On Fri, Apr 05, 2019 at 02:10:13PM +0800, Murphy Zhou wrote:
> On Fri, Apr 5, 2019 at 14:08 Murphy Zhou <jencce.kernel@xxxxxxxxx> wrote:
>
> >
> >
> > On Fri, Apr 5, 2019 at 06:47 NeilBrown <neilb@xxxxxxxx> wrote:
> >
> >> On Thu, Apr 04 2019, Murphy Zhou wrote:
> >>
> >> > After this commit
> >> > f875a79 nfsd: allow nfsv3 readdir request to be larger.
> >> > nfsv3 readdir request size can be larger than PAGE_SIZE. So if the
> >> > directory been read is large enough, we can use multiple pages
> >> > in rq_respages. Update buffer count and page pointers like we do
> >> > in readdirplus to make this happen.
> >> >
> >> > Now listing a directory within 3000 files will panic because we
> >> > are counting in a wrong way and would write on random page.
> >> >
> >> > Fixes: f875a79 "nfsd: allow nfsv3 readdir request to be larger"
> >> > Signed-off-by: Murphy Zhou <jencce.kernel@xxxxxxxxx>
> >> > ---
> >> >
> >> > v2:
> >> > fix nfs3svc_decode_readdirargs to set page pointers like
> >> decode_readdirplusargs
> >> > do not test pointers in encode_entry as we've fixed them when decoding
> >> >
> >> > fs/nfsd/nfs3proc.c | 17 +++++++++++++++--
> >> > fs/nfsd/nfs3xdr.c | 11 +++++++++--
> >> > 2 files changed, 24 insertions(+), 4 deletions(-)
> >> >
> >> > diff --git a/fs/nfsd/nfs3proc.c b/fs/nfsd/nfs3proc.c
> >> > index 8f933e8..9bc32af 100644
> >> > --- a/fs/nfsd/nfs3proc.c
> >> > +++ b/fs/nfsd/nfs3proc.c
> >> > @@ -442,7 +442,9 @@
> >> > struct nfsd3_readdirargs *argp = rqstp->rq_argp;
> >> > struct nfsd3_readdirres *resp = rqstp->rq_resp;
> >> > __be32 nfserr;
> >> > - int count;
> >> > + int count = 0;
> >> > + struct page **p;
> >> > + caddr_t page_addr = NULL;
> >> >
> >> > dprintk("nfsd: READDIR(3) %s %d bytes at %d\n",
> >> > SVCFH_fmt(&argp->fh),
> >> > @@ -462,7 +464,18 @@
> >> > nfserr = nfsd_readdir(rqstp, &resp->fh, (loff_t*) &argp->cookie,
> >> > &resp->common,
> >> nfs3svc_encode_entry);
> >> > memcpy(resp->verf, argp->verf, 8);
> >> > - resp->count = resp->buffer - argp->buffer;
> >> > + count = 0;
> >>
> >> Thanks - looks good.
> >> Setting 'count' to zero a second time looks a bit clumsy, but that is a
> >> minor detail.
> >
> >
> Save some space in the stack :)
We're not *that* desperate for stack space. But, I can live with it.
Queuing up for 5.1.
--b.
>
> Thanks for your review!
>
> M
>
> >
> >>
> >> Reviewed-by: NeilBrown <neilb@xxxxxxxx>
> >>
> >> Thanks,
> >> NeilBrown
> >>
> >>
> >> > + for (p = rqstp->rq_respages + 1; p < rqstp->rq_next_page; p++) {
> >> > + page_addr = page_address(*p);
> >> > +
> >> > + if (((caddr_t)resp->buffer >= page_addr) &&
> >> > + ((caddr_t)resp->buffer < page_addr + PAGE_SIZE)) {
> >> > + count += (caddr_t)resp->buffer - page_addr;
> >> > + break;
> >> > + }
> >> > + count += PAGE_SIZE;
> >> > + }
> >> > + resp->count = count >> 2;
> >> > if (resp->offset) {
> >> > loff_t offset = argp->cookie;
> >> >
> >> > diff --git a/fs/nfsd/nfs3xdr.c b/fs/nfsd/nfs3xdr.c
> >> > index 93fea24..8d78912 100644
> >> > --- a/fs/nfsd/nfs3xdr.c
> >> > +++ b/fs/nfsd/nfs3xdr.c
> >> > @@ -573,6 +573,7 @@ void fill_post_wcc(struct svc_fh *fhp)
> >> > nfs3svc_decode_readdirargs(struct svc_rqst *rqstp, __be32 *p)
> >> > {
> >> > struct nfsd3_readdirargs *args = rqstp->rq_argp;
> >> > + int len;
> >> > u32 max_blocksize = svc_max_payload(rqstp);
> >> >
> >> > p = decode_fh(p, &args->fh);
> >> > @@ -582,8 +583,14 @@ void fill_post_wcc(struct svc_fh *fhp)
> >> > args->verf = p; p += 2;
> >> > args->dircount = ~0;
> >> > args->count = ntohl(*p++);
> >> > - args->count = min_t(u32, args->count, max_blocksize);
> >> > - args->buffer = page_address(*(rqstp->rq_next_page++));
> >> > + len = args->count = min_t(u32, args->count, max_blocksize);
> >> > +
> >> > + while (len > 0) {
> >> > + struct page *p = *(rqstp->rq_next_page++);
> >> > + if (!args->buffer)
> >> > + args->buffer = page_address(p);
> >> > + len -= PAGE_SIZE;
> >> > + }
> >> >
> >> > return xdr_argsize_check(rqstp, p);
> >> > }
> >> > --
> >> > 1.8.3.1
> >>
> >
