NFSv4.1/pnfs: Allow pNFS device drivers to customise layout segment insertion

This is needed in order to allow merging of contiguous layout segments,
and also to correct the ordering of layouts for those device drivers that
don't necessarily want to place the read-write layouts first.

Signed-off-by: Trond Myklebust <trond.myklebust@primarydata.com>

diff --git a/fs/nfs/pnfs.c b/fs/nfs/pnfs.c
index 68cc4b16976977f956d8a3046dd341c804596b87..914c1daf08dfe32c418caf9f0cfdcf1f9e774c23 100644 (file)
--- a/fs/nfs/pnfs.c
+++ b/fs/nfs/pnfs.c
@@ -1188,16 +1188,41 @@ pnfs_lseg_range_cmp(const struct pnfs_layout_range *l1,
        return (int)(l1->iomode == IOMODE_READ) - (int)(l2->iomode == IOMODE_READ);
 }
 
-static void
-pnfs_layout_insert_lseg(struct pnfs_layout_hdr *lo,
-                  struct pnfs_layout_segment *lseg)
+static bool
+pnfs_lseg_range_is_after(const struct pnfs_layout_range *l1,
+               const struct pnfs_layout_range *l2)
+{
+       return pnfs_lseg_range_cmp(l1, l2) > 0;
+}
+
+static bool
+pnfs_lseg_no_merge(struct pnfs_layout_segment *lseg,
+               struct pnfs_layout_segment *old)
+{
+       return false;
+}
+
+void
+pnfs_generic_layout_insert_lseg(struct pnfs_layout_hdr *lo,
+                  struct pnfs_layout_segment *lseg,
+                  bool (*is_after)(const struct pnfs_layout_range *,
+                          const struct pnfs_layout_range *),
+                  bool (*do_merge)(struct pnfs_layout_segment *,
+                          struct pnfs_layout_segment *),
+                  struct list_head *free_me)
 {
-       struct pnfs_layout_segment *lp;
+       struct pnfs_layout_segment *lp, *tmp;
 
        dprintk("%s:Begin\n", __func__);
 
-       list_for_each_entry(lp, &lo->plh_segs, pls_list) {
-               if (pnfs_lseg_range_cmp(&lseg->pls_range, &lp->pls_range) > 0)
+       list_for_each_entry_safe(lp, tmp, &lo->plh_segs, pls_list) {
+               if (test_bit(NFS_LSEG_VALID, &lp->pls_flags) == 0)
+                       continue;
+               if (do_merge(lseg, lp)) {
+                       mark_lseg_invalid(lp, free_me);
+                       continue;
+               }
+               if (is_after(&lseg->pls_range, &lp->pls_range))
                        continue;
                list_add_tail(&lseg->pls_list, &lp->pls_list);
                dprintk("%s: inserted lseg %p "
@@ -1219,6 +1244,24 @@ out:
 
        dprintk("%s:Return\n", __func__);
 }
+EXPORT_SYMBOL_GPL(pnfs_generic_layout_insert_lseg);
+
+static void
+pnfs_layout_insert_lseg(struct pnfs_layout_hdr *lo,
+                  struct pnfs_layout_segment *lseg,
+                  struct list_head *free_me)
+{
+       struct inode *inode = lo->plh_inode;
+       struct pnfs_layoutdriver_type *ld = NFS_SERVER(inode)->pnfs_curr_ld;
+
+       if (ld->add_lseg != NULL)
+               ld->add_lseg(lo, lseg, free_me);
+       else
+               pnfs_generic_layout_insert_lseg(lo, lseg,
+                               pnfs_lseg_range_is_after,
+                               pnfs_lseg_no_merge,
+                               free_me);
+}
 
 static struct pnfs_layout_hdr *
 alloc_init_layout_hdr(struct inode *ino,
@@ -1311,8 +1354,6 @@ pnfs_find_lseg(struct pnfs_layout_hdr *lo,
                        ret = pnfs_get_lseg(lseg);
                        break;
                }
-               if (lseg->pls_range.offset > range->offset)
-                       break;
        }
 
        dprintk("%s:Return lseg %p ref %d\n",
@@ -1637,7 +1678,7 @@ pnfs_layout_process(struct nfs4_layoutget *lgp)
        clear_bit(NFS_LAYOUT_INVALID_STID, &lo->plh_flags);
 
        pnfs_get_lseg(lseg);
-       pnfs_layout_insert_lseg(lo, lseg);
+       pnfs_layout_insert_lseg(lo, lseg, &free_me);
 
        if (res->return_on_close)
                set_bit(NFS_LSEG_ROC, &lseg->pls_flags);

diff --git a/fs/nfs/pnfs.h b/fs/nfs/pnfs.h
index 4df87ef3dccc4622b425aba5e9c9410e9ffcf727..869069d8b99629fa769f61f467d04b685ab2f45b 100644 (file)
--- a/fs/nfs/pnfs.h
+++ b/fs/nfs/pnfs.h
@@ -128,6 +128,9 @@ struct pnfs_layoutdriver_type {
 
        struct pnfs_layout_segment * (*alloc_lseg) (struct pnfs_layout_hdr *layoutid, struct nfs4_layoutget_res *lgr, gfp_t gfp_flags);
        void (*free_lseg) (struct pnfs_layout_segment *lseg);
+       void (*add_lseg) (struct pnfs_layout_hdr *layoutid,
+                       struct pnfs_layout_segment *lseg,
+                       struct list_head *free_me);
 
        void (*return_range) (struct pnfs_layout_hdr *lo,
                              struct pnfs_layout_range *range);
@@ -285,6 +288,14 @@ struct pnfs_layout_segment *pnfs_update_layout(struct inode *ino,
                                               gfp_t gfp_flags);
 void pnfs_clear_layoutreturn_waitbit(struct pnfs_layout_hdr *lo);
 
+void pnfs_generic_layout_insert_lseg(struct pnfs_layout_hdr *lo,
+                  struct pnfs_layout_segment *lseg,
+                  bool (*is_after)(const struct pnfs_layout_range *lseg_range,
+                          const struct pnfs_layout_range *old),
+                  bool (*do_merge)(struct pnfs_layout_segment *lseg,
+                          struct pnfs_layout_segment *old),
+                  struct list_head *free_me);
+
 void nfs4_deviceid_mark_client_invalid(struct nfs_client *clp);
 int pnfs_read_done_resend_to_mds(struct nfs_pgio_header *);
 int pnfs_write_done_resend_to_mds(struct nfs_pgio_header *);