io_uring: don't allow discontig pages for IORING_SETUP_NO_MMAP

io_sqes_map() is used rather than io_mem_alloc(), if the application
passes in memory for mapping rather than have the kernel allocate it and
then mmap(2) the ranges. This then calls __io_uaddr_map() to perform the
page mapping and pinning, which checks if we end up with the same pages,
if more than one page is mapped. But this check is incorrect and only
checks if the first and last pages are the same, where it really should
be checking if the mapped pages are contigous. This allows mapping a
single normal page, or a huge page range.

Down the line we can add support for remapping pages to be virtually
contigous, which is really all that io_uring cares about.

Cc: [email protected]
Fixes: 03d89a2 ("io_uring: support for user allocated memory for rings/sqes")
Reported-by: Jann Horn <[email protected]>
Signed-off-by: Jens Axboe <[email protected]>

Author: axboe

io_uring/io_uring.c

@@ -2697,6 +2697,7 @@ static void *__io_uaddr_map(struct page ***pages, unsigned short *npages,
 {
 	struct page **page_array;
 	unsigned int nr_pages;
+	void *page_addr;
 	int ret, i;
 
 	*npages = 0;
@@ -2718,27 +2719,29 @@ static void *__io_uaddr_map(struct page ***pages, unsigned short *npages,
 		io_pages_free(&page_array, ret > 0 ? ret : 0);
 		return ret < 0 ? ERR_PTR(ret) : ERR_PTR(-EFAULT);
 	}
-	/*
-	 * Should be a single page. If the ring is small enough that we can
-	 * use a normal page, that is fine. If we need multiple pages, then
-	 * userspace should use a huge page. That's the only way to guarantee
-	 * that we get contigious memory, outside of just being lucky or
-	 * (currently) having low memory fragmentation.
-	 */
-	if (page_array[0] != page_array[ret - 1])
-		goto err;
 
-	/*
-	 * Can't support mapping user allocated ring memory on 32-bit archs
-	 * where it could potentially reside in highmem. Just fail those with
-	 * -EINVAL, just like we did on kernels that didn't support this
-	 * feature.
-	 */
+	page_addr = page_address(page_array[0]);
 	for (i = 0; i < nr_pages; i++) {
-		if (PageHighMem(page_array[i])) {
-			ret = -EINVAL;
+		ret = -EINVAL;
+
+		/*
+		 * Can't support mapping user allocated ring memory on 32-bit
+		 * archs where it could potentially reside in highmem. Just
+		 * fail those with -EINVAL, just like we did on kernels that
+		 * didn't support this feature.
+		 */
+		if (PageHighMem(page_array[i]))
 			goto err;
-		}
+
+		/*
+		 * No support for discontig pages for now, should either be a
+		 * single normal page, or a huge page. Later on we can add
+		 * support for remapping discontig pages, for now we will
+		 * just fail them with EINVAL.
+		 */
+		if (page_address(page_array[i]) != page_addr)
+			goto err;
+		page_addr += PAGE_SIZE;
 	}
 
 	*pages = page_array;