[PATCH v3 08/10] mm/shmem_luo: Allow preserving tmpfs files that are still linked

[David Matlack]

Allow preserving tmpfs files that are still linked. To preserve any
file, require permission to write to it and the file to be opened for
reading, and reject append-only files.

shmem_luo already accepts any shmem file without links, which includes
unlinked tmpfs files and files created with O_TMPFILE. Accepting linked
files means that userspace doesn't have to unlink files to preserve them
and can keep using them by path until the live update.

The path of the file isn't preserved. Like other files from tmpfs
mounts, a linked file is retrieved onto a tmpfs mount chosen by
userspace, which can then link it at the path it wants.

While a file is preserved, its size is frozen, so truncating it,
fallocate() and writes past its end fail with -EPERM. For a linked file,
this affects everyone who opens the file by its path, so a caller that
can only read the file, e.g. a world-readable file in /dev/shm, could
use it to break the owner of the file. So require permission to write to
the file to preserve it, for all files. Like FS_IOC_ENABLE_VERITY, don't
require a file opened for writing, so that binaries that are being
executed can still be preserved through read-only fds. Also don't
require a writable mount, since preserving a file doesn't write to the
filesystem. If the size of the file can't change anyway, because it is
immutable or on a read-only superblock, also let its owner preserve it.
Don't let others do that, since the size could change again once the
file is made writable. Like FS_IOC_ENABLE_VERITY, reject append-only
files, since their writers can only append to them, and freezing their
size would stop all appends.

Also like FS_IOC_ENABLE_VERITY, require the file to be opened for
reading. The retrieved file can be read, so otherwise a caller that can
only write to a file could preserve it to read its contents.

Preserving a file also fills its holes, but don't require any permission
for that. Anyone who can read a tmpfs file can already fill its holes by
faulting them in through a read-only mapping, also on read-only mounts
and for immutable files.

Document the frozen size and the permission requirements.

Signed-off-by: David Matlack <dmatlack@xxxxxxxxxx>
---
 mm/shmem_luo.c | 67 +++++++++++++++++++++++++++++++++++++++++++++++---
 1 file changed, 64 insertions(+), 3 deletions(-)

diff --git a/mm/shmem_luo.c b/mm/shmem_luo.c
index ecbeb37d33a8..0c012f242fc5 100644
--- a/mm/shmem_luo.c
+++ b/mm/shmem_luo.c
@@ -28,6 +28,24 @@
  * the file are preserved. All others are reset to default. The preserved
  * properties are described below.
  *
+ * A tmpfs file can be preserved even if it is linked, but its path is not
+ * preserved. Userspace chooses where to put the file when retrieving it, as
+ * described in "Retrieving Tmpfs Files" below.
+ *
+ * Preserving a file requires it to be opened for reading, since the retrieved
+ * file can be read. It also requires permission to write to the file, since
+ * preserving it affects all of its users, as described below. But the file
+ * doesn't need to be opened for writing. For example, a binary that is being
+ * executed can be preserved through a read-only file descriptor. The owner of
+ * a file whose size can't change anyway, because it is immutable or on a
+ * read-only superblock, may also preserve it. The file can be preserved
+ * through a read-only mount. Append-only files can't be preserved.
+ *
+ * While a file is preserved, its size can't change and no holes can be punched
+ * into it. Truncating the file, ``fallocate()`` and writes past the end of the
+ * file fail with ``-EPERM``. This applies to all users of the file, including
+ * users that open it by its path.
+ *
  * .. note::
  *    The LUO API is not stabilized yet, so the preserved properties of a memfd
  *    or tmpfs file are also not stable and are subject to backwards
@@ -298,6 +316,45 @@ static void shmem_luo_unpreserve_folios(struct kho_vmalloc *kho_vmalloc,
 	vfree(folios_ser);
 }
 
+/*
+ * Check whether the caller may preserve @file. The retrieved file can be read,
+ * so require @file to be opened for reading, like FS_IOC_ENABLE_VERITY does.
+ *
+ * Preserving the file freezes its size until it is unpreserved, for all users
+ * of the file, so require permission to write to the file. Like
+ * FS_IOC_ENABLE_VERITY, don't require a file opened for writing, so that e.g.
+ * a binary that is being executed can be preserved. Also don't require a
+ * writable mount, since preserving the file doesn't write to the filesystem.
+ * If the size of the file can't change anyway, because it is on a read-only
+ * superblock or immutable, its owner may also preserve it. The size could only
+ * change again after the file is made writable, which is up to the owner or
+ * root. Reject append-only files: Their writers can only append to them, and
+ * freezing the size would stop all appends.
+ *
+ * Preserving the file also fills its holes, but that doesn't need any more
+ * permission: Anyone who can read the file can fill its holes by faulting
+ * them in through a read-only mapping.
+ */
+static int shmem_luo_may_preserve(struct file *file)
+{
+	struct mnt_idmap *idmap = file_mnt_idmap(file);
+	struct inode *inode = file_inode(file);
+	int err;
+
+	if (!(file->f_mode & FMODE_READ))
+		return -EBADF;
+
+	if (IS_APPEND(inode))
+		return -EPERM;
+
+	err = inode_permission(idmap, inode, MAY_WRITE);
+	if (err && (sb_rdonly(inode->i_sb) || IS_IMMUTABLE(inode)) &&
+	    inode_owner_or_capable(idmap, inode))
+		return 0;
+
+	return err;
+}
+
 static int shmem_luo_preserve(struct liveupdate_file_op_args *args)
 {
 	struct inode *inode = file_inode(args->file);
@@ -307,6 +364,11 @@ static int shmem_luo_preserve(struct liveupdate_file_op_args *args)
 	int err = 0, seals;
 
 	inode_lock(inode);
+
+	err = shmem_luo_may_preserve(args->file);
+	if (err)
+		goto err_unlock_inode;
+
 	shmem_freeze(inode, true);
 
 	/* Allocate the main serialization structure in preserved memory */
@@ -362,6 +424,7 @@ static int shmem_luo_preserve(struct liveupdate_file_op_args *args)
 	kho_unpreserve_free(ser);
 err_unlock:
 	shmem_freeze(inode, false);
+err_unlock_inode:
 	inode_unlock(inode);
 	return err;
 }
@@ -710,9 +773,7 @@ static int shmem_luo_retrieve(struct liveupdate_file_op_args *args)
 static bool shmem_luo_can_preserve(struct liveupdate_file_handler *handler,
 				   struct file *file)
 {
-	struct inode *inode = file_inode(file);
-
-	return shmem_file(file) && !inode->i_nlink;
+	return shmem_file(file);
 }
 
 static unsigned long shmem_luo_get_id(struct file *file)
-- 
2.56.0.385.gac671c819a-goog