Skip to contentSkip to conversation

multipathd: the local path change is not considered

PATCH In review Brian Bunker 3 participants 5 messages Jan 17, 2024 – Jan 18, 2024 Device mapper

lore
Share & tools
lore
#0 lore raw

PATCH 1/1 multipathd: the local path change is not considered

When update_prio is called, there is a check to see if the local path has a priority change. Then all the remaining paths are simliarly checked.

Only the result of paths not matching the local path's priority are considered in the calculation of 'changed'. In the case of failed paths becoming again available this can lead to multipath not reloading.

The result will look like this:
3624a93703c9f34490f6140a100011010 dm-2 PURE,FlashArray size=3.0T features='1 retain_attached_hw_handler' hwhandler='1 alua' wp=rw

|-+- policy='service-time 0' prio=50 status=active
| |- 1:0:0:1  sdb 8:16  active ready running
| |- 9:0:0:1  sdc 8:32  active ready running
| |- 11:0:0:1 sdd 8:48  active ready running
| `- 13:0:0:1 sdh 8:112 active ready running
`-+- policy='service-time 0' prio=50 status=enabled
  |- 8:0:0:1  sde 8:64  active ready running
  |- 10:0:0:1 sdf 8:80  active ready running
  |- 12:0:0:1 sdg 8:96  active ready running

`- 14:0:0:1 sdi 8:128 active ready running

Where the path's priorities are updated, but the path group is not activated.

---
 multipathd/main.c | 13 ++++++++-----
 1 file changed, 8 insertions(+), 5 deletions(-)
multipathd/main.c +8 -5
index 230c9d10..62b87bf8 100644
@@ -2127,7 +2127,7 @@ int update_prio(struct path *pp, int refresh_all)
 	int oldpriority;
 	struct path *pp1;
 	struct pathgroup * pgp;
-	int i, j, changed = 0;
+	int i, j, local_changed = 0, other_changed = 0;
 	struct config *conf;
 
 	oldpriority = pp->priority;
@@ -2138,8 +2138,11 @@ int update_prio(struct path *pp, int refresh_all)
 		pthread_cleanup_pop(1);
 	}
 
-	if (pp->priority == oldpriority && !refresh_all)
-		return 0;
+	if (pp->priority != oldpriority)
+		local_changed = 1;
+	else
+		if (!refresh_all)
+			return 0;
 
 	vector_foreach_slot (pp->mpp->pg, pgp, i) {
 		vector_foreach_slot (pgp->paths, pp1, j) {
@@ -2151,10 +2154,10 @@ int update_prio(struct path *pp, int refresh_all)
 			pathinfo(pp1, conf, DI_PRIO);
 			pthread_cleanup_pop(1);
 			if (pp1->priority != oldpriority)
-				changed = 1;
+				other_changed = 1;
 		}
 	}
-	return changed;
+	return (local_changed || other_changed);
 }
 
 static int reload_map(struct vectors *vecs, struct multipath *mpp,
Signature · 1 line
2.43.0
#1 lore raw
Quoted 27 lines · Brian Bunker

When update_prio is called, there is a check to see if the local path has a priority change. Then all the remaining paths are simliarly checked.

Only the result of paths not matching the local path's priority are considered in the calculation of 'changed'. In the case of failed paths
becoming again available this can lead to multipath not reloading.

The result will look like this:
3624a93703c9f34490f6140a100011010 dm-2 PURE,FlashArray size=3.0T features='1 retain_attached_hw_handler' hwhandler='1 alua' wp=rw

Quoted 5 lines

-+- policy='service-time 0' prio=50 status=active

Quoted 3 lines

- 1:0:0:1  sdb 8:16  active ready running
- 9:0:0:1  sdc 8:32  active ready running
- 11:0:0:1 sdd 8:48  active ready running

`- 13:0:0:1 sdh 8:112 active ready running

`-+- policy='service-time 0' prio=50 status=enabled

  |- 8:0:0:1  sde 8:64  active ready running
  |- 10:0:0:1 sdf 8:80  active ready running
  |- 12:0:0:1 sdg 8:96  active ready running

  `- 14:0:0:1 sdi 8:128 active ready running

Where the path's priorities are updated, but the path group is not activated.

Quoted 7 lines

---
 multipathd/main.c | 13 ++++++++-----

 1 file changed, 8 insertions(+), 5 deletions(-)

Thanks for the report and fix!

multipathd/main.c int update_prio(struct path *pp, int · line 2131

Quoted 11 lines
multipathd/main.c +1 -1
index 230c9d10..62b87bf8 100644
@@ -2127,7 +2127,7 @@ int update_prio(struct path *pp, int
 refresh_all)
  	int oldpriority;
  	struct path *pp1;
  	struct pathgroup * pgp;
-	int i, j, changed = 0;
+	int i, j, local_changed = 0, other_changed = 0;

Why use two variables here? AFAICS, you can implement the same logic with just the "changed" variable.

Other than that, this looks good.

When you submit a v2, please add [email protected] and myself to cc, and add

Fixes 6ccd7b8 ("multipathd: only refresh priorities in update_prio()")

Regards,
Martin

Quoted 32 lines
multipathd/main.c +0 -0
  	struct config *conf;


  oldpriority = pp->priority;

multipathd/main.c +0 -0
@@ -2138,8 +2138,11 @@ int update_prio(struct path *pp, int
 refresh_all)
  		pthread_cleanup_pop(1);
  	}

multipathd/main.c +5 -2
-	if (pp->priority == oldpriority && !refresh_all)
-		return 0;
+	if (pp->priority != oldpriority)
+		local_changed = 1;
+	else
+		if (!refresh_all)
+			return 0;


  vector_foreach_slot (pp->mpp->pg, pgp, i) { vector_foreach_slot (pgp->paths, pp1, j) {

multipathd/main.c +2 -2
@@ -2151,10 +2154,10 @@ int update_prio(struct path *pp, int
 refresh_all)
  			pathinfo(pp1, conf, DI_PRIO);
  			pthread_cleanup_pop(1);
  			if (pp1->priority != oldpriority)
-				changed = 1;
+				other_changed = 1;
  		}
  	}
-	return changed;
+	return (local_changed || other_changed);
  }


 static int reload_map(struct vectors *vecs, struct multipath *mpp,

#2 lore raw
Quoted 67 lines · Martin Wilck
Quoted 29 lines · Brian Bunker

When update_prio is called, there is a check to see if the local path
has a priority change. Then all the remaining paths are simliarly checked.

Only the result of paths not matching the local path's priority are considered in the calculation of 'changed'. In the case of failed paths
becoming again available this can lead to multipath not reloading.

The result will look like this:
3624a93703c9f34490f6140a100011010 dm-2 PURE,FlashArray size=3.0T features='1 retain_attached_hw_handler' hwhandler='1 alua'
wp=rw

Quoted 5 lines

-+- policy='service-time 0' prio=50 status=active

Quoted 3 lines

- 1:0:0:1  sdb 8:16  active ready running
- 9:0:0:1  sdc 8:32  active ready running
- 11:0:0:1 sdd 8:48  active ready running

`- 13:0:0:1 sdh 8:112 active ready running

`-+- policy='service-time 0' prio=50 status=enabled

  |- 8:0:0:1  sde 8:64  active ready running
  |- 10:0:0:1 sdf 8:80  active ready running
  |- 12:0:0:1 sdg 8:96  active ready running

  `- 14:0:0:1 sdi 8:128 active ready running

Where the path's priorities are updated, but the path group is not activated.

Quoted 7 lines

---
 multipathd/main.c | 13 ++++++++-----

 1 file changed, 8 insertions(+), 5 deletions(-)

Thanks for the report and fix!

multipathd/main.c int update_prio(struct path *pp, int · line 2131

Quoted 11 lines
multipathd/main.c +1 -1
index 230c9d10..62b87bf8 100644
@@ -2127,7 +2127,7 @@ int update_prio(struct path *pp, int
 refresh_all)
  	int oldpriority;
  	struct path *pp1;
  	struct pathgroup * pgp;
-	int i, j, changed = 0;
+	int i, j, local_changed = 0, other_changed = 0;

Why use two variables here? AFAICS, you can implement the same logic with just the "changed" variable.

Other than that, this looks good.

When you submit a v2, please add [email protected] and myself to cc,
and add

Fixes 6ccd7b8 ("multipathd: only refresh priorities in

update_prio()")

Regards,
Martin

Quoted 15 lines
Quoted 8 lines
multipathd/main.c +0 -0
  	struct config *conf;


  oldpriority = pp->priority;

multipathd/main.c +0 -0
@@ -2138,8 +2138,11 @@ int update_prio(struct path *pp, int
 refresh_all)
  		pthread_cleanup_pop(1);
  	}

multipathd/main.c

Quoted 7 lines
multipathd/main.c +5 -2
-	if (pp->priority == oldpriority && !refresh_all)
-		return 0;
+	if (pp->priority != oldpriority)
+		local_changed = 1;
+	else
+		if (!refresh_all)
+			return 0;

Just realized that this would unnecessarily run through the loop below if (pp->priority != oldpriority && !refresh_all).

IMO the patch should simply look like this:

multipathd/main.c +0 -0
index fbc3f8d..ca8e65c 100644
@@ -2115,8 +2115,10 @@ int update_prio(struct path *pp, int

refresh_all)

                pthread_cleanup_pop(1);
        }

- if (pp->priority == oldpriority && !refresh_all)
- return 0;
+ if (pp->priority != oldpriority)
+ changed = 1;
+ if (!refresh_all)
+ return changed;

        vector_foreach_slot (pp->mpp->pg, pgp, i) {
                vector_foreach_slot (pgp->paths, pp1, j) {

If you don't object, I'll apply it in this form, leaving you as author.

Martin

#3 lore raw
Quoted 12 lines · Martin Wilck
Quoted 12 lines
Quoted 12 lines

  struct config *conf;

  oldpriority = pp->priority;

(quoted diff) +0 -0
@@ -2138,8 +2138,11 @@ int update_prio(struct path *pp, int
 refresh_all)
  		pthread_cleanup_pop(1);
  	}

(quoted diff) +1 -2
-	if (pp->priority == oldpriority && !refresh_all)
-		return 0;
+	if (pp->priority != oldpriority)
Quoted 8 lines
Quoted 4 lines

Review

Quoted 4 lines
(quoted diff) +4 -0
+		local_changed = 1;
+	else
+		if (!refresh_all)
+			return 0;

Just realized that this would unnecessarily run through the loop below
if (pp->priority != oldpriority && !refresh_all).

Sorry, this was nonsense. The commit msg of 6ccd7b8 explains that we must run the loop if "changed" is true, even if refresh_all is not set.

Martin

#4 lore raw
In reply to Martin Wilck
Quoted 55 lines · Martin Wilck
Quoted 27 lines · Brian Bunker

When update_prio is called, there is a check to see if the local path has a priority change. Then all the remaining paths are simliarly checked.

Only the result of paths not matching the local path's priority are considered in the calculation of 'changed'. In the case of failed paths
becoming again available this can lead to multipath not reloading.

The result will look like this:
3624a93703c9f34490f6140a100011010 dm-2 PURE,FlashArray size=3.0T features='1 retain_attached_hw_handler' hwhandler='1 alua' wp=rw

Quoted 5 lines

-+- policy='service-time 0' prio=50 status=active

Quoted 3 lines

- 1:0:0:1  sdb 8:16  active ready running
- 9:0:0:1  sdc 8:32  active ready running
- 11:0:0:1 sdd 8:48  active ready running

`- 13:0:0:1 sdh 8:112 active ready running

`-+- policy='service-time 0' prio=50 status=enabled

  |- 8:0:0:1  sde 8:64  active ready running
  |- 10:0:0:1 sdf 8:80  active ready running
  |- 12:0:0:1 sdg 8:96  active ready running

  `- 14:0:0:1 sdi 8:128 active ready running

Where the path's priorities are updated, but the path group is not activated.

Quoted 7 lines

---
 multipathd/main.c | 13 ++++++++-----

 1 file changed, 8 insertions(+), 5 deletions(-)

Thanks for the report and fix!

multipathd/main.c int update_prio(struct path *pp, int · line 2131

Quoted 11 lines
multipathd/main.c +1 -1
index 230c9d10..62b87bf8 100644
@@ -2127,7 +2127,7 @@ int update_prio(struct path *pp, int
 refresh_all)
  	int oldpriority;
  	struct path *pp1;
  	struct pathgroup * pgp;
-	int i, j, changed = 0;
+	int i, j, local_changed = 0, other_changed = 0;

Why use two variables here? AFAICS, you can implement the same logic with just the "changed" variable.

Other than that, this looks good.

Quoted 7 lines

When you submit a v2, please add [email protected] and myself to cc, and add

Fixes 6ccd7b8 ("multipathd: only refresh priorities in update_prio()")

Regards,
Martin

ACK for the code with using "changed" for both the local and other changes.

Thanks for catching this.

-Ben

Quoted 33 lines
Quoted 32 lines
multipathd/main.c +0 -0
  	struct config *conf;


  oldpriority = pp->priority;

multipathd/main.c +0 -0
@@ -2138,8 +2138,11 @@ int update_prio(struct path *pp, int
 refresh_all)
  		pthread_cleanup_pop(1);
  	}

multipathd/main.c +5 -2
-	if (pp->priority == oldpriority && !refresh_all)
-		return 0;
+	if (pp->priority != oldpriority)
+		local_changed = 1;
+	else
+		if (!refresh_all)
+			return 0;


  vector_foreach_slot (pp->mpp->pg, pgp, i) { vector_foreach_slot (pgp->paths, pp1, j) {

multipathd/main.c +2 -2
@@ -2151,10 +2154,10 @@ int update_prio(struct path *pp, int
 refresh_all)
  			pathinfo(pp1, conf, DI_PRIO);
  			pthread_cleanup_pop(1);
  			if (pp1->priority != oldpriority)
-				changed = 1;
+				other_changed = 1;
  		}
  	}
-	return changed;
+	return (local_changed || other_changed);
  }


 static int reload_map(struct vectors *vecs, struct multipath *mpp,