Skip to contentSkip to conversation

btrfs: unwind device add when sysfs registration fails

PATCH Superseded Adarsh Das 3 participants 6 messages Oct 3 – Oct 13 Btrfs file system

lore
Share & tools
lore

Adarsh Das

author
#0 lore raw

PATCH 1/1 btrfs: unwind device add when sysfs registration fails

btrfs_init_new_device() ignored the return value of btrfs_sysfs_add_device(). When sysfs link creation failed (e.g. -EEXIST), the add path continued, aborted the transaction with the same errno, and tripped WARN_ON(btrfs_abort_should_print_stack()) while leaving the device half-registered in memory.

Unwind through error_sysfs when sysfs registration fails, matching other failure paths after the device lists are updated.

---
 fs/btrfs/volumes.c | 6 +++++-
 1 file changed, 5 insertions(+), 1 deletion(-)
fs/btrfs/volumes.c +5 -1
index 85ea9c5d4536..1517ed3da22c 100644
@@ -3066,7 +3066,11 @@ int btrfs_init_new_device(struct btrfs_fs_info *fs_info, const char *device_path
 	mutex_unlock(&fs_info->chunk_mutex);
 
 	/* Add sysfs device entry */
-	btrfs_sysfs_add_device(device);
+	ret = btrfs_sysfs_add_device(device);
+	if (ret) {
+		mutex_unlock(&fs_devices->device_list_mutex);
+		goto error_sysfs;
+	}
 
 	mutex_unlock(&fs_devices->device_list_mutex);
 
#1 lore raw
Quoted 25 lines · Adarsh Das

btrfs_init_new_device() ignored the return value of btrfs_sysfs_add_device(). When sysfs link creation failed (e.g. -EEXIST), the add path continued, aborted the transaction with the same errno, and tripped WARN_ON(btrfs_abort_should_print_stack()) while leaving the device half-registered in memory.

Unwind through error_sysfs when sysfs registration fails, matching other failure paths after the device lists are updated.

---
  fs/btrfs/volumes.c | 6 +++++-
  1 file changed, 5 insertions(+), 1 deletion(-)
fs/btrfs/volumes.c +0 -0
index 85ea9c5d4536..1517ed3da22c 100644
@@ -3066,7 +3066,11 @@ int btrfs_init_new_device(struct btrfs_fs_info *fs_info, const char *device_path
  	mutex_unlock(&fs_info->chunk_mutex);
 

fs/btrfs/volumes.c int btrfs_init_new_device(struct btrfs_fs_info *fs_info, const char *device_path · line 3076

Quoted 10 lines
fs/btrfs/volumes.c +5 -1
  	/* Add sysfs device entry */
-	btrfs_sysfs_add_device(device);
+	ret = btrfs_sysfs_add_device(device);
+	if (ret) {
+		mutex_unlock(&fs_devices->device_list_mutex);
+		goto error_sysfs;
+	}
 
  	mutex_unlock(&fs_devices->device_list_mutex);
 

Are you sure this is safe? If btrfs_sysfs_add_device() fails i.e. at sysfs_create_link() and you do 'goto error_sysfs;' which then calls btrfs_sysfs_remove_device(). That one then calls sysfs_remove_link() for a link that was never created. This is only working at the moment, because sysfs_remove_link() doesn't check for errors of kernfs_remove_by_name() (which in this case returns ENOENT).

author
#2 lore raw

PATCH v2 1/1 btrfs: unwind device add when sysfs registration fails

btrfs_init_new_device() ignored the return value of btrfs_sysfs_add_device(). When sysfs link creation failed (e.g. -EEXIST), the add path continued, aborted the transaction with the same errno, and tripped WARN_ON(btrfs_abort_should_print_stack()) while leaving the device half-registered in memory.

Check the return value and unwind when registration fails, without aborting the filesystem for errnos like -EEXIST. If add failed before anything was published in sysfs, tear down the in-memory device state only and do not call btrfs_sysfs_remove_device(). When the block-device link was created but devid kobject registration fails, drop the link inside btrfs_sysfs_add_device() before returning the error.

---
v2:
- Unwind without sysfs remove when add failed (review on v1)
- Roll back block-device link if devid kobject add fails
v1: https://lore.kernel.org/all/[email protected]/

---
 fs/btrfs/sysfs.c   | 3 +++
 fs/btrfs/volumes.c | 7 ++++++-
 2 files changed, 9 insertions(+), 1 deletion(-)
fs/btrfs/sysfs.c +3 -0
index 39cb01ee441a..2f034689e71b 100644
@@ -2165,6 +2165,9 @@ int btrfs_sysfs_add_device(struct btrfs_device *device)
 	ret = kobject_init_and_add(&device->devid_kobj, &devid_ktype,
 				   devinfo_kobj, "%llu", device->devid);
 	if (ret) {
+		if (device->bdev)
+			sysfs_remove_link(devices_kobj,
+					  bdev_kobj(device->bdev)->name);
 		kobject_put(&device->devid_kobj);
 		btrfs_warn(device->fs_info,
 			   "devinfo init for devid %llu failed: %d",
fs/btrfs/volumes.c +6 -1
index 85ea9c5d4536..56b38f012093 100644
@@ -3066,7 +3066,11 @@ int btrfs_init_new_device(struct btrfs_fs_info *fs_info, const char *device_path
 	mutex_unlock(&fs_info->chunk_mutex);
 
 	/* Add sysfs device entry */
-	btrfs_sysfs_add_device(device);
+	ret = btrfs_sysfs_add_device(device);
+	if (ret) {
+		mutex_unlock(&fs_devices->device_list_mutex);
+		goto error_unwind_device;
+	}
 
 	mutex_unlock(&fs_devices->device_list_mutex);
 
@@ -3147,6 +3151,7 @@ int btrfs_init_new_device(struct btrfs_fs_info *fs_info, const char *device_path
 
 error_sysfs:
 	btrfs_sysfs_remove_device(device);
+error_unwind_device:
 	mutex_lock(&fs_info->fs_devices->device_list_mutex);
 	if (seeding_dev)
 		btrfs_assign_next_active_device(device, seed_devices->latest_dev);
#4 lore raw
In reply to Adarsh Das
Quoted 60 lines · Adarsh Das

btrfs_init_new_device() ignored the return value of btrfs_sysfs_add_device(). When sysfs link creation failed (e.g. -EEXIST), the add path continued, aborted the transaction with the same errno, and tripped WARN_ON(btrfs_abort_should_print_stack()) while leaving the device half-registered in memory.

Check the return value and unwind when registration fails, without aborting the filesystem for errnos like -EEXIST. If add failed before anything was published in sysfs, tear down the in-memory device state only and do not call btrfs_sysfs_remove_device(). When the block-device link was created but devid kobject registration fails, drop the link inside btrfs_sysfs_add_device() before returning the error.

---
v2:
- Unwind without sysfs remove when add failed (review on v1)
- Roll back block-device link if devid kobject add fails
v1: https://lore.kernel.org/all/[email protected]/

---
 fs/btrfs/sysfs.c   | 3 +++
 fs/btrfs/volumes.c | 7 ++++++-
 2 files changed, 9 insertions(+), 1 deletion(-)
fs/btrfs/sysfs.c +3 -0
index 39cb01ee441a..2f034689e71b 100644
@@ -2165,6 +2165,9 @@ int btrfs_sysfs_add_device(struct btrfs_device *device)
        ret = kobject_init_and_add(&device->devid_kobj, &devid_ktype,
                                   devinfo_kobj, "%llu", device->devid);
        if (ret) {
+               if (device->bdev)
+                       sysfs_remove_link(devices_kobj,
+                                         bdev_kobj(device->bdev)->name);
                kobject_put(&device->devid_kobj);
                btrfs_warn(device->fs_info,
                           "devinfo init for devid %llu failed: %d",
fs/btrfs/volumes.c +5 -1
index 85ea9c5d4536..56b38f012093 100644
@@ -3066,7 +3066,11 @@ int btrfs_init_new_device(struct btrfs_fs_info *fs_info, const char *device_path
        mutex_unlock(&fs_info->chunk_mutex);
 
        /* Add sysfs device entry */
-       btrfs_sysfs_add_device(device);
+       ret = btrfs_sysfs_add_device(device);
+       if (ret) {
+               mutex_unlock(&fs_devices->device_list_mutex);
+               goto error_unwind_device;
+       }
 
        mutex_unlock(&fs_devices->device_list_mutex);
 

fs/btrfs/volumes.c int btrfs_init_new_device(struct btrfs_fs_info *fs_info, const char *device_path · line 3157

Quoted 7 lines
fs/btrfs/volumes.c +1 -0
@@ -3147,6 +3151,7 @@ int btrfs_init_new_device(struct btrfs_fs_info *fs_info, const char *device_path
 
 error_sysfs:
        btrfs_sysfs_remove_device(device);
+error_unwind_device:
        mutex_lock(&fs_info->fs_devices->device_list_mutex);
        if (seeding_dev)
                btrfs_assign_next_active_device(device, seed_devices->latest_dev);

I'm wondering why not this?:

@@ -3066,7 +3066,9 @@ int btrfs_init_new_device(struct btrfs_fs_info *fs_info, const char *device_path

        mutex_unlock(&fs_info->chunk_mutex);

        /* Add sysfs device entry */
-       btrfs_sysfs_add_device(device);
+       ret = btrfs_sysfs_add_device(device);
+       if (ret)

+ goto error_device;

        mutex_unlock(&fs_devices->device_list_mutex);

@@ -3148,6 +3150,7 @@ int btrfs_init_new_device(struct btrfs_fs_info *fs_info, const char *device_path
error_sysfs:

        btrfs_sysfs_remove_device(device);
        mutex_lock(&fs_info->fs_devices->device_list_mutex);
+error_device:
        mutex_lock(&fs_info->chunk_mutex);
        list_del_rcu(&device->dev_list);
        list_del(&device->dev_alloc_list);

--nX

#5 lore raw
Quoted 92 lines · Daniel Vacek
Quoted 60 lines · Adarsh Das

btrfs_init_new_device() ignored the return value of btrfs_sysfs_add_device(). When sysfs link creation failed (e.g. -EEXIST), the add path continued, aborted the transaction with the same errno, and tripped WARN_ON(btrfs_abort_should_print_stack()) while leaving the device half-registered in memory.

Check the return value and unwind when registration fails, without aborting the filesystem for errnos like -EEXIST. If add failed before anything was published in sysfs, tear down the in-memory device state only and do not call btrfs_sysfs_remove_device(). When the block-device link was created but devid kobject registration fails, drop the link inside btrfs_sysfs_add_device() before returning the error.

---
v2:
- Unwind without sysfs remove when add failed (review on v1)
- Roll back block-device link if devid kobject add fails
v1: https://lore.kernel.org/all/[email protected]/

---
  fs/btrfs/sysfs.c   | 3 +++
  fs/btrfs/volumes.c | 7 ++++++-
  2 files changed, 9 insertions(+), 1 deletion(-)
fs/btrfs/sysfs.c +3 -0
index 39cb01ee441a..2f034689e71b 100644
@@ -2165,6 +2165,9 @@ int btrfs_sysfs_add_device(struct btrfs_device *device)
         ret = kobject_init_and_add(&device->devid_kobj, &devid_ktype,
                                    devinfo_kobj, "%llu", device->devid);
         if (ret) {
+               if (device->bdev)
+                       sysfs_remove_link(devices_kobj,
+                                         bdev_kobj(device->bdev)->name);
                 kobject_put(&device->devid_kobj);
                 btrfs_warn(device->fs_info,
                            "devinfo init for devid %llu failed: %d",
fs/btrfs/volumes.c +5 -1
index 85ea9c5d4536..56b38f012093 100644
@@ -3066,7 +3066,11 @@ int btrfs_init_new_device(struct btrfs_fs_info *fs_info, const char *device_path
         mutex_unlock(&fs_info->chunk_mutex);
 
         /* Add sysfs device entry */
-       btrfs_sysfs_add_device(device);
+       ret = btrfs_sysfs_add_device(device);
+       if (ret) {
+               mutex_unlock(&fs_devices->device_list_mutex);
+               goto error_unwind_device;
+       }
 
         mutex_unlock(&fs_devices->device_list_mutex);
 

fs/btrfs/volumes.c int btrfs_init_new_device(struct btrfs_fs_info *fs_info, const char *device_path · line 3157

Quoted 7 lines
fs/btrfs/volumes.c +1 -0
@@ -3147,6 +3151,7 @@ int btrfs_init_new_device(struct btrfs_fs_info *fs_info, const char *device_path
 
  error_sysfs:
         btrfs_sysfs_remove_device(device);
+error_unwind_device:
         mutex_lock(&fs_info->fs_devices->device_list_mutex);
         if (seeding_dev)
                 btrfs_assign_next_active_device(device, seed_devices->latest_dev);

I'm wondering why not this?:

fs/btrfs/volumes.c +4 -1
@@ -3066,7 +3066,9 @@ int btrfs_init_new_device(struct btrfs_fs_info
 *fs_info, const char *device_path
         mutex_unlock(&fs_info->chunk_mutex);
 
         /* Add sysfs device entry */
-       btrfs_sysfs_add_device(device);
+       ret = btrfs_sysfs_add_device(device);
+       if (ret)
+               goto error_device;
 
         mutex_unlock(&fs_devices->device_list_mutex);
@@ -3148,6 +3150,7 @@ int btrfs_init_new_device(struct btrfs_fs_info
 *fs_info, const char *device_path
  error_sysfs:
         btrfs_sysfs_remove_device(device);
         mutex_lock(&fs_info->fs_devices->device_list_mutex);
+error_device:
         mutex_lock(&fs_info->chunk_mutex);
         list_del_rcu(&device->dev_list);
         list_del(&device->dev_alloc_list);
Quoted 1 line

--nX

This would remove the need for the lock of the device_list_mutex indeed.