NFSv4.1: Clean ups and bugfixes for the pNFS read/writeback/commit code

Move more pnfs-isms out of the generic commit code.

Bugfixes:

- filelayout_scan_commit_lists doesn't need to get/put the lseg.
  In fact since it is run under the inode->i_lock, the lseg_put()
  can deadlock.

- Ensure that we distinguish between what needs to be done for
  commit-to-data server and what needs to be done for commit-to-MDS
  using the new flag PG_COMMIT_TO_DS. Otherwise we may end up calling
  put_lseg() on a bucket for a struct nfs_page that got written
  through the MDS.

- Fix a case where we were using list_del() on an nfs_page->wb_list
  instead of list_del_init().

- filelayout_initiate_commit needs to call filelayout_commit_release
  on error instead of the mds_ops->rpc_release(). Otherwise it won't
  clear the commit lock.

Cleanups:

- Let the files layout manage the commit lists for the pNFS case.
  Don't expose stuff like pnfs_choose_commit_list, and the fact
  that the commit buckets hold references to the layout segment
  in common code.

- Cast out the put_lseg() calls for the struct nfs_read/write_data->lseg
  into the pNFS layer from whence they came.

- Let the pNFS layer manage the NFS_INO_PNFS_COMMIT bit.

Signed-off-by: Trond Myklebust <Trond.Myklebust@netapp.com>
Cc: Fred Isaman <iisaman@netapp.com>

fs/nfs/internal.h

@@ -308,8 +308,6 @@ extern void nfs_pageio_reset_read_mds(struct nfs_pageio_descriptor *pgio);
 extern void nfs_readdata_release(struct nfs_read_data *rdata);
 
 /* write.c */
-extern int nfs_scan_commit_list(struct list_head *src, struct list_head *dst,
-				int max);
 extern int nfs_generic_flush(struct nfs_pageio_descriptor *desc,
 		struct list_head *head);
 extern void nfs_pageio_init_write_mds(struct nfs_pageio_descriptor *pgio,
@@ -334,6 +332,8 @@ void nfs_retry_commit(struct list_head *page_list,
 void nfs_commit_clear_lock(struct nfs_inode *nfsi);
 void nfs_commitdata_release(void *data);
 void nfs_commit_release_pages(struct nfs_write_data *data);
+void nfs_request_add_commit_list(struct nfs_page *req, struct list_head *head);
+void nfs_request_remove_commit_list(struct nfs_page *req);
 
 #ifdef CONFIG_MIGRATION
 extern int nfs_migrate_page(struct address_space *,

fs/nfs/nfs4filelayout.c

@@ -224,6 +224,7 @@ static void filelayout_read_release(void *data)
 {
 	struct nfs_read_data *rdata = (struct nfs_read_data *)data;
 
+	put_lseg(rdata->lseg);
 	rdata->mds_ops->rpc_release(data);
 }
 
@@ -310,6 +311,7 @@ static void filelayout_write_release(void *data)
 {
 	struct nfs_write_data *wdata = (struct nfs_write_data *)data;
 
+	put_lseg(wdata->lseg);
 	wdata->mds_ops->rpc_release(data);
 }
 
@@ -320,6 +322,7 @@ static void filelayout_commit_release(void *data)
 	nfs_commit_release_pages(wdata);
 	if (atomic_dec_and_test(&NFS_I(wdata->inode)->commits_outstanding))
 		nfs_commit_clear_lock(NFS_I(wdata->inode));
+	put_lseg(wdata->lseg);
 	nfs_commitdata_release(wdata);
 }
 
@@ -779,11 +782,16 @@ static u32 select_bucket_index(struct nfs4_filelayout_segment *fl, u32 j)
 
 /* The generic layer is about to remove the req from the commit list.
  * If this will make the bucket empty, it will need to put the lseg reference.
- * Note inode lock is held, so we can't do the put here.
  */
-static struct pnfs_layout_segment *
-filelayout_remove_commit_req(struct nfs_page *req)
+static void
+filelayout_clear_request_commit(struct nfs_page *req)
 {
+	struct pnfs_layout_segment *freeme = NULL;
+	struct inode *inode = req->wb_context->dentry->d_inode;
+
+	spin_lock(&inode->i_lock);
+	if (!test_and_clear_bit(PG_COMMIT_TO_DS, &req->wb_flags))
+		goto out;
 	if (list_is_singular(&req->wb_list)) {
 		struct inode *inode = req->wb_context->dentry->d_inode;
 		struct pnfs_layout_segment *lseg;
@@ -792,11 +800,16 @@ filelayout_remove_commit_req(struct nfs_page *req)
 		 * since there is only one relevant lseg...
 		 */
 		list_for_each_entry(lseg, &NFS_I(inode)->layout->plh_segs, pls_list) {
-			if (lseg->pls_range.iomode == IOMODE_RW)
-				return lseg;
+			if (lseg->pls_range.iomode == IOMODE_RW) {
+				freeme = lseg;
+				break;
+			}
 		}
 	}
-	return NULL;
+out:
+	nfs_request_remove_commit_list(req);
+	spin_unlock(&inode->i_lock);
+	put_lseg(freeme);
 }
 
 static struct list_head *
@@ -829,9 +842,20 @@ filelayout_choose_commit_list(struct nfs_page *req,
 		 */
 		get_lseg(lseg);
 	}
+	set_bit(PG_COMMIT_TO_DS, &req->wb_flags);
 	return list;
 }
 
+static void
+filelayout_mark_request_commit(struct nfs_page *req,
+		struct pnfs_layout_segment *lseg)
+{
+	struct list_head *list;
+
+	list = filelayout_choose_commit_list(req, lseg);
+	nfs_request_add_commit_list(req, list);
+}
+
 static u32 calc_ds_index_from_commit(struct pnfs_layout_segment *lseg, u32 i)
 {
 	struct nfs4_filelayout_segment *flseg = FILELAYOUT_LSEG(lseg);
@@ -872,7 +896,7 @@ static int filelayout_initiate_commit(struct nfs_write_data *data, int how)
 		set_bit(lo_fail_bit(IOMODE_RW), &lseg->pls_layout->plh_flags);
 		set_bit(lo_fail_bit(IOMODE_READ), &lseg->pls_layout->plh_flags);
 		prepare_to_resend_writes(data);
-		data->mds_ops->rpc_release(data);
+		filelayout_commit_release(data);
 		return -EAGAIN;
 	}
 	dprintk("%s ino %lu, how %d\n", __func__, data->inode->i_ino, how);
@@ -895,7 +919,7 @@ find_only_write_lseg_locked(struct inode *inode)
 
 	list_for_each_entry(lseg, &NFS_I(inode)->layout->plh_segs, pls_list)
 		if (lseg->pls_range.iomode == IOMODE_RW)
-			return get_lseg(lseg);
+			return lseg;
 	return NULL;
 }
 
@@ -905,10 +929,33 @@ static struct pnfs_layout_segment *find_only_write_lseg(struct inode *inode)
 
 	spin_lock(&inode->i_lock);
 	rv = find_only_write_lseg_locked(inode);
+	if (rv)
+		get_lseg(rv);
 	spin_unlock(&inode->i_lock);
 	return rv;
 }
 
+static int
+filelayout_scan_ds_commit_list(struct nfs4_fl_commit_bucket *bucket, int max)
+{
+	struct list_head *src = &bucket->written;
+	struct list_head *dst = &bucket->committing;
+	struct nfs_page *req, *tmp;
+	int ret = 0;
+
+	list_for_each_entry_safe(req, tmp, src, wb_list) {
+		if (!nfs_lock_request(req))
+			continue;
+		nfs_request_remove_commit_list(req);
+		clear_bit(PG_COMMIT_TO_DS, &req->wb_flags);
+		nfs_list_add_request(req, dst);
+		ret++;
+		if (ret == max)
+			break;
+	}
+	return ret;
+}
+
 /* Move reqs from written to committing lists, returning count of number moved.
  * Note called with i_lock held.
  */
@@ -920,21 +967,16 @@ static int filelayout_scan_commit_lists(struct inode *inode, int max)
 
 	lseg = find_only_write_lseg_locked(inode);
 	if (!lseg)
-		return 0;
+		goto out_done;
 	fl = FILELAYOUT_LSEG(lseg);
 	if (fl->commit_through_mds)
-		goto out_put;
-	for (i = 0; i < fl->number_of_buckets; i++) {
-		if (list_empty(&fl->commit_buckets[i].written))
-			continue;
-		cnt = nfs_scan_commit_list(&fl->commit_buckets[i].written,
-					   &fl->commit_buckets[i].committing,
-					   max);
+		goto out_done;
+	for (i = 0; i < fl->number_of_buckets && max != 0; i++) {
+		cnt = filelayout_scan_ds_commit_list(&fl->commit_buckets[i], max);
 		max -= cnt;
 		rv += cnt;
 	}
-out_put:
-	put_lseg(lseg);
+out_done:
 	return rv;
 }
 
@@ -1033,8 +1075,8 @@ static struct pnfs_layoutdriver_type filelayout_type = {
 	.free_lseg		= filelayout_free_lseg,
 	.pg_read_ops		= &filelayout_pg_read_ops,
 	.pg_write_ops		= &filelayout_pg_write_ops,
-	.choose_commit_list	= filelayout_choose_commit_list,
-	.remove_commit_req	= filelayout_remove_commit_req,
+	.mark_request_commit	= filelayout_mark_request_commit,
+	.clear_request_commit	= filelayout_clear_request_commit,
 	.scan_commit_lists	= filelayout_scan_commit_lists,
 	.commit_pagelist	= filelayout_commit_pagelist,
 	.read_pagelist		= filelayout_read_pagelist,

fs/nfs/pnfs.c

@@ -1210,6 +1210,7 @@ void pnfs_ld_write_done(struct nfs_write_data *data)
 		}
 		data->task.tk_status = pnfs_write_done_resend_to_mds(data->inode, &data->pages);
 	}
+	put_lseg(data->lseg);
 	data->mds_ops->rpc_release(data);
 }
 EXPORT_SYMBOL_GPL(pnfs_ld_write_done);
@@ -1223,6 +1224,7 @@ pnfs_write_through_mds(struct nfs_pageio_descriptor *desc,
 		nfs_list_add_request(data->req, &desc->pg_list);
 	nfs_pageio_reset_write_mds(desc);
 	desc->pg_recoalesce = 1;
+	put_lseg(data->lseg);
 	nfs_writedata_release(data);
 }
 
@@ -1323,6 +1325,7 @@ void pnfs_ld_read_done(struct nfs_read_data *data)
 		data->mds_ops->rpc_call_done(&data->task, data);
 	} else
 		pnfs_ld_handle_read_error(data);
+	put_lseg(data->lseg);
 	data->mds_ops->rpc_release(data);
 }
 EXPORT_SYMBOL_GPL(pnfs_ld_read_done);

fs/nfs/pnfs.h

@@ -94,9 +94,9 @@ struct pnfs_layoutdriver_type {
 	const struct nfs_pageio_ops *pg_read_ops;
 	const struct nfs_pageio_ops *pg_write_ops;
 
-	struct list_head * (*choose_commit_list) (struct nfs_page *req,
+	void (*mark_request_commit) (struct nfs_page *req,
 					struct pnfs_layout_segment *lseg);
-	struct pnfs_layout_segment *(*remove_commit_req) (struct nfs_page *req);
+	void (*clear_request_commit) (struct nfs_page *req);
 	int (*scan_commit_lists) (struct inode *inode, int max);
 	int (*commit_pagelist)(struct inode *inode, struct list_head *mds_pages, int how);
 
@@ -269,39 +269,42 @@ pnfs_commit_list(struct inode *inode, struct list_head *mds_pages, int how)
 	return NFS_SERVER(inode)->pnfs_curr_ld->commit_pagelist(inode, mds_pages, how);
 }
 
-static inline struct list_head *
-pnfs_choose_commit_list(struct nfs_page *req, struct pnfs_layout_segment *lseg)
+static inline bool
+pnfs_mark_request_commit(struct nfs_page *req, struct pnfs_layout_segment *lseg)
 {
 	struct inode *inode = req->wb_context->dentry->d_inode;
-	struct list_head *rv;
+	struct pnfs_layoutdriver_type *ld = NFS_SERVER(inode)->pnfs_curr_ld;
 
-	if (lseg && NFS_SERVER(inode)->pnfs_curr_ld->choose_commit_list)
-		rv = NFS_SERVER(inode)->pnfs_curr_ld->choose_commit_list(req, lseg);
-	else
-		rv = &NFS_I(inode)->commit_list;
-	return rv;
+	if (lseg == NULL || ld->mark_request_commit == NULL)
+		return false;
+	ld->mark_request_commit(req, lseg);
+	return true;
 }
 
-static inline struct pnfs_layout_segment *
+static inline bool
 pnfs_clear_request_commit(struct nfs_page *req)
 {
 	struct inode *inode = req->wb_context->dentry->d_inode;
+	struct pnfs_layoutdriver_type *ld = NFS_SERVER(inode)->pnfs_curr_ld;
 
-	if (NFS_SERVER(inode)->pnfs_curr_ld &&
-	    NFS_SERVER(inode)->pnfs_curr_ld->remove_commit_req)
-		return NFS_SERVER(inode)->pnfs_curr_ld->remove_commit_req(req);
-	else
-		return NULL;
+	if (ld == NULL || ld->clear_request_commit == NULL)
+		return false;
+	ld->clear_request_commit(req);
+	return true;
 }
 
 static inline int
 pnfs_scan_commit_lists(struct inode *inode, int max)
 {
-	if (NFS_SERVER(inode)->pnfs_curr_ld &&
-	    NFS_SERVER(inode)->pnfs_curr_ld->scan_commit_lists)
-		return NFS_SERVER(inode)->pnfs_curr_ld->scan_commit_lists(inode, max);
-	else
+	struct pnfs_layoutdriver_type *ld = NFS_SERVER(inode)->pnfs_curr_ld;
+	int ret;
+
+	if (ld == NULL || ld->scan_commit_lists == NULL)
 		return 0;
+	ret = ld->scan_commit_lists(inode, max);
+	if (ret != 0)
+		set_bit(NFS_INO_PNFS_COMMIT, &NFS_I(inode)->flags);
+	return ret;
 }
 
 /* Should the pNFS client commit and return the layout upon a setattr */
@@ -403,18 +406,16 @@ pnfs_commit_list(struct inode *inode, struct list_head *mds_pages, int how)
 	return PNFS_NOT_ATTEMPTED;
 }
 
-static inline struct list_head *
-pnfs_choose_commit_list(struct nfs_page *req, struct pnfs_layout_segment *lseg)
+static inline bool
+pnfs_mark_request_commit(struct nfs_page *req, struct pnfs_layout_segment *lseg)
 {
-	struct inode *inode = req->wb_context->dentry->d_inode;
-
-	return &NFS_I(inode)->commit_list;
+	return false;
 }
 
-static inline struct pnfs_layout_segment *
+static inline bool
 pnfs_clear_request_commit(struct nfs_page *req)
 {
-	return NULL;
+	return false;
 }
 
 static inline int

fs/nfs/read.c

@@ -66,7 +66,6 @@ void nfs_readdata_free(struct nfs_read_data *p)
 
 void nfs_readdata_release(struct nfs_read_data *rdata)
 {
-	put_lseg(rdata->lseg);
 	put_nfs_open_context(rdata->args.context);
 	nfs_readdata_free(rdata);
 }