Fix AUFS support / s_files (for Docker AUFS storage driver support)

fs/aufs/branch.c

@@ -568,23 +568,25 @@ static unsigned long long au_farray_cb(void *a,
 {
 	unsigned long long n;
 	struct file **p, *f;
+	struct au_sphlhead *files;
+	struct au_finfo *finfo;
 	struct super_block *sb = arg;
 
 	n = 0;
 	p = a;
-	lg_global_lock(&files_lglock);
-	do_file_list_for_each_entry(sb, f) {
-		if (au_fi(f)
-		    && file_count(f)
+	files = &au_sbi(sb)->si_files;
+	spin_lock(&files->spin);
+	hlist_for_each_entry(finfo, &files->head, fi_hlist) {
+		f = finfo->fi_file;
+		if (file_count(f)
 		    && !special_file(file_inode(f)->i_mode)) {
 			get_file(f);
 			*p++ = f;
 			n++;
 			AuDebugOn(n > max);
 		}
-	} while_file_list_for_each_entry;
-	lg_global_unlock(&files_lglock);
-
+	}
+	spin_unlock(&files->spin);
 	return n;
 }
 
@@ -1273,7 +1275,13 @@ static int au_br_mod_files_ro(struct super_block *sb, aufs_bindex_t bindex)
 			continue;
 
 		/* todo: already flushed? */
-		/* cf. fs/super.c:mark_files_ro() */
+		 /*
+		* fs/super.c:mark_files_ro() is gone, but aufs keeps its
+		* approach which resets f_mode and calls mnt_drop_write() and
+		* file_release_write() for each file, because the branch
+		* attribute in aufs world is totally different from the native
+		* fs rw/ro mode.
+		*/
 		/* fi_read_lock(file); */
 		hfile = &au_fi(file)->fi_htop;
 		hf = hfile->hf_file;

fs/aufs/f_op.c

@@ -35,6 +35,8 @@ int au_do_open_nondir(struct file *file, int flags)
 		au_set_fbstart(file, bindex);
 		au_set_h_fptr(file, bindex, h_file);
 		au_update_figen(file);
+		finfo->fi_file = file;
+		au_sphl_add(&finfo->fi_hlist, &au_sbi(dentry->d_sb)->si_files);
 		/* todo: necessary? */
 		/* file->f_ra = h_file->f_ra; */
 	}
@@ -65,6 +67,7 @@ int aufs_release_nondir(struct inode *inode __maybe_unused, struct file *file)
 	aufs_bindex_t bindex;
 
 	finfo = au_fi(file);
+	au_sphl_del(&finfo->fi_hlist, &au_sbi(file->f_dentry->d_sb)->si_files);
 	bindex = finfo->fi_btop;
 	if (bindex >= 0)
 		au_set_h_fptr(file, bindex, NULL);

fs/aufs/file.h

@@ -48,6 +48,8 @@ struct au_finfo {
 		atomic_t			fi_mmapped;
 	};
 	struct au_fidir		*fi_hdir;	/* for dir only */
+	struct hlist_node   fi_hlist;
+	struct file     *fi_file;   /* very ugly */
 } ____cacheline_aligned_in_smp;
 
 /* ---------------------------------------------------------------------- */

fs/aufs/sbinfo.c

@@ -105,6 +105,8 @@ int au_si_alloc(struct super_block *sb)
 	init_waitqueue_head(&sbinfo->si_plink_wq);
 	spin_lock_init(&sbinfo->si_plink_maint_lock);
 
+	au_sphl_init(&sbinfo->si_files);
+
 	/* leave other members for sysaufs and si_mnt. */
 	sbinfo->si_sb = sb;
 	sb->s_fs_info = sbinfo;

fs/aufs/super.h

@@ -96,7 +96,8 @@ struct au_sbinfo {
 	} au_si_pid;
 
 	/*
-	 * dirty approach to protect sb->sb_inodes and ->s_files from remount.
+	* dirty approach to protect sb->sb_inodes and ->s_files (gone) from
+	* remount.
 	 */
 	atomic_long_t		si_ninodes, si_nfiles;
 
@@ -177,6 +178,9 @@ struct au_sbinfo {
 	spinlock_t		si_plink_maint_lock;
 	pid_t			si_plink_maint_pid;
 
+	/* file list */
+	struct au_sphlhead  si_files;
+
 	/*
 	 * sysfs and lifetime management.
 	 * this is not a small structure and it may be a waste of memory in case

fs/aufs/sysrq.c

@@ -17,6 +17,8 @@ static void sysrq_sb(struct super_block *sb)
 	char *plevel;
 	struct au_sbinfo *sbinfo;
 	struct file *file;
+	struct au_sphlhead *files;
+	struct au_finfo *finfo;
 
 	plevel = au_plevel;
 	au_plevel = KERN_WARNING;
@@ -73,15 +75,16 @@ static void sysrq_sb(struct super_block *sb)
 	}
 #endif
 	pr("files\n");
-	lg_global_lock(&files_lglock);
-	do_file_list_for_each_entry(sb, file) {
+	files = &au_sbi(sb)->si_files;
+	spin_lock(&files->spin);
+	hlist_for_each_entry(finfo, &files->head, fi_hlist) {
 		umode_t mode;
-
+		file = finfo->fi_file;
 		mode = file_inode(file)->i_mode;
 		if (!special_file(mode))
 			au_dpri_file(file);
-	} while_file_list_for_each_entry;
-	lg_global_unlock(&files_lglock);
+		}
+		spin_unlock(&files->spin);
 	pr("done\n");
 
 #undef pr

fs/aufs/vfsub.h

@@ -23,37 +23,6 @@ extern struct lglock vfsmount_lock;
 extern void __mnt_drop_write(struct vfsmount *);
 extern spinlock_t inode_sb_list_lock;
 
-/* copied from linux/fs/file_table.c */
-extern struct lglock files_lglock;
-#ifdef CONFIG_SMP
-/*
- * These macros iterate all files on all CPUs for a given superblock.
- * files_lglock must be held globally.
- */
-#define do_file_list_for_each_entry(__sb, __file)		\
-{								\
-	int i;							\
-	for_each_possible_cpu(i) {				\
-		struct list_head *list;				\
-		list = per_cpu_ptr((__sb)->s_files, i);		\
-		list_for_each_entry((__file), list, f_u.fu_list)
-
-#define while_file_list_for_each_entry				\
-	}							\
-}
-
-#else
-
-#define do_file_list_for_each_entry(__sb, __file)		\
-{								\
-	struct list_head *list;					\
-	list = &(sb)->s_files;					\
-	list_for_each_entry((__file), list, f_u.fu_list)
-
-#define while_file_list_for_each_entry				\
-}
-#endif
-
 /* ---------------------------------------------------------------------- */
 
 /* lock subclass for lower inode */