From mboxrd@z Thu Jan 1 00:00:00 1970 Return-Path: Received: from mail-iw0-f174.google.com ([209.85.214.174]:40091 "EHLO mail-iw0-f174.google.com" rhost-flags-OK-OK-OK-OK) by vger.kernel.org with ESMTP id S1755250Ab1ELPSf convert rfc822-to-8bit (ORCPT ); Thu, 12 May 2011 11:18:35 -0400 Received: by iwn34 with SMTP id 34so1495628iwn.19 for ; Thu, 12 May 2011 08:18:35 -0700 (PDT) In-Reply-To: <1304960811-3863-1-git-send-email-bhalevy@panasas.com> References: <4DC81E8C.6040901@panasas.com> <1304960811-3863-1-git-send-email-bhalevy@panasas.com> Date: Thu, 12 May 2011 11:18:35 -0400 Message-ID: Subject: Re: [PATCH v2 03/29] pnfs: Use byte-range for layoutget From: Fred Isaman To: Benny Halevy Cc: Trond Myklebust , Boaz Harrosh , linux-nfs@vger.kernel.org Content-Type: text/plain; charset=ISO-8859-1 Sender: linux-nfs-owner@vger.kernel.org List-ID: MIME-Version: 1.0 On Mon, May 9, 2011 at 1:06 PM, Benny Halevy wrote: > Add offset and count parameters to pnfs_update_layout and use them to get > the layout in the pageio path. > > 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. > > Signed-off-by: Benny Halevy > --- >  fs/nfs/pnfs.c  |  149 +++++++++++++++++++++++++++++++++++++++++++++----------- >  fs/nfs/pnfs.h  |    4 +- >  fs/nfs/read.c  |   10 +++- >  fs/nfs/write.c |    8 ++- >  4 files changed, 136 insertions(+), 35 deletions(-) > > diff --git a/fs/nfs/pnfs.c b/fs/nfs/pnfs.c > index d9ab972..e689bdf 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) >  { > @@ -466,7 +525,7 @@ 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) >  { >        struct inode *ino = lo->plh_inode; >        struct nfs_server *server = NFS_SERVER(ino); > @@ -497,11 +556,11 @@ 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; Do you want to align offet to page boundary? >        lgp->args.type = server->pnfs_curr_ld->id; >        lgp->args.inode = ino; >        lgp->args.ctx = get_nfs_open_context(ctx); > @@ -515,7 +574,7 @@ 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 */ > @@ -622,10 +681,24 @@ 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; > + > +       /* higher offset > lower offset */ > +       d = l1->offset - l2->offset; > +       if (d) > +               return d; > + > +       /* longer length > shorter length */ > +       d = l1->length - l2->length; > +       if (d) > +               return d; > + Assuming iomodes the same, don't we prefer seeing the longer to the shorter? Wouldn't we prefer a short rw to a long ro? >        /* read > read/write */ > -       return (int)(iomode2 == IOMODE_READ) - (int)(iomode1 == IOMODE_READ); > +       return (int)(l2->iomode == IOMODE_READ) - > +               (int)(l1->iomode == IOMODE_READ); >  } > >  static void > @@ -639,7 +712,7 @@ pnfs_insert_layout(struct pnfs_layout_hdr *lo, > >        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(&lp->pls_range, &lseg->pls_range) > 0) >                        continue; >                list_add_tail(&lseg->pls_list, &lp->pls_list); >                dprintk("%s: inserted lseg %p " > @@ -718,16 +791,28 @@ pnfs_find_alloc_layout(struct inode *ino) >  * READ                RW      true >  */ >  static int > -is_matching_lseg(struct pnfs_layout_segment *lseg, u32 iomode) > +is_matching_lseg(struct pnfs_layout_segment *lseg, > +                struct pnfs_layout_range *range) arguments to this should probably both be of struct pnfs_layout_range >  { > -       return (iomode != IOMODE_RW || lseg->pls_range.iomode == IOMODE_RW); > +       struct pnfs_layout_range range1; > + > +       if ((range->iomode == IOMODE_RW && > +            lseg->pls_range.iomode != IOMODE_RW) || > +           !lo_seg_intersecting(&lseg->pls_range, range)) > +               return 0; > + > +       /* range1 covers only the first byte in the range */ > +       range1 = *range; > +       range1.length = 1; > +       return lo_seg_contained(&lseg->pls_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; > > @@ -736,11 +821,11 @@ 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, range)) { >                        ret = get_lseg(lseg); >                        break; >                } > -               if (cmp_layout(iomode, lseg->pls_range.iomode) > 0) > +               if (cmp_layout(range, &lseg->pls_range) > 0) >                        break; >        } > > @@ -756,8 +841,15 @@ 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) >  { > +       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; > @@ -785,7 +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; > > @@ -807,7 +899,7 @@ pnfs_update_layout(struct inode *ino, >                spin_unlock(&clp->cl_lock); >        } > > -       lseg = send_layoutget(lo, ctx, iomode); > +       lseg = send_layoutget(lo, ctx, &arg); >        if (!lseg && first) { >                spin_lock(&clp->cl_lock); >                list_del_init(&lo->plh_layouts); > @@ -834,17 +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); >        if (!lseg || IS_ERR(lseg)) { > @@ -899,8 +980,13 @@ 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); > -       } > +       } 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); >  } > > @@ -921,8 +1007,13 @@ 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); > -       } > +       } 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 4cb0a0d..14a2af9 100644 > --- a/fs/nfs/pnfs.h > +++ b/fs/nfs/pnfs.h > @@ -129,7 +129,7 @@ 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); > +                  loff_t pos, u64 count, enum pnfs_iomode access_type); >  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 *, > @@ -248,7 +248,7 @@ 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) > +                  loff_t pos, u64 count, enum pnfs_iomode access_type) >  { >        return NULL; >  } > diff --git a/fs/nfs/read.c b/fs/nfs/read.c > index 7cded2b..10eff1c 100644 > --- a/fs/nfs/read.c > +++ b/fs/nfs/read.c > @@ -288,13 +288,17 @@ 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); > +       lseg = pnfs_update_layout(desc->pg_inode, req->wb_context, > +                                 req_offset(req), desc->pg_count, > +                                 IOMODE_READ); >        ClearPageError(page); >        offset = 0; >        nbytes = desc->pg_count; >        do { >                int ret2; > > +               /* FIXME: need a new layout segment? */ > + No need for FIXME, since we assume strongly throughout that the layouts are page aligned. Fred >                data = list_entry(list.next, struct nfs_read_data, pages); >                list_del_init(&data->pages); > > @@ -351,7 +355,9 @@ 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); > +               lseg = pnfs_update_layout(desc->pg_inode, req->wb_context, > +                                         req_offset(req), desc->pg_count, > +                                         IOMODE_READ); > >        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 e4cbc11..318e0a3 100644 > --- a/fs/nfs/write.c > +++ b/fs/nfs/write.c > @@ -940,7 +940,9 @@ 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); > +       lseg = pnfs_update_layout(desc->pg_inode, req->wb_context, > +                                 req_offset(req), desc->pg_count, > +                                 IOMODE_RW); >        ClearPageError(page); >        offset = 0; >        nbytes = desc->pg_count; > @@ -1014,7 +1016,9 @@ 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); > +               lseg = pnfs_update_layout(desc->pg_inode, req->wb_context, > +                                         req_offset(req), desc->pg_count, > +                                         IOMODE_RW); > >        if ((desc->pg_ioflags & FLUSH_COND_STABLE) && >            (desc->pg_moreio || NFS_I(desc->pg_inode)->ncommit)) > -- > 1.7.3.4 > > -- > 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 >