User should not be able to write block device if it is read-only at
block level (e.g force_ro attribute). This is ensured in the regular
fops write operation (blkdev_write_iter) but not when writing via
user mapping (mmap), allowing user to actually write a read-only
block device via a PROT_WRITE mapping.
Example: This can lead to integrity issue of eMMC boot partition
(e.g mmcblk0boot0) which is read-only by default.
To fix this issue, simply deny shared writable mapping if the block
is readonly.
Note: Block remains writable if switch to read-only is performed
after the initial mapping, but this is expected behavior according
to commit 
a32e236eb93e ("Partially revert "block: fail op_is_write()
requests to read-only partitions"")'.
Signed-off-by: Loic Poulain <loic.poulain@linaro.org>
Reviewed-by: Christoph Hellwig <hch@lst.de>
Link: https://lore.kernel.org/r/20230510074223.991297-1-loic.poulain@linaro.org
Signed-off-by: Jens Axboe <axboe@kernel.dk>
        return error;
 }
 
+static int blkdev_mmap(struct file *file, struct vm_area_struct *vma)
+{
+       struct inode *bd_inode = bdev_file_inode(file);
+
+       if (bdev_read_only(I_BDEV(bd_inode)))
+               return generic_file_readonly_mmap(file, vma);
+
+       return generic_file_mmap(file, vma);
+}
+
 const struct file_operations def_blk_fops = {
        .open           = blkdev_open,
        .release        = blkdev_close,
        .read_iter      = blkdev_read_iter,
        .write_iter     = blkdev_write_iter,
        .iopoll         = iocb_bio_iopoll,
-       .mmap           = generic_file_mmap,
+       .mmap           = blkdev_mmap,
        .fsync          = blkdev_fsync,
        .unlocked_ioctl = blkdev_ioctl,
 #ifdef CONFIG_COMPAT