[PATCH 15/42] Unionfs: dentry revalidation

[Erez Zadok <ezk@cs.sunysb.edu>]

Includes d_release methods and cache-coherency support for dentries.

Signed-off-by: Erez Zadok <ezk@cs.sunysb.edu>
---
 fs/unionfs/dentry.c |  498 +++++++++++++++++++++++++++++++++++++++++++++++++++
 1 files changed, 498 insertions(+), 0 deletions(-)
 create mode 100644 fs/unionfs/dentry.c

diff --git a/fs/unionfs/dentry.c b/fs/unionfs/dentry.c
new file mode 100644
index 0000000..7d27987
--- /dev/null
+++ b/fs/unionfs/dentry.c
@@ -0,0 +1,498 @@
+/*
+ * Copyright (c) 2003-2007 Erez Zadok
+ * Copyright (c) 2003-2006 Charles P. Wright
+ * Copyright (c) 2005-2007 Josef 'Jeff' Sipek
+ * Copyright (c) 2005-2006 Junjiro Okajima
+ * Copyright (c) 2005      Arun M. Krishnakumar
+ * Copyright (c) 2004-2006 David P. Quigley
+ * Copyright (c) 2003-2004 Mohammad Nayyer Zubair
+ * Copyright (c) 2003      Puja Gupta
+ * Copyright (c) 2003      Harikesavan Krishnan
+ * Copyright (c) 2003-2007 Stony Brook University
+ * Copyright (c) 2003-2007 The Research Foundation of SUNY
+ *
+ * This program is free software; you can redistribute it and/or modify
+ * it under the terms of the GNU General Public License version 2 as
+ * published by the Free Software Foundation.
+ */
+
+#include "union.h"
+
+/*
+ * Revalidate a single dentry.
+ * Assume that dentry's info node is locked.
+ * Assume that parent(s) are all valid already, but
+ * the child may not yet be valid.
+ * Returns true if valid, false otherwise.
+ */
+static bool __unionfs_d_revalidate_one(struct dentry *dentry,
+				       struct nameidata *nd)
+{
+	bool valid = true;	/* default is valid */
+	struct dentry *lower_dentry;
+	int bindex, bstart, bend;
+	int sbgen, dgen;
+	int positive = 0;
+	int locked = 0;
+	int interpose_flag;
+	struct nameidata lowernd; /* TODO: be gentler to the stack */
+
+	if (nd)
+		memcpy(&lowernd, nd, sizeof(struct nameidata));
+	else
+		memset(&lowernd, 0, sizeof(struct nameidata));
+
+	verify_locked(dentry);
+
+	/* if the dentry is unhashed, do NOT revalidate */
+	if (d_deleted(dentry))
+		goto out;
+
+	BUG_ON(dbstart(dentry) == -1);
+	if (dentry->d_inode)
+		positive = 1;
+	dgen = atomic_read(&UNIONFS_D(dentry)->generation);
+	sbgen = atomic_read(&UNIONFS_SB(dentry->d_sb)->generation);
+	/*
+	 * If we are working on an unconnected dentry, then there is no
+	 * revalidation to be done, because this file does not exist within
+	 * the namespace, and Unionfs operates on the namespace, not data.
+	 */
+	if (unlikely(sbgen != dgen)) {
+		struct dentry *result;
+		int pdgen;
+
+		/* The root entry should always be valid */
+		BUG_ON(IS_ROOT(dentry));
+
+		/* We can't work correctly if our parent isn't valid. */
+		pdgen = atomic_read(&UNIONFS_D(dentry->d_parent)->generation);
+		BUG_ON(pdgen != sbgen);	/* should never happen here */
+
+		/* Free the pointers for our inodes and this dentry. */
+		bstart = dbstart(dentry);
+		bend = dbend(dentry);
+		if (bstart >= 0) {
+			struct dentry *lower_dentry;
+			for (bindex = bstart; bindex <= bend; bindex++) {
+				lower_dentry =
+					unionfs_lower_dentry_idx(dentry,
+								 bindex);
+				dput(lower_dentry);
+			}
+		}
+		set_dbstart(dentry, -1);
+		set_dbend(dentry, -1);
+
+		interpose_flag = INTERPOSE_REVAL_NEG;
+		if (positive) {
+			interpose_flag = INTERPOSE_REVAL;
+			/*
+			 * During BRM, the VFS could already hold a lock on
+			 * a file being read, so don't lock it again
+			 * (deadlock), but if you lock it in this function,
+			 * then release it here too.
+			 */
+			if (!mutex_is_locked(&dentry->d_inode->i_mutex)) {
+				mutex_lock(&dentry->d_inode->i_mutex);
+				locked = 1;
+			}
+
+			bstart = ibstart(dentry->d_inode);
+			bend = ibend(dentry->d_inode);
+			if (bstart >= 0) {
+				struct inode *lower_inode;
+				for (bindex = bstart; bindex <= bend;
+				     bindex++) {
+					lower_inode =
+						unionfs_lower_inode_idx(
+							dentry->d_inode,
+							bindex);
+					iput(lower_inode);
+				}
+			}
+			kfree(UNIONFS_I(dentry->d_inode)->lower_inodes);
+			UNIONFS_I(dentry->d_inode)->lower_inodes = NULL;
+			ibstart(dentry->d_inode) = -1;
+			ibend(dentry->d_inode) = -1;
+			if (locked)
+				mutex_unlock(&dentry->d_inode->i_mutex);
+		}
+
+		result = unionfs_lookup_backend(dentry, &lowernd,
+						interpose_flag);
+		if (result) {
+			if (IS_ERR(result)) {
+				valid = false;
+				goto out;
+			}
+			/*
+			 * current unionfs_lookup_backend() doesn't return
+			 * a valid dentry
+			 */
+			dput(dentry);
+			dentry = result;
+		}
+
+		if (unlikely(positive && UNIONFS_I(dentry->d_inode)->stale)) {
+			make_bad_inode(dentry->d_inode);
+			d_drop(dentry);
+			valid = false;
+			goto out;
+		}
+		goto out;
+	}
+
+	/* The revalidation must occur across all branches */
+	bstart = dbstart(dentry);
+	bend = dbend(dentry);
+	BUG_ON(bstart == -1);
+	for (bindex = bstart; bindex <= bend; bindex++) {
+		lower_dentry = unionfs_lower_dentry_idx(dentry, bindex);
+		if (!lower_dentry || !lower_dentry->d_op
+		    || !lower_dentry->d_op->d_revalidate)
+			continue;
+		/*
+		 * Don't pass nameidata to lower file system, because we
+		 * don't want an arbitrary lower file being opened or
+		 * returned to us: it may be useless to us because of the
+		 * fanout nature of unionfs (cf. file/directory open-file
+		 * invariants).  We will open lower files as and when needed
+		 * later on.
+		 */
+		if (!lower_dentry->d_op->d_revalidate(lower_dentry, NULL))
+			valid = false;
+	}
+
+	if (!dentry->d_inode)
+		valid = false;
+
+	if (valid) {
+		/*
+		 * If we get here, and we copy the meta-data from the lower
+		 * inode to our inode, then it is vital that we have already
+		 * purged all unionfs-level file data.  We do that in the
+		 * caller (__unionfs_d_revalidate_chain) by calling
+		 * purge_inode_data.
+		 */
+		unionfs_copy_attr_all(dentry->d_inode,
+				      unionfs_lower_inode(dentry->d_inode));
+		fsstack_copy_inode_size(dentry->d_inode,
+					unionfs_lower_inode(dentry->d_inode));
+	}
+
+out:
+	return valid;
+}
+
+/*
+ * Determine if the lower inode objects have changed from below the unionfs
+ * inode.  Return true if changed, false otherwise.
+ *
+ * We check if the mtime or ctime have changed.  However, the inode times
+ * can be changed by anyone without much protection, including
+ * asynchronously.  This can sometimes cause unionfs to find that the lower
+ * file system doesn't change its inode times quick enough, resulting in a
+ * false positive indication (which is harmless, it just makes unionfs do
+ * extra work in re-validating the objects).  To minimize the chances of
+ * these situations, we still consider such small time changes valid, but we
+ * don't print debugging messages unless the time changes are greater than
+ * UNIONFS_MIN_CC_TIME (which defaults to 3 seconds, as with NFS's acregmin)
+ * because significant changes are more likely due to users manually
+ * touching lower files.
+ */
+bool is_newer_lower(const struct dentry *dentry)
+{
+	int bindex;
+	struct inode *inode;
+	struct inode *lower_inode;
+
+	/* ignore if we're called on semi-initialized dentries/inodes */
+	if (!dentry || !UNIONFS_D(dentry))
+		return false;
+	inode = dentry->d_inode;
+	if (!inode || !UNIONFS_I(inode) ||
+	    ibstart(inode) < 0 || ibend(inode) < 0)
+		return false;
+
+	for (bindex = ibstart(inode); bindex <= ibend(inode); bindex++) {
+		lower_inode = unionfs_lower_inode_idx(inode, bindex);
+		if (!lower_inode)
+			continue;
+
+		/* check if mtime/ctime have changed */
+		if (unlikely(timespec_compare(&inode->i_mtime,
+					      &lower_inode->i_mtime) < 0)) {
+			if ((lower_inode->i_mtime.tv_sec -
+			     inode->i_mtime.tv_sec) > UNIONFS_MIN_CC_TIME) {
+				pr_info("unionfs: new lower inode mtime "
+					"(bindex=%d, name=%s)\n", bindex,
+					dentry->d_name.name);
+				show_dinode_times(dentry);
+			}
+			return true;
+		}
+		if (unlikely(timespec_compare(&inode->i_ctime,
+					      &lower_inode->i_ctime) < 0)) {
+			if ((lower_inode->i_ctime.tv_sec -
+			     inode->i_ctime.tv_sec) > UNIONFS_MIN_CC_TIME) {
+				pr_info("unionfs: new lower inode ctime "
+					"(bindex=%d, name=%s)\n", bindex,
+					dentry->d_name.name);
+				show_dinode_times(dentry);
+			}
+			return true;
+		}
+	}
+	return false;		/* default: lower is not newer */
+}
+
+/*
+ * Purge/remove/unmap all date pages of a unionfs inode.  This is called
+ * when the lower inode has changed, and we have to force processes to get
+ * the new data.
+ *
+ * XXX: Our implementation works in that as long as a user process will have
+ * caused Unionfs to be called, directly or indirectly, even to just do
+ * ->d_revalidate; then we will have purged the current Unionfs data and the
+ * process will see the new data.  For example, a process that continually
+ * re-reads the same file's data will see the NEW data as soon as the lower
+ * file had changed, upon the next read(2) syscall (even if the file is
+ * still open!)  However, this doesn't work when the process re-reads the
+ * open file's data via mmap(2) (unless the user unmaps/closes the file and
+ * remaps/reopens it).  Once we respond to ->readpage(s), then the kernel
+ * maps the page into the process's address space and there doesn't appear
+ * to be a way to force the kernel to invalidate those pages/mappings, and
+ * force the process to re-issue ->readpage.  If there's a way to invalidate
+ * active mappings and force a ->readpage, let us know please
+ * (invalidate_inode_pages2 doesn't do the trick).
+ */
+static inline void purge_inode_data(struct inode *inode)
+{
+	/* remove all non-private mappings */
+	unmap_mapping_range(inode->i_mapping, 0, 0, 0);
+
+	if (inode->i_data.nrpages)
+		truncate_inode_pages(&inode->i_data, 0);
+}
+
+/*
+ * Revalidate a parent chain of dentries, then the actual node.
+ * Assumes that dentry is locked, but will lock all parents if/when needed.
+ *
+ * If 'willwrite' is true, and the lower inode times are not in sync, then
+ * *don't* purge_inode_data, as it could deadlock if ->write calls us and we
+ * try to truncate a locked page.  Besides, if unionfs is about to write
+ * data to a file, then there's the data unionfs is about to write is more
+ * authoritative than what's below, therefore we can safely overwrite the
+ * lower inode times and data.
+ */
+bool __unionfs_d_revalidate_chain(struct dentry *dentry, struct nameidata *nd,
+				  bool willwrite)
+{
+	bool valid = false;	/* default is invalid */
+	struct dentry **chain = NULL; /* chain of dentries to reval */
+	int chain_len = 0;
+	struct dentry *dtmp;
+	int sbgen, dgen, i;
+	int saved_bstart, saved_bend, bindex;
+
+	/* find length of chain needed to revalidate */
+	/* XXX: should I grab some global (dcache?) lock? */
+	chain_len = 0;
+	sbgen = atomic_read(&UNIONFS_SB(dentry->d_sb)->generation);
+	dtmp = dentry->d_parent;
+	dgen = atomic_read(&UNIONFS_D(dtmp)->generation);
+	/* XXX: should we check if is_newer_lower all the way up? */
+	if (unlikely(is_newer_lower(dtmp))) {
+		/*
+		 * Special case: the root dentry's generation number must
+		 * always be valid, but its lower inode times don't have to
+		 * be, so sync up the times only.
+		 */
+		if (IS_ROOT(dtmp)) {
+			unionfs_copy_attr_times(dtmp->d_inode);
+		} else {
+			/*
+			 * reset generation number to zero, guaranteed to be
+			 * "old"
+			 */
+			dgen = 0;
+			atomic_set(&UNIONFS_D(dtmp)->generation, dgen);
+		}
+		purge_inode_data(dtmp->d_inode);
+	}
+	while (sbgen != dgen) {
+		/* The root entry should always be valid */
+		BUG_ON(IS_ROOT(dtmp));
+		chain_len++;
+		dtmp = dtmp->d_parent;
+		dgen = atomic_read(&UNIONFS_D(dtmp)->generation);
+	}
+	if (chain_len == 0)
+		goto out_this;	/* shortcut if parents are OK */
+
+	/*
+	 * Allocate array of dentries to reval.  We could use linked lists,
+	 * but the number of entries we need to alloc here is often small,
+	 * and short lived, so locality will be better.
+	 */
+	chain = kzalloc(chain_len * sizeof(struct dentry *), GFP_KERNEL);
+	if (unlikely(!chain)) {
+		printk(KERN_CRIT "unionfs: no more memory in %s\n",
+		       __FUNCTION__);
+		goto out;
+	}
+
+	/*
+	 * lock all dentries in chain, in child to parent order.
+	 * if failed, then sleep for a little, then retry.
+	 */
+	dtmp = dentry->d_parent;
+	for (i = chain_len-1; i >= 0; i--) {
+		chain[i] = dget(dtmp);
+		dtmp = dtmp->d_parent;
+	}
+
+	/*
+	 * call __unionfs_d_revalidate_one() on each dentry, but in parent
+	 * to child order.
+	 */
+	for (i = 0; i < chain_len; i++) {
+		unionfs_lock_dentry(chain[i]);
+		saved_bstart = dbstart(chain[i]);
+		saved_bend = dbend(chain[i]);
+		sbgen = atomic_read(&UNIONFS_SB(dentry->d_sb)->generation);
+		dgen = atomic_read(&UNIONFS_D(chain[i])->generation);
+
+		valid = __unionfs_d_revalidate_one(chain[i], nd);
+		/* XXX: is this the correct mntput condition?! */
+		if (valid && chain_len > 0 &&
+		    sbgen != dgen && chain[i]->d_inode &&
+		    S_ISDIR(chain[i]->d_inode->i_mode)) {
+			for (bindex = saved_bstart; bindex <= saved_bend;
+			     bindex++)
+				unionfs_mntput(chain[i], bindex);
+		}
+		unionfs_unlock_dentry(chain[i]);
+
+		if (unlikely(!valid))
+			goto out_free;
+	}
+
+
+out_this:
+	/* finally, lock this dentry and revalidate it */
+	verify_locked(dentry);
+	dgen = atomic_read(&UNIONFS_D(dentry)->generation);
+
+	if (unlikely(is_newer_lower(dentry))) {
+		/* root dentry special case as aforementioned */
+		if (IS_ROOT(dentry)) {
+			unionfs_copy_attr_times(dentry->d_inode);
+		} else {
+			/*
+			 * reset generation number to zero, guaranteed to be
+			 * "old"
+			 */
+			dgen = 0;
+			atomic_set(&UNIONFS_D(dentry)->generation, dgen);
+		}
+		if (!willwrite)
+			purge_inode_data(dentry->d_inode);
+	}
+	valid = __unionfs_d_revalidate_one(dentry, nd);
+
+	/*
+	 * If __unionfs_d_revalidate_one() succeeded above, then it will
+	 * have incremented the refcnt of the mnt's, but also the branch
+	 * indices of the dentry will have been updated (to take into
+	 * account any branch insertions/deletion.  So the current
+	 * dbstart/dbend match the current, and new, indices of the mnts
+	 * which __unionfs_d_revalidate_one has incremented.  Note: the "if"
+	 * test below does not depend on whether chain_len was 0 or greater.
+	 */
+	if (valid && sbgen != dgen)
+		for (bindex = dbstart(dentry);
+		     bindex <= dbend(dentry);
+		     bindex++)
+			unionfs_mntput(dentry, bindex);
+
+out_free:
+	/* unlock/dput all dentries in chain and return status */
+	if (chain_len > 0) {
+		for (i = 0; i < chain_len; i++)
+			dput(chain[i]);
+		kfree(chain);
+	}
+out:
+	return valid;
+}
+
+static int unionfs_d_revalidate(struct dentry *dentry, struct nameidata *nd)
+{
+	int err;
+
+	unionfs_read_lock(dentry->d_sb);
+
+	unionfs_lock_dentry(dentry);
+	err = __unionfs_d_revalidate_chain(dentry, nd, false);
+	if (likely(err > 0)) { /* true==1: dentry is valid */
+		unionfs_check_dentry(dentry);
+		unionfs_check_nd(nd);
+	}
+	unionfs_unlock_dentry(dentry);
+
+	unionfs_read_unlock(dentry->d_sb);
+
+	return err;
+}
+
+/*
+ * At this point no one can reference this dentry, so we don't have to be
+ * careful about concurrent access.
+ */
+static void unionfs_d_release(struct dentry *dentry)
+{
+	int bindex, bstart, bend;
+
+	unionfs_read_lock(dentry->d_sb);
+
+	unionfs_check_dentry(dentry);
+	/* this could be a negative dentry, so check first */
+	if (unlikely(!UNIONFS_D(dentry))) {
+		printk(KERN_ERR "unionfs: dentry without private data: %.*s\n",
+		       dentry->d_name.len, dentry->d_name.name);
+		goto out;
+	} else if (dbstart(dentry) < 0)
+		goto out_free;  /* due to a (normal) failed lookup */
+
+	/* Release all the lower dentries */
+	bstart = dbstart(dentry);
+	bend = dbend(dentry);
+	for (bindex = bstart; bindex <= bend; bindex++) {
+		dput(unionfs_lower_dentry_idx(dentry, bindex));
+		unionfs_set_lower_dentry_idx(dentry, bindex, NULL);
+		/* NULL lower mnt is ok if this is a negative dentry */
+		if (!dentry->d_inode && !unionfs_lower_mnt_idx(dentry, bindex))
+			continue;
+		unionfs_mntput(dentry, bindex);
+		unionfs_set_lower_mnt_idx(dentry, bindex, NULL);
+	}
+	/* free private data (unionfs_dentry_info) here */
+	kfree(UNIONFS_D(dentry)->lower_paths);
+	UNIONFS_D(dentry)->lower_paths = NULL;
+
+out_free:
+	/* No need to unlock it, because it is disappeared. */
+	free_dentry_private_data(dentry);
+
+out:
+	unionfs_read_unlock(dentry->d_sb);
+	return;
+}
+
+struct dentry_operations unionfs_dops = {
+	.d_revalidate	= unionfs_d_revalidate,
+	.d_release	= unionfs_d_release,
+};
-- 
1.5.2.2

--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/