aufs: virtual or vertical directory 1/2, intro

This commit is just to prepare for the succeeding commit, and split to
suppress the size of a single commit.
Signed-off-by: J. R. Okajima <hooanon05g@gmail.com>

Documentation/filesystems/aufs/design/02struct.txt

@@ -128,6 +128,36 @@ As well as XINO files, aufs has a feature to truncate/refresh XIB to
 reduce the number of consumed disk blocks for these files.
 
 
+Virtual or Vertical Dir, and Readdir in Userspace
+----------------------------------------------------------------------
+In order to support multiple layers (branches), aufs readdir operation
+constructs a virtual dir block on memory. For readdir, aufs calls
+vfs_readdir() internally for each dir on branches, merges their entries
+with eliminating the whiteout-ed ones, and sets it to file (dir)
+object. So the file object has its entry list until it is closed. The
+entry list will be updated when the file position is zero and becomes
+obsoleted. This decision is made in aufs automatically.
+
+The dynamically allocated memory block for the name of entries has a
+unit of 512 bytes (by default) and stores the names contiguously (no
+padding). Another block for each entry is handled by kmem_cache too.
+During building dir blocks, aufs creates hash list and judging whether
+the entry is whiteouted by its upper branch or already listed.
+The merged result is cached in the corresponding inode object and
+maintained by a customizable life-time option.
+
+Some people may call it can be a security hole or invite DoS attack
+since the opened and once readdir-ed dir (file object) holds its entry
+list and becomes a pressure for system memory. But I'd say it is similar
+to files under /proc or /sys. The virtual files in them also holds a
+memory page (generally) while they are opened. When an idea to reduce
+memory for them is introduced, it will be applied to aufs too.
+For those who really hate this situation, I've developed readdir(3)
+library which operates this merging in userspace. You just need to set
+LD_PRELOAD environment variable, and aufs will not consume no memory in
+kernel space for readdir(3).
+
+
 Workqueue
 ----------------------------------------------------------------------
 Aufs sometimes requires privilege access to a branch. For instance,

fs/aufs/Makefile

@@ -16,7 +16,7 @@ aufs-y := module.o sbinfo.o super.o branch.o xino.o sysaufs.o opts.o \
 	dinfo.o dentry.o \
 	dynop.o \
 	finfo.o file.o \
-	dir.o \
+	dir.o vdir.o \
 	iinfo.o inode.o i_op.o
 
 # all are boolean

fs/aufs/debug.c

@@ -55,6 +55,45 @@ char *au_plevel = KERN_DEBUG;
 
 /* ---------------------------------------------------------------------- */
 
+void au_dpri_whlist(struct au_nhash *whlist)
+{
+	unsigned long ul, n;
+	struct hlist_head *head;
+	struct au_vdir_wh *pos;
+
+	n = whlist->nh_num;
+	head = whlist->nh_head;
+	for (ul = 0; ul < n; ul++) {
+		hlist_for_each_entry(pos, head, wh_hash)
+			dpri("b%d, %.*s, %d\n",
+			     pos->wh_bindex,
+			     pos->wh_str.len, pos->wh_str.name,
+			     pos->wh_str.len);
+		head++;
+	}
+}
+
+void au_dpri_vdir(struct au_vdir *vdir)
+{
+	unsigned long ul;
+	union au_vdir_deblk_p p;
+	unsigned char *o;
+
+	if (!vdir || IS_ERR(vdir)) {
+		dpri("err %ld\n", PTR_ERR(vdir));
+		return;
+	}
+
+	dpri("deblk %u, nblk %lu, deblk %p, last{%lu, %p}, ver %llu\n",
+	     vdir->vd_deblk_sz, vdir->vd_nblk, vdir->vd_deblk,
+	     vdir->vd_last.ul, vdir->vd_last.p.deblk, vdir->vd_version);
+	for (ul = 0; ul < vdir->vd_nblk; ul++) {
+		p.deblk = vdir->vd_deblk[ul];
+		o = p.deblk;
+		dpri("[%lu]: %p\n", ul, o);
+	}
+}
+
 static int do_pri_inode(aufs_bindex_t bindex, struct inode *inode, int hn,
 			struct dentry *wh)
 {
@@ -206,6 +245,9 @@ static int do_pri_file(aufs_bindex_t bindex, struct file *file)
 void au_dpri_file(struct file *file)
 {
 	struct au_finfo *finfo;
+	struct au_fidir *fidir;
+	struct au_hfile *hfile;
+	aufs_bindex_t bindex;
 	int err;
 
 	err = do_pri_file(-1, file);
@@ -219,7 +261,16 @@ void au_dpri_file(struct file *file)
 		return;
 	if (finfo->fi_btop < 0)
 		return;
+	fidir = finfo->fi_hdir;
+	if (!fidir)
 		do_pri_file(finfo->fi_btop, finfo->fi_htop.hf_file);
+	else
+		for (bindex = finfo->fi_btop;
+		     bindex >= 0 && bindex <= fidir->fd_bbot;
+		     bindex++) {
+			hfile = fidir->fd_hfile + bindex;
+			do_pri_file(bindex, hfile ? hfile->hf_file : NULL);
+		}
 }
 
 static int do_pri_br(aufs_bindex_t bindex, struct au_branch *br)

fs/aufs/debug.h

@@ -97,6 +97,10 @@ struct dentry;
 #ifdef CONFIG_AUFS_DEBUG
 extern struct mutex au_dbg_mtx;
 extern char *au_plevel;
+struct au_nhash;
+void au_dpri_whlist(struct au_nhash *whlist);
+struct au_vdir;
+void au_dpri_vdir(struct au_vdir *vdir);
 struct inode;
 void au_dpri_inode(struct inode *inode);
 void au_dpri_dalias(struct inode *inode);
@@ -111,6 +115,20 @@ void __au_dbg_verify_dinode(struct dentry *dentry, const char *func, int line);
 void au_dbg_verify_gen(struct dentry *parent, unsigned int sigen);
 void au_dbg_verify_kthread(void);
 
+#define AuDbgWhlist(w) do { \
+	mutex_lock(&au_dbg_mtx); \
+	AuDbg(#w "\n"); \
+	au_dpri_whlist(w); \
+	mutex_unlock(&au_dbg_mtx); \
+} while (0)
+
+#define AuDbgVdir(v) do { \
+	mutex_lock(&au_dbg_mtx); \
+	AuDbg(#v "\n"); \
+	au_dpri_vdir(v); \
+	mutex_unlock(&au_dbg_mtx); \
+} while (0)
+
 #define AuDbgInode(i) do { \
 	mutex_lock(&au_dbg_mtx); \
 	AuDbg(#i "\n"); \
@@ -156,6 +174,8 @@ AuStubVoid(au_dbg_verify_dinode, struct dentry *dentry)
 AuStubVoid(au_dbg_verify_gen, struct dentry *parent, unsigned int sigen)
 AuStubVoid(au_dbg_verify_kthread, void)
 
+#define AuDbgWhlist(w)		do {} while (0)
+#define AuDbgVdir(v)		do {} while (0)
 #define AuDbgInode(i)		do {} while (0)
 #define AuDbgDAlias(i)		do {} while (0)
 #define AuDbgDentry(d)		do {} while (0)

fs/aufs/dir.c

@@ -39,6 +39,49 @@ void au_sub_nlink(struct inode *dir, struct inode *h_dir)
 	set_nlink(dir, nlink);
 }
 
+loff_t au_dir_size(struct file *file, struct dentry *dentry)
+{
+	loff_t sz;
+	aufs_bindex_t bindex, bbot;
+	struct file *h_file;
+	struct dentry *h_dentry;
+
+	sz = 0;
+	if (file) {
+		AuDebugOn(!d_is_dir(file->f_path.dentry));
+
+		bbot = au_fbbot_dir(file);
+		for (bindex = au_fbtop(file);
+		     bindex <= bbot && sz < KMALLOC_MAX_SIZE;
+		     bindex++) {
+			h_file = au_hf_dir(file, bindex);
+			if (h_file && file_inode(h_file))
+				sz += vfsub_f_size_read(h_file);
+		}
+	} else {
+		AuDebugOn(!dentry);
+		AuDebugOn(!d_is_dir(dentry));
+
+		bbot = au_dbtaildir(dentry);
+		for (bindex = au_dbtop(dentry);
+		     bindex <= bbot && sz < KMALLOC_MAX_SIZE;
+		     bindex++) {
+			h_dentry = au_h_dptr(dentry, bindex);
+			if (h_dentry && d_is_positive(h_dentry))
+				sz += i_size_read(d_inode(h_dentry));
+		}
+	}
+	if (sz < KMALLOC_MAX_SIZE)
+		sz = roundup_pow_of_two(sz);
+	if (sz > KMALLOC_MAX_SIZE)
+		sz = KMALLOC_MAX_SIZE;
+	else if (sz < NAME_MAX) {
+		BUILD_BUG_ON(AUFS_RDBLK_DEF < NAME_MAX);
+		sz = AUFS_RDBLK_DEF;
+	}
+	return sz;
+}
+
 struct au_dir_ts_arg {
 	struct dentry *dentry;
 	aufs_bindex_t brid;

fs/aufs/dir.h

@@ -16,10 +16,67 @@
 
 /* ---------------------------------------------------------------------- */
 
+/* need to be faster and smaller */
+
+struct au_nhash {
+	unsigned int		nh_num;
+	struct hlist_head	*nh_head;
+};
+
+struct au_vdir_destr {
+	unsigned char	len;
+	unsigned char	name[0];
+} __packed;
+
+struct au_vdir_dehstr {
+	struct hlist_node	hash;
+	struct au_vdir_destr	*str;
+	struct rcu_head		rcu;
+} ____cacheline_aligned_in_smp;
+
+struct au_vdir_de {
+	ino_t			de_ino;
+	unsigned char		de_type;
+	/* caution: packed */
+	struct au_vdir_destr	de_str;
+} __packed;
+
+struct au_vdir_wh {
+	struct hlist_node	wh_hash;
+	aufs_bindex_t		wh_bindex;
+	/* caution: packed */
+	struct au_vdir_destr	wh_str;
+} __packed;
+
+union au_vdir_deblk_p {
+	unsigned char		*deblk;
+	struct au_vdir_de	*de;
+};
+
+struct au_vdir {
+	unsigned char	**vd_deblk;
+	unsigned long	vd_nblk;
+	struct {
+		unsigned long		ul;
+		union au_vdir_deblk_p	p;
+	} vd_last;
+
+	u64		vd_version;
+	unsigned int	vd_deblk_sz;
+	unsigned long	vd_jiffy;
+	struct rcu_head	rcu;
+} ____cacheline_aligned_in_smp;
+
+/* ---------------------------------------------------------------------- */
+
 /* dir.c */
 void au_add_nlink(struct inode *dir, struct inode *h_dir);
 void au_sub_nlink(struct inode *dir, struct inode *h_dir);
+loff_t au_dir_size(struct file *file, struct dentry *dentry);
 void au_dir_ts(struct inode *dir, aufs_bindex_t bsrc);
 
+/* vdir.c */
+void au_vdir_free(struct au_vdir *vdir);
+
 #endif /* __KERNEL__ */
 #endif /* __AUFS_DIR_H__ */

fs/aufs/iinfo.c

@@ -198,6 +198,7 @@ int au_iinfo_init(struct inode *inode)
 		iinfo->ii_generation.ig_generation = au_sigen(sb);
 		iinfo->ii_btop = -1;
 		iinfo->ii_bbot = -1;
+		iinfo->ii_vdir = NULL;
 		return 0;
 	}
 	return -ENOMEM;
@@ -251,6 +252,9 @@ void au_iinfo_fin(struct inode *inode)
 	}
 
 	iinfo = au_ii(inode);
+	if (iinfo->ii_vdir)
+		au_vdir_free(iinfo->ii_vdir);
+
 	bindex = iinfo->ii_btop;
 	if (bindex >= 0) {
 		hi = au_hinode(iinfo, bindex);

fs/aufs/inode.h

@@ -52,6 +52,7 @@ struct au_iigen {
 	__u32		ig_generation, ig_flags;
 };
 
+struct au_vdir;
 struct au_iinfo {
 	struct au_iigen		ii_generation;
 	struct super_block	*ii_hsb1;	/* no get/put */
@@ -60,6 +61,7 @@ struct au_iinfo {
 	aufs_bindex_t		ii_btop, ii_bbot;
 	__u32			ii_higen;
 	struct au_hinode	*ii_hinode;
+	struct au_vdir		*ii_vdir;
 };
 
 struct au_icntnr {
@@ -349,6 +351,12 @@ static inline aufs_bindex_t au_ibbot(struct inode *inode)
 	return au_ii(inode)->ii_bbot;
 }
 
+static inline struct au_vdir *au_ivdir(struct inode *inode)
+{
+	IiMustAnyLock(inode);
+	return au_ii(inode)->ii_vdir;
+}
+
 static inline struct dentry *au_hi_wh(struct inode *inode, aufs_bindex_t bindex)
 {
 	IiMustAnyLock(inode);
@@ -367,6 +375,12 @@ static inline void au_set_ibbot(struct inode *inode, aufs_bindex_t bindex)
 	au_ii(inode)->ii_bbot = bindex;
 }
 
+static inline void au_set_ivdir(struct inode *inode, struct au_vdir *vdir)
+{
+	IiMustWriteLock(inode);
+	au_ii(inode)->ii_vdir = vdir;
+}
+
 static inline struct au_hinode *au_hi(struct inode *inode, aufs_bindex_t bindex)
 {
 	IiMustAnyLock(inode);

fs/aufs/module.c

@@ -101,6 +101,10 @@ static int __init au_cache_init(void)
 		au_cache[AuCache_FINFO] = AuCacheCtor(au_finfo,
 						      au_fi_init_once);
 	if (au_cache[AuCache_FINFO])
+		au_cache[AuCache_VDIR] = AuCache(au_vdir);
+	if (au_cache[AuCache_VDIR])
+		au_cache[AuCache_DEHSTR] = AuCache(au_vdir_dehstr);
+	if (au_cache[AuCache_DEHSTR])
 		return 0;
 
 	au_cache_fin();

fs/aufs/module.h

@@ -15,6 +15,7 @@
 #include <linux/slab.h>
 #include "debug.h"
 #include "dentry.h"
+#include "dir.h"
 #include "file.h"
 #include "inode.h"
 
@@ -106,6 +107,8 @@ enum {
 	AuCache_DINFO,
 	AuCache_ICNTNR,
 	AuCache_FINFO,
+	AuCache_VDIR,
+	AuCache_DEHSTR,
 	AuCache_HNOTIFY, /* must be last */
 	AuCache_Last
 };
@@ -139,6 +142,8 @@ extern struct kmem_cache *au_cache[AuCache_Last];
 AuCacheFuncs(dinfo, DINFO);
 AuCacheFuncs(icntnr, ICNTNR);
 AuCacheFuncs(finfo, FINFO);
+AuCacheFuncs(vdir, VDIR);
+AuCacheFuncs(vdir_dehstr, DEHSTR);
 #ifdef CONFIG_AUFS_HNOTIFY
 AuCacheFuncs(hnotify, HNOTIFY);
 #endif

fs/aufs/opts.c

@@ -17,6 +17,8 @@
 enum {
 	Opt_br,
 	Opt_add,
+	Opt_rdcache, Opt_rdblk, Opt_rdhash,
+	Opt_rdblk_def, Opt_rdhash_def,
 	Opt_xino, Opt_noxino,
 	Opt_trunc_xino, Opt_trunc_xino_v, Opt_notrunc_xino,
 	Opt_trunc_xino_path, Opt_itrunc_xino,
@@ -60,6 +62,12 @@ static match_table_t options = {
 	{Opt_dio, "dio"},
 	{Opt_nodio, "nodio"},
 
+	{Opt_rdcache, "rdcache=%d"},
+	{Opt_rdblk, "rdblk=%d"},
+	{Opt_rdblk_def, "rdblk=def"},
+	{Opt_rdhash, "rdhash=%d"},
+	{Opt_rdhash_def, "rdhash=def"},
+
 	{Opt_wbr_create, "create=%s"},
 	{Opt_wbr_create, "create_policy=%s"},
 	{Opt_wbr_copyup, "cpup=%s"},
@@ -436,6 +444,21 @@ static void dump_opts(struct au_opts *opts)
 				  u.add->bindex, u.add->pathname, u.add->perm,
 				  u.add->path.dentry);
 			break;
+		case Opt_rdcache:
+			AuDbg("rdcache %d\n", opt->rdcache);
+			break;
+		case Opt_rdblk:
+			AuDbg("rdblk %u\n", opt->rdblk);
+			break;
+		case Opt_rdblk_def:
+			AuDbg("rdblk_def\n");
+			break;
+		case Opt_rdhash:
+			AuDbg("rdhash %u\n", opt->rdhash);
+			break;
+		case Opt_rdhash_def:
+			AuDbg("rdhash_def\n");
+			break;
 		case Opt_xino:
 			u.xino = &opt->xino;
 			AuDbg("xino {%s %pD}\n", u.xino->path, u.xino->file);
@@ -723,6 +746,49 @@ int au_opts_parse(struct super_block *sb, char *str, struct au_opts *opts)
 			opt->type = token;
 			break;
 
+		case Opt_rdcache:
+			if (unlikely(match_int(&a->args[0], &n))) {
+				pr_err("bad integer in %s\n", opt_str);
+				break;
+			}
+			if (unlikely(n > AUFS_RDCACHE_MAX)) {
+				pr_err("rdcache must be smaller than %d\n",
+				       AUFS_RDCACHE_MAX);
+				break;
+			}
+			opt->rdcache = n;
+			err = 0;
+			opt->type = token;
+			break;
+		case Opt_rdblk:
+			if (unlikely(match_int(&a->args[0], &n)
+				     || n < 0
+				     || n > KMALLOC_MAX_SIZE)) {
+				pr_err("bad integer in %s\n", opt_str);
+				break;
+			}
+			if (unlikely(n && n < NAME_MAX)) {
+				pr_err("rdblk must be larger than %d\n",
+				       NAME_MAX);
+				break;
+			}
+			opt->rdblk = n;
+			err = 0;
+			opt->type = token;
+			break;
+		case Opt_rdhash:
+			if (unlikely(match_int(&a->args[0], &n)
+				     || n < 0
+				     || n * sizeof(struct hlist_head)
+				     > KMALLOC_MAX_SIZE)) {
+				pr_err("bad integer in %s\n", opt_str);
+				break;
+			}
+			opt->rdhash = n;
+			err = 0;
+			opt->type = token;
+			break;
+
 		case Opt_trunc_xino:
 		case Opt_notrunc_xino:
 		case Opt_noxino:
@@ -733,6 +799,8 @@ int au_opts_parse(struct super_block *sb, char *str, struct au_opts *opts)
 		case Opt_list_plink:
 		case Opt_dio:
 		case Opt_nodio:
+		case Opt_rdblk_def:
+		case Opt_rdhash_def:
 			err = 0;
 			opt->type = token;
 			break;
@@ -891,6 +959,23 @@ static int au_opt_simple(struct super_block *sb, struct au_opt *opt,
 		sbinfo->si_wbr_copyup_ops = au_wbr_copyup_ops + opt->wbr_copyup;
 		break;
 
+	case Opt_rdcache:
+		sbinfo->si_rdcache
+			= msecs_to_jiffies(opt->rdcache * MSEC_PER_SEC);
+		break;
+	case Opt_rdblk:
+		sbinfo->si_rdblk = opt->rdblk;
+		break;
+	case Opt_rdblk_def:
+		sbinfo->si_rdblk = AUFS_RDBLK_DEF;
+		break;
+	case Opt_rdhash:
+		sbinfo->si_rdhash = opt->rdhash;
+		break;
+	case Opt_rdhash_def:
+		sbinfo->si_rdhash = AUFS_RDHASH_DEF;
+		break;
+
 	case Opt_trunc_xino:
 		au_opt_set(sbinfo->si_mntflags, TRUNC_XINO);
 		break;

fs/aufs/opts.h

@@ -120,6 +120,9 @@ struct au_opt {
 		struct au_opt_xino	xino;
 		struct au_opt_xino_itrunc xino_itrunc;
 		struct au_opt_add	add;
+		int			rdcache;
+		unsigned int		rdblk;
+		unsigned int		rdhash;
 		int			udba;
 		struct au_opt_wbr_create wbr_create;
 		int			wbr_copyup;

fs/aufs/sbinfo.c

@@ -73,6 +73,10 @@ int au_si_alloc(struct super_block *sb)
 	mutex_init(&sbinfo->si_xib_mtx);
 	/* leave si_xib_last_pindex and si_xib_next_bit */
 
+	sbinfo->si_rdcache = msecs_to_jiffies(AUFS_RDCACHE_DEF * MSEC_PER_SEC);
+	sbinfo->si_rdblk = AUFS_RDBLK_DEF;
+	sbinfo->si_rdhash = AUFS_RDHASH_DEF;
+
 	for (i = 0; i < AuPlink_NHASH; i++)
 		INIT_HLIST_BL_HEAD(sbinfo->si_plink + i);
 	init_waitqueue_head(&sbinfo->si_plink_wq);

fs/aufs/super.h

@@ -112,6 +112,11 @@ struct au_sbinfo {
 	atomic_t		si_xigen_next;
 #endif
 
+	/* vdir parameters */
+	unsigned long		si_rdcache;	/* max cache time in jiffies */
+	unsigned int		si_rdblk;	/* deblk size */
+	unsigned int		si_rdhash;	/* hash size */
+
 	/* pseudo_link list */
 	struct hlist_bl_head	si_plink[AuPlink_NHASH];
 	wait_queue_head_t	si_plink_wq;

fs/aufs/vdir.c

+// SPDX-License-Identifier: GPL-2.0
+/*
+ * Copyright (C) 2005-2019 Junjiro R. Okajima
+ */
+
+/*
+ * virtual or vertical directory
+ */
+
+#include "aufs.h"
+
+void au_vdir_free(struct au_vdir *vdir)
+{
+	unsigned char **deblk;
+
+	deblk = vdir->vd_deblk;
+	while (vdir->vd_nblk--)
+		au_kfree_try_rcu(*deblk++);
+	au_kfree_try_rcu(vdir->vd_deblk);
+	au_cache_free_vdir(vdir);
+}

include/uapi/linux/aufs_type.h

@@ -73,6 +73,10 @@ typedef int16_t aufs_bindex_t;
 #define AUFS_XINO_DEFPATH	"/tmp/" AUFS_XINO_FNAME
 #define AUFS_XINO_DEF_SEC	30 /* seconds */
 #define AUFS_XINO_DEF_TRUNC	45 /* percentage */
+#define AUFS_RDCACHE_DEF	10 /* seconds */
+#define AUFS_RDCACHE_MAX	3600 /* seconds */
+#define AUFS_RDBLK_DEF		512 /* bytes */
+#define AUFS_RDHASH_DEF		32
 #define AUFS_WKQ_NAME		AUFS_NAME "d"
 #define AUFS_MFS_DEF_SEC	30 /* seconds */
 #define AUFS_MFS_MAX_SEC	3600 /* seconds */