From: "J. Bruce Fields" <bfields@xxxxxxxxxx>
Rewrite readdir encoding using the new xdr primitives. This allows us
to return more than 4k of data in a readdir reply, improving performance
on large directories.
Signed-off-by: J. Bruce Fields <bfields@xxxxxxxxxx>
---
fs/nfsd/nfs4xdr.c | 157 +++++++++++++++++++++++++++++-------------------------
fs/nfsd/xdr4.h | 5 +-
2 files changed, 85 insertions(+), 77 deletions(-)
diff --git a/fs/nfsd/nfs4xdr.c b/fs/nfsd/nfs4xdr.c
index a378435..84ad067 100644
--- a/fs/nfsd/nfs4xdr.c
+++ b/fs/nfsd/nfs4xdr.c
@@ -1870,6 +1870,11 @@ static void write64(struct svcxdr_ptr *ptr, u64 n)
writemem(ptr, i, sizeof(i));
}
+static void write_nfserr(struct svcxdr_ptr *ptr, __be32 err)
+{
+ writemem(ptr, &err, sizeof(err));
+}
+
static bool write_opaque(struct svcxdr_stream *xdr, const void *data, int len)
{
struct svcxdr_ptr ptr;
@@ -2672,7 +2677,7 @@ static inline int attributes_need_mount(u32 *bmval)
static __be32
nfsd4_encode_dirent_fattr(struct nfsd4_readdir *cd,
- const char *name, int namlen, __be32 **p, int buflen)
+ const char *name, int namlen, struct svcxdr_stream *xdr)
{
struct svc_export *exp = cd->rd_fhp->fh_export;
struct dentry *dentry;
@@ -2724,7 +2729,7 @@ nfsd4_encode_dirent_fattr(struct nfsd4_readdir *cd,
}
out_encode:
- nfserr = nfsd4_encode_fattr_to_buffer(p, buflen, NULL, exp, dentry, cd->rd_bmval,
+ nfserr = nfsd4_encode_fattr(xdr, NULL, exp, dentry, cd->rd_bmval,
cd->rd_rqstp, ignore_crossmnt);
out_put:
dput(dentry);
@@ -2732,18 +2737,20 @@ out_put:
return nfserr;
}
-static __be32 *
-nfsd4_encode_rdattr_error(__be32 *p, int buflen, __be32 nfserr)
+static bool
+nfsd4_encode_rdattr_error(struct svcxdr_stream *xdr, __be32 nfserr)
{
- if (buflen < 6)
- return NULL;
- *p++ = htonl(2);
- *p++ = htonl(FATTR4_WORD0_RDATTR_ERROR); /* bmval0 */
- *p++ = htonl(0); /* bmval1 */
+ struct svcxdr_ptr ptr;
- *p++ = htonl(4); /* attribute length */
- *p++ = nfserr; /* no htonl */
- return p;
+ if (!svcxdr_reserve_space(xdr, &ptr, 5*4))
+ return false;
+ write32(&ptr, 2);
+ write32(&ptr, FATTR4_WORD0_RDATTR_ERROR); /* bmval0 */
+ write32(&ptr, 0); /* bmval1 */
+
+ write32(&ptr, 4); /* attribute length */
+ write_nfserr(&ptr, nfserr);
+ return true;
}
static int
@@ -2752,9 +2759,10 @@ nfsd4_encode_dirent(void *ccdv, const char *name, int namlen,
{
struct readdir_cd *ccd = ccdv;
struct nfsd4_readdir *cd = container_of(ccd, struct nfsd4_readdir, common);
- int buflen;
- __be32 *p = cd->buffer;
- __be32 *cookiep;
+ struct svcxdr_stream *xdr = cd->xdr;
+ struct svcxdr_ptr start = xdr->ptr;
+ struct svcxdr_ptr cookiep;
+ struct svcxdr_ptr ptr;
__be32 nfserr = nfserr_toosmall;
/* In nfsv4, "." and ".." never make it onto the wire.. */
@@ -2763,27 +2771,30 @@ nfsd4_encode_dirent(void *ccdv, const char *name, int namlen,
return 0;
}
- if (cd->offset)
- xdr_encode_hyper(cd->offset, (u64) offset);
+ if (cd->offset.p) {
+ ptr = cd->offset;
+ write64(&ptr, (u64)offset);
+ }
- buflen = cd->buflen - 4 - XDR_QUADLEN(namlen);
- if (buflen < 0)
+ if (!svcxdr_reserve_space(xdr, &ptr, 3*4))
goto fail;
+ write32(&ptr, 1); /* mark entry present */
+ cookiep = ptr;
+ write64(&ptr, NFS_OFFSET_MAX); /* offset of next entry */
- *p++ = xdr_one; /* mark entry present */
- cookiep = p;
- p = xdr_encode_hyper(p, NFS_OFFSET_MAX); /* offset of next entry */
- p = xdr_encode_array(p, name, namlen); /* name length & name */
+ if (!write_opaque(xdr, name, namlen))
+ goto fail;
- nfserr = nfsd4_encode_dirent_fattr(cd, name, namlen, &p, buflen);
+ nfserr = nfsd4_encode_dirent_fattr(cd, name, namlen, xdr);
switch (nfserr) {
case nfs_ok:
break;
case nfserr_resource:
nfserr = nfserr_toosmall;
goto fail;
- case nfserr_noent:
- goto skip_entry;
+ case nfserr_noent: /* skip entry: */
+ svcxdr_reset(xdr, &start);
+ goto out;
default:
/*
* If the client requested the RDATTR_ERROR attribute,
@@ -2794,19 +2805,17 @@ nfsd4_encode_dirent(void *ccdv, const char *name, int namlen,
*/
if (!(cd->rd_bmval[0] & FATTR4_WORD0_RDATTR_ERROR))
goto fail;
- p = nfsd4_encode_rdattr_error(p, buflen, nfserr);
- if (p == NULL) {
+ if (!nfsd4_encode_rdattr_error(xdr, nfserr)) {
nfserr = nfserr_toosmall;
goto fail;
}
}
- cd->buflen -= (p - cd->buffer);
- cd->buffer = p;
cd->offset = cookiep;
-skip_entry:
+out:
cd->common.err = nfs_ok;
return 0;
fail:
+ svcxdr_reset(xdr, &start);
cd->common.err = nfserr;
return -EINVAL;
}
@@ -3231,46 +3240,45 @@ nfsd4_encode_readdir(struct nfsd4_compoundres *resp, __be32 nfserr, struct nfsd4
{
int maxcount;
loff_t offset;
- __be32 *page, *savep, *tailbase;
- __be32 *p;
+ struct svcxdr_stream xdr;
+ struct svcxdr_ptr p;
+ struct svcxdr_ptr savep;
if (nfserr)
return nfserr;
- if (resp->xbuf->page_len)
- return nfserr_resource;
if (!*resp->rqstp->rq_next_page)
return nfserr_resource;
- RESERVE_SPACE(NFS4_VERIFIER_SIZE);
- savep = p;
-
- /* XXX: Following NFSv3, we ignore the READDIR verifier for now. */
- WRITE32(0);
- WRITE32(0);
- ADJUST_ARGS();
- resp->xbuf->head[0].iov_len = ((char*)resp->p) - (char*)resp->xbuf->head[0].iov_base;
- tailbase = p;
-
- maxcount = PAGE_SIZE;
+ nfserr = svcxdr_stream_init_from_resp(&xdr, resp);
+ if (nfserr)
+ return nfserr;
+ maxcount = svc_max_payload(resp->rqstp);
if (maxcount > readdir->rd_maxcount)
maxcount = readdir->rd_maxcount;
-
/*
- * Convert from bytes to words, account for the two words already
- * written, make sure to leave two words at the end for the next
- * pointer and eof field.
+ * Allow space for the two words (entry follows, eof) at the
+ * end. Note we generally assume rd_maxcount is at fault.
+ * Which is the most likely case. If someone sent a readdir at
+ * the end of an extremely long compound this might result in us
+ * returning the wrong error:
*/
- maxcount = (maxcount >> 2) - 4;
- if (maxcount < 0) {
- nfserr = nfserr_toosmall;
+ maxcount -= 8;
+ nfserr = nfserr_toosmall;
+ if (maxcount < 0)
goto err_no_verf;
- }
- page = page_address(*(resp->rqstp->rq_next_page++));
+ xdr.maxbytes = maxcount;
+ if (!svcxdr_reserve_space(&xdr, &p, 8))
+ goto err_no_verf;
+ savep = p;
+
+ /* XXX: Following NFSv3, we ignore the READDIR verifier for now. */
+ write32(&p, 0);
+ write32(&p, 0);
+
+ readdir->xdr = &xdr;
readdir->common.err = 0;
- readdir->buflen = maxcount;
- readdir->buffer = page;
- readdir->offset = NULL;
+ readdir->offset.p = NULL;
offset = readdir->rd_cookie;
nfserr = nfsd_readdir(readdir->rd_rqstp, readdir->rd_fhp,
@@ -3278,30 +3286,31 @@ nfsd4_encode_readdir(struct nfsd4_compoundres *resp, __be32 nfserr, struct nfsd4
&readdir->common, nfsd4_encode_dirent);
if (nfserr == nfs_ok &&
readdir->common.err == nfserr_toosmall &&
- readdir->buffer == page)
+ xdr.ptr.p == xdr.last_commit.p + 2) /* We made no progress at all */
nfserr = nfserr_toosmall;
if (nfserr)
goto err_no_verf;
- if (readdir->offset)
- xdr_encode_hyper(readdir->offset, offset);
-
- p = readdir->buffer;
- *p++ = 0; /* no more entries */
- *p++ = htonl(readdir->common.err == nfserr_eof);
- resp->xbuf->page_len = ((char*)p) -
- (char*)page_address(*(resp->rqstp->rq_next_page-1));
+ if (readdir->offset.p) {
+ p = readdir->offset;
+ write64(&p, offset);
+ }
+ xdr.maxbytes += 8;
+ if (!svcxdr_reserve_space(&xdr, &p, 8)) {
+ WARN_ON_ONCE(1);
+ goto err_no_verf;
+ }
+ write32(&p, 0); /* no more entries */
+ write32(&p, readdir->common.err == nfserr_eof);
- /* Use rest of head for padding and remaining ops: */
- resp->xbuf->tail[0].iov_base = tailbase;
- resp->xbuf->tail[0].iov_len = 0;
- resp->p = resp->xbuf->tail[0].iov_base;
- resp->end = resp->p + (PAGE_SIZE - resp->xbuf->head[0].iov_len)/4;
+ svcxdr_stream_update_resp(&xdr, resp);
return 0;
err_no_verf:
- p = savep;
- ADJUST_ARGS();
+ /*
+ * returning without svcxdr_stream_update_resp will leave
+ * xdr_buf unchanged, as if we'd encoded nothing:
+ */
return nfserr;
}
diff --git a/fs/nfsd/xdr4.h b/fs/nfsd/xdr4.h
index 36b8c2c..9c58f05 100644
--- a/fs/nfsd/xdr4.h
+++ b/fs/nfsd/xdr4.h
@@ -299,9 +299,8 @@ struct nfsd4_readdir {
struct svc_fh * rd_fhp; /* response */
struct readdir_cd common;
- __be32 * buffer;
- int buflen;
- __be32 * offset;
+ struct svcxdr_stream *xdr;
+ struct svcxdr_ptr offset;
};
struct nfsd4_release_lockowner {
--
1.7.11.7
--
To unsubscribe from this list: send the line "unsubscribe linux-nfs" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at http://vger.kernel.org/majordomo-info.html
