On Mon, Oct 31, 2022 at 7:53 AM Dawei Li <set_pte_at@xxxxxxxxxxx> wrote: > > Racing conflict could be: > task A task B > list_for_each_entry > strcmp(h->name)) > list_for_each_entry > strcmp(h->name) > kzalloc kzalloc > ...... ..... > device_create device_create > list_add > list_add > > The root cause is that task B has no idea about the fact someone > else(A) has inserted heap with same name when it calls list_add, > so a potential collision occurs. > > v1: https://lore.kernel.org/all/TYCP286MB2323950197F60FC3473123B7CA349@xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx/ > > v1->v2: Narrow down locking scope, check the existence of heap before > insertion, as suggested by Andrew Davis. > > Fixes: c02a81fba74f ("dma-buf: Add dma-buf heaps framework") > > base-commit: 447fb14bf07905b880c9ed1ea92c53d6dd0649d7 > > Signed-off-by: Dawei Li <set_pte_at@xxxxxxxxxxx> > --- > drivers/dma-buf/dma-heap.c | 37 ++++++++++++++++++++++++++++--------- > 1 file changed, 28 insertions(+), 9 deletions(-) > > diff --git a/drivers/dma-buf/dma-heap.c b/drivers/dma-buf/dma-heap.c > index 8f5848aa144f..1c787a147e3a 100644 > --- a/drivers/dma-buf/dma-heap.c > +++ b/drivers/dma-buf/dma-heap.c > @@ -216,9 +216,21 @@ const char *dma_heap_get_name(struct dma_heap *heap) > return heap->name; > } > > +static inline bool dma_heap_exist(const char *name) > +{ > + struct dma_heap *h; > + > + list_for_each_entry(h, &heap_list, list) { > + if (!strcmp(h->name, name)) > + return true; > + } > + > + return false; > +} > + > struct dma_heap *dma_heap_add(const struct dma_heap_export_info *exp_info) > { > - struct dma_heap *heap, *h, *err_ret; > + struct dma_heap *heap, *err_ret; > struct device *dev_ret; > unsigned int minor; > int ret; > @@ -235,13 +247,11 @@ struct dma_heap *dma_heap_add(const struct dma_heap_export_info *exp_info) > > /* check the name is unique */ > mutex_lock(&heap_list_lock); > - list_for_each_entry(h, &heap_list, list) { > - if (!strcmp(h->name, exp_info->name)) { > - mutex_unlock(&heap_list_lock); > - pr_err("dma_heap: Already registered heap named %s\n", > - exp_info->name); > - return ERR_PTR(-EINVAL); > - } > + if (dma_heap_exist(exp_info->name)) { > + mutex_unlock(&heap_list_lock); > + pr_err("dma_heap: Already registered heap named %s\n", > + exp_info->name); > + return ERR_PTR(-EINVAL); Hi Dawei, What Andrew was suggesting was that you entirely move the check from here to the critical section down below, not duplicate the check. I don't think we want to check this twice. We should be able to do this by taking the heap_list_lock only once. > } > mutex_unlock(&heap_list_lock); > > @@ -283,13 +293,22 @@ struct dma_heap *dma_heap_add(const struct dma_heap_export_info *exp_info) > err_ret = ERR_CAST(dev_ret); > goto err2; > } > + > /* Add heap to the list */ > mutex_lock(&heap_list_lock); > + if (unlikely(dma_heap_exist(exp_info->name))) { > + mutex_unlock(&heap_list_lock); > + pr_err("dma_heap: Already registered heap named %s\n", > + exp_info->name); > + err_ret = ERR_PTR(-EINVAL); > + goto err3; > + } > list_add(&heap->list, &heap_list); > mutex_unlock(&heap_list_lock); > > return heap; > - > +err3: > + device_destroy(dma_heap_class, heap->heap_devt); > err2: > cdev_del(&heap->heap_cdev); > err1: > -- > 2.25.1 >
