diff mbox series

[v3,1/2] ceph: always queue a writeback when revoking the Fb caps

Message ID 20240117042758.700349-2-xiubli@redhat.com
State New
Headers show
Series ceph: fix caps revocation stuck | expand

Commit Message

Xiubo Li Jan. 17, 2024, 4:27 a.m. UTC 
From: Xiubo Li <xiubli@redhat.com>

In case there is 'Fw' dirty caps and 'CHECK_CAPS_FLUSH' is set we
will always ignore queue a writeback. Queue a writeback is very
important because it will block kclient flushing the snapcaps to
MDS and which will block MDS waiting for revoking the 'Fb' caps.

URL: https://tracker.ceph.com/issues/50223
Signed-off-by: Xiubo Li <xiubli@redhat.com>
---
 fs/ceph/caps.c | 48 ++++++++++++++++++++++++------------------------
 1 file changed, 24 insertions(+), 24 deletions(-)
diff mbox series

Patch

diff --git a/fs/ceph/caps.c b/fs/ceph/caps.c
index b8a8e0d3c6b7..c0db0e9e82d2 100644
--- a/fs/ceph/caps.c
+++ b/fs/ceph/caps.c
@@ -2156,6 +2156,30 @@  void ceph_check_caps(struct ceph_inode_info *ci, int flags)
 		      ceph_cap_string(cap->implemented),
 		      ceph_cap_string(revoking));
 
+		/* completed revocation? going down and there are no caps? */
+		if (revoking) {
+			if ((revoking & cap_used) == 0) {
+				doutc(cl, "completed revocation of %s\n",
+				      ceph_cap_string(cap->implemented & ~cap->issued));
+				goto ack;
+			}
+
+			/*
+			 * If the "i_wrbuffer_ref" was increased by mmap or generic
+			 * cache write just before the ceph_check_caps() is called,
+			 * the Fb capability revoking will fail this time. Then we
+			 * must wait for the BDI's delayed work to flush the dirty
+			 * pages and to release the "i_wrbuffer_ref", which will cost
+			 * at most 5 seconds. That means the MDS needs to wait at
+			 * most 5 seconds to finished the Fb capability's revocation.
+			 *
+			 * Let's queue a writeback for it.
+			 */
+			if (S_ISREG(inode->i_mode) && ci->i_wrbuffer_ref &&
+			    (revoking & CEPH_CAP_FILE_BUFFER))
+				queue_writeback = true;
+		}
+
 		if (cap == ci->i_auth_cap &&
 		    (cap->issued & CEPH_CAP_FILE_WR)) {
 			/* request larger max_size from MDS? */
@@ -2183,30 +2207,6 @@  void ceph_check_caps(struct ceph_inode_info *ci, int flags)
 			}
 		}
 
-		/* completed revocation? going down and there are no caps? */
-		if (revoking) {
-			if ((revoking & cap_used) == 0) {
-				doutc(cl, "completed revocation of %s\n",
-				      ceph_cap_string(cap->implemented & ~cap->issued));
-				goto ack;
-			}
-
-			/*
-			 * If the "i_wrbuffer_ref" was increased by mmap or generic
-			 * cache write just before the ceph_check_caps() is called,
-			 * the Fb capability revoking will fail this time. Then we
-			 * must wait for the BDI's delayed work to flush the dirty
-			 * pages and to release the "i_wrbuffer_ref", which will cost
-			 * at most 5 seconds. That means the MDS needs to wait at
-			 * most 5 seconds to finished the Fb capability's revocation.
-			 *
-			 * Let's queue a writeback for it.
-			 */
-			if (S_ISREG(inode->i_mode) && ci->i_wrbuffer_ref &&
-			    (revoking & CEPH_CAP_FILE_BUFFER))
-				queue_writeback = true;
-		}
-
 		/* want more caps from mds? */
 		if (want & ~cap->mds_wanted) {
 			if (want & ~(cap->mds_wanted | cap->issued))