aufs: dinfo core

The structure is very similar to aufs inode info (in previous commit).
Signed-off-by: J. R. Okajima <hooanon05g@gmail.com>

Documentation/filesystems/aufs/design/03lookup.txt

+
+# Copyright (C) 2005-2019 Junjiro R. Okajima
+
+Lookup in a Branch
+----------------------------------------------------------------------
+Since aufs has a character of sub-VFS (see Introduction), it operates
+lookup for branches as VFS does. It may be a heavy work. But almost all
+lookup operation in aufs is the simplest case, ie. lookup only an entry
+directly connected to its parent. Digging down the directory hierarchy
+is unnecessary. VFS has a function lookup_one_len() for that use, and
+aufs calls it.
+
+When a branch is a remote filesystem, aufs basically relies upon its
+->d_revalidate(), also aufs forces the hardest revalidate tests for
+them.
+For d_revalidate, aufs implements three levels of revalidate tests. See
+"Revalidate Dentry and UDBA" in detail.

fs/aufs/Makefile

@@ -10,6 +10,7 @@ ccflags-y += -include ${srctree}/include/uapi/linux/aufs_type.h
 
 obj-$(CONFIG_AUFS_FS) += aufs.o
 aufs-y := module.o super.o \
+	dinfo.o \
 	iinfo.o
 
 # all are boolean

fs/aufs/aufs.h

@@ -22,6 +22,8 @@
 
 #include "debug.h"
 
+#include "dcsub.h"
+#include "dentry.h"
 #include "fstype.h"
 #include "inode.h"
 #include "module.h"

fs/aufs/dcsub.h

+/* SPDX-License-Identifier: GPL-2.0 */
+/*
+ * Copyright (C) 2005-2019 Junjiro R. Okajima
+ */
+
+/*
+ * sub-routines for dentry cache
+ */
+
+#ifndef __AUFS_DCSUB_H__
+#define __AUFS_DCSUB_H__
+
+#ifdef __KERNEL__
+
+#include <linux/dcache.h>
+
+/*
+ * by the commit
+ * 360f547 2015-01-25 dcache: let the dentry count go down to zero without
+ *			taking d_lock
+ * the type of d_lockref.count became int, but the inlined function d_count()
+ * still returns unsigned int.
+ * I don't know why. Maybe it is for every d_count() users?
+ * Anyway au_dcount() lives on.
+ */
+static inline int au_dcount(struct dentry *d)
+{
+	return (int)d_count(d);
+}
+
+#endif /* __KERNEL__ */
+#endif /* __AUFS_DCSUB_H__ */

fs/aufs/debug.c

@@ -97,3 +97,87 @@ void au_dpri_inode(struct inode *inode)
 	for (bindex = iinfo->ii_btop; bindex <= iinfo->ii_bbot; bindex++)
 		do_pri_inode(bindex, iinfo->ii_hinode[0 + bindex].hi_inode);
 }
+
+void au_dpri_dalias(struct inode *inode)
+{
+	struct dentry *d;
+
+	spin_lock(&inode->i_lock);
+	hlist_for_each_entry(d, &inode->i_dentry, d_u.d_alias)
+		au_dpri_dentry(d);
+	spin_unlock(&inode->i_lock);
+}
+
+static int do_pri_dentry(aufs_bindex_t bindex, struct dentry *dentry)
+{
+	if (!dentry || IS_ERR(dentry)) {
+		dpri("d%d: err %ld\n", bindex, PTR_ERR(dentry));
+		return -1;
+	}
+	/* do not call dget_parent() here */
+	/* note: access d_xxx without d_lock */
+	dpri("d%d: %p, %pd2?, %s, cnt %d, flags 0x%x, %shashed\n",
+	     bindex, dentry, dentry,
+	     dentry->d_sb ? au_sbtype(dentry->d_sb) : "??",
+	     au_dcount(dentry), dentry->d_flags,
+	     d_unhashed(dentry) ? "un" : "");
+	do_pri_inode(bindex, d_inode(dentry));
+	return 0;
+}
+
+void au_dpri_dentry(struct dentry *dentry)
+{
+	struct au_dinfo *dinfo;
+	aufs_bindex_t bindex;
+	int err;
+
+	err = do_pri_dentry(-1, dentry);
+	if (err || !au_test_aufs(dentry->d_sb))
+		return;
+
+	dinfo = au_di(dentry);
+	if (!dinfo)
+		return;
+	dpri("d-1: btop %d, bbot %d\n",
+	     dinfo->di_btop, dinfo->di_bbot);
+	if (dinfo->di_btop < 0)
+		return;
+	for (bindex = dinfo->di_btop; bindex <= dinfo->di_bbot; bindex++)
+		do_pri_dentry(bindex, au_hdentry(dinfo, bindex)->hd_dentry);
+}
+
+/* ---------------------------------------------------------------------- */
+
+void __au_dbg_verify_dinode(struct dentry *dentry, const char *func, int line)
+{
+	struct inode *h_inode, *inode = d_inode(dentry);
+	struct dentry *h_dentry;
+	aufs_bindex_t bindex, bbot, bi;
+
+	if (!inode /* || au_di(dentry)->di_lsc == AuLsc_DI_TMP */)
+		return;
+
+	bbot = au_dbbot(dentry);
+	bi = au_ibbot(inode);
+	if (bi < bbot)
+		bbot = bi;
+	bindex = au_dbtop(dentry);
+	bi = au_ibtop(inode);
+	if (bi > bindex)
+		bindex = bi;
+
+	for (; bindex <= bbot; bindex++) {
+		h_dentry = au_h_dptr(dentry, bindex);
+		if (!h_dentry)
+			continue;
+		h_inode = au_h_iptr(inode, bindex);
+		if (unlikely(h_inode != d_inode(h_dentry))) {
+			au_debug_on();
+			AuDbg("b%d, %s:%d\n", bindex, func, line);
+			AuDbgDentry(dentry);
+			AuDbgInode(inode);
+			au_debug_off();
+			BUG();
+		}
+	}
+}

fs/aufs/debug.h

@@ -60,6 +60,11 @@ extern struct mutex au_dbg_mtx;
 extern char *au_plevel;
 struct inode;
 void au_dpri_inode(struct inode *inode);
+void au_dpri_dalias(struct inode *inode);
+void au_dpri_dentry(struct dentry *dentry);
+
+#define au_dbg_verify_dinode(d) __au_dbg_verify_dinode(d, __func__, __LINE__)
+void __au_dbg_verify_dinode(struct dentry *dentry, const char *func, int line);
 
 #define AuDbgInode(i) do { \
 	mutex_lock(&au_dbg_mtx); \
@@ -67,8 +72,26 @@ void au_dpri_inode(struct inode *inode);
 	au_dpri_inode(i); \
 	mutex_unlock(&au_dbg_mtx); \
 } while (0)
+
+#define AuDbgDAlias(i) do { \
+	mutex_lock(&au_dbg_mtx); \
+	AuDbg(#i "\n"); \
+	au_dpri_dalias(i); \
+	mutex_unlock(&au_dbg_mtx); \
+} while (0)
+
+#define AuDbgDentry(d) do { \
+	mutex_lock(&au_dbg_mtx); \
+	AuDbg(#d "\n"); \
+	au_dpri_dentry(d); \
+	mutex_unlock(&au_dbg_mtx); \
+} while (0)
 #else
+AuStubVoid(au_dbg_verify_dinode, struct dentry *dentry)
+
 #define AuDbgInode(i)		do {} while (0)
+#define AuDbgDAlias(i)		do {} while (0)
+#define AuDbgDentry(d)		do {} while (0)
 #endif /* CONFIG_AUFS_DEBUG */
 
 #endif /* __KERNEL__ */

fs/aufs/dentry.h

+/* SPDX-License-Identifier: GPL-2.0 */
+/*
+ * Copyright (C) 2005-2019 Junjiro R. Okajima
+ */
+
+/*
+ * lookup and dentry operations
+ */
+
+#ifndef __AUFS_DENTRY_H__
+#define __AUFS_DENTRY_H__
+
+#ifdef __KERNEL__
+
+#include <linux/dcache.h>
+#include "rwsem.h"
+
+struct au_hdentry {
+	struct dentry		*hd_dentry;
+};
+
+struct au_dinfo {
+	struct au_rwsem		di_rwsem;
+	aufs_bindex_t		di_btop, di_bbot;
+	struct au_hdentry	*di_hdentry;
+	struct rcu_head		rcu;
+} ____cacheline_aligned_in_smp;
+
+/* ---------------------------------------------------------------------- */
+
+/* dinfo.c */
+void au_di_init_once(void *_di);
+struct au_dinfo *au_di_alloc(struct super_block *sb, unsigned int lsc);
+void au_di_free(struct au_dinfo *dinfo);
+int au_di_init(struct dentry *dentry);
+void au_di_fin(struct dentry *dentry);
+
+void di_read_lock(struct dentry *d, int flags, unsigned int lsc);
+void di_read_unlock(struct dentry *d, int flags);
+void di_downgrade_lock(struct dentry *d, int flags);
+void di_write_lock(struct dentry *d, unsigned int lsc);
+void di_write_unlock(struct dentry *d);
+
+struct dentry *au_h_dptr(struct dentry *dentry, aufs_bindex_t bindex);
+
+void au_set_h_dptr(struct dentry *dentry, aufs_bindex_t bindex,
+		   struct dentry *h_dentry);
+void au_update_dbtop(struct dentry *dentry);
+void au_update_dbbot(struct dentry *dentry);
+
+/* ---------------------------------------------------------------------- */
+
+static inline struct au_dinfo *au_di(struct dentry *dentry)
+{
+	return dentry->d_fsdata;
+}
+
+/* ---------------------------------------------------------------------- */
+
+/* lock subclass for dinfo */
+enum {
+	AuLsc_DI_CHILD,		/* child first */
+	AuLsc_DI_CHILD2,	/* rename(2), link(2), and cpup at hnotify */
+	AuLsc_DI_CHILD3,	/* copyup dirs */
+	AuLsc_DI_PARENT,
+	AuLsc_DI_PARENT2,
+	AuLsc_DI_PARENT3,
+	AuLsc_DI_TMP		/* temp for replacing dinfo */
+};
+
+/*
+ * di_read_lock_child, di_write_lock_child,
+ * di_read_lock_child2, di_write_lock_child2,
+ * di_read_lock_child3, di_write_lock_child3,
+ * di_read_lock_parent, di_write_lock_parent,
+ * di_read_lock_parent2, di_write_lock_parent2,
+ * di_read_lock_parent3, di_write_lock_parent3,
+ */
+#define AuReadLockFunc(name, lsc) \
+static inline void di_read_lock_##name(struct dentry *d, int flags) \
+{ di_read_lock(d, flags, AuLsc_DI_##lsc); }
+
+#define AuWriteLockFunc(name, lsc) \
+static inline void di_write_lock_##name(struct dentry *d) \
+{ di_write_lock(d, AuLsc_DI_##lsc); }
+
+#define AuRWLockFuncs(name, lsc) \
+	AuReadLockFunc(name, lsc) \
+	AuWriteLockFunc(name, lsc)
+
+AuRWLockFuncs(child, CHILD);
+AuRWLockFuncs(child2, CHILD2);
+AuRWLockFuncs(child3, CHILD3);
+AuRWLockFuncs(parent, PARENT);
+AuRWLockFuncs(parent2, PARENT2);
+AuRWLockFuncs(parent3, PARENT3);
+
+#undef AuReadLockFunc
+#undef AuWriteLockFunc
+#undef AuRWLockFuncs
+
+#define DiMustNoWaiters(d)	AuRwMustNoWaiters(&au_di(d)->di_rwsem)
+#define DiMustAnyLock(d)	AuRwMustAnyLock(&au_di(d)->di_rwsem)
+#define DiMustWriteLock(d)	AuRwMustWriteLock(&au_di(d)->di_rwsem)
+
+/* ---------------------------------------------------------------------- */
+
+static inline void au_h_dentry_init(struct au_hdentry *hdentry)
+{
+	hdentry->hd_dentry = NULL;
+}
+
+static inline struct au_hdentry *au_hdentry(struct au_dinfo *di,
+					    aufs_bindex_t bindex)
+{
+	return di->di_hdentry + bindex;
+}
+
+static inline void au_hdput(struct au_hdentry *hd)
+{
+	if (hd)
+		dput(hd->hd_dentry);
+}
+
+static inline aufs_bindex_t au_dbtop(struct dentry *dentry)
+{
+	DiMustAnyLock(dentry);
+	return au_di(dentry)->di_btop;
+}
+
+static inline aufs_bindex_t au_dbbot(struct dentry *dentry)
+{
+	DiMustAnyLock(dentry);
+	return au_di(dentry)->di_bbot;
+}
+
+/* todo: hard/soft set? */
+static inline void au_set_dbtop(struct dentry *dentry, aufs_bindex_t bindex)
+{
+	DiMustWriteLock(dentry);
+	au_di(dentry)->di_btop = bindex;
+}
+
+static inline void au_set_dbbot(struct dentry *dentry, aufs_bindex_t bindex)
+{
+	DiMustWriteLock(dentry);
+	au_di(dentry)->di_bbot = bindex;
+}
+
+#endif /* __KERNEL__ */
+#endif /* __AUFS_DENTRY_H__ */

fs/aufs/dinfo.c

+// SPDX-License-Identifier: GPL-2.0
+/*
+ * Copyright (C) 2005-2019 Junjiro R. Okajima
+ */
+
+/*
+ * dentry private data
+ */
+
+#include "aufs.h"
+
+void au_di_init_once(void *_dinfo)
+{
+	struct au_dinfo *dinfo = _dinfo;
+
+	au_rw_init(&dinfo->di_rwsem);
+}
+
+struct au_dinfo *au_di_alloc(struct super_block *sb, unsigned int lsc)
+{
+	struct au_dinfo *dinfo;
+	int nbr;
+
+	dinfo = au_cache_alloc_dinfo();
+	if (unlikely(!dinfo))
+		goto out;
+
+	nbr = 1; /* re-commit later */
+	dinfo->di_hdentry = kcalloc(nbr, sizeof(*dinfo->di_hdentry), GFP_NOFS);
+	if (dinfo->di_hdentry) {
+		au_rw_write_lock_nested(&dinfo->di_rwsem, lsc);
+		dinfo->di_btop = -1;
+		dinfo->di_bbot = -1;
+		goto out;
+	}
+
+	au_cache_free_dinfo(dinfo);
+	dinfo = NULL;
+
+out:
+	return dinfo;
+}
+
+void au_di_free(struct au_dinfo *dinfo)
+{
+	struct au_hdentry *p;
+	aufs_bindex_t bbot, bindex;
+
+	/* dentry may not be revalidated */
+	bindex = dinfo->di_btop;
+	if (bindex >= 0) {
+		bbot = dinfo->di_bbot;
+		p = au_hdentry(dinfo, bindex);
+		while (bindex++ <= bbot)
+			au_hdput(p++);
+	}
+	au_kfree_try_rcu(dinfo->di_hdentry);
+	au_cache_free_dinfo(dinfo);
+}
+
+int au_di_init(struct dentry *dentry)
+{
+	int err;
+	struct super_block *sb;
+	struct au_dinfo *dinfo;
+
+	err = 0;
+	sb = dentry->d_sb;
+	dinfo = au_di_alloc(sb, AuLsc_DI_CHILD);
+	if (dinfo)
+		dentry->d_fsdata = dinfo;
+	else
+		err = -ENOMEM;
+
+	return err;
+}
+
+void au_di_fin(struct dentry *dentry)
+{
+	struct au_dinfo *dinfo;
+
+	dinfo = au_di(dentry);
+	AuRwDestroy(&dinfo->di_rwsem);
+	au_di_free(dinfo);
+}
+
+/* ---------------------------------------------------------------------- */
+
+static void do_ii_write_lock(struct inode *inode, unsigned int lsc)
+{
+	switch (lsc) {
+	case AuLsc_DI_CHILD:
+		ii_write_lock_child(inode);
+		break;
+	case AuLsc_DI_CHILD2:
+		ii_write_lock_child2(inode);
+		break;
+	case AuLsc_DI_CHILD3:
+		ii_write_lock_child3(inode);
+		break;
+	case AuLsc_DI_PARENT:
+		ii_write_lock_parent(inode);
+		break;
+	case AuLsc_DI_PARENT2:
+		ii_write_lock_parent2(inode);
+		break;
+	case AuLsc_DI_PARENT3:
+		ii_write_lock_parent3(inode);
+		break;
+	default:
+		BUG();
+	}
+}
+
+static void do_ii_read_lock(struct inode *inode, unsigned int lsc)
+{
+	switch (lsc) {
+	case AuLsc_DI_CHILD:
+		ii_read_lock_child(inode);
+		break;
+	case AuLsc_DI_CHILD2:
+		ii_read_lock_child2(inode);
+		break;
+	case AuLsc_DI_CHILD3:
+		ii_read_lock_child3(inode);
+		break;
+	case AuLsc_DI_PARENT:
+		ii_read_lock_parent(inode);
+		break;
+	case AuLsc_DI_PARENT2:
+		ii_read_lock_parent2(inode);
+		break;
+	case AuLsc_DI_PARENT3:
+		ii_read_lock_parent3(inode);
+		break;
+	default:
+		BUG();
+	}
+}
+
+void di_read_lock(struct dentry *d, int flags, unsigned int lsc)
+{
+	struct inode *inode;
+
+	au_rw_read_lock_nested(&au_di(d)->di_rwsem, lsc);
+	if (d_really_is_positive(d)) {
+		inode = d_inode(d);
+		if (au_ftest_lock(flags, IW))
+			do_ii_write_lock(inode, lsc);
+		else if (au_ftest_lock(flags, IR))
+			do_ii_read_lock(inode, lsc);
+	}
+}
+
+void di_read_unlock(struct dentry *d, int flags)
+{
+	struct inode *inode;
+
+	if (d_really_is_positive(d)) {
+		inode = d_inode(d);
+		if (au_ftest_lock(flags, IW)) {
+			au_dbg_verify_dinode(d);
+			ii_write_unlock(inode);
+		} else if (au_ftest_lock(flags, IR)) {
+			au_dbg_verify_dinode(d);
+			ii_read_unlock(inode);
+		}
+	}
+	au_rw_read_unlock(&au_di(d)->di_rwsem);
+}
+
+void di_downgrade_lock(struct dentry *d, int flags)
+{
+	if (d_really_is_positive(d) && au_ftest_lock(flags, IR))
+		ii_downgrade_lock(d_inode(d));
+	au_rw_dgrade_lock(&au_di(d)->di_rwsem);
+}
+
+void di_write_lock(struct dentry *d, unsigned int lsc)
+{
+	au_rw_write_lock_nested(&au_di(d)->di_rwsem, lsc);
+	if (d_really_is_positive(d))
+		do_ii_write_lock(d_inode(d), lsc);
+}
+
+void di_write_unlock(struct dentry *d)
+{
+	au_dbg_verify_dinode(d);
+	if (d_really_is_positive(d))
+		ii_write_unlock(d_inode(d));
+	au_rw_write_unlock(&au_di(d)->di_rwsem);
+}
+
+/* ---------------------------------------------------------------------- */
+
+struct dentry *au_h_dptr(struct dentry *dentry, aufs_bindex_t bindex)
+{
+	struct dentry *d;
+
+	DiMustAnyLock(dentry);
+
+	if (au_dbtop(dentry) < 0 || bindex < au_dbtop(dentry))
+		return NULL;
+	AuDebugOn(bindex < 0);
+	d = au_hdentry(au_di(dentry), bindex)->hd_dentry;
+	AuDebugOn(d && au_dcount(d) <= 0);
+	return d;
+}
+
+/* ---------------------------------------------------------------------- */
+
+void au_set_h_dptr(struct dentry *dentry, aufs_bindex_t bindex,
+		   struct dentry *h_dentry)
+{
+	struct au_dinfo *dinfo;
+	struct au_hdentry *hd;
+
+	DiMustWriteLock(dentry);
+
+	dinfo = au_di(dentry);
+	hd = au_hdentry(dinfo, bindex);
+	au_hdput(hd);
+	hd->hd_dentry = h_dentry;
+}
+
+void au_update_dbtop(struct dentry *dentry)
+{
+	aufs_bindex_t bindex, bbot;
+	struct dentry *h_dentry;
+
+	bbot = au_dbbot(dentry);
+	for (bindex = au_dbtop(dentry); bindex <= bbot; bindex++) {
+		h_dentry = au_h_dptr(dentry, bindex);
+		if (!h_dentry)
+			continue;
+		if (d_is_positive(h_dentry)) {
+			au_set_dbtop(dentry, bindex);
+			return;
+		}
+		au_set_h_dptr(dentry, bindex, NULL);
+	}
+}
+
+void au_update_dbbot(struct dentry *dentry)
+{
+	aufs_bindex_t bindex, btop;
+	struct dentry *h_dentry;
+
+	btop = au_dbtop(dentry);
+	for (bindex = au_dbbot(dentry); bindex >= btop; bindex--) {
+		h_dentry = au_h_dptr(dentry, bindex);
+		if (!h_dentry)
+			continue;
+		if (d_is_positive(h_dentry)) {
+			au_set_dbbot(dentry, bindex);
+			return;
+		}
+		au_set_h_dptr(dentry, bindex, NULL);
+	}
+}

fs/aufs/module.c

@@ -88,6 +88,8 @@ static void au_cache_fin(void)
 
 static int __init au_cache_init(void)
 {
+	au_cache[AuCache_DINFO] = AuCacheCtor(au_dinfo, au_di_init_once);
+	if (au_cache[AuCache_DINFO])
 		/* SLAB_DESTROY_BY_RCU */
 		au_cache[AuCache_ICNTNR] = AuCacheCtor(au_icntnr,
 						       au_icntnr_init_once);

fs/aufs/module.h

@@ -14,6 +14,7 @@
 
 #include <linux/slab.h>
 #include "debug.h"
+#include "dentry.h"
 #include "inode.h"
 
 void *au_krealloc(void *p, unsigned int new_sz, gfp_t gfp, int may_shrink);
@@ -80,6 +81,7 @@ static inline int au_kmidx_sub(size_t sz, size_t new_sz)
 
 /* kmem cache */
 enum {
+	AuCache_DINFO,
 	AuCache_ICNTNR,
 	AuCache_Last
 };
@@ -110,6 +112,7 @@ extern struct kmem_cache *au_cache[AuCache_Last];
 	{ /* au_cache_free_##name##_norcu(p); */			\
 		au_cache_free_##name##_rcu(p); }
 
+AuCacheFuncs(dinfo, DINFO);
 AuCacheFuncs(icntnr, ICNTNR);
 
 #endif /* __KERNEL__ */

fs/aufs/super.h

@@ -12,6 +12,18 @@
 
 #ifdef __KERNEL__
 
+/* flags for __si_read_lock()/aufs_read_lock()/di_read_lock() */
+#define AuLock_DW		1		/* write-lock dentry */
+#define AuLock_IR		(1 << 1)	/* read-lock inode */
+#define AuLock_IW		(1 << 2)	/* write-lock inode */
+#define au_ftest_lock(flags, name)	((flags) & AuLock_##name)
+#define au_fset_lock(flags, name) \
+	do { (flags) |= AuLock_##name; } while (0)
+#define au_fclr_lock(flags, name) \
+	do { (flags) &= ~AuLock_##name; } while (0)
+
+/* ---------------------------------------------------------------------- */
+
 /* super.c */
 struct super_block;
 struct inode *au_iget_locked(struct super_block *sb, ino_t ino);