aufs: copy-up 1/7, attributes

Copy the inode attributes between branches.
See also the document in this commit.
Signed-off-by: J. R. Okajima <hooanon05g@gmail.com>

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

@@ -178,3 +178,12 @@ simply renames the whiteouted name back and returns an error. If all are
 succeeded, aufs registers a function to remove the whiteouted unique
 temporary name completely and asynchronously to the system global
 workqueue.
+
+
+Copy-up
+----------------------------------------------------------------------
+It is a well-known feature or concept.
+When user modifies a file on a readonly branch, aufs operate "copy-up"
+internally and makes change to the new file on the upper writable branch.
+When the trigger systemcall does not update the timestamps of the parent
+dir, aufs reverts it after copy-up.

fs/aufs/Makefile

@@ -14,6 +14,7 @@ aufs-y := module.o sbinfo.o super.o branch.o xino.o sysaufs.o opts.o \
 	wkq.o vfsub.o dcsub.o \
 	cpup.o whout.o \
 	dinfo.o dentry.o \
+	dir.o \
 	iinfo.o inode.o i_op.o
 
 # all are boolean

fs/aufs/aufs.h

@@ -26,6 +26,7 @@
 #include "cpup.h"
 #include "dcsub.h"
 #include "dentry.h"
+#include "dir.h"
 #include "fstype.h"
 #include "hbl.h"
 #include "inode.h"

fs/aufs/branch.c

@@ -470,8 +470,11 @@ int au_br_add(struct super_block *sb, struct au_opt_add *add)
 	au_br_do_add(sb, add_branch, add_bindex);
 
 	h_dentry = add->path.dentry;
-	if (!add_bindex)
+	if (!add_bindex) {
+		au_cpup_attr_all(root_inode, /*force*/1);
 		sb->s_maxbytes = h_dentry->d_sb->s_maxbytes;
+	} else
+		au_add_nlink(root_inode, d_inode(h_dentry));
 
 out:
 	return err;

fs/aufs/cpup.c

@@ -7,8 +7,81 @@
  * copy-up functions, see wbr_policy.c for copy-down
  */
 
+#include <linux/fs_stack.h>
 #include "aufs.h"
 
+void au_cpup_attr_flags(struct inode *dst, unsigned int iflags)
+{
+	const unsigned int mask = S_DEAD | S_SWAPFILE | S_PRIVATE
+		| S_NOATIME | S_NOCMTIME | S_AUTOMOUNT;
+
+	BUILD_BUG_ON(sizeof(iflags) != sizeof(dst->i_flags));
+
+	dst->i_flags |= iflags & ~mask;
+	if (au_test_fs_notime(dst->i_sb))
+		dst->i_flags |= S_NOATIME | S_NOCMTIME;
+}
+
+void au_cpup_attr_timesizes(struct inode *inode)
+{
+	struct inode *h_inode;
+
+	h_inode = au_h_iptr(inode, au_ibtop(inode));
+	fsstack_copy_attr_times(inode, h_inode);
+	fsstack_copy_inode_size(inode, h_inode);
+}
+
+void au_cpup_attr_nlink(struct inode *inode, int force)
+{
+	struct inode *h_inode;
+	struct super_block *sb;
+	aufs_bindex_t bindex, bbot;
+
+	sb = inode->i_sb;
+	bindex = au_ibtop(inode);
+	h_inode = au_h_iptr(inode, bindex);
+	if (!force
+	    && !S_ISDIR(h_inode->i_mode)
+	    && au_opt_test(au_mntflags(sb), PLINK)
+	    && au_plink_test(inode))
+		return;
+
+	/*
+	 * 0 can happen in revalidating.
+	 * h_inode->i_mutex may not be held here, but it is harmless since once
+	 * i_nlink reaches 0, it will never become positive except O_TMPFILE
+	 * case.
+	 * todo: O_TMPFILE+linkat(AT_SYMLINK_FOLLOW) bypassing aufs may cause
+	 *	 the incorrect link count.
+	 */
+	set_nlink(inode, h_inode->i_nlink);
+
+	/*
+	 * fewer nlink makes find(1) noisy, but larger nlink doesn't.
+	 * it may includes whplink directory.
+	 */
+	if (S_ISDIR(h_inode->i_mode)) {
+		bbot = au_ibbot(inode);
+		for (bindex++; bindex <= bbot; bindex++) {
+			h_inode = au_h_iptr(inode, bindex);
+			if (h_inode)
+				au_add_nlink(inode, h_inode);
+		}
+	}
+}
+
+void au_cpup_attr_changeable(struct inode *inode)
+{
+	struct inode *h_inode;
+
+	h_inode = au_h_iptr(inode, au_ibtop(inode));
+	inode->i_mode = h_inode->i_mode;
+	inode->i_uid = h_inode->i_uid;
+	inode->i_gid = h_inode->i_gid;
+	au_cpup_attr_timesizes(inode);
+	au_cpup_attr_flags(inode, h_inode->i_flags);
+}
+
 void au_cpup_igen(struct inode *inode, struct inode *h_inode)
 {
 	struct au_iinfo *iinfo = au_ii(inode);
@@ -18,3 +91,50 @@ void au_cpup_igen(struct inode *inode, struct inode *h_inode)
 	iinfo->ii_higen = h_inode->i_generation;
 	iinfo->ii_hsb1 = h_inode->i_sb;
 }
+
+void au_cpup_attr_all(struct inode *inode, int force)
+{
+	struct inode *h_inode;
+
+	h_inode = au_h_iptr(inode, au_ibtop(inode));
+	au_cpup_attr_changeable(inode);
+	if (inode->i_nlink > 0)
+		au_cpup_attr_nlink(inode, force);
+	inode->i_rdev = h_inode->i_rdev;
+	inode->i_blkbits = h_inode->i_blkbits;
+	au_cpup_igen(inode, h_inode);
+}
+
+/* ---------------------------------------------------------------------- */
+
+/* Note: dt_dentry and dt_h_dentry are not dget/dput-ed */
+
+/* keep the timestamps of the parent dir when cpup */
+void au_dtime_store(struct au_dtime *dt, struct dentry *dentry,
+		    struct path *h_path)
+{
+	struct inode *h_inode;
+
+	dt->dt_dentry = dentry;
+	dt->dt_h_path = *h_path;
+	h_inode = d_inode(h_path->dentry);
+	dt->dt_atime = h_inode->i_atime;
+	dt->dt_mtime = h_inode->i_mtime;
+	/* smp_mb(); */
+}
+
+void au_dtime_revert(struct au_dtime *dt)
+{
+	struct iattr attr;
+	int err;
+
+	attr.ia_atime = dt->dt_atime;
+	attr.ia_mtime = dt->dt_mtime;
+	attr.ia_valid = ATTR_FORCE | ATTR_MTIME | ATTR_MTIME_SET
+		| ATTR_ATIME | ATTR_ATIME_SET;
+
+	/* no delegation since this is a directory */
+	err = vfsub_notify_change(&dt->dt_h_path, &attr, /*delegated*/NULL);
+	if (unlikely(err))
+		pr_warn("restoring timestamps failed(%d). ignored\n", err);
+}

fs/aufs/cpup.h

@@ -12,9 +12,28 @@
 
 #ifdef __KERNEL__
 
+#include <linux/path.h>
+
 struct inode;
 
+void au_cpup_attr_flags(struct inode *dst, unsigned int iflags);
+void au_cpup_attr_timesizes(struct inode *inode);
+void au_cpup_attr_nlink(struct inode *inode, int force);
+void au_cpup_attr_changeable(struct inode *inode);
 void au_cpup_igen(struct inode *inode, struct inode *h_inode);
+void au_cpup_attr_all(struct inode *inode, int force);
+
+/* ---------------------------------------------------------------------- */
+
+/* keep timestamps when copyup */
+struct au_dtime {
+	struct dentry *dt_dentry;
+	struct path dt_h_path;
+	struct timespec64 dt_atime, dt_mtime;
+};
+void au_dtime_store(struct au_dtime *dt, struct dentry *dentry,
+		    struct path *h_path);
+void au_dtime_revert(struct au_dtime *dt);
 
 #endif /* __KERNEL__ */
 #endif /* __AUFS_CPUP_H__ */

fs/aufs/dir.c

+// SPDX-License-Identifier: GPL-2.0
+/*
+ * Copyright (C) 2005-2019 Junjiro R. Okajima
+ */
+
+/*
+ * directory operations
+ */
+
+#include "aufs.h"
+
+void au_add_nlink(struct inode *dir, struct inode *h_dir)
+{
+	unsigned int nlink;
+
+	AuDebugOn(!S_ISDIR(dir->i_mode) || !S_ISDIR(h_dir->i_mode));
+
+	nlink = dir->i_nlink;
+	nlink += h_dir->i_nlink - 2;
+	if (h_dir->i_nlink < 2)
+		nlink += 2;
+	smp_mb(); /* for i_nlink */
+	/* 0 can happen in revaliding */
+	set_nlink(dir, nlink);
+}
+
+void au_sub_nlink(struct inode *dir, struct inode *h_dir)
+{
+	unsigned int nlink;
+
+	AuDebugOn(!S_ISDIR(dir->i_mode) || !S_ISDIR(h_dir->i_mode));
+
+	nlink = dir->i_nlink;
+	nlink -= h_dir->i_nlink - 2;
+	if (h_dir->i_nlink < 2)
+		nlink -= 2;
+	smp_mb(); /* for i_nlink */
+	/* nlink == 0 means the branch-fs is broken */
+	set_nlink(dir, nlink);
+}
+
+struct au_dir_ts_arg {
+	struct dentry *dentry;
+	aufs_bindex_t brid;
+};
+
+static void au_do_dir_ts(void *arg)
+{
+	struct au_dir_ts_arg *a = arg;
+	struct au_dtime dt;
+	struct path h_path;
+	struct inode *dir, *h_dir;
+	struct super_block *sb;
+	struct au_branch *br;
+	struct au_hinode *hdir;
+	int err;
+	aufs_bindex_t btop, bindex;
+
+	sb = a->dentry->d_sb;
+	if (d_really_is_negative(a->dentry))
+		goto out;
+	/* no dir->i_mutex lock */
+	si_read_lock(sb, /*flags*/0); /* noflush */
+	di_write_lock(a->dentry, AuLsc_DI_CHILD);
+
+	dir = d_inode(a->dentry);
+	btop = au_ibtop(dir);
+	bindex = au_br_index(sb, a->brid);
+	if (bindex < btop)
+		goto out_unlock;
+
+	br = au_sbr(sb, bindex);
+	h_path.dentry = au_h_dptr(a->dentry, bindex);
+	if (!h_path.dentry)
+		goto out_unlock;
+	h_path.mnt = au_br_mnt(br);
+	au_dtime_store(&dt, a->dentry, &h_path);
+
+	br = au_sbr(sb, btop);
+	if (!au_br_writable(br->br_perm))
+		goto out_unlock;
+	h_path.dentry = au_h_dptr(a->dentry, btop);
+	h_path.mnt = au_br_mnt(br);
+	err = vfsub_mnt_want_write(h_path.mnt);
+	if (err)
+		goto out_unlock;
+	hdir = au_hi(dir, btop);
+	inode_lock_nested(hdir->hi_inode, AuLsc_I_PARENT);
+	h_dir = au_h_iptr(dir, btop);
+	if (h_dir->i_nlink
+	    && timespec64_compare(&h_dir->i_mtime, &dt.dt_mtime) < 0) {
+		dt.dt_h_path = h_path;
+		au_dtime_revert(&dt);
+	}
+	inode_unlock(hdir->hi_inode);
+	vfsub_mnt_drop_write(h_path.mnt);
+	au_cpup_attr_timesizes(dir);
+
+out_unlock:
+	di_write_unlock(a->dentry);
+	si_read_unlock(sb);
+out:
+	dput(a->dentry);
+	au_nwt_done(&au_sbi(sb)->si_nowait);
+	au_kfree_try_rcu(arg);
+}
+
+void au_dir_ts(struct inode *dir, aufs_bindex_t bindex)
+{
+	int perm, wkq_err;
+	aufs_bindex_t btop;
+	struct au_dir_ts_arg *arg;
+	struct dentry *dentry;
+	struct super_block *sb;
+
+	IMustLock(dir);
+
+	dentry = d_find_any_alias(dir);
+	AuDebugOn(!dentry);
+	sb = dentry->d_sb;
+	btop = au_ibtop(dir);
+	if (btop == bindex) {
+		au_cpup_attr_timesizes(dir);
+		goto out;
+	}
+
+	perm = au_sbr_perm(sb, btop);
+	if (!au_br_writable(perm))
+		goto out;
+
+	arg = kmalloc(sizeof(*arg), GFP_NOFS);
+	if (!arg)
+		goto out;
+
+	arg->dentry = dget(dentry); /* will be dput-ted by au_do_dir_ts() */
+	arg->brid = au_sbr_id(sb, bindex);
+	wkq_err = au_wkq_nowait(au_do_dir_ts, arg, sb, /*flags*/0);
+	if (unlikely(wkq_err)) {
+		pr_err("wkq %d\n", wkq_err);
+		dput(dentry);
+		au_kfree_try_rcu(arg);
+	}
+
+out:
+	dput(dentry);
+}

fs/aufs/dir.h

+/* SPDX-License-Identifier: GPL-2.0 */
+/*
+ * Copyright (C) 2005-2019 Junjiro R. Okajima
+ */
+
+/*
+ * directory operations
+ */
+
+#ifndef __AUFS_DIR_H__
+#define __AUFS_DIR_H__
+
+#ifdef __KERNEL__
+
+#include <linux/fs.h>
+
+/* ---------------------------------------------------------------------- */
+
+/* dir.c */
+void au_add_nlink(struct inode *dir, struct inode *h_dir);
+void au_sub_nlink(struct inode *dir, struct inode *h_dir);
+void au_dir_ts(struct inode *dir, aufs_bindex_t bsrc);
+
+#endif /* __KERNEL__ */
+#endif /* __AUFS_DIR_H__ */

fs/aufs/fstype.h

@@ -247,6 +247,16 @@ static inline int au_test_fs_no_limit_nlink(struct super_block *sb)
 		|| au_test_ubifs(sb);
 }
 
+/*
+ * filesystems which sets S_NOATIME and S_NOCMTIME.
+ */
+static inline int au_test_fs_notime(struct super_block *sb)
+{
+	return au_test_nfs(sb)
+		|| au_test_ubifs(sb)
+		;
+}
+
 /* ---------------------------------------------------------------------- */
 
 /*

fs/aufs/vfsub.c

@@ -180,6 +180,67 @@ out:
 
 /* ---------------------------------------------------------------------- */
 
+struct notify_change_args {
+	int *errp;
+	struct path *path;
+	struct iattr *ia;
+	struct inode **delegated_inode;
+};
+
+static void call_notify_change(void *args)
+{
+	struct notify_change_args *a = args;
+	struct inode *h_inode;
+
+	h_inode = d_inode(a->path->dentry);
+	IMustLock(h_inode);
+
+	*a->errp = -EPERM;
+	if (!IS_IMMUTABLE(h_inode) && !IS_APPEND(h_inode)) {
+		lockdep_off();
+		*a->errp = notify_change(a->path->dentry, a->ia,
+					 a->delegated_inode);
+		lockdep_on();
+	}
+	AuTraceErr(*a->errp);
+}
+
+int vfsub_notify_change(struct path *path, struct iattr *ia,
+			struct inode **delegated_inode)
+{
+	int err;
+	struct notify_change_args args = {
+		.errp			= &err,
+		.path			= path,
+		.ia			= ia,
+		.delegated_inode	= delegated_inode
+	};
+
+	call_notify_change(&args);
+
+	return err;
+}
+
+int vfsub_sio_notify_change(struct path *path, struct iattr *ia,
+			    struct inode **delegated_inode)
+{
+	int err, wkq_err;
+	struct notify_change_args args = {
+		.errp			= &err,
+		.path			= path,
+		.ia			= ia,
+		.delegated_inode	= delegated_inode
+	};
+
+	wkq_err = au_wkq_wait(call_notify_change, &args);
+	if (unlikely(wkq_err))
+		err = wkq_err;
+
+	return err;
+}
+
+/* ---------------------------------------------------------------------- */
+
 struct unlink_args {
 	int *errp;
 	struct inode *dir;

fs/aufs/vfsub.h

@@ -95,6 +95,10 @@ static inline loff_t vfsub_f_size_read(struct file *file)
 
 /* ---------------------------------------------------------------------- */
 
+int vfsub_sio_notify_change(struct path *path, struct iattr *ia,
+			    struct inode **delegated_inode);
+int vfsub_notify_change(struct path *path, struct iattr *ia,
+			struct inode **delegated_inode);
 int vfsub_unlink(struct inode *dir, struct path *path,
 		 struct inode **delegated_inode, int force);