Skip to contentSkip to conversation

pNFS: Mark layout for return if return-on-close was not sent

PATCH Superseded [email protected] 2 participants 5 messages Jan 5, 2021 – Jan 6, 2021

Subsystems (2)

v1 of 2 Superseded Latest is v2 →

lore
Share & tools
lore

[email protected]

author maintainer
#0 lore raw

PATCH 1/4 pNFS: Mark layout for return if return-on-close was not sent

From: Trond Myklebust <[email protected]>

If the layout return-on-close failed because the layoutreturn was never sent, then we should mark the layout for return again.

Fixes 9c47b18cf722 ("pNFS: Ensure we do clear the return-on-close layout stateid on fatal errors")
---
 fs/nfs/pnfs.c | 6 ++++++
 1 file changed, 6 insertions(+)
fs/nfs/pnfs.c +6 -0
index 07f59dc8cb2e..ccc89fab1802 100644
@@ -1560,12 +1560,18 @@ void pnfs_roc_release(struct nfs4_layoutreturn_args *args,
 		int ret)
 {
 	struct pnfs_layout_hdr *lo = args->layout;
+	struct inode *inode = args->inode;
 	const nfs4_stateid *arg_stateid = NULL;
 	const nfs4_stateid *res_stateid = NULL;
 	struct nfs4_xdr_opaque_data *ld_private = args->ld_private;
 
 	switch (ret) {
 	case -NFS4ERR_NOMATCHING_LAYOUT:
+		spin_lock(&inode->i_lock);
+		if (pnfs_layout_is_valid(lo) &&
+		    nfs4_stateid_match_other(&args->stateid, &lo->plh_stateid))
+			pnfs_set_plh_return_info(lo, args->range.iomode, 0);
+		spin_unlock(&inode->i_lock);
 		break;
 	case 0:
 		if (res->lrs_present)
Signature · 1 line
2.29.2
author maintainer
#1 lore raw

PATCH 2/4 pNFS: We want return-on-close to complete when evicting the inode

From: Trond Myklebust <[email protected]>

If the inode is being evicted, it should be safe to run return-on-close, so we should do it to ensure we don't inadvertently leak layout segments.

Fixes 1c5bd76d17cc ("pNFS: Enable layoutreturn operation for return-on-close")
---
 fs/nfs/nfs4proc.c | 26 ++++++++++----------------
 fs/nfs/pnfs.c     |  8 +++-----
 fs/nfs/pnfs.h     |  6 ++----
 3 files changed, 15 insertions(+), 25 deletions(-)
fs/nfs/nfs4proc.c +10 -16
index 0ce04e0e5d82..bbca5c46e400 100644
@@ -3536,10 +3536,8 @@ static void nfs4_close_done(struct rpc_task *task, void *data)
 	trace_nfs4_close(state, &calldata->arg, &calldata->res, task->tk_status);
 
 	/* Handle Layoutreturn errors */
-	if (pnfs_roc_done(task, calldata->inode,
-				&calldata->arg.lr_args,
-				&calldata->res.lr_res,
-				&calldata->res.lr_ret) == -EAGAIN)
+	if (pnfs_roc_done(task, &calldata->arg.lr_args, &calldata->res.lr_res,
+			  &calldata->res.lr_ret) == -EAGAIN)
 		goto out_restart;
 
 	/* hmm. we are done with the inode, and in the process of freeing
@@ -6384,10 +6382,8 @@ static void nfs4_delegreturn_done(struct rpc_task *task, void *calldata)
 	trace_nfs4_delegreturn_exit(&data->args, &data->res, task->tk_status);
 
 	/* Handle Layoutreturn errors */
-	if (pnfs_roc_done(task, data->inode,
-				&data->args.lr_args,
-				&data->res.lr_res,
-				&data->res.lr_ret) == -EAGAIN)
+	if (pnfs_roc_done(task, &data->args.lr_args, &data->res.lr_res,
+			  &data->res.lr_ret) == -EAGAIN)
 		goto out_restart;
 
 	switch (task->tk_status) {
@@ -6441,10 +6437,10 @@ static void nfs4_delegreturn_release(void *calldata)
 	struct nfs4_delegreturndata *data = calldata;
 	struct inode *inode = data->inode;
 
+	if (data->lr.roc)
+		pnfs_roc_release(&data->lr.arg, &data->lr.res,
+				 data->res.lr_ret);
 	if (inode) {
-		if (data->lr.roc)
-			pnfs_roc_release(&data->lr.arg, &data->lr.res,
-					data->res.lr_ret);
 		nfs_post_op_update_inode_force_wcc(inode, &data->fattr);
 		nfs_iput_and_deactive(inode);
 	}
@@ -6520,16 +6516,14 @@ static int _nfs4_proc_delegreturn(struct inode *inode, const struct cred *cred,
 	nfs_fattr_init(data->res.fattr);
 	data->timestamp = jiffies;
 	data->rpc_status = 0;
-	data->lr.roc = pnfs_roc(inode, &data->lr.arg, &data->lr.res, cred);
 	data->inode = nfs_igrab_and_active(inode);
-	if (data->inode) {
+	if (data->inode || issync) {
+		data->lr.roc = pnfs_roc(inode, &data->lr.arg, &data->lr.res,
+					cred);
 		if (data->lr.roc) {
 			data->args.lr_args = &data->lr.arg;
 			data->res.lr_res = &data->lr.res;
 		}
-	} else if (data->lr.roc) {
-		pnfs_roc_release(&data->lr.arg, &data->lr.res, 0);
-		data->lr.roc = false;
 	}
 
 	task_setup_data.callback_data = data;
fs/nfs/pnfs.c +3 -5
index ccc89fab1802..a18b1992b2fb 100644
@@ -1509,10 +1509,8 @@ bool pnfs_roc(struct inode *ino,
 	return false;
 }
 
-int pnfs_roc_done(struct rpc_task *task, struct inode *inode,
-		struct nfs4_layoutreturn_args **argpp,
-		struct nfs4_layoutreturn_res **respp,
-		int *ret)
+int pnfs_roc_done(struct rpc_task *task, struct nfs4_layoutreturn_args **argpp,
+		  struct nfs4_layoutreturn_res **respp, int *ret)
 {
 	struct nfs4_layoutreturn_args *arg = *argpp;
 	int retval = -EAGAIN;
@@ -1545,7 +1543,7 @@ int pnfs_roc_done(struct rpc_task *task, struct inode *inode,
 		return 0;
 	case -NFS4ERR_OLD_STATEID:
 		if (!nfs4_layout_refresh_old_stateid(&arg->stateid,
-					&arg->range, inode))
+						     &arg->range, arg->inode))
 			break;
 		*ret = -NFS4ERR_NOMATCHING_LAYOUT;
 		return -EAGAIN;
fs/nfs/pnfs.h +2 -4
index bbd3de1025f2..4d5ee568411d 100644
@@ -297,10 +297,8 @@ bool pnfs_roc(struct inode *ino,
 		struct nfs4_layoutreturn_args *args,
 		struct nfs4_layoutreturn_res *res,
 		const struct cred *cred);
-int pnfs_roc_done(struct rpc_task *task, struct inode *inode,
-		struct nfs4_layoutreturn_args **argpp,
-		struct nfs4_layoutreturn_res **respp,
-		int *ret);
+int pnfs_roc_done(struct rpc_task *task, struct nfs4_layoutreturn_args **argpp,
+		  struct nfs4_layoutreturn_res **respp, int *ret);
 void pnfs_roc_release(struct nfs4_layoutreturn_args *args,
 		struct nfs4_layoutreturn_res *res,
 		int ret);
Signature · 1 line
2.29.2
author maintainer
#2 lore raw

PATCH 3/4 pNFS: Clean up pnfs_layoutreturn_free_lsegs()

From: Trond Myklebust <[email protected]>

Remove the check for whether or not the stateid is NULL, and fix up the callers.

---
 fs/nfs/pnfs.c | 8 +++-----
 1 file changed, 3 insertions(+), 5 deletions(-)
fs/nfs/pnfs.c +3 -5
index a18b1992b2fb..a0d8a67a4dd9 100644
@@ -1152,7 +1152,7 @@ void pnfs_layoutreturn_free_lsegs(struct pnfs_layout_hdr *lo,
 	LIST_HEAD(freeme);
 
 	spin_lock(&inode->i_lock);
-	if (!pnfs_layout_is_valid(lo) || !arg_stateid ||
+	if (!pnfs_layout_is_valid(lo) ||
 	    !nfs4_stateid_match_other(&lo->plh_stateid, arg_stateid))
 		goto out_unlock;
 	if (stateid) {
@@ -1559,7 +1559,6 @@ void pnfs_roc_release(struct nfs4_layoutreturn_args *args,
 {
 	struct pnfs_layout_hdr *lo = args->layout;
 	struct inode *inode = args->inode;
-	const nfs4_stateid *arg_stateid = NULL;
 	const nfs4_stateid *res_stateid = NULL;
 	struct nfs4_xdr_opaque_data *ld_private = args->ld_private;
 
@@ -1576,11 +1575,10 @@ void pnfs_roc_release(struct nfs4_layoutreturn_args *args,
 			res_stateid = &res->stateid;
 		fallthrough;
 	default:
-		arg_stateid = &args->stateid;
+		pnfs_layoutreturn_free_lsegs(lo, &args->stateid, &args->range,
+					     res_stateid);
 	}
 	trace_nfs4_layoutreturn_on_close(args->inode, &args->stateid, ret);
-	pnfs_layoutreturn_free_lsegs(lo, arg_stateid, &args->range,
-			res_stateid);
 	if (ld_private && ld_private->ops && ld_private->ops->free)
 		ld_private->ops->free(ld_private);
 	pnfs_put_layout_hdr(lo);
Signature · 1 line
2.29.2
author maintainer
#3 lore raw

PATCH 4/4 pNFS: Stricter ordering of layoutget and layoutreturn

From: Trond Myklebust <[email protected]>

If a layout return is in progress, we should wait for it to complete, in case the layout segment we are picking up gets returned too.

Fixes 30cb3ee299cb ("pNFS: Handle NFS4ERR_OLD_STATEID on layoutreturn by bumping the state seqid")
---
 fs/nfs/pnfs.c | 43 +++++++++++++++++++++----------------------
 1 file changed, 21 insertions(+), 22 deletions(-)
fs/nfs/pnfs.c +21 -22
index a0d8a67a4dd9..e4a27afeee90 100644
@@ -2017,6 +2017,27 @@ pnfs_update_layout(struct inode *ino,
 		goto lookup_again;
 	}
 
+	/*
+	 * Because we free lsegs when sending LAYOUTRETURN, we need to wait
+	 * for LAYOUTRETURN.
+	 */
+	if (test_bit(NFS_LAYOUT_RETURN, &lo->plh_flags)) {
+		spin_unlock(&ino->i_lock);
+		dprintk("%s wait for layoutreturn\n", __func__);
+		lseg = ERR_PTR(pnfs_prepare_to_retry_layoutget(lo));
+		if (!IS_ERR(lseg)) {
+			pnfs_put_layout_hdr(lo);
+			dprintk("%s retrying\n", __func__);
+			trace_pnfs_update_layout(ino, pos, count, iomode, lo,
+						 lseg,
+						 PNFS_UPDATE_LAYOUT_RETRY);
+			goto lookup_again;
+		}
+		trace_pnfs_update_layout(ino, pos, count, iomode, lo, lseg,
+					 PNFS_UPDATE_LAYOUT_RETURN);
+		goto out_put_layout_hdr;
+	}
+
 	lseg = pnfs_find_lseg(lo, &arg, strict_iomode);
 	if (lseg) {
 		trace_pnfs_update_layout(ino, pos, count, iomode, lo, lseg,
@@ -2069,28 +2090,6 @@ pnfs_update_layout(struct inode *ino,
 		nfs4_stateid_copy(&stateid, &lo->plh_stateid);
 	}
 
-	/*
-	 * Because we free lsegs before sending LAYOUTRETURN, we need to wait
-	 * for LAYOUTRETURN even if first is true.
-	 */
-	if (test_bit(NFS_LAYOUT_RETURN, &lo->plh_flags)) {
-		spin_unlock(&ino->i_lock);
-		dprintk("%s wait for layoutreturn\n", __func__);
-		lseg = ERR_PTR(pnfs_prepare_to_retry_layoutget(lo));
-		if (!IS_ERR(lseg)) {
-			if (first)
-				pnfs_clear_first_layoutget(lo);
-			pnfs_put_layout_hdr(lo);
-			dprintk("%s retrying\n", __func__);
-			trace_pnfs_update_layout(ino, pos, count, iomode, lo,
-					lseg, PNFS_UPDATE_LAYOUT_RETRY);
-			goto lookup_again;
-		}
-		trace_pnfs_update_layout(ino, pos, count, iomode, lo, lseg,
-				PNFS_UPDATE_LAYOUT_RETURN);
-		goto out_put_layout_hdr;
-	}
-
 	if (pnfs_layoutgets_blocked(lo)) {
 		trace_pnfs_update_layout(ino, pos, count, iomode, lo, lseg,
 				PNFS_UPDATE_LAYOUT_BLOCKED);
Signature · 1 line
2.29.2
#4 lore raw

Re: [PATCH 2/4] pNFS: We want return-on-close to complete when evicting the inode

In reply to [email protected]

Hi Trond,

Quoted 109 lines · On Tue, Jan 5, 2021 at 10:30 AM <[email protected]> wrote:

From: Trond Myklebust <[email protected]>

If the inode is being evicted, it should be safe to run return-on-close, so we should do it to ensure we don't inadvertently leak layout segments.

Fixes 1c5bd76d17cc ("pNFS: Enable layoutreturn operation for return-on-close")
---
 fs/nfs/nfs4proc.c | 26 ++++++++++----------------
 fs/nfs/pnfs.c     |  8 +++-----
 fs/nfs/pnfs.h     |  6 ++----
 3 files changed, 15 insertions(+), 25 deletions(-)
fs/nfs/nfs4proc.c +10 -16
index 0ce04e0e5d82..bbca5c46e400 100644
@@ -3536,10 +3536,8 @@ static void nfs4_close_done(struct rpc_task *task, void *data)
        trace_nfs4_close(state, &calldata->arg, &calldata->res, task->tk_status);
 
        /* Handle Layoutreturn errors */
-       if (pnfs_roc_done(task, calldata->inode,
-                               &calldata->arg.lr_args,
-                               &calldata->res.lr_res,
-                               &calldata->res.lr_ret) == -EAGAIN)
+       if (pnfs_roc_done(task, &calldata->arg.lr_args, &calldata->res.lr_res,
+                         &calldata->res.lr_ret) == -EAGAIN)
                goto out_restart;
 
        /* hmm. we are done with the inode, and in the process of freeing
@@ -6384,10 +6382,8 @@ static void nfs4_delegreturn_done(struct rpc_task *task, void *calldata)
        trace_nfs4_delegreturn_exit(&data->args, &data->res, task->tk_status);
 
        /* Handle Layoutreturn errors */
-       if (pnfs_roc_done(task, data->inode,
-                               &data->args.lr_args,
-                               &data->res.lr_res,
-                               &data->res.lr_ret) == -EAGAIN)
+       if (pnfs_roc_done(task, &data->args.lr_args, &data->res.lr_res,
+                         &data->res.lr_ret) == -EAGAIN)
                goto out_restart;
 
        switch (task->tk_status) {
@@ -6441,10 +6437,10 @@ static void nfs4_delegreturn_release(void *calldata)
        struct nfs4_delegreturndata *data = calldata;
        struct inode *inode = data->inode;
 
+       if (data->lr.roc)
+               pnfs_roc_release(&data->lr.arg, &data->lr.res,
+                                data->res.lr_ret);
        if (inode) {
-               if (data->lr.roc)
-                       pnfs_roc_release(&data->lr.arg, &data->lr.res,
-                                       data->res.lr_ret);
                nfs_post_op_update_inode_force_wcc(inode, &data->fattr);
                nfs_iput_and_deactive(inode);
        }
@@ -6520,16 +6516,14 @@ static int _nfs4_proc_delegreturn(struct inode *inode, const struct cred *cred,
        nfs_fattr_init(data->res.fattr);
        data->timestamp = jiffies;
        data->rpc_status = 0;
-       data->lr.roc = pnfs_roc(inode, &data->lr.arg, &data->lr.res, cred);
        data->inode = nfs_igrab_and_active(inode);
-       if (data->inode) {
+       if (data->inode || issync) {
+               data->lr.roc = pnfs_roc(inode, &data->lr.arg, &data->lr.res,
+                                       cred);
                if (data->lr.roc) {
                        data->args.lr_args = &data->lr.arg;
                        data->res.lr_res = &data->lr.res;
                }
-       } else if (data->lr.roc) {
-               pnfs_roc_release(&data->lr.arg, &data->lr.res, 0);
-               data->lr.roc = false;
        }
 
        task_setup_data.callback_data = data;
fs/nfs/pnfs.c +3 -5
index ccc89fab1802..a18b1992b2fb 100644
@@ -1509,10 +1509,8 @@ bool pnfs_roc(struct inode *ino,
        return false;
 }
 
-int pnfs_roc_done(struct rpc_task *task, struct inode *inode,
-               struct nfs4_layoutreturn_args **argpp,
-               struct nfs4_layoutreturn_res **respp,
-               int *ret)
+int pnfs_roc_done(struct rpc_task *task, struct nfs4_layoutreturn_args **argpp,
+                 struct nfs4_layoutreturn_res **respp, int *ret)
 {
        struct nfs4_layoutreturn_args *arg = *argpp;
        int retval = -EAGAIN;
@@ -1545,7 +1543,7 @@ int pnfs_roc_done(struct rpc_task *task, struct inode *inode,
                return 0;
        case -NFS4ERR_OLD_STATEID:
                if (!nfs4_layout_refresh_old_stateid(&arg->stateid,
-                                       &arg->range, inode))
+                                                    &arg->range, arg->inode))
                        break;
                *ret = -NFS4ERR_NOMATCHING_LAYOUT;
                return -EAGAIN;
fs/nfs/pnfs.h +0 -0
index bbd3de1025f2..4d5ee568411d 100644

fs/nfs/pnfs.h bool pnfs_roc(struct inode *ino, · line 301

Quoted 9 lines
fs/nfs/pnfs.h +2 -4
@@ -297,10 +297,8 @@ bool pnfs_roc(struct inode *ino,
                struct nfs4_layoutreturn_args *args,
                struct nfs4_layoutreturn_res *res,
                const struct cred *cred);
-int pnfs_roc_done(struct rpc_task *task, struct inode *inode,
-               struct nfs4_layoutreturn_args **argpp,
-               struct nfs4_layoutreturn_res **respp,
-               int *ret);
+int pnfs_roc_done(struct rpc_task *task, struct nfs4_layoutreturn_args **argpp,
+                 struct nfs4_layoutreturn_res **respp, int *ret);

I think you also need to make this change to the fallback definition (for CONFIG_NFS_V4=n)

Thanks,
Anna

Quoted 6 lines
fs/nfs/pnfs.h +0 -1
 void pnfs_roc_release(struct nfs4_layoutreturn_args *args,
                struct nfs4_layoutreturn_res *res,
                int ret);
--

2.29.2