs_compare_fh(&m1->fh_versions[i],
+					&m2->fh_versions[j]) == 0) {
+				found_fh = true;
+				break;
+			}
+		}
+		if (!found_fh)
+			return false;
+	}
+	return true;
+}
+
+static struct nfs4_ff_layout_mirror *
+ff_layout_add_mirror(struct pnfs_layout_hdr *lo,
+		struct nfs4_ff_layout_mirror *mirror)
+{
+	struct nfs4_flexfile_layout *ff_layout = FF_LAYOUT_FROM_HDR(lo);
+	struct nfs4_ff_layout_mirror *pos;
+	struct inode *inode = lo->plh_inode;
+
+	spin_lock(&inode->i_lock);
+	list_for_each_entry(pos, &ff_layout->mirrors, mirrors) {
+		if (mirror->mirror_ds != pos->mirror_ds)
+			continue;
+		if (!ff_mirror_match_fh(mirror, pos))
+			continue;
+		if (atomic_inc_not_zero(&pos->ref)) {
+			spin_unlock(&inode->i_lock);
+			return pos;
+		}
+	}
+	list_add(&mirror->mirrors, &ff_layout->mirrors);
+	mirror->layout = lo;
+	spin_unlock(&inode->i_lock);
+	return mirror;
+}
+
+static void
+ff_layout_remove_mirror(struct nfs4_ff_layout_mirror *mirror)
+{
+	struct inode *inode;
+	if (mirror->layout == NULL)
+		return;
+	inode = mirror->layout->plh_inode;
+	spin_lock(&inode->i_lock);
+	list_del(&mirror->mirrors);
+	spin_unlock(&inode->i_lock);
+	mirror->layout = NULL;
+}
+
+static struct nfs4_ff_layout_mirror *ff_layout_alloc_mirror(gfp_t gfp_flags)
+{
+	struct nfs4_ff_layout_mirror *mirror;
+
+	mirror = kzalloc(sizeof(*mirror), gfp_flags);
+	if (mirror != NULL) {
+		spin_lock_init(&mirror->lock);
+		atomic_set(&mirror->ref, 1);
+		INIT_LIST_HEAD(&mirror->mirrors);
+	}
+	return mirror;
+}
+
+static void ff_layout_free_mirror(struct nfs4_ff_layout_mirror *mirror)
+{
+	ff_layout_remove_mirror(mirror);
+	kfree(mirror->fh_versions);
+	if (mirror->cred)
+		put_rpccred(mirror->cred);
+	nfs4_ff_layout_put_deviceid(mirror->mirror_ds);
+	kfree(mirror);
+}
+
+static void ff_layout_put_mirror(struct nfs4_ff_layout_mirror *mirror)
+{
+	if (mirror != NULL && atomic_dec_and_test(&mirror->ref))
+		ff_layout_free_mirror(mirror);
+}
+
 static void ff_layout_free_mirror_array(struct nfs4_ff_layout_segment *fls)
 {
 	int i;
···
 			/* normally mirror_ds is freed in
 			 * .free_deviceid_node but we still do it here
 			 * for .alloc_lseg error path */
-			if (fls->mirror_array[i]) {
-				kfree(fls->mirror_array[i]->fh_versions);
-				nfs4_ff_layout_put_deviceid(fls->mirror_array[i]->mirror_ds);
-				kfree(fls->mirror_array[i]);
-			}
+			ff_layout_put_mirror(fls->mirror_array[i]);
 		}
 		kfree(fls->mirror_array);
 		fls->mirror_array = NULL;
···
 		ff_layout_free_mirror_array(fls);
 		kfree(fls);
 	}
+}
+
+static bool
+ff_lseg_range_is_after(const struct pnfs_layout_range *l1,
+		const struct pnfs_layout_range *l2)
+{
+	u64 end1, end2;
+
+	if (l1->iomode != l2->iomode)
+		return l1->iomode != IOMODE_READ;
+	end1 = pnfs_calc_offset_end(l1->offset, l1->length);
+	end2 = pnfs_calc_offset_end(l2->offset, l2->length);
+	if (end1 < l2->offset)
+		return false;
+	if (end2 < l1->offset)
+		return true;
+	return l2->offset <= l1->offset;
+}
+
+static bool
+ff_lseg_merge(struct pnfs_layout_segment *new,
+		struct pnfs_layout_segment *old)
+{
+	u64 new_end, old_end;
+
+	if (new->pls_range.iomode != old->pls_range.iomode)
+		return false;
+	old_end = pnfs_calc_offset_end(old->pls_range.offset,
+			old->pls_range.length);
+	if (old_end < new->pls_range.offset)
+		return false;
+	new_end = pnfs_calc_offset_end(new->pls_range.offset,
+			new->pls_range.length);
+	if (new_end < old->pls_range.offset)
+		return false;
+
+	/* Mergeable: copy info from 'old' to 'new' */
+	if (new_end < old_end)
+		new_end = old_end;
+	if (new->pls_range.offset < old->pls_range.offset)
+		new->pls_range.offset = old->pls_range.offset;
+	new->pls_range.length = pnfs_calc_offset_length(new->pls_range.offset,
+			new_end);
+	if (test_bit(NFS_LSEG_ROC, &old->pls_flags))
+		set_bit(NFS_LSEG_ROC, &new->pls_flags);
+	if (test_bit(NFS_LSEG_LAYOUTRETURN, &old->pls_flags))
+		set_bit(NFS_LSEG_LAYOUTRETURN, &new->pls_flags);
+	return true;
+}
+
+static void
+ff_layout_add_lseg(struct pnfs_layout_hdr *lo,
+		struct pnfs_layout_segment *lseg,
+		struct list_head *free_me)
+{
+	pnfs_generic_layout_insert_lseg(lo, lseg,
+			ff_lseg_range_is_after,
+			ff_lseg_merge,
+			free_me);
 }
 
 static void ff_layout_sort_mirrors(struct nfs4_ff_layout_segment *fls)
···
 		goto out_err_free;
 
 	for (i = 0; i < fls->mirror_array_cnt; i++) {
+		struct nfs4_ff_layout_mirror *mirror;
 		struct nfs4_deviceid devid;
 		struct nfs4_deviceid_node *idnode;
 		u32 ds_count;
···
 		if (ds_count != 1)
 			goto out_err_free;
 
-		fls->mirror_array[i] =
-			kzalloc(sizeof(struct nfs4_ff_layout_mirror),
-				gfp_flags);
+		fls->mirror_array[i] = ff_layout_alloc_mirror(gfp_flags);
 		if (fls->mirror_array[i] == NULL) {
 			rc = -ENOMEM;
 			goto out_err_free;
 		}
 
-		spin_lock_init(&fls->mirror_array[i]->lock);
 		fls->mirror_array[i]->ds_count = ds_count;
-		fls->mirror_array[i]->lseg = &fls->generic_hdr;
 
 		/* deviceid */
 		rc = decode_deviceid(&stream, &devid);
···
 		if (rc)
 			goto out_err_free;
 
+		mirror = ff_layout_add_mirror(lh, fls->mirror_array[i]);
+		if (mirror != fls->mirror_array[i]) {
+			ff_layout_free_mirror(fls->mirror_array[i]);
+			fls->mirror_array[i] = mirror;
+		}
+
 		dprintk("%s: uid %d gid %d\n", __func__,
 			fls->mirror_array[i]->uid,
 			fls->mirror_array[i]->gid);
···
 ff_layout_free_lseg(struct pnfs_layout_segment *lseg)
 {
 	struct nfs4_ff_layout_segment *fls = FF_LAYOUT_LSEG(lseg);
-	int i;
 
 	dprintk("--> %s\n", __func__);
-
-	for (i = 0; i < fls->mirror_array_cnt; i++) {
-		if (fls->mirror_array[i]) {
-			nfs4_ff_layout_put_deviceid(fls->mirror_array[i]->mirror_ds);
-			fls->mirror_array[i]->mirror_ds = NULL;
-			if (fls->mirror_array[i]->cred) {
-				put_rpccred(fls->mirror_array[i]->cred);
-				fls->mirror_array[i]->cred = NULL;
-			}
-		}
-	}
 
 	if (lseg->pls_range.iomode == IOMODE_RW) {
 		struct nfs4_flexfile_layout *ffl;
···
 }
 
 static void
-nfs4_ff_start_busy_timer(struct nfs4_ff_busy_timer *timer)
+nfs4_ff_start_busy_timer(struct nfs4_ff_busy_timer *timer, ktime_t now)
 {
 	/* first IO request? */
 	if (atomic_inc_return(&timer->n_ops) == 1) {
-		timer->start_time = ktime_get();
+		timer->start_time = now;
 	}
 }
 
 static ktime_t
-nfs4_ff_end_busy_timer(struct nfs4_ff_busy_timer *timer)
+nfs4_ff_end_busy_timer(struct nfs4_ff_busy_timer *timer, ktime_t now)
 {
-	ktime_t start, now;
+	ktime_t start;
 
 	if (atomic_dec_return(&timer->n_ops) < 0)
 		WARN_ON_ONCE(1);
 
-	now = ktime_get();
 	start = timer->start_time;
 	timer->start_time = now;
 	return ktime_sub(now, start);
 }
 
-static ktime_t
-nfs4_ff_layout_calc_completion_time(struct rpc_task *task)
-{
-	return ktime_sub(ktime_get(), task->tk_start);
-}
-
 static bool
 nfs4_ff_layoutstat_start_io(struct nfs4_ff_layout_mirror *mirror,
-			    struct nfs4_ff_layoutstat *layoutstat)
+			    struct nfs4_ff_layoutstat *layoutstat,
+			    ktime_t now)
 {
 	static const ktime_t notime = {0};
-	ktime_t now = ktime_get();
+	s64 report_interval = FF_LAYOUTSTATS_REPORT_INTERVAL;
 
-	nfs4_ff_start_busy_timer(&layoutstat->busy_timer);
+	nfs4_ff_start_busy_timer(&layoutstat->busy_timer, now);
 	if (ktime_equal(mirror->start_time, notime))
 		mirror->start_time = now;
 	if (ktime_equal(mirror->last_report_time, notime))
 		mirror->last_report_time = now;
+	if (layoutstats_timer != 0)
+		report_interval = (s64)layoutstats_timer * 1000LL;
 	if (ktime_to_ms(ktime_sub(now, mirror->last_report_time)) >=
-			FF_LAYOUTSTATS_REPORT_INTERVAL) {
+			report_interval) {
 		mirror->last_report_time = now;
 		return true;
 	}
···
 nfs4_ff_layout_stat_io_update_completed(struct nfs4_ff_layoutstat *layoutstat,
 		__u64 requested,
 		__u64 completed,
-		ktime_t time_completed)
+		ktime_t time_completed,
+		ktime_t time_started)
 {
 	struct nfs4_ff_io_stat *iostat = &layoutstat->io_stat;
+	ktime_t completion_time = ktime_sub(time_completed, time_started);
 	ktime_t timer;
 
 	iostat->ops_completed++;
 	iostat->bytes_completed += completed;
 	iostat->bytes_not_delivered += requested - completed;
 
-	timer = nfs4_ff_end_busy_timer(&layoutstat->busy_timer);
+	timer = nfs4_ff_end_busy_timer(&layoutstat->busy_timer, time_completed);
 	iostat->total_busy_time =
 			ktime_add(iostat->total_busy_time, timer);
 	iostat->aggregate_completion_time =
-			ktime_add(iostat->aggregate_completion_time, time_completed);
+			ktime_add(iostat->aggregate_completion_time,
+					completion_time);
 }
 
 static void
-nfs4_ff_layout_stat_io_start_read(struct nfs4_ff_layout_mirror *mirror,
-		__u64 requested)
+nfs4_ff_layout_stat_io_start_read(struct inode *inode,
+		struct nfs4_ff_layout_mirror *mirror,
+		__u64 requested, ktime_t now)
 {
 	bool report;
 
 	spin_lock(&mirror->lock);
-	report = nfs4_ff_layoutstat_start_io(mirror, &mirror->read_stat);
+	report = nfs4_ff_layoutstat_start_io(mirror, &mirror->read_stat, now);
 	nfs4_ff_layout_stat_io_update_requested(&mirror->read_stat, requested);
 	spin_unlock(&mirror->lock);
 
 	if (report)
-		pnfs_report_layoutstat(mirror->lseg->pls_layout->plh_inode);
+		pnfs_report_layoutstat(inode, GFP_KERNEL);
 }
 
 static void
···
 	spin_lock(&mirror->lock);
 	nfs4_ff_layout_stat_io_update_completed(&mirror->read_stat,
 			requested, completed,
-			nfs4_ff_layout_calc_completion_time(task));
+			ktime_get(), task->tk_start);
 	spin_unlock(&mirror->lock);
 }
 
 static void
-nfs4_ff_layout_stat_io_start_write(struct nfs4_ff_layout_mirror *mirror,
-		__u64 requested)
+nfs4_ff_layout_stat_io_start_write(struct inode *inode,
+		struct nfs4_ff_layout_mirror *mirror,
+		__u64 requested, ktime_t now)
 {
 	bool report;
 
 	spin_lock(&mirror->lock);
-	report = nfs4_ff_layoutstat_start_io(mirror , &mirror->write_stat);
+	report = nfs4_ff_layoutstat_start_io(mirror , &mirror->write_stat, now);
 	nfs4_ff_layout_stat_io_update_requested(&mirror->write_stat, requested);
 	spin_unlock(&mirror->lock);
 
 	if (report)
-		pnfs_report_layoutstat(mirror->lseg->pls_layout->plh_inode);
+		pnfs_report_layoutstat(inode, GFP_NOIO);
 }
 
 static void
···
 
 	spin_lock(&mirror->lock);
 	nfs4_ff_layout_stat_io_update_completed(&mirror->write_stat,
-			requested, completed,
-			nfs4_ff_layout_calc_completion_time(task));
+			requested, completed, ktime_get(), task->tk_start);
 	spin_unlock(&mirror->lock);
 }
 
···
 		return FF_LAYOUT_MIRROR_COUNT(pgio->pg_lseg);
 
 	/* no lseg means that pnfs is not in use, so no mirroring here */
-	pnfs_put_lseg(pgio->pg_lseg);
-	pgio->pg_lseg = NULL;
 	nfs_pageio_reset_write_mds(pgio);
 	return 1;
 }
···
 	if (task->tk_status >= 0)
 		return 0;
 
-	if (task->tk_status != -EJUKEBOX) {
+	switch (task->tk_status) {
+	/* File access problems. Don't mark the device as unavailable */
+	case -EACCES:
+	case -ESTALE:
+	case -EISDIR:
+	case -EBADHANDLE:
+	case -ELOOP:
+	case -ENOSPC:
+		break;
+	case -EJUKEBOX:
+		nfs_inc_stats(lseg->pls_layout->plh_inode, NFSIOS_DELAY);
+		goto out_retry;
+	default:
 		dprintk("%s DS connection error %d\n", __func__,
 			task->tk_status);
 		nfs4_mark_deviceid_unavailable(devid);
-		if (ff_layout_has_available_ds(lseg))
-			return -NFS4ERR_RESET_TO_PNFS;
-		else
-			return -NFS4ERR_RESET_TO_MDS;
 	}
-
-	if (task->tk_status == -EJUKEBOX)
-		nfs_inc_stats(lseg->pls_layout->plh_inode, NFSIOS_DELAY);
+	/* FIXME: Need to prevent infinite looping here. */
+	return -NFS4ERR_RESET_TO_PNFS;
+out_retry:
 	task->tk_status = 0;
 	rpc_restart_call(task);
 	rpc_delay(task, NFS_JUKEBOX_RETRY_TIME);
···
 
 static void ff_layout_io_track_ds_error(struct pnfs_layout_segment *lseg,
 					int idx, u64 offset, u64 length,
-					u32 status, int opnum)
+					u32 status, int opnum, int error)
 {
 	struct nfs4_ff_layout_mirror *mirror;
 	int err;
+
+	if (status == 0) {
+		switch (error) {
+		case -ETIMEDOUT:
+		case -EPFNOSUPPORT:
+		case -EPROTONOSUPPORT:
+		case -EOPNOTSUPP:
+		case -ECONNREFUSED:
+		case -ECONNRESET:
+		case -EHOSTDOWN:
+		case -EHOSTUNREACH:
+		case -ENETUNREACH:
+		case -EADDRINUSE:
+		case -ENOBUFS:
+		case -EPIPE:
+		case -EPERM:
+			status = NFS4ERR_NXIO;
+			break;
+		case -EACCES:
+			status = NFS4ERR_ACCESS;
+			break;
+		default:
+			return;
+		}
+	}
 
 	mirror = FF_LAYOUT_COMP(lseg, idx);
 	err = ff_layout_track_ds_error(FF_LAYOUT_FROM_HDR(lseg->pls_layout),
 				       mirror, offset, length, status, opnum,
 				       GFP_NOIO);
+	pnfs_error_mark_layout_for_return(lseg->pls_layout->plh_inode, lseg);
 	dprintk("%s: err %d op %d status %u\n", __func__, err, opnum, status);
 }
 
···
 static int ff_layout_read_done_cb(struct rpc_task *task,
 				struct nfs_pgio_header *hdr)
 {
-	struct inode *inode;
 	int err;
 
 	trace_nfs4_pnfs_read(hdr, task->tk_status);
-	if (task->tk_status == -ETIMEDOUT && !hdr->res.op_status)
-		hdr->res.op_status = NFS4ERR_NXIO;
-	if (task->tk_status < 0 && hdr->res.op_status)
+	if (task->tk_status < 0)
 		ff_layout_io_track_ds_error(hdr->lseg, hdr->pgio_mirror_idx,
 					    hdr->args.offset, hdr->args.count,
-					    hdr->res.op_status, OP_READ);
+					    hdr->res.op_status, OP_READ,
+					    task->tk_status);
 	err = ff_layout_async_handle_error(task, hdr->args.context->state,
 					   hdr->ds_clp, hdr->lseg,
 					   hdr->pgio_mirror_idx);
···
 		pnfs_read_resend_pnfs(hdr);
 		return task->tk_status;
 	case -NFS4ERR_RESET_TO_MDS:
-		inode = hdr->lseg->pls_layout->plh_inode;
-		pnfs_error_mark_layout_for_return(inode, hdr->lseg);
 		ff_layout_reset_read(hdr);
 		return task->tk_status;
 	case -EAGAIN:
···
 static int ff_layout_read_prepare_common(struct rpc_task *task,
 					 struct nfs_pgio_header *hdr)
 {
-	nfs4_ff_layout_stat_io_start_read(
+	nfs4_ff_layout_stat_io_start_read(hdr->inode,
 			FF_LAYOUT_COMP(hdr->lseg, hdr->pgio_mirror_idx),
-			hdr->args.count);
+			hdr->args.count,
+			task->tk_start);
 
 	if (unlikely(test_bit(NFS_CONTEXT_BAD, &hdr->args.context->flags))) {
 		rpc_exit(task, -EIO);
···
 static int ff_layout_write_done_cb(struct rpc_task *task,
 				struct nfs_pgio_header *hdr)
 {
-	struct inode *inode;
 	int err;
 
 	trace_nfs4_pnfs_write(hdr, task->tk_status);
-	if (task->tk_status == -ETIMEDOUT && !hdr->res.op_status)
-		hdr->res.op_status = NFS4ERR_NXIO;
-	if (task->tk_status < 0 && hdr->res.op_status)
+	if (task->tk_status < 0)
 		ff_layout_io_track_ds_error(hdr->lseg, hdr->pgio_mirror_idx,
 					    hdr->args.offset, hdr->args.count,
-					    hdr->res.op_status, OP_WRITE);
+					    hdr->res.op_status, OP_WRITE,
+					    task->tk_status);
 	err = ff_layout_async_handle_error(task, hdr->args.context->state,
 					   hdr->ds_clp, hdr->lseg,
 					   hdr->pgio_mirror_idx);
 
 	switch (err) {
 	case -NFS4ERR_RESET_TO_PNFS:
+		pnfs_set_retry_layoutget(hdr->lseg->pls_layout);
+		ff_layout_reset_write(hdr, true);
+		return task->tk_status;
 	case -NFS4ERR_RESET_TO_MDS:
-		inode = hdr->lseg->pls_layout->plh_inode;
-		pnfs_error_mark_layout_for_return(inode, hdr->lseg);
-		if (err == -NFS4ERR_RESET_TO_PNFS) {
-			pnfs_set_retry_layoutget(hdr->lseg->pls_layout);
-			ff_layout_reset_write(hdr, true);
-		} else {
-			pnfs_clear_retry_layoutget(hdr->lseg->pls_layout);
-			ff_layout_reset_write(hdr, false);
-		}
+		pnfs_clear_retry_layoutget(hdr->lseg->pls_layout);
+		ff_layout_reset_write(hdr, false);
 		return task->tk_status;
 	case -EAGAIN:
 		rpc_restart_call_prepare(task);
···
 	    hdr->res.verf->committed == NFS_DATA_SYNC)
 		ff_layout_set_layoutcommit(hdr);
 
+	/* zero out fattr since we don't care DS attr at all */
+	hdr->fattr.valid = 0;
+	if (task->tk_status >= 0)
+		nfs_writeback_update_inode(hdr);
+
 	return 0;
 }
 
 static int ff_layout_commit_done_cb(struct rpc_task *task,
 				     struct nfs_commit_data *data)
 {
-	struct inode *inode;
 	int err;
 
 	trace_nfs4_pnfs_commit_ds(data, task->tk_status);
-	if (task->tk_status == -ETIMEDOUT && !data->res.op_status)
-		data->res.op_status = NFS4ERR_NXIO;
-	if (task->tk_status < 0 && data->res.op_status)
+	if (task->tk_status < 0)
 		ff_layout_io_track_ds_error(data->lseg, data->ds_commit_index,
 					    data->args.offset, data->args.count,
-					    data->res.op_status, OP_COMMIT);
+					    data->res.op_status, OP_COMMIT,
+					    task->tk_status);
 	err = ff_layout_async_handle_error(task, NULL, data->ds_clp,
 					   data->lseg, data->ds_commit_index);
 
 	switch (err) {
 	case -NFS4ERR_RESET_TO_PNFS:
+		pnfs_set_retry_layoutget(data->lseg->pls_layout);
+		pnfs_generic_prepare_to_resend_writes(data);
+		return -EAGAIN;
 	case -NFS4ERR_RESET_TO_MDS:
-		inode = data->lseg->pls_layout->plh_inode;
-		pnfs_error_mark_layout_for_return(inode, data->lseg);
-		if (err == -NFS4ERR_RESET_TO_PNFS)
-			pnfs_set_retry_layoutget(data->lseg->pls_layout);
-		else
-			pnfs_clear_retry_layoutget(data->lseg->pls_layout);
+		pnfs_clear_retry_layoutget(data->lseg->pls_layout);
 		pnfs_generic_prepare_to_resend_writes(data);
 		return -EAGAIN;
 	case -EAGAIN:
···
 static int ff_layout_write_prepare_common(struct rpc_task *task,
 					  struct nfs_pgio_header *hdr)
 {
-	nfs4_ff_layout_stat_io_start_write(
+	nfs4_ff_layout_stat_io_start_write(hdr->inode,
 			FF_LAYOUT_COMP(hdr->lseg, hdr->pgio_mirror_idx),
-			hdr->args.count);
+			hdr->args.count,
+			task->tk_start);
 
 	if (unlikely(test_bit(NFS_CONTEXT_BAD, &hdr->args.context->flags))) {
 		rpc_exit(task, -EIO);
···
 static void ff_layout_commit_prepare_common(struct rpc_task *task,
 		struct nfs_commit_data *cdata)
 {
-	nfs4_ff_layout_stat_io_start_write(
+	nfs4_ff_layout_stat_io_start_write(cdata->inode,
 			FF_LAYOUT_COMP(cdata->lseg, cdata->ds_commit_index),
-			0);
+			0, task->tk_start);
 }
 
 static void ff_layout_commit_prepare_v3(struct rpc_task *task, void *data)
···
 	*start = cpu_to_be32((xdr->p - start - 1) * 4);
 }
 
-static bool
+static int
 ff_layout_mirror_prepare_stats(struct nfs42_layoutstat_args *args,
-			       struct pnfs_layout_segment *pls,
-			       int *dev_count, int dev_limit)
+			       struct pnfs_layout_hdr *lo,
+			       int dev_limit)
 {
+	struct nfs4_flexfile_layout *ff_layout = FF_LAYOUT_FROM_HDR(lo);
 	struct nfs4_ff_layout_mirror *mirror;
 	struct nfs4_deviceid_node *dev;
 	struct nfs42_layoutstat_devinfo *devinfo;
-	int i;
+	int i = 0;
 
-	for (i = 0; i < FF_LAYOUT_MIRROR_COUNT(pls); i++) {
-		if (*dev_count >= dev_limit)
+	list_for_each_entry(mirror, &ff_layout->mirrors, mirrors) {
+		if (i >= dev_limit)
 			break;
-		mirror = FF_LAYOUT_COMP(pls, i);
-		if (!mirror || !mirror->mirror_ds)
+		if (!mirror->mirror_ds)
 			continue;
-		dev = FF_LAYOUT_DEVID_NODE(pls, i);
-		devinfo = &args->devinfo[*dev_count];
+		/* mirror refcount put in cleanup_layoutstats */
+		if (!atomic_inc_not_zero(&mirror->ref))
+			continue;
+		dev = &mirror->mirror_ds->id_node; 
+		devinfo = &args->devinfo[i];
 		memcpy(&devinfo->dev_id, &dev->deviceid, NFS4_DEVICEID4_SIZE);
-		devinfo->offset = pls->pls_range.offset;
-		devinfo->length = pls->pls_range.length;
-		/* well, we don't really know if IO is continuous or not! */
-		devinfo->read_count = mirror->read_stat.io_stat.bytes_completed;
+		devinfo->offset = 0;
+		devinfo->length = NFS4_MAX_UINT64;
+		devinfo->read_count = mirror->read_stat.io_stat.ops_completed;
 		devinfo->read_bytes = mirror->read_stat.io_stat.bytes_completed;
-		devinfo->write_count = mirror->write_stat.io_stat.bytes_completed;
+		devinfo->write_count = mirror->write_stat.io_stat.ops_completed;
 		devinfo->write_bytes = mirror->write_stat.io_stat.bytes_completed;
 		devinfo->layout_type = LAYOUT_FLEX_FILES;
 		devinfo->layoutstats_encode = ff_layout_encode_layoutstats;
 		devinfo->layout_private = mirror;
-		/* lseg refcount put in cleanup_layoutstats */
-		pnfs_get_lseg(pls);
 
-		++(*dev_count);
+		i++;
 	}
-
-	return *dev_count < dev_limit;
+	return i;
 }
 
 static int
 ff_layout_prepare_layoutstats(struct nfs42_layoutstat_args *args)
 {
-	struct pnfs_layout_segment *pls;
+	struct nfs4_flexfile_layout *ff_layout;
+	struct nfs4_ff_layout_mirror *mirror;
 	int dev_count = 0;
 
 	spin_lock(&args->inode->i_lock);
-	list_for_each_entry(pls, &NFS_I(args->inode)->layout->plh_segs, pls_list) {
-		dev_count += FF_LAYOUT_MIRROR_COUNT(pls);
+	ff_layout = FF_LAYOUT_FROM_HDR(NFS_I(args->inode)->layout);
+	list_for_each_entry(mirror, &ff_layout->mirrors, mirrors) {
+		if (atomic_read(&mirror->ref) != 0)
+			dev_count ++;
 	}
 	spin_unlock(&args->inode->i_lock);
 	/* For now, send at most PNFS_LAYOUTSTATS_MAXDEV statistics */
···
 			__func__, dev_count, PNFS_LAYOUTSTATS_MAXDEV);
 		dev_count = PNFS_LAYOUTSTATS_MAXDEV;
 	}
-	args->devinfo = kmalloc(dev_count * sizeof(*args->devinfo), GFP_KERNEL);
+	args->devinfo = kmalloc_array(dev_count, sizeof(*args->devinfo), GFP_NOIO);
 	if (!args->devinfo)
 		return -ENOMEM;
 
-	dev_count = 0;
 	spin_lock(&args->inode->i_lock);
-	list_for_each_entry(pls, &NFS_I(args->inode)->layout->plh_segs, pls_list) {
-		if (!ff_layout_mirror_prepare_stats(args, pls, &dev_count,
-						    PNFS_LAYOUTSTATS_MAXDEV)) {
-			break;
-		}
-	}
+	args->num_dev = ff_layout_mirror_prepare_stats(args,
+			&ff_layout->generic_hdr, dev_count);
 	spin_unlock(&args->inode->i_lock);
-	args->num_dev = dev_count;
 
 	return 0;
 }
···
 	for (i = 0; i < data->args.num_dev; i++) {
 		mirror = data->args.devinfo[i].layout_private;
 		data->args.devinfo[i].layout_private = NULL;
-		pnfs_put_lseg(mirror->lseg);
+		ff_layout_put_mirror(mirror);
 	}
 }
 
···
 	.free_layout_hdr	= ff_layout_free_layout_hdr,
 	.alloc_lseg		= ff_layout_alloc_lseg,
 	.free_lseg		= ff_layout_free_lseg,
+	.add_lseg		= ff_layout_add_lseg,
 	.pg_read_ops		= &ff_layout_pg_read_ops,
 	.pg_write_ops		= &ff_layout_pg_write_ops,
 	.get_ds_info		= ff_layout_get_ds_info,

fs/nfs/flexfilelayout/flexfilelayout.h

···
 };
 
 struct nfs4_ff_layout_mirror {
-	struct pnfs_layout_segment	*lseg; /* back pointer */
+	struct pnfs_layout_hdr		*layout;
+	struct list_head		mirrors;
 	u32				ds_count;
 	u32				efficiency;
 	struct nfs4_ff_layout_ds	*mirror_ds;
···
 	u32				uid;
 	u32				gid;
 	struct rpc_cred			*cred;
+	atomic_t			ref;
 	spinlock_t			lock;
 	struct nfs4_ff_layoutstat	read_stat;
 	struct nfs4_ff_layoutstat	write_stat;
···
 struct nfs4_flexfile_layout {
 	struct pnfs_layout_hdr generic_hdr;
 	struct pnfs_ds_commit_info commit_info;
+	struct list_head	mirrors;
 	struct list_head	error_list; /* nfs4_ff_layout_ds_err */
 };
 

fs/nfs/flexfilelayout/flexfilelayoutdev.c

···
 	return NULL;
 }
 
+static void ff_layout_mark_devid_invalid(struct pnfs_layout_segment *lseg,
+		struct nfs4_deviceid_node *devid)
+{
+	nfs4_mark_deviceid_unavailable(devid);
+	if (!ff_layout_has_available_ds(lseg))
+		pnfs_error_mark_layout_for_return(lseg->pls_layout->plh_inode,
+				lseg);
+}
+
+static bool ff_layout_mirror_valid(struct pnfs_layout_segment *lseg,
+		struct nfs4_ff_layout_mirror *mirror)
+{
+	if (mirror == NULL || mirror->mirror_ds == NULL) {
+		pnfs_error_mark_layout_for_return(lseg->pls_layout->plh_inode,
+					lseg);
+		return false;
+	}
+	if (mirror->mirror_ds->ds == NULL) {
+		struct nfs4_deviceid_node *devid;
+		devid = &mirror->mirror_ds->id_node;
+		ff_layout_mark_devid_invalid(lseg, devid);
+		return false;
+	}
+	return true;
+}
+
 static u64
 end_offset(u64 start, u64 len)
 {
···
 {
 	struct nfs4_ff_layout_mirror *mirror = FF_LAYOUT_COMP(lseg, mirror_idx);
 	struct nfs_fh *fh = NULL;
-	struct nfs4_deviceid_node *devid;
 
-	if (mirror == NULL || mirror->mirror_ds == NULL ||
-	    mirror->mirror_ds->ds == NULL) {
-		printk(KERN_ERR "NFS: %s: No data server for mirror offset index %d\n",
+	if (!ff_layout_mirror_valid(lseg, mirror)) {
+		pr_err_ratelimited("NFS: %s: No data server for mirror offset index %d\n",
 			__func__, mirror_idx);
-		if (mirror && mirror->mirror_ds) {
-			devid = &mirror->mirror_ds->id_node;
-			pnfs_generic_mark_devid_invalid(devid);
-		}
 		goto out;
 	}
 
···
 	unsigned int max_payload;
 	rpc_authflavor_t flavor;
 
-	if (mirror == NULL || mirror->mirror_ds == NULL ||
-	    mirror->mirror_ds->ds == NULL) {
-		printk(KERN_ERR "NFS: %s: No data server for offset index %d\n",
+	if (!ff_layout_mirror_valid(lseg, mirror)) {
+		pr_err_ratelimited("NFS: %s: No data server for offset index %d\n",
 			__func__, ds_idx);
-		if (mirror && mirror->mirror_ds) {
-			devid = &mirror->mirror_ds->id_node;
-			pnfs_generic_mark_devid_invalid(devid);
-		}
 		goto out;
 	}
 
···
 					   range->offset, range->length))
 			continue;
 		/* offset(8) + length(8) + stateid(NFS4_STATEID_SIZE)
-		 * + deviceid(NFS4_DEVICEID4_SIZE) + status(4) + opnum(4)
+		 * + array length + deviceid(NFS4_DEVICEID4_SIZE)
+		 * + status(4) + opnum(4)
 		 */
 		p = xdr_reserve_space(xdr,
-				24 + NFS4_STATEID_SIZE + NFS4_DEVICEID4_SIZE);
+				28 + NFS4_STATEID_SIZE + NFS4_DEVICEID4_SIZE);
 		if (unlikely(!p))
 			return -ENOBUFS;
 		p = xdr_encode_hyper(p, err->offset);
 		p = xdr_encode_hyper(p, err->length);
 		p = xdr_encode_opaque_fixed(p, &err->stateid,
 					    NFS4_STATEID_SIZE);
+		/* Encode 1 error */
+		*p++ = cpu_to_be32(1);
 		p = xdr_encode_opaque_fixed(p, &err->deviceid,
 					    NFS4_DEVICEID4_SIZE);
 		*p++ = cpu_to_be32(err->status);
···
 	return 0;
 }
 
-bool ff_layout_has_available_ds(struct pnfs_layout_segment *lseg)
+static bool ff_read_layout_has_available_ds(struct pnfs_layout_segment *lseg)
 {
 	struct nfs4_ff_layout_mirror *mirror;
 	struct nfs4_deviceid_node *devid;
-	int idx;
+	u32 idx;
 
 	for (idx = 0; idx < FF_LAYOUT_MIRROR_COUNT(lseg); idx++) {
 		mirror = FF_LAYOUT_COMP(lseg, idx);
···
 	}
 
 	return false;
+}
+
+static bool ff_rw_layout_has_available_ds(struct pnfs_layout_segment *lseg)
+{
+	struct nfs4_ff_layout_mirror *mirror;
+	struct nfs4_deviceid_node *devid;
+	u32 idx;
+
+	for (idx = 0; idx < FF_LAYOUT_MIRROR_COUNT(lseg); idx++) {
+		mirror = FF_LAYOUT_COMP(lseg, idx);
+		if (!mirror || !mirror->mirror_ds)
+			return false;
+		devid = &mirror->mirror_ds->id_node;
+		if (ff_layout_test_devid_unavailable(devid))
+			return false;
+	}
+
+	return FF_LAYOUT_MIRROR_COUNT(lseg) != 0;
+}
+
+bool ff_layout_has_available_ds(struct pnfs_layout_segment *lseg)
+{
+	if (lseg->pls_range.iomode == IOMODE_READ)
+		return  ff_read_layout_has_available_ds(lseg);
+	/* Note: RW layout needs all mirrors available */
+	return ff_rw_layout_has_available_ds(lseg);
 }
 
 module_param(dataserver_retrans, uint, 0644);

fs/nfs/inode.c

···
 {
 	struct inode *inode = d_inode(dentry);
 	struct nfs_fattr *fattr;
-	int error = -ENOMEM;
+	int error = 0;
 
 	nfs_inc_stats(inode, NFSIOS_VFSSETATTR);
 
···
 		attr->ia_valid &= ~ATTR_MODE;
 
 	if (attr->ia_valid & ATTR_SIZE) {
-		loff_t i_size;
-
 		BUG_ON(!S_ISREG(inode->i_mode));
 
-		i_size = i_size_read(inode);
-		if (attr->ia_size == i_size)
+		error = inode_newsize_ok(inode, attr->ia_size);
+		if (error)
+			return error;
+
+		if (attr->ia_size == i_size_read(inode))
 			attr->ia_valid &= ~ATTR_SIZE;
-		else if (attr->ia_size < i_size && IS_SWAPFILE(inode))
-			return -ETXTBSY;
 	}
 
 	/* Optimization: if the end result is no change, don't RPC */
···
 		nfs_sync_inode(inode);
 
 	fattr = nfs_alloc_fattr();
-	if (fattr == NULL)
+	if (fattr == NULL) {
+		error = -ENOMEM;
 		goto out;
+	}
+
 	/*
 	 * Return any delegations if we're going to change ACLs
 	 */
···
  * @ctx: pointer to context
  * @is_sync: is this a synchronous close
  *
- * always ensure that the attributes are up to date if we're mounted
- * with close-to-open semantics
+ * Ensure that the attributes are up to date if we're mounted
+ * with close-to-open semantics and we have cached data that will
+ * need to be revalidated on open.
  */
 void nfs_close_context(struct nfs_open_context *ctx, int is_sync)
 {
+	struct nfs_inode *nfsi;
 	struct inode *inode;
 	struct nfs_server *server;
 
···
 	if (!is_sync)
 		return;
 	inode = d_inode(ctx->dentry);
-	if (!list_empty(&NFS_I(inode)->open_files))
+	nfsi = NFS_I(inode);
+	if (inode->i_mapping->nrpages == 0)
+		return;
+	if (nfsi->cache_validity & NFS_INO_INVALID_DATA)
+		return;
+	if (!list_empty(&nfsi->open_files))
 		return;
 	server = NFS_SERVER(inode);
 	if (server->flags & NFS_MOUNT_NOCTO)
···
 }
 EXPORT_SYMBOL_GPL(put_nfs_open_context);
 
+static void put_nfs_open_context_sync(struct nfs_open_context *ctx)
+{
+	__put_nfs_open_context(ctx, 1);
+}
+
 /*
  * Ensure that mmap has a recent RPC credential for use when writing out
  * shared pages
···
 	return ctx;
 }
 
-static void nfs_file_clear_open_context(struct file *filp)
+void nfs_file_clear_open_context(struct file *filp)
 {
 	struct nfs_open_context *ctx = nfs_file_open_context(filp);
 
···
 		spin_lock(&inode->i_lock);
 		list_move_tail(&ctx->list, &NFS_I(inode)->open_files);
 		spin_unlock(&inode->i_lock);
-		__put_nfs_open_context(ctx, filp->f_flags & O_DIRECT ? 0 : 1);
+		put_nfs_open_context_sync(ctx);
 	}
 }
 
···
 	nfs_file_set_open_context(filp, ctx);
 	put_nfs_open_context(ctx);
 	nfs_fscache_open_file(inode, filp);
-	return 0;
-}
-
-int nfs_release(struct inode *inode, struct file *filp)
-{
-	nfs_file_clear_open_context(filp);
 	return 0;
 }
 
···
 	return 0;
 }
 
-static int nfs_ctime_need_update(const struct inode *inode, const struct nfs_fattr *fattr)
-{
-	if (!(fattr->valid & NFS_ATTR_FATTR_CTIME))
-		return 0;
-	return timespec_compare(&fattr->ctime, &inode->i_ctime) > 0;
-}
-
 static atomic_long_t nfs_attr_generation_counter;
 
 static unsigned long nfs_read_attr_generation_counter(void)
···
 	const struct nfs_inode *nfsi = NFS_I(inode);
 
 	return ((long)fattr->gencount - (long)nfsi->attr_gencount) > 0 ||
-		nfs_ctime_need_update(inode, fattr) ||
 		((long)nfsi->attr_gencount - (long)nfs_read_attr_generation_counter() > 0);
 }
 
···
 static int nfs_post_op_update_inode_locked(struct inode *inode, struct nfs_fattr *fattr)
 {
 	unsigned long invalid = NFS_INO_INVALID_ATTR|NFS_INO_REVAL_PAGECACHE;
+
+	/*
+	 * Don't revalidate the pagecache if we hold a delegation, but do
+	 * force an attribute update
+	 */
+	if (NFS_PROTO(inode)->have_delegation(inode, FMODE_READ))
+		invalid = NFS_INO_INVALID_ATTR|NFS_INO_REVAL_FORCED;
 
 	if (S_ISDIR(inode->i_mode))
 		invalid |= NFS_INO_INVALID_DATA;

fs/nfs/internal.h

···
 }
 #endif
 
-#ifdef CONFIG_NFS_V4_1
-int nfs_sockaddr_match_ipaddr(const struct sockaddr *, const struct sockaddr *);
-#endif
-
 /* callback_xdr.c */
 extern struct svc_version nfs4_callback_version1;
 extern struct svc_version nfs4_callback_version4;
···
 /* file.c */
 int nfs_file_fsync_commit(struct file *, loff_t, loff_t, int);
 loff_t nfs_file_llseek(struct file *, loff_t, int);
-int nfs_file_flush(struct file *, fl_owner_t);
 ssize_t nfs_file_read(struct kiocb *, struct iov_iter *);
 ssize_t nfs_file_splice_read(struct file *, loff_t *, struct pipe_inode_info *,
 			     size_t, unsigned int);
···
 void nfs_commitdata_release(struct nfs_commit_data *data);
 void nfs_request_add_commit_list(struct nfs_page *req, struct list_head *dst,
 				 struct nfs_commit_info *cinfo);
+void nfs_request_add_commit_list_locked(struct nfs_page *req,
+		struct list_head *dst,
+		struct nfs_commit_info *cinfo);
 void nfs_request_remove_commit_list(struct nfs_page *req,
 				    struct nfs_commit_info *cinfo);
 void nfs_init_cinfo(struct nfs_commit_info *cinfo,
···
  * Record the page as unstable and mark its inode as dirty.
  */
 static inline
-void nfs_mark_page_unstable(struct page *page)
+void nfs_mark_page_unstable(struct page *page, struct nfs_commit_info *cinfo)
 {
-	struct inode *inode = page_file_mapping(page)->host;
+	if (!cinfo->dreq) {
+		struct inode *inode = page_file_mapping(page)->host;
 
-	inc_zone_page_state(page, NR_UNSTABLE_NFS);
-	inc_wb_stat(&inode_to_bdi(inode)->wb, WB_RECLAIMABLE);
-	 __mark_inode_dirty(inode, I_DIRTY_DATASYNC);
+		inc_zone_page_state(page, NR_UNSTABLE_NFS);
+		inc_wb_stat(&inode_to_bdi(inode)->wb, WB_RECLAIMABLE);
+		__mark_inode_dirty(inode, I_DIRTY_DATASYNC);
+	}
 }
 
 /*

fs/nfs/nfs3xdr.c

···
 {
 	encode_diropargs3(xdr, args->fromfh, args->fromname, args->fromlen);
 	encode_symlinkdata3(xdr, args);
+	xdr->buf->flags |= XDRBUF_WRITE;
 }
 
 /*

fs/nfs/nfs42.h

···
 loff_t nfs42_proc_llseek(struct file *, loff_t, int);
 int nfs42_proc_layoutstats_generic(struct nfs_server *,
 				   struct nfs42_layoutstat_data *);
-/* nfs4.2xdr.h */
-extern struct rpc_procinfo nfs4_2_procedures[];
 
 #endif /* __LINUX_FS_NFS_NFS4_2_H */

fs/nfs/nfs42xdr.c

···
 	return -EIO;
 }
 
-static int decode_layoutstats(struct xdr_stream *xdr,
-			      struct nfs42_layoutstat_res *res)
+static int decode_layoutstats(struct xdr_stream *xdr)
 {
 	return decode_op_hdr(xdr, OP_LAYOUTSTATS);
 }
···
 		goto out;
 	WARN_ON(res->num_dev > PNFS_LAYOUTSTATS_MAXDEV);
 	for (i = 0; i < res->num_dev; i++) {
-		status = decode_layoutstats(xdr, res);
+		status = decode_layoutstats(xdr);
 		if (status)
 			goto out;
 	}

fs/nfs/nfs4_fs.h

···
 int nfs41_discover_server_trunking(struct nfs_client *clp,
 			struct nfs_client **, struct rpc_cred *);
 extern void nfs4_schedule_session_recovery(struct nfs4_session *, int);
-extern void nfs41_server_notify_target_slotid_update(struct nfs_client *clp);
-extern void nfs41_server_notify_highest_slotid_update(struct nfs_client *clp);
-
+extern void nfs41_notify_server(struct nfs_client *);
 #else
 static inline void nfs4_schedule_session_recovery(struct nfs4_session *session, int err)
 {

fs/nfs/nfs4client.c

···
 		return false;
 
 	/* Match only the IP address, not the port number */
-	if (!nfs_sockaddr_match_ipaddr(addr, clap))
-		return false;
-
-	return true;
+	return rpc_cmp_addr(addr, clap);
 }
 
 /*

fs/nfs/nfs4file.c

···
 #include <linux/fs.h>
 #include <linux/falloc.h>
 #include <linux/nfs_fs.h>
+#include "delegation.h"
 #include "internal.h"
+#include "iostat.h"
 #include "fscache.h"
 #include "pnfs.h"
 
···
 	struct inode *dir;
 	unsigned openflags = filp->f_flags;
 	struct iattr attr;
-	int opened = 0;
 	int err;
 
 	/*
···
 		nfs_sync_inode(inode);
 	}
 
-	inode = NFS_PROTO(dir)->open_context(dir, ctx, openflags, &attr, &opened);
+	inode = NFS_PROTO(dir)->open_context(dir, ctx, openflags, &attr, NULL);
 	if (IS_ERR(inode)) {
 		err = PTR_ERR(inode);
 		switch (err) {
···
 	d_drop(dentry);
 	err = -EOPENSTALE;
 	goto out_put_ctx;
+}
+
+/*
+ * Flush all dirty pages, and check for write errors.
+ */
+static int
+nfs4_file_flush(struct file *file, fl_owner_t id)
+{
+	struct inode	*inode = file_inode(file);
+
+	dprintk("NFS: flush(%pD2)\n", file);
+
+	nfs_inc_stats(inode, NFSIOS_VFSFLUSH);
+	if ((file->f_mode & FMODE_WRITE) == 0)
+		return 0;
+
+	/*
+	 * If we're holding a write delegation, then check if we're required
+	 * to flush the i/o on close. If not, then just start the i/o now.
+	 */
+	if (!nfs4_delegation_flush_on_close(inode))
+		return filemap_fdatawrite(file->f_mapping);
+
+	/* Flush writes to the server and return any errors */
+	return vfs_fsync(file, 0);
 }
 
 static int
···
 	.write_iter	= nfs_file_write,
 	.mmap		= nfs_file_mmap,
 	.open		= nfs4_file_open,
-	.flush		= nfs_file_flush,
+	.flush		= nfs4_file_flush,
 	.release	= nfs_file_release,
 	.fsync		= nfs4_file_fsync,
 	.lock		= nfs_lock,

fs/nfs/nfs4idmap.c

···
 	.read		= user_read,
 };
 
-static int nfs_idmap_init_keyring(void)
+int nfs_idmap_init(void)
 {
 	struct cred *cred;
 	struct key *keyring;
···
 	return ret;
 }
 
-static void nfs_idmap_quit_keyring(void)
+void nfs_idmap_quit(void)
 {
 	key_revoke(id_resolver_cache->thread_keyring);
 	unregister_key_type(&key_type_id_resolver);
···
 			&idmap->idmap_pdo);
 	rpc_destroy_pipe_data(idmap->idmap_pipe);
 	kfree(idmap);
-}
-
-int nfs_idmap_init(void)
-{
-	return nfs_idmap_init_keyring();
-}
-
-void nfs_idmap_quit(void)
-{
-	nfs_idmap_quit_keyring();
 }
 
 static int nfs_idmap_prepare_message(char *desc, struct idmap *idmap,

fs/nfs/nfs4proc.c

···
 	spin_unlock(&tbl->slot_tbl_lock);
 	res->sr_slot = NULL;
 	if (send_new_highest_used_slotid)
-		nfs41_server_notify_highest_slotid_update(session->clp);
+		nfs41_notify_server(session->clp);
 }
 
 int nfs41_sequence_done(struct rpc_task *task, struct nfs4_sequence_res *res)
···
 	return ret;
 }
 
-static int can_open_delegated(struct nfs_delegation *delegation, fmode_t fmode)
+static int can_open_delegated(struct nfs_delegation *delegation, fmode_t fmode,
+		enum open_claim_type4 claim)
 {
 	if (delegation == NULL)
 		return 0;
···
 		return 0;
 	if (test_bit(NFS_DELEGATION_RETURNING, &delegation->flags))
 		return 0;
+	switch (claim) {
+	case NFS4_OPEN_CLAIM_NULL:
+	case NFS4_OPEN_CLAIM_FH:
+		break;
+	case NFS4_OPEN_CLAIM_PREVIOUS:
+		if (!test_bit(NFS_DELEGATION_NEED_RECLAIM, &delegation->flags))
+			break;
+	default:
+		return 0;
+	}
 	nfs_mark_delegation_referenced(delegation);
 	return 1;
 }
···
 }
 
 static void nfs_clear_open_stateid_locked(struct nfs4_state *state,
+		nfs4_stateid *arg_stateid,
 		nfs4_stateid *stateid, fmode_t fmode)
 {
 	clear_bit(NFS_O_RDWR_STATE, &state->flags);
···
 	if (stateid == NULL)
 		return;
 	/* Handle races with OPEN */
-	if (!nfs4_stateid_match_other(stateid, &state->open_stateid) ||
-	    !nfs4_stateid_is_newer(stateid, &state->open_stateid)) {
+	if (!nfs4_stateid_match_other(arg_stateid, &state->open_stateid) ||
+	    (nfs4_stateid_match_other(stateid, &state->open_stateid) &&
+	    !nfs4_stateid_is_newer(stateid, &state->open_stateid))) {
 		nfs_resync_open_stateid_locked(state);
 		return;
 	}
···
 	nfs4_stateid_copy(&state->open_stateid, stateid);
 }
 
-static void nfs_clear_open_stateid(struct nfs4_state *state, nfs4_stateid *stateid, fmode_t fmode)
+static void nfs_clear_open_stateid(struct nfs4_state *state,
+	nfs4_stateid *arg_stateid,
+	nfs4_stateid *stateid, fmode_t fmode)
 {
 	write_seqlock(&state->seqlock);
-	nfs_clear_open_stateid_locked(state, stateid, fmode);
+	nfs_clear_open_stateid_locked(state, arg_stateid, stateid, fmode);
 	write_sequnlock(&state->seqlock);
 	if (test_bit(NFS_STATE_RECLAIM_NOGRACE, &state->flags))
 		nfs4_schedule_state_manager(state->owner->so_server->nfs_client);
···
 	struct nfs_delegation *delegation;
 	int open_mode = opendata->o_arg.open_flags;
 	fmode_t fmode = opendata->o_arg.fmode;
+	enum open_claim_type4 claim = opendata->o_arg.claim;
 	nfs4_stateid stateid;
 	int ret = -EAGAIN;
 
···
 		spin_unlock(&state->owner->so_lock);
 		rcu_read_lock();
 		delegation = rcu_dereference(nfsi->delegation);
-		if (!can_open_delegated(delegation, fmode)) {
+		if (!can_open_delegated(delegation, fmode, claim)) {
 			rcu_read_unlock();
 			break;
 		}
···
 	struct nfs4_opendata *data = calldata;
 	struct nfs4_state_owner *sp = data->owner;
 	struct nfs_client *clp = sp->so_server->nfs_client;
+	enum open_claim_type4 claim = data->o_arg.claim;
 
 	if (nfs_wait_on_sequence(data->o_arg.seqid, task) != 0)
 		goto out_wait;
···
 			goto out_no_action;
 		rcu_read_lock();
 		delegation = rcu_dereference(NFS_I(data->state->inode)->delegation);
-		if (data->o_arg.claim != NFS4_OPEN_CLAIM_DELEGATE_CUR &&
-		    data->o_arg.claim != NFS4_OPEN_CLAIM_DELEG_CUR_FH &&
-		    can_open_delegated(delegation, data->o_arg.fmode))
+		if (can_open_delegated(delegation, data->o_arg.fmode, claim))
 			goto unlock_no_action;
 		rcu_read_unlock();
 	}
 	/* Update client id. */
 	data->o_arg.clientid = clp->cl_clientid;
-	switch (data->o_arg.claim) {
+	switch (claim) {
+	default:
+		break;
 	case NFS4_OPEN_CLAIM_PREVIOUS:
 	case NFS4_OPEN_CLAIM_DELEG_CUR_FH:
 	case NFS4_OPEN_CLAIM_DELEG_PREV_FH:
···
  * fields corresponding to attributes that were used to store the verifier.
  * Make sure we clobber those fields in the later setattr call
  */
-static inline void nfs4_exclusive_attrset(struct nfs4_opendata *opendata, struct iattr *sattr)
+static inline void nfs4_exclusive_attrset(struct nfs4_opendata *opendata,
+				struct iattr *sattr, struct nfs4_label **label)
 {
-	if ((opendata->o_res.attrset[1] & FATTR4_WORD1_TIME_ACCESS) &&
+	const u32 *attrset = opendata->o_res.attrset;
+
+	if ((attrset[1] & FATTR4_WORD1_TIME_ACCESS) &&
 	    !(sattr->ia_valid & ATTR_ATIME_SET))
 		sattr->ia_valid |= ATTR_ATIME;
 
-	if ((opendata->o_res.attrset[1] & FATTR4_WORD1_TIME_MODIFY) &&
+	if ((attrset[1] & FATTR4_WORD1_TIME_MODIFY) &&
 	    !(sattr->ia_valid & ATTR_MTIME_SET))
 		sattr->ia_valid |= ATTR_MTIME;
+
+	/* Except MODE, it seems harmless of setting twice. */
+	if ((attrset[1] & FATTR4_WORD1_MODE))
+		sattr->ia_valid &= ~ATTR_MODE;
+
+	if (attrset[2] & FATTR4_WORD2_SECURITY_LABEL)
+		*label = NULL;
 }
 
 static int _nfs4_open_and_get_state(struct nfs4_opendata *opendata,
···
 		goto err_free_label;
 	state = ctx->state;
 
-	if ((opendata->o_arg.open_flags & O_EXCL) &&
+	if ((opendata->o_arg.open_flags & (O_CREAT|O_EXCL)) == (O_CREAT|O_EXCL) &&
 	    (opendata->o_arg.createmode != NFS4_CREATE_GUARDED)) {
-		nfs4_exclusive_attrset(opendata, sattr);
+		nfs4_exclusive_attrset(opendata, sattr, &label);
 
 		nfs_fattr_init(opendata->o_res.f_attr);
 		status = nfs4_do_setattr(state->inode, cred,
···
 			nfs_setsecurity(state->inode, opendata->o_res.f_attr, olabel);
 		}
 	}
-	if (opendata->file_created)
+	if (opened && opendata->file_created)
 		*opened |= FILE_CREATED;
 
 	if (pnfs_use_threshold(ctx_th, opendata->f_attr.mdsthreshold, server)) {
···
 	switch (task->tk_status) {
 		case 0:
 			res_stateid = &calldata->res.stateid;
-			if (calldata->arg.fmode == 0 && calldata->roc)
+			if (calldata->roc)
 				pnfs_roc_set_barrier(state->inode,
 						     calldata->roc_barrier);
 			renew_lease(server, calldata->timestamp);
···
 				goto out_release;
 			}
 	}
-	nfs_clear_open_stateid(state, res_stateid, calldata->arg.fmode);
+	nfs_clear_open_stateid(state, &calldata->arg.stateid,
+			res_stateid, calldata->arg.fmode);
 out_release:
 	nfs_release_seqid(calldata->arg.seqid);
 	nfs_refresh_inode(calldata->inode, calldata->res.fattr);
···
 		goto out_no_action;
 	}
 
-	if (calldata->arg.fmode == 0) {
+	if (calldata->arg.fmode == 0)
 		task->tk_msg.rpc_proc = &nfs4_procedures[NFSPROC4_CLNT_CLOSE];
-		if (calldata->roc &&
-		    pnfs_roc_drain(inode, &calldata->roc_barrier, task)) {
-			nfs_release_seqid(calldata->arg.seqid);
-			goto out_wait;
-		    }
-	}
+	if (calldata->roc)
+		pnfs_roc_get_barrier(inode, &calldata->roc_barrier);
+
 	calldata->arg.share_access =
 		nfs4_map_atomic_open_share(NFS_SERVER(inode),
 				calldata->arg.fmode, 0);
···
 
 static int _nfs4_server_capabilities(struct nfs_server *server, struct nfs_fh *fhandle)
 {
+	u32 bitmask[3] = {}, minorversion = server->nfs_client->cl_minorversion;
 	struct nfs4_server_caps_arg args = {
 		.fhandle = fhandle,
+		.bitmask = bitmask,
 	};
 	struct nfs4_server_caps_res res = {};
 	struct rpc_message msg = {
···
 	};
 	int status;
 
+	bitmask[0] = FATTR4_WORD0_SUPPORTED_ATTRS |
+		     FATTR4_WORD0_FH_EXPIRE_TYPE |
+		     FATTR4_WORD0_LINK_SUPPORT |
+		     FATTR4_WORD0_SYMLINK_SUPPORT |
+		     FATTR4_WORD0_ACLSUPPORT;
+	if (minorversion)
+		bitmask[2] = FATTR4_WORD2_SUPPATTR_EXCLCREAT;
+
 	status = nfs4_call_sync(server->client, server, &msg, &args.seq_args, &res.seq_res, 0);
 	if (status == 0) {
 		/* Sanity check the server answers */
-		switch (server->nfs_client->cl_minorversion) {
+		switch (minorversion) {
 		case 0:
 			res.attr_bitmask[1] &= FATTR4_WORD1_NFS40_MASK;
 			res.attr_bitmask[2] = 0;
···
 		server->cache_consistency_bitmask[0] &= FATTR4_WORD0_CHANGE|FATTR4_WORD0_SIZE;
 		server->cache_consistency_bitmask[1] &= FATTR4_WORD1_TIME_METADATA|FATTR4_WORD1_TIME_MODIFY;
 		server->cache_consistency_bitmask[2] = 0;
+		memcpy(server->exclcreat_bitmask, res.exclcreat_bitmask,
+			sizeof(server->exclcreat_bitmask));
 		server->acl_bitmask = res.acl_bitmask;
 		server->fh_expire_type = res.fh_expire_type;
 	}
···
 	struct nfs4_label l, *ilabel = NULL;
 	struct nfs_open_context *ctx;
 	struct nfs4_state *state;
-	int opened = 0;
 	int status = 0;
 
 	ctx = alloc_nfs_open_context(dentry, FMODE_READ);
···
 	ilabel = nfs4_label_init_security(dir, dentry, sattr, &l);
 
 	sattr->ia_mode &= ~current_umask();
-	state = nfs4_do_open(dir, ctx, flags, sattr, ilabel, &opened);
+	state = nfs4_do_open(dir, ctx, flags, sattr, ilabel, NULL);
 	if (IS_ERR(state)) {
 		status = PTR_ERR(state);
 		goto out;
···
 	int result;
 	size_t len;
 	char *str;
-	bool retried = false;
 
 	if (clp->cl_owner_id != NULL)
 		return 0;
-retry:
+
 	rcu_read_lock();
-	len = 10 + strlen(clp->cl_ipaddr) + 1 +
+	len = 14 + strlen(clp->cl_ipaddr) + 1 +
 		strlen(rpc_peeraddr2str(clp->cl_rpcclient, RPC_DISPLAY_ADDR)) +
 		1 +
 		strlen(rpc_peeraddr2str(clp->cl_rpcclient, RPC_DISPLAY_PROTO)) +
···
 			rpc_peeraddr2str(clp->cl_rpcclient, RPC_DISPLAY_PROTO));
 	rcu_read_unlock();
 
-	/* Did something change? */
-	if (result >= len) {
-		kfree(str);
-		if (retried)
-			return -EINVAL;
-		retried = true;
-		goto retry;
-	}
 	clp->cl_owner_id = str;
 	return 0;
 }
···
 			clp->rpc_ops->version, clp->cl_minorversion,
 			nfs4_client_id_uniquifier,
 			clp->cl_rpcclient->cl_nodename);
-	if (result >= len) {
-		kfree(str);
-		return -EINVAL;
-	}
 	clp->cl_owner_id = str;
 	return 0;
 }
···
 	result = scnprintf(str, len, "Linux NFSv%u.%u %s",
 			clp->rpc_ops->version, clp->cl_minorversion,
 			clp->cl_rpcclient->cl_nodename);
-	if (result >= len) {
-		kfree(str);
-		return -EINVAL;
-	}
 	clp->cl_owner_id = str;
 	return 0;
 }
···
 
 	d_data = (struct nfs4_delegreturndata *)data;
 
-	if (d_data->roc &&
-	    pnfs_roc_drain(d_data->inode, &d_data->roc_barrier, task))
-		return;
+	if (d_data->roc)
+		pnfs_roc_get_barrier(d_data->inode, &d_data->roc_barrier);
 
 	nfs4_setup_sequence(d_data->res.server,
 			&d_data->args.seq_args,
···
 	case 0:
 		goto out;
 	/*
+	 * NFS4ERR_BADLAYOUT means the MDS cannot return a layout of
+	 * length lgp->args.minlength != 0 (see RFC5661 section 18.43.3).
+	 */
+	case -NFS4ERR_BADLAYOUT:
+		goto out_overflow;
+	/*
 	 * NFS4ERR_LAYOUTTRYLATER is a conflict with another client
-	 * (or clients) writing to the same RAID stripe
+	 * (or clients) writing to the same RAID stripe except when
+	 * the minlength argument is 0 (see RFC5661 section 18.43.3).
 	 */
 	case -NFS4ERR_LAYOUTTRYLATER:
+		if (lgp->args.minlength == 0)
+			goto out_overflow;
 	/*
 	 * NFS4ERR_RECALLCONFLICT is when conflict with self (must recall
 	 * existing layout before getting a new one).
···
 		rpc_restart_call_prepare(task);
 out:
 	dprintk("<-- %s\n", __func__);
+	return;
+out_overflow:
+	task->tk_status = -EOVERFLOW;
+	goto out;
 }
 
 static size_t max_response_pages(struct nfs_server *server)
···
 	.reboot_recovery_ops = &nfs41_reboot_recovery_ops,
 	.nograce_recovery_ops = &nfs41_nograce_recovery_ops,
 	.state_renewal_ops = &nfs41_state_renewal_ops,
+	.mig_recovery_ops = &nfs41_mig_recovery_ops,
 };
 #endif
 

fs/nfs/nfs4state.c

···
 }
 EXPORT_SYMBOL_GPL(nfs4_schedule_session_recovery);
 
-static void nfs41_ping_server(struct nfs_client *clp)
+void nfs41_notify_server(struct nfs_client *clp)
 {
 	/* Use CHECK_LEASE to ping the server with a SEQUENCE */
 	set_bit(NFS4CLNT_CHECK_LEASE, &clp->cl_state);
 	nfs4_schedule_state_manager(clp);
-}
-
-void nfs41_server_notify_target_slotid_update(struct nfs_client *clp)
-{
-	nfs41_ping_server(clp);
-}
-
-void nfs41_server_notify_highest_slotid_update(struct nfs_client *clp)
-{
-	nfs41_ping_server(clp);
 }
 
 static void nfs4_reset_all_state(struct nfs_client *clp)

fs/nfs/nfs4trace.h

···
 DEFINE_NFS4_GETATTR_EVENT(nfs4_lookup_root);
 DEFINE_NFS4_GETATTR_EVENT(nfs4_fsinfo);
 
+DECLARE_EVENT_CLASS(nfs4_inode_callback_event,
+		TP_PROTO(
+			const struct nfs_client *clp,
+			const struct nfs_fh *fhandle,
+			const struct inode *inode,
+			int error
+		),
+
+		TP_ARGS(clp, fhandle, inode, error),
+
+		TP_STRUCT__entry(
+			__field(int, error)
+			__field(dev_t, dev)
+			__field(u32, fhandle)
+			__field(u64, fileid)
+			__string(dstaddr, clp ?
+				rpc_peeraddr2str(clp->cl_rpcclient,
+					RPC_DISPLAY_ADDR) : "unknown")
+		),
+
+		TP_fast_assign(
+			__entry->error = error;
+			__entry->fhandle = nfs_fhandle_hash(fhandle);
+			if (inode != NULL) {
+				__entry->fileid = NFS_FILEID(inode);
+				__entry->dev = inode->i_sb->s_dev;
+			} else {
+				__entry->fileid = 0;
+				__entry->dev = 0;
+			}
+			__assign_str(dstaddr, clp ?
+				rpc_peeraddr2str(clp->cl_rpcclient,
+					RPC_DISPLAY_ADDR) : "unknown")
+		),
+
+		TP_printk(
+			"error=%d (%s) fileid=%02x:%02x:%llu fhandle=0x%08x "
+			"dstaddr=%s",
+			__entry->error,
+			show_nfsv4_errors(__entry->error),
+			MAJOR(__entry->dev), MINOR(__entry->dev),
+			(unsigned long long)__entry->fileid,
+			__entry->fhandle,
+			__get_str(dstaddr)
+		)
+);
+
+#define DEFINE_NFS4_INODE_CALLBACK_EVENT(name) \
+	DEFINE_EVENT(nfs4_inode_callback_event, name, \
+			TP_PROTO( \
+				const struct nfs_client *clp, \
+				const struct nfs_fh *fhandle, \
+				const struct inode *inode, \
+				int error \
+			), \
+			TP_ARGS(clp, fhandle, inode, error))
+DEFINE_NFS4_INODE_CALLBACK_EVENT(nfs4_cb_getattr);
+DEFINE_NFS4_INODE_CALLBACK_EVENT(nfs4_cb_layoutrecall_inode);
+
+
 DECLARE_EVENT_CLASS(nfs4_idmap_event,
 		TP_PROTO(
 			const char *name,
···
 
 DEFINE_NFS4_INODE_EVENT(nfs4_layoutcommit);
 DEFINE_NFS4_INODE_EVENT(nfs4_layoutreturn);
+DEFINE_NFS4_INODE_EVENT(nfs4_layoutreturn_on_close);
 
 #endif /* CONFIG_NFS_V4_1 */
 

fs/nfs/nfs4xdr.c

···
 #define decode_layoutcommit_maxsz (op_decode_hdr_maxsz + 3)
 #define encode_layoutreturn_maxsz (8 + op_encode_hdr_maxsz + \
 				encode_stateid_maxsz + \
-				1 /* FIXME: opaque lrf_body always empty at the moment */)
+				1 + \
+				XDR_QUADLEN(NFS4_OPAQUE_LIMIT))
 #define decode_layoutreturn_maxsz (op_decode_hdr_maxsz + \
 				1 + decode_stateid_maxsz)
 #define encode_secinfo_no_name_maxsz (op_encode_hdr_maxsz + 1)
···
 
 static void encode_attrs(struct xdr_stream *xdr, const struct iattr *iap,
 				const struct nfs4_label *label,
-				const struct nfs_server *server)
+				const struct nfs_server *server,
+				bool excl_check)
 {
 	char owner_name[IDMAP_NAMESZ];
 	char owner_group[IDMAP_NAMESZ];
···
 		bmval[1] |= FATTR4_WORD1_TIME_MODIFY_SET;
 		len += 4;
 	}
+
+	if (excl_check) {
+		const u32 *excl_bmval = server->exclcreat_bitmask;
+		bmval[0] &= excl_bmval[0];
+		bmval[1] &= excl_bmval[1];
+		bmval[2] &= excl_bmval[2];
+
+		if (!(excl_bmval[2] & FATTR4_WORD2_SECURITY_LABEL))
+			label = NULL;
+	}
+
 	if (label) {
 		len += 4 + 4 + 4 + (XDR_QUADLEN(label->len) << 2);
 		bmval[2] |= FATTR4_WORD2_SECURITY_LABEL;
···
 	case NF4LNK:
 		p = reserve_space(xdr, 4);
 		*p = cpu_to_be32(create->u.symlink.len);
-		xdr_write_pages(xdr, create->u.symlink.pages, 0, create->u.symlink.len);
+		xdr_write_pages(xdr, create->u.symlink.pages, 0,
+				create->u.symlink.len);
+		xdr->buf->flags |= XDRBUF_WRITE;
 		break;
 
 	case NF4BLK: case NF4CHR:
···
 	}
 
 	encode_string(xdr, create->name->len, create->name->name);
-	encode_attrs(xdr, create->attrs, create->label, create->server);
+	encode_attrs(xdr, create->attrs, create->label, create->server, false);
 }
 
 static void encode_getattr_one(struct xdr_stream *xdr, uint32_t bitmap, struct compound_hdr *hdr)
···
 
 static inline void encode_createmode(struct xdr_stream *xdr, const struct nfs_openargs *arg)
 {
-	struct iattr dummy;
 	__be32 *p;
 
 	p = reserve_space(xdr, 4);
 	switch(arg->createmode) {
 	case NFS4_CREATE_UNCHECKED:
 		*p = cpu_to_be32(NFS4_CREATE_UNCHECKED);
-		encode_attrs(xdr, arg->u.attrs, arg->label, arg->server);
+		encode_attrs(xdr, arg->u.attrs, arg->label, arg->server, false);
 		break;
 	case NFS4_CREATE_GUARDED:
 		*p = cpu_to_be32(NFS4_CREATE_GUARDED);
-		encode_attrs(xdr, arg->u.attrs, arg->label, arg->server);
+		encode_attrs(xdr, arg->u.attrs, arg->label, arg->server, false);
 		break;
 	case NFS4_CREATE_EXCLUSIVE:
 		*p = cpu_to_be32(NFS4_CREATE_EXCLUSIVE);
···
 	case NFS4_CREATE_EXCLUSIVE4_1:
 		*p = cpu_to_be32(NFS4_CREATE_EXCLUSIVE4_1);
 		encode_nfs4_verifier(xdr, &arg->u.verifier);
-		dummy.ia_valid = 0;
-		encode_attrs(xdr, &dummy, arg->label, arg->server);
+		encode_attrs(xdr, arg->u.attrs, arg->label, arg->server, true);
 	}
 }
 
···
 {
 	encode_op_hdr(xdr, OP_SETATTR, decode_setattr_maxsz, hdr);
 	encode_nfs4_stateid(xdr, &arg->stateid);
-	encode_attrs(xdr, arg->iap, arg->label, server);
+	encode_attrs(xdr, arg->iap, arg->label, server, false);
 }
 
 static void encode_setclientid(struct xdr_stream *xdr, const struct nfs4_setclientid *setclientid, struct compound_hdr *hdr)
···
 				     struct xdr_stream *xdr,
 				     struct nfs4_server_caps_arg *args)
 {
+	const u32 *bitmask = args->bitmask;
 	struct compound_hdr hdr = {
 		.minorversion = nfs4_xdr_minorversion(&args->seq_args),
 	};
···
 	encode_compound_hdr(xdr, req, &hdr);
 	encode_sequence(xdr, &args->seq_args, &hdr);
 	encode_putfh(xdr, args->fhandle, &hdr);
-	encode_getattr_one(xdr, FATTR4_WORD0_SUPPORTED_ATTRS|
-			   FATTR4_WORD0_FH_EXPIRE_TYPE|
-			   FATTR4_WORD0_LINK_SUPPORT|
-			   FATTR4_WORD0_SYMLINK_SUPPORT|
-			   FATTR4_WORD0_ACLSUPPORT, &hdr);
+	encode_getattr_three(xdr, bitmask[0], bitmask[1], bitmask[2], &hdr);
 	encode_nops(&hdr);
 }
 
···
 out_overflow:
 	print_overflow_msg(__func__, xdr);
 	return -EIO;
+}
+
+static int decode_attr_exclcreat_supported(struct xdr_stream *xdr,
+				 uint32_t *bitmap, uint32_t *bitmask)
+{
+	if (likely(bitmap[2] & FATTR4_WORD2_SUPPATTR_EXCLCREAT)) {
+		int ret;
+		ret = decode_attr_bitmap(xdr, bitmask);
+		if (unlikely(ret < 0))
+			return ret;
+		bitmap[2] &= ~FATTR4_WORD2_SUPPATTR_EXCLCREAT;
+	} else
+		bitmask[0] = bitmask[1] = bitmask[2] = 0;
+	dprintk("%s: bitmask=%08x:%08x:%08x\n", __func__,
+		bitmask[0], bitmask[1], bitmask[2]);
+	return 0;
 }
 
 static int decode_attr_filehandle(struct xdr_stream *xdr, uint32_t *bitmap, struct nfs_fh *fh)
···
 		goto xdr_error;
 	if ((status = decode_attr_aclsupport(xdr, bitmap, &res->acl_bitmask)) != 0)
 		goto xdr_error;
+	if ((status = decode_attr_exclcreat_supported(xdr, bitmap,
+				res->exclcreat_bitmask)) != 0)
+		goto xdr_error;
 	status = verify_attr_len(xdr, savep, attrlen);
 xdr_error:
 	dprintk("%s: xdr returned %d!\n", __func__, -status);
···
 }
 
 /* This is too sick! */
-static int decode_space_limit(struct xdr_stream *xdr, u64 *maxsize)
+static int decode_space_limit(struct xdr_stream *xdr,
+		unsigned long *pagemod_limit)
 {
 	__be32 *p;
 	uint32_t limit_type, nblocks, blocksize;
+	u64 maxsize = 0;
 
 	p = xdr_inline_decode(xdr, 12);
 	if (unlikely(!p))
 		goto out_overflow;
 	limit_type = be32_to_cpup(p++);
 	switch (limit_type) {
-	case 1:
-		xdr_decode_hyper(p, maxsize);
+	case NFS4_LIMIT_SIZE:
+		xdr_decode_hyper(p, &maxsize);
 		break;
-	case 2:
+	case NFS4_LIMIT_BLOCKS:
 		nblocks = be32_to_cpup(p++);
 		blocksize = be32_to_cpup(p);
-		*maxsize = (uint64_t)nblocks * (uint64_t)blocksize;
+		maxsize = (uint64_t)nblocks * (uint64_t)blocksize;
 	}
+	maxsize >>= PAGE_CACHE_SHIFT;
+	*pagemod_limit = min_t(u64, maxsize, ULONG_MAX);
 	return 0;
 out_overflow:
 	print_overflow_msg(__func__, xdr);
···
 		break;
 	case NFS4_OPEN_DELEGATE_WRITE:
 		res->delegation_type = FMODE_WRITE|FMODE_READ;
-		if (decode_space_limit(xdr, &res->maxsize) < 0)
+		if (decode_space_limit(xdr, &res->pagemod_limit) < 0)
 				return -EIO;
 	}
 	return decode_ace(xdr, NULL, res->server->nfs_client);

fs/nfs/pagelist.c

···
 void nfs_set_pgio_error(struct nfs_pgio_header *hdr, int error, loff_t pos)
 {
 	spin_lock(&hdr->lock);
-	if (pos < hdr->io_start + hdr->good_bytes) {
-		set_bit(NFS_IOHDR_ERROR, &hdr->flags);
+	if (!test_and_set_bit(NFS_IOHDR_ERROR, &hdr->flags)
+	    || pos < hdr->io_start + hdr->good_bytes) {
 		clear_bit(NFS_IOHDR_EOF, &hdr->flags);
 		hdr->good_bytes = pos - hdr->io_start;
 		hdr->error = error;

fs/nfs/pnfs.c

···
 	if (test_and_set_bit(NFS_LAYOUT_RETURN, &lo->plh_flags))
 		return false;
 	lo->plh_return_iomode = 0;
-	lo->plh_block_lgets++;
 	pnfs_get_layout_hdr(lo);
 	clear_bit(NFS_LAYOUT_RETURN_BEFORE_CLOSE, &lo->plh_flags);
 	return true;
···
 	return !pnfs_seqid_is_newer(seqid, lo->plh_barrier);
 }
 
-static bool
-pnfs_layout_returning(const struct pnfs_layout_hdr *lo,
-		      struct pnfs_layout_range *range)
-{
-	return test_bit(NFS_LAYOUT_RETURN, &lo->plh_flags) &&
-		(lo->plh_return_iomode == IOMODE_ANY ||
-		 lo->plh_return_iomode == range->iomode);
-}
-
 /* lget is set to 1 if called from inside send_layoutget call chain */
 static bool
-pnfs_layoutgets_blocked(const struct pnfs_layout_hdr *lo,
-			struct pnfs_layout_range *range, int lget)
+pnfs_layoutgets_blocked(const struct pnfs_layout_hdr *lo)
 {
 	return lo->plh_block_lgets ||
-		test_bit(NFS_LAYOUT_BULK_RECALL, &lo->plh_flags) ||
-		(list_empty(&lo->plh_segs) &&
-		 (atomic_read(&lo->plh_outstanding) > lget)) ||
-		pnfs_layout_returning(lo, range);
+		test_bit(NFS_LAYOUT_BULK_RECALL, &lo->plh_flags);
 }
 
 int
···
 
 	dprintk("--> %s\n", __func__);
 	spin_lock(&lo->plh_inode->i_lock);
-	if (pnfs_layoutgets_blocked(lo, range, 1)) {
+	if (pnfs_layoutgets_blocked(lo)) {
 		status = -EAGAIN;
 	} else if (!nfs4_valid_open_stateid(open_state)) {
 		status = -EBADF;
···
 	struct nfs_server *server = NFS_SERVER(ino);
 	struct nfs4_layoutget *lgp;
 	struct pnfs_layout_segment *lseg;
+	loff_t i_size;
 
 	dprintk("--> %s\n", __func__);
 
···
 	if (lgp == NULL)
 		return NULL;
 
+	i_size = i_size_read(ino);
+
 	lgp->args.minlength = PAGE_CACHE_SIZE;
 	if (lgp->args.minlength > range->length)
 		lgp->args.minlength = range->length;
+	if (range->iomode == IOMODE_READ) {
+		if (range->offset >= i_size)
+			lgp->args.minlength = 0;
+		else if (i_size - range->offset < lgp->args.minlength)
+			lgp->args.minlength = i_size - range->offset;
+	}
 	lgp->args.maxcount = PNFS_LAYOUT_MAXSIZE;
 	lgp->args.range = *range;
 	lgp->args.type = server->pnfs_curr_ld->id;
···
 	if (unlikely(lrp == NULL)) {
 		status = -ENOMEM;
 		spin_lock(&ino->i_lock);
-		lo->plh_block_lgets--;
 		pnfs_clear_layoutreturn_waitbit(lo);
-		rpc_wake_up(&NFS_SERVER(ino)->roc_rpcwaitq);
 		spin_unlock(&ino->i_lock);
 		pnfs_put_layout_hdr(lo);
 		goto out;
···
 	struct pnfs_layout_segment *lseg, *tmp;
 	nfs4_stateid stateid;
 	LIST_HEAD(tmp_list);
-	bool found = false, layoutreturn = false;
+	bool found = false, layoutreturn = false, roc = false;
 
 	spin_lock(&ino->i_lock);
 	lo = nfsi->layout;
-	if (!lo || !test_and_clear_bit(NFS_LAYOUT_ROC, &lo->plh_flags) ||
-	    test_bit(NFS_LAYOUT_BULK_RECALL, &lo->plh_flags))
+	if (!lo || test_bit(NFS_LAYOUT_BULK_RECALL, &lo->plh_flags))
 		goto out_noroc;
 
-	/* Don't return layout if we hold a delegation */
+	/* no roc if we hold a delegation */
 	if (nfs4_check_delegation(ino, FMODE_READ))
 		goto out_noroc;
 
···
 			goto out_noroc;
 	}
 
+	stateid = lo->plh_stateid;
+	/* always send layoutreturn if being marked so */
+	if (test_and_clear_bit(NFS_LAYOUT_RETURN_BEFORE_CLOSE,
+				   &lo->plh_flags))
+		layoutreturn = pnfs_prepare_layoutreturn(lo);
+
 	pnfs_clear_retry_layoutget(lo);
 	list_for_each_entry_safe(lseg, tmp, &lo->plh_segs, pls_list)
-		if (test_bit(NFS_LSEG_ROC, &lseg->pls_flags)) {
+		/* If we are sending layoutreturn, invalidate all valid lsegs */
+		if (layoutreturn || test_bit(NFS_LSEG_ROC, &lseg->pls_flags)) {
 			mark_lseg_invalid(lseg, &tmp_list);
 			found = true;
 		}
-	if (!found)
-		goto out_noroc;
-	lo->plh_block_lgets++;
-	pnfs_get_layout_hdr(lo); /* matched in pnfs_roc_release */
+	/* pnfs_prepare_layoutreturn() grabs lo ref and it will be put
+	 * in pnfs_roc_release(). We don't really send a layoutreturn but
+	 * still want others to view us like we are sending one!
+	 *
+	 * If pnfs_prepare_layoutreturn() fails, it means someone else is doing
+	 * LAYOUTRETURN, so we proceed like there are no layouts to return.
+	 *
+	 * ROC in three conditions:
+	 * 1. there are ROC lsegs
+	 * 2. we don't send layoutreturn
+	 * 3. no others are sending layoutreturn
+	 */
+	if (found && !layoutreturn && pnfs_prepare_layoutreturn(lo))
+		roc = true;
+
+out_noroc:
 	spin_unlock(&ino->i_lock);
 	pnfs_free_lseg_list(&tmp_list);
 	pnfs_layoutcommit_inode(ino, true);
-	return true;
-
-out_noroc:
-	if (lo) {
-		stateid = lo->plh_stateid;
-		if (test_and_clear_bit(NFS_LAYOUT_RETURN_BEFORE_CLOSE,
-					   &lo->plh_flags))
-			layoutreturn = pnfs_prepare_layoutreturn(lo);
-	}
-	spin_unlock(&ino->i_lock);
-	if (layoutreturn) {
-		pnfs_layoutcommit_inode(ino, true);
+	if (layoutreturn)
 		pnfs_send_layoutreturn(lo, stateid, IOMODE_ANY, true);
-	}
-	return false;
+	return roc;
 }
 
 void pnfs_roc_release(struct inode *ino)
···
 
 	spin_lock(&ino->i_lock);
 	lo = NFS_I(ino)->layout;
-	lo->plh_block_lgets--;
+	pnfs_clear_layoutreturn_waitbit(lo);
 	if (atomic_dec_and_test(&lo->plh_refcount)) {
 		pnfs_detach_layout_hdr(lo);
 		spin_unlock(&ino->i_lock);
···
 	if (pnfs_seqid_is_newer(barrier, lo->plh_barrier))
 		lo->plh_barrier = barrier;
 	spin_unlock(&ino->i_lock);
+	trace_nfs4_layoutreturn_on_close(ino, 0);
 }
 
-bool pnfs_roc_drain(struct inode *ino, u32 *barrier, struct rpc_task *task)
+void pnfs_roc_get_barrier(struct inode *ino, u32 *barrier)
 {
 	struct nfs_inode *nfsi = NFS_I(ino);
 	struct pnfs_layout_hdr *lo;
-	struct pnfs_layout_segment *lseg;
-	nfs4_stateid stateid;
 	u32 current_seqid;
-	bool layoutreturn = false;
 
 	spin_lock(&ino->i_lock);
-	list_for_each_entry(lseg, &nfsi->layout->plh_segs, pls_list) {
-		if (!test_bit(NFS_LSEG_ROC, &lseg->pls_flags))
-			continue;
-		if (test_bit(NFS_LSEG_VALID, &lseg->pls_flags))
-			continue;
-		rpc_sleep_on(&NFS_SERVER(ino)->roc_rpcwaitq, task, NULL);
-		spin_unlock(&ino->i_lock);
-		return true;
-	}
 	lo = nfsi->layout;
 	current_seqid = be32_to_cpu(lo->plh_stateid.seqid);
 
···
 	 * a barrier, we choose the worst-case barrier.
 	 */
 	*barrier = current_seqid + atomic_read(&lo->plh_outstanding);
-	stateid = lo->plh_stateid;
-	if (test_and_clear_bit(NFS_LAYOUT_RETURN_BEFORE_CLOSE,
-					   &lo->plh_flags))
-		layoutreturn = pnfs_prepare_layoutreturn(lo);
-	if (test_bit(NFS_LAYOUT_RETURN, &lo->plh_flags))
-		rpc_sleep_on(&NFS_SERVER(ino)->roc_rpcwaitq, task, NULL);
-
 	spin_unlock(&ino->i_lock);
-	if (layoutreturn) {
-		pnfs_send_layoutreturn(lo, stateid, IOMODE_ANY, false);
-		return true;
-	}
-	return false;
 }
 
 /*
···
 	return (int)(l1->iomode == IOMODE_READ) - (int)(l2->iomode == IOMODE_READ);
 }
 
-static void
-pnfs_layout_insert_lseg(struct pnfs_layout_hdr *lo,
-		   struct pnfs_layout_segment *lseg)
+static bool
+pnfs_lseg_range_is_after(const struct pnfs_layout_range *l1,
+		const struct pnfs_layout_range *l2)
 {
-	struct pnfs_layout_segment *lp;
+	return pnfs_lseg_range_cmp(l1, l2) > 0;
+}
+
+static bool
+pnfs_lseg_no_merge(struct pnfs_layout_segment *lseg,
+		struct pnfs_layout_segment *old)
+{
+	return false;
+}
+
+void
+pnfs_generic_layout_insert_lseg(struct pnfs_layout_hdr *lo,
+		   struct pnfs_layout_segment *lseg,
+		   bool (*is_after)(const struct pnfs_layout_range *,
+			   const struct pnfs_layout_range *),
+		   bool (*do_merge)(struct pnfs_layout_segment *,
+			   struct pnfs_layout_segment *),
+		   struct list_head *free_me)
+{
+	struct pnfs_layout_segment *lp, *tmp;
 
 	dprintk("%s:Begin\n", __func__);
 
-	list_for_each_entry(lp, &lo->plh_segs, pls_list) {
-		if (pnfs_lseg_range_cmp(&lseg->pls_range, &lp->pls_range) > 0)
+	list_for_each_entry_safe(lp, tmp, &lo->plh_segs, pls_list) {
+		if (test_bit(NFS_LSEG_VALID, &lp->pls_flags) == 0)
+			continue;
+		if (do_merge(lseg, lp)) {
+			mark_lseg_invalid(lp, free_me);
+			continue;
+		}
+		if (is_after(&lseg->pls_range, &lp->pls_range))
 			continue;
 		list_add_tail(&lseg->pls_list, &lp->pls_list);
 		dprintk("%s: inserted lseg %p "
···
 	pnfs_get_layout_hdr(lo);
 
 	dprintk("%s:Return\n", __func__);
+}
+EXPORT_SYMBOL_GPL(pnfs_generic_layout_insert_lseg);
+
+static void
+pnfs_layout_insert_lseg(struct pnfs_layout_hdr *lo,
+		   struct pnfs_layout_segment *lseg,
+		   struct list_head *free_me)
+{
+	struct inode *inode = lo->plh_inode;
+	struct pnfs_layoutdriver_type *ld = NFS_SERVER(inode)->pnfs_curr_ld;
+
+	if (ld->add_lseg != NULL)
+		ld->add_lseg(lo, lseg, free_me);
+	else
+		pnfs_generic_layout_insert_lseg(lo, lseg,
+				pnfs_lseg_range_is_after,
+				pnfs_lseg_no_merge,
+				free_me);
 }
 
 static struct pnfs_layout_hdr *
···
 			ret = pnfs_get_lseg(lseg);
 			break;
 		}
-		if (lseg->pls_range.offset > range->offset)
-			break;
 	}
 
 	dprintk("%s:Return lseg %p ref %d\n",
···
 
 static bool pnfs_prepare_to_retry_layoutget(struct pnfs_layout_hdr *lo)
 {
+	if (!pnfs_should_retry_layoutget(lo))
+		return false;
 	/*
 	 * send layoutcommit as it can hold up layoutreturn due to lseg
 	 * reference
···
 	bool first;
 
 	if (!pnfs_enabled_sb(NFS_SERVER(ino)))
+		goto out;
+
+	if (iomode == IOMODE_READ && i_size_read(ino) == 0)
 		goto out;
 
 	if (pnfs_within_mdsthreshold(ctx, ino, iomode))
···
 	 * Because we free lsegs before sending LAYOUTRETURN, we need to wait
 	 * for LAYOUTRETURN even if first is true.
 	 */
-	if (!lseg && pnfs_should_retry_layoutget(lo) &&
-	    test_bit(NFS_LAYOUT_RETURN, &lo->plh_flags)) {
+	if (test_bit(NFS_LAYOUT_RETURN, &lo->plh_flags)) {
 		spin_unlock(&ino->i_lock);
 		dprintk("%s wait for layoutreturn\n", __func__);
 		if (pnfs_prepare_to_retry_layoutget(lo)) {
···
 		goto out_put_layout_hdr;
 	}
 
-	if (pnfs_layoutgets_blocked(lo, &arg, 0))
+	if (pnfs_layoutgets_blocked(lo))
 		goto out_unlock;
 	atomic_inc(&lo->plh_outstanding);
 	spin_unlock(&ino->i_lock);
···
 }
 EXPORT_SYMBOL_GPL(pnfs_update_layout);
 
+static bool
+pnfs_sanity_check_layout_range(struct pnfs_layout_range *range)
+{
+	switch (range->iomode) {
+	case IOMODE_READ:
+	case IOMODE_RW:
+		break;
+	default:
+		return false;
+	}
+	if (range->offset == NFS4_MAX_UINT64)
+		return false;
+	if (range->length == 0)
+		return false;
+	if (range->length != NFS4_MAX_UINT64 &&
+	    range->length > NFS4_MAX_UINT64 - range->offset)
+		return false;
+	return true;
+}
+
 struct pnfs_layout_segment *
 pnfs_layout_process(struct nfs4_layoutget *lgp)
 {
···
 	struct pnfs_layout_segment *lseg;
 	struct inode *ino = lo->plh_inode;
 	LIST_HEAD(free_me);
-	int status = 0;
+	int status = -EINVAL;
+
+	if (!pnfs_sanity_check_layout_range(&res->range))
+		goto out;
 
 	/* Inject layout blob into I/O device driver */
 	lseg = NFS_SERVER(ino)->pnfs_curr_ld->alloc_lseg(lo, res, lgp->gfp_flags);
···
 	lseg->pls_range = res->range;
 
 	spin_lock(&ino->i_lock);
-	if (test_bit(NFS_LAYOUT_BULK_RECALL, &lo->plh_flags)) {
-		dprintk("%s forget reply due to recall\n", __func__);
-		goto out_forget_reply;
-	}
-
-	if (pnfs_layoutgets_blocked(lo, &lgp->args.range, 1)) {
+	if (pnfs_layoutgets_blocked(lo)) {
 		dprintk("%s forget reply due to state\n", __func__);
 		goto out_forget_reply;
 	}
···
 	clear_bit(NFS_LAYOUT_INVALID_STID, &lo->plh_flags);
 
 	pnfs_get_lseg(lseg);
-	pnfs_layout_insert_lseg(lo, lseg);
+	pnfs_layout_insert_lseg(lo, lseg, &free_me);
 
-	if (res->return_on_close) {
+	if (res->return_on_close)
 		set_bit(NFS_LSEG_ROC, &lseg->pls_flags);
-		set_bit(NFS_LAYOUT_ROC, &lo->plh_flags);
-	}
 
 	spin_unlock(&ino->i_lock);
 	pnfs_free_lseg_list(&free_me);
···
 				lseg->pls_range.length);
 			set_bit(NFS_LSEG_LAYOUTRETURN, &lseg->pls_flags);
 			mark_lseg_invalid(lseg, tmp_list);
+			set_bit(NFS_LAYOUT_RETURN_BEFORE_CLOSE,
+					&lo->plh_flags);
 		}
 }
 
···
 
 #if IS_ENABLED(CONFIG_NFS_V4_2)
 int
-pnfs_report_layoutstat(struct inode *inode)
+pnfs_report_layoutstat(struct inode *inode, gfp_t gfp_flags)
 {
 	struct pnfs_layoutdriver_type *ld = NFS_SERVER(inode)->pnfs_curr_ld;
 	struct nfs_server *server = NFS_SERVER(inode);
···
 	pnfs_get_layout_hdr(hdr);
 	spin_unlock(&inode->i_lock);
 
-	data = kzalloc(sizeof(*data), GFP_KERNEL);
+	data = kzalloc(sizeof(*data), gfp_flags);
 	if (!data) {
 		status = -ENOMEM;
 		goto out_put;
···
 }
 EXPORT_SYMBOL_GPL(pnfs_report_layoutstat);
 #endif
+
+unsigned int layoutstats_timer;
+module_param(layoutstats_timer, uint, 0644);
+EXPORT_SYMBOL_GPL(layoutstats_timer);

fs/nfs/pnfs.h

···
 	NFS_LAYOUT_RO_FAILED = 0,	/* get ro layout failed stop trying */
 	NFS_LAYOUT_RW_FAILED,		/* get rw layout failed stop trying */
 	NFS_LAYOUT_BULK_RECALL,		/* bulk recall affecting layout */
-	NFS_LAYOUT_ROC,			/* some lseg had roc bit set */
 	NFS_LAYOUT_RETURN,		/* Return this layout ASAP */
 	NFS_LAYOUT_RETURN_BEFORE_CLOSE,	/* Return this layout before close */
 	NFS_LAYOUT_INVALID_STID,	/* layout stateid id is invalid */
···
 
 	struct pnfs_layout_segment * (*alloc_lseg) (struct pnfs_layout_hdr *layoutid, struct nfs4_layoutget_res *lgr, gfp_t gfp_flags);
 	void (*free_lseg) (struct pnfs_layout_segment *lseg);
+	void (*add_lseg) (struct pnfs_layout_hdr *layoutid,
+			struct pnfs_layout_segment *lseg,
+			struct list_head *free_me);
 
 	void (*return_range) (struct pnfs_layout_hdr *lo,
 			      struct pnfs_layout_range *range);
···
 
 struct pnfs_layout_hdr {
 	atomic_t		plh_refcount;
+	atomic_t		plh_outstanding; /* number of RPCs out */
 	struct list_head	plh_layouts;   /* other client layouts */
 	struct list_head	plh_bulk_destroy;
 	struct list_head	plh_segs;      /* layout segments list */
-	nfs4_stateid		plh_stateid;
-	atomic_t		plh_outstanding; /* number of RPCs out */
 	unsigned long		plh_block_lgets; /* block LAYOUTGET if >0 */
-	u32			plh_barrier; /* ignore lower seqids */
 	unsigned long		plh_retry_timestamp;
 	unsigned long		plh_flags;
+	nfs4_stateid		plh_stateid;
+	u32			plh_barrier; /* ignore lower seqids */
 	enum pnfs_iomode	plh_return_iomode;
 	loff_t			plh_lwb; /* last write byte for layoutcommit */
 	struct rpc_cred		*plh_lc_cred; /* layoutcommit cred */
···
 bool pnfs_roc(struct inode *ino);
 void pnfs_roc_release(struct inode *ino);
 void pnfs_roc_set_barrier(struct inode *ino, u32 barrier);
-bool pnfs_roc_drain(struct inode *ino, u32 *barrier, struct rpc_task *task);
+void pnfs_roc_get_barrier(struct inode *ino, u32 *barrier);
 void pnfs_set_layoutcommit(struct inode *, struct pnfs_layout_segment *, loff_t);
 void pnfs_cleanup_layoutcommit(struct nfs4_layoutcommit_data *data);
 int pnfs_layoutcommit_inode(struct inode *inode, bool sync);
···
 					       enum pnfs_iomode iomode,
 					       gfp_t gfp_flags);
 void pnfs_clear_layoutreturn_waitbit(struct pnfs_layout_hdr *lo);
+
+void pnfs_generic_layout_insert_lseg(struct pnfs_layout_hdr *lo,
+		   struct pnfs_layout_segment *lseg,
+		   bool (*is_after)(const struct pnfs_layout_range *lseg_range,
+			   const struct pnfs_layout_range *old),
+		   bool (*do_merge)(struct pnfs_layout_segment *lseg,
+			   struct pnfs_layout_segment *old),
+		   struct list_head *free_me);
 
 void nfs4_deviceid_mark_client_invalid(struct nfs_client *clp);
 int pnfs_read_done_resend_to_mds(struct nfs_pgio_header *);
···
 					nfss->pnfs_curr_ld->id == src->l_type);
 }
 
+static inline u64
+pnfs_calc_offset_end(u64 offset, u64 len)
+{
+	if (len == NFS4_MAX_UINT64 || len >= NFS4_MAX_UINT64 - offset)
+		return NFS4_MAX_UINT64;
+	return offset + len - 1;
+}
+
+static inline u64
+pnfs_calc_offset_length(u64 offset, u64 end)
+{
+	if (end == NFS4_MAX_UINT64 || end <= offset)
+		return NFS4_MAX_UINT64;
+	return 1 + end - offset;
+}
+
+extern unsigned int layoutstats_timer;
+
 #ifdef NFS_DEBUG
 void nfs4_print_deviceid(const struct nfs4_deviceid *dev_id);
 #else
 static inline void nfs4_print_deviceid(const struct nfs4_deviceid *dev_id)
 {
 }
+
 #endif /* NFS_DEBUG */
 #else  /* CONFIG_NFS_V4_1 */
 
···
 {
 }
 
-static inline bool
-pnfs_roc_drain(struct inode *ino, u32 *barrier, struct rpc_task *task)
+static inline void
+pnfs_roc_get_barrier(struct inode *ino, u32 *barrier)
 {
-	return false;
 }
 
 static inline void set_pnfs_layoutdriver(struct nfs_server *s,
···
 #endif /* CONFIG_NFS_V4_1 */
 
 #if IS_ENABLED(CONFIG_NFS_V4_2)
-int pnfs_report_layoutstat(struct inode *inode);
+int pnfs_report_layoutstat(struct inode *inode, gfp_t gfp_flags);
 #else
 static inline int
-pnfs_report_layoutstat(struct inode *inode)
+pnfs_report_layoutstat(struct inode *inode, gfp_t gfp_flags)
 {
 	return 0;
 }

fs/nfs/pnfs_nfs.c

···
 	if (ret) {
 		cinfo->ds->nwritten -= ret;
 		cinfo->ds->ncommitting += ret;
-		bucket->clseg = bucket->wlseg;
-		if (list_empty(src))
+		if (bucket->clseg == NULL)
+			bucket->clseg = pnfs_get_lseg(bucket->wlseg);
+		if (list_empty(src)) {
+			pnfs_put_lseg_locked(bucket->wlseg);
 			bucket->wlseg = NULL;
-		else
-			pnfs_get_lseg(bucket->clseg);
+		}
 	}
 	return ret;
 }
···
 	struct pnfs_ds_commit_info *fl_cinfo = cinfo->ds;
 	struct pnfs_commit_bucket *bucket;
 	struct pnfs_layout_segment *freeme;
+	LIST_HEAD(pages);
 	int i;
 
+	spin_lock(cinfo->lock);
 	for (i = idx; i < fl_cinfo->nbuckets; i++) {
 		bucket = &fl_cinfo->buckets[i];
 		if (list_empty(&bucket->committing))
 			continue;
-		nfs_retry_commit(&bucket->committing, bucket->clseg, cinfo, i);
-		spin_lock(cinfo->lock);
 		freeme = bucket->clseg;
 		bucket->clseg = NULL;
+		list_splice_init(&bucket->committing, &pages);
 		spin_unlock(cinfo->lock);
+		nfs_retry_commit(&pages, freeme, cinfo, i);
 		pnfs_put_lseg(freeme);
+		spin_lock(cinfo->lock);
 	}
+	spin_unlock(cinfo->lock);
 }
 
 static unsigned int
···
 		if (!data)
 			break;
 		data->ds_commit_index = i;
-		spin_lock(cinfo->lock);
-		data->lseg = bucket->clseg;
-		bucket->clseg = NULL;
-		spin_unlock(cinfo->lock);
 		list_add(&data->pages, list);
 		nreq++;
 	}
···
 	/* Clean up on error */
 	pnfs_generic_retry_commit(cinfo, i);
 	return nreq;
+}
+
+static inline
+void pnfs_fetch_commit_bucket_list(struct list_head *pages,
+		struct nfs_commit_data *data,
+		struct nfs_commit_info *cinfo)
+{
+	struct pnfs_commit_bucket *bucket;
+
+	bucket = &cinfo->ds->buckets[data->ds_commit_index];
+	spin_lock(cinfo->lock);
+	list_splice_init(&bucket->committing, pages);
+	data->lseg = bucket->clseg;
+	bucket->clseg = NULL;
+	spin_unlock(cinfo->lock);
+
 }
 
 /* This follows nfs_commit_list pretty closely */
···
 	if (!list_empty(mds_pages)) {
 		data = nfs_commitdata_alloc();
 		if (data != NULL) {
-			data->lseg = NULL;
+			data->ds_commit_index = -1;
 			list_add(&data->pages, &list);
 			nreq++;
 		} else {
···
 
 	list_for_each_entry_safe(data, tmp, &list, pages) {
 		list_del_init(&data->pages);
-		if (!data->lseg) {
+		if (data->ds_commit_index < 0) {
 			nfs_init_commit(data, mds_pages, NULL, cinfo);
 			nfs_initiate_commit(NFS_CLIENT(inode), data,
 					    NFS_PROTO(data->inode),
 					    data->mds_ops, how, 0);
 		} else {
-			struct pnfs_commit_bucket *buckets;
+			LIST_HEAD(pages);
 
-			buckets = cinfo->ds->buckets;
-			nfs_init_commit(data,
-					&buckets[data->ds_commit_index].committing,
-					data->lseg,
-					cinfo);
+			pnfs_fetch_commit_bucket_list(&pages, data, cinfo);
+			nfs_init_commit(data, &pages, data->lseg, cinfo);
 			initiate_commit(data, how);
 		}
 	}
···
 	return false;
 }
 
+/*
+ * Checks if 'dsaddrs1' contains a subset of 'dsaddrs2'. If it does,
+ * declare a match.
+ */
 static bool
 _same_data_server_addrs_locked(const struct list_head *dsaddrs1,
 			       const struct list_head *dsaddrs2)
 {
 	struct nfs4_pnfs_ds_addr *da1, *da2;
+	struct sockaddr *sa1, *sa2;
+	bool match = false;
 
-	/* step through both lists, comparing as we go */
-	for (da1 = list_first_entry(dsaddrs1, typeof(*da1), da_node),
-	     da2 = list_first_entry(dsaddrs2, typeof(*da2), da_node);
-	     da1 != NULL && da2 != NULL;
-	     da1 = list_entry(da1->da_node.next, typeof(*da1), da_node),
-	     da2 = list_entry(da2->da_node.next, typeof(*da2), da_node)) {
-		if (!same_sockaddr((struct sockaddr *)&da1->da_addr,
-				   (struct sockaddr *)&da2->da_addr))
-			return false;
+	list_for_each_entry(da1, dsaddrs1, da_node) {
+		sa1 = (struct sockaddr *)&da1->da_addr;
+		match = false;
+		list_for_each_entry(da2, dsaddrs2, da_node) {
+			sa2 = (struct sockaddr *)&da2->da_addr;
+			match = same_sockaddr(sa1, sa2);
+			if (match)
+				break;
+		}
+		if (!match)
+			break;
 	}
-	if (da1 == NULL && da2 == NULL)
-		return true;
-
-	return false;
+	return match;
 }
 
 /*
···
 	}
 	set_bit(PG_COMMIT_TO_DS, &req->wb_flags);
 	cinfo->ds->nwritten++;
-	spin_unlock(cinfo->lock);
 
-	nfs_request_add_commit_list(req, list, cinfo);
+	nfs_request_add_commit_list_locked(req, list, cinfo);
+	spin_unlock(cinfo->lock);
+	nfs_mark_page_unstable(req->wb_page, cinfo);
 }
 EXPORT_SYMBOL_GPL(pnfs_layout_mark_request_commit);
 

fs/nfs/super.c

···
 	ret = nfs_register_sysctl();
 	if (ret < 0)
 		goto error_2;
-	register_shrinker(&acl_shrinker);
+	ret = register_shrinker(&acl_shrinker);
+	if (ret < 0)
+		goto error_3;
 	return 0;
-
+error_3:
+	nfs_unregister_sysctl();
 error_2:
 	unregister_nfs4_fs();
 error_1:

fs/nfs/write.c

···
 }
 
 /**
+ * nfs_request_add_commit_list_locked - add request to a commit list
+ * @req: pointer to a struct nfs_page
+ * @dst: commit list head
+ * @cinfo: holds list lock and accounting info
+ *
+ * This sets the PG_CLEAN bit, updates the cinfo count of
+ * number of outstanding requests requiring a commit as well as
+ * the MM page stats.
+ *
+ * The caller must hold the cinfo->lock, and the nfs_page lock.
+ */
+void
+nfs_request_add_commit_list_locked(struct nfs_page *req, struct list_head *dst,
+			    struct nfs_commit_info *cinfo)
+{
+	set_bit(PG_CLEAN, &req->wb_flags);
+	nfs_list_add_request(req, dst);
+	cinfo->mds->ncommit++;
+}
+EXPORT_SYMBOL_GPL(nfs_request_add_commit_list_locked);
+
+/**
  * nfs_request_add_commit_list - add request to a commit list
  * @req: pointer to a struct nfs_page
  * @dst: commit list head
···
 nfs_request_add_commit_list(struct nfs_page *req, struct list_head *dst,
 			    struct nfs_commit_info *cinfo)
 {
-	set_bit(PG_CLEAN, &(req)->wb_flags);
 	spin_lock(cinfo->lock);
-	nfs_list_add_request(req, dst);
-	cinfo->mds->ncommit++;
+	nfs_request_add_commit_list_locked(req, dst, cinfo);
 	spin_unlock(cinfo->lock);
-	if (!cinfo->dreq)
-		nfs_mark_page_unstable(req->wb_page);
+	nfs_mark_page_unstable(req->wb_page, cinfo);
 }
 EXPORT_SYMBOL_GPL(nfs_request_add_commit_list);
 
···
 	return res;
 }
 
-static int nfs_commit_unstable_pages(struct inode *inode, struct writeback_control *wbc)
+int nfs_write_inode(struct inode *inode, struct writeback_control *wbc)
 {
 	struct nfs_inode *nfsi = NFS_I(inode);
 	int flags = FLUSH_SYNC;
···
 out_mark_dirty:
 	__mark_inode_dirty(inode, I_DIRTY_DATASYNC);
 	return ret;
-}
-
-int nfs_write_inode(struct inode *inode, struct writeback_control *wbc)
-{
-	return nfs_commit_unstable_pages(inode, wbc);
 }
 EXPORT_SYMBOL_GPL(nfs_write_inode);
 

fs/nfsd/blocklayoutxdr.c

···
 	}
 
 	nr_iomaps = be32_to_cpup(p++);
-	expected = sizeof(__be32) + nr_iomaps * NFS4_BLOCK_EXTENT_SIZE;
+	expected = sizeof(__be32) + nr_iomaps * PNFS_BLOCK_EXTENT_SIZE;
 	if (len != expected) {
 		dprintk("%s: extent array size mismatch: %u/%u\n",
 			__func__, len, expected);

fs/nfsd/blocklayoutxdr.h

···
 struct iomap;
 struct xdr_stream;
 
-enum pnfs_block_extent_state {
-	PNFS_BLOCK_READWRITE_DATA	= 0,
-	PNFS_BLOCK_READ_DATA		= 1,
-	PNFS_BLOCK_INVALID_DATA		= 2,
-	PNFS_BLOCK_NONE_DATA		= 3,
-};
-
 struct pnfs_block_extent {
 	struct nfsd4_deviceid		vol_id;
 	u64				foff;
 	u64				len;
 	u64				soff;
 	enum pnfs_block_extent_state	es;
-};
-#define NFS4_BLOCK_EXTENT_SIZE		44
-
-enum pnfs_block_volume_type {
-	PNFS_BLOCK_VOLUME_SIMPLE	= 0,
-	PNFS_BLOCK_VOLUME_SLICE		= 1,
-	PNFS_BLOCK_VOLUME_CONCAT	= 2,
-	PNFS_BLOCK_VOLUME_STRIPE	= 3,
 };
 
 /*

include/linux/nfs4.h

···
 	NOTIFY_DEVICEID4_DELETE = 1 << 2,
 };
 
+enum pnfs_block_volume_type {
+	PNFS_BLOCK_VOLUME_SIMPLE	= 0,
+	PNFS_BLOCK_VOLUME_SLICE		= 1,
+	PNFS_BLOCK_VOLUME_CONCAT	= 2,
+	PNFS_BLOCK_VOLUME_STRIPE	= 3,
+};
+
+enum pnfs_block_extent_state {
+	PNFS_BLOCK_READWRITE_DATA	= 0,
+	PNFS_BLOCK_READ_DATA		= 1,
+	PNFS_BLOCK_INVALID_DATA		= 2,
+	PNFS_BLOCK_NONE_DATA		= 3,
+};
+
+/* on the wire size of a block layout extent */
+#define PNFS_BLOCK_EXTENT_SIZE \
+	(7 * sizeof(__be32) + NFS4_DEVICEID4_SIZE)
+
 #define NFL4_UFLG_MASK			0x0000003F
 #define NFL4_UFLG_DENSE			0x00000001
 #define NFL4_UFLG_COMMIT_THRU_MDS	0x00000002

include/linux/nfs_fs.h

···
 extern void nfs_access_set_mask(struct nfs_access_entry *, u32);
 extern int nfs_permission(struct inode *, int);
 extern int nfs_open(struct inode *, struct file *);
-extern int nfs_release(struct inode *, struct file *);
 extern int nfs_attribute_timeout(struct inode *inode);
 extern int nfs_attribute_cache_expired(struct inode *inode);
 extern int nfs_revalidate_inode(struct nfs_server *server, struct inode *inode);
···
 extern struct nfs_open_context *alloc_nfs_open_context(struct dentry *dentry, fmode_t f_mode);
 extern void nfs_inode_attach_open_context(struct nfs_open_context *ctx);
 extern void nfs_file_set_open_context(struct file *filp, struct nfs_open_context *ctx);
+extern void nfs_file_clear_open_context(struct file *flip);
 extern struct nfs_lock_context *nfs_get_lock_context(struct nfs_open_context *ctx);
 extern void nfs_put_lock_context(struct nfs_lock_context *l_ctx);
 extern u64 nfs_compat_user_ino64(u64 fileid);

include/linux/nfs_fs_sb.h

···
 						   set of attributes supported
 						   on this filesystem excluding
 						   the label support bit. */
+	u32			exclcreat_bitmask[3];
+						/* V4 bitmask representing the
+						   set of attributes supported
+						   on this filesystem for the
+						   exclusive create. */
 	u32			cache_consistency_bitmask[3];
 						/* V4 bitmask representing the subset
 						   of change attribute, size, ctime

include/linux/nfs_xdr.h

···
 	struct stateowner_id	id;
 	union {
 		struct {
-			struct iattr *  attrs;    /* UNCHECKED, GUARDED */
+			struct iattr *  attrs;    /* UNCHECKED, GUARDED, EXCLUSIVE4_1 */
 			nfs4_verifier   verifier; /* EXCLUSIVE */
 		};
 		nfs4_stateid	delegation;		/* CLAIM_DELEGATE_CUR */
···
 	const struct nfs_server *server;	 /* Needed for ID mapping */
 	const u32 *		bitmask;
 	const u32 *		open_bitmap;
-	__u32			claim;
+	enum open_claim_type4	claim;
 	enum createmode4	createmode;
 	const struct nfs4_label *label;
 };
···
 	const struct nfs_server *server;
 	fmode_t			delegation_type;
 	nfs4_stateid		delegation;
+	unsigned long		pagemod_limit;
 	__u32			do_recall;
-	__u64			maxsize;
 	__u32			attrset[NFS4_BITMAP_SIZE];
 	struct nfs4_string	*owner;
 	struct nfs4_string	*group_owner;
···
 struct nfs4_server_caps_arg {
 	struct nfs4_sequence_args	seq_args;
 	struct nfs_fh		       *fhandle;
+	const u32 *			bitmask;
 };
 
 struct nfs4_server_caps_res {
 	struct nfs4_sequence_res	seq_res;
 	u32				attr_bitmask[3];
+	u32				exclcreat_bitmask[3];
 	u32				acl_bitmask;
 	u32				has_links;
 	u32				has_symlinks;

include/linux/sunrpc/addr.h

···
 #define IPV6_SCOPE_DELIMITER		'%'
 #define IPV6_SCOPE_ID_LEN		sizeof("%nnnnnnnnnn")
 
-static inline bool __rpc_cmp_addr4(const struct sockaddr *sap1,
-				   const struct sockaddr *sap2)
+static inline bool rpc_cmp_addr4(const struct sockaddr *sap1,
+				 const struct sockaddr *sap2)
 {
 	const struct sockaddr_in *sin1 = (const struct sockaddr_in *)sap1;
 	const struct sockaddr_in *sin2 = (const struct sockaddr_in *)sap2;
···
 }
 
 #if IS_ENABLED(CONFIG_IPV6)
-static inline bool __rpc_cmp_addr6(const struct sockaddr *sap1,
-				   const struct sockaddr *sap2)
+static inline bool rpc_cmp_addr6(const struct sockaddr *sap1,
+				 const struct sockaddr *sap2)
 {
 	const struct sockaddr_in6 *sin1 = (const struct sockaddr_in6 *)sap1;
 	const struct sockaddr_in6 *sin2 = (const struct sockaddr_in6 *)sap2;
···
 	return true;
 }
 #else	/* !(IS_ENABLED(CONFIG_IPV6) */
-static inline bool __rpc_cmp_addr6(const struct sockaddr *sap1,
+static inline bool rpc_cmp_addr6(const struct sockaddr *sap1,
 				   const struct sockaddr *sap2)
 {
 	return false;
···
 	if (sap1->sa_family == sap2->sa_family) {
 		switch (sap1->sa_family) {
 		case AF_INET:
-			return __rpc_cmp_addr4(sap1, sap2);
+			return rpc_cmp_addr4(sap1, sap2);
 		case AF_INET6:
-			return __rpc_cmp_addr6(sap1, sap2);
+			return rpc_cmp_addr6(sap1, sap2);
 		}
 	}
 	return false;
+}
+
+/**
+ * rpc_cmp_addr_port - compare the address and port number of two sockaddrs.
+ * @sap1: first sockaddr
+ * @sap2: second sockaddr
+ */
+static inline bool rpc_cmp_addr_port(const struct sockaddr *sap1,
+				     const struct sockaddr *sap2)
+{
+	if (!rpc_cmp_addr(sap1, sap2))
+		return false;
+	return rpc_get_port(sap1) == rpc_get_port(sap2);
 }
 
 /**

include/linux/sunrpc/auth.h

···
 #include <linux/atomic.h>
 #include <linux/rcupdate.h>
 #include <linux/uidgid.h>
+#include <linux/utsname.h>
 
-/* size of the nodename buffer */
-#define UNX_MAXNODENAME	32
+/*
+ * Size of the nodename buffer. RFC1831 specifies a hard limit of 255 bytes,
+ * but Linux hostnames are actually limited to __NEW_UTS_LEN bytes.
+ */
+#define UNX_MAXNODENAME	__NEW_UTS_LEN
 
 struct rpcsec_gss_info;
 

include/linux/sunrpc/xprtrdma.h

···
  * a single chunk type per message is supported currently.
  */
 #define RPCRDMA_MIN_SLOT_TABLE	(2U)
-#define RPCRDMA_DEF_SLOT_TABLE	(32U)
+#define RPCRDMA_DEF_SLOT_TABLE	(128U)
 #define RPCRDMA_MAX_SLOT_TABLE	(256U)
 
 #define RPCRDMA_DEF_INLINE  (1024)	/* default inline max */

include/rdma/ib_verbs.h

···
 }
 
 /**
- * ib_reg_phys_mr - Prepares a virtually addressed memory region for use
- *   by an HCA.
- * @pd: The protection domain associated assigned to the registered region.
- * @phys_buf_array: Specifies a list of physical buffers to use in the
- *   memory region.
- * @num_phys_buf: Specifies the size of the phys_buf_array.
- * @mr_access_flags: Specifies the memory access rights.
- * @iova_start: The offset of the region's starting I/O virtual address.
- */
-struct ib_mr *ib_reg_phys_mr(struct ib_pd *pd,
-			     struct ib_phys_buf *phys_buf_array,
-			     int num_phys_buf,
-			     int mr_access_flags,
-			     u64 *iova_start);
-
-/**
- * ib_rereg_phys_mr - Modifies the attributes of an existing memory region.
- *   Conceptually, this call performs the functions deregister memory region
- *   followed by register physical memory region.  Where possible,
- *   resources are reused instead of deallocated and reallocated.
- * @mr: The memory region to modify.
- * @mr_rereg_mask: A bit-mask used to indicate which of the following
- *   properties of the memory region are being modified.
- * @pd: If %IB_MR_REREG_PD is set in mr_rereg_mask, this field specifies
- *   the new protection domain to associated with the memory region,
- *   otherwise, this parameter is ignored.
- * @phys_buf_array: If %IB_MR_REREG_TRANS is set in mr_rereg_mask, this
- *   field specifies a list of physical buffers to use in the new
- *   translation, otherwise, this parameter is ignored.
- * @num_phys_buf: If %IB_MR_REREG_TRANS is set in mr_rereg_mask, this
- *   field specifies the size of the phys_buf_array, otherwise, this
- *   parameter is ignored.
- * @mr_access_flags: If %IB_MR_REREG_ACCESS is set in mr_rereg_mask, this
- *   field specifies the new memory access rights, otherwise, this
- *   parameter is ignored.
- * @iova_start: The offset of the region's starting I/O virtual address.
- */
-int ib_rereg_phys_mr(struct ib_mr *mr,
-		     int mr_rereg_mask,
-		     struct ib_pd *pd,
-		     struct ib_phys_buf *phys_buf_array,
-		     int num_phys_buf,
-		     int mr_access_flags,
-		     u64 *iova_start);
-
-/**
  * ib_query_mr - Retrieves information about a specific memory region.
  * @mr: The memory region to retrieve information about.
  * @mr_attr: The attributes of the specified memory region.

include/uapi/linux/nfs4.h

···
 
 #include <linux/types.h>
 
-#define NFS4_BITMAP_SIZE	2
+#define NFS4_BITMAP_SIZE	3
 #define NFS4_VERIFIER_SIZE	8
 #define NFS4_STATEID_SEQID_SIZE 4
 #define NFS4_STATEID_OTHER_SIZE 12

net/sunrpc/auth_unix.c

···
 };
 #define uc_uid			uc_base.cr_uid
 
-#define UNX_WRITESLACK		(21 + (UNX_MAXNODENAME >> 2))
+#define UNX_WRITESLACK		(21 + XDR_QUADLEN(UNX_MAXNODENAME))
 
 #if IS_ENABLED(CONFIG_SUNRPC_DEBUG)
 # define RPCDBG_FACILITY	RPCDBG_AUTH

net/sunrpc/xprtrdma/fmr_ops.c

···
 fmr_op_open(struct rpcrdma_ia *ia, struct rpcrdma_ep *ep,
 	    struct rpcrdma_create_data_internal *cdata)
 {
+	struct ib_device_attr *devattr = &ia->ri_devattr;
+	struct ib_mr *mr;
+
+	/* Obtain an lkey to use for the regbufs, which are
+	 * protected from remote access.
+	 */
+	if (devattr->device_cap_flags & IB_DEVICE_LOCAL_DMA_LKEY) {
+		ia->ri_dma_lkey = ia->ri_device->local_dma_lkey;
+	} else {
+		mr = ib_get_dma_mr(ia->ri_pd, IB_ACCESS_LOCAL_WRITE);
+		if (IS_ERR(mr)) {
+			pr_err("%s: ib_get_dma_mr for failed with %lX\n",
+			       __func__, PTR_ERR(mr));
+			return -ENOMEM;
+		}
+		ia->ri_dma_lkey = ia->ri_dma_mr->lkey;
+		ia->ri_dma_mr = mr;
+	}
+
 	return 0;
 }
 

net/sunrpc/xprtrdma/frwr_ops.c

···
 	struct ib_device_attr *devattr = &ia->ri_devattr;
 	int depth, delta;
 
+	/* Obtain an lkey to use for the regbufs, which are
+	 * protected from remote access.
+	 */
+	ia->ri_dma_lkey = ia->ri_device->local_dma_lkey;
+
 	ia->ri_max_frmr_depth =
 			min_t(unsigned int, RPCRDMA_MAX_DATA_SEGS,
 			      devattr->max_fast_reg_page_list_len);

net/sunrpc/xprtrdma/physical_ops.c

···
 physical_op_open(struct rpcrdma_ia *ia, struct rpcrdma_ep *ep,
 		 struct rpcrdma_create_data_internal *cdata)
 {
+	struct ib_device_attr *devattr = &ia->ri_devattr;
+	struct ib_mr *mr;
+
+	/* Obtain an rkey to use for RPC data payloads.
+	 */
+	mr = ib_get_dma_mr(ia->ri_pd,
+			   IB_ACCESS_LOCAL_WRITE |
+			   IB_ACCESS_REMOTE_WRITE |
+			   IB_ACCESS_REMOTE_READ);
+	if (IS_ERR(mr)) {
+		pr_err("%s: ib_get_dma_mr for failed with %lX\n",
+		       __func__, PTR_ERR(mr));
+		return -ENOMEM;
+	}
+	ia->ri_dma_mr = mr;
+
+	/* Obtain an lkey to use for regbufs.
+	 */
+	if (devattr->device_cap_flags & IB_DEVICE_LOCAL_DMA_LKEY)
+		ia->ri_dma_lkey = ia->ri_device->local_dma_lkey;
+	else
+		ia->ri_dma_lkey = ia->ri_dma_mr->lkey;
+
 	return 0;
 }
 
···
 	struct rpcrdma_ia *ia = &r_xprt->rx_ia;
 
 	rpcrdma_map_one(ia->ri_device, seg, rpcrdma_data_dir(writing));
-	seg->mr_rkey = ia->ri_bind_mem->rkey;
+	seg->mr_rkey = ia->ri_dma_mr->rkey;
 	seg->mr_base = seg->mr_dma;
 	seg->mr_nsegs = 1;
 	return 1;

net/sunrpc/xprtrdma/rpc_rdma.c

···
 };
 #endif
 
+/* The client can send a request inline as long as the RPCRDMA header
+ * plus the RPC call fit under the transport's inline limit. If the
+ * combined call message size exceeds that limit, the client must use
+ * the read chunk list for this operation.
+ */
+static bool rpcrdma_args_inline(struct rpc_rqst *rqst)
+{
+	unsigned int callsize = RPCRDMA_HDRLEN_MIN + rqst->rq_snd_buf.len;
+
+	return callsize <= RPCRDMA_INLINE_WRITE_THRESHOLD(rqst);
+}
+
+/* The client can't know how large the actual reply will be. Thus it
+ * plans for the largest possible reply for that particular ULP
+ * operation. If the maximum combined reply message size exceeds that
+ * limit, the client must provide a write list or a reply chunk for
+ * this request.
+ */
+static bool rpcrdma_results_inline(struct rpc_rqst *rqst)
+{
+	unsigned int repsize = RPCRDMA_HDRLEN_MIN + rqst->rq_rcv_buf.buflen;
+
+	return repsize <= RPCRDMA_INLINE_READ_THRESHOLD(rqst);
+}
+
+static int
+rpcrdma_tail_pullup(struct xdr_buf *buf)
+{
+	size_t tlen = buf->tail[0].iov_len;
+	size_t skip = tlen & 3;
+
+	/* Do not include the tail if it is only an XDR pad */
+	if (tlen < 4)
+		return 0;
+
+	/* xdr_write_pages() adds a pad at the beginning of the tail
+	 * if the content in "buf->pages" is unaligned. Force the
+	 * tail's actual content to land at the next XDR position
+	 * after the head instead.
+	 */
+	if (skip) {
+		unsigned char *src, *dst;
+		unsigned int count;
+
+		src = buf->tail[0].iov_base;
+		dst = buf->head[0].iov_base;
+		dst += buf->head[0].iov_len;
+
+		src += skip;
+		tlen -= skip;
+
+		dprintk("RPC:       %s: skip=%zu, memmove(%p, %p, %zu)\n",
+			__func__, skip, dst, src, tlen);
+
+		for (count = tlen; count; count--)
+			*dst++ = *src++;
+	}
+
+	return tlen;
+}
+
 /*
  * Chunk assembly from upper layer xdr_buf.
  *
···
 	/* Message overflows the seg array */
 	if (len && n == nsegs)
 		return -EIO;
+
+	/* When encoding the read list, the tail is always sent inline */
+	if (type == rpcrdma_readch)
+		return n;
 
 	if (xdrbuf->tail[0].iov_len) {
 		/* the rpcrdma protocol allows us to omit any trailing
···
  * pre-registered memory buffer for this request. For small amounts
  * of data, this is efficient. The cutoff value is tunable.
  */
-static int
-rpcrdma_inline_pullup(struct rpc_rqst *rqst, int pad)
+static void rpcrdma_inline_pullup(struct rpc_rqst *rqst)
 {
 	int i, npages, curlen;
 	int copy_len;
···
 	destp = rqst->rq_svec[0].iov_base;
 	curlen = rqst->rq_svec[0].iov_len;
 	destp += curlen;
-	/*
-	 * Do optional padding where it makes sense. Alignment of write
-	 * payload can help the server, if our setting is accurate.
-	 */
-	pad -= (curlen + 36/*sizeof(struct rpcrdma_msg_padded)*/);
-	if (pad < 0 || rqst->rq_slen - curlen < RPCRDMA_INLINE_PAD_THRESH)
-		pad = 0;	/* don't pad this request */
 
-	dprintk("RPC:       %s: pad %d destp 0x%p len %d hdrlen %d\n",
-		__func__, pad, destp, rqst->rq_slen, curlen);
+	dprintk("RPC:       %s: destp 0x%p len %d hdrlen %d\n",
+		__func__, destp, rqst->rq_slen, curlen);
 
 	copy_len = rqst->rq_snd_buf.page_len;
 
···
 		page_base = 0;
 	}
 	/* header now contains entire send message */
-	return pad;
 }
 
 /*
···
 	struct rpcrdma_xprt *r_xprt = rpcx_to_rdmax(xprt);
 	struct rpcrdma_req *req = rpcr_to_rdmar(rqst);
 	char *base;
-	size_t rpclen, padlen;
+	size_t rpclen;
 	ssize_t hdrlen;
 	enum rpcrdma_chunktype rtype, wtype;
 	struct rpcrdma_msg *headerp;
···
 	/*
 	 * Chunks needed for results?
 	 *
+	 * o Read ops return data as write chunk(s), header as inline.
 	 * o If the expected result is under the inline threshold, all ops
-	 *   return as inline (but see later).
+	 *   return as inline.
 	 * o Large non-read ops return as a single reply chunk.
-	 * o Large read ops return data as write chunk(s), header as inline.
-	 *
-	 * Note: the NFS code sending down multiple result segments implies
-	 * the op is one of read, readdir[plus], readlink or NFSv4 getacl.
 	 */
-
-	/*
-	 * This code can handle read chunks, write chunks OR reply
-	 * chunks -- only one type. If the request is too big to fit
-	 * inline, then we will choose read chunks. If the request is
-	 * a READ, then use write chunks to separate the file data
-	 * into pages; otherwise use reply chunks.
-	 */
-	if (rqst->rq_rcv_buf.buflen <= RPCRDMA_INLINE_READ_THRESHOLD(rqst))
-		wtype = rpcrdma_noch;
-	else if (rqst->rq_rcv_buf.page_len == 0)
-		wtype = rpcrdma_replych;
-	else if (rqst->rq_rcv_buf.flags & XDRBUF_READ)
+	if (rqst->rq_rcv_buf.flags & XDRBUF_READ)
 		wtype = rpcrdma_writech;
+	else if (rpcrdma_results_inline(rqst))
+		wtype = rpcrdma_noch;
 	else
 		wtype = rpcrdma_replych;
 
···
 	 *
 	 * o If the total request is under the inline threshold, all ops
 	 *   are sent as inline.
-	 * o Large non-write ops are sent with the entire message as a
-	 *   single read chunk (protocol 0-position special case).
 	 * o Large write ops transmit data as read chunk(s), header as
 	 *   inline.
+	 * o Large non-write ops are sent with the entire message as a
+	 *   single read chunk (protocol 0-position special case).
 	 *
-	 * Note: the NFS code sending down multiple argument segments
-	 * implies the op is a write.
-	 * TBD check NFSv4 setacl
+	 * This assumes that the upper layer does not present a request
+	 * that both has a data payload, and whose non-data arguments
+	 * by themselves are larger than the inline threshold.
 	 */
-	if (rqst->rq_snd_buf.len <= RPCRDMA_INLINE_WRITE_THRESHOLD(rqst))
+	if (rpcrdma_args_inline(rqst)) {
 		rtype = rpcrdma_noch;
-	else if (rqst->rq_snd_buf.page_len == 0)
-		rtype = rpcrdma_areadch;
-	else
+	} else if (rqst->rq_snd_buf.flags & XDRBUF_WRITE) {
 		rtype = rpcrdma_readch;
+	} else {
+		r_xprt->rx_stats.nomsg_call_count++;
+		headerp->rm_type = htonl(RDMA_NOMSG);
+		rtype = rpcrdma_areadch;
+		rpclen = 0;
+	}
 
 	/* The following simplification is not true forever */
 	if (rtype != rpcrdma_noch && wtype == rpcrdma_replych)
···
 	}
 
 	hdrlen = RPCRDMA_HDRLEN_MIN;
-	padlen = 0;
 
 	/*
 	 * Pull up any extra send data into the preregistered buffer.
···
 	 */
 	if (rtype == rpcrdma_noch) {
 
-		padlen = rpcrdma_inline_pullup(rqst,
-						RPCRDMA_INLINE_PAD_VALUE(rqst));
+		rpcrdma_inline_pullup(rqst);
 
-		if (padlen) {
-			headerp->rm_type = rdma_msgp;
-			headerp->rm_body.rm_padded.rm_align =
-				cpu_to_be32(RPCRDMA_INLINE_PAD_VALUE(rqst));
-			headerp->rm_body.rm_padded.rm_thresh =
-				cpu_to_be32(RPCRDMA_INLINE_PAD_THRESH);
-			headerp->rm_body.rm_padded.rm_pempty[0] = xdr_zero;
-			headerp->rm_body.rm_padded.rm_pempty[1] = xdr_zero;
-			headerp->rm_body.rm_padded.rm_pempty[2] = xdr_zero;
-			hdrlen += 2 * sizeof(u32); /* extra words in padhdr */
-			if (wtype != rpcrdma_noch) {
-				dprintk("RPC:       %s: invalid chunk list\n",
-					__func__);
-				return -EIO;
-			}
-		} else {
-			headerp->rm_body.rm_nochunks.rm_empty[0] = xdr_zero;
-			headerp->rm_body.rm_nochunks.rm_empty[1] = xdr_zero;
-			headerp->rm_body.rm_nochunks.rm_empty[2] = xdr_zero;
-			/* new length after pullup */
-			rpclen = rqst->rq_svec[0].iov_len;
-			/*
-			 * Currently we try to not actually use read inline.
-			 * Reply chunks have the desirable property that
-			 * they land, packed, directly in the target buffers
-			 * without headers, so they require no fixup. The
-			 * additional RDMA Write op sends the same amount
-			 * of data, streams on-the-wire and adds no overhead
-			 * on receive. Therefore, we request a reply chunk
-			 * for non-writes wherever feasible and efficient.
-			 */
-			if (wtype == rpcrdma_noch)
-				wtype = rpcrdma_replych;
-		}
-	}
-
+		headerp->rm_body.rm_nochunks.rm_empty[0] = xdr_zero;
+		headerp->rm_body.rm_nochunks.rm_empty[1] = xdr_zero;
+		headerp->rm_body.rm_nochunks.rm_empty[2] = xdr_zero;
+		/* new length after pullup */
+		rpclen = rqst->rq_svec[0].iov_len;
+	} else if (rtype == rpcrdma_readch)
+		rpclen += rpcrdma_tail_pullup(&rqst->rq_snd_buf);
 	if (rtype != rpcrdma_noch) {
 		hdrlen = rpcrdma_create_chunks(rqst, &rqst->rq_snd_buf,
 					       headerp, rtype);
···
 	if (hdrlen < 0)
 		return hdrlen;
 
-	dprintk("RPC:       %s: %s: hdrlen %zd rpclen %zd padlen %zd"
+	dprintk("RPC:       %s: %s: hdrlen %zd rpclen %zd"
 		" headerp 0x%p base 0x%p lkey 0x%x\n",
-		__func__, transfertypes[wtype], hdrlen, rpclen, padlen,
+		__func__, transfertypes[wtype], hdrlen, rpclen,
 		headerp, base, rdmab_lkey(req->rl_rdmabuf));
 
 	/*
···
 	req->rl_send_iov[0].length = hdrlen;
 	req->rl_send_iov[0].lkey = rdmab_lkey(req->rl_rdmabuf);
 
+	req->rl_niovs = 1;
+	if (rtype == rpcrdma_areadch)
+		return 0;
+
 	req->rl_send_iov[1].addr = rdmab_addr(req->rl_sendbuf);
 	req->rl_send_iov[1].length = rpclen;
 	req->rl_send_iov[1].lkey = rdmab_lkey(req->rl_sendbuf);
 
 	req->rl_niovs = 2;
-
-	if (padlen) {
-		struct rpcrdma_ep *ep = &r_xprt->rx_ep;
-
-		req->rl_send_iov[2].addr = rdmab_addr(ep->rep_padbuf);
-		req->rl_send_iov[2].length = padlen;
-		req->rl_send_iov[2].lkey = rdmab_lkey(ep->rep_padbuf);
-
-		req->rl_send_iov[3].addr = req->rl_send_iov[1].addr + rpclen;
-		req->rl_send_iov[3].length = rqst->rq_slen - rpclen;
-		req->rl_send_iov[3].lkey = rdmab_lkey(req->rl_sendbuf);
-
-		req->rl_niovs = 4;
-	}
-
 	return 0;
 }
 

net/sunrpc/xprtrdma/transport.c

···
 }
 
 static void
-xprt_rdma_format_addresses(struct rpc_xprt *xprt)
+xprt_rdma_format_addresses(struct rpc_xprt *xprt, struct sockaddr *sap)
 {
-	struct sockaddr *sap = (struct sockaddr *)
-					&rpcx_to_rdmad(xprt).addr;
 	char buf[128];
 
 	switch (sap->sa_family) {
···
 	struct rpc_xprt *xprt;
 	struct rpcrdma_xprt *new_xprt;
 	struct rpcrdma_ep *new_ep;
-	struct sockaddr_in *sin;
+	struct sockaddr *sap;
 	int rc;
 
 	if (args->addrlen > sizeof(xprt->addr)) {
···
 	 * Set up RDMA-specific connect data.
 	 */
 
-	/* Put server RDMA address in local cdata */
-	memcpy(&cdata.addr, args->dstaddr, args->addrlen);
+	sap = (struct sockaddr *)&cdata.addr;
+	memcpy(sap, args->dstaddr, args->addrlen);
 
 	/* Ensure xprt->addr holds valid server TCP (not RDMA)
 	 * address, for any side protocols which peek at it */
 	xprt->prot = IPPROTO_TCP;
 	xprt->addrlen = args->addrlen;
-	memcpy(&xprt->addr, &cdata.addr, xprt->addrlen);
+	memcpy(&xprt->addr, sap, xprt->addrlen);
 
-	sin = (struct sockaddr_in *)&cdata.addr;
-	if (ntohs(sin->sin_port) != 0)
+	if (rpc_get_port(sap))
 		xprt_set_bound(xprt);
 
-	dprintk("RPC:       %s: %pI4:%u\n",
-		__func__, &sin->sin_addr.s_addr, ntohs(sin->sin_port));
-
-	/* Set max requests */
 	cdata.max_requests = xprt->max_reqs;
 
-	/* Set some length limits */
 	cdata.rsize = RPCRDMA_MAX_SEGS * PAGE_SIZE; /* RDMA write max */
 	cdata.wsize = RPCRDMA_MAX_SEGS * PAGE_SIZE; /* RDMA read max */
 
···
 
 	new_xprt = rpcx_to_rdmax(xprt);
 
-	rc = rpcrdma_ia_open(new_xprt, (struct sockaddr *) &cdata.addr,
-				xprt_rdma_memreg_strategy);
+	rc = rpcrdma_ia_open(new_xprt, sap, xprt_rdma_memreg_strategy);
 	if (rc)
 		goto out1;
 
···
 	INIT_DELAYED_WORK(&new_xprt->rx_connect_worker,
 			  xprt_rdma_connect_worker);
 
-	xprt_rdma_format_addresses(xprt);
+	xprt_rdma_format_addresses(xprt, sap);
 	xprt->max_payload = new_xprt->rx_ia.ri_ops->ro_maxpages(new_xprt);
 	if (xprt->max_payload == 0)
 		goto out4;
···
 	if (!try_module_get(THIS_MODULE))
 		goto out4;
 
+	dprintk("RPC:       %s: %s:%s\n", __func__,
+		xprt->address_strings[RPC_DISPLAY_ADDR],
+		xprt->address_strings[RPC_DISPLAY_PORT]);
 	return xprt;
 
 out4:
···
 	if (xprt_connected(xprt))
 		idle_time = (long)(jiffies - xprt->last_used) / HZ;
 
-	seq_printf(seq,
-	  "\txprt:\trdma %u %lu %lu %lu %ld %lu %lu %lu %Lu %Lu "
-	  "%lu %lu %lu %Lu %Lu %Lu %Lu %lu %lu %lu\n",
-
-	   0,	/* need a local port? */
-	   xprt->stat.bind_count,
-	   xprt->stat.connect_count,
-	   xprt->stat.connect_time,
-	   idle_time,
-	   xprt->stat.sends,
-	   xprt->stat.recvs,
-	   xprt->stat.bad_xids,
-	   xprt->stat.req_u,
-	   xprt->stat.bklog_u,
-
-	   r_xprt->rx_stats.read_chunk_count,
-	   r_xprt->rx_stats.write_chunk_count,
-	   r_xprt->rx_stats.reply_chunk_count,
-	   r_xprt->rx_stats.total_rdma_request,
-	   r_xprt->rx_stats.total_rdma_reply,
-	   r_xprt->rx_stats.pullup_copy_count,
-	   r_xprt->rx_stats.fixup_copy_count,
-	   r_xprt->rx_stats.hardway_register_count,
-	   r_xprt->rx_stats.failed_marshal_count,
-	   r_xprt->rx_stats.bad_reply_count);
+	seq_puts(seq, "\txprt:\trdma ");
+	seq_printf(seq, "%u %lu %lu %lu %ld %lu %lu %lu %llu %llu ",
+		   0,	/* need a local port? */
+		   xprt->stat.bind_count,
+		   xprt->stat.connect_count,
+		   xprt->stat.connect_time,
+		   idle_time,
+		   xprt->stat.sends,
+		   xprt->stat.recvs,
+		   xprt->stat.bad_xids,
+		   xprt->stat.req_u,
+		   xprt->stat.bklog_u);
+	seq_printf(seq, "%lu %lu %lu %llu %llu %llu %llu %lu %lu %lu %lu\n",
+		   r_xprt->rx_stats.read_chunk_count,
+		   r_xprt->rx_stats.write_chunk_count,
+		   r_xprt->rx_stats.reply_chunk_count,
+		   r_xprt->rx_stats.total_rdma_request,
+		   r_xprt->rx_stats.total_rdma_reply,
+		   r_xprt->rx_stats.pullup_copy_count,
+		   r_xprt->rx_stats.fixup_copy_count,
+		   r_xprt->rx_stats.hardway_register_count,
+		   r_xprt->rx_stats.failed_marshal_count,
+		   r_xprt->rx_stats.bad_reply_count,
+		   r_xprt->rx_stats.nomsg_call_count);
 }
 
 static int

net/sunrpc/xprtrdma/verbs.c

···
 #include <linux/prefetch.h>
 #include <linux/sunrpc/addr.h>
 #include <asm/bitops.h>
+#include <linux/module.h> /* try_module_get()/module_put() */
 
 #include "xprt_rdma.h"
 
···
 	return 0;
 }
 
+static void rpcrdma_destroy_id(struct rdma_cm_id *id)
+{
+	if (id) {
+		module_put(id->device->owner);
+		rdma_destroy_id(id);
+	}
+}
+
 static struct rdma_cm_id *
 rpcrdma_create_id(struct rpcrdma_xprt *xprt,
 			struct rpcrdma_ia *ia, struct sockaddr *addr)
···
 	}
 	wait_for_completion_interruptible_timeout(&ia->ri_done,
 				msecs_to_jiffies(RDMA_RESOLVE_TIMEOUT) + 1);
+
+	/* FIXME:
+	 * Until xprtrdma supports DEVICE_REMOVAL, the provider must
+	 * be pinned while there are active NFS/RDMA mounts to prevent
+	 * hangs and crashes at umount time.
+	 */
+	if (!ia->ri_async_rc && !try_module_get(id->device->owner)) {
+		dprintk("RPC:       %s: Failed to get device module\n",
+			__func__);
+		ia->ri_async_rc = -ENODEV;
+	}
 	rc = ia->ri_async_rc;
 	if (rc)
 		goto out;
···
 	if (rc) {
 		dprintk("RPC:       %s: rdma_resolve_route() failed %i\n",
 			__func__, rc);
-		goto out;
+		goto put;
 	}
 	wait_for_completion_interruptible_timeout(&ia->ri_done,
 				msecs_to_jiffies(RDMA_RESOLVE_TIMEOUT) + 1);
 	rc = ia->ri_async_rc;
 	if (rc)
-		goto out;
+		goto put;
 
 	return id;
-
+put:
+	module_put(id->device->owner);
 out:
 	rdma_destroy_id(id);
 	return ERR_PTR(rc);
···
 int
 rpcrdma_ia_open(struct rpcrdma_xprt *xprt, struct sockaddr *addr, int memreg)
 {
-	int rc, mem_priv;
 	struct rpcrdma_ia *ia = &xprt->rx_ia;
 	struct ib_device_attr *devattr = &ia->ri_devattr;
+	int rc;
+
+	ia->ri_dma_mr = NULL;
 
 	ia->ri_id = rpcrdma_create_id(xprt, ia, addr);
 	if (IS_ERR(ia->ri_id)) {
···
 		goto out3;
 	}
 
-	if (devattr->device_cap_flags & IB_DEVICE_LOCAL_DMA_LKEY) {
-		ia->ri_have_dma_lkey = 1;
-		ia->ri_dma_lkey = ia->ri_device->local_dma_lkey;
-	}
-
 	if (memreg == RPCRDMA_FRMR) {
 		/* Requires both frmr reg and local dma lkey */
 		if (((devattr->device_cap_flags &
···
 		if (!ia->ri_device->alloc_fmr) {
 			dprintk("RPC:       %s: MTHCAFMR registration "
 				"not supported by HCA\n", __func__);
-			memreg = RPCRDMA_ALLPHYSICAL;
+			goto out3;
 		}
 	}
 
-	/*
-	 * Optionally obtain an underlying physical identity mapping in
-	 * order to do a memory window-based bind. This base registration
-	 * is protected from remote access - that is enabled only by binding
-	 * for the specific bytes targeted during each RPC operation, and
-	 * revoked after the corresponding completion similar to a storage
-	 * adapter.
-	 */
 	switch (memreg) {
 	case RPCRDMA_FRMR:
 		ia->ri_ops = &rpcrdma_frwr_memreg_ops;
 		break;
 	case RPCRDMA_ALLPHYSICAL:
 		ia->ri_ops = &rpcrdma_physical_memreg_ops;
-		mem_priv = IB_ACCESS_LOCAL_WRITE |
-				IB_ACCESS_REMOTE_WRITE |
-				IB_ACCESS_REMOTE_READ;
-		goto register_setup;
+		break;
 	case RPCRDMA_MTHCAFMR:
 		ia->ri_ops = &rpcrdma_fmr_memreg_ops;
-		if (ia->ri_have_dma_lkey)
-			break;
-		mem_priv = IB_ACCESS_LOCAL_WRITE;
-	register_setup:
-		ia->ri_bind_mem = ib_get_dma_mr(ia->ri_pd, mem_priv);
-		if (IS_ERR(ia->ri_bind_mem)) {
-			printk(KERN_ALERT "%s: ib_get_dma_mr for "
-				"phys register failed with %lX\n",
-				__func__, PTR_ERR(ia->ri_bind_mem));
-			rc = -ENOMEM;
-			goto out3;
-		}
 		break;
 	default:
 		printk(KERN_ERR "RPC: Unsupported memory "
···
 	ib_dealloc_pd(ia->ri_pd);
 	ia->ri_pd = NULL;
 out2:
-	rdma_destroy_id(ia->ri_id);
+	rpcrdma_destroy_id(ia->ri_id);
 	ia->ri_id = NULL;
 out1:
 	return rc;
···
 void
 rpcrdma_ia_close(struct rpcrdma_ia *ia)
 {
-	int rc;
-
 	dprintk("RPC:       %s: entering\n", __func__);
-	if (ia->ri_bind_mem != NULL) {
-		rc = ib_dereg_mr(ia->ri_bind_mem);
-		dprintk("RPC:       %s: ib_dereg_mr returned %i\n",
-			__func__, rc);
-	}
-
 	if (ia->ri_id != NULL && !IS_ERR(ia->ri_id)) {
 		if (ia->ri_id->qp)
 			rdma_destroy_qp(ia->ri_id);
-		rdma_destroy_id(ia->ri_id);
+		rpcrdma_destroy_id(ia->ri_id);
 		ia->ri_id = NULL;
 	}
 
···
 	struct ib_cq_init_attr cq_attr = {};
 	int rc, err;
 
+	if (devattr->max_sge < RPCRDMA_MAX_IOVS) {
+		dprintk("RPC:       %s: insufficient sge's available\n",
+			__func__);
+		return -ENOMEM;
+	}
+
 	/* check provider's send/recv wr limits */
 	if (cdata->max_requests > devattr->max_qp_wr)
 		cdata->max_requests = devattr->max_qp_wr;
···
 	if (rc)
 		return rc;
 	ep->rep_attr.cap.max_recv_wr = cdata->max_requests;
-	ep->rep_attr.cap.max_send_sge = (cdata->padding ? 4 : 2);
+	ep->rep_attr.cap.max_send_sge = RPCRDMA_MAX_IOVS;
 	ep->rep_attr.cap.max_recv_sge = 1;
 	ep->rep_attr.cap.max_inline_data = 0;
 	ep->rep_attr.sq_sig_type = IB_SIGNAL_REQ_WR;
 	ep->rep_attr.qp_type = IB_QPT_RC;
 	ep->rep_attr.port_num = ~0;
-
-	if (cdata->padding) {
-		ep->rep_padbuf = rpcrdma_alloc_regbuf(ia, cdata->padding,
-						      GFP_KERNEL);
-		if (IS_ERR(ep->rep_padbuf))
-			return PTR_ERR(ep->rep_padbuf);
-	} else
-		ep->rep_padbuf = NULL;
 
 	dprintk("RPC:       %s: requested max: dtos: send %d recv %d; "
 		"iovs: send %d recv %d\n",
···
 		dprintk("RPC:       %s: ib_destroy_cq returned %i\n",
 			__func__, err);
 out1:
-	rpcrdma_free_regbuf(ia, ep->rep_padbuf);
+	if (ia->ri_dma_mr)
+		ib_dereg_mr(ia->ri_dma_mr);
 	return rc;
 }
 
···
 		ia->ri_id->qp = NULL;
 	}
 
-	rpcrdma_free_regbuf(ia, ep->rep_padbuf);
-
 	rpcrdma_clean_cq(ep->rep_attr.recv_cq);
 	rc = ib_destroy_cq(ep->rep_attr.recv_cq);
 	if (rc)
···
 	if (rc)
 		dprintk("RPC:       %s: ib_destroy_cq returned %i\n",
 			__func__, rc);
+
+	if (ia->ri_dma_mr) {
+		rc = ib_dereg_mr(ia->ri_dma_mr);
+		dprintk("RPC:       %s: ib_dereg_mr returned %i\n",
+			__func__, rc);
+	}
 }
 
 /*
···
 		if (ia->ri_device != id->device) {
 			printk("RPC:       %s: can't reconnect on "
 				"different device!\n", __func__);
-			rdma_destroy_id(id);
+			rpcrdma_destroy_id(id);
 			rc = -ENETUNREACH;
 			goto out;
 		}
···
 		if (rc) {
 			dprintk("RPC:       %s: rdma_create_qp failed %i\n",
 				__func__, rc);
-			rdma_destroy_id(id);
+			rpcrdma_destroy_id(id);
 			rc = -ENETUNREACH;
 			goto out;
 		}
···
 		write_unlock(&ia->ri_qplock);
 
 		rdma_destroy_qp(old);
-		rdma_destroy_id(old);
+		rpcrdma_destroy_id(old);
 	} else {
 		dprintk("RPC:       %s: connecting...\n", __func__);
 		rc = rdma_create_qp(ia->ri_id, ia->ri_pd, &ep->rep_attr);
···
 		(unsigned long long)seg->mr_dma, seg->mr_dmalen);
 }
 
-static int
-rpcrdma_register_internal(struct rpcrdma_ia *ia, void *va, int len,
-				struct ib_mr **mrp, struct ib_sge *iov)
-{
-	struct ib_phys_buf ipb;
-	struct ib_mr *mr;
-	int rc;
-
-	/*
-	 * All memory passed here was kmalloc'ed, therefore phys-contiguous.
-	 */
-	iov->addr = ib_dma_map_single(ia->ri_device,
-			va, len, DMA_BIDIRECTIONAL);
-	if (ib_dma_mapping_error(ia->ri_device, iov->addr))
-		return -ENOMEM;
-
-	iov->length = len;
-
-	if (ia->ri_have_dma_lkey) {
-		*mrp = NULL;
-		iov->lkey = ia->ri_dma_lkey;
-		return 0;
-	} else if (ia->ri_bind_mem != NULL) {
-		*mrp = NULL;
-		iov->lkey = ia->ri_bind_mem->lkey;
-		return 0;
-	}
-
-	ipb.addr = iov->addr;
-	ipb.size = iov->length;
-	mr = ib_reg_phys_mr(ia->ri_pd, &ipb, 1,
-			IB_ACCESS_LOCAL_WRITE, &iov->addr);
-
-	dprintk("RPC:       %s: phys convert: 0x%llx "
-			"registered 0x%llx length %d\n",
-			__func__, (unsigned long long)ipb.addr,
-			(unsigned long long)iov->addr, len);
-
-	if (IS_ERR(mr)) {
-		*mrp = NULL;
-		rc = PTR_ERR(mr);
-		dprintk("RPC:       %s: failed with %i\n", __func__, rc);
-	} else {
-		*mrp = mr;
-		iov->lkey = mr->lkey;
-		rc = 0;
-	}
-
-	return rc;
-}
-
-static int
-rpcrdma_deregister_internal(struct rpcrdma_ia *ia,
-				struct ib_mr *mr, struct ib_sge *iov)
-{
-	int rc;
-
-	ib_dma_unmap_single(ia->ri_device,
-			    iov->addr, iov->length, DMA_BIDIRECTIONAL);
-
-	if (NULL == mr)
-		return 0;
-
-	rc = ib_dereg_mr(mr);
-	if (rc)
-		dprintk("RPC:       %s: ib_dereg_mr failed %i\n", __func__, rc);
-	return rc;
-}
-
 /**
  * rpcrdma_alloc_regbuf - kmalloc and register memory for SEND/RECV buffers
  * @ia: controlling rpcrdma_ia
···
 rpcrdma_alloc_regbuf(struct rpcrdma_ia *ia, size_t size, gfp_t flags)
 {
 	struct rpcrdma_regbuf *rb;
-	int rc;
+	struct ib_sge *iov;
 
-	rc = -ENOMEM;
 	rb = kmalloc(sizeof(*rb) + size, flags);
 	if (rb == NULL)
 		goto out;
 
-	rb->rg_size = size;
-	rb->rg_owner = NULL;
-	rc = rpcrdma_register_internal(ia, rb->rg_base, size,
-				       &rb->rg_mr, &rb->rg_iov);
-	if (rc)
+	iov = &rb->rg_iov;
+	iov->addr = ib_dma_map_single(ia->ri_device,
+				      (void *)rb->rg_base, size,
+				      DMA_BIDIRECTIONAL);
+	if (ib_dma_mapping_error(ia->ri_device, iov->addr))
 		goto out_free;
 
+	iov->length = size;
+	iov->lkey = ia->ri_dma_lkey;
+	rb->rg_size = size;
+	rb->rg_owner = NULL;
 	return rb;
 
 out_free:
 	kfree(rb);
 out:
-	return ERR_PTR(rc);
+	return ERR_PTR(-ENOMEM);
 }
 
 /**
···
 void
 rpcrdma_free_regbuf(struct rpcrdma_ia *ia, struct rpcrdma_regbuf *rb)
 {
-	if (rb) {
-		rpcrdma_deregister_internal(ia, rb->rg_mr, &rb->rg_iov);
-		kfree(rb);
-	}
+	struct ib_sge *iov;
+
+	if (!rb)
+		return;
+
+	iov = &rb->rg_iov;
+	ib_dma_unmap_single(ia->ri_device,
+			    iov->addr, iov->length, DMA_BIDIRECTIONAL);
+	kfree(rb);
 }
 
 /*
···
 		struct rpcrdma_ep *ep,
 		struct rpcrdma_req *req)
 {
+	struct ib_device *device = ia->ri_device;
 	struct ib_send_wr send_wr, *send_wr_fail;
 	struct rpcrdma_rep *rep = req->rl_reply;
-	int rc;
+	struct ib_sge *iov = req->rl_send_iov;
+	int i, rc;
 
 	if (rep) {
 		rc = rpcrdma_ep_post_recv(ia, ep, rep);
···
 
 	send_wr.next = NULL;
 	send_wr.wr_id = RPCRDMA_IGNORE_COMPLETION;
-	send_wr.sg_list = req->rl_send_iov;
+	send_wr.sg_list = iov;
 	send_wr.num_sge = req->rl_niovs;
 	send_wr.opcode = IB_WR_SEND;
-	if (send_wr.num_sge == 4)	/* no need to sync any pad (constant) */
-		ib_dma_sync_single_for_device(ia->ri_device,
-					      req->rl_send_iov[3].addr,
-					      req->rl_send_iov[3].length,
-					      DMA_TO_DEVICE);
-	ib_dma_sync_single_for_device(ia->ri_device,
-				      req->rl_send_iov[1].addr,
-				      req->rl_send_iov[1].length,
-				      DMA_TO_DEVICE);
-	ib_dma_sync_single_for_device(ia->ri_device,
-				      req->rl_send_iov[0].addr,
-				      req->rl_send_iov[0].length,
-				      DMA_TO_DEVICE);
+
+	for (i = 0; i < send_wr.num_sge; i++)
+		ib_dma_sync_single_for_device(device, iov[i].addr,
+					      iov[i].length, DMA_TO_DEVICE);
+	dprintk("RPC:       %s: posting %d s/g entries\n",
+		__func__, send_wr.num_sge);
 
 	if (DECR_CQCOUNT(ep) > 0)
 		send_wr.send_flags = 0;

net/sunrpc/xprtrdma/xprt_rdma.h

···
 	struct ib_device	*ri_device;
 	struct rdma_cm_id 	*ri_id;
 	struct ib_pd		*ri_pd;
-	struct ib_mr		*ri_bind_mem;
+	struct ib_mr		*ri_dma_mr;
 	u32			ri_dma_lkey;
-	int			ri_have_dma_lkey;
 	struct completion	ri_done;
 	int			ri_async_rc;
 	unsigned int		ri_max_frmr_depth;
···
 	int			rep_connected;
 	struct ib_qp_init_attr	rep_attr;
 	wait_queue_head_t 	rep_connect_wait;
-	struct rpcrdma_regbuf	*rep_padbuf;
 	struct rdma_conn_param	rep_remote_cma;
 	struct sockaddr_storage	rep_remote_addr;
 	struct delayed_work	rep_connect_worker;
···
 struct rpcrdma_regbuf {
 	size_t			rg_size;
 	struct rpcrdma_req	*rg_owner;
-	struct ib_mr		*rg_mr;
 	struct ib_sge		rg_iov;
 	__be32			rg_base[0] __attribute__ ((aligned(256)));
 };
···
  * struct rpcrdma_buffer. N is the max number of outstanding requests.
  */
 
-/* temporary static scatter/gather max */
-#define RPCRDMA_MAX_DATA_SEGS	(64)	/* max scatter/gather */
+#define RPCRDMA_MAX_DATA_SEGS	((1 * 1024 * 1024) / PAGE_SIZE)
 #define RPCRDMA_MAX_SEGS 	(RPCRDMA_MAX_DATA_SEGS + 2) /* head+tail = 2 */
 
 struct rpcrdma_buffer;
···
 	char		*mr_offset;	/* kva if no page, else offset */
 };
 
+#define RPCRDMA_MAX_IOVS	(2)
+
 struct rpcrdma_req {
-	unsigned int	rl_niovs;	/* 0, 2 or 4 */
-	unsigned int	rl_nchunks;	/* non-zero if chunks */
-	unsigned int	rl_connect_cookie;	/* retry detection */
-	struct rpcrdma_buffer *rl_buffer; /* home base for this structure */
+	unsigned int		rl_niovs;
+	unsigned int		rl_nchunks;
+	unsigned int		rl_connect_cookie;
+	struct rpcrdma_buffer	*rl_buffer;
 	struct rpcrdma_rep	*rl_reply;/* holder for reply buffer */
-	struct ib_sge	rl_send_iov[4];	/* for active requests */
-	struct rpcrdma_regbuf *rl_rdmabuf;
-	struct rpcrdma_regbuf *rl_sendbuf;
-	struct rpcrdma_mr_seg rl_segments[RPCRDMA_MAX_SEGS];
+	struct ib_sge		rl_send_iov[RPCRDMA_MAX_IOVS];
+	struct rpcrdma_regbuf	*rl_rdmabuf;
+	struct rpcrdma_regbuf	*rl_sendbuf;
+	struct rpcrdma_mr_seg	rl_segments[RPCRDMA_MAX_SEGS];
 };
 
 static inline struct rpcrdma_req *
···
 	unsigned long		hardway_register_count;
 	unsigned long		failed_marshal_count;
 	unsigned long		bad_reply_count;
+	unsigned long		nomsg_call_count;
 };
 
 /*

net/sunrpc/xprtsock.c

···
 	if (atomic_read(&transport->xprt.swapper))
 		sk_clear_memalloc(sk);
 
+	kernel_sock_shutdown(sock, SHUT_RDWR);
+
 	write_lock_bh(&sk->sk_callback_lock);
 	transport->inet = NULL;
 	transport->sock = NULL;
···
 	sk->sk_user_data = NULL;
 
 	xs_restore_old_callbacks(transport, sk);
+	xprt_clear_connected(xprt);
 	write_unlock_bh(&sk->sk_callback_lock);
 	xs_sock_reset_connection_flags(xprt);
 
···
 		sk->sk_data_ready = xs_local_data_ready;
 		sk->sk_write_space = xs_udp_write_space;
 		sk->sk_error_report = xs_error_report;
-		sk->sk_allocation = GFP_ATOMIC;
+		sk->sk_allocation = GFP_NOIO;
 
 		xprt_clear_connected(xprt);
 
···
 		sk->sk_user_data = xprt;
 		sk->sk_data_ready = xs_udp_data_ready;
 		sk->sk_write_space = xs_udp_write_space;
-		sk->sk_allocation = GFP_ATOMIC;
+		sk->sk_allocation = GFP_NOIO;
 
 		xprt_set_connected(xprt);
 
···
 		sk->sk_state_change = xs_tcp_state_change;
 		sk->sk_write_space = xs_tcp_write_space;
 		sk->sk_error_report = xs_error_report;
-		sk->sk_allocation = GFP_ATOMIC;
+		sk->sk_allocation = GFP_NOIO;
 
 		/* socket options */
 		sock_reset_flag(sk, SOCK_LINGER);
···
 
 	WARN_ON_ONCE(!xprt_lock_connect(xprt, task, transport));
 
-	/* Start by resetting any existing state */
-	xs_reset_transport(transport);
-
-	if (transport->sock != NULL && !RPC_IS_SOFTCONN(task)) {
+	if (transport->sock != NULL) {
 		dprintk("RPC:       xs_connect delayed xprt %p for %lu "
 				"seconds\n",
 				xprt, xprt->reestablish_timeout / HZ);
+
+		/* Start by resetting any existing state */
+		xs_reset_transport(transport);
+
 		queue_delayed_work(rpciod_workqueue,
 				   &transport->connect_worker,
 				   xprt->reestablish_timeout);