1 files changed, 198 insertions, 46 deletions

diff --git a/fs/nfs/pnfs.c b/fs/nfs/pnfs.c
index 542ea8dfd1bc..b8d78f393365 100644
--- a/fs/nfs/pnfs.c
+++ b/fs/nfs/pnfs.c
@@ -268,11 +268,11 @@ pnfs_free_layout_hdr(struct pnfs_layout_hdr *lo)
 	struct nfs_server *server = NFS_SERVER(lo->plh_inode);
 	struct pnfs_layoutdriver_type *ld = server->pnfs_curr_ld;
 
-	if (!list_empty(&lo->plh_layouts)) {
+	if (test_and_clear_bit(NFS_LAYOUT_HASHED, &lo->plh_flags)) {
 		struct nfs_client *clp = server->nfs_client;
 
 		spin_lock(&clp->cl_lock);
-		list_del_init(&lo->plh_layouts);
+		list_del_rcu(&lo->plh_layouts);
 		spin_unlock(&clp->cl_lock);
 	}
 	put_cred(lo->plh_lc_cred);
@@ -309,6 +309,16 @@ pnfs_put_layout_hdr(struct pnfs_layout_hdr *lo)
 	}
 }
 
+static struct inode *
+pnfs_grab_inode_layout_hdr(struct pnfs_layout_hdr *lo)
+{
+	struct inode *inode = igrab(lo->plh_inode);
+	if (inode)
+		return inode;
+	set_bit(NFS_LAYOUT_INODE_FREEING, &lo->plh_flags);
+	return NULL;
+}
+
 static void
 pnfs_set_plh_return_info(struct pnfs_layout_hdr *lo, enum pnfs_iomode iomode,
 			 u32 seq)
@@ -496,6 +506,7 @@ pnfs_init_lseg(struct pnfs_layout_hdr *lo, struct pnfs_layout_segment *lseg,
 {
 	INIT_LIST_HEAD(&lseg->pls_list);
 	INIT_LIST_HEAD(&lseg->pls_lc_list);
+	INIT_LIST_HEAD(&lseg->pls_commits);
 	refcount_set(&lseg->pls_refcount, 1);
 	set_bit(NFS_LSEG_VALID, &lseg->pls_flags);
 	lseg->pls_layout = lo;
@@ -782,9 +793,10 @@ pnfs_layout_bulk_destroy_byserver_locked(struct nfs_client *clp,
 		/* If the sb is being destroyed, just bail */
 		if (!nfs_sb_active(server->super))
 			break;
-		inode = igrab(lo->plh_inode);
+		inode = pnfs_grab_inode_layout_hdr(lo);
 		if (inode != NULL) {
-			list_del_init(&lo->plh_layouts);
+			if (test_and_clear_bit(NFS_LAYOUT_HASHED, &lo->plh_flags))
+				list_del_rcu(&lo->plh_layouts);
 			if (pnfs_layout_add_bulk_destroy_list(inode,
 						layout_list))
 				continue;
@@ -794,7 +806,6 @@ pnfs_layout_bulk_destroy_byserver_locked(struct nfs_client *clp,
 		} else {
 			rcu_read_unlock();
 			spin_unlock(&clp->cl_lock);
-			set_bit(NFS_LAYOUT_INODE_FREEING, &lo->plh_flags);
 		}
 		nfs_sb_deactive(server->super);
 		spin_lock(&clp->cl_lock);
@@ -903,10 +914,21 @@ pnfs_destroy_all_layouts(struct nfs_client *clp)
 	pnfs_destroy_layouts_byclid(clp, false);
 }
 
+static void
+pnfs_set_layout_cred(struct pnfs_layout_hdr *lo, const struct cred *cred)
+{
+	const struct cred *old;
+
+	if (cred && cred_fscmp(lo->plh_lc_cred, cred) != 0) {
+		old = xchg(&lo->plh_lc_cred, get_cred(cred));
+		put_cred(old);
+	}
+}
+
 /* update lo->plh_stateid with new if is more recent */
 void
 pnfs_set_layout_stateid(struct pnfs_layout_hdr *lo, const nfs4_stateid *new,
-			bool update_barrier)
+			const struct cred *cred, bool update_barrier)
 {
 	u32 oldseq, newseq, new_barrier = 0;
 
@@ -914,6 +936,7 @@ pnfs_set_layout_stateid(struct pnfs_layout_hdr *lo, const nfs4_stateid *new,
 	newseq = be32_to_cpu(new->seqid);
 
 	if (!pnfs_layout_is_valid(lo)) {
+		pnfs_set_layout_cred(lo, cred);
 		nfs4_stateid_copy(&lo->plh_stateid, new);
 		lo->plh_barrier = newseq;
 		pnfs_clear_layoutreturn_info(lo);
@@ -1061,7 +1084,7 @@ pnfs_alloc_init_layoutget_args(struct inode *ino,
 	lgp->args.ctx = get_nfs_open_context(ctx);
 	nfs4_stateid_copy(&lgp->args.stateid, stateid);
 	lgp->gfp_flags = gfp_flags;
-	lgp->cred = get_cred(ctx->cred);
+	lgp->cred = ctx->cred;
 	return lgp;
 }
 
@@ -1072,7 +1095,6 @@ void pnfs_layoutget_free(struct nfs4_layoutget *lgp)
 	nfs4_free_pages(lgp->args.layout.pages, max_pages);
 	if (lgp->args.inode)
 		pnfs_put_layout_hdr(NFS_I(lgp->args.inode)->layout);
-	put_cred(lgp->cred);
 	put_nfs_open_context(lgp->args.ctx);
 	kfree(lgp);
 }
@@ -1109,7 +1131,7 @@ void pnfs_layoutreturn_free_lsegs(struct pnfs_layout_hdr *lo,
 
 		pnfs_mark_matching_lsegs_invalid(lo, &freeme, range, seq);
 		pnfs_free_returned_lsegs(lo, &freeme, range, seq);
-		pnfs_set_layout_stateid(lo, stateid, true);
+		pnfs_set_layout_stateid(lo, stateid, NULL, true);
 	} else
 		pnfs_mark_layout_stateid_invalid(lo, &freeme);
 out_unlock:
@@ -1122,6 +1144,7 @@ out_unlock:
 static bool
 pnfs_prepare_layoutreturn(struct pnfs_layout_hdr *lo,
 		nfs4_stateid *stateid,
+		const struct cred **cred,
 		enum pnfs_iomode *iomode)
 {
 	/* Serialise LAYOUTGET/LAYOUTRETURN */
@@ -1132,18 +1155,17 @@ pnfs_prepare_layoutreturn(struct pnfs_layout_hdr *lo,
 	set_bit(NFS_LAYOUT_RETURN, &lo->plh_flags);
 	pnfs_get_layout_hdr(lo);
 	if (test_bit(NFS_LAYOUT_RETURN_REQUESTED, &lo->plh_flags)) {
-		if (stateid != NULL) {
-			nfs4_stateid_copy(stateid, &lo->plh_stateid);
-			if (lo->plh_return_seq != 0)
-				stateid->seqid = cpu_to_be32(lo->plh_return_seq);
-		}
+		nfs4_stateid_copy(stateid, &lo->plh_stateid);
+		*cred = get_cred(lo->plh_lc_cred);
+		if (lo->plh_return_seq != 0)
+			stateid->seqid = cpu_to_be32(lo->plh_return_seq);
 		if (iomode != NULL)
 			*iomode = lo->plh_return_iomode;
 		pnfs_clear_layoutreturn_info(lo);
 		return true;
 	}
-	if (stateid != NULL)
-		nfs4_stateid_copy(stateid, &lo->plh_stateid);
+	nfs4_stateid_copy(stateid, &lo->plh_stateid);
+	*cred = get_cred(lo->plh_lc_cred);
 	if (iomode != NULL)
 		*iomode = IOMODE_ANY;
 	return true;
@@ -1167,20 +1189,26 @@ pnfs_init_layoutreturn_args(struct nfs4_layoutreturn_args *args,
 }
 
 static int
-pnfs_send_layoutreturn(struct pnfs_layout_hdr *lo, const nfs4_stateid *stateid,
-		       enum pnfs_iomode iomode, bool sync)
+pnfs_send_layoutreturn(struct pnfs_layout_hdr *lo,
+		       const nfs4_stateid *stateid,
+		       const struct cred **pcred,
+		       enum pnfs_iomode iomode,
+		       bool sync)
 {
 	struct inode *ino = lo->plh_inode;
 	struct pnfs_layoutdriver_type *ld = NFS_SERVER(ino)->pnfs_curr_ld;
 	struct nfs4_layoutreturn *lrp;
+	const struct cred *cred = *pcred;
 	int status = 0;
 
+	*pcred = NULL;
 	lrp = kzalloc(sizeof(*lrp), GFP_NOFS);
 	if (unlikely(lrp == NULL)) {
 		status = -ENOMEM;
 		spin_lock(&ino->i_lock);
 		pnfs_clear_layoutreturn_waitbit(lo);
 		spin_unlock(&ino->i_lock);
+		put_cred(cred);
 		pnfs_put_layout_hdr(lo);
 		goto out;
 	}
@@ -1188,7 +1216,7 @@ pnfs_send_layoutreturn(struct pnfs_layout_hdr *lo, const nfs4_stateid *stateid,
 	pnfs_init_layoutreturn_args(&lrp->args, lo, stateid, iomode);
 	lrp->args.ld_private = &lrp->ld_private;
 	lrp->clp = NFS_SERVER(ino)->nfs_client;
-	lrp->cred = lo->plh_lc_cred;
+	lrp->cred = cred;
 	if (ld->prepare_layoutreturn)
 		ld->prepare_layoutreturn(&lrp->args);
 
@@ -1233,15 +1261,16 @@ static void pnfs_layoutreturn_before_put_layout_hdr(struct pnfs_layout_hdr *lo)
 		return;
 	spin_lock(&inode->i_lock);
 	if (pnfs_layout_need_return(lo)) {
+		const struct cred *cred;
 		nfs4_stateid stateid;
 		enum pnfs_iomode iomode;
 		bool send;
 
-		send = pnfs_prepare_layoutreturn(lo, &stateid, &iomode);
+		send = pnfs_prepare_layoutreturn(lo, &stateid, &cred, &iomode);
 		spin_unlock(&inode->i_lock);
 		if (send) {
 			/* Send an async layoutreturn so we dont deadlock */
-			pnfs_send_layoutreturn(lo, &stateid, iomode, false);
+			pnfs_send_layoutreturn(lo, &stateid, &cred, iomode, false);
 		}
 	} else
 		spin_unlock(&inode->i_lock);
@@ -1261,6 +1290,7 @@ _pnfs_return_layout(struct inode *ino)
 	struct pnfs_layout_hdr *lo = NULL;
 	struct nfs_inode *nfsi = NFS_I(ino);
 	LIST_HEAD(tmp_list);
+	const struct cred *cred;
 	nfs4_stateid stateid;
 	int status = 0;
 	bool send, valid_layout;
@@ -1305,10 +1335,10 @@ _pnfs_return_layout(struct inode *ino)
 		goto out_put_layout_hdr;
 	}
 
-	send = pnfs_prepare_layoutreturn(lo, &stateid, NULL);
+	send = pnfs_prepare_layoutreturn(lo, &stateid, &cred, NULL);
 	spin_unlock(&ino->i_lock);
 	if (send)
-		status = pnfs_send_layoutreturn(lo, &stateid, IOMODE_ANY, true);
+		status = pnfs_send_layoutreturn(lo, &stateid, &cred, IOMODE_ANY, true);
 out_put_layout_hdr:
 	pnfs_free_lseg_list(&tmp_list);
 	pnfs_put_layout_hdr(lo);
@@ -1354,6 +1384,7 @@ bool pnfs_roc(struct inode *ino,
 	struct nfs4_state *state;
 	struct pnfs_layout_hdr *lo;
 	struct pnfs_layout_segment *lseg, *next;
+	const struct cred *lc_cred;
 	nfs4_stateid stateid;
 	enum pnfs_iomode iomode = 0;
 	bool layoutreturn = false, roc = false;
@@ -1423,16 +1454,20 @@ retry:
 	 * 2. we don't send layoutreturn
 	 */
 	/* lo ref dropped in pnfs_roc_release() */
-	layoutreturn = pnfs_prepare_layoutreturn(lo, &stateid, &iomode);
+	layoutreturn = pnfs_prepare_layoutreturn(lo, &stateid, &lc_cred, &iomode);
 	/* If the creds don't match, we can't compound the layoutreturn */
-	if (!layoutreturn || cred_fscmp(cred, lo->plh_lc_cred) != 0)
+	if (!layoutreturn)
 		goto out_noroc;
+	if (cred_fscmp(cred, lc_cred) != 0)
+		goto out_noroc_put_cred;
 
 	roc = layoutreturn;
 	pnfs_init_layoutreturn_args(args, lo, &stateid, iomode);
 	res->lrs_present = 0;
 	layoutreturn = false;
 
+out_noroc_put_cred:
+	put_cred(lc_cred);
 out_noroc:
 	spin_unlock(&ino->i_lock);
 	rcu_read_unlock();
@@ -1445,7 +1480,7 @@ out_noroc:
 		return true;
 	}
 	if (layoutreturn)
-		pnfs_send_layoutreturn(lo, &stateid, iomode, true);
+		pnfs_send_layoutreturn(lo, &stateid, &lc_cred, iomode, true);
 	pnfs_put_layout_hdr(lo);
 	return false;
 }
@@ -1859,15 +1894,14 @@ static void pnfs_clear_first_layoutget(struct pnfs_layout_hdr *lo)
 static void _add_to_server_list(struct pnfs_layout_hdr *lo,
 				struct nfs_server *server)
 {
-	if (list_empty(&lo->plh_layouts)) {
+	if (!test_and_set_bit(NFS_LAYOUT_HASHED, &lo->plh_flags)) {
 		struct nfs_client *clp = server->nfs_client;
 
 		/* The lo must be on the clp list if there is any
 		 * chance of a CB_LAYOUTRECALL(FILE) coming in.
 		 */
 		spin_lock(&clp->cl_lock);
-		if (list_empty(&lo->plh_layouts))
-			list_add_tail(&lo->plh_layouts, &server->layouts);
+		list_add_tail_rcu(&lo->plh_layouts, &server->layouts);
 		spin_unlock(&clp->cl_lock);
 	}
 }
@@ -1989,6 +2023,7 @@ lookup_again:
 			goto lookup_again;
 		}
 
+		spin_unlock(&ino->i_lock);
 		first = true;
 		status = nfs4_select_rw_stateid(ctx->state,
 					iomode == IOMODE_RW ? FMODE_WRITE : FMODE_READ,
@@ -1998,12 +2033,12 @@ lookup_again:
 			trace_pnfs_update_layout(ino, pos, count,
 					iomode, lo, lseg,
 					PNFS_UPDATE_LAYOUT_INVALID_OPEN);
-			spin_unlock(&ino->i_lock);
 			nfs4_schedule_stateid_recovery(server, ctx->state);
 			pnfs_clear_first_layoutget(lo);
 			pnfs_put_layout_hdr(lo);
 			goto lookup_again;
 		}
+		spin_lock(&ino->i_lock);
 	} else {
 		nfs4_stateid_copy(&stateid, &lo->plh_stateid);
 	}
@@ -2323,14 +2358,14 @@ pnfs_layout_process(struct nfs4_layoutget *lgp)
 
 	if (!pnfs_layout_is_valid(lo)) {
 		/* We have a completely new layout */
-		pnfs_set_layout_stateid(lo, &res->stateid, true);
+		pnfs_set_layout_stateid(lo, &res->stateid, lgp->cred, true);
 	} else if (nfs4_stateid_match_other(&lo->plh_stateid, &res->stateid)) {
 		/* existing state ID, make sure the sequence number matches. */
 		if (pnfs_layout_stateid_blocked(lo, &res->stateid)) {
 			dprintk("%s forget reply due to sequence\n", __func__);
 			goto out_forget;
 		}
-		pnfs_set_layout_stateid(lo, &res->stateid, false);
+		pnfs_set_layout_stateid(lo, &res->stateid, lgp->cred, false);
 	} else {
 		/*
 		 * We got an entirely new state ID.  Mark all segments for the
@@ -2423,43 +2458,159 @@ pnfs_mark_matching_lsegs_return(struct pnfs_layout_hdr *lo,
 	return -ENOENT;
 }
 
-void pnfs_error_mark_layout_for_return(struct inode *inode,
-				       struct pnfs_layout_segment *lseg)
+static void
+pnfs_mark_layout_for_return(struct inode *inode,
+			    const struct pnfs_layout_range *range)
 {
-	struct pnfs_layout_hdr *lo = NFS_I(inode)->layout;
-	struct pnfs_layout_range range = {
-		.iomode = lseg->pls_range.iomode,
-		.offset = 0,
-		.length = NFS4_MAX_UINT64,
-	};
+	struct pnfs_layout_hdr *lo;
 	bool return_now = false;
 
 	spin_lock(&inode->i_lock);
+	lo = NFS_I(inode)->layout;
 	if (!pnfs_layout_is_valid(lo)) {
 		spin_unlock(&inode->i_lock);
 		return;
 	}
-	pnfs_set_plh_return_info(lo, range.iomode, 0);
+	pnfs_set_plh_return_info(lo, range->iomode, 0);
 	/*
 	 * mark all matching lsegs so that we are sure to have no live
 	 * segments at hand when sending layoutreturn. See pnfs_put_lseg()
 	 * for how it works.
 	 */
-	if (pnfs_mark_matching_lsegs_return(lo, &lo->plh_return_segs, &range, 0) != -EBUSY) {
+	if (pnfs_mark_matching_lsegs_return(lo, &lo->plh_return_segs, range, 0) != -EBUSY) {
+		const struct cred *cred;
 		nfs4_stateid stateid;
 		enum pnfs_iomode iomode;
 
-		return_now = pnfs_prepare_layoutreturn(lo, &stateid, &iomode);
+		return_now = pnfs_prepare_layoutreturn(lo, &stateid, &cred, &iomode);
 		spin_unlock(&inode->i_lock);
 		if (return_now)
-			pnfs_send_layoutreturn(lo, &stateid, iomode, false);
+			pnfs_send_layoutreturn(lo, &stateid, &cred, iomode, false);
 	} else {
 		spin_unlock(&inode->i_lock);
 		nfs_commit_inode(inode, 0);
 	}
 }
+
+void pnfs_error_mark_layout_for_return(struct inode *inode,
+				       struct pnfs_layout_segment *lseg)
+{
+	struct pnfs_layout_range range = {
+		.iomode = lseg->pls_range.iomode,
+		.offset = 0,
+		.length = NFS4_MAX_UINT64,
+	};
+
+	pnfs_mark_layout_for_return(inode, &range);
+}
 EXPORT_SYMBOL_GPL(pnfs_error_mark_layout_for_return);
 
+static bool
+pnfs_layout_can_be_returned(struct pnfs_layout_hdr *lo)
+{
+	return pnfs_layout_is_valid(lo) &&
+		!test_bit(NFS_LAYOUT_INODE_FREEING, &lo->plh_flags) &&
+		!test_bit(NFS_LAYOUT_RETURN, &lo->plh_flags);
+}
+
+static struct pnfs_layout_segment *
+pnfs_find_first_lseg(struct pnfs_layout_hdr *lo,
+		     const struct pnfs_layout_range *range,
+		     enum pnfs_iomode iomode)
+{
+	struct pnfs_layout_segment *lseg;
+
+	list_for_each_entry(lseg, &lo->plh_segs, pls_list) {
+		if (!test_bit(NFS_LSEG_VALID, &lseg->pls_flags))
+			continue;
+		if (test_bit(NFS_LSEG_LAYOUTRETURN, &lseg->pls_flags))
+			continue;
+		if (lseg->pls_range.iomode != iomode && iomode != IOMODE_ANY)
+			continue;
+		if (pnfs_lseg_range_intersecting(&lseg->pls_range, range))
+			return lseg;
+	}
+	return NULL;
+}
+
+/* Find open file states whose mode matches that of the range */
+static bool
+pnfs_should_return_unused_layout(struct pnfs_layout_hdr *lo,
+				 const struct pnfs_layout_range *range)
+{
+	struct list_head *head;
+	struct nfs_open_context *ctx;
+	fmode_t mode = 0;
+
+	if (!pnfs_layout_can_be_returned(lo) ||
+	    !pnfs_find_first_lseg(lo, range, range->iomode))
+		return false;
+
+	head = &NFS_I(lo->plh_inode)->open_files;
+	list_for_each_entry_rcu(ctx, head, list) {
+		if (ctx->state)
+			mode |= ctx->state->state & (FMODE_READ|FMODE_WRITE);
+	}
+
+	switch (range->iomode) {
+	default:
+		break;
+	case IOMODE_READ:
+		mode &= ~FMODE_WRITE;
+		break;
+	case IOMODE_RW:
+		if (pnfs_find_first_lseg(lo, range, IOMODE_READ))
+			mode &= ~FMODE_READ;
+	}
+	return mode == 0;
+}
+
+static int
+pnfs_layout_return_unused_byserver(struct nfs_server *server, void *data)
+{
+	const struct pnfs_layout_range *range = data;
+	struct pnfs_layout_hdr *lo;
+	struct inode *inode;
+restart:
+	rcu_read_lock();
+	list_for_each_entry_rcu(lo, &server->layouts, plh_layouts) {
+		if (!pnfs_layout_can_be_returned(lo) ||
+		    test_bit(NFS_LAYOUT_RETURN_REQUESTED, &lo->plh_flags))
+			continue;
+		inode = lo->plh_inode;
+		spin_lock(&inode->i_lock);
+		if (!pnfs_should_return_unused_layout(lo, range)) {
+			spin_unlock(&inode->i_lock);
+			continue;
+		}
+		spin_unlock(&inode->i_lock);
+		inode = pnfs_grab_inode_layout_hdr(lo);
+		if (!inode)
+			continue;
+		rcu_read_unlock();
+		pnfs_mark_layout_for_return(inode, range);
+		iput(inode);
+		cond_resched();
+		goto restart;
+	}
+	rcu_read_unlock();
+	return 0;
+}
+
+void
+pnfs_layout_return_unused_byclid(struct nfs_client *clp,
+				 enum pnfs_iomode iomode)
+{
+	struct pnfs_layout_range range = {
+		.iomode = iomode,
+		.offset = 0,
+		.length = NFS4_MAX_UINT64,
+	};
+
+	nfs_client_for_each_server(clp, pnfs_layout_return_unused_byserver,
+			&range);
+}
+
 void
 pnfs_generic_pg_check_layout(struct nfs_pageio_descriptor *pgio)
 {
@@ -2475,7 +2626,7 @@ EXPORT_SYMBOL_GPL(pnfs_generic_pg_check_layout);
  * Check for any intersection between the request and the pgio->pg_lseg,
  * and if none, put this pgio->pg_lseg away.
  */
-static void
+void
 pnfs_generic_pg_check_range(struct nfs_pageio_descriptor *pgio, struct nfs_page *req)
 {
 	if (pgio->pg_lseg && !pnfs_lseg_request_intersecting(pgio->pg_lseg, req)) {
@@ -2483,6 +2634,7 @@ pnfs_generic_pg_check_range(struct nfs_pageio_descriptor *pgio, struct nfs_page
 		pgio->pg_lseg = NULL;
 	}
 }
+EXPORT_SYMBOL_GPL(pnfs_generic_pg_check_range);
 
 void
 pnfs_generic_pg_init_read(struct nfs_pageio_descriptor *pgio, struct nfs_page *req)
@@ -3000,10 +3152,10 @@ pnfs_layoutcommit_inode(struct inode *inode, bool sync)
 	end_pos = nfsi->layout->plh_lwb;
 
 	nfs4_stateid_copy(&data->args.stateid, &nfsi->layout->plh_stateid);
+	data->cred = get_cred(nfsi->layout->plh_lc_cred);
 	spin_unlock(&inode->i_lock);
 
 	data->args.inode = inode;
-	data->cred = get_cred(nfsi->layout->plh_lc_cred);
 	nfs_fattr_init(&data->fattr);
 	data->args.bitmask = NFS_SERVER(inode)->cache_consistency_bitmask;
 	data->res.fattr = &data->fattr;