Skip to contentSkip to conversation

ceph: init the i_list/g_list for cap flush

PATCH In review [email protected] 3 participants 4 messages Aug 25, 2021 – Aug 26, 2021

Subsystems (2)
lore
Share & tools
lore
#0 lore raw

PATCH 1/1 ceph: init the i_list/g_list for cap flush

From: Xiubo Li <[email protected]>

Always init the i_list/g_list in the begining to make sure it won't crash the kernel if someone want to delete the cap_flush from the lists.

Cc: [email protected]
URL: https://tracker.ceph.com/issues/52401

---
 fs/ceph/caps.c | 2 +-
 fs/ceph/snap.c | 2 ++
 2 files changed, 3 insertions(+), 1 deletion(-)
fs/ceph/caps.c +1 -1
index 4f0dbc640b0b..60f60260cf42 100644
@@ -3666,7 +3666,7 @@ static void handle_cap_flush_ack(struct inode *inode, u64 flush_tid,
 	while (!list_empty(&to_remove)) {
 		cf = list_first_entry(&to_remove,
 				      struct ceph_cap_flush, i_list);
-		list_del(&cf->i_list);
+		list_del_init(&cf->i_list);
 		if (!cf->is_capsnap)
 			ceph_free_cap_flush(cf);
 	}
fs/ceph/snap.c +2 -0
index 62fab59bbf96..b41e6724c591 100644
@@ -488,6 +488,8 @@ static void ceph_queue_cap_snap(struct ceph_inode_info *ci)
 		return;
 	}
 	capsnap->cap_flush.is_capsnap = true;
+	INIT_LIST_HEAD(&capsnap->cap_flush.i_list);
+	INIT_LIST_HEAD(&capsnap->cap_flush.g_list);
 
 	spin_lock(&ci->i_ceph_lock);
 	used = __ceph_caps_used(ci);
Signature · 1 line
2.27.0
#1 lore raw
Quoted 32 lines · On Wed, 2021-08-25 at 13:22 +0800, [email protected] wrote:

From: Xiubo Li <[email protected]>

Always init the i_list/g_list in the begining to make sure it won't crash the kernel if someone want to delete the cap_flush from the lists.

Cc: [email protected]
URL: https://tracker.ceph.com/issues/52401

---
 fs/ceph/caps.c | 2 +-
 fs/ceph/snap.c | 2 ++
 2 files changed, 3 insertions(+), 1 deletion(-)
fs/ceph/caps.c +1 -1
index 4f0dbc640b0b..60f60260cf42 100644
@@ -3666,7 +3666,7 @@ static void handle_cap_flush_ack(struct inode *inode, u64 flush_tid,
 	while (!list_empty(&to_remove)) {
 		cf = list_first_entry(&to_remove,
 				      struct ceph_cap_flush, i_list);
-		list_del(&cf->i_list);
+		list_del_init(&cf->i_list);
 		if (!cf->is_capsnap)
 			ceph_free_cap_flush(cf);
 	}
fs/ceph/snap.c +0 -0
index 62fab59bbf96..b41e6724c591 100644

fs/ceph/snap.c static void ceph_queue_cap_snap(struct ceph_inode_info *ci) · line 495

Quoted 8 lines
fs/ceph/snap.c +2 -0
@@ -488,6 +488,8 @@ static void ceph_queue_cap_snap(struct ceph_inode_info *ci)
 		return;
 	}
 	capsnap->cap_flush.is_capsnap = true;
+	INIT_LIST_HEAD(&capsnap->cap_flush.i_list);
+	INIT_LIST_HEAD(&capsnap->cap_flush.g_list);
 
 	spin_lock(&ci->i_ceph_lock);
 	used = __ceph_caps_used(ci);

I'm not certain the second hunk is strictly needed. These either end up on the list or they just get freed. That said, they shouldn't hurt anything and it is more consistent. Merged into testing.

Ilya, since this is marked for stable, this probably ought to go to Linus in the last v5.14 pile.

Thanks,

Signature · 1 line
Jeff Layton <[email protected]>
maintainer
#2 lore raw
Quoted 47 lines · Jeff Layton
Quoted 32 lines · On Wed, 2021-08-25 at 13:22 +0800, [email protected] wrote:

From: Xiubo Li <[email protected]>

Always init the i_list/g_list in the begining to make sure it won't crash the kernel if someone want to delete the cap_flush from the lists.

Cc: [email protected]
URL: https://tracker.ceph.com/issues/52401

---
 fs/ceph/caps.c | 2 +-
 fs/ceph/snap.c | 2 ++
 2 files changed, 3 insertions(+), 1 deletion(-)
fs/ceph/caps.c +1 -1
index 4f0dbc640b0b..60f60260cf42 100644
@@ -3666,7 +3666,7 @@ static void handle_cap_flush_ack(struct inode *inode, u64 flush_tid,
      while (!list_empty(&to_remove)) {
              cf = list_first_entry(&to_remove,
                                    struct ceph_cap_flush, i_list);
-             list_del(&cf->i_list);
+             list_del_init(&cf->i_list);
              if (!cf->is_capsnap)
                      ceph_free_cap_flush(cf);
      }
fs/ceph/snap.c +0 -0
index 62fab59bbf96..b41e6724c591 100644

fs/ceph/snap.c static void ceph_queue_cap_snap(struct ceph_inode_info *ci) · line 495

Quoted 8 lines
fs/ceph/snap.c +2 -0
@@ -488,6 +488,8 @@ static void ceph_queue_cap_snap(struct ceph_inode_info *ci)
              return;
      }
      capsnap->cap_flush.is_capsnap = true;
+     INIT_LIST_HEAD(&capsnap->cap_flush.i_list);
+     INIT_LIST_HEAD(&capsnap->cap_flush.g_list);
 
      spin_lock(&ci->i_ceph_lock);
      used = __ceph_caps_used(ci);

I'm not certain the second hunk is strictly needed. These either end up on the list or they just get freed. That said, they shouldn't hurt anything and it is more consistent. Merged into testing.

Quoted 2 lines

Ilya, since this is marked for stable, this probably ought to go to Linus in the last v5.14 pile.

I'm inclined to fold this into "ceph: correctly handle releasing an embedded cap flush" which is already queued for 5.14.

Thanks,

Ilya

author
#3 lore raw
Quoted 46 lines · Ilya Dryomov
Quoted 45 lines · Jeff Layton
Quoted 32 lines · On Wed, 2021-08-25 at 13:22 +0800, [email protected] wrote:

From: Xiubo Li <[email protected]>

Always init the i_list/g_list in the begining to make sure it won't crash the kernel if someone want to delete the cap_flush from the lists.

Cc: [email protected]
URL: https://tracker.ceph.com/issues/52401

---
  fs/ceph/caps.c | 2 +-
  fs/ceph/snap.c | 2 ++
  2 files changed, 3 insertions(+), 1 deletion(-)
fs/ceph/caps.c +1 -1
index 4f0dbc640b0b..60f60260cf42 100644
@@ -3666,7 +3666,7 @@ static void handle_cap_flush_ack(struct inode *inode, u64 flush_tid,
       while (!list_empty(&to_remove)) {
               cf = list_first_entry(&to_remove,
                                     struct ceph_cap_flush, i_list);
-             list_del(&cf->i_list);
+             list_del_init(&cf->i_list);
               if (!cf->is_capsnap)
                       ceph_free_cap_flush(cf);
       }
fs/ceph/snap.c +0 -0
index 62fab59bbf96..b41e6724c591 100644

fs/ceph/snap.c static void ceph_queue_cap_snap(struct ceph_inode_info *ci) · line 495

Quoted 8 lines
fs/ceph/snap.c +2 -0
@@ -488,6 +488,8 @@ static void ceph_queue_cap_snap(struct ceph_inode_info *ci)
               return;
       }
       capsnap->cap_flush.is_capsnap = true;
+     INIT_LIST_HEAD(&capsnap->cap_flush.i_list);
+     INIT_LIST_HEAD(&capsnap->cap_flush.g_list);
 
       spin_lock(&ci->i_ceph_lock);
       used = __ceph_caps_used(ci);

I'm not certain the second hunk is strictly needed. These either end up on the list or they just get freed. That said, they shouldn't hurt anything and it is more consistent. Merged into testing.

Quoted 4 lines
Quoted 2 lines

Ilya, since this is marked for stable, this probably ought to go to Linus in the last v5.14 pile.

I'm inclined to fold this into "ceph: correctly handle releasing an embedded cap flush" which is already queued for 5.14.

Yeah, that's will be better.

Quoted 5 lines

Thanks,

Ilya