public inbox for linux-kernel@vger.kernel.org
 help / color / mirror / Atom feed
From: Theodore Tso <tytso@mit.edu>
To: Linus Torvalds <torvalds@linux-foundation.org>
Cc: Markus Trippelsdorf <markus@trippelsdorf.de>,
	linux-kernel@vger.kernel.org, eugene@ibrix.com,
	msnitzer@ibrix.com, akpm@linux-foundation.org
Subject: [PATCH] Re: ext3: fix ext3_dx_readdir hash collision handling - Regression
Date: Sat, 25 Oct 2008 07:56:31 -0400	[thread overview]
Message-ID: <20081025115631.GA17661@mit.edu> (raw)
In-Reply-To: <alpine.LFD.2.00.0810240853310.3287@nehalem.linux-foundation.org>

I believe I've found the problem.  I'm still not 100% sure why I'm not
seeing this on a 32-bit kernel, so I want to do a bit more testing to
make sure I completely understand what was happening.  However, once I
reproduced it on a a 64-bit kernel, it was pretty clear what the
problem was.  The problem only happens when the directory is changing
out from under us (which would be the case when the directory is
getting deleted via "/bin/rm -r" or "git clean").  When that happens,
the following check causes us to reload the rbtree we use to return
the directory entries in hash sort error:

		/*
		 * Fill the rbtree if we have no more entries,
		 * or the inode has changed since we last read in the
		 * cached entries.
		 */
		if ((!info->curr_node) ||
		    (filp->f_version != inode->i_version)) {
			info->curr_node = NULL;
			free_rb_tree_fname(&info->root);
			filp->f_version = inode->i_version;
			ret = ext3_htree_fill_tree(filp, info->curr_hash,
						   info->curr_minor_hash,
						   &info->next_hash);


Unfortunately, the commit which which fixes the duplicated entries
when there are hash collisions wasn't properly setting info->curr_hash
and info->curr_minor_hash after going to the next node, which results
the first entry being returned twice on the subsequent getdents()
system call.  What I don't understand was why I wasn't seeing this on
a 32-bit kernel (maybe I screwed up my test environment; this problem
should have occurred regardless of the 32-bit or 64-bit environment).

In any case, if folks could test to see whether this patch fixes
things, I'd appreciate it.  (Patch for ext3 and ext4 follows).

							- Ted

diff --git a/fs/ext3/dir.c b/fs/ext3/dir.c
index 4c82531..5853f44 100644
--- a/fs/ext3/dir.c
+++ b/fs/ext3/dir.c
@@ -456,17 +456,8 @@ static int ext3_dx_readdir(struct file * filp,
 	if (info->extra_fname) {
 		if (call_filldir(filp, dirent, filldir, info->extra_fname))
 			goto finished;
-
 		info->extra_fname = NULL;
-		info->curr_node = rb_next(info->curr_node);
-		if (!info->curr_node) {
-			if (info->next_hash == ~0) {
-				filp->f_pos = EXT3_HTREE_EOF;
-				goto finished;
-			}
-			info->curr_hash = info->next_hash;
-			info->curr_minor_hash = 0;
-		}
+		goto next_node;
 	} else if (!info->curr_node)
 		info->curr_node = rb_first(&info->root);
 
@@ -498,9 +489,14 @@ static int ext3_dx_readdir(struct file * filp,
 		info->curr_minor_hash = fname->minor_hash;
 		if (call_filldir(filp, dirent, filldir, fname))
 			break;
-
+	next_node:
 		info->curr_node = rb_next(info->curr_node);
-		if (!info->curr_node) {
+		if (info->curr_node) {
+			fname = rb_entry(info->curr_node, struct fname,
+					 rb_hash);
+			info->curr_hash = fname->hash;
+			info->curr_minor_hash = fname->minor_hash;
+		} else {
 			if (info->next_hash == ~0) {
 				filp->f_pos = EXT3_HTREE_EOF;
 				break;
diff --git a/fs/ext4/dir.c b/fs/ext4/dir.c
index 3ca6a2b..fed5b61 100644
--- a/fs/ext4/dir.c
+++ b/fs/ext4/dir.c
@@ -459,17 +459,8 @@ static int ext4_dx_readdir(struct file *filp,
 	if (info->extra_fname) {
 		if (call_filldir(filp, dirent, filldir, info->extra_fname))
 			goto finished;
-
 		info->extra_fname = NULL;
-		info->curr_node = rb_next(info->curr_node);
-		if (!info->curr_node) {
-			if (info->next_hash == ~0) {
-				filp->f_pos = EXT4_HTREE_EOF;
-				goto finished;
-			}
-			info->curr_hash = info->next_hash;
-			info->curr_minor_hash = 0;
-		}
+		goto next_node;
 	} else if (!info->curr_node)
 		info->curr_node = rb_first(&info->root);
 
@@ -501,9 +492,14 @@ static int ext4_dx_readdir(struct file *filp,
 		info->curr_minor_hash = fname->minor_hash;
 		if (call_filldir(filp, dirent, filldir, fname))
 			break;
-
+	next_node:
 		info->curr_node = rb_next(info->curr_node);
-		if (!info->curr_node) {
+		if (info->curr_node) {
+			fname = rb_entry(info->curr_node, struct fname,
+					 rb_hash);
+			info->curr_hash = fname->hash;
+			info->curr_minor_hash = fname->minor_hash;
+		} else {
 			if (info->next_hash == ~0) {
 				filp->f_pos = EXT4_HTREE_EOF;
 				break;

  parent reply	other threads:[~2008-10-25 11:56 UTC|newest]

Thread overview: 13+ messages / expand[flat|nested]  mbox.gz  Atom feed  top
2008-10-22  9:32 ext3: fix ext3_dx_readdir hash collision handling - Regression Markus Trippelsdorf
2008-10-23  3:28 ` Theodore Tso
2008-10-23  6:37   ` Markus Trippelsdorf
2008-10-24  0:01     ` Theodore Tso
     [not found]       ` <20081024042851.GA2360@gentoox2.trippelsdorf.de>
2008-10-24 10:41         ` Theodore Tso
2008-10-24 16:08         ` Linus Torvalds
2008-10-24 20:44           ` Theodore Tso
2008-10-25 11:56           ` Theodore Tso [this message]
2008-10-25 12:25             ` [PATCH] " Markus Trippelsdorf
2008-10-25 21:15             ` Linus Torvalds
2008-10-26  2:39               ` [GIT PULL] " Theodore Tso
2008-10-26  2:42                 ` [PATCH 1/2] ext3: Fix duplicate entries returned from getdents() system call Theodore Ts'o
2008-10-26  2:42                   ` [PATCH 2/2] ext4: " Theodore Ts'o

Reply instructions:

You may reply publicly to this message via plain-text email
using any one of the following methods:

* Save the following mbox file, import it into your mail client,
  and reply-to-all from there: mbox

  Avoid top-posting and favor interleaved quoting:
  https://en.wikipedia.org/wiki/Posting_style#Interleaved_style

* Reply using the --to, --cc, and --in-reply-to
  switches of git-send-email(1):

  git send-email \
    --in-reply-to=20081025115631.GA17661@mit.edu \
    --to=tytso@mit.edu \
    --cc=akpm@linux-foundation.org \
    --cc=eugene@ibrix.com \
    --cc=linux-kernel@vger.kernel.org \
    --cc=markus@trippelsdorf.de \
    --cc=msnitzer@ibrix.com \
    --cc=torvalds@linux-foundation.org \
    /path/to/YOUR_REPLY

  https://kernel.org/pub/software/scm/git/docs/git-send-email.html

* If your mail client supports setting the In-Reply-To header
  via mailto: links, try the mailto: link
Be sure your reply has a Subject: header at the top and a blank line before the message body.
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox