9pfs: local: read native symlinks when security-model=mapped

Directories attached using virtfs with security-model=mapped
may contain native symlinks

This can happen e.g. when booting from a rootfs directory tree
(usually with a writable overlay set up on the host side)

Currently, when security-model=mapped[-xattr|-file],
QEMU assumes that host-side "symlinks" are in the mapped format,
i.e. are regular files storing the linked path,
so it tries to open with O_NOFOLLOW
and fails with ELOOP on native symlinks

This patch introduces a fallback for such cases:
reuse security-model=[none|passthrough] else if branch logic
where readlink will be called for the path basename

Resolves: https://gitlab.com/qemu-project/qemu/-/issues/173
Signed-off-by: Andrey Erokhin <language.lawyer@gmail.com>
Reviewed-by: Greg Kurz <groug@kaod.org>
Link: https://lore.kernel.org/qemu-devel/3c35955d-a57e-4203-81c5-395146e23f83@gmail.com
Signed-off-by: Christian Schoenebeck <qemu_oss@crudebyte.com>

diff --git a/hw/9pfs/9p-local.c b/hw/9pfs/9p-local.c
index c3745f2839f26c69d85486ddb757c06d0711bda6..24cb1da90ae082be661c35d8355959cb278a5c94 100644 (file)
--- a/hw/9pfs/9p-local.c
+++ b/hw/9pfs/9p-local.c
@@ -469,12 +469,16 @@ static ssize_t local_readlink(FsContext *fs_ctx, V9fsPath *fs_path,
 
         fd = local_open_nofollow(fs_ctx, fs_path->data, O_RDONLY, 0);
         if (fd == -1) {
+            if (errno == ELOOP) {
+                goto native_symlink;
+            }
             return -1;
         }
         tsize = RETRY_ON_EINTR(read(fd, (void *)buf, bufsz));
         close_preserve_errno(fd);
     } else if ((fs_ctx->export_flags & V9FS_SM_PASSTHROUGH) ||
                (fs_ctx->export_flags & V9FS_SM_NONE)) {
+    native_symlink:;
         char *dirpath = g_path_get_dirname(fs_path->data);
         char *name = g_path_get_basename(fs_path->data);
         int dirfd;