pnfsblock: xdr decode pnfs_block_layout4
authorFred Isaman <iisaman@citi.umich.edu>
Sun, 31 Jul 2011 00:52:47 +0000 (20:52 -0400)
committerTrond Myklebust <Trond.Myklebust@netapp.com>
Sun, 31 Jul 2011 16:18:16 +0000 (12:18 -0400)
XDR decodes the block layout payload sent in LAYOUTGET result, storing
the result in an extent list.

[pnfsblock: get rid of deprecated xdr macros]
Signed-off-by: Jim Rees <rees@umich.edu>
Signed-off-by: Fred Isaman <iisaman@citi.umich.edu>
[pnfsblock: fix bug getting pnfs_layout_type in translate_devid().]
Signed-off-by: Tao Guo <guotao@nrchpc.ac.cn>
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>
fs/nfs/blocklayout/blocklayoutdev.c

index b23fe601d1c95d2876ebd8a16c84cfb3cc9afc53..a83b393fb01c3231e7575e47768b909800bb88a0 100644 (file)
 
 #define NFSDBG_FACILITY         NFSDBG_PNFS_LD
 
+static int decode_sector_number(__be32 **rp, sector_t *sp)
+{
+       uint64_t s;
+
+       *rp = xdr_decode_hyper(*rp, &s);
+       if (s & 0x1ff) {
+               printk(KERN_WARNING "%s: sector not aligned\n", __func__);
+               return -1;
+       }
+       *sp = s >> SECTOR_SHIFT;
+       return 0;
+}
+
 /* Open a block_device by device number. */
 struct block_device *nfs4_blkdev_get(dev_t dev)
 {
@@ -197,10 +210,201 @@ out:
        return rv;
 }
 
+/* Map deviceid returned by the server to constructed block_device */
+static struct block_device *translate_devid(struct pnfs_layout_hdr *lo,
+                                           struct nfs4_deviceid *id)
+{
+       struct block_device *rv = NULL;
+       struct block_mount_id *mid;
+       struct pnfs_block_dev *dev;
+
+       dprintk("%s enter, lo=%p, id=%p\n", __func__, lo, id);
+       mid = BLK_ID(lo);
+       spin_lock(&mid->bm_lock);
+       list_for_each_entry(dev, &mid->bm_devlist, bm_node) {
+               if (memcmp(id->data, dev->bm_mdevid.data,
+                          NFS4_DEVICEID4_SIZE) == 0) {
+                       rv = dev->bm_mdev;
+                       goto out;
+               }
+       }
+ out:
+       spin_unlock(&mid->bm_lock);
+       dprintk("%s returning %p\n", __func__, rv);
+       return rv;
+}
+
+/* Tracks info needed to ensure extents in layout obey constraints of spec */
+struct layout_verification {
+       u32 mode;       /* R or RW */
+       u64 start;      /* Expected start of next non-COW extent */
+       u64 inval;      /* Start of INVAL coverage */
+       u64 cowread;    /* End of COW read coverage */
+};
+
+/* Verify the extent meets the layout requirements of the pnfs-block draft,
+ * section 2.3.1.
+ */
+static int verify_extent(struct pnfs_block_extent *be,
+                        struct layout_verification *lv)
+{
+       if (lv->mode == IOMODE_READ) {
+               if (be->be_state == PNFS_BLOCK_READWRITE_DATA ||
+                   be->be_state == PNFS_BLOCK_INVALID_DATA)
+                       return -EIO;
+               if (be->be_f_offset != lv->start)
+                       return -EIO;
+               lv->start += be->be_length;
+               return 0;
+       }
+       /* lv->mode == IOMODE_RW */
+       if (be->be_state == PNFS_BLOCK_READWRITE_DATA) {
+               if (be->be_f_offset != lv->start)
+                       return -EIO;
+               if (lv->cowread > lv->start)
+                       return -EIO;
+               lv->start += be->be_length;
+               lv->inval = lv->start;
+               return 0;
+       } else if (be->be_state == PNFS_BLOCK_INVALID_DATA) {
+               if (be->be_f_offset != lv->start)
+                       return -EIO;
+               lv->start += be->be_length;
+               return 0;
+       } else if (be->be_state == PNFS_BLOCK_READ_DATA) {
+               if (be->be_f_offset > lv->start)
+                       return -EIO;
+               if (be->be_f_offset < lv->inval)
+                       return -EIO;
+               if (be->be_f_offset < lv->cowread)
+                       return -EIO;
+               /* It looks like you might want to min this with lv->start,
+                * but you really don't.
+                */
+               lv->inval = lv->inval + be->be_length;
+               lv->cowread = be->be_f_offset + be->be_length;
+               return 0;
+       } else
+               return -EIO;
+}
+
+/* XDR decode pnfs_block_layout4 structure */
 int
 nfs4_blk_process_layoutget(struct pnfs_layout_hdr *lo,
                           struct nfs4_layoutget_res *lgr, gfp_t gfp_flags)
 {
-       /* STUB */
-       return -EIO;
+       struct pnfs_block_layout *bl = BLK_LO2EXT(lo);
+       int i, status = -EIO;
+       uint32_t count;
+       struct pnfs_block_extent *be = NULL, *save;
+       struct xdr_stream stream;
+       struct xdr_buf buf;
+       struct page *scratch;
+       __be32 *p;
+       struct layout_verification lv = {
+               .mode = lgr->range.iomode,
+               .start = lgr->range.offset >> SECTOR_SHIFT,
+               .inval = lgr->range.offset >> SECTOR_SHIFT,
+               .cowread = lgr->range.offset >> SECTOR_SHIFT,
+       };
+       LIST_HEAD(extents);
+
+       dprintk("---> %s\n", __func__);
+
+       scratch = alloc_page(gfp_flags);
+       if (!scratch)
+               return -ENOMEM;
+
+       xdr_init_decode_pages(&stream, &buf, lgr->layoutp->pages, lgr->layoutp->len);
+       xdr_set_scratch_buffer(&stream, page_address(scratch), PAGE_SIZE);
+
+       p = xdr_inline_decode(&stream, 4);
+       if (unlikely(!p))
+               goto out_err;
+
+       count = be32_to_cpup(p++);
+
+       dprintk("%s enter, number of extents %i\n", __func__, count);
+       p = xdr_inline_decode(&stream, (28 + NFS4_DEVICEID4_SIZE) * count);
+       if (unlikely(!p))
+               goto out_err;
+
+       /* Decode individual extents, putting them in temporary
+        * staging area until whole layout is decoded to make error
+        * recovery easier.
+        */
+       for (i = 0; i < count; i++) {
+               be = bl_alloc_extent();
+               if (!be) {
+                       status = -ENOMEM;
+                       goto out_err;
+               }
+               memcpy(&be->be_devid, p, NFS4_DEVICEID4_SIZE);
+               p += XDR_QUADLEN(NFS4_DEVICEID4_SIZE);
+               be->be_mdev = translate_devid(lo, &be->be_devid);
+               if (!be->be_mdev)
+                       goto out_err;
+
+               /* The next three values are read in as bytes,
+                * but stored as 512-byte sector lengths
+                */
+               if (decode_sector_number(&p, &be->be_f_offset) < 0)
+                       goto out_err;
+               if (decode_sector_number(&p, &be->be_length) < 0)
+                       goto out_err;
+               if (decode_sector_number(&p, &be->be_v_offset) < 0)
+                       goto out_err;
+               be->be_state = be32_to_cpup(p++);
+               if (be->be_state == PNFS_BLOCK_INVALID_DATA)
+                       be->be_inval = &bl->bl_inval;
+               if (verify_extent(be, &lv)) {
+                       dprintk("%s verify failed\n", __func__);
+                       goto out_err;
+               }
+               list_add_tail(&be->be_node, &extents);
+       }
+       if (lgr->range.offset + lgr->range.length !=
+                       lv.start << SECTOR_SHIFT) {
+               dprintk("%s Final length mismatch\n", __func__);
+               be = NULL;
+               goto out_err;
+       }
+       if (lv.start < lv.cowread) {
+               dprintk("%s Final uncovered COW extent\n", __func__);
+               be = NULL;
+               goto out_err;
+       }
+       /* Extents decoded properly, now try to merge them in to
+        * existing layout extents.
+        */
+       spin_lock(&bl->bl_ext_lock);
+       list_for_each_entry_safe(be, save, &extents, be_node) {
+               list_del(&be->be_node);
+               status = bl_add_merge_extent(bl, be);
+               if (status) {
+                       spin_unlock(&bl->bl_ext_lock);
+                       /* This is a fairly catastrophic error, as the
+                        * entire layout extent lists are now corrupted.
+                        * We should have some way to distinguish this.
+                        */
+                       be = NULL;
+                       goto out_err;
+               }
+       }
+       spin_unlock(&bl->bl_ext_lock);
+       status = 0;
+ out:
+       __free_page(scratch);
+       dprintk("%s returns %i\n", __func__, status);
+       return status;
+
+ out_err:
+       bl_put_extent(be);
+       while (!list_empty(&extents)) {
+               be = list_first_entry(&extents, struct pnfs_block_extent,
+                                     be_node);
+               list_del(&be->be_node);
+               bl_put_extent(be);
+       }
+       goto out;
 }