| Message ID | 20171024190723.GD27853@fieldses.org (mailing list archive) |
|---|---|
| State | New, archived |
| Headers | show |
Quoting "J. Bruce Fields" <bfields@fieldses.org>: > On Tue, Oct 24, 2017 at 02:18:52PM -0400, Jeff Layton wrote: >> On Tue, 2017-10-24 at 13:53 -0400, J. Bruce Fields wrote: >> > On Tue, Oct 24, 2017 at 01:26:49PM -0400, Weston Andros Adamson wrote: >> > > Is there a reason to BUG() in these places? Couldn't we >> WARN_ON_ONCE and return an error? >> > >> > I think the BUG() will just kill an nfsd thread that isn't holding any >> > interesting locks. >> > >> >> Not necessarily. If panic_on_oops is set (and it usually is in >> "production" setups), it'll crash the box there. > > Maybe they're getting what they asked for? > >> > The failures look unlikely. (Except for that read_u32... return, I >> > wonder if we're missing a check there.) >> >> Agreed, looks like you only hit an error if the read attempts to go out >> of bounds. In principle that shouldn't ever happen (and I haven't seen >> any reports of it). >> >> Still...I agree with Dros that it's better to handle this without >> oopsing if we can. We can return an error from either of those >> functions. A sane error and a WARN_ONCE would be better here. > > OK, OK, OK. > > There are also some more BUGs that could use looking into if anyone > wants to. > > --b. > > commit eb754930662f > Author: J. Bruce Fields <bfields@redhat.com> > Date: Tue Oct 24 14:58:11 2017 -0400 > > rpc: remove some BUG()s > > It would be kinder to WARN() and recover in several spots here instead > of BUG()ing. > > Also, it looks like the read_u32_from_xdr_buf() call could actually > fail, though it might require a broken (or malicious) client, so convert > that to just an error return. > > Reported-by: Weston Andros Adamson <dros@monkey.org> > Signed-off-by: J. Bruce Fields <bfields@redhat.com> > > diff --git a/net/sunrpc/auth_gss/svcauth_gss.c > b/net/sunrpc/auth_gss/svcauth_gss.c > index 7b1ee5a0b03c..73165e9ca5bf 100644 > --- a/net/sunrpc/auth_gss/svcauth_gss.c > +++ b/net/sunrpc/auth_gss/svcauth_gss.c > @@ -855,11 +855,13 @@ unwrap_integ_data(struct svc_rqst *rqstp, > struct xdr_buf *buf, u32 seq, struct g > return stat; > if (integ_len > buf->len) > return stat; > - if (xdr_buf_subsegment(buf, &integ_buf, 0, integ_len)) > - BUG(); > + if (xdr_buf_subsegment(buf, &integ_buf, 0, integ_len)) { > + WARN_ON_ONCE(1); > + return stat; > + } > /* copy out mic... */ > if (read_u32_from_xdr_buf(buf, integ_len, &mic.len)) > - BUG(); > + return stat; > if (mic.len > RPC_MAX_AUTH_SIZE) > return stat; > mic.data = kmalloc(mic.len, GFP_KERNEL); > @@ -1611,8 +1613,10 @@ svcauth_gss_wrap_resp_integ(struct svc_rqst *rqstp) > BUG_ON(integ_len % 4); > *p++ = htonl(integ_len); > *p++ = htonl(gc->gc_seq); > - if (xdr_buf_subsegment(resbuf, &integ_buf, integ_offset, integ_len)) > - BUG(); > + if (xdr_buf_subsegment(resbuf, &integ_buf, integ_offset, integ_len)) { > + WARN_ON_ONCE(1); > + goto out_err; > + } > if (resbuf->tail[0].iov_base == NULL) { > if (resbuf->head[0].iov_len + RPC_MAX_AUTH_SIZE > PAGE_SIZE) > goto out_err; What about the following BUG() at net/sunrpc/svc_xprt.c:1058: /* * Remove a dead transport */ static void svc_delete_xprt(struct svc_xprt *xprt) { struct svc_serv *serv = xprt->xpt_server; struct svc_deferred_req *dr; /* Only do this once */ if (test_and_set_bit(XPT_DEAD, &xprt->xpt_flags)) BUG(); dprintk("svc: svc_delete_xprt(%p)\n", xprt); xprt->xpt_ops->xpo_detach(xprt); spin_lock_bh(&serv->sv_lock); list_del_init(&xprt->xpt_list); WARN_ON_ONCE(!list_empty(&xprt->xpt_ready)); if (test_bit(XPT_TEMP, &xprt->xpt_flags)) serv->sv_tmpcnt--; spin_unlock_bh(&serv->sv_lock); while ((dr = svc_deferred_dequeue(xprt)) != NULL) kfree(dr); call_xpt_users(xprt); svc_xprt_put(xprt); } I'm suspicious about that comment above the _if_ condition: /* Only do this once */ Would it help to replace that BUG with a WARN_ON_ONCE? Thanks -- Gustavo A. R. Silva -- To unsubscribe from this list: send the line "unsubscribe linux-nfs" in the body of a message to majordomo@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html
On Tue, 2017-10-24 at 15:12 -0500, Gustavo A. R. Silva wrote: > Quoting "J. Bruce Fields" <bfields@fieldses.org>: > > > On Tue, Oct 24, 2017 at 02:18:52PM -0400, Jeff Layton wrote: > > > On Tue, 2017-10-24 at 13:53 -0400, J. Bruce Fields wrote: > > > > On Tue, Oct 24, 2017 at 01:26:49PM -0400, Weston Andros Adamson wrote: > > > > > Is there a reason to BUG() in these places? Couldn't we > > > > > > WARN_ON_ONCE and return an error? > > > > > > > > I think the BUG() will just kill an nfsd thread that isn't holding any > > > > interesting locks. > > > > > > > > > > Not necessarily. If panic_on_oops is set (and it usually is in > > > "production" setups), it'll crash the box there. > > > > Maybe they're getting what they asked for? > > > > > > The failures look unlikely. (Except for that read_u32... return, I > > > > wonder if we're missing a check there.) > > > > > > Agreed, looks like you only hit an error if the read attempts to go out > > > of bounds. In principle that shouldn't ever happen (and I haven't seen > > > any reports of it). > > > > > > Still...I agree with Dros that it's better to handle this without > > > oopsing if we can. We can return an error from either of those > > > functions. A sane error and a WARN_ONCE would be better here. > > > > OK, OK, OK. > > > > There are also some more BUGs that could use looking into if anyone > > wants to. > > > > --b. > > > > commit eb754930662f > > Author: J. Bruce Fields <bfields@redhat.com> > > Date: Tue Oct 24 14:58:11 2017 -0400 > > > > rpc: remove some BUG()s > > > > It would be kinder to WARN() and recover in several spots here instead > > of BUG()ing. > > > > Also, it looks like the read_u32_from_xdr_buf() call could actually > > fail, though it might require a broken (or malicious) client, so convert > > that to just an error return. > > > > Reported-by: Weston Andros Adamson <dros@monkey.org> > > Signed-off-by: J. Bruce Fields <bfields@redhat.com> > > > > diff --git a/net/sunrpc/auth_gss/svcauth_gss.c > > b/net/sunrpc/auth_gss/svcauth_gss.c > > index 7b1ee5a0b03c..73165e9ca5bf 100644 > > --- a/net/sunrpc/auth_gss/svcauth_gss.c > > +++ b/net/sunrpc/auth_gss/svcauth_gss.c > > @@ -855,11 +855,13 @@ unwrap_integ_data(struct svc_rqst *rqstp, > > struct xdr_buf *buf, u32 seq, struct g > > return stat; > > if (integ_len > buf->len) > > return stat; > > - if (xdr_buf_subsegment(buf, &integ_buf, 0, integ_len)) > > - BUG(); > > + if (xdr_buf_subsegment(buf, &integ_buf, 0, integ_len)) { > > + WARN_ON_ONCE(1); > > + return stat; > > + } > > /* copy out mic... */ > > if (read_u32_from_xdr_buf(buf, integ_len, &mic.len)) > > - BUG(); > > + return stat; > > if (mic.len > RPC_MAX_AUTH_SIZE) > > return stat; > > mic.data = kmalloc(mic.len, GFP_KERNEL); > > @@ -1611,8 +1613,10 @@ svcauth_gss_wrap_resp_integ(struct svc_rqst *rqstp) > > BUG_ON(integ_len % 4); > > *p++ = htonl(integ_len); > > *p++ = htonl(gc->gc_seq); > > - if (xdr_buf_subsegment(resbuf, &integ_buf, integ_offset, integ_len)) > > - BUG(); > > + if (xdr_buf_subsegment(resbuf, &integ_buf, integ_offset, integ_len)) { > > + WARN_ON_ONCE(1); > > + goto out_err; > > + } > > if (resbuf->tail[0].iov_base == NULL) { > > if (resbuf->head[0].iov_len + RPC_MAX_AUTH_SIZE > PAGE_SIZE) > > goto out_err; > > What about the following BUG() at net/sunrpc/svc_xprt.c:1058: > > /* > * Remove a dead transport > */ > static void svc_delete_xprt(struct svc_xprt *xprt) > { > struct svc_serv *serv = xprt->xpt_server; > struct svc_deferred_req *dr; > > /* Only do this once */ > if (test_and_set_bit(XPT_DEAD, &xprt->xpt_flags)) > BUG(); > > dprintk("svc: svc_delete_xprt(%p)\n", xprt); > xprt->xpt_ops->xpo_detach(xprt); > > spin_lock_bh(&serv->sv_lock); > list_del_init(&xprt->xpt_list); > WARN_ON_ONCE(!list_empty(&xprt->xpt_ready)); > if (test_bit(XPT_TEMP, &xprt->xpt_flags)) > serv->sv_tmpcnt--; > spin_unlock_bh(&serv->sv_lock); > > while ((dr = svc_deferred_dequeue(xprt)) != NULL) > kfree(dr); > > call_xpt_users(xprt); > svc_xprt_put(xprt); > } > > I'm suspicious about that comment above the _if_ condition: /* Only do > this once */ > > Would it help to replace that BUG with a WARN_ON_ONCE? > Maybe. How safe is it to run all of that more than once after XPT_DEAD is set? You'd probably just end up crashing at some later point down the line anyway... I suppose you could do a WARN() there and then return without actually doing anything. That might let you limp along if you did hit this.
diff --git a/net/sunrpc/auth_gss/svcauth_gss.c b/net/sunrpc/auth_gss/svcauth_gss.c index 7b1ee5a0b03c..73165e9ca5bf 100644 --- a/net/sunrpc/auth_gss/svcauth_gss.c +++ b/net/sunrpc/auth_gss/svcauth_gss.c @@ -855,11 +855,13 @@ unwrap_integ_data(struct svc_rqst *rqstp, struct xdr_buf *buf, u32 seq, struct g return stat; if (integ_len > buf->len) return stat; - if (xdr_buf_subsegment(buf, &integ_buf, 0, integ_len)) - BUG(); + if (xdr_buf_subsegment(buf, &integ_buf, 0, integ_len)) { + WARN_ON_ONCE(1); + return stat; + } /* copy out mic... */ if (read_u32_from_xdr_buf(buf, integ_len, &mic.len)) - BUG(); + return stat; if (mic.len > RPC_MAX_AUTH_SIZE) return stat; mic.data = kmalloc(mic.len, GFP_KERNEL); @@ -1611,8 +1613,10 @@ svcauth_gss_wrap_resp_integ(struct svc_rqst *rqstp) BUG_ON(integ_len % 4); *p++ = htonl(integ_len); *p++ = htonl(gc->gc_seq); - if (xdr_buf_subsegment(resbuf, &integ_buf, integ_offset, integ_len)) - BUG(); + if (xdr_buf_subsegment(resbuf, &integ_buf, integ_offset, integ_len)) { + WARN_ON_ONCE(1); + goto out_err; + } if (resbuf->tail[0].iov_base == NULL) { if (resbuf->head[0].iov_len + RPC_MAX_AUTH_SIZE > PAGE_SIZE) goto out_err;