pnfsblock: note written INVAL areas for layoutcommit

Signed-off-by: Peng Tao <peng_tao@emc.com>
Signed-off-by: Fred Isaman <iisaman@citi.umich.edu>
Signed-off-by: Benny Halevy <bhalevy@panasas.com>
Signed-off-by: Benny Halevy <bhalevy@tonian.com>
Signed-off-by: Jim Rees <rees@umich.edu>
Signed-off-by: Trond Myklebust <Trond.Myklebust@netapp.com>

3 files changed, 129 insertions, 0 deletions

diff --git a/fs/nfs/blocklayout/blocklayout.c b/fs/nfs/blocklayout/blocklayout.c
index 2e373826db80..21efef7c2fd2 100644
--- a/fs/nfs/blocklayout/blocklayout.c
+++ b/fs/nfs/blocklayout/blocklayout.c
@@ -329,6 +329,30 @@ out:
         return PNFS_NOT_ATTEMPTED;
 }
 
+static void mark_extents_written(struct pnfs_block_layout *bl,
+                                 __u64 offset, __u32 count)
+{
+        sector_t isect, end;
+        struct pnfs_block_extent *be;
+
+        dprintk("%s(%llu, %u)\n", __func__, offset, count);
+        if (count == 0)
+                return;
+        isect = (offset & (long)(PAGE_CACHE_MASK)) >> SECTOR_SHIFT;
+        end = (offset + count + PAGE_CACHE_SIZE - 1) & (long)(PAGE_CACHE_MASK);
+        end >>= SECTOR_SHIFT;
+        while (isect < end) {
+                sector_t len;
+                be = bl_find_get_extent(bl, isect, NULL);
+                BUG_ON(!be); /* FIXME */
+                len = min(end, be->be_f_offset + be->be_length) - isect;
+                if (be->be_state == PNFS_BLOCK_INVALID_DATA)
+                        bl_mark_for_commit(be, isect, len); /* What if fails? */
+                isect += len;
+                bl_put_extent(be);
+        }
+}
+
 /* This is basically copied from mpage_end_io_read */
 static void bl_end_io_write(struct bio *bio, int err)
 {
@@ -355,6 +379,14 @@ static void bl_write_cleanup(struct work_struct *work)
         dprintk("%s enter\n", __func__);
         task = container_of(work, struct rpc_task, u.tk_work);
         wdata = container_of(task, struct nfs_write_data, task);
+        if (!wdata->task.tk_status) {
+                /* Marks for LAYOUTCOMMIT */
+                /* BUG - this should be called after each bio, not after
+                 * all finish, unless have some way of storing success/failure
+                 */
+                mark_extents_written(BLK_LSEG2EXT(wdata->lseg),
+                                     wdata->args.offset, wdata->args.count);
+        }
         pnfs_ld_write_done(wdata);
 }
 
diff --git a/fs/nfs/blocklayout/blocklayout.h b/fs/nfs/blocklayout/blocklayout.h
index 6a703b79c33d..f27d827960a3 100644
--- a/fs/nfs/blocklayout/blocklayout.h
+++ b/fs/nfs/blocklayout/blocklayout.h
@@ -201,5 +201,7 @@ void clean_pnfs_block_layoutupdate(struct pnfs_block_layout *bl,
                                    int status);
 int bl_add_merge_extent(struct pnfs_block_layout *bl,
                          struct pnfs_block_extent *new);
+int bl_mark_for_commit(struct pnfs_block_extent *be,
+                        sector_t offset, sector_t length);
 
 #endif /* FS_NFS_NFS4BLOCKLAYOUT_H */
diff --git a/fs/nfs/blocklayout/extents.c b/fs/nfs/blocklayout/extents.c
index 7521940dcca5..19fa7b0b8c00 100644
--- a/fs/nfs/blocklayout/extents.c
+++ b/fs/nfs/blocklayout/extents.c
@@ -217,6 +217,48 @@ int bl_is_sector_init(struct pnfs_inval_markings *marks, sector_t isect)
         return rv;
 }
 
+/* Assume start, end already sector aligned */
+static int
+_range_has_tag(struct my_tree *tree, u64 start, u64 end, int32_t tag)
+{
+        struct pnfs_inval_tracking *pos;
+        u64 expect = 0;
+
+        dprintk("%s(%llu, %llu, %i) enter\n", __func__, start, end, tag);
+        list_for_each_entry_reverse(pos, &tree->mtt_stub, it_link) {
+                if (pos->it_sector >= end)
+                        continue;
+                if (!expect) {
+                        if ((pos->it_sector == end - tree->mtt_step_size) &&
+                            (pos->it_tags & (1 << tag))) {
+                                expect = pos->it_sector - tree->mtt_step_size;
+                                if (pos->it_sector < tree->mtt_step_size || expect < start)
+                                        return 1;
+                                continue;
+                        } else {
+                                return 0;
+                        }
+                }
+                if (pos->it_sector != expect || !(pos->it_tags & (1 << tag)))
+                        return 0;
+                expect -= tree->mtt_step_size;
+                if (expect < start)
+                        return 1;
+        }
+        return 0;
+}
+
+static int is_range_written(struct pnfs_inval_markings *marks,
+                            sector_t start, sector_t end)
+{
+        int rv;
+
+        spin_lock(&marks->im_lock);
+        rv = _range_has_tag(&marks->im_tree, start, end, EXTENT_WRITTEN);
+        spin_unlock(&marks->im_lock);
+        return rv;
+}
+
 /* Marks sectors in [offest, offset_length) as having been initialized.
  * All lengths are step-aligned, where step is min(pagesize, blocksize).
  * Notes where partial block is initialized, and helps prepare it for
@@ -396,6 +438,59 @@ static void add_to_commitlist(struct pnfs_block_layout *bl,
         print_clist(clist, bl->bl_count);
 }
 
+/* Note the range described by offset, length is guaranteed to be contained
+ * within be.
+ */
+int bl_mark_for_commit(struct pnfs_block_extent *be,
+                    sector_t offset, sector_t length)
+{
+        sector_t new_end, end = offset + length;
+        struct pnfs_block_short_extent *new;
+        struct pnfs_block_layout *bl = container_of(be->be_inval,
+                                                    struct pnfs_block_layout,
+                                                    bl_inval);
+
+        new = kmalloc(sizeof(*new), GFP_NOFS);
+        if (!new)
+                return -ENOMEM;
+
+        mark_written_sectors(be->be_inval, offset, length);
+        /* We want to add the range to commit list, but it must be
+         * block-normalized, and verified that the normalized range has
+         * been entirely written to disk.
+         */
+        new->bse_f_offset = offset;
+        offset = normalize(offset, bl->bl_blocksize);
+        if (offset < new->bse_f_offset) {
+                if (is_range_written(be->be_inval, offset, new->bse_f_offset))
+                        new->bse_f_offset = offset;
+                else
+                        new->bse_f_offset = offset + bl->bl_blocksize;
+        }
+        new_end = normalize_up(end, bl->bl_blocksize);
+        if (end < new_end) {
+                if (is_range_written(be->be_inval, end, new_end))
+                        end = new_end;
+                else
+                        end = new_end - bl->bl_blocksize;
+        }
+        if (end <= new->bse_f_offset) {
+                kfree(new);
+                return 0;
+        }
+        new->bse_length = end - new->bse_f_offset;
+        new->bse_devid = be->be_devid;
+        new->bse_mdev = be->be_mdev;
+
+        spin_lock(&bl->bl_ext_lock);
+        /* new will be freed, either by add_to_commitlist if it decides not
+         * to use it, or after LAYOUTCOMMIT uses it in the commitlist.
+         */
+        add_to_commitlist(bl, new);
+        spin_unlock(&bl->bl_ext_lock);
+        return 0;
+}
+
 static void print_bl_extent(struct pnfs_block_extent *be)
 {
         dprintk("PRINT EXTENT extent %p\n", be);