pnfs: only tear down lsegs that precede seqid in LAYOUTRETURN args

LAYOUTRETURN is "special" in that servers and clients are expected to
work with old stateids. When the client sends a LAYOUTRETURN with an old
stateid in it then the server is expected to only tear down layout
segments that were present when that seqid was current. Ensure that the
client handles its accounting accordingly.

Signed-off-by: Jeff Layton <jeff.layton@primarydata.com>
Signed-off-by: Anna Schumaker <Anna.Schumaker@Netapp.com>

1 files changed, 42 insertions, 22 deletions

diff --git a/fs/nfs/pnfs.c b/fs/nfs/pnfs.c
index 3ad6bd377d35..21ff53fe7a9e 100644
--- a/fs/nfs/pnfs.c
+++ b/fs/nfs/pnfs.c
@@ -270,7 +270,7 @@ pnfs_mark_layout_stateid_invalid(struct pnfs_layout_hdr *lo,
 	};
 
 	set_bit(NFS_LAYOUT_INVALID_STID, &lo->plh_flags);
-	return pnfs_mark_matching_lsegs_invalid(lo, lseg_list, &range);
+	return pnfs_mark_matching_lsegs_invalid(lo, lseg_list, &range, 0);
 }
 
 static int
@@ -308,7 +308,7 @@ pnfs_layout_io_set_failed(struct pnfs_layout_hdr *lo, u32 iomode)
 
 	spin_lock(&inode->i_lock);
 	pnfs_layout_set_fail_bit(lo, pnfs_iomode_to_fail_bit(iomode));
-	pnfs_mark_matching_lsegs_invalid(lo, &head, &range);
+	pnfs_mark_matching_lsegs_invalid(lo, &head, &range, 0);
 	spin_unlock(&inode->i_lock);
 	pnfs_free_lseg_list(&head);
 	dprintk("%s Setting layout IOMODE_%s fail bit\n", __func__,
@@ -522,13 +522,35 @@ static int mark_lseg_invalid(struct pnfs_layout_segment *lseg,
 	return rv;
 }
 
-/* Returns count of number of matching invalid lsegs remaining in list
- * after call.
+/*
+ * Compare 2 layout stateid sequence ids, to see which is newer,
+ * taking into account wraparound issues.
+ */
+static bool pnfs_seqid_is_newer(u32 s1, u32 s2)
+{
+	return (s32)(s1 - s2) > 0;
+}
+
+/**
+ * pnfs_mark_matching_lsegs_invalid - tear down lsegs or mark them for later
+ * @lo: layout header containing the lsegs
+ * @tmp_list: list head where doomed lsegs should go
+ * @recall_range: optional recall range argument to match (may be NULL)
+ * @seq: only invalidate lsegs obtained prior to this sequence (may be 0)
+ *
+ * Walk the list of lsegs in the layout header, and tear down any that should
+ * be destroyed. If "recall_range" is specified then the segment must match
+ * that range. If "seq" is non-zero, then only match segments that were handed
+ * out at or before that sequence.
+ *
+ * Returns number of matching invalid lsegs remaining in list after scanning
+ * it and purging them.
  */
 int
 pnfs_mark_matching_lsegs_invalid(struct pnfs_layout_hdr *lo,
 			    struct list_head *tmp_list,
-			    const struct pnfs_layout_range *recall_range)
+			    const struct pnfs_layout_range *recall_range,
+			    u32 seq)
 {
 	struct pnfs_layout_segment *lseg, *next;
 	int remaining = 0;
@@ -540,10 +562,12 @@ pnfs_mark_matching_lsegs_invalid(struct pnfs_layout_hdr *lo,
 	list_for_each_entry_safe(lseg, next, &lo->plh_segs, pls_list)
 		if (!recall_range ||
 		    should_free_lseg(&lseg->pls_range, recall_range)) {
-			dprintk("%s: freeing lseg %p iomode %d "
+			if (seq && pnfs_seqid_is_newer(lseg->pls_seq, seq))
+				continue;
+			dprintk("%s: freeing lseg %p iomode %d seq %u"
 				"offset %llu length %llu\n", __func__,
-				lseg, lseg->pls_range.iomode, lseg->pls_range.offset,
-				lseg->pls_range.length);
+				lseg, lseg->pls_range.iomode, lseg->pls_seq,
+				lseg->pls_range.offset, lseg->pls_range.length);
 			if (!mark_lseg_invalid(lseg, tmp_list))
 				remaining++;
 		}
@@ -730,15 +754,6 @@ pnfs_destroy_all_layouts(struct nfs_client *clp)
 	pnfs_destroy_layouts_byclid(clp, false);
 }
 
-/*
- * Compare 2 layout stateid sequence ids, to see which is newer,
- * taking into account wraparound issues.
- */
-static bool pnfs_seqid_is_newer(u32 s1, u32 s2)
-{
-	return (s32)(s1 - s2) > 0;
-}
-
 /* update lo->plh_stateid with new if is more recent */
 void
 pnfs_set_layout_stateid(struct pnfs_layout_hdr *lo, const nfs4_stateid *new,
@@ -1014,7 +1029,7 @@ _pnfs_return_layout(struct inode *ino)
 	pnfs_get_layout_hdr(lo);
 	empty = list_empty(&lo->plh_segs);
 	pnfs_clear_layoutcommit(ino, &tmp_list);
-	pnfs_mark_matching_lsegs_invalid(lo, &tmp_list, NULL);
+	pnfs_mark_matching_lsegs_invalid(lo, &tmp_list, NULL, 0);
 
 	if (NFS_SERVER(ino)->pnfs_curr_ld->return_range) {
 		struct pnfs_layout_range range = {
@@ -1721,7 +1736,7 @@ pnfs_layout_process(struct nfs4_layoutget *lgp)
 		 * inode invalid, and don't bother validating the stateid
 		 * sequence number.
 		 */
-		pnfs_mark_matching_lsegs_invalid(lo, &free_me, NULL);
+		pnfs_mark_matching_lsegs_invalid(lo, &free_me, NULL, 0);
 
 		nfs4_stateid_copy(&lo->plh_stateid, &res->stateid);
 		lo->plh_barrier = be32_to_cpu(res->stateid.seqid);
@@ -1775,7 +1790,8 @@ pnfs_set_plh_return_info(struct pnfs_layout_hdr *lo, enum pnfs_iomode iomode,
 int
 pnfs_mark_matching_lsegs_return(struct pnfs_layout_hdr *lo,
 				struct list_head *tmp_list,
-				const struct pnfs_layout_range *return_range)
+				const struct pnfs_layout_range *return_range,
+				u32 seq)
 {
 	struct pnfs_layout_segment *lseg, *next;
 	int remaining = 0;
@@ -1798,8 +1814,11 @@ pnfs_mark_matching_lsegs_return(struct pnfs_layout_hdr *lo,
 				continue;
 			remaining++;
 			set_bit(NFS_LSEG_LAYOUTRETURN, &lseg->pls_flags);
-			pnfs_set_plh_return_info(lo, return_range->iomode, lseg->pls_seq);
 		}
+
+	if (remaining)
+		pnfs_set_plh_return_info(lo, return_range->iomode, seq);
+
 	return remaining;
 }
 
@@ -1822,7 +1841,8 @@ void pnfs_error_mark_layout_for_return(struct inode *inode,
 	 * segments at hand when sending layoutreturn. See pnfs_put_lseg()
 	 * for how it works.
 	 */
-	if (!pnfs_mark_matching_lsegs_return(lo, &free_me, &range)) {
+	if (!pnfs_mark_matching_lsegs_return(lo, &free_me,
+						&range, lseg->pls_seq)) {
 		nfs4_stateid stateid;
 		enum pnfs_iomode iomode = lo->plh_return_iomode;