From patchwork Thu May 22 19:32:11 2014 Content-Type: text/plain; charset="utf-8" MIME-Version: 1.0 Content-Transfer-Encoding: 7bit X-Patchwork-Submitter: Bruce Fields X-Patchwork-Id: 4225981 Return-Path: X-Original-To: patchwork-linux-nfs@patchwork.kernel.org Delivered-To: patchwork-parsemail@patchwork2.web.kernel.org Received: from mail.kernel.org (mail.kernel.org [198.145.19.201]) by patchwork2.web.kernel.org (Postfix) with ESMTP id CB9F5BF90B for ; Thu, 22 May 2014 19:33:08 +0000 (UTC) Received: from mail.kernel.org (localhost [127.0.0.1]) by mail.kernel.org (Postfix) with ESMTP id B658820386 for ; Thu, 22 May 2014 19:33:07 +0000 (UTC) Received: from vger.kernel.org (vger.kernel.org [209.132.180.67]) by mail.kernel.org (Postfix) with ESMTP id 92BF22038D for ; Thu, 22 May 2014 19:33:06 +0000 (UTC) Received: (majordomo@vger.kernel.org) by vger.kernel.org via listexpand id S1752760AbaEVTcq (ORCPT ); Thu, 22 May 2014 15:32:46 -0400 Received: from fieldses.org ([174.143.236.118]:43457 "EHLO fieldses.org" rhost-flags-OK-OK-OK-OK) by vger.kernel.org with ESMTP id S1752671AbaEVTco (ORCPT ); Thu, 22 May 2014 15:32:44 -0400 Received: from bfields by fieldses.org with local (Exim 4.76) (envelope-from ) id 1WnYjL-0006rO-Jf; Thu, 22 May 2014 15:32:43 -0400 From: "J. Bruce Fields" To: linux-nfs@vger.kernel.org Cc: Christoph Hellwig , "J. Bruce Fields" Subject: [PATCH 36/52] nfsd4: allow large readdirs Date: Thu, 22 May 2014 15:32:11 -0400 Message-Id: <1400787148-25941-37-git-send-email-bfields@redhat.com> X-Mailer: git-send-email 1.7.9.5 In-Reply-To: <1400787148-25941-1-git-send-email-bfields@redhat.com> References: <1400787148-25941-1-git-send-email-bfields@redhat.com> Sender: linux-nfs-owner@vger.kernel.org Precedence: bulk List-ID: X-Mailing-List: linux-nfs@vger.kernel.org X-Spam-Status: No, score=-7.5 required=5.0 tests=BAYES_00, RCVD_IN_DNSWL_HI, RP_MATCHES_RCVD, UNPARSEABLE_RELAY autolearn=ham version=3.3.1 X-Spam-Checker-Version: SpamAssassin 3.3.1 (2010-03-16) on mail.kernel.org X-Virus-Scanned: ClamAV using ClamSMTP From: "J. Bruce Fields" Currently we limit readdir results to a single page. This can result in a performance regression compared to NFSv3 when reading large directories. Signed-off-by: J. Bruce Fields --- fs/nfsd/nfs4proc.c | 9 ++-- fs/nfsd/nfs4xdr.c | 137 +++++++++++++++++++++++++++++------------------------ fs/nfsd/xdr4.h | 5 +- 3 files changed, 82 insertions(+), 69 deletions(-) diff --git a/fs/nfsd/nfs4proc.c b/fs/nfsd/nfs4proc.c index e1196ed..f56f5e3 100644 --- a/fs/nfsd/nfs4proc.c +++ b/fs/nfsd/nfs4proc.c @@ -1502,13 +1502,14 @@ static inline u32 nfsd4_read_rsize(struct svc_rqst *rqstp, struct nfsd4_op *op) static inline u32 nfsd4_readdir_rsize(struct svc_rqst *rqstp, struct nfsd4_op *op) { + u32 maxcount = svc_max_payload(rqstp); u32 rlen = op->u.readdir.rd_maxcount; - if (rlen > PAGE_SIZE) - rlen = PAGE_SIZE; + if (rlen > maxcount) + rlen = maxcount; - return (op_encode_hdr_size + op_encode_verifier_maxsz) - * sizeof(__be32) + rlen; + return (op_encode_hdr_size + op_encode_verifier_maxsz + + XDR_QUADLEN(rlen)) * sizeof(__be32); } static inline u32 nfsd4_remove_rsize(struct svc_rqst *rqstp, struct nfsd4_op *op) diff --git a/fs/nfsd/nfs4xdr.c b/fs/nfsd/nfs4xdr.c index 311e281..a2524b3 100644 --- a/fs/nfsd/nfs4xdr.c +++ b/fs/nfsd/nfs4xdr.c @@ -2575,8 +2575,8 @@ 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) +nfsd4_encode_dirent_fattr(struct xdr_stream *xdr, struct nfsd4_readdir *cd, + const char *name, int namlen) { struct svc_export *exp = cd->rd_fhp->fh_export; struct dentry *dentry; @@ -2628,8 +2628,7 @@ nfsd4_encode_dirent_fattr(struct nfsd4_readdir *cd, } out_encode: - nfserr = nfsd4_encode_fattr_to_buf(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); @@ -2638,9 +2637,12 @@ out_put: } static __be32 * -nfsd4_encode_rdattr_error(__be32 *p, int buflen, __be32 nfserr) +nfsd4_encode_rdattr_error(struct xdr_stream *xdr, __be32 nfserr) { - if (buflen < 6) + __be32 *p; + + p = xdr_reserve_space(xdr, 6); + if (!p) return NULL; *p++ = htonl(2); *p++ = htonl(FATTR4_WORD0_RDATTR_ERROR); /* bmval0 */ @@ -2657,10 +2659,13 @@ 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 xdr_stream *xdr = cd->xdr; + int start_offset = xdr->buf->len; + int cookie_offset; + int entry_bytes; __be32 nfserr = nfserr_toosmall; + __be64 wire_offset; + __be32 *p; /* In nfsv4, "." and ".." never make it onto the wire.. */ if (name && isdotent(name, namlen)) { @@ -2668,19 +2673,24 @@ nfsd4_encode_dirent(void *ccdv, const char *name, int namlen, return 0; } - if (cd->offset) - xdr_encode_hyper(cd->offset, (u64) offset); + if (cd->cookie_offset) { + wire_offset = cpu_to_be64(offset); + write_bytes_to_xdr_buf(xdr->buf, cd->cookie_offset, + &wire_offset, 8); + } - buflen = cd->buflen - 4 - XDR_QUADLEN(namlen); - if (buflen < 0) + p = xdr_reserve_space(xdr, 4); + if (!p) goto fail; - *p++ = xdr_one; /* mark entry present */ - cookiep = p; + cookie_offset = xdr->buf->len; + p = xdr_reserve_space(xdr, 3*4 + namlen); + if (!p) + goto fail; p = xdr_encode_hyper(p, NFS_OFFSET_MAX); /* offset of next entry */ p = xdr_encode_array(p, name, namlen); /* name length & name */ - nfserr = nfsd4_encode_dirent_fattr(cd, name, namlen, &p, buflen); + nfserr = nfsd4_encode_dirent_fattr(xdr, cd, name, namlen); switch (nfserr) { case nfs_ok: break; @@ -2699,19 +2709,23 @@ 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); + p = nfsd4_encode_rdattr_error(xdr, nfserr); if (p == NULL) { nfserr = nfserr_toosmall; goto fail; } } - cd->buflen -= (p - cd->buffer); - cd->buffer = p; - cd->offset = cookiep; + nfserr = nfserr_toosmall; + entry_bytes = xdr->buf->len - start_offset; + if (entry_bytes > cd->rd_maxcount) + goto fail; + cd->rd_maxcount -= entry_bytes; + cd->cookie_offset = cookie_offset; skip_entry: cd->common.err = nfs_ok; return 0; fail: + xdr_truncate_encode(xdr, start_offset); cd->common.err = nfserr; return -EINVAL; } @@ -3206,10 +3220,11 @@ static __be32 nfsd4_encode_readdir(struct nfsd4_compoundres *resp, __be32 nfserr, struct nfsd4_readdir *readdir) { int maxcount; + int bytes_left; loff_t offset; + __be64 wire_offset; struct xdr_stream *xdr = &resp->xdr; int starting_len = xdr->buf->len; - __be32 *page, *tailbase; __be32 *p; if (nfserr) @@ -3219,38 +3234,38 @@ nfsd4_encode_readdir(struct nfsd4_compoundres *resp, __be32 nfserr, struct nfsd4 if (!p) return nfserr_resource; - if (resp->xdr.buf->page_len) - return nfserr_resource; - if (!*resp->rqstp->rq_next_page) - return nfserr_resource; - /* XXX: Following NFSv3, we ignore the READDIR verifier for now. */ WRITE32(0); WRITE32(0); resp->xdr.buf->head[0].iov_len = ((char *)resp->xdr.p) - (char *)resp->xdr.buf->head[0].iov_base; - tailbase = p; - - maxcount = PAGE_SIZE; - 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. + * Number of bytes left for directory entries allowing for the + * final 8 bytes of the readdir and a following failed op: + */ + bytes_left = xdr->buf->buflen - xdr->buf->len + - COMPOUND_ERR_SLACK_SPACE - 8; + if (bytes_left < 0) { + nfserr = nfserr_resource; + goto err_no_verf; + } + maxcount = min_t(u32, readdir->rd_maxcount, INT_MAX); + /* + * Note the rfc defines rd_maxcount as the size of the + * READDIR4resok structure, which includes the verifier above + * and the 8 bytes encoded at the end of this function: */ - maxcount = (maxcount >> 2) - 4; - if (maxcount < 0) { - nfserr = nfserr_toosmall; + if (maxcount < 16) { + nfserr = nfserr_toosmall; goto err_no_verf; } + maxcount = min_t(int, maxcount-16, bytes_left); - page = page_address(*(resp->rqstp->rq_next_page++)); + readdir->xdr = xdr; + readdir->rd_maxcount = maxcount; readdir->common.err = 0; - readdir->buflen = maxcount; - readdir->buffer = page; - readdir->offset = NULL; + readdir->cookie_offset = 0; offset = readdir->rd_cookie; nfserr = nfsd_readdir(readdir->rd_rqstp, readdir->rd_fhp, @@ -3258,33 +3273,31 @@ 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) - nfserr = nfserr_toosmall; + xdr->buf->len == starting_len + 8) { + /* nothing encoded; which limit did we hit?: */ + if (maxcount - 16 < bytes_left) + /* It was the fault of rd_maxcount: */ + nfserr = nfserr_toosmall; + else + /* We ran out of buffer space: */ + nfserr = nfserr_resource; + } if (nfserr) goto err_no_verf; - if (readdir->offset) - xdr_encode_hyper(readdir->offset, offset); + if (readdir->cookie_offset) { + wire_offset = cpu_to_be64(offset); + write_bytes_to_xdr_buf(xdr->buf, readdir->cookie_offset, + &wire_offset, 8); + } - p = readdir->buffer; + p = xdr_reserve_space(xdr, 8); + if (!p) { + WARN_ON_ONCE(1); + goto err_no_verf; + } *p++ = 0; /* no more entries */ *p++ = htonl(readdir->common.err == nfserr_eof); - resp->xdr.buf->page_len = ((char *)p) - - (char*)page_address(*(resp->rqstp->rq_next_page-1)); - xdr->buf->len += xdr->buf->page_len; - - xdr->iov = xdr->buf->tail; - - xdr->page_ptr++; - xdr->buf->buflen -= PAGE_SIZE; - xdr->iov = xdr->buf->tail; - - /* Use rest of head for padding and remaining ops: */ - resp->xdr.buf->tail[0].iov_base = tailbase; - resp->xdr.buf->tail[0].iov_len = 0; - resp->xdr.p = resp->xdr.buf->tail[0].iov_base; - resp->xdr.end = resp->xdr.p + - (PAGE_SIZE - resp->xdr.buf->head[0].iov_len)/4; return 0; err_no_verf: diff --git a/fs/nfsd/xdr4.h b/fs/nfsd/xdr4.h index 41e5229..18cbb6d 100644 --- a/fs/nfsd/xdr4.h +++ b/fs/nfsd/xdr4.h @@ -287,9 +287,8 @@ struct nfsd4_readdir { struct svc_fh * rd_fhp; /* response */ struct readdir_cd common; - __be32 * buffer; - int buflen; - __be32 * offset; + struct xdr_stream *xdr; + int cookie_offset; }; struct nfsd4_release_lockowner {