[PATCH 2/2] xen/privcmd: add PRIVCMD_MMAPBATCH_V2 ioctl

[David Vrabel <david.vrabel@citrix.com>]

From: David Vrabel <david.vrabel@citrix.com>

PRIVCMD_MMAPBATCH_V2 extends PRIVCMD_MMAPBATCH with an additional
field for reporting the error code for every frame that could not be
mapped.  libxc prefers PRIVCMD_MMAPBATCH_V2 over PRIVCMD_MMAPBATCH.

Signed-off-by: David Vrabel <david.vrabel@citrix.com>
---
 drivers/xen/privcmd.c |   99 +++++++++++++++++++++++++++++++++++++++---------
 include/xen/privcmd.h |   23 +++++++++++-
 2 files changed, 102 insertions(+), 20 deletions(-)

diff --git a/drivers/xen/privcmd.c b/drivers/xen/privcmd.c
index ccee0f1..c0e89e7 100644
--- a/drivers/xen/privcmd.c
+++ b/drivers/xen/privcmd.c
@@ -76,7 +76,7 @@ static void free_page_list(struct list_head *pages)
  */
 static int gather_array(struct list_head *pagelist,
 			unsigned nelem, size_t size,
-			void __user *data)
+			const void __user *data)
 {
 	unsigned pageidx;
 	void *pagedata;
@@ -248,18 +248,37 @@ struct mmap_batch_state {
 	struct vm_area_struct *vma;
 	int err;
 
-	xen_pfn_t __user *user;
+	xen_pfn_t __user *user_mfn;
+	int __user *user_err;
 };
 
 static int mmap_batch_fn(void *data, void *state)
 {
 	xen_pfn_t *mfnp = data;
 	struct mmap_batch_state *st = state;
+	int ret;
 
-	if (xen_remap_domain_mfn_range(st->vma, st->va & PAGE_MASK, *mfnp, 1,
-				       st->vma->vm_page_prot, st->domain) < 0) {
-		*mfnp |= 0xf0000000U;
-		st->err++;
+	ret = xen_remap_domain_mfn_range(st->vma, st->va & PAGE_MASK, *mfnp, 1,
+					 st->vma->vm_page_prot, st->domain);
+	if (ret < 0) {
+		/*
+		 * Error reporting is a mess but userspace relies on
+		 * it behaving this way.
+		 *
+		 * V2 needs to a) return the result of each frame's
+		 * remap; and b) return -ENOENT if any frame failed
+		 * with -ENOENT.
+		 *
+		 * In this first pass the error code is saved by
+		 * overwriting the mfn and an error is indicated in
+		 * st->err.
+		 *
+		 * The second pass by mmap_return_errors() will write
+		 * the error codes to user space and get the right
+		 * ioctl return value.
+		 */
+		*(int *)mfnp = ret;
+		st->err = ret;
 	}
 	st->va += PAGE_SIZE;
 
@@ -270,16 +289,33 @@ static int mmap_return_errors(void *data, void *state)
 {
 	xen_pfn_t *mfnp = data;
 	struct mmap_batch_state *st = state;
+	int ret;
+
+	if (st->user_err) {
+		int err = *(int *)mfnp;
+
+		if (err == -ENOENT)
+			st->err = err;
 
-	return put_user(*mfnp, st->user++);
+		return __put_user(err, st->user_err++);
+	} else {
+		xen_pfn_t mfn;
+
+		ret = __get_user(mfn, st->user_mfn);
+		if (ret < 0)
+			return ret;
+
+		mfn |= PRIVCMD_MMAPBATCH_MFN_ERROR;
+		return __put_user(mfn, st->user_mfn++);
+	}
 }
 
 static struct vm_operations_struct privcmd_vm_ops;
 
-static long privcmd_ioctl_mmap_batch(void __user *udata)
+static long privcmd_ioctl_mmap_batch(void __user *udata, int version)
 {
 	int ret;
-	struct privcmd_mmapbatch m;
+	struct privcmd_mmapbatch_v2 m;
 	struct mm_struct *mm = current->mm;
 	struct vm_area_struct *vma;
 	unsigned long nr_pages;
@@ -289,15 +325,31 @@ static long privcmd_ioctl_mmap_batch(void __user *udata)
 	if (!xen_initial_domain())
 		return -EPERM;
 
-	if (copy_from_user(&m, udata, sizeof(m)))
-		return -EFAULT;
+	switch (version) {
+	case 1:
+		if (copy_from_user(&m, udata, sizeof(struct privcmd_mmapbatch)))
+			return -EFAULT;
+		/* Returns per-frame error in m.arr. */
+		m.err = NULL;
+		if (!access_ok(VERIFY_WRITE, m.arr, m.num * sizeof(*m.arr)))
+			return -EFAULT;
+		break;
+	case 2:
+		if (copy_from_user(&m, udata, sizeof(struct privcmd_mmapbatch_v2)))
+			return -EFAULT;
+		/* Returns per-frame error code in m.err. */
+		if (!access_ok(VERIFY_WRITE, m.err, m.num * (sizeof(*m.err))))
+			return -EFAULT;
+		break;
+	default:
+		return -EINVAL;
+	}
 
 	nr_pages = m.num;
 	if ((m.num <= 0) || (nr_pages > (LONG_MAX >> PAGE_SHIFT)))
 		return -EINVAL;
 
-	ret = gather_array(&pagelist, m.num, sizeof(xen_pfn_t),
-			   m.arr);
+	ret = gather_array(&pagelist, m.num, sizeof(xen_pfn_t), m.arr);
 
 	if (ret || list_empty(&pagelist))
 		goto out;
@@ -325,12 +377,17 @@ static long privcmd_ioctl_mmap_batch(void __user *udata)
 
 	up_write(&mm->mmap_sem);
 
-	if (state.err > 0) {
-		state.user = m.arr;
+	if (state.err) {
+		state.err = 0;
+		state.user_mfn = (xen_pfn_t *)m.arr;
+		state.user_err = m.err;
 		ret = traverse_pages(m.num, sizeof(xen_pfn_t),
-			       &pagelist,
-			       mmap_return_errors, &state);
-	}
+				     &pagelist,
+				     mmap_return_errors, &state);
+		if (ret >= 0)
+			ret = state.err;
+	} else if (m.err)
+		__clear_user(m.err, m.num * sizeof(*m.err));
 
 out:
 	free_page_list(&pagelist);
@@ -354,7 +411,11 @@ static long privcmd_ioctl(struct file *file,
 		break;
 
 	case IOCTL_PRIVCMD_MMAPBATCH:
-		ret = privcmd_ioctl_mmap_batch(udata);
+		ret = privcmd_ioctl_mmap_batch(udata, 1);
+		break;
+
+	case IOCTL_PRIVCMD_MMAPBATCH_V2:
+		ret = privcmd_ioctl_mmap_batch(udata, 2);
 		break;
 
 	default:
diff --git a/include/xen/privcmd.h b/include/xen/privcmd.h
index 17857fb..f60d75c 100644
--- a/include/xen/privcmd.h
+++ b/include/xen/privcmd.h
@@ -59,13 +59,32 @@ struct privcmd_mmapbatch {
 	int num;     /* number of pages to populate */
 	domid_t dom; /* target domain */
 	__u64 addr;  /* virtual address */
-	xen_pfn_t __user *arr; /* array of mfns - top nibble set on err */
+	xen_pfn_t __user *arr; /* array of mfns - or'd with
+				  PRIVCMD_MMAPBATCH_MFN_ERROR on err */
+};
+
+#define PRIVCMD_MMAPBATCH_MFN_ERROR 0xf0000000U
+
+struct privcmd_mmapbatch_v2 {
+	unsigned int num; /* number of pages to populate */
+	domid_t dom;      /* target domain */
+	__u64 addr;       /* virtual address */
+	const xen_pfn_t __user *arr; /* array of mfns */
+	int __user *err;  /* array of error codes */
 };
 
 /*
  * @cmd: IOCTL_PRIVCMD_HYPERCALL
  * @arg: &privcmd_hypercall_t
  * Return: Value returned from execution of the specified hypercall.
+ *
+ * @cmd: IOCTL_PRIVCMD_MMAPBATCH_V2
+ * @arg: &struct privcmd_mmapbatch_v2
+ * Return: 0 on success (i.e., arg->err contains valid error codes for
+ * each frame).  On an error other than a failed frame remap, -1 is
+ * returned and errno is set to EINVAL, EFAULT etc.  As an exception,
+ * if the operation was otherwise successful but any frame failed with
+ * -ENOENT, then -1 is returned and errno is set to ENOENT.
  */
 #define IOCTL_PRIVCMD_HYPERCALL					\
 	_IOC(_IOC_NONE, 'P', 0, sizeof(struct privcmd_hypercall))
@@ -73,5 +92,7 @@ struct privcmd_mmapbatch {
 	_IOC(_IOC_NONE, 'P', 2, sizeof(struct privcmd_mmap))
 #define IOCTL_PRIVCMD_MMAPBATCH					\
 	_IOC(_IOC_NONE, 'P', 3, sizeof(struct privcmd_mmapbatch))
+#define IOCTL_PRIVCMD_MMAPBATCH_V2				\
+	_IOC(_IOC_NONE, 'P', 4, sizeof(struct privcmd_mmapbatch_v2))
 
 #endif /* __LINUX_PUBLIC_PRIVCMD_H__ */
-- 
1.7.2.5