aio: add aio_kernel_() interface

This adds an interface that lets kernel callers submit aio iocbs without
going through the user space syscalls.  This lets kernel callers avoid
the management limits and overhead of the context.  It will also let us
integrate aio operations with other kernel apis that the user space
interface doesn't have access to.

Signed-off-by: Dave Kleikamp <dave.kleikamp@oracle.com>
Cc: Zach Brown <zab@zabbo.net>

diff --git a/fs/aio.c b/fs/aio.c
index 6b868f0e0c4c019c3b68712ca230deb979711a98..8e700bacac78daf84336a6cbd12194be9dc91a1e 100644 (file)
--- a/fs/aio.c
+++ b/fs/aio.c
@@ -855,6 +855,10 @@ void aio_complete(struct kiocb *iocb, long res, long res2)
                iocb->ki_ctx = ERR_PTR(-EXDEV);
                wake_up_process(iocb->ki_obj.tsk);
                return;
+       } else if (is_kernel_kiocb(iocb)) {
+               iocb->ki_obj.complete(iocb->ki_user_data, res);
+               aio_kernel_free(iocb);
+               return;
        }
 
        /*
@@ -1281,6 +1285,80 @@ rw_common:
        return 0;
 }
 
+/*
+ * This allocates an iocb that will be used to submit and track completion of
+ * an IO that is issued from kernel space.
+ *
+ * The caller is expected to call the appropriate aio_kernel_init_() functions
+ * and then call aio_kernel_submit().  From that point forward progress is
+ * guaranteed by the file system aio method.  Eventually the caller's
+ * completion callback will be called.
+ *
+ * These iocbs are special.  They don't have a context, we don't limit the
+ * number pending, and they can't be canceled.
+ */
+struct kiocb *aio_kernel_alloc(gfp_t gfp)
+{
+       return kzalloc(sizeof(struct kiocb), gfp);
+}
+EXPORT_SYMBOL_GPL(aio_kernel_alloc);
+
+void aio_kernel_free(struct kiocb *iocb)
+{
+       kfree(iocb);
+}
+EXPORT_SYMBOL_GPL(aio_kernel_free);
+
+/*
+ * ptr and count can be a buff and bytes or an iov and segs.
+ */
+void aio_kernel_init_rw(struct kiocb *iocb, struct file *filp,
+                       size_t nr, loff_t off)
+{
+       iocb->ki_filp = filp;
+       iocb->ki_nbytes = nr;
+       iocb->ki_pos = off;
+       iocb->ki_ctx = (void *)-1;
+}
+EXPORT_SYMBOL_GPL(aio_kernel_init_rw);
+
+void aio_kernel_init_callback(struct kiocb *iocb,
+                             void (*complete)(u64 user_data, long res),
+                             u64 user_data)
+{
+       iocb->ki_obj.complete = complete;
+       iocb->ki_user_data = user_data;
+}
+EXPORT_SYMBOL_GPL(aio_kernel_init_callback);
+
+/*
+ * The iocb is our responsibility once this is called.  The caller must not
+ * reference it.
+ *
+ * Callers must be prepared for their iocb completion callback to be called the
+ * moment they enter this function.  The completion callback may be called from
+ * any context.
+ *
+ * Returns: 0: the iocb completion callback will be called with the op result
+ * negative errno: the operation was not submitted and the iocb was freed
+ */
+int aio_kernel_submit(struct kiocb *iocb, unsigned short op, void *ptr)
+{
+       int ret;
+
+       BUG_ON(!is_kernel_kiocb(iocb));
+       BUG_ON(!iocb->ki_obj.complete);
+       BUG_ON(!iocb->ki_filp);
+
+       ret = aio_run_iocb(iocb, op, ptr, 0);
+
+       if (ret)
+               aio_kernel_free(iocb);
+
+       return ret;
+}
+EXPORT_SYMBOL_GPL(aio_kernel_submit);
+
 static int io_submit_one(struct kioctx *ctx, struct iocb __user *user_iocb,
                         struct iocb *iocb, bool compat)
 {

diff --git a/include/linux/aio.h b/include/linux/aio.h
index d9c92daa3944e43a13f285a7baaa1443868cce3d..9f4b10a81f8a020d698bb0c239537e1bff9faca1 100644 (file)
--- a/include/linux/aio.h
+++ b/include/linux/aio.h
@@ -31,13 +31,15 @@ typedef int (kiocb_cancel_fn)(struct kiocb *);
 
 struct kiocb {
        struct file             *ki_filp;
-       struct kioctx           *ki_ctx;        /* NULL for sync ops */
+       struct kioctx           *ki_ctx;        /* NULL for sync ops,
+                                                * -1 for kernel caller */
        kiocb_cancel_fn         *ki_cancel;
        void                    *private;
 
        union {
                void __user             *user;
                struct task_struct      *tsk;
+               void                    (*complete)(u64 user_data, long res);
        } ki_obj;
 
        __u64                   ki_user_data;   /* user's data for completion */
@@ -59,6 +61,11 @@ static inline bool is_sync_kiocb(struct kiocb *kiocb)
        return kiocb->ki_ctx == NULL;
 }
 
+static inline bool is_kernel_kiocb(struct kiocb *kiocb)
+{
+       return kiocb->ki_ctx == (void *)-1;
+}
+
 static inline void init_sync_kiocb(struct kiocb *kiocb, struct file *filp)
 {
        *kiocb = (struct kiocb) {
@@ -77,6 +84,14 @@ extern void exit_aio(struct mm_struct *mm);
 extern long do_io_submit(aio_context_t ctx_id, long nr,
                         struct iocb __user *__user *iocbpp, bool compat);
 void kiocb_set_cancel_fn(struct kiocb *req, kiocb_cancel_fn *cancel);
+struct kiocb *aio_kernel_alloc(gfp_t gfp);
+void aio_kernel_free(struct kiocb *iocb);
+void aio_kernel_init_rw(struct kiocb *iocb, struct file *filp, size_t nr,
+                       loff_t off);
+void aio_kernel_init_callback(struct kiocb *iocb,
+                             void (*complete)(u64 user_data, long res),
+                             u64 user_data);
+int aio_kernel_submit(struct kiocb *iocb, unsigned short op, void *ptr);
 #else
 static inline ssize_t wait_on_sync_kiocb(struct kiocb *iocb) { return 0; }
 static inline void aio_complete(struct kiocb *iocb, long res, long res2) { }