From patchwork Wed May 25 21:27:33 2011 Content-Type: text/plain; charset="utf-8" MIME-Version: 1.0 Content-Transfer-Encoding: 7bit X-Patchwork-Submitter: Benny Halevy X-Patchwork-Id: 817802 Received: from vger.kernel.org (vger.kernel.org [209.132.180.67]) by demeter1.kernel.org (8.14.4/8.14.3) with ESMTP id p4PLQZIt012631 for ; Wed, 25 May 2011 21:27:37 GMT Received: (majordomo@vger.kernel.org) by vger.kernel.org via listexpand id S1754884Ab1EYV1g (ORCPT ); Wed, 25 May 2011 17:27:36 -0400 Received: from daytona.panasas.com ([67.152.220.89]:57092 "EHLO daytona.panasas.com" rhost-flags-OK-OK-OK-OK) by vger.kernel.org with ESMTP id S1752869Ab1EYV1g (ORCPT ); Wed, 25 May 2011 17:27:36 -0400 Received: from fs1.bhalevy.com ([172.17.33.63]) by daytona.panasas.com with Microsoft SMTPSVC(6.0.3790.4675); Wed, 25 May 2011 17:27:34 -0400 From: Benny Halevy To: Trond Myklebust Cc: Boaz Harrosh , linux-nfs@vger.kernel.org Subject: [PATCH v8 09/32] pnfs: Use byte-range for layoutget Date: Thu, 26 May 2011 00:27:33 +0300 Message-Id: <1306358853-17112-1-git-send-email-bhalevy@panasas.com> X-Mailer: git-send-email 1.7.3.4 In-Reply-To: <4DDD7392.6040505@panasas.com> References: <4DDD7392.6040505@panasas.com> X-OriginalArrivalTime: 25 May 2011 21:27:35.0009 (UTC) FILETIME=[90439D10:01CC1B22] Sender: linux-nfs-owner@vger.kernel.org Precedence: bulk List-ID: X-Mailing-List: linux-nfs@vger.kernel.org X-Greylist: IP, sender and recipient auto-whitelisted, not delayed by milter-greylist-4.2.6 (demeter1.kernel.org [140.211.167.41]); Wed, 25 May 2011 21:27:37 +0000 (UTC) Add offset and count parameters to pnfs_update_layout and use them to get the layout in the pageio path. Order cache layout segments in the following order: * offset (ascending) * length (descending) * iomode (RW before READ) Test byte range against the layout segment in use in pnfs_{read,write}_pg_test so not to coalesce pages not using the same layout segment. [fix lseg ordering] [clean up pnfs_find_lseg lseg arg] [remove unnecessary FIXME] [fix ordering in pnfs_insert_layout] [clean up pnfs_insert_layout] Signed-off-by: Benny Halevy --- fs/nfs/pnfs.c | 165 ++++++++++++++++++++++++++++++++++++++++++------------- fs/nfs/pnfs.h | 6 ++- fs/nfs/read.c | 8 ++- fs/nfs/write.c | 8 ++- 4 files changed, 142 insertions(+), 45 deletions(-) diff --git a/fs/nfs/pnfs.c b/fs/nfs/pnfs.c index f57f528..c2f09e9 100644 --- a/fs/nfs/pnfs.c +++ b/fs/nfs/pnfs.c @@ -261,6 +261,65 @@ put_lseg(struct pnfs_layout_segment *lseg) } EXPORT_SYMBOL_GPL(put_lseg); +static inline u64 +end_offset(u64 start, u64 len) +{ + u64 end; + + end = start + len; + return end >= start ? end : NFS4_MAX_UINT64; +} + +/* last octet in a range */ +static inline u64 +last_byte_offset(u64 start, u64 len) +{ + u64 end; + + BUG_ON(!len); + end = start + len; + return end > start ? end - 1 : NFS4_MAX_UINT64; +} + +/* + * is l2 fully contained in l1? + * start1 end1 + * [----------------------------------) + * start2 end2 + * [----------------) + */ +static inline int +lo_seg_contained(struct pnfs_layout_range *l1, + struct pnfs_layout_range *l2) +{ + u64 start1 = l1->offset; + u64 end1 = end_offset(start1, l1->length); + u64 start2 = l2->offset; + u64 end2 = end_offset(start2, l2->length); + + return (start1 <= start2) && (end1 >= end2); +} + +/* + * is l1 and l2 intersecting? + * start1 end1 + * [----------------------------------) + * start2 end2 + * [----------------) + */ +static inline int +lo_seg_intersecting(struct pnfs_layout_range *l1, + struct pnfs_layout_range *l2) +{ + u64 start1 = l1->offset; + u64 end1 = end_offset(start1, l1->length); + u64 start2 = l2->offset; + u64 end2 = end_offset(start2, l2->length); + + return (end1 == NFS4_MAX_UINT64 || end1 > start2) && + (end2 == NFS4_MAX_UINT64 || end2 > start1); +} + static bool should_free_lseg(u32 lseg_iomode, u32 recall_iomode) { @@ -467,7 +526,7 @@ pnfs_choose_layoutget_stateid(nfs4_stateid *dst, struct pnfs_layout_hdr *lo, static struct pnfs_layout_segment * send_layoutget(struct pnfs_layout_hdr *lo, struct nfs_open_context *ctx, - u32 iomode, + struct pnfs_layout_range *range, gfp_t gfp_flags) { struct inode *ino = lo->plh_inode; @@ -499,11 +558,11 @@ send_layoutget(struct pnfs_layout_hdr *lo, goto out_err_free; } - lgp->args.minlength = NFS4_MAX_UINT64; + lgp->args.minlength = PAGE_CACHE_SIZE; + if (lgp->args.minlength > range->length) + lgp->args.minlength = range->length; lgp->args.maxcount = PNFS_LAYOUT_MAXSIZE; - lgp->args.range.iomode = iomode; - lgp->args.range.offset = 0; - lgp->args.range.length = NFS4_MAX_UINT64; + lgp->args.range = *range; lgp->args.type = server->pnfs_curr_ld->id; lgp->args.inode = ino; lgp->args.ctx = get_nfs_open_context(ctx); @@ -518,7 +577,7 @@ send_layoutget(struct pnfs_layout_hdr *lo, nfs4_proc_layoutget(lgp); if (!lseg) { /* remember that LAYOUTGET failed and suspend trying */ - set_bit(lo_fail_bit(iomode), &lo->plh_flags); + set_bit(lo_fail_bit(range->iomode), &lo->plh_flags); } /* free xdr pages */ @@ -625,10 +684,23 @@ bool pnfs_roc_drain(struct inode *ino, u32 *barrier) * are seen first. */ static s64 -cmp_layout(u32 iomode1, u32 iomode2) +cmp_layout(struct pnfs_layout_range *l1, + struct pnfs_layout_range *l2) { + s64 d; + + /* high offset > low offset */ + d = l1->offset - l2->offset; + if (d) + return d; + + /* short length > long length */ + d = l2->length - l1->length; + if (d) + return d; + /* read > read/write */ - return (int)(iomode2 == IOMODE_READ) - (int)(iomode1 == IOMODE_READ); + return (int)(l1->iomode == IOMODE_READ) - (int)(l2->iomode == IOMODE_READ); } static void @@ -636,13 +708,12 @@ pnfs_insert_layout(struct pnfs_layout_hdr *lo, struct pnfs_layout_segment *lseg) { struct pnfs_layout_segment *lp; - int found = 0; dprintk("%s:Begin\n", __func__); assert_spin_locked(&lo->plh_inode->i_lock); list_for_each_entry(lp, &lo->plh_segs, pls_list) { - if (cmp_layout(lp->pls_range.iomode, lseg->pls_range.iomode) > 0) + if (cmp_layout(&lseg->pls_range, &lp->pls_range) > 0) continue; list_add_tail(&lseg->pls_list, &lp->pls_list); dprintk("%s: inserted lseg %p " @@ -652,16 +723,14 @@ pnfs_insert_layout(struct pnfs_layout_hdr *lo, lseg->pls_range.offset, lseg->pls_range.length, lp, lp->pls_range.iomode, lp->pls_range.offset, lp->pls_range.length); - found = 1; - break; - } - if (!found) { - list_add_tail(&lseg->pls_list, &lo->plh_segs); - dprintk("%s: inserted lseg %p " - "iomode %d offset %llu length %llu at tail\n", - __func__, lseg, lseg->pls_range.iomode, - lseg->pls_range.offset, lseg->pls_range.length); + goto out; } + list_add_tail(&lseg->pls_list, &lo->plh_segs); + dprintk("%s: inserted lseg %p " + "iomode %d offset %llu length %llu at tail\n", + __func__, lseg, lseg->pls_range.iomode, + lseg->pls_range.offset, lseg->pls_range.length); +out: get_layout_hdr(lo); dprintk("%s:Return\n", __func__); @@ -721,16 +790,28 @@ pnfs_find_alloc_layout(struct inode *ino, gfp_t gfp_flags) * READ RW true */ static int -is_matching_lseg(struct pnfs_layout_segment *lseg, u32 iomode) +is_matching_lseg(struct pnfs_layout_range *ls_range, + struct pnfs_layout_range *range) { - return (iomode != IOMODE_RW || lseg->pls_range.iomode == IOMODE_RW); + struct pnfs_layout_range range1; + + if ((range->iomode == IOMODE_RW && + ls_range->iomode != IOMODE_RW) || + !lo_seg_intersecting(ls_range, range)) + return 0; + + /* range1 covers only the first byte in the range */ + range1 = *range; + range1.length = 1; + return lo_seg_contained(ls_range, &range1); } /* * lookup range in layout */ static struct pnfs_layout_segment * -pnfs_find_lseg(struct pnfs_layout_hdr *lo, u32 iomode) +pnfs_find_lseg(struct pnfs_layout_hdr *lo, + struct pnfs_layout_range *range) { struct pnfs_layout_segment *lseg, *ret = NULL; @@ -739,11 +820,11 @@ pnfs_find_lseg(struct pnfs_layout_hdr *lo, u32 iomode) assert_spin_locked(&lo->plh_inode->i_lock); list_for_each_entry(lseg, &lo->plh_segs, pls_list) { if (test_bit(NFS_LSEG_VALID, &lseg->pls_flags) && - is_matching_lseg(lseg, iomode)) { + is_matching_lseg(&lseg->pls_range, range)) { ret = get_lseg(lseg); break; } - if (cmp_layout(iomode, lseg->pls_range.iomode) > 0) + if (cmp_layout(range, &lseg->pls_range) > 0) break; } @@ -759,9 +840,16 @@ pnfs_find_lseg(struct pnfs_layout_hdr *lo, u32 iomode) struct pnfs_layout_segment * pnfs_update_layout(struct inode *ino, struct nfs_open_context *ctx, + loff_t pos, + u64 count, enum pnfs_iomode iomode, gfp_t gfp_flags) { + struct pnfs_layout_range arg = { + .iomode = iomode, + .offset = pos, + .length = count, + }; struct nfs_inode *nfsi = NFS_I(ino); struct nfs_client *clp = NFS_SERVER(ino)->nfs_client; struct pnfs_layout_hdr *lo; @@ -789,7 +877,7 @@ pnfs_update_layout(struct inode *ino, goto out_unlock; /* Check to see if the layout for the given range already exists */ - lseg = pnfs_find_lseg(lo, iomode); + lseg = pnfs_find_lseg(lo, &arg); if (lseg) goto out_unlock; @@ -811,7 +899,7 @@ pnfs_update_layout(struct inode *ino, spin_unlock(&clp->cl_lock); } - lseg = send_layoutget(lo, ctx, iomode, gfp_flags); + lseg = send_layoutget(lo, ctx, &arg, gfp_flags); if (!lseg && first) { spin_lock(&clp->cl_lock); list_del_init(&lo->plh_layouts); @@ -838,17 +926,6 @@ pnfs_layout_process(struct nfs4_layoutget *lgp) struct nfs_client *clp = NFS_SERVER(ino)->nfs_client; int status = 0; - /* Verify we got what we asked for. - * Note that because the xdr parsing only accepts a single - * element array, this can fail even if the server is behaving - * correctly. - */ - if (lgp->args.range.iomode > res->range.iomode || - res->range.offset != 0 || - res->range.length != NFS4_MAX_UINT64) { - status = -EINVAL; - goto out; - } /* Inject layout blob into I/O device driver */ lseg = NFS_SERVER(ino)->pnfs_curr_ld->alloc_lseg(lo, res, lgp->gfp_flags); if (!lseg || IS_ERR(lseg)) { @@ -903,9 +980,14 @@ static int pnfs_read_pg_test(struct nfs_pageio_descriptor *pgio, /* This is first coelesce call for a series of nfs_pages */ pgio->pg_lseg = pnfs_update_layout(pgio->pg_inode, prev->wb_context, + req_offset(req), + pgio->pg_count, IOMODE_READ, GFP_KERNEL); - } + } else if (pgio->pg_lseg && + req_offset(req) > end_offset(pgio->pg_lseg->pls_range.offset, + pgio->pg_lseg->pls_range.length)) + return 0; return NFS_SERVER(pgio->pg_inode)->pnfs_curr_ld->pg_test(pgio, prev, req); } @@ -926,9 +1008,14 @@ static int pnfs_write_pg_test(struct nfs_pageio_descriptor *pgio, /* This is first coelesce call for a series of nfs_pages */ pgio->pg_lseg = pnfs_update_layout(pgio->pg_inode, prev->wb_context, + req_offset(req), + pgio->pg_count, IOMODE_RW, GFP_NOFS); - } + } else if (pgio->pg_lseg && + req_offset(req) > end_offset(pgio->pg_lseg->pls_range.offset, + pgio->pg_lseg->pls_range.length)) + return 0; return NFS_SERVER(pgio->pg_inode)->pnfs_curr_ld->pg_test(pgio, prev, req); } diff --git a/fs/nfs/pnfs.h b/fs/nfs/pnfs.h index b7f88d4..7121139 100644 --- a/fs/nfs/pnfs.h +++ b/fs/nfs/pnfs.h @@ -131,7 +131,8 @@ void get_layout_hdr(struct pnfs_layout_hdr *lo); void put_lseg(struct pnfs_layout_segment *lseg); struct pnfs_layout_segment * pnfs_update_layout(struct inode *ino, struct nfs_open_context *ctx, - enum pnfs_iomode access_type, gfp_t gfp_flags); + loff_t pos, u64 count, enum pnfs_iomode access_type, + gfp_t gfp_flags); void set_pnfs_layoutdriver(struct nfs_server *, u32 id); void unset_pnfs_layoutdriver(struct nfs_server *); enum pnfs_try_status pnfs_try_to_write_data(struct nfs_write_data *, @@ -271,7 +272,8 @@ static inline void put_lseg(struct pnfs_layout_segment *lseg) static inline struct pnfs_layout_segment * pnfs_update_layout(struct inode *ino, struct nfs_open_context *ctx, - enum pnfs_iomode access_type, gfp_t gfp_flags) + loff_t pos, u64 count, enum pnfs_iomode access_type, + gfp_t gfp_flags) { return NULL; } diff --git a/fs/nfs/read.c b/fs/nfs/read.c index 2bcf0dc..540c8bc 100644 --- a/fs/nfs/read.c +++ b/fs/nfs/read.c @@ -288,7 +288,9 @@ static int nfs_pagein_multi(struct nfs_pageio_descriptor *desc) atomic_set(&req->wb_complete, requests); BUG_ON(desc->pg_lseg != NULL); - lseg = pnfs_update_layout(desc->pg_inode, req->wb_context, IOMODE_READ, GFP_KERNEL); + lseg = pnfs_update_layout(desc->pg_inode, req->wb_context, + req_offset(req), desc->pg_count, + IOMODE_READ, GFP_KERNEL); ClearPageError(page); offset = 0; nbytes = desc->pg_count; @@ -351,7 +353,9 @@ static int nfs_pagein_one(struct nfs_pageio_descriptor *desc) } req = nfs_list_entry(data->pages.next); if ((!lseg) && list_is_singular(&data->pages)) - lseg = pnfs_update_layout(desc->pg_inode, req->wb_context, IOMODE_READ, GFP_KERNEL); + lseg = pnfs_update_layout(desc->pg_inode, req->wb_context, + req_offset(req), desc->pg_count, + IOMODE_READ, GFP_KERNEL); ret = nfs_read_rpcsetup(req, data, &nfs_read_full_ops, desc->pg_count, 0, lseg); diff --git a/fs/nfs/write.c b/fs/nfs/write.c index 49c715b..7edb72f 100644 --- a/fs/nfs/write.c +++ b/fs/nfs/write.c @@ -939,7 +939,9 @@ static int nfs_flush_multi(struct nfs_pageio_descriptor *desc) atomic_set(&req->wb_complete, requests); BUG_ON(desc->pg_lseg); - lseg = pnfs_update_layout(desc->pg_inode, req->wb_context, IOMODE_RW, GFP_NOFS); + lseg = pnfs_update_layout(desc->pg_inode, req->wb_context, + req_offset(req), desc->pg_count, + IOMODE_RW, GFP_NOFS); ClearPageError(page); offset = 0; nbytes = desc->pg_count; @@ -1013,7 +1015,9 @@ static int nfs_flush_one(struct nfs_pageio_descriptor *desc) } req = nfs_list_entry(data->pages.next); if ((!lseg) && list_is_singular(&data->pages)) - lseg = pnfs_update_layout(desc->pg_inode, req->wb_context, IOMODE_RW, GFP_NOFS); + lseg = pnfs_update_layout(desc->pg_inode, req->wb_context, + req_offset(req), desc->pg_count, + IOMODE_RW, GFP_NOFS); if ((desc->pg_ioflags & FLUSH_COND_STABLE) && (desc->pg_moreio || NFS_I(desc->pg_inode)->ncommit))