diff options
author | Eric Van Hensbergen <ericvh@gmail.com> | 2006-03-02 02:54:33 -0800 |
---|---|---|
committer | Linus Torvalds <torvalds@g5.osdl.org> | 2006-03-02 08:33:07 -0800 |
commit | 46f6dac259717551916405ee3388de89fb152bca (patch) | |
tree | 0e6c8b4505339e90c05ec8326220286fd19fe1f4 /fs/9p/vfs_dentry.c | |
parent | 74b8054c730785cd9db093e48f53337e521b6270 (diff) | |
download | lwn-46f6dac259717551916405ee3388de89fb152bca.tar.gz lwn-46f6dac259717551916405ee3388de89fb152bca.zip |
[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 <ericvh@gmail.com>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
Diffstat (limited to 'fs/9p/vfs_dentry.c')
-rw-r--r-- | fs/9p/vfs_dentry.c | 45 |
1 files changed, 8 insertions, 37 deletions
diff --git a/fs/9p/vfs_dentry.c b/fs/9p/vfs_dentry.c index 2dd806dac9f1..12c9cc926b71 100644 --- a/fs/9p/vfs_dentry.c +++ b/fs/9p/vfs_dentry.c @@ -43,47 +43,18 @@ #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; } /** @@ -118,6 +89,6 @@ void v9fs_dentry_release(struct dentry *dentry) } struct dentry_operations v9fs_dentry_operations = { - .d_revalidate = v9fs_dentry_validate, + .d_delete = v9fs_dentry_delete, .d_release = v9fs_dentry_release, }; |