cxl: fix leak of ctx->mapping when releasing kernel API contexts

When a context is created via the kernel API, ctx->mapping is allocated
within the kernel and thus needs to be freed when the context is freed.
reclaim_ctx() attempts to do this for contexts with the ctx->kernelapi flag
set, but afu_release() (which can be called from the kernel API through
cxl_fd_release()) sets ctx->mapping to NULL before calling
cxl_context_free() to free the context.

Add a check to afu_release() so that the mappings in contexts created via
the kernel API are left alone so reclaim_ctx() can free them.

Reported-by: Matthew R. Ochs <mrochs@linux.vnet.ibm.com>
Fixes: 6f7f0b3df6d4 ("cxl: Add AFU virtual PHB and kernel API")
Signed-off-by: Andrew Donnellan <andrew.donnellan@au1.ibm.com>
Acked-by: Ian Munsie <imunsie@au1.ibm.com>
Reviewed-by: Matthew R. Ochs <mrochs@linux.vnet.ibm.com>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>

diff --git a/drivers/misc/cxl/file.c b/drivers/misc/cxl/file.c
index a30bf285b5bdd75c3f2b357d89dbab251bb98c7c..7ccd2998be92b8b3f7cdca2a0acbf3f9586d0f34 100644 (file)
--- a/drivers/misc/cxl/file.c
+++ b/drivers/misc/cxl/file.c
@@ -120,9 +120,16 @@ int afu_release(struct inode *inode, struct file *file)
                 __func__, ctx->pe);
        cxl_context_detach(ctx);
 
-       mutex_lock(&ctx->mapping_lock);
-       ctx->mapping = NULL;
-       mutex_unlock(&ctx->mapping_lock);
+
+       /*
+        * Delete the context's mapping pointer, unless it's created by the
+        * kernel API, in which case leave it so it can be freed by reclaim_ctx()
+        */
+       if (!ctx->kernelapi) {
+               mutex_lock(&ctx->mapping_lock);
+               ctx->mapping = NULL;
+               mutex_unlock(&ctx->mapping_lock);
+       }
 
        put_device(&ctx->afu->dev);