Skip to contentSkip to conversation

scsi: two fixes in scsi_add_host_with_dma

PATCH Superseded Ming Lei 3 participants 14 messages May 31, 2021 – Jun 2, 2021 SCSI

lore
Share & tools
lore

Ming Lei

author
#0 lore raw

v3 cover scsi: two fixes in scsi_add_host_with_dma

Hello Martin,

Fix two memory leaks and one double free issue in alloc/add host code path.

V3:
- fix memory leak caused in scsi_host_alloc
- comment typo suggested by John

V2:
- add patch 2 for addressing shost leak in case of adding host failure, reported by John Garry.

Ming Lei (3):
scsi: core: use put_device() to release host
scsi: core: fix failure handling of scsi_add_host_with_dma
scsi: core: put ->shost_gendev.parent in failure handling path

 drivers/scsi/hosts.c | 25 ++++++++++++-------------
 1 file changed, 12 insertions(+), 13 deletions(-)
Cc Bart Van Assche <[email protected]>
Cc John Garry <[email protected]>
Cc Hannes Reinecke <[email protected]>
Signature · 1 line
2.29.2
author
#1 lore raw

PATCH v3 1/3 scsi: core: use put_device() to release host

After device is initialized via device_initialize(), or its name is set via dev_set_name(), the device has to be freed via put_device(), otherwise device name will be leaked because it is allocated dynamically in dev_set_name().

Fixes the issue by replacing kfree(shost) via two put_device() since both .shost_dev and .shost_gendev share same lifetime. Meantime move get_device(shost->shost_gendev) from scsi_add_host_with_dma to scsi_host_alloc(), so that we can grab parent's refcnt explicitly when assigning .shost_dev->parent. With this way code becomes more readable.

Also call put_device(dev->parent) in scsi_host_cls_release() so that code readability can be improved.

Cc Bart Van Assche <[email protected]>
Cc John Garry <[email protected]>
Cc Hannes Reinecke <[email protected]>
---
 drivers/scsi/hosts.c | 10 +++++-----
 1 file changed, 5 insertions(+), 5 deletions(-)
drivers/scsi/hosts.c +5 -5
index 624e2582c3df..ada11e3ae577 100644
@@ -55,7 +55,7 @@ static DEFINE_IDA(host_index_ida);
 
 static void scsi_host_cls_release(struct device *dev)
 {
-	put_device(&class_to_shost(dev)->shost_gendev);
+	put_device(dev->parent);
 }
 
 static struct class shost_class = {
@@ -261,8 +261,6 @@ int scsi_add_host_with_dma(struct Scsi_Host *shost, struct device *dev,
 	if (error)
 		goto out_del_gendev;
 
-	get_device(&shost->shost_gendev);
-
 	if (shost->transportt->host_size) {
 		shost->shost_data = kzalloc(shost->transportt->host_size,
 					 GFP_KERNEL);
@@ -473,7 +471,7 @@ struct Scsi_Host *scsi_host_alloc(struct scsi_host_template *sht, int privsize)
 	shost->shost_gendev.type = &scsi_host_type;
 
 	device_initialize(&shost->shost_dev);
-	shost->shost_dev.parent = &shost->shost_gendev;
+	shost->shost_dev.parent = get_device(&shost->shost_gendev);
 	shost->shost_dev.class = &shost_class;
 	dev_set_name(&shost->shost_dev, "host%d", shost->host_no);
 	shost->shost_dev.groups = scsi_sysfs_shost_attr_groups;
@@ -503,7 +501,9 @@ struct Scsi_Host *scsi_host_alloc(struct scsi_host_template *sht, int privsize)
  fail_index_remove:
 	ida_simple_remove(&host_index_ida, shost->host_no);
  fail_kfree:
-	kfree(shost);
+	put_device(&shost->shost_dev);
+	put_device(&shost->shost_gendev);
+
 	return NULL;
 }
 EXPORT_SYMBOL(scsi_host_alloc);
Signature · 1 line
2.29.2
author
#2 lore raw

PATCH v4 1/3 scsi: core: use put_device() to release host

From 2a70cffeca09156faf68d933157586b5bf9278fa Mon Sep 17 00:00:00 2001
From: Ming Lei <[email protected]>
Date: Mon, 31 May 2021 11:02:01 +0800
Subject: [PATCH V4 1/3] scsi: core: use put_device() to release host

After device is initialized via device_initialize(), or its name is set via dev_set_name(), the device has to be freed via put_device(), otherwise device name will be leaked because it is allocated dynamically in dev_set_name().

Fixes the issue by replacing kfree(shost) via two put_device() since both .shost_dev and .shost_gendev share same lifetime. Meantime move get_device(shost->shost_gendev) from scsi_add_host_with_dma to scsi_host_alloc(), so that we can grab parent's refcnt explicitly when assigning .shost_dev->parent. With this way code becomes more readable.

Also call put_device(dev->parent) in scsi_host_cls_release() so that code readability can be improved.

Cc Bart Van Assche <[email protected]>
Cc John Garry <[email protected]>
Cc Hannes Reinecke <[email protected]>

---
V4:
- avoid to touch un-initialized device instance

 drivers/scsi/hosts.c | 17 +++++++++--------
 1 file changed, 9 insertions(+), 8 deletions(-)
drivers/scsi/hosts.c +9 -8
index 624e2582c3df..181a5d6e3c7b 100644
@@ -55,7 +55,7 @@ static DEFINE_IDA(host_index_ida);
 
 static void scsi_host_cls_release(struct device *dev)
 {
-	put_device(&class_to_shost(dev)->shost_gendev);
+	put_device(dev->parent);
 }
 
 static struct class shost_class = {
@@ -261,8 +261,6 @@ int scsi_add_host_with_dma(struct Scsi_Host *shost, struct device *dev,
 	if (error)
 		goto out_del_gendev;
 
-	get_device(&shost->shost_gendev);
-
 	if (shost->transportt->host_size) {
 		shost->shost_data = kzalloc(shost->transportt->host_size,
 					 GFP_KERNEL);
@@ -391,8 +389,10 @@ struct Scsi_Host *scsi_host_alloc(struct scsi_host_template *sht, int privsize)
 	mutex_init(&shost->scan_mutex);
 
 	index = ida_simple_get(&host_index_ida, 0, 0, GFP_KERNEL);
-	if (index < 0)
-		goto fail_kfree;
+	if (index < 0) {
+		kfree(shost);
+		return NULL;
+	}
 	shost->host_no = index;
 
 	shost->dma_channel = 0xff;
@@ -473,7 +473,7 @@ struct Scsi_Host *scsi_host_alloc(struct scsi_host_template *sht, int privsize)
 	shost->shost_gendev.type = &scsi_host_type;
 
 	device_initialize(&shost->shost_dev);
-	shost->shost_dev.parent = &shost->shost_gendev;
+	shost->shost_dev.parent = get_device(&shost->shost_gendev);
 	shost->shost_dev.class = &shost_class;
 	dev_set_name(&shost->shost_dev, "host%d", shost->host_no);
 	shost->shost_dev.groups = scsi_sysfs_shost_attr_groups;
@@ -502,8 +502,9 @@ struct Scsi_Host *scsi_host_alloc(struct scsi_host_template *sht, int privsize)
 	kthread_stop(shost->ehandler);
  fail_index_remove:
 	ida_simple_remove(&host_index_ida, shost->host_no);
- fail_kfree:
-	kfree(shost);
+	/* free host instance */
+	put_device(&shost->shost_dev);
+	put_device(&shost->shost_gendev);
 	return NULL;
 }
 EXPORT_SYMBOL(scsi_host_alloc);
Signature · 1 line
2.29.2
author
#3 lore raw

PATCH v3 2/3 scsi: core: fix failure handling of scsi_add_host_with_dma

In reply to Ming Lei

When scsi_add_host_with_dma() return failure, the caller will call scsi_host_put(shost) to release everything allocated for this host instance. So we can't free allocated stuff in scsi_add_host_with_dma(), otherwise double free will be caused.

Strictly speaking, these host resources allocation should have been moved to scsi_host_alloc(), but the allocation may need driver's info which can be built between calling scsi_host_alloc() and scsi_add_host(), so just keep the allocations in scsi_add_host_with_dma().

Fixes the problem by relying on host device's release handler to release everything.

Cc Bart Van Assche <[email protected]>
Cc John Garry <[email protected]>
Cc Hannes Reinecke <[email protected]>
---
 drivers/scsi/hosts.c | 14 ++++++--------
 1 file changed, 6 insertions(+), 8 deletions(-)
drivers/scsi/hosts.c +6 -8
index ada11e3ae577..6cbc3eb16525 100644
@@ -279,23 +279,22 @@ int scsi_add_host_with_dma(struct Scsi_Host *shost, struct device *dev,
 
 		if (!shost->work_q) {
 			error = -EINVAL;
-			goto out_free_shost_data;
+			goto out_del_dev;
 		}
 	}
 
 	error = scsi_sysfs_add_host(shost);
 	if (error)
-		goto out_destroy_host;
+		goto out_del_dev;
 
 	scsi_proc_host_add(shost);
 	scsi_autopm_put_host(shost);
 	return error;
 
- out_destroy_host:
-	if (shost->work_q)
-		destroy_workqueue(shost->work_q);
- out_free_shost_data:
-	kfree(shost->shost_data);
+	/*
+	 * any host allocation in this function will be freed in
+	 * scsi_host_dev_release, so don't free them in the failure path
+	 */
  out_del_dev:
 	device_del(&shost->shost_dev);
  out_del_gendev:
@@ -305,7 +304,6 @@ int scsi_add_host_with_dma(struct Scsi_Host *shost, struct device *dev,
 	pm_runtime_disable(&shost->shost_gendev);
 	pm_runtime_set_suspended(&shost->shost_gendev);
 	pm_runtime_put_noidle(&shost->shost_gendev);
-	scsi_mq_destroy_tags(shost);
  fail:
 	return error;
 }
Signature · 1 line
2.29.2
Reviewed-by on v3 2/3
#4 lore raw

Re: [PATCH V3 2/3] scsi: core: fix failure handling of scsi_add_host_with_dma

Quoted 58 lines · Ming Lei

When scsi_add_host_with_dma() return failure, the caller will call scsi_host_put(shost) to release everything allocated for this host instance. So we can't free allocated stuff in scsi_add_host_with_dma(), otherwise double free will be caused.

Strictly speaking, these host resources allocation should have been moved to scsi_host_alloc(), but the allocation may need driver's info which can be built between calling scsi_host_alloc() and scsi_add_host(), so just keep the allocations in scsi_add_host_with_dma().

Fixes the problem by relying on host device's release handler to release everything.

Cc Bart Van Assche <[email protected]>
Cc John Garry <[email protected]>
Cc Hannes Reinecke <[email protected]>
---
  drivers/scsi/hosts.c | 14 ++++++--------
  1 file changed, 6 insertions(+), 8 deletions(-)
drivers/scsi/hosts.c +6 -7
index ada11e3ae577..6cbc3eb16525 100644
@@ -279,23 +279,22 @@ int scsi_add_host_with_dma(struct Scsi_Host *shost, struct device *dev,
 
  		if (!shost->work_q) {
  			error = -EINVAL;
-			goto out_free_shost_data;
+			goto out_del_dev;
  		}
  	}
 
  	error = scsi_sysfs_add_host(shost);
  	if (error)
-		goto out_destroy_host;
+		goto out_del_dev;
 
  	scsi_proc_host_add(shost);
  	scsi_autopm_put_host(shost);
  	return error;
 
- out_destroy_host:
-	if (shost->work_q)
-		destroy_workqueue(shost->work_q);
- out_free_shost_data:
-	kfree(shost->shost_data);
+	/*
+	 * any host allocation in this function will be freed in
+	 * scsi_host_dev_release, so don't free them in the failure path
+	 */
   out_del_dev:
  	device_del(&shost->shost_dev);
   out_del_gendev:

drivers/scsi/hosts.c int scsi_add_host_with_dma(struct Scsi_Host *shost, struct device *dev, · line 309

Quoted 7 lines
drivers/scsi/hosts.c +0 -1
@@ -305,7 +304,6 @@ int scsi_add_host_with_dma(struct Scsi_Host *shost, struct device *dev,
  	pm_runtime_disable(&shost->shost_gendev);
  	pm_runtime_set_suspended(&shost->shost_gendev);
  	pm_runtime_put_noidle(&shost->shost_gendev);
-	scsi_mq_destroy_tags(shost);
   fail:
  	return error;
  }

Cheers,

Hannes

Signature · 4 lines
Dr. Hannes Reinecke                Kernel Storage Architect
[email protected]                              +49 911 74053 688
SUSE Software Solutions GmbH, Maxfeldstr. 5, 90409 Nürnberg
HRB 36809 (AG Nürnberg), Geschäftsführer: Felix Imendörffer
author
#5 lore raw

PATCH v3 3/3 scsi: core: put ->shost_gendev.parent in failure handling path

In reply to Ming Lei

get_device(shost->shost_gendev.parent) is called in scsi_add_host_with_dma(), but its counter pair isn't called in the failure path, so fix it by calling put_device(shost->shost_gendev.parent) in its failure path.

Cc Bart Van Assche <[email protected]>
Cc Hannes Reinecke <[email protected]>
---
 drivers/scsi/hosts.c | 1 +
 1 file changed, 1 insertion(+)
drivers/scsi/hosts.c +1 -0
index 6cbc3eb16525..6cc43c51b7b3 100644
@@ -298,6 +298,7 @@ int scsi_add_host_with_dma(struct Scsi_Host *shost, struct device *dev,
  out_del_dev:
 	device_del(&shost->shost_dev);
  out_del_gendev:
+	put_device(shost->shost_gendev.parent);
 	device_del(&shost->shost_gendev);
  out_disable_runtime_pm:
 	device_disable_async_suspend(&shost->shost_gendev);
Signature · 1 line
2.29.2
#6 lore raw

Re: [PATCH V3 3/3] scsi: core: put ->shost_gendev.parent in failure handling path

Quoted 19 lines · Ming Lei

get_device(shost->shost_gendev.parent) is called in scsi_add_host_with_dma(), but its counter pair isn't called in the failure path, so fix it by calling put_device(shost->shost_gendev.parent) in its failure path.

Cc Bart Van Assche <[email protected]>
Cc Hannes Reinecke <[email protected]>
---
  drivers/scsi/hosts.c | 1 +
  1 file changed, 1 insertion(+)
drivers/scsi/hosts.c +0 -0
index 6cbc3eb16525..6cc43c51b7b3 100644

drivers/scsi/hosts.c int scsi_add_host_with_dma(struct Scsi_Host *shost, struct device *dev, · line 304

Quoted 7 lines
drivers/scsi/hosts.c +1 -0
@@ -298,6 +298,7 @@ int scsi_add_host_with_dma(struct Scsi_Host *shost, struct device *dev,
   out_del_dev:
  	device_del(&shost->shost_dev);
   out_del_gendev:
+	put_device(shost->shost_gendev.parent);
  	device_del(&shost->shost_gendev);
   out_disable_runtime_pm:
  	device_disable_async_suspend(&shost->shost_gendev);

This really needs to be folded into the first patch as it's really a bugfix for that.

Cheers,

Hannes

Signature · 4 lines
Dr. Hannes Reinecke                Kernel Storage Architect
[email protected]                              +49 911 74053 688
SUSE Software Solutions GmbH, Maxfeldstr. 5, 90409 Nürnberg
HRB 36809 (AG Nürnberg), Geschäftsführer: Felix Imendörffer
author
#7 lore raw

Re: [PATCH V3 3/3] scsi: core: put ->shost_gendev.parent in failure handling path

Quoted 14 lines · Hannes Reinecke
Quoted 13 lines · Ming Lei

get_device(shost->shost_gendev.parent) is called in scsi_add_host_with_dma(), but its counter pair isn't called in the failure path, so fix it by calling put_device(shost->shost_gendev.parent) in its failure path.

Cc Bart Van Assche <[email protected]>
Cc Hannes Reinecke <[email protected]>
---
  drivers/scsi/hosts.c | 1 +
  1 file changed, 1 insertion(+)
Quoted 15 lines

drivers/scsi/hosts.c int scsi_add_host_with_dma(struct Scsi_Host *shost, struct device *dev, · line 304

Quoted 13 lines
drivers/scsi/hosts.c +1 -0
index 6cbc3eb16525..6cc43c51b7b3 100644
@@ -298,6 +298,7 @@ int scsi_add_host_with_dma(struct Scsi_Host *shost, struct device *dev,
   out_del_dev:
  	device_del(&shost->shost_dev);
   out_del_gendev:
+	put_device(shost->shost_gendev.parent);
  	device_del(&shost->shost_gendev);
   out_disable_runtime_pm:
  	device_disable_async_suspend(&shost->shost_gendev);

This really needs to be folded into the first patch as it's really a bugfix for that.

All three are bug fixes, and this one may leak .shost_gendev's parent, but the 1st one leaks .shost_gendev->kobj.name, so we needn't to fold the 3rd into the 1st one.

Thanks,
Ming

#8 lore raw

Re: [PATCH V3 3/3] scsi: core: put ->shost_gendev.parent in failure handling path

Quoted 31 lines · Ming Lei
Quoted 14 lines · Hannes Reinecke
Quoted 13 lines · Ming Lei

get_device(shost->shost_gendev.parent) is called in scsi_add_host_with_dma(), but its counter pair isn't called in the failure path, so fix it by calling put_device(shost->shost_gendev.parent) in its failure path.

Cc Bart Van Assche <[email protected]>
Cc Hannes Reinecke <[email protected]>
---
   drivers/scsi/hosts.c | 1 +
   1 file changed, 1 insertion(+)
Quoted 15 lines

drivers/scsi/hosts.c int scsi_add_host_with_dma(struct Scsi_Host *shost, struct device *dev, · line 304

Quoted 13 lines
drivers/scsi/hosts.c +1 -0
index 6cbc3eb16525..6cc43c51b7b3 100644
@@ -298,6 +298,7 @@ int scsi_add_host_with_dma(struct Scsi_Host *shost, struct device *dev,
    out_del_dev:
   	device_del(&shost->shost_dev);
    out_del_gendev:
+	put_device(shost->shost_gendev.parent);
   	device_del(&shost->shost_gendev);
    out_disable_runtime_pm:
   	device_disable_async_suspend(&shost->shost_gendev);

This really needs to be folded into the first patch as it's really a bugfix for that.

Quoted 4 lines

All three are bug fixes, and this one may leak .shost_gendev's parent, but the 1st one leaks .shost_gendev->kobj.name, so we needn't to fold the 3rd into the 1st one.

I beg to disagree.
The first patch removes the 'get_device()' from
scsi_add_host_with_dma(), but does not remove the corresponding 'put_device()' in the error path.
So the first patch introduces a reference imbalance which is fixed by the third patch. Hence my suggestion to merge those two patches.

Cheers,

Hannes

Signature · 4 lines
Dr. Hannes Reinecke                Kernel Storage Architect
[email protected]                              +49 911 74053 688
SUSE Software Solutions GmbH, Maxfeldstr. 5, 90409 Nürnberg
HRB 36809 (AG Nürnberg), Geschäftsführer: Felix Imendörffer
author
#9 lore raw

Re: [PATCH V3 3/3] scsi: core: put ->shost_gendev.parent in failure handling path

Quoted 32 lines · Hannes Reinecke
Quoted 31 lines · Ming Lei
Quoted 29 lines · Hannes Reinecke
Quoted 19 lines · Ming Lei

get_device(shost->shost_gendev.parent) is called in scsi_add_host_with_dma(), but its counter pair isn't called in the failure path, so fix it by calling put_device(shost->shost_gendev.parent) in its failure path.

Cc Bart Van Assche <[email protected]>
Cc Hannes Reinecke <[email protected]>
---
   drivers/scsi/hosts.c | 1 +
   1 file changed, 1 insertion(+)
drivers/scsi/hosts.c +0 -0
index 6cbc3eb16525..6cc43c51b7b3 100644

drivers/scsi/hosts.c int scsi_add_host_with_dma(struct Scsi_Host *shost, struct device *dev, · line 304

Quoted 7 lines
drivers/scsi/hosts.c +1 -0
@@ -298,6 +298,7 @@ int scsi_add_host_with_dma(struct Scsi_Host *shost, struct device *dev,
    out_del_dev:
   	device_del(&shost->shost_dev);
    out_del_gendev:
+	put_device(shost->shost_gendev.parent);
   	device_del(&shost->shost_gendev);
    out_disable_runtime_pm:
   	device_disable_async_suspend(&shost->shost_gendev);

This really needs to be folded into the first patch as it's really a bugfix for that.

Quoted 8 lines
Quoted 4 lines

All three are bug fixes, and this one may leak .shost_gendev's parent, but the 1st one leaks .shost_gendev->kobj.name, so we needn't to fold the 3rd into the 1st one.

I beg to disagree.
The first patch removes the 'get_device()' from scsi_add_host_with_dma(), but does not remove the corresponding 'put_device()' in the error path.

There isn't such 'put_device' in the error path of scsi_add_host_with_dma(), is there?

Quoted 2 lines

So the first patch introduces a reference imbalance which is fixed by the third patch. Hence my suggestion to merge those two patches.

No, that isn't true, please look at current code of scsi_add_host_with_dma().

Thanks,
Ming

#10 lore raw
In reply to Ming Lei
Quoted 20 lines · Ming Lei

Hello Martin,

Fix two memory leaks and one double free issue in alloc/add host code path.

V3:
- fix memory leak caused in scsi_host_alloc
- comment typo suggested by John

V2:
- add patch 2 for addressing shost leak in case of adding host failure, reported by John Garry.

Ming Lei (3):
scsi: core: use put_device() to release host
scsi: core: fix failure handling of scsi_add_host_with_dma
scsi: core: put ->shost_gendev.parent in failure handling path

Quoted 6 lines
  drivers/scsi/hosts.c | 25 ++++++++++++-------------
  1 file changed, 12 insertions(+), 13 deletions(-)
Cc Bart Van Assche <[email protected]>
Cc John Garry <[email protected]>
Cc Hannes Reinecke <[email protected]>

I tested this again with the same experiment as before to error in
scsi_add_host_with_dma(), and we don't call scsi_host_dev_release() now. The shost_gendev dev refcount is 2 upon exit in scsi_add_host_with_dma().

We don't call scsi_host_cls_release() either, so I guess a ref count is leaked for shost_dev - I see its refcount is 1 at exit in scsi_add_host_with_dma(). We have the device_initialize(), device_add(), device_del() in the alloc and add host functions, but I don't know who is responsible for the final "device put".

Thanks,
John

author
#11 lore raw
Quoted 33 lines · John Garry
Quoted 20 lines · Ming Lei

Hello Martin,

Fix two memory leaks and one double free issue in alloc/add host code path.

V3:
- fix memory leak caused in scsi_host_alloc
- comment typo suggested by John

V2:
- add patch 2 for addressing shost leak in case of adding host failure, reported by John Garry.

Ming Lei (3):
scsi: core: use put_device() to release host
scsi: core: fix failure handling of scsi_add_host_with_dma
scsi: core: put ->shost_gendev.parent in failure handling path

Quoted 6 lines
  drivers/scsi/hosts.c | 25 ++++++++++++-------------
  1 file changed, 12 insertions(+), 13 deletions(-)
Cc Bart Van Assche <[email protected]>
Cc John Garry <[email protected]>
Cc Hannes Reinecke <[email protected]>

I tested this again with the same experiment as before to error in scsi_add_host_with_dma(), and we don't call scsi_host_dev_release() now. The shost_gendev dev refcount is 2 upon exit in scsi_add_host_with_dma().

Quoted 5 lines

We don't call scsi_host_cls_release() either, so I guess a ref count is leaked for shost_dev - I see its refcount is 1 at exit in scsi_add_host_with_dma(). We have the device_initialize(), device_add(), device_del() in the alloc and add host functions, but I don't know who is responsible for the final "device put".

Hammm, we still need to put ->shost_dev before returning the error, and the following delta patch can fix the issue, and it should have been wrapped into the 1st one.

drivers/scsi/hosts.c +2 -0
index 22a58e453a0c..532165462a42 100644
@@ -306,6 +306,8 @@ int scsi_add_host_with_dma(struct Scsi_Host *shost, struct device *dev,
 	pm_runtime_set_suspended(&shost->shost_gendev);
 	pm_runtime_put_noidle(&shost->shost_gendev);
  fail:
+	/* drop ref of ->shost_dev so that caller can release this host */
+	put_device(&shost->shost_dev);
 	return error;
 }
 EXPORT_SYMBOL(scsi_add_host_with_dma);

Thanks,
Ming

#12 lore raw
Quoted 14 lines · Ming Lei
Quoted 5 lines

We don't call scsi_host_cls_release() either, so I guess a ref count is leaked for shost_dev - I see its refcount is 1 at exit in scsi_add_host_with_dma(). We have the device_initialize(), device_add(), device_del() in the alloc and add host functions, but I don't know who is responsible for the final "device put".

Hammm, we still need to put ->shost_dev before returning the error, and the following delta patch can fix the issue, and it should have been wrapped into the 1st one.

drivers/scsi/hosts.c +0 -0
index 22a58e453a0c..532165462a42 100644

drivers/scsi/hosts.c int scsi_add_host_with_dma(struct Scsi_Host *shost, struct device *dev, · line 313

Quoted 8 lines
drivers/scsi/hosts.c +2 -0
@@ -306,6 +306,8 @@ int scsi_add_host_with_dma(struct Scsi_Host *shost, struct device *dev,
  	pm_runtime_set_suspended(&shost->shost_gendev);
  	pm_runtime_put_noidle(&shost->shost_gendev);
   fail:
+	/* drop ref of ->shost_dev so that caller can release this host */
+	put_device(&shost->shost_dev);
  	return error;
  }
  EXPORT_SYMBOL(scsi_add_host_with_dma);

That looks better now.

And we can see the equivalent on the normal removal path in scsi_remove_host() -> device_unregister(&shost->shost_dev), which does a device_del()+put_device().

So could we actually just have:
out_del_dev:

	unregister_dev(&shost->shost_dev)

I am not sure if we are required to keep that shost_dev reference all the way until the exit, as you do.

Thanks,
John

author
#13 lore raw
Quoted 26 lines · John Garry
Quoted 14 lines · Ming Lei
Quoted 5 lines

We don't call scsi_host_cls_release() either, so I guess a ref count is leaked for shost_dev - I see its refcount is 1 at exit in scsi_add_host_with_dma(). We have the device_initialize(), device_add(), device_del() in the alloc and add host functions, but I don't know who is responsible for the final "device put".

Hammm, we still need to put ->shost_dev before returning the error, and the following delta patch can fix the issue, and it should have been wrapped into the 1st one.

drivers/scsi/hosts.c +0 -0
index 22a58e453a0c..532165462a42 100644

drivers/scsi/hosts.c int scsi_add_host_with_dma(struct Scsi_Host *shost, struct device *dev, · line 313

Quoted 8 lines
drivers/scsi/hosts.c +2 -0
@@ -306,6 +306,8 @@ int scsi_add_host_with_dma(struct Scsi_Host *shost, struct device *dev,
  	pm_runtime_set_suspended(&shost->shost_gendev);
  	pm_runtime_put_noidle(&shost->shost_gendev);
   fail:
+	/* drop ref of ->shost_dev so that caller can release this host */
+	put_device(&shost->shost_dev);
  	return error;
  }
  EXPORT_SYMBOL(scsi_add_host_with_dma);

That looks better now.

Quoted 8 lines

And we can see the equivalent on the normal removal path in scsi_remove_host() -> device_unregister(&shost->shost_dev), which does a device_del()+put_device().

So could we actually just have:
out_del_dev:

	unregister_dev(&shost->shost_dev)

No, we still have to call put_device(&shost->shost_dev) only in case of failure before adding &shost->shost_dev.

Quoted 2 lines

I am not sure if we are required to keep that shost_dev reference all the way until the exit, as you do.

That has been done in this way, the problem is that both .shost_dev and .shost_gendev share same lifetime and memory(same struct Scsi_Host instance), this kind of pattern isn't one usual driver core use case, and we have to handle it carefully.

Thanks,
Ming