[PATCH] v9fs: simplify fid mapping

v9fs has been plagued by an over-complicated approach trying to map Linux
dentry semantics to Plan 9 fid semantics. Our previous approach called for
aggressive flushing of the dcache resulting in several problems (including
wierd cwd behavior when running /bin/pwd).

This patch dramatically simplifies our handling of this fid management. Fids
will not be clunked as promptly, but the new approach is more functionally
correct. We now clunk un-open fids only when their dentry ref_count reaches 0
(and d_delete is called).

Another simplification is we no longer seek to match fids to the process-id or
uid of the action initiator. The uid-matching will need to be revisited when
we fix the security model.

Signed-off-by: Eric Van Hensbergen <[e-mail]>
Signed-off-by: Andrew Morton <[e-mail]>
Signed-off-by: Linus Torvalds <[e-mail]>

fs/9p/fid.c

···
 /*
  * V9FS FID Management
  *
- *  Copyright (C) 2005 by Eric Van Hensbergen <[e-mail]>
+ *  Copyright (C) 2005, 2006 by Eric Van Hensbergen <[e-mail]>
  *
  *  This program is free software; you can redistribute it and/or modify
  *  it under the terms of the GNU General Public License as published by
···
 	}
 
 	fid->uid = current->uid;
-	fid->pid = current->pid;
 	list_add(&fid->list, fid_list);
 	return 0;
 }
···
 	new->rdir_fcall = NULL;
 	INIT_LIST_HEAD(&new->list);
 
-		return new;
+	return new;
 }
 
 /**
···
 }
 
 /**
- * v9fs_fid_walk_up - walks from the process current directory
- * 	up to the specified dentry.
- */
-static struct v9fs_fid *v9fs_fid_walk_up(struct dentry *dentry)
-{
-	int fidnum, cfidnum, err;
-	struct v9fs_fid *cfid, *fid;
-	struct dentry *cde;
-	struct v9fs_session_info *v9ses;
-
-	v9ses = v9fs_inode2v9ses(current->fs->pwd->d_inode);
-	cfid = v9fs_fid_lookup(current->fs->pwd);
-	if (cfid == NULL) {
-		dprintk(DEBUG_ERROR, "process cwd doesn't have a fid\n");
-		return ERR_PTR(-ENOENT);
-	}
-
-	cfidnum = cfid->fid;
-	cde = current->fs->pwd;
-	/* TODO: take advantage of multiwalk */
-
-	fidnum = v9fs_get_idpool(&v9ses->fidpool);
-	if (fidnum < 0) {
-		dprintk(DEBUG_ERROR, "could not get a new fid num\n");
-		err = -ENOENT;
-		goto clunk_fid;
-	}
-
-	while (cde != dentry) {
-		if (cde == cde->d_parent) {
-			dprintk(DEBUG_ERROR, "can't find dentry\n");
-			err = -ENOENT;
-			goto clunk_fid;
-		}
-
-		err = v9fs_t_walk(v9ses, cfidnum, fidnum, "..", NULL);
-		if (err < 0) {
-			dprintk(DEBUG_ERROR, "problem walking to parent\n");
-			goto clunk_fid;
-		}
-
-		cfidnum = fidnum;
-		cde = cde->d_parent;
-	}
-
-	fid = v9fs_fid_create(v9ses, fidnum);
-	if (fid) {
-		err = v9fs_fid_insert(fid, dentry);
-		if (err < 0) {
-			kfree(fid);
-			goto clunk_fid;
-		}
-	}
-
-	return fid;
-
-clunk_fid:
-	v9fs_t_clunk(v9ses, fidnum);
-	return ERR_PTR(err);
-}
-
-/**
  * v9fs_fid_lookup - retrieve the right fid from a  particular dentry
  * @dentry: dentry to look for fid in
  * @type: intent of lookup (operation or traversal)
  *
- * search list of fids associated with a dentry for a fid with a matching
- * thread id or uid.  If that fails, look up the dentry's parents to see if you
- * can find a matching fid.
+ * find a fid in the dentry
+ *
+ * TODO: only match fids that have the same uid as current user
  *
  */
 
···
 		return_fid = list_entry(fid_list->next, struct v9fs_fid, list);
 
 	if (!return_fid) {
-		struct dentry *par = current->fs->pwd->d_parent;
-		int count = 1;
-		while (par != NULL) {
-			if (par == dentry)
-				break;
-			count++;
-			if (par == par->d_parent) {
-				dprintk(DEBUG_ERROR,
-					"got to root without finding dentry\n");
-				break;
-			}
-			par = par->d_parent;
-		}
-
-/* XXX - there may be some duplication we can get rid of */
-		if (par == dentry) {
-			return_fid = v9fs_fid_walk_up(dentry);
-			if (IS_ERR(return_fid))
-				return_fid = NULL;
-		}
+		dprintk(DEBUG_ERROR, "Couldn't find a fid in dentry\n");
 	}
 
 	return return_fid;

fs/9p/fid.h

···
 	struct v9fs_fcall *rdir_fcall;
 
 	/* management stuff */
-	pid_t pid;		/* thread associated with this fid */
 	uid_t uid;		/* user associated with this fid */
 
 	/* private data */

fs/9p/v9fs.c

···
 	}
 
 	if (v9ses->afid != ~0) {
+		dprintk(DEBUG_ERROR, "afid not equal to ~0\n");
 		if (v9fs_t_clunk(v9ses, v9ses->afid))
 			dprintk(DEBUG_ERROR, "clunk failed\n");
 	}

fs/9p/vfs_dentry.c

···
 #include "fid.h"
 
 /**
- * v9fs_dentry_validate - VFS dcache hook to validate cache
- * @dentry:  dentry that is being validated
- * @nd: path data
+ * v9fs_dentry_delete - called when dentry refcount equals 0
+ * @dentry:  dentry in question
  *
- * dcache really shouldn't be used for 9P2000 as at all due to
- * potential attached semantics to directory traversal (walk).
- *
- * FUTURE: look into how to use dcache to allow multi-stage
- * walks in Plan 9 & potential for better dcache operation which
- * would remain valid for Plan 9 semantics.  Older versions
- * had validation via stat for those interested.  However, since
- * stat has the same approximate overhead as walk there really
- * is no difference.  The only improvement would be from a
- * time-decay cache like NFS has and that undermines the
- * synchronous nature of 9P2000.
+ * By returning 1 here we should remove cacheing of unused
+ * dentry components.
  *
  */
 
-static int v9fs_dentry_validate(struct dentry *dentry, struct nameidata *nd)
+int v9fs_dentry_delete(struct dentry *dentry)
 {
-	struct dentry *dc = current->fs->pwd;
-
-	dprintk(DEBUG_VFS, "dentry: %s (%p)\n", dentry->d_iname, dentry);
-	if (v9fs_fid_lookup(dentry)) {
-		dprintk(DEBUG_VFS, "VALID\n");
-		return 1;
-	}
-
-	while (dc != NULL) {
-		if (dc == dentry) {
-			dprintk(DEBUG_VFS, "VALID\n");
-			return 1;
-		}
-		if (dc == dc->d_parent)
-			break;
-
-		dc = dc->d_parent;
-	}
-
-	dprintk(DEBUG_VFS, "INVALID\n");
-	return 0;
+	dprintk(DEBUG_VFS, " dentry: %s (%p)\n", dentry->d_iname, dentry);
+	return 1;
 }
 
 /**
···
 }
 
 struct dentry_operations v9fs_dentry_operations = {
-	.d_revalidate = v9fs_dentry_validate,
+	.d_delete = v9fs_dentry_delete,
 	.d_release = v9fs_dentry_release,
 };