]> git.ipfire.org Git - thirdparty/kernel/linux.git/commitdiff
ovl: concurrent copy up of regular files
authorAmir Goldstein <amir73il@gmail.com>
Tue, 17 Jan 2017 04:34:57 +0000 (06:34 +0200)
committerMiklos Szeredi <mszeredi@redhat.com>
Tue, 7 Feb 2017 14:47:14 +0000 (15:47 +0100)
Now that copy up of regular file is done using O_TMPFILE,
we don't need to hold rename_lock throughout copy up.

Use the copy up waitqueue to synchronize concurrent copy up
of the same file. Different regular files can be copied up
concurrently.

The upper dir inode_lock is taken instead of rename_lock,
because it is needed for lookup and later for linking the
temp file, but it is released while copying up data.

Suggested-by: Al Viro <viro@zeniv.linux.org.uk>
Signed-off-by: Amir Goldstein <amir73il@gmail.com>
Signed-off-by: Miklos Szeredi <mszeredi@redhat.com>
fs/overlayfs/copy_up.c

index 6e39e90b560523350b1eb01dc451a3f49ddcaacf..48eb8812ac5b2220e9b2420cfc8ba659a9cf3bfd 100644 (file)
@@ -291,7 +291,16 @@ static int ovl_copy_up_locked(struct dentry *workdir, struct dentry *upperdir,
                BUG_ON(upperpath.dentry != NULL);
                upperpath.dentry = temp;
 
-               err = ovl_copy_up_data(lowerpath, &upperpath, stat->size);
+               if (tmpfile) {
+                       inode_unlock(udir);
+                       err = ovl_copy_up_data(lowerpath, &upperpath,
+                                              stat->size);
+                       inode_lock_nested(udir, I_MUTEX_PARENT);
+               } else {
+                       err = ovl_copy_up_data(lowerpath, &upperpath,
+                                              stat->size);
+               }
+
                if (err)
                        goto out_cleanup;
        }
@@ -353,8 +362,6 @@ static int ovl_copy_up_one(struct dentry *parent, struct dentry *dentry,
        struct dentry *upperdir;
        const char *link = NULL;
        struct ovl_fs *ofs = dentry->d_sb->s_fs_info;
-       /* Should we copyup with O_TMPFILE or with workdir? */
-       bool tmpfile = S_ISREG(stat->mode) && ofs->tmpfile;
 
        if (WARN_ON(!workdir))
                return -EROFS;
@@ -374,6 +381,25 @@ static int ovl_copy_up_one(struct dentry *parent, struct dentry *dentry,
                        return PTR_ERR(link);
        }
 
+       /* Should we copyup with O_TMPFILE or with workdir? */
+       if (S_ISREG(stat->mode) && ofs->tmpfile) {
+               err = ovl_copy_up_start(dentry);
+               /* err < 0: interrupted, err > 0: raced with another copy-up */
+               if (unlikely(err)) {
+                       pr_debug("ovl_copy_up_start(%pd2) = %i\n", dentry, err);
+                       if (err > 0)
+                               err = 0;
+                       goto out_done;
+               }
+
+               inode_lock_nested(upperdir->d_inode, I_MUTEX_PARENT);
+               err = ovl_copy_up_locked(workdir, upperdir, dentry, lowerpath,
+                                        stat, link, &pstat, true);
+               inode_unlock(upperdir->d_inode);
+               ovl_copy_up_end(dentry);
+               goto out_done;
+       }
+
        err = -EIO;
        if (lock_rename(workdir, upperdir) != NULL) {
                pr_err("overlayfs: failed to lock workdir+upperdir\n");
@@ -386,9 +412,10 @@ static int ovl_copy_up_one(struct dentry *parent, struct dentry *dentry,
        }
 
        err = ovl_copy_up_locked(workdir, upperdir, dentry, lowerpath,
-                                stat, link, &pstat, tmpfile);
+                                stat, link, &pstat, false);
 out_unlock:
        unlock_rename(workdir, upperdir);
+out_done:
        do_delayed_call(&done);
 
        return err;