On Tue, 2022-03-01 at 21:57 +0800, Xiubo Li wrote:
> On 3/1/22 9:20 PM, Jeff Layton wrote:
> > On Tue, 2022-03-01 at 19:30 +0800, xiubli@xxxxxxxxxx wrote:
> > > From: Xiubo Li <xiubli@xxxxxxxxxx>
> > >
> > > ------------[ cut here ]------------
> > > kernel BUG at fs/ceph/dir.c:537!
> > > invalid opcode: 0000 [#1] PREEMPT SMP KASAN NOPTI
> > > CPU: 16 PID: 21641 Comm: ls Tainted: G E 5.17.0-rc2+ #92
> > > Hardware name: Red Hat RHEV Hypervisor, BIOS 1.11.0-2.el7 04/01/2014
> > >
> > > The corresponding code in ceph_readdir() is:
> > >
> > > BUG_ON(rde->offset < ctx->pos);
> > >
> > > Signed-off-by: Xiubo Li <xiubli@xxxxxxxxxx>
> > > ---
> > > fs/ceph/dir.c | 13 +++++++------
> > > fs/ceph/inode.c | 5 +++--
> > > fs/ceph/mds_client.c | 2 +-
> > > 3 files changed, 11 insertions(+), 9 deletions(-)
> > >
> > > diff --git a/fs/ceph/dir.c b/fs/ceph/dir.c
> > > index a449f4a07c07..6be0c1f793c2 100644
> > > --- a/fs/ceph/dir.c
> > > +++ b/fs/ceph/dir.c
> > > @@ -534,6 +534,13 @@ static int ceph_readdir(struct file *file, struct dir_context *ctx)
> > > .ctext_len = rde->altname_len };
> > > u32 olen = oname.len;
> > >
> > > + err = ceph_fname_to_usr(&fname, &tname, &oname, NULL);
> > > + if (err) {
> > > + pr_err("%s unable to decode %.*s, got %d\n", __func__,
> > > + rde->name_len, rde->name, err);
> > > + goto out;
> > > + }
> > > +
> > > BUG_ON(rde->offset < ctx->pos);
> > > BUG_ON(!rde->inode.in);
> > >
> > > @@ -542,12 +549,6 @@ static int ceph_readdir(struct file *file, struct dir_context *ctx)
> > > i, rinfo->dir_nr, ctx->pos,
> > > rde->name_len, rde->name, &rde->inode.in);
> > >
> > > - err = ceph_fname_to_usr(&fname, &tname, &oname, NULL);
> > > - if (err) {
> > > - dout("Unable to decode %.*s. Skipping it.\n", rde->name_len, rde->name);
> > > - continue;
> > > - }
> > > -
> > > if (!dir_emit(ctx, oname.name, oname.len,
> > > ceph_present_ino(inode->i_sb, le64_to_cpu(rde->inode.in->ino)),
> > > le32_to_cpu(rde->inode.in->mode) >> 12)) {
> > > diff --git a/fs/ceph/inode.c b/fs/ceph/inode.c
> > > index 8b0832271fdf..2bc2f02b84e8 100644
> > > --- a/fs/ceph/inode.c
> > > +++ b/fs/ceph/inode.c
> > > @@ -1898,8 +1898,9 @@ int ceph_readdir_prepopulate(struct ceph_mds_request *req,
> > >
> > > err = ceph_fname_to_usr(&fname, &tname, &oname, &is_nokey);
> > > if (err) {
> > > - dout("Unable to decode %.*s. Skipping it.", rde->name_len, rde->name);
> > > - continue;
> > > + pr_err("%s unable to decode %.*s, got %d\n", __func__,
> > > + rde->name_len, rde->name, err);
> > > + goto out;
> > > }
> > >
> >
> > Is this really an improvement?
>
> Yeah, if we just continue without setting the rde->offset it will crash
> in "BUG_ON(rde->offset < ctx->pos);" in ceph_readdir().
>
>
Ok.
> > Suppose I have one dentry with a corrupt
> > name. Do I want to fail a readdir request which might allow me to get at
> > other dentries in that directory that isn't corrupt?
>
> It's a little hard to handle the code in ceph_readdir():
>
> 503 /* search start position */
> 504 if (rinfo->dir_nr > 0) {
> 505 int step, nr = rinfo->dir_nr;
> 506 while (nr > 0) {
> 507 step = nr >> 1;
> 508 if (rinfo->dir_entries[i + step].offset <
> ctx->pos) {
> 509 i += step + 1;
> 510 nr -= step + 1;
> 511 } else {
> 512 nr = step;
> 513 }
> 514 }
> 515 }
>
> In this case how to set the rde->offset ?
>
>
Yeah, that is the nasty part. I would probably just pretend that the
corrupt dentry doesn't exist. Offsets are set by ceph_make_fpos, AFAICT,
and I don't think skipping one should affect the position of the other.
OTOH, this is just a nice-to-have thing. If it's too nasty to deal with,
we can just return an error for now and aim to do better error handling
here later.
> >
> > Maybe we should try to emit some placeholder there?
> >
> >
> > > dname.name = oname.name;
> > > diff --git a/fs/ceph/mds_client.c b/fs/ceph/mds_client.c
> > > index 914a6e68bb56..94b4c6508044 100644
> > > --- a/fs/ceph/mds_client.c
> > > +++ b/fs/ceph/mds_client.c
> > > @@ -3474,7 +3474,7 @@ static void handle_reply(struct ceph_mds_session *session, struct ceph_msg *msg)
> > > if (err == 0) {
> > > if (result == 0 && (req->r_op == CEPH_MDS_OP_READDIR ||
> > > req->r_op == CEPH_MDS_OP_LSSNAP))
> > > - ceph_readdir_prepopulate(req, req->r_session);
> > > + err = ceph_readdir_prepopulate(req, req->r_session);
> > > }
> > > current->journal_info = NULL;
> > > mutex_unlock(&req->r_fill_mutex);
>
--
Jeff Layton <jlayton@xxxxxxxxxx>
