| Message ID | 20240106-ericvh-fix-cache-dups-v1-1-538c2074f363@kernel.org (mailing list archive) |
|---|---|
| State | New |
| Headers | show |
| Series | fs/9p: simplify inode lookup operations | expand |
Ooops -- realized after I sent the series out that this one can be dropped since I completely switch up the format of qid2ino to a macro in a future patch. On Fri, Jan 5, 2024 at 8:11 PM Eric Van Hensbergen <ericvh@kernel.org> wrote: > > The qid2ino code had a check to allow a reduced precision qid to inode number mapping, but it only checked if the sizes were > equal. Change to check if the ino_t is smaller than qid->path > before using the reduced precision version. > > Signed-off-by: Eric Van Hensbergen <ericvh@kernel.org> > --- > fs/9p/vfs_inode.c | 17 ++++++++++++----- > 1 file changed, 12 insertions(+), 5 deletions(-) > > diff --git a/fs/9p/vfs_inode.c b/fs/9p/vfs_inode.c > index b845ee18a80b..6d149ba12bc0 100644 > --- a/fs/9p/vfs_inode.c > +++ b/fs/9p/vfs_inode.c > @@ -1192,18 +1192,25 @@ v9fs_stat2inode(struct p9_wstat *stat, struct inode *inode, > * v9fs_qid2ino - convert qid into inode number > * @qid: qid to hash > * > - * BUG: potential for inode number collisions? > + * We add 2 to qid->path to keep reserved inode > + * numbers reserved. > + * > + * Its possible in the future we will make qid->path > + * just a hash to lookup inodes, but this breaks too > + * much right now. > + * > */ > > ino_t v9fs_qid2ino(struct p9_qid *qid) > { > - u64 path = qid->path + 2; > ino_t i = 0; > > - if (sizeof(ino_t) == sizeof(path)) > - memcpy(&i, &path, sizeof(ino_t)); > + WARN_ON((qid->path+2)==0); > + > + if (sizeof(ino_t) < sizeof(qid->path)) > + i = (ino_t) ((qid->path+2) ^ (qid->path >> 32)); > else > - i = (ino_t) (path ^ (path >> 32)); > + i = (ino_t) qid->path; > > return i; > } > > -- > 2.41.0 >
diff --git a/fs/9p/vfs_inode.c b/fs/9p/vfs_inode.c index b845ee18a80b..6d149ba12bc0 100644 --- a/fs/9p/vfs_inode.c +++ b/fs/9p/vfs_inode.c @@ -1192,18 +1192,25 @@ v9fs_stat2inode(struct p9_wstat *stat, struct inode *inode, * v9fs_qid2ino - convert qid into inode number * @qid: qid to hash * - * BUG: potential for inode number collisions? + * We add 2 to qid->path to keep reserved inode + * numbers reserved. + * + * Its possible in the future we will make qid->path + * just a hash to lookup inodes, but this breaks too + * much right now. + * */ ino_t v9fs_qid2ino(struct p9_qid *qid) { - u64 path = qid->path + 2; ino_t i = 0; - if (sizeof(ino_t) == sizeof(path)) - memcpy(&i, &path, sizeof(ino_t)); + WARN_ON((qid->path+2)==0); + + if (sizeof(ino_t) < sizeof(qid->path)) + i = (ino_t) ((qid->path+2) ^ (qid->path >> 32)); else - i = (ino_t) (path ^ (path >> 32)); + i = (ino_t) qid->path; return i; }
The qid2ino code had a check to allow a reduced precision qid to inode number mapping, but it only checked if the sizes were equal. Change to check if the ino_t is smaller than qid->path before using the reduced precision version. Signed-off-by: Eric Van Hensbergen <ericvh@kernel.org> --- fs/9p/vfs_inode.c | 17 ++++++++++++----- 1 file changed, 12 insertions(+), 5 deletions(-)